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)
193 bv = bio_first_bvec_all(bio);
194 inode = bv->bv_page->mapping->host;
195 mirror_num = bio_ctrl->mirror_num;
197 /* Caller should ensure the bio has at least some range added */
198 ASSERT(bio->bi_iter.bi_size);
200 btrfs_bio(bio)->file_offset = page_offset(bv->bv_page) + bv->bv_offset;
202 if (!is_data_inode(inode))
203 btrfs_submit_metadata_bio(inode, bio, mirror_num);
204 else if (btrfs_op(bio) == BTRFS_MAP_WRITE)
205 btrfs_submit_data_write_bio(inode, bio, mirror_num);
207 btrfs_submit_data_read_bio(inode, bio, mirror_num,
208 bio_ctrl->compress_type);
210 /* The bio is owned by the bi_end_io handler now */
211 bio_ctrl->bio = NULL;
215 * Submit or fail the current bio in an extent_page_data structure.
217 static void submit_write_bio(struct extent_page_data *epd, int ret)
219 struct bio *bio = epd->bio_ctrl.bio;
226 bio->bi_status = errno_to_blk_status(ret);
228 /* The bio is owned by the bi_end_io handler now */
229 epd->bio_ctrl.bio = NULL;
231 submit_one_bio(&epd->bio_ctrl);
235 int __init extent_state_cache_init(void)
237 extent_state_cache = kmem_cache_create("btrfs_extent_state",
238 sizeof(struct extent_state), 0,
239 SLAB_MEM_SPREAD, NULL);
240 if (!extent_state_cache)
245 int __init extent_io_init(void)
247 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
248 sizeof(struct extent_buffer), 0,
249 SLAB_MEM_SPREAD, NULL);
250 if (!extent_buffer_cache)
253 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
254 offsetof(struct btrfs_bio, bio),
256 goto free_buffer_cache;
258 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
264 bioset_exit(&btrfs_bioset);
267 kmem_cache_destroy(extent_buffer_cache);
268 extent_buffer_cache = NULL;
272 void __cold extent_state_cache_exit(void)
274 btrfs_extent_state_leak_debug_check();
275 kmem_cache_destroy(extent_state_cache);
278 void __cold extent_io_exit(void)
281 * Make sure all delayed rcu free are flushed before we
285 kmem_cache_destroy(extent_buffer_cache);
286 bioset_exit(&btrfs_bioset);
290 * For the file_extent_tree, we want to hold the inode lock when we lookup and
291 * update the disk_i_size, but lockdep will complain because our io_tree we hold
292 * the tree lock and get the inode lock when setting delalloc. These two things
293 * are unrelated, so make a class for the file_extent_tree so we don't get the
294 * two locking patterns mixed up.
296 static struct lock_class_key file_extent_tree_class;
298 void extent_io_tree_init(struct btrfs_fs_info *fs_info,
299 struct extent_io_tree *tree, unsigned int owner,
302 tree->fs_info = fs_info;
303 tree->state = RB_ROOT;
304 tree->dirty_bytes = 0;
305 spin_lock_init(&tree->lock);
306 tree->private_data = private_data;
308 if (owner == IO_TREE_INODE_FILE_EXTENT)
309 lockdep_set_class(&tree->lock, &file_extent_tree_class);
312 void extent_io_tree_release(struct extent_io_tree *tree)
314 spin_lock(&tree->lock);
316 * Do a single barrier for the waitqueue_active check here, the state
317 * of the waitqueue should not change once extent_io_tree_release is
321 while (!RB_EMPTY_ROOT(&tree->state)) {
322 struct rb_node *node;
323 struct extent_state *state;
325 node = rb_first(&tree->state);
326 state = rb_entry(node, struct extent_state, rb_node);
327 rb_erase(&state->rb_node, &tree->state);
328 RB_CLEAR_NODE(&state->rb_node);
330 * btree io trees aren't supposed to have tasks waiting for
331 * changes in the flags of extent states ever.
333 ASSERT(!waitqueue_active(&state->wq));
334 free_extent_state(state);
336 cond_resched_lock(&tree->lock);
338 spin_unlock(&tree->lock);
341 static struct extent_state *alloc_extent_state(gfp_t mask)
343 struct extent_state *state;
346 * The given mask might be not appropriate for the slab allocator,
347 * drop the unsupported bits
349 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
350 state = kmem_cache_alloc(extent_state_cache, mask);
354 state->failrec = NULL;
355 RB_CLEAR_NODE(&state->rb_node);
356 btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states);
357 refcount_set(&state->refs, 1);
358 init_waitqueue_head(&state->wq);
359 trace_alloc_extent_state(state, mask, _RET_IP_);
363 void free_extent_state(struct extent_state *state)
367 if (refcount_dec_and_test(&state->refs)) {
368 WARN_ON(extent_state_in_tree(state));
369 btrfs_leak_debug_del(&leak_lock, &state->leak_list);
370 trace_free_extent_state(state, _RET_IP_);
371 kmem_cache_free(extent_state_cache, state);
376 * Search @tree for an entry that contains @offset. Such entry would have
377 * entry->start <= offset && entry->end >= offset.
379 * @tree: the tree to search
380 * @offset: offset that should fall within an entry in @tree
381 * @node_ret: pointer where new node should be anchored (used when inserting an
383 * @parent_ret: points to entry which would have been the parent of the entry,
386 * Return a pointer to the entry that contains @offset byte address and don't change
387 * @node_ret and @parent_ret.
389 * If no such entry exists, return pointer to entry that ends before @offset
390 * and fill parameters @node_ret and @parent_ret, ie. does not return NULL.
392 static inline struct rb_node *tree_search_for_insert(struct extent_io_tree *tree,
394 struct rb_node ***node_ret,
395 struct rb_node **parent_ret)
397 struct rb_root *root = &tree->state;
398 struct rb_node **node = &root->rb_node;
399 struct rb_node *prev = NULL;
400 struct tree_entry *entry;
404 entry = rb_entry(prev, struct tree_entry, rb_node);
406 if (offset < entry->start)
407 node = &(*node)->rb_left;
408 else if (offset > entry->end)
409 node = &(*node)->rb_right;
419 /* Search neighbors until we find the first one past the end */
420 while (prev && offset > entry->end) {
421 prev = rb_next(prev);
422 entry = rb_entry(prev, struct tree_entry, rb_node);
429 * Inexact rb-tree search, return the next entry if @offset is not found
431 static inline struct rb_node *tree_search(struct extent_io_tree *tree, u64 offset)
433 return tree_search_for_insert(tree, offset, NULL, NULL);
437 * Search offset in the tree or fill neighbor rbtree node pointers.
439 * @tree: the tree to search
440 * @offset: offset that should fall within an entry in @tree
441 * @next_ret: pointer to the first entry whose range ends after @offset
442 * @prev_ret: pointer to the first entry whose range begins before @offset
444 * Return a pointer to the entry that contains @offset byte address. If no
445 * such entry exists, then return NULL and fill @prev_ret and @next_ret.
446 * Otherwise return the found entry and other pointers are left untouched.
448 static struct rb_node *tree_search_prev_next(struct extent_io_tree *tree,
450 struct rb_node **prev_ret,
451 struct rb_node **next_ret)
453 struct rb_root *root = &tree->state;
454 struct rb_node **node = &root->rb_node;
455 struct rb_node *prev = NULL;
456 struct rb_node *orig_prev = NULL;
457 struct tree_entry *entry;
464 entry = rb_entry(prev, struct tree_entry, rb_node);
466 if (offset < entry->start)
467 node = &(*node)->rb_left;
468 else if (offset > entry->end)
469 node = &(*node)->rb_right;
475 while (prev && offset > entry->end) {
476 prev = rb_next(prev);
477 entry = rb_entry(prev, struct tree_entry, rb_node);
482 entry = rb_entry(prev, struct tree_entry, rb_node);
483 while (prev && offset < entry->start) {
484 prev = rb_prev(prev);
485 entry = rb_entry(prev, struct tree_entry, rb_node);
493 * utility function to look for merge candidates inside a given range.
494 * Any extents with matching state are merged together into a single
495 * extent in the tree. Extents with EXTENT_IO in their state field
496 * are not merged because the end_io handlers need to be able to do
497 * operations on them without sleeping (or doing allocations/splits).
499 * This should be called with the tree lock held.
501 static void merge_state(struct extent_io_tree *tree,
502 struct extent_state *state)
504 struct extent_state *other;
505 struct rb_node *other_node;
507 if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY))
510 other_node = rb_prev(&state->rb_node);
512 other = rb_entry(other_node, struct extent_state, rb_node);
513 if (other->end == state->start - 1 &&
514 other->state == state->state) {
515 if (tree->private_data &&
516 is_data_inode(tree->private_data))
517 btrfs_merge_delalloc_extent(tree->private_data,
519 state->start = other->start;
520 rb_erase(&other->rb_node, &tree->state);
521 RB_CLEAR_NODE(&other->rb_node);
522 free_extent_state(other);
525 other_node = rb_next(&state->rb_node);
527 other = rb_entry(other_node, struct extent_state, rb_node);
528 if (other->start == state->end + 1 &&
529 other->state == state->state) {
530 if (tree->private_data &&
531 is_data_inode(tree->private_data))
532 btrfs_merge_delalloc_extent(tree->private_data,
534 state->end = other->end;
535 rb_erase(&other->rb_node, &tree->state);
536 RB_CLEAR_NODE(&other->rb_node);
537 free_extent_state(other);
542 static void set_state_bits(struct extent_io_tree *tree,
543 struct extent_state *state, u32 bits,
544 struct extent_changeset *changeset);
547 * insert an extent_state struct into the tree. 'bits' are set on the
548 * struct before it is inserted.
550 * This may return -EEXIST if the extent is already there, in which case the
551 * state struct is freed.
553 * The tree lock is not taken internally. This is a utility function and
554 * probably isn't what you want to call (see set/clear_extent_bit).
556 static int insert_state(struct extent_io_tree *tree,
557 struct extent_state *state,
558 u32 bits, struct extent_changeset *changeset)
560 struct rb_node **node;
561 struct rb_node *parent;
562 const u64 end = state->end;
564 set_state_bits(tree, state, bits, changeset);
566 node = &tree->state.rb_node;
568 struct tree_entry *entry;
571 entry = rb_entry(parent, struct tree_entry, rb_node);
573 if (end < entry->start) {
574 node = &(*node)->rb_left;
575 } else if (end > entry->end) {
576 node = &(*node)->rb_right;
578 btrfs_err(tree->fs_info,
579 "found node %llu %llu on insert of %llu %llu",
580 entry->start, entry->end, state->start, end);
585 rb_link_node(&state->rb_node, parent, node);
586 rb_insert_color(&state->rb_node, &tree->state);
588 merge_state(tree, state);
593 * Insert state to @tree to the location given by @node and @parent.
595 static void insert_state_fast(struct extent_io_tree *tree,
596 struct extent_state *state, struct rb_node **node,
597 struct rb_node *parent, unsigned bits,
598 struct extent_changeset *changeset)
600 set_state_bits(tree, state, bits, changeset);
601 rb_link_node(&state->rb_node, parent, node);
602 rb_insert_color(&state->rb_node, &tree->state);
603 merge_state(tree, state);
607 * split a given extent state struct in two, inserting the preallocated
608 * struct 'prealloc' as the newly created second half. 'split' indicates an
609 * offset inside 'orig' where it should be split.
612 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
613 * are two extent state structs in the tree:
614 * prealloc: [orig->start, split - 1]
615 * orig: [ split, orig->end ]
617 * The tree locks are not taken by this function. They need to be held
620 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
621 struct extent_state *prealloc, u64 split)
623 struct rb_node *parent = NULL;
624 struct rb_node **node;
626 if (tree->private_data && is_data_inode(tree->private_data))
627 btrfs_split_delalloc_extent(tree->private_data, orig, split);
629 prealloc->start = orig->start;
630 prealloc->end = split - 1;
631 prealloc->state = orig->state;
634 parent = &orig->rb_node;
637 struct tree_entry *entry;
640 entry = rb_entry(parent, struct tree_entry, rb_node);
642 if (prealloc->end < entry->start) {
643 node = &(*node)->rb_left;
644 } else if (prealloc->end > entry->end) {
645 node = &(*node)->rb_right;
647 free_extent_state(prealloc);
652 rb_link_node(&prealloc->rb_node, parent, node);
653 rb_insert_color(&prealloc->rb_node, &tree->state);
658 static struct extent_state *next_state(struct extent_state *state)
660 struct rb_node *next = rb_next(&state->rb_node);
662 return rb_entry(next, struct extent_state, rb_node);
668 * utility function to clear some bits in an extent state struct.
669 * it will optionally wake up anyone waiting on this state (wake == 1).
671 * If no bits are set on the state struct after clearing things, the
672 * struct is freed and removed from the tree
674 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
675 struct extent_state *state,
677 struct extent_changeset *changeset)
679 struct extent_state *next;
680 u32 bits_to_clear = bits & ~EXTENT_CTLBITS;
683 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
684 u64 range = state->end - state->start + 1;
685 WARN_ON(range > tree->dirty_bytes);
686 tree->dirty_bytes -= range;
689 if (tree->private_data && is_data_inode(tree->private_data))
690 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
692 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
694 state->state &= ~bits_to_clear;
697 if (state->state == 0) {
698 next = next_state(state);
699 if (extent_state_in_tree(state)) {
700 rb_erase(&state->rb_node, &tree->state);
701 RB_CLEAR_NODE(&state->rb_node);
702 free_extent_state(state);
707 merge_state(tree, state);
708 next = next_state(state);
713 static struct extent_state *
714 alloc_extent_state_atomic(struct extent_state *prealloc)
717 prealloc = alloc_extent_state(GFP_ATOMIC);
722 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
724 btrfs_panic(tree->fs_info, err,
725 "locking error: extent tree was modified by another thread while locked");
729 * clear some bits on a range in the tree. This may require splitting
730 * or inserting elements in the tree, so the gfp mask is used to
731 * indicate which allocations or sleeping are allowed.
733 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
734 * the given range from the tree regardless of state (ie for truncate).
736 * the range [start, end] is inclusive.
738 * This takes the tree lock, and returns 0 on success and < 0 on error.
740 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
741 u32 bits, int wake, int delete,
742 struct extent_state **cached_state,
743 gfp_t mask, struct extent_changeset *changeset)
745 struct extent_state *state;
746 struct extent_state *cached;
747 struct extent_state *prealloc = NULL;
748 struct rb_node *node;
753 btrfs_debug_check_extent_io_range(tree, start, end);
754 trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits);
756 if (bits & EXTENT_DELALLOC)
757 bits |= EXTENT_NORESERVE;
760 bits |= ~EXTENT_CTLBITS;
762 if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY))
765 if (!prealloc && gfpflags_allow_blocking(mask)) {
767 * Don't care for allocation failure here because we might end
768 * up not needing the pre-allocated extent state at all, which
769 * is the case if we only have in the tree extent states that
770 * cover our input range and don't cover too any other range.
771 * If we end up needing a new extent state we allocate it later.
773 prealloc = alloc_extent_state(mask);
776 spin_lock(&tree->lock);
778 cached = *cached_state;
781 *cached_state = NULL;
785 if (cached && extent_state_in_tree(cached) &&
786 cached->start <= start && cached->end > start) {
788 refcount_dec(&cached->refs);
793 free_extent_state(cached);
796 * this search will find the extents that end after
799 node = tree_search(tree, start);
802 state = rb_entry(node, struct extent_state, rb_node);
804 if (state->start > end)
806 WARN_ON(state->end < start);
807 last_end = state->end;
809 /* the state doesn't have the wanted bits, go ahead */
810 if (!(state->state & bits)) {
811 state = next_state(state);
816 * | ---- desired range ---- |
818 * | ------------- state -------------- |
820 * We need to split the extent we found, and may flip
821 * bits on second half.
823 * If the extent we found extends past our range, we
824 * just split and search again. It'll get split again
825 * the next time though.
827 * If the extent we found is inside our range, we clear
828 * the desired bit on it.
831 if (state->start < start) {
832 prealloc = alloc_extent_state_atomic(prealloc);
834 err = split_state(tree, state, prealloc, start);
836 extent_io_tree_panic(tree, err);
841 if (state->end <= end) {
842 state = clear_state_bit(tree, state, bits, wake, changeset);
848 * | ---- desired range ---- |
850 * We need to split the extent, and clear the bit
853 if (state->start <= end && state->end > end) {
854 prealloc = alloc_extent_state_atomic(prealloc);
856 err = split_state(tree, state, prealloc, end + 1);
858 extent_io_tree_panic(tree, err);
863 clear_state_bit(tree, prealloc, bits, wake, changeset);
869 state = clear_state_bit(tree, state, bits, wake, changeset);
871 if (last_end == (u64)-1)
873 start = last_end + 1;
874 if (start <= end && state && !need_resched())
880 spin_unlock(&tree->lock);
881 if (gfpflags_allow_blocking(mask))
886 spin_unlock(&tree->lock);
888 free_extent_state(prealloc);
894 static void wait_on_state(struct extent_io_tree *tree,
895 struct extent_state *state)
896 __releases(tree->lock)
897 __acquires(tree->lock)
900 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
901 spin_unlock(&tree->lock);
903 spin_lock(&tree->lock);
904 finish_wait(&state->wq, &wait);
908 * waits for one or more bits to clear on a range in the state tree.
909 * The range [start, end] is inclusive.
910 * The tree lock is taken by this function
912 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
915 struct extent_state *state;
916 struct rb_node *node;
918 btrfs_debug_check_extent_io_range(tree, start, end);
920 spin_lock(&tree->lock);
924 * this search will find all the extents that end after
927 node = tree_search(tree, start);
932 state = rb_entry(node, struct extent_state, rb_node);
934 if (state->start > end)
937 if (state->state & bits) {
938 start = state->start;
939 refcount_inc(&state->refs);
940 wait_on_state(tree, state);
941 free_extent_state(state);
944 start = state->end + 1;
949 if (!cond_resched_lock(&tree->lock)) {
950 node = rb_next(node);
955 spin_unlock(&tree->lock);
958 static void set_state_bits(struct extent_io_tree *tree,
959 struct extent_state *state,
960 u32 bits, struct extent_changeset *changeset)
962 u32 bits_to_set = bits & ~EXTENT_CTLBITS;
965 if (tree->private_data && is_data_inode(tree->private_data))
966 btrfs_set_delalloc_extent(tree->private_data, state, bits);
968 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
969 u64 range = state->end - state->start + 1;
970 tree->dirty_bytes += range;
972 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
974 state->state |= bits_to_set;
977 static void cache_state_if_flags(struct extent_state *state,
978 struct extent_state **cached_ptr,
981 if (cached_ptr && !(*cached_ptr)) {
982 if (!flags || (state->state & flags)) {
984 refcount_inc(&state->refs);
989 static void cache_state(struct extent_state *state,
990 struct extent_state **cached_ptr)
992 return cache_state_if_flags(state, cached_ptr,
993 EXTENT_LOCKED | EXTENT_BOUNDARY);
997 * set some bits on a range in the tree. This may require allocations or
998 * sleeping, so the gfp mask is used to indicate what is allowed.
1000 * If any of the exclusive bits are set, this will fail with -EEXIST if some
1001 * part of the range already has the desired bits set. The start of the
1002 * existing range is returned in failed_start in this case.
1004 * [start, end] is inclusive This takes the tree lock.
1006 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits,
1007 u32 exclusive_bits, u64 *failed_start,
1008 struct extent_state **cached_state, gfp_t mask,
1009 struct extent_changeset *changeset)
1011 struct extent_state *state;
1012 struct extent_state *prealloc = NULL;
1013 struct rb_node *node;
1015 struct rb_node *parent;
1020 btrfs_debug_check_extent_io_range(tree, start, end);
1021 trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits);
1024 ASSERT(failed_start);
1026 ASSERT(failed_start == NULL);
1028 if (!prealloc && gfpflags_allow_blocking(mask)) {
1030 * Don't care for allocation failure here because we might end
1031 * up not needing the pre-allocated extent state at all, which
1032 * is the case if we only have in the tree extent states that
1033 * cover our input range and don't cover too any other range.
1034 * If we end up needing a new extent state we allocate it later.
1036 prealloc = alloc_extent_state(mask);
1039 spin_lock(&tree->lock);
1040 if (cached_state && *cached_state) {
1041 state = *cached_state;
1042 if (state->start <= start && state->end > start &&
1043 extent_state_in_tree(state)) {
1044 node = &state->rb_node;
1049 * this search will find all the extents that end after
1052 node = tree_search_for_insert(tree, start, &p, &parent);
1054 prealloc = alloc_extent_state_atomic(prealloc);
1056 prealloc->start = start;
1057 prealloc->end = end;
1058 insert_state_fast(tree, prealloc, p, parent, bits, changeset);
1059 cache_state(prealloc, cached_state);
1063 state = rb_entry(node, struct extent_state, rb_node);
1065 last_start = state->start;
1066 last_end = state->end;
1069 * | ---- desired range ---- |
1072 * Just lock what we found and keep going
1074 if (state->start == start && state->end <= end) {
1075 if (state->state & exclusive_bits) {
1076 *failed_start = state->start;
1081 set_state_bits(tree, state, bits, changeset);
1082 cache_state(state, cached_state);
1083 merge_state(tree, state);
1084 if (last_end == (u64)-1)
1086 start = last_end + 1;
1087 state = next_state(state);
1088 if (start < end && state && state->start == start &&
1095 * | ---- desired range ---- |
1098 * | ------------- state -------------- |
1100 * We need to split the extent we found, and may flip bits on
1103 * If the extent we found extends past our
1104 * range, we just split and search again. It'll get split
1105 * again the next time though.
1107 * If the extent we found is inside our range, we set the
1108 * desired bit on it.
1110 if (state->start < start) {
1111 if (state->state & exclusive_bits) {
1112 *failed_start = start;
1118 * If this extent already has all the bits we want set, then
1119 * skip it, not necessary to split it or do anything with it.
1121 if ((state->state & bits) == bits) {
1122 start = state->end + 1;
1123 cache_state(state, cached_state);
1127 prealloc = alloc_extent_state_atomic(prealloc);
1129 err = split_state(tree, state, prealloc, start);
1131 extent_io_tree_panic(tree, err);
1136 if (state->end <= end) {
1137 set_state_bits(tree, state, bits, changeset);
1138 cache_state(state, cached_state);
1139 merge_state(tree, state);
1140 if (last_end == (u64)-1)
1142 start = last_end + 1;
1143 state = next_state(state);
1144 if (start < end && state && state->start == start &&
1151 * | ---- desired range ---- |
1152 * | state | or | state |
1154 * There's a hole, we need to insert something in it and
1155 * ignore the extent we found.
1157 if (state->start > start) {
1159 if (end < last_start)
1162 this_end = last_start - 1;
1164 prealloc = alloc_extent_state_atomic(prealloc);
1168 * Avoid to free 'prealloc' if it can be merged with
1171 prealloc->start = start;
1172 prealloc->end = this_end;
1173 err = insert_state(tree, prealloc, bits, changeset);
1175 extent_io_tree_panic(tree, err);
1177 cache_state(prealloc, cached_state);
1179 start = this_end + 1;
1183 * | ---- desired range ---- |
1185 * We need to split the extent, and set the bit
1188 if (state->start <= end && state->end > end) {
1189 if (state->state & exclusive_bits) {
1190 *failed_start = start;
1195 prealloc = alloc_extent_state_atomic(prealloc);
1197 err = split_state(tree, state, prealloc, end + 1);
1199 extent_io_tree_panic(tree, err);
1201 set_state_bits(tree, prealloc, bits, changeset);
1202 cache_state(prealloc, cached_state);
1203 merge_state(tree, prealloc);
1211 spin_unlock(&tree->lock);
1212 if (gfpflags_allow_blocking(mask))
1217 spin_unlock(&tree->lock);
1219 free_extent_state(prealloc);
1226 * convert_extent_bit - convert all bits in a given range from one bit to
1228 * @tree: the io tree to search
1229 * @start: the start offset in bytes
1230 * @end: the end offset in bytes (inclusive)
1231 * @bits: the bits to set in this range
1232 * @clear_bits: the bits to clear in this range
1233 * @cached_state: state that we're going to cache
1235 * This will go through and set bits for the given range. If any states exist
1236 * already in this range they are set with the given bit and cleared of the
1237 * clear_bits. This is only meant to be used by things that are mergeable, ie
1238 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1239 * boundary bits like LOCK.
1241 * All allocations are done with GFP_NOFS.
1243 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1244 u32 bits, u32 clear_bits,
1245 struct extent_state **cached_state)
1247 struct extent_state *state;
1248 struct extent_state *prealloc = NULL;
1249 struct rb_node *node;
1251 struct rb_node *parent;
1255 bool first_iteration = true;
1257 btrfs_debug_check_extent_io_range(tree, start, end);
1258 trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits,
1264 * Best effort, don't worry if extent state allocation fails
1265 * here for the first iteration. We might have a cached state
1266 * that matches exactly the target range, in which case no
1267 * extent state allocations are needed. We'll only know this
1268 * after locking the tree.
1270 prealloc = alloc_extent_state(GFP_NOFS);
1271 if (!prealloc && !first_iteration)
1275 spin_lock(&tree->lock);
1276 if (cached_state && *cached_state) {
1277 state = *cached_state;
1278 if (state->start <= start && state->end > start &&
1279 extent_state_in_tree(state)) {
1280 node = &state->rb_node;
1286 * this search will find all the extents that end after
1289 node = tree_search_for_insert(tree, start, &p, &parent);
1291 prealloc = alloc_extent_state_atomic(prealloc);
1296 prealloc->start = start;
1297 prealloc->end = end;
1298 insert_state_fast(tree, prealloc, p, parent, bits, NULL);
1299 cache_state(prealloc, cached_state);
1303 state = rb_entry(node, struct extent_state, rb_node);
1305 last_start = state->start;
1306 last_end = state->end;
1309 * | ---- desired range ---- |
1312 * Just lock what we found and keep going
1314 if (state->start == start && state->end <= end) {
1315 set_state_bits(tree, state, bits, NULL);
1316 cache_state(state, cached_state);
1317 state = clear_state_bit(tree, state, clear_bits, 0, NULL);
1318 if (last_end == (u64)-1)
1320 start = last_end + 1;
1321 if (start < end && state && state->start == start &&
1328 * | ---- desired range ---- |
1331 * | ------------- state -------------- |
1333 * We need to split the extent we found, and may flip bits on
1336 * If the extent we found extends past our
1337 * range, we just split and search again. It'll get split
1338 * again the next time though.
1340 * If the extent we found is inside our range, we set the
1341 * desired bit on it.
1343 if (state->start < start) {
1344 prealloc = alloc_extent_state_atomic(prealloc);
1349 err = split_state(tree, state, prealloc, start);
1351 extent_io_tree_panic(tree, err);
1355 if (state->end <= end) {
1356 set_state_bits(tree, state, bits, NULL);
1357 cache_state(state, cached_state);
1358 state = clear_state_bit(tree, state, clear_bits, 0, NULL);
1359 if (last_end == (u64)-1)
1361 start = last_end + 1;
1362 if (start < end && state && state->start == start &&
1369 * | ---- desired range ---- |
1370 * | state | or | state |
1372 * There's a hole, we need to insert something in it and
1373 * ignore the extent we found.
1375 if (state->start > start) {
1377 if (end < last_start)
1380 this_end = last_start - 1;
1382 prealloc = alloc_extent_state_atomic(prealloc);
1389 * Avoid to free 'prealloc' if it can be merged with
1392 prealloc->start = start;
1393 prealloc->end = this_end;
1394 err = insert_state(tree, prealloc, bits, NULL);
1396 extent_io_tree_panic(tree, err);
1397 cache_state(prealloc, cached_state);
1399 start = this_end + 1;
1403 * | ---- desired range ---- |
1405 * We need to split the extent, and set the bit
1408 if (state->start <= end && state->end > end) {
1409 prealloc = alloc_extent_state_atomic(prealloc);
1415 err = split_state(tree, state, prealloc, end + 1);
1417 extent_io_tree_panic(tree, err);
1419 set_state_bits(tree, prealloc, bits, NULL);
1420 cache_state(prealloc, cached_state);
1421 clear_state_bit(tree, prealloc, clear_bits, 0, NULL);
1429 spin_unlock(&tree->lock);
1431 first_iteration = false;
1435 spin_unlock(&tree->lock);
1437 free_extent_state(prealloc);
1442 /* wrappers around set/clear extent bit */
1443 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1444 u32 bits, struct extent_changeset *changeset)
1447 * We don't support EXTENT_LOCKED yet, as current changeset will
1448 * record any bits changed, so for EXTENT_LOCKED case, it will
1449 * either fail with -EEXIST or changeset will record the whole
1452 BUG_ON(bits & EXTENT_LOCKED);
1454 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1458 int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end,
1461 return set_extent_bit(tree, start, end, bits, 0, NULL, NULL,
1465 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1466 u32 bits, int wake, int delete,
1467 struct extent_state **cached)
1469 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1470 cached, GFP_NOFS, NULL);
1473 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1474 u32 bits, struct extent_changeset *changeset)
1477 * Don't support EXTENT_LOCKED case, same reason as
1478 * set_record_extent_bits().
1480 BUG_ON(bits & EXTENT_LOCKED);
1482 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1487 * either insert or lock state struct between start and end use mask to tell
1488 * us if waiting is desired.
1490 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1491 struct extent_state **cached_state)
1497 err = set_extent_bit(tree, start, end, EXTENT_LOCKED,
1498 EXTENT_LOCKED, &failed_start,
1499 cached_state, GFP_NOFS, NULL);
1500 if (err == -EEXIST) {
1501 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1502 start = failed_start;
1505 WARN_ON(start > end);
1510 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1515 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1516 &failed_start, NULL, GFP_NOFS, NULL);
1517 if (err == -EEXIST) {
1518 if (failed_start > start)
1519 clear_extent_bit(tree, start, failed_start - 1,
1520 EXTENT_LOCKED, 1, 0, NULL);
1526 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1528 unsigned long index = start >> PAGE_SHIFT;
1529 unsigned long end_index = end >> PAGE_SHIFT;
1532 while (index <= end_index) {
1533 page = find_get_page(inode->i_mapping, index);
1534 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1535 clear_page_dirty_for_io(page);
1541 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1543 struct address_space *mapping = inode->i_mapping;
1544 unsigned long index = start >> PAGE_SHIFT;
1545 unsigned long end_index = end >> PAGE_SHIFT;
1546 struct folio *folio;
1548 while (index <= end_index) {
1549 folio = filemap_get_folio(mapping, index);
1550 filemap_dirty_folio(mapping, folio);
1551 folio_account_redirty(folio);
1552 index += folio_nr_pages(folio);
1557 /* find the first state struct with 'bits' set after 'start', and
1558 * return it. tree->lock must be held. NULL will returned if
1559 * nothing was found after 'start'
1561 static struct extent_state *
1562 find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits)
1564 struct rb_node *node;
1565 struct extent_state *state;
1568 * this search will find all the extents that end after
1571 node = tree_search(tree, start);
1576 state = rb_entry(node, struct extent_state, rb_node);
1577 if (state->end >= start && (state->state & bits))
1580 node = rb_next(node);
1589 * Find the first offset in the io tree with one or more @bits set.
1591 * Note: If there are multiple bits set in @bits, any of them will match.
1593 * Return 0 if we find something, and update @start_ret and @end_ret.
1594 * Return 1 if we found nothing.
1596 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1597 u64 *start_ret, u64 *end_ret, u32 bits,
1598 struct extent_state **cached_state)
1600 struct extent_state *state;
1603 spin_lock(&tree->lock);
1604 if (cached_state && *cached_state) {
1605 state = *cached_state;
1606 if (state->end == start - 1 && extent_state_in_tree(state)) {
1607 while ((state = next_state(state)) != NULL) {
1608 if (state->state & bits)
1611 free_extent_state(*cached_state);
1612 *cached_state = NULL;
1615 free_extent_state(*cached_state);
1616 *cached_state = NULL;
1619 state = find_first_extent_bit_state(tree, start, bits);
1622 cache_state_if_flags(state, cached_state, 0);
1623 *start_ret = state->start;
1624 *end_ret = state->end;
1628 spin_unlock(&tree->lock);
1633 * Find a contiguous area of bits
1635 * @tree: io tree to check
1636 * @start: offset to start the search from
1637 * @start_ret: the first offset we found with the bits set
1638 * @end_ret: the final contiguous range of the bits that were set
1639 * @bits: bits to look for
1641 * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges
1642 * to set bits appropriately, and then merge them again. During this time it
1643 * will drop the tree->lock, so use this helper if you want to find the actual
1644 * contiguous area for given bits. We will search to the first bit we find, and
1645 * then walk down the tree until we find a non-contiguous area. The area
1646 * returned will be the full contiguous area with the bits set.
1648 int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start,
1649 u64 *start_ret, u64 *end_ret, u32 bits)
1651 struct extent_state *state;
1654 spin_lock(&tree->lock);
1655 state = find_first_extent_bit_state(tree, start, bits);
1657 *start_ret = state->start;
1658 *end_ret = state->end;
1659 while ((state = next_state(state)) != NULL) {
1660 if (state->start > (*end_ret + 1))
1662 *end_ret = state->end;
1666 spin_unlock(&tree->lock);
1671 * Find the first range that has @bits not set. This range could start before
1674 * @tree: the tree to search
1675 * @start: offset at/after which the found extent should start
1676 * @start_ret: records the beginning of the range
1677 * @end_ret: records the end of the range (inclusive)
1678 * @bits: the set of bits which must be unset
1680 * Since unallocated range is also considered one which doesn't have the bits
1681 * set it's possible that @end_ret contains -1, this happens in case the range
1682 * spans (last_range_end, end of device]. In this case it's up to the caller to
1683 * trim @end_ret to the appropriate size.
1685 void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start,
1686 u64 *start_ret, u64 *end_ret, u32 bits)
1688 struct extent_state *state;
1689 struct rb_node *node, *prev = NULL, *next;
1691 spin_lock(&tree->lock);
1693 /* Find first extent with bits cleared */
1695 node = tree_search_prev_next(tree, start, &prev, &next);
1696 if (!node && !next && !prev) {
1698 * Tree is completely empty, send full range and let
1699 * caller deal with it
1704 } else if (!node && !next) {
1706 * We are past the last allocated chunk, set start at
1707 * the end of the last extent.
1709 state = rb_entry(prev, struct extent_state, rb_node);
1710 *start_ret = state->end + 1;
1717 * At this point 'node' either contains 'start' or start is
1720 state = rb_entry(node, struct extent_state, rb_node);
1722 if (in_range(start, state->start, state->end - state->start + 1)) {
1723 if (state->state & bits) {
1725 * |--range with bits sets--|
1729 start = state->end + 1;
1732 * 'start' falls within a range that doesn't
1733 * have the bits set, so take its start as
1734 * the beginning of the desired range
1736 * |--range with bits cleared----|
1740 *start_ret = state->start;
1745 * |---prev range---|---hole/unset---|---node range---|
1751 * |---hole/unset--||--first node--|
1756 state = rb_entry(prev, struct extent_state,
1758 *start_ret = state->end + 1;
1767 * Find the longest stretch from start until an entry which has the
1771 state = rb_entry(node, struct extent_state, rb_node);
1772 if (state->end >= start && !(state->state & bits)) {
1773 *end_ret = state->end;
1775 *end_ret = state->start - 1;
1779 node = rb_next(node);
1784 spin_unlock(&tree->lock);
1788 * find a contiguous range of bytes in the file marked as delalloc, not
1789 * more than 'max_bytes'. start and end are used to return the range,
1791 * true is returned if we find something, false if nothing was in the tree
1793 bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start,
1794 u64 *end, u64 max_bytes,
1795 struct extent_state **cached_state)
1797 struct rb_node *node;
1798 struct extent_state *state;
1799 u64 cur_start = *start;
1801 u64 total_bytes = 0;
1803 spin_lock(&tree->lock);
1806 * this search will find all the extents that end after
1809 node = tree_search(tree, cur_start);
1816 state = rb_entry(node, struct extent_state, rb_node);
1817 if (found && (state->start != cur_start ||
1818 (state->state & EXTENT_BOUNDARY))) {
1821 if (!(state->state & EXTENT_DELALLOC)) {
1827 *start = state->start;
1828 *cached_state = state;
1829 refcount_inc(&state->refs);
1833 cur_start = state->end + 1;
1834 node = rb_next(node);
1835 total_bytes += state->end - state->start + 1;
1836 if (total_bytes >= max_bytes)
1842 spin_unlock(&tree->lock);
1847 * Process one page for __process_pages_contig().
1849 * Return >0 if we hit @page == @locked_page.
1850 * Return 0 if we updated the page status.
1851 * Return -EGAIN if the we need to try again.
1852 * (For PAGE_LOCK case but got dirty page or page not belong to mapping)
1854 static int process_one_page(struct btrfs_fs_info *fs_info,
1855 struct address_space *mapping,
1856 struct page *page, struct page *locked_page,
1857 unsigned long page_ops, u64 start, u64 end)
1861 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
1862 len = end + 1 - start;
1864 if (page_ops & PAGE_SET_ORDERED)
1865 btrfs_page_clamp_set_ordered(fs_info, page, start, len);
1866 if (page_ops & PAGE_SET_ERROR)
1867 btrfs_page_clamp_set_error(fs_info, page, start, len);
1868 if (page_ops & PAGE_START_WRITEBACK) {
1869 btrfs_page_clamp_clear_dirty(fs_info, page, start, len);
1870 btrfs_page_clamp_set_writeback(fs_info, page, start, len);
1872 if (page_ops & PAGE_END_WRITEBACK)
1873 btrfs_page_clamp_clear_writeback(fs_info, page, start, len);
1875 if (page == locked_page)
1878 if (page_ops & PAGE_LOCK) {
1881 ret = btrfs_page_start_writer_lock(fs_info, page, start, len);
1884 if (!PageDirty(page) || page->mapping != mapping) {
1885 btrfs_page_end_writer_lock(fs_info, page, start, len);
1889 if (page_ops & PAGE_UNLOCK)
1890 btrfs_page_end_writer_lock(fs_info, page, start, len);
1894 static int __process_pages_contig(struct address_space *mapping,
1895 struct page *locked_page,
1896 u64 start, u64 end, unsigned long page_ops,
1899 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
1900 pgoff_t start_index = start >> PAGE_SHIFT;
1901 pgoff_t end_index = end >> PAGE_SHIFT;
1902 pgoff_t index = start_index;
1903 unsigned long nr_pages = end_index - start_index + 1;
1904 unsigned long pages_processed = 0;
1905 struct page *pages[16];
1909 if (page_ops & PAGE_LOCK) {
1910 ASSERT(page_ops == PAGE_LOCK);
1911 ASSERT(processed_end && *processed_end == start);
1914 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1915 mapping_set_error(mapping, -EIO);
1917 while (nr_pages > 0) {
1920 found_pages = find_get_pages_contig(mapping, index,
1921 min_t(unsigned long,
1922 nr_pages, ARRAY_SIZE(pages)), pages);
1923 if (found_pages == 0) {
1925 * Only if we're going to lock these pages, we can find
1926 * nothing at @index.
1928 ASSERT(page_ops & PAGE_LOCK);
1933 for (i = 0; i < found_pages; i++) {
1936 process_ret = process_one_page(fs_info, mapping,
1937 pages[i], locked_page, page_ops,
1939 if (process_ret < 0) {
1940 for (; i < found_pages; i++)
1948 nr_pages -= found_pages;
1949 index += found_pages;
1953 if (err && processed_end) {
1955 * Update @processed_end. I know this is awful since it has
1956 * two different return value patterns (inclusive vs exclusive).
1958 * But the exclusive pattern is necessary if @start is 0, or we
1959 * underflow and check against processed_end won't work as
1962 if (pages_processed)
1963 *processed_end = min(end,
1964 ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1);
1966 *processed_end = start;
1971 static noinline void __unlock_for_delalloc(struct inode *inode,
1972 struct page *locked_page,
1975 unsigned long index = start >> PAGE_SHIFT;
1976 unsigned long end_index = end >> PAGE_SHIFT;
1978 ASSERT(locked_page);
1979 if (index == locked_page->index && end_index == index)
1982 __process_pages_contig(inode->i_mapping, locked_page, start, end,
1986 static noinline int lock_delalloc_pages(struct inode *inode,
1987 struct page *locked_page,
1991 unsigned long index = delalloc_start >> PAGE_SHIFT;
1992 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1993 u64 processed_end = delalloc_start;
1996 ASSERT(locked_page);
1997 if (index == locked_page->index && index == end_index)
2000 ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start,
2001 delalloc_end, PAGE_LOCK, &processed_end);
2002 if (ret == -EAGAIN && processed_end > delalloc_start)
2003 __unlock_for_delalloc(inode, locked_page, delalloc_start,
2009 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
2010 * more than @max_bytes.
2012 * @start: The original start bytenr to search.
2013 * Will store the extent range start bytenr.
2014 * @end: The original end bytenr of the search range
2015 * Will store the extent range end bytenr.
2017 * Return true if we find a delalloc range which starts inside the original
2018 * range, and @start/@end will store the delalloc range start/end.
2020 * Return false if we can't find any delalloc range which starts inside the
2021 * original range, and @start/@end will be the non-delalloc range start/end.
2024 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
2025 struct page *locked_page, u64 *start,
2028 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2029 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2030 const u64 orig_start = *start;
2031 const u64 orig_end = *end;
2032 /* The sanity tests may not set a valid fs_info. */
2033 u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE;
2037 struct extent_state *cached_state = NULL;
2041 /* Caller should pass a valid @end to indicate the search range end */
2042 ASSERT(orig_end > orig_start);
2044 /* The range should at least cover part of the page */
2045 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
2046 orig_end <= page_offset(locked_page)));
2048 /* step one, find a bunch of delalloc bytes starting at start */
2049 delalloc_start = *start;
2051 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
2052 max_bytes, &cached_state);
2053 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
2054 *start = delalloc_start;
2056 /* @delalloc_end can be -1, never go beyond @orig_end */
2057 *end = min(delalloc_end, orig_end);
2058 free_extent_state(cached_state);
2063 * start comes from the offset of locked_page. We have to lock
2064 * pages in order, so we can't process delalloc bytes before
2067 if (delalloc_start < *start)
2068 delalloc_start = *start;
2071 * make sure to limit the number of pages we try to lock down
2073 if (delalloc_end + 1 - delalloc_start > max_bytes)
2074 delalloc_end = delalloc_start + max_bytes - 1;
2076 /* step two, lock all the pages after the page that has start */
2077 ret = lock_delalloc_pages(inode, locked_page,
2078 delalloc_start, delalloc_end);
2079 ASSERT(!ret || ret == -EAGAIN);
2080 if (ret == -EAGAIN) {
2081 /* some of the pages are gone, lets avoid looping by
2082 * shortening the size of the delalloc range we're searching
2084 free_extent_state(cached_state);
2085 cached_state = NULL;
2087 max_bytes = PAGE_SIZE;
2096 /* step three, lock the state bits for the whole range */
2097 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
2099 /* then test to make sure it is all still delalloc */
2100 ret = test_range_bit(tree, delalloc_start, delalloc_end,
2101 EXTENT_DELALLOC, 1, cached_state);
2103 unlock_extent_cached(tree, delalloc_start, delalloc_end,
2105 __unlock_for_delalloc(inode, locked_page,
2106 delalloc_start, delalloc_end);
2110 free_extent_state(cached_state);
2111 *start = delalloc_start;
2112 *end = delalloc_end;
2117 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2118 struct page *locked_page,
2119 u32 clear_bits, unsigned long page_ops)
2121 clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL);
2123 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
2124 start, end, page_ops, NULL);
2128 * count the number of bytes in the tree that have a given bit(s)
2129 * set. This can be fairly slow, except for EXTENT_DIRTY which is
2130 * cached. The total number found is returned.
2132 u64 count_range_bits(struct extent_io_tree *tree,
2133 u64 *start, u64 search_end, u64 max_bytes,
2134 u32 bits, int contig)
2136 struct rb_node *node;
2137 struct extent_state *state;
2138 u64 cur_start = *start;
2139 u64 total_bytes = 0;
2143 if (WARN_ON(search_end <= cur_start))
2146 spin_lock(&tree->lock);
2147 if (cur_start == 0 && bits == EXTENT_DIRTY) {
2148 total_bytes = tree->dirty_bytes;
2152 * this search will find all the extents that end after
2155 node = tree_search(tree, cur_start);
2160 state = rb_entry(node, struct extent_state, rb_node);
2161 if (state->start > search_end)
2163 if (contig && found && state->start > last + 1)
2165 if (state->end >= cur_start && (state->state & bits) == bits) {
2166 total_bytes += min(search_end, state->end) + 1 -
2167 max(cur_start, state->start);
2168 if (total_bytes >= max_bytes)
2171 *start = max(cur_start, state->start);
2175 } else if (contig && found) {
2178 node = rb_next(node);
2183 spin_unlock(&tree->lock);
2188 * set the private field for a given byte offset in the tree. If there isn't
2189 * an extent_state there already, this does nothing.
2191 int set_state_failrec(struct extent_io_tree *tree, u64 start,
2192 struct io_failure_record *failrec)
2194 struct rb_node *node;
2195 struct extent_state *state;
2198 spin_lock(&tree->lock);
2200 * this search will find all the extents that end after
2203 node = tree_search(tree, start);
2208 state = rb_entry(node, struct extent_state, rb_node);
2209 if (state->start != start) {
2213 state->failrec = failrec;
2215 spin_unlock(&tree->lock);
2219 struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start)
2221 struct rb_node *node;
2222 struct extent_state *state;
2223 struct io_failure_record *failrec;
2225 spin_lock(&tree->lock);
2227 * this search will find all the extents that end after
2230 node = tree_search(tree, start);
2232 failrec = ERR_PTR(-ENOENT);
2235 state = rb_entry(node, struct extent_state, rb_node);
2236 if (state->start != start) {
2237 failrec = ERR_PTR(-ENOENT);
2241 failrec = state->failrec;
2243 spin_unlock(&tree->lock);
2248 * searches a range in the state tree for a given mask.
2249 * If 'filled' == 1, this returns 1 only if every extent in the tree
2250 * has the bits set. Otherwise, 1 is returned if any bit in the
2251 * range is found set.
2253 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
2254 u32 bits, int filled, struct extent_state *cached)
2256 struct extent_state *state = NULL;
2257 struct rb_node *node;
2260 spin_lock(&tree->lock);
2261 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
2262 cached->end > start)
2263 node = &cached->rb_node;
2265 node = tree_search(tree, start);
2266 while (node && start <= end) {
2267 state = rb_entry(node, struct extent_state, rb_node);
2269 if (filled && state->start > start) {
2274 if (state->start > end)
2277 if (state->state & bits) {
2281 } else if (filled) {
2286 if (state->end == (u64)-1)
2289 start = state->end + 1;
2292 node = rb_next(node);
2299 spin_unlock(&tree->lock);
2303 int free_io_failure(struct extent_io_tree *failure_tree,
2304 struct extent_io_tree *io_tree,
2305 struct io_failure_record *rec)
2310 set_state_failrec(failure_tree, rec->start, NULL);
2311 ret = clear_extent_bits(failure_tree, rec->start,
2312 rec->start + rec->len - 1,
2313 EXTENT_LOCKED | EXTENT_DIRTY);
2317 ret = clear_extent_bits(io_tree, rec->start,
2318 rec->start + rec->len - 1,
2328 * this bypasses the standard btrfs submit functions deliberately, as
2329 * the standard behavior is to write all copies in a raid setup. here we only
2330 * want to write the one bad copy. so we do the mapping for ourselves and issue
2331 * submit_bio directly.
2332 * to avoid any synchronization issues, wait for the data after writing, which
2333 * actually prevents the read that triggered the error from finishing.
2334 * currently, there can be no more than two copies of every data bit. thus,
2335 * exactly one rewrite is required.
2337 static int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
2338 u64 length, u64 logical, struct page *page,
2339 unsigned int pg_offset, int mirror_num)
2341 struct btrfs_device *dev;
2342 struct bio_vec bvec;
2346 struct btrfs_io_context *bioc = NULL;
2349 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
2350 BUG_ON(!mirror_num);
2352 if (btrfs_repair_one_zone(fs_info, logical))
2355 map_length = length;
2358 * Avoid races with device replace and make sure our bioc has devices
2359 * associated to its stripes that don't go away while we are doing the
2360 * read repair operation.
2362 btrfs_bio_counter_inc_blocked(fs_info);
2363 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2365 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2366 * to update all raid stripes, but here we just want to correct
2367 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2368 * stripe's dev and sector.
2370 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2371 &map_length, &bioc, 0);
2373 goto out_counter_dec;
2374 ASSERT(bioc->mirror_num == 1);
2376 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2377 &map_length, &bioc, mirror_num);
2379 goto out_counter_dec;
2380 BUG_ON(mirror_num != bioc->mirror_num);
2383 sector = bioc->stripes[bioc->mirror_num - 1].physical >> 9;
2384 dev = bioc->stripes[bioc->mirror_num - 1].dev;
2385 btrfs_put_bioc(bioc);
2387 if (!dev || !dev->bdev ||
2388 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2390 goto out_counter_dec;
2393 bio_init(&bio, dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC);
2394 bio.bi_iter.bi_sector = sector;
2395 __bio_add_page(&bio, page, length, pg_offset);
2397 btrfsic_check_bio(&bio);
2398 ret = submit_bio_wait(&bio);
2400 /* try to remap that extent elsewhere? */
2401 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2402 goto out_bio_uninit;
2405 btrfs_info_rl_in_rcu(fs_info,
2406 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2408 rcu_str_deref(dev->name), sector);
2414 btrfs_bio_counter_dec(fs_info);
2418 int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num)
2420 struct btrfs_fs_info *fs_info = eb->fs_info;
2421 u64 start = eb->start;
2422 int i, num_pages = num_extent_pages(eb);
2425 if (sb_rdonly(fs_info->sb))
2428 for (i = 0; i < num_pages; i++) {
2429 struct page *p = eb->pages[i];
2431 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2432 start - page_offset(p), mirror_num);
2441 static int next_mirror(const struct io_failure_record *failrec, int cur_mirror)
2443 if (cur_mirror == failrec->num_copies)
2444 return cur_mirror + 1 - failrec->num_copies;
2445 return cur_mirror + 1;
2448 static int prev_mirror(const struct io_failure_record *failrec, int cur_mirror)
2450 if (cur_mirror == 1)
2451 return failrec->num_copies;
2452 return cur_mirror - 1;
2456 * each time an IO finishes, we do a fast check in the IO failure tree
2457 * to see if we need to process or clean up an io_failure_record
2459 int clean_io_failure(struct btrfs_fs_info *fs_info,
2460 struct extent_io_tree *failure_tree,
2461 struct extent_io_tree *io_tree, u64 start,
2462 struct page *page, u64 ino, unsigned int pg_offset)
2465 struct io_failure_record *failrec;
2466 struct extent_state *state;
2471 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2476 failrec = get_state_failrec(failure_tree, start);
2477 if (IS_ERR(failrec))
2480 BUG_ON(!failrec->this_mirror);
2482 if (sb_rdonly(fs_info->sb))
2485 spin_lock(&io_tree->lock);
2486 state = find_first_extent_bit_state(io_tree,
2489 spin_unlock(&io_tree->lock);
2491 if (!state || state->start > failrec->start ||
2492 state->end < failrec->start + failrec->len - 1)
2495 mirror = failrec->this_mirror;
2497 mirror = prev_mirror(failrec, mirror);
2498 repair_io_failure(fs_info, ino, start, failrec->len,
2499 failrec->logical, page, pg_offset, mirror);
2500 } while (mirror != failrec->failed_mirror);
2503 free_io_failure(failure_tree, io_tree, failrec);
2508 * Can be called when
2509 * - hold extent lock
2510 * - under ordered extent
2511 * - the inode is freeing
2513 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2515 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2516 struct io_failure_record *failrec;
2517 struct extent_state *state, *next;
2519 if (RB_EMPTY_ROOT(&failure_tree->state))
2522 spin_lock(&failure_tree->lock);
2523 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2525 if (state->start > end)
2528 ASSERT(state->end <= end);
2530 next = next_state(state);
2532 failrec = state->failrec;
2533 free_extent_state(state);
2538 spin_unlock(&failure_tree->lock);
2541 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
2542 struct btrfs_bio *bbio,
2543 unsigned int bio_offset)
2545 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2546 u64 start = bbio->file_offset + bio_offset;
2547 struct io_failure_record *failrec;
2548 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2549 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2550 const u32 sectorsize = fs_info->sectorsize;
2553 failrec = get_state_failrec(failure_tree, start);
2554 if (!IS_ERR(failrec)) {
2555 btrfs_debug(fs_info,
2556 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu",
2557 failrec->logical, failrec->start, failrec->len);
2559 * when data can be on disk more than twice, add to failrec here
2560 * (e.g. with a list for failed_mirror) to make
2561 * clean_io_failure() clean all those errors at once.
2563 ASSERT(failrec->this_mirror == bbio->mirror_num);
2564 ASSERT(failrec->len == fs_info->sectorsize);
2568 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2570 return ERR_PTR(-ENOMEM);
2572 failrec->start = start;
2573 failrec->len = sectorsize;
2574 failrec->failed_mirror = bbio->mirror_num;
2575 failrec->this_mirror = bbio->mirror_num;
2576 failrec->logical = (bbio->iter.bi_sector << SECTOR_SHIFT) + bio_offset;
2578 btrfs_debug(fs_info,
2579 "new io failure record logical %llu start %llu",
2580 failrec->logical, start);
2582 failrec->num_copies = btrfs_num_copies(fs_info, failrec->logical, sectorsize);
2583 if (failrec->num_copies == 1) {
2585 * We only have a single copy of the data, so don't bother with
2586 * all the retry and error correction code that follows. No
2587 * matter what the error is, it is very likely to persist.
2589 btrfs_debug(fs_info,
2590 "cannot repair logical %llu num_copies %d",
2591 failrec->logical, failrec->num_copies);
2593 return ERR_PTR(-EIO);
2596 /* Set the bits in the private failure tree */
2597 ret = set_extent_bits(failure_tree, start, start + sectorsize - 1,
2598 EXTENT_LOCKED | EXTENT_DIRTY);
2600 ret = set_state_failrec(failure_tree, start, failrec);
2601 /* Set the bits in the inode's tree */
2602 ret = set_extent_bits(tree, start, start + sectorsize - 1,
2604 } else if (ret < 0) {
2606 return ERR_PTR(ret);
2612 int btrfs_repair_one_sector(struct inode *inode, struct btrfs_bio *failed_bbio,
2613 u32 bio_offset, struct page *page, unsigned int pgoff,
2614 submit_bio_hook_t *submit_bio_hook)
2616 u64 start = failed_bbio->file_offset + bio_offset;
2617 struct io_failure_record *failrec;
2618 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2619 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2620 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2621 struct bio *failed_bio = &failed_bbio->bio;
2622 const int icsum = bio_offset >> fs_info->sectorsize_bits;
2623 struct bio *repair_bio;
2624 struct btrfs_bio *repair_bbio;
2626 btrfs_debug(fs_info,
2627 "repair read error: read error at %llu", start);
2629 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2631 failrec = btrfs_get_io_failure_record(inode, failed_bbio, bio_offset);
2632 if (IS_ERR(failrec))
2633 return PTR_ERR(failrec);
2636 * There are two premises:
2637 * a) deliver good data to the caller
2638 * b) correct the bad sectors on disk
2640 * Since we're only doing repair for one sector, we only need to get
2641 * a good copy of the failed sector and if we succeed, we have setup
2642 * everything for repair_io_failure to do the rest for us.
2644 failrec->this_mirror = next_mirror(failrec, failrec->this_mirror);
2645 if (failrec->this_mirror == failrec->failed_mirror) {
2646 btrfs_debug(fs_info,
2647 "failed to repair num_copies %d this_mirror %d failed_mirror %d",
2648 failrec->num_copies, failrec->this_mirror, failrec->failed_mirror);
2649 free_io_failure(failure_tree, tree, failrec);
2653 repair_bio = btrfs_bio_alloc(1);
2654 repair_bbio = btrfs_bio(repair_bio);
2655 repair_bbio->file_offset = start;
2656 repair_bio->bi_opf = REQ_OP_READ;
2657 repair_bio->bi_end_io = failed_bio->bi_end_io;
2658 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
2659 repair_bio->bi_private = failed_bio->bi_private;
2661 if (failed_bbio->csum) {
2662 const u32 csum_size = fs_info->csum_size;
2664 repair_bbio->csum = repair_bbio->csum_inline;
2665 memcpy(repair_bbio->csum,
2666 failed_bbio->csum + csum_size * icsum, csum_size);
2669 bio_add_page(repair_bio, page, failrec->len, pgoff);
2670 repair_bbio->iter = repair_bio->bi_iter;
2672 btrfs_debug(btrfs_sb(inode->i_sb),
2673 "repair read error: submitting new read to mirror %d",
2674 failrec->this_mirror);
2677 * At this point we have a bio, so any errors from submit_bio_hook()
2678 * will be handled by the endio on the repair_bio, so we can't return an
2681 submit_bio_hook(inode, repair_bio, failrec->this_mirror, 0);
2685 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
2687 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
2689 ASSERT(page_offset(page) <= start &&
2690 start + len <= page_offset(page) + PAGE_SIZE);
2693 if (fsverity_active(page->mapping->host) &&
2695 !PageUptodate(page) &&
2696 start < i_size_read(page->mapping->host) &&
2697 !fsverity_verify_page(page)) {
2698 btrfs_page_set_error(fs_info, page, start, len);
2700 btrfs_page_set_uptodate(fs_info, page, start, len);
2703 btrfs_page_clear_uptodate(fs_info, page, start, len);
2704 btrfs_page_set_error(fs_info, page, start, len);
2707 if (!btrfs_is_subpage(fs_info, page))
2710 btrfs_subpage_end_reader(fs_info, page, start, len);
2713 static void end_sector_io(struct page *page, u64 offset, bool uptodate)
2715 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2716 const u32 sectorsize = inode->root->fs_info->sectorsize;
2717 struct extent_state *cached = NULL;
2719 end_page_read(page, uptodate, offset, sectorsize);
2721 set_extent_uptodate(&inode->io_tree, offset,
2722 offset + sectorsize - 1, &cached, GFP_ATOMIC);
2723 unlock_extent_cached_atomic(&inode->io_tree, offset,
2724 offset + sectorsize - 1, &cached);
2727 static void submit_data_read_repair(struct inode *inode,
2728 struct btrfs_bio *failed_bbio,
2729 u32 bio_offset, const struct bio_vec *bvec,
2730 unsigned int error_bitmap)
2732 const unsigned int pgoff = bvec->bv_offset;
2733 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2734 struct page *page = bvec->bv_page;
2735 const u64 start = page_offset(bvec->bv_page) + bvec->bv_offset;
2736 const u64 end = start + bvec->bv_len - 1;
2737 const u32 sectorsize = fs_info->sectorsize;
2738 const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits;
2741 BUG_ON(bio_op(&failed_bbio->bio) == REQ_OP_WRITE);
2743 /* This repair is only for data */
2744 ASSERT(is_data_inode(inode));
2746 /* We're here because we had some read errors or csum mismatch */
2747 ASSERT(error_bitmap);
2750 * We only get called on buffered IO, thus page must be mapped and bio
2751 * must not be cloned.
2753 ASSERT(page->mapping && !bio_flagged(&failed_bbio->bio, BIO_CLONED));
2755 /* Iterate through all the sectors in the range */
2756 for (i = 0; i < nr_bits; i++) {
2757 const unsigned int offset = i * sectorsize;
2758 bool uptodate = false;
2761 if (!(error_bitmap & (1U << i))) {
2763 * This sector has no error, just end the page read
2764 * and unlock the range.
2770 ret = btrfs_repair_one_sector(inode, failed_bbio,
2771 bio_offset + offset, page, pgoff + offset,
2772 btrfs_submit_data_read_bio);
2775 * We have submitted the read repair, the page release
2776 * will be handled by the endio function of the
2777 * submitted repair bio.
2778 * Thus we don't need to do any thing here.
2783 * Continue on failed repair, otherwise the remaining sectors
2784 * will not be properly unlocked.
2787 end_sector_io(page, start + offset, uptodate);
2791 /* lots and lots of room for performance fixes in the end_bio funcs */
2793 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2795 struct btrfs_inode *inode;
2796 const bool uptodate = (err == 0);
2799 ASSERT(page && page->mapping);
2800 inode = BTRFS_I(page->mapping->host);
2801 btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate);
2804 const struct btrfs_fs_info *fs_info = inode->root->fs_info;
2807 ASSERT(end + 1 - start <= U32_MAX);
2808 len = end + 1 - start;
2810 btrfs_page_clear_uptodate(fs_info, page, start, len);
2811 btrfs_page_set_error(fs_info, page, start, len);
2812 ret = err < 0 ? err : -EIO;
2813 mapping_set_error(page->mapping, ret);
2818 * after a writepage IO is done, we need to:
2819 * clear the uptodate bits on error
2820 * clear the writeback bits in the extent tree for this IO
2821 * end_page_writeback if the page has no more pending IO
2823 * Scheduling is not allowed, so the extent state tree is expected
2824 * to have one and only one object corresponding to this IO.
2826 static void end_bio_extent_writepage(struct bio *bio)
2828 int error = blk_status_to_errno(bio->bi_status);
2829 struct bio_vec *bvec;
2832 struct bvec_iter_all iter_all;
2833 bool first_bvec = true;
2835 ASSERT(!bio_flagged(bio, BIO_CLONED));
2836 bio_for_each_segment_all(bvec, bio, iter_all) {
2837 struct page *page = bvec->bv_page;
2838 struct inode *inode = page->mapping->host;
2839 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2840 const u32 sectorsize = fs_info->sectorsize;
2842 /* Our read/write should always be sector aligned. */
2843 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
2845 "partial page write in btrfs with offset %u and length %u",
2846 bvec->bv_offset, bvec->bv_len);
2847 else if (!IS_ALIGNED(bvec->bv_len, sectorsize))
2849 "incomplete page write with offset %u and length %u",
2850 bvec->bv_offset, bvec->bv_len);
2852 start = page_offset(page) + bvec->bv_offset;
2853 end = start + bvec->bv_len - 1;
2856 btrfs_record_physical_zoned(inode, start, bio);
2860 end_extent_writepage(page, error, start, end);
2862 btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len);
2869 * Record previously processed extent range
2871 * For endio_readpage_release_extent() to handle a full extent range, reducing
2872 * the extent io operations.
2874 struct processed_extent {
2875 struct btrfs_inode *inode;
2876 /* Start of the range in @inode */
2878 /* End of the range in @inode */
2884 * Try to release processed extent range
2886 * May not release the extent range right now if the current range is
2887 * contiguous to processed extent.
2889 * Will release processed extent when any of @inode, @uptodate, the range is
2890 * no longer contiguous to the processed range.
2892 * Passing @inode == NULL will force processed extent to be released.
2894 static void endio_readpage_release_extent(struct processed_extent *processed,
2895 struct btrfs_inode *inode, u64 start, u64 end,
2898 struct extent_state *cached = NULL;
2899 struct extent_io_tree *tree;
2901 /* The first extent, initialize @processed */
2902 if (!processed->inode)
2906 * Contiguous to processed extent, just uptodate the end.
2908 * Several things to notice:
2910 * - bio can be merged as long as on-disk bytenr is contiguous
2911 * This means we can have page belonging to other inodes, thus need to
2912 * check if the inode still matches.
2913 * - bvec can contain range beyond current page for multi-page bvec
2914 * Thus we need to do processed->end + 1 >= start check
2916 if (processed->inode == inode && processed->uptodate == uptodate &&
2917 processed->end + 1 >= start && end >= processed->end) {
2918 processed->end = end;
2922 tree = &processed->inode->io_tree;
2924 * Now we don't have range contiguous to the processed range, release
2925 * the processed range now.
2927 if (processed->uptodate && tree->track_uptodate)
2928 set_extent_uptodate(tree, processed->start, processed->end,
2929 &cached, GFP_ATOMIC);
2930 unlock_extent_cached_atomic(tree, processed->start, processed->end,
2934 /* Update processed to current range */
2935 processed->inode = inode;
2936 processed->start = start;
2937 processed->end = end;
2938 processed->uptodate = uptodate;
2941 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
2943 ASSERT(PageLocked(page));
2944 if (!btrfs_is_subpage(fs_info, page))
2947 ASSERT(PagePrivate(page));
2948 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
2952 * Find extent buffer for a givne bytenr.
2954 * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking
2957 static struct extent_buffer *find_extent_buffer_readpage(
2958 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
2960 struct extent_buffer *eb;
2963 * For regular sectorsize, we can use page->private to grab extent
2966 if (fs_info->nodesize >= PAGE_SIZE) {
2967 ASSERT(PagePrivate(page) && page->private);
2968 return (struct extent_buffer *)page->private;
2971 /* For subpage case, we need to lookup buffer radix tree */
2973 eb = radix_tree_lookup(&fs_info->buffer_radix,
2974 bytenr >> fs_info->sectorsize_bits);
2981 * after a readpage IO is done, we need to:
2982 * clear the uptodate bits on error
2983 * set the uptodate bits if things worked
2984 * set the page up to date if all extents in the tree are uptodate
2985 * clear the lock bit in the extent tree
2986 * unlock the page if there are no other extents locked for it
2988 * Scheduling is not allowed, so the extent state tree is expected
2989 * to have one and only one object corresponding to this IO.
2991 static void end_bio_extent_readpage(struct bio *bio)
2993 struct bio_vec *bvec;
2994 struct btrfs_bio *bbio = btrfs_bio(bio);
2995 struct extent_io_tree *tree, *failure_tree;
2996 struct processed_extent processed = { 0 };
2998 * The offset to the beginning of a bio, since one bio can never be
2999 * larger than UINT_MAX, u32 here is enough.
3003 struct bvec_iter_all iter_all;
3005 ASSERT(!bio_flagged(bio, BIO_CLONED));
3006 bio_for_each_segment_all(bvec, bio, iter_all) {
3007 bool uptodate = !bio->bi_status;
3008 struct page *page = bvec->bv_page;
3009 struct inode *inode = page->mapping->host;
3010 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3011 const u32 sectorsize = fs_info->sectorsize;
3012 unsigned int error_bitmap = (unsigned int)-1;
3013 bool repair = false;
3018 btrfs_debug(fs_info,
3019 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
3020 bio->bi_iter.bi_sector, bio->bi_status,
3022 tree = &BTRFS_I(inode)->io_tree;
3023 failure_tree = &BTRFS_I(inode)->io_failure_tree;
3026 * We always issue full-sector reads, but if some block in a
3027 * page fails to read, blk_update_request() will advance
3028 * bv_offset and adjust bv_len to compensate. Print a warning
3029 * for unaligned offsets, and an error if they don't add up to
3032 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
3034 "partial page read in btrfs with offset %u and length %u",
3035 bvec->bv_offset, bvec->bv_len);
3036 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
3039 "incomplete page read with offset %u and length %u",
3040 bvec->bv_offset, bvec->bv_len);
3042 start = page_offset(page) + bvec->bv_offset;
3043 end = start + bvec->bv_len - 1;
3046 mirror = bbio->mirror_num;
3047 if (likely(uptodate)) {
3048 if (is_data_inode(inode)) {
3049 error_bitmap = btrfs_verify_data_csum(bbio,
3050 bio_offset, page, start, end);
3054 if (btrfs_validate_metadata_buffer(bbio,
3055 page, start, end, mirror))
3060 if (likely(uptodate)) {
3061 loff_t i_size = i_size_read(inode);
3062 pgoff_t end_index = i_size >> PAGE_SHIFT;
3064 clean_io_failure(BTRFS_I(inode)->root->fs_info,
3065 failure_tree, tree, start, page,
3066 btrfs_ino(BTRFS_I(inode)), 0);
3069 * Zero out the remaining part if this range straddles
3072 * Here we should only zero the range inside the bvec,
3073 * not touch anything else.
3075 * NOTE: i_size is exclusive while end is inclusive.
3077 if (page->index == end_index && i_size <= end) {
3078 u32 zero_start = max(offset_in_page(i_size),
3079 offset_in_page(start));
3081 zero_user_segment(page, zero_start,
3082 offset_in_page(end) + 1);
3084 } else if (is_data_inode(inode)) {
3086 * Only try to repair bios that actually made it to a
3087 * device. If the bio failed to be submitted mirror
3088 * is 0 and we need to fail it without retrying.
3090 * This also includes the high level bios for compressed
3091 * extents - these never make it to a device and repair
3092 * is already handled on the lower compressed bio.
3097 struct extent_buffer *eb;
3099 eb = find_extent_buffer_readpage(fs_info, page, start);
3100 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3101 eb->read_mirror = mirror;
3102 atomic_dec(&eb->io_pages);
3107 * submit_data_read_repair() will handle all the good
3108 * and bad sectors, we just continue to the next bvec.
3110 submit_data_read_repair(inode, bbio, bio_offset, bvec,
3113 /* Update page status and unlock */
3114 end_page_read(page, uptodate, start, len);
3115 endio_readpage_release_extent(&processed, BTRFS_I(inode),
3116 start, end, PageUptodate(page));
3119 ASSERT(bio_offset + len > bio_offset);
3123 /* Release the last extent */
3124 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
3125 btrfs_bio_free_csum(bbio);
3130 * Populate every free slot in a provided array with pages.
3132 * @nr_pages: number of pages to allocate
3133 * @page_array: the array to fill with pages; any existing non-null entries in
3134 * the array will be skipped
3136 * Return: 0 if all pages were able to be allocated;
3137 * -ENOMEM otherwise, and the caller is responsible for freeing all
3138 * non-null page pointers in the array.
3140 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array)
3142 unsigned int allocated;
3144 for (allocated = 0; allocated < nr_pages;) {
3145 unsigned int last = allocated;
3147 allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array);
3149 if (allocated == nr_pages)
3153 * During this iteration, no page could be allocated, even
3154 * though alloc_pages_bulk_array() falls back to alloc_page()
3155 * if it could not bulk-allocate. So we must be out of memory.
3157 if (allocated == last)
3160 memalloc_retry_wait(GFP_NOFS);
3166 * Initialize the members up to but not including 'bio'. Use after allocating a
3167 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
3168 * 'bio' because use of __GFP_ZERO is not supported.
3170 static inline void btrfs_bio_init(struct btrfs_bio *bbio)
3172 memset(bbio, 0, offsetof(struct btrfs_bio, bio));
3176 * Allocate a btrfs_io_bio, with @nr_iovecs as maximum number of iovecs.
3178 * The bio allocation is backed by bioset and does not fail.
3180 struct bio *btrfs_bio_alloc(unsigned int nr_iovecs)
3184 ASSERT(0 < nr_iovecs && nr_iovecs <= BIO_MAX_VECS);
3185 bio = bio_alloc_bioset(NULL, nr_iovecs, 0, GFP_NOFS, &btrfs_bioset);
3186 btrfs_bio_init(btrfs_bio(bio));
3190 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size)
3193 struct btrfs_bio *bbio;
3195 ASSERT(offset <= UINT_MAX && size <= UINT_MAX);
3197 /* this will never fail when it's backed by a bioset */
3198 bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset);
3201 bbio = btrfs_bio(bio);
3202 btrfs_bio_init(bbio);
3204 bio_trim(bio, offset >> 9, size >> 9);
3205 bbio->iter = bio->bi_iter;
3210 * Attempt to add a page to bio
3212 * @bio_ctrl: record both the bio, and its bio_flags
3213 * @page: page to add to the bio
3214 * @disk_bytenr: offset of the new bio or to check whether we are adding
3215 * a contiguous page to the previous one
3216 * @size: portion of page that we want to write
3217 * @pg_offset: starting offset in the page
3218 * @compress_type: compression type of the current bio to see if we can merge them
3220 * Attempt to add a page to bio considering stripe alignment etc.
3222 * Return >= 0 for the number of bytes added to the bio.
3223 * Can return 0 if the current bio is already at stripe/zone boundary.
3224 * Return <0 for error.
3226 static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl,
3228 u64 disk_bytenr, unsigned int size,
3229 unsigned int pg_offset,
3230 enum btrfs_compression_type compress_type)
3232 struct bio *bio = bio_ctrl->bio;
3233 u32 bio_size = bio->bi_iter.bi_size;
3235 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
3236 bool contig = false;
3240 /* The limit should be calculated when bio_ctrl->bio is allocated */
3241 ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary);
3242 if (bio_ctrl->compress_type != compress_type)
3246 if (bio->bi_iter.bi_size == 0) {
3247 /* We can always add a page into an empty bio. */
3249 } else if (bio_ctrl->compress_type == BTRFS_COMPRESS_NONE) {
3250 struct bio_vec *bvec = bio_last_bvec_all(bio);
3253 * The contig check requires the following conditions to be met:
3254 * 1) The pages are belonging to the same inode
3255 * This is implied by the call chain.
3257 * 2) The range has adjacent logical bytenr
3259 * 3) The range has adjacent file offset
3260 * This is required for the usage of btrfs_bio->file_offset.
3262 if (bio_end_sector(bio) == sector &&
3263 page_offset(bvec->bv_page) + bvec->bv_offset +
3264 bvec->bv_len == page_offset(page) + pg_offset)
3268 * For compression, all IO should have its logical bytenr
3269 * set to the starting bytenr of the compressed extent.
3271 contig = bio->bi_iter.bi_sector == sector;
3277 real_size = min(bio_ctrl->len_to_oe_boundary,
3278 bio_ctrl->len_to_stripe_boundary) - bio_size;
3279 real_size = min(real_size, size);
3282 * If real_size is 0, never call bio_add_*_page(), as even size is 0,
3283 * bio will still execute its endio function on the page!
3288 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
3289 ret = bio_add_zone_append_page(bio, page, real_size, pg_offset);
3291 ret = bio_add_page(bio, page, real_size, pg_offset);
3296 static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl,
3297 struct btrfs_inode *inode, u64 file_offset)
3299 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3300 struct btrfs_io_geometry geom;
3301 struct btrfs_ordered_extent *ordered;
3302 struct extent_map *em;
3303 u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT);
3307 * Pages for compressed extent are never submitted to disk directly,
3308 * thus it has no real boundary, just set them to U32_MAX.
3310 * The split happens for real compressed bio, which happens in
3311 * btrfs_submit_compressed_read/write().
3313 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
3314 bio_ctrl->len_to_oe_boundary = U32_MAX;
3315 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3318 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
3321 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio),
3323 free_extent_map(em);
3327 if (geom.len > U32_MAX)
3328 bio_ctrl->len_to_stripe_boundary = U32_MAX;
3330 bio_ctrl->len_to_stripe_boundary = (u32)geom.len;
3332 if (bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) {
3333 bio_ctrl->len_to_oe_boundary = U32_MAX;
3337 /* Ordered extent not yet created, so we're good */
3338 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
3340 bio_ctrl->len_to_oe_boundary = U32_MAX;
3344 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
3345 ordered->disk_bytenr + ordered->disk_num_bytes - logical);
3346 btrfs_put_ordered_extent(ordered);
3350 static int alloc_new_bio(struct btrfs_inode *inode,
3351 struct btrfs_bio_ctrl *bio_ctrl,
3352 struct writeback_control *wbc,
3354 bio_end_io_t end_io_func,
3355 u64 disk_bytenr, u32 offset, u64 file_offset,
3356 enum btrfs_compression_type compress_type)
3358 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3362 bio = btrfs_bio_alloc(BIO_MAX_VECS);
3364 * For compressed page range, its disk_bytenr is always @disk_bytenr
3365 * passed in, no matter if we have added any range into previous bio.
3367 if (compress_type != BTRFS_COMPRESS_NONE)
3368 bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
3370 bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT;
3371 bio_ctrl->bio = bio;
3372 bio_ctrl->compress_type = compress_type;
3373 bio->bi_end_io = end_io_func;
3375 ret = calc_bio_boundaries(bio_ctrl, inode, file_offset);
3381 * For Zone append we need the correct block_device that we are
3382 * going to write to set in the bio to be able to respect the
3383 * hardware limitation. Look it up here:
3385 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
3386 struct btrfs_device *dev;
3388 dev = btrfs_zoned_get_device(fs_info, disk_bytenr,
3389 fs_info->sectorsize);
3395 bio_set_dev(bio, dev->bdev);
3398 * Otherwise pick the last added device to support
3399 * cgroup writeback. For multi-device file systems this
3400 * means blk-cgroup policies have to always be set on the
3401 * last added/replaced device. This is a bit odd but has
3402 * been like that for a long time.
3404 bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev);
3406 wbc_init_bio(wbc, bio);
3408 ASSERT(bio_op(bio) != REQ_OP_ZONE_APPEND);
3412 bio_ctrl->bio = NULL;
3413 bio->bi_status = errno_to_blk_status(ret);
3419 * @opf: bio REQ_OP_* and REQ_* flags as one value
3420 * @wbc: optional writeback control for io accounting
3421 * @page: page to add to the bio
3422 * @disk_bytenr: logical bytenr where the write will be
3423 * @size: portion of page that we want to write to
3424 * @pg_offset: offset of the new bio or to check whether we are adding
3425 * a contiguous page to the previous one
3426 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
3427 * @end_io_func: end_io callback for new bio
3428 * @mirror_num: desired mirror to read/write
3429 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
3430 * @compress_type: compress type for current bio
3432 static int submit_extent_page(blk_opf_t opf,
3433 struct writeback_control *wbc,
3434 struct btrfs_bio_ctrl *bio_ctrl,
3435 struct page *page, u64 disk_bytenr,
3436 size_t size, unsigned long pg_offset,
3437 bio_end_io_t end_io_func,
3438 enum btrfs_compression_type compress_type,
3439 bool force_bio_submit)
3442 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3443 unsigned int cur = pg_offset;
3447 ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE &&
3448 pg_offset + size <= PAGE_SIZE);
3449 if (force_bio_submit)
3450 submit_one_bio(bio_ctrl);
3452 while (cur < pg_offset + size) {
3453 u32 offset = cur - pg_offset;
3456 /* Allocate new bio if needed */
3457 if (!bio_ctrl->bio) {
3458 ret = alloc_new_bio(inode, bio_ctrl, wbc, opf,
3459 end_io_func, disk_bytenr, offset,
3460 page_offset(page) + cur,
3466 * We must go through btrfs_bio_add_page() to ensure each
3467 * page range won't cross various boundaries.
3469 if (compress_type != BTRFS_COMPRESS_NONE)
3470 added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr,
3471 size - offset, pg_offset + offset,
3474 added = btrfs_bio_add_page(bio_ctrl, page,
3475 disk_bytenr + offset, size - offset,
3476 pg_offset + offset, compress_type);
3478 /* Metadata page range should never be split */
3479 if (!is_data_inode(&inode->vfs_inode))
3480 ASSERT(added == 0 || added == size - offset);
3482 /* At least we added some page, update the account */
3484 wbc_account_cgroup_owner(wbc, page, added);
3486 /* We have reached boundary, submit right now */
3487 if (added < size - offset) {
3488 /* The bio should contain some page(s) */
3489 ASSERT(bio_ctrl->bio->bi_iter.bi_size);
3490 submit_one_bio(bio_ctrl);
3497 static int attach_extent_buffer_page(struct extent_buffer *eb,
3499 struct btrfs_subpage *prealloc)
3501 struct btrfs_fs_info *fs_info = eb->fs_info;
3505 * If the page is mapped to btree inode, we should hold the private
3506 * lock to prevent race.
3507 * For cloned or dummy extent buffers, their pages are not mapped and
3508 * will not race with any other ebs.
3511 lockdep_assert_held(&page->mapping->private_lock);
3513 if (fs_info->nodesize >= PAGE_SIZE) {
3514 if (!PagePrivate(page))
3515 attach_page_private(page, eb);
3517 WARN_ON(page->private != (unsigned long)eb);
3521 /* Already mapped, just free prealloc */
3522 if (PagePrivate(page)) {
3523 btrfs_free_subpage(prealloc);
3528 /* Has preallocated memory for subpage */
3529 attach_page_private(page, prealloc);
3531 /* Do new allocation to attach subpage */
3532 ret = btrfs_attach_subpage(fs_info, page,
3533 BTRFS_SUBPAGE_METADATA);
3537 int set_page_extent_mapped(struct page *page)
3539 struct btrfs_fs_info *fs_info;
3541 ASSERT(page->mapping);
3543 if (PagePrivate(page))
3546 fs_info = btrfs_sb(page->mapping->host->i_sb);
3548 if (btrfs_is_subpage(fs_info, page))
3549 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
3551 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
3555 void clear_page_extent_mapped(struct page *page)
3557 struct btrfs_fs_info *fs_info;
3559 ASSERT(page->mapping);
3561 if (!PagePrivate(page))
3564 fs_info = btrfs_sb(page->mapping->host->i_sb);
3565 if (btrfs_is_subpage(fs_info, page))
3566 return btrfs_detach_subpage(fs_info, page);
3568 detach_page_private(page);
3571 static struct extent_map *
3572 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
3573 u64 start, u64 len, struct extent_map **em_cached)
3575 struct extent_map *em;
3577 if (em_cached && *em_cached) {
3579 if (extent_map_in_tree(em) && start >= em->start &&
3580 start < extent_map_end(em)) {
3581 refcount_inc(&em->refs);
3585 free_extent_map(em);
3589 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
3590 if (em_cached && !IS_ERR(em)) {
3592 refcount_inc(&em->refs);
3598 * basic readpage implementation. Locked extent state structs are inserted
3599 * into the tree that are removed when the IO is done (by the end_io
3601 * XXX JDM: This needs looking at to ensure proper page locking
3602 * return 0 on success, otherwise return error
3604 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
3605 struct btrfs_bio_ctrl *bio_ctrl,
3606 blk_opf_t read_flags, u64 *prev_em_start)
3608 struct inode *inode = page->mapping->host;
3609 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3610 u64 start = page_offset(page);
3611 const u64 end = start + PAGE_SIZE - 1;
3614 u64 last_byte = i_size_read(inode);
3617 struct extent_map *em;
3619 size_t pg_offset = 0;
3621 size_t blocksize = inode->i_sb->s_blocksize;
3622 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3624 ret = set_page_extent_mapped(page);
3626 unlock_extent(tree, start, end);
3627 btrfs_page_set_error(fs_info, page, start, PAGE_SIZE);
3632 if (page->index == last_byte >> PAGE_SHIFT) {
3633 size_t zero_offset = offset_in_page(last_byte);
3636 iosize = PAGE_SIZE - zero_offset;
3637 memzero_page(page, zero_offset, iosize);
3640 begin_page_read(fs_info, page);
3641 while (cur <= end) {
3642 unsigned long this_bio_flag = 0;
3643 bool force_bio_submit = false;
3646 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
3647 if (cur >= last_byte) {
3648 struct extent_state *cached = NULL;
3650 iosize = PAGE_SIZE - pg_offset;
3651 memzero_page(page, pg_offset, iosize);
3652 set_extent_uptodate(tree, cur, cur + iosize - 1,
3654 unlock_extent_cached(tree, cur,
3655 cur + iosize - 1, &cached);
3656 end_page_read(page, true, cur, iosize);
3659 em = __get_extent_map(inode, page, pg_offset, cur,
3660 end - cur + 1, em_cached);
3662 unlock_extent(tree, cur, end);
3663 end_page_read(page, false, cur, end + 1 - cur);
3667 extent_offset = cur - em->start;
3668 BUG_ON(extent_map_end(em) <= cur);
3671 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3672 this_bio_flag = em->compress_type;
3674 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3675 cur_end = min(extent_map_end(em) - 1, end);
3676 iosize = ALIGN(iosize, blocksize);
3677 if (this_bio_flag != BTRFS_COMPRESS_NONE)
3678 disk_bytenr = em->block_start;
3680 disk_bytenr = em->block_start + extent_offset;
3681 block_start = em->block_start;
3682 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3683 block_start = EXTENT_MAP_HOLE;
3686 * If we have a file range that points to a compressed extent
3687 * and it's followed by a consecutive file range that points
3688 * to the same compressed extent (possibly with a different
3689 * offset and/or length, so it either points to the whole extent
3690 * or only part of it), we must make sure we do not submit a
3691 * single bio to populate the pages for the 2 ranges because
3692 * this makes the compressed extent read zero out the pages
3693 * belonging to the 2nd range. Imagine the following scenario:
3696 * [0 - 8K] [8K - 24K]
3699 * points to extent X, points to extent X,
3700 * offset 4K, length of 8K offset 0, length 16K
3702 * [extent X, compressed length = 4K uncompressed length = 16K]
3704 * If the bio to read the compressed extent covers both ranges,
3705 * it will decompress extent X into the pages belonging to the
3706 * first range and then it will stop, zeroing out the remaining
3707 * pages that belong to the other range that points to extent X.
3708 * So here we make sure we submit 2 bios, one for the first
3709 * range and another one for the third range. Both will target
3710 * the same physical extent from disk, but we can't currently
3711 * make the compressed bio endio callback populate the pages
3712 * for both ranges because each compressed bio is tightly
3713 * coupled with a single extent map, and each range can have
3714 * an extent map with a different offset value relative to the
3715 * uncompressed data of our extent and different lengths. This
3716 * is a corner case so we prioritize correctness over
3717 * non-optimal behavior (submitting 2 bios for the same extent).
3719 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3720 prev_em_start && *prev_em_start != (u64)-1 &&
3721 *prev_em_start != em->start)
3722 force_bio_submit = true;
3725 *prev_em_start = em->start;
3727 free_extent_map(em);
3730 /* we've found a hole, just zero and go on */
3731 if (block_start == EXTENT_MAP_HOLE) {
3732 struct extent_state *cached = NULL;
3734 memzero_page(page, pg_offset, iosize);
3736 set_extent_uptodate(tree, cur, cur + iosize - 1,
3738 unlock_extent_cached(tree, cur,
3739 cur + iosize - 1, &cached);
3740 end_page_read(page, true, cur, iosize);
3742 pg_offset += iosize;
3745 /* the get_extent function already copied into the page */
3746 if (test_range_bit(tree, cur, cur_end,
3747 EXTENT_UPTODATE, 1, NULL)) {
3748 unlock_extent(tree, cur, cur + iosize - 1);
3749 end_page_read(page, true, cur, iosize);
3751 pg_offset += iosize;
3754 /* we have an inline extent but it didn't get marked up
3755 * to date. Error out
3757 if (block_start == EXTENT_MAP_INLINE) {
3758 unlock_extent(tree, cur, cur + iosize - 1);
3759 end_page_read(page, false, cur, iosize);
3761 pg_offset += iosize;
3765 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
3766 bio_ctrl, page, disk_bytenr, iosize,
3767 pg_offset, end_bio_extent_readpage,
3768 this_bio_flag, force_bio_submit);
3771 * We have to unlock the remaining range, or the page
3772 * will never be unlocked.
3774 unlock_extent(tree, cur, end);
3775 end_page_read(page, false, cur, end + 1 - cur);
3779 pg_offset += iosize;
3785 int btrfs_read_folio(struct file *file, struct folio *folio)
3787 struct page *page = &folio->page;
3788 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3789 u64 start = page_offset(page);
3790 u64 end = start + PAGE_SIZE - 1;
3791 struct btrfs_bio_ctrl bio_ctrl = { 0 };
3794 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3796 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
3798 * If btrfs_do_readpage() failed we will want to submit the assembled
3799 * bio to do the cleanup.
3801 submit_one_bio(&bio_ctrl);
3805 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
3807 struct extent_map **em_cached,
3808 struct btrfs_bio_ctrl *bio_ctrl,
3811 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
3814 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
3816 for (index = 0; index < nr_pages; index++) {
3817 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
3818 REQ_RAHEAD, prev_em_start);
3819 put_page(pages[index]);
3824 * helper for __extent_writepage, doing all of the delayed allocation setup.
3826 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3827 * to write the page (copy into inline extent). In this case the IO has
3828 * been started and the page is already unlocked.
3830 * This returns 0 if all went well (page still locked)
3831 * This returns < 0 if there were errors (page still locked)
3833 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
3834 struct page *page, struct writeback_control *wbc)
3836 const u64 page_end = page_offset(page) + PAGE_SIZE - 1;
3837 u64 delalloc_start = page_offset(page);
3838 u64 delalloc_to_write = 0;
3839 /* How many pages are started by btrfs_run_delalloc_range() */
3840 unsigned long nr_written = 0;
3842 int page_started = 0;
3844 while (delalloc_start < page_end) {
3845 u64 delalloc_end = page_end;
3848 found = find_lock_delalloc_range(&inode->vfs_inode, page,
3852 delalloc_start = delalloc_end + 1;
3855 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3856 delalloc_end, &page_started, &nr_written, wbc);
3858 btrfs_page_set_error(inode->root->fs_info, page,
3859 page_offset(page), PAGE_SIZE);
3863 * delalloc_end is already one less than the total length, so
3864 * we don't subtract one from PAGE_SIZE
3866 delalloc_to_write += (delalloc_end - delalloc_start +
3867 PAGE_SIZE) >> PAGE_SHIFT;
3868 delalloc_start = delalloc_end + 1;
3870 if (wbc->nr_to_write < delalloc_to_write) {
3873 if (delalloc_to_write < thresh * 2)
3874 thresh = delalloc_to_write;
3875 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3879 /* Did btrfs_run_dealloc_range() already unlock and start the IO? */
3882 * We've unlocked the page, so we can't update the mapping's
3883 * writeback index, just update nr_to_write.
3885 wbc->nr_to_write -= nr_written;
3893 * Find the first byte we need to write.
3895 * For subpage, one page can contain several sectors, and
3896 * __extent_writepage_io() will just grab all extent maps in the page
3897 * range and try to submit all non-inline/non-compressed extents.
3899 * This is a big problem for subpage, we shouldn't re-submit already written
3901 * This function will lookup subpage dirty bit to find which range we really
3904 * Return the next dirty range in [@start, @end).
3905 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
3907 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
3908 struct page *page, u64 *start, u64 *end)
3910 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
3911 struct btrfs_subpage_info *spi = fs_info->subpage_info;
3912 u64 orig_start = *start;
3913 /* Declare as unsigned long so we can use bitmap ops */
3914 unsigned long flags;
3915 int range_start_bit;
3919 * For regular sector size == page size case, since one page only
3920 * contains one sector, we return the page offset directly.
3922 if (!btrfs_is_subpage(fs_info, page)) {
3923 *start = page_offset(page);
3924 *end = page_offset(page) + PAGE_SIZE;
3928 range_start_bit = spi->dirty_offset +
3929 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
3931 /* We should have the page locked, but just in case */
3932 spin_lock_irqsave(&subpage->lock, flags);
3933 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
3934 spi->dirty_offset + spi->bitmap_nr_bits);
3935 spin_unlock_irqrestore(&subpage->lock, flags);
3937 range_start_bit -= spi->dirty_offset;
3938 range_end_bit -= spi->dirty_offset;
3940 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
3941 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
3945 * helper for __extent_writepage. This calls the writepage start hooks,
3946 * and does the loop to map the page into extents and bios.
3948 * We return 1 if the IO is started and the page is unlocked,
3949 * 0 if all went well (page still locked)
3950 * < 0 if there were errors (page still locked)
3952 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
3954 struct writeback_control *wbc,
3955 struct extent_page_data *epd,
3959 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3960 u64 cur = page_offset(page);
3961 u64 end = cur + PAGE_SIZE - 1;
3964 struct extent_map *em;
3968 enum req_op op = REQ_OP_WRITE;
3969 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
3970 bool has_error = false;
3973 ret = btrfs_writepage_cow_fixup(page);
3975 /* Fixup worker will requeue */
3976 redirty_page_for_writepage(wbc, page);
3982 * we don't want to touch the inode after unlocking the page,
3983 * so we update the mapping writeback index now
3987 while (cur <= end) {
3990 u64 dirty_range_start = cur;
3991 u64 dirty_range_end;
3994 if (cur >= i_size) {
3995 btrfs_writepage_endio_finish_ordered(inode, page, cur,
3998 * This range is beyond i_size, thus we don't need to
3999 * bother writing back.
4000 * But we still need to clear the dirty subpage bit, or
4001 * the next time the page gets dirtied, we will try to
4002 * writeback the sectors with subpage dirty bits,
4003 * causing writeback without ordered extent.
4005 btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur);
4009 find_next_dirty_byte(fs_info, page, &dirty_range_start,
4011 if (cur < dirty_range_start) {
4012 cur = dirty_range_start;
4016 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
4018 btrfs_page_set_error(fs_info, page, cur, end - cur + 1);
4019 ret = PTR_ERR_OR_ZERO(em);
4026 extent_offset = cur - em->start;
4027 em_end = extent_map_end(em);
4028 ASSERT(cur <= em_end);
4030 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
4031 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
4032 block_start = em->block_start;
4033 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4034 disk_bytenr = em->block_start + extent_offset;
4037 * Note that em_end from extent_map_end() and dirty_range_end from
4038 * find_next_dirty_byte() are all exclusive
4040 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
4042 if (btrfs_use_zone_append(inode, em->block_start))
4043 op = REQ_OP_ZONE_APPEND;
4045 free_extent_map(em);
4049 * compressed and inline extents are written through other
4052 if (compressed || block_start == EXTENT_MAP_HOLE ||
4053 block_start == EXTENT_MAP_INLINE) {
4057 btrfs_writepage_endio_finish_ordered(inode,
4058 page, cur, cur + iosize - 1, true);
4059 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
4064 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
4065 if (!PageWriteback(page)) {
4066 btrfs_err(inode->root->fs_info,
4067 "page %lu not writeback, cur %llu end %llu",
4068 page->index, cur, end);
4072 * Although the PageDirty bit is cleared before entering this
4073 * function, subpage dirty bit is not cleared.
4074 * So clear subpage dirty bit here so next time we won't submit
4075 * page for range already written to disk.
4077 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
4079 ret = submit_extent_page(op | write_flags, wbc,
4080 &epd->bio_ctrl, page,
4081 disk_bytenr, iosize,
4082 cur - page_offset(page),
4083 end_bio_extent_writepage,
4090 btrfs_page_set_error(fs_info, page, cur, iosize);
4091 if (PageWriteback(page))
4092 btrfs_page_clear_writeback(fs_info, page, cur,
4100 * If we finish without problem, we should not only clear page dirty,
4101 * but also empty subpage dirty bits
4104 btrfs_page_assert_not_dirty(fs_info, page);
4112 * the writepage semantics are similar to regular writepage. extent
4113 * records are inserted to lock ranges in the tree, and as dirty areas
4114 * are found, they are marked writeback. Then the lock bits are removed
4115 * and the end_io handler clears the writeback ranges
4117 * Return 0 if everything goes well.
4118 * Return <0 for error.
4120 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
4121 struct extent_page_data *epd)
4123 struct folio *folio = page_folio(page);
4124 struct inode *inode = page->mapping->host;
4125 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4126 const u64 page_start = page_offset(page);
4127 const u64 page_end = page_start + PAGE_SIZE - 1;
4131 loff_t i_size = i_size_read(inode);
4132 unsigned long end_index = i_size >> PAGE_SHIFT;
4134 trace___extent_writepage(page, inode, wbc);
4136 WARN_ON(!PageLocked(page));
4138 btrfs_page_clear_error(btrfs_sb(inode->i_sb), page,
4139 page_offset(page), PAGE_SIZE);
4141 pg_offset = offset_in_page(i_size);
4142 if (page->index > end_index ||
4143 (page->index == end_index && !pg_offset)) {
4144 folio_invalidate(folio, 0, folio_size(folio));
4145 folio_unlock(folio);
4149 if (page->index == end_index)
4150 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
4152 ret = set_page_extent_mapped(page);
4158 if (!epd->extent_locked) {
4159 ret = writepage_delalloc(BTRFS_I(inode), page, wbc);
4166 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size,
4173 /* make sure the mapping tag for page dirty gets cleared */
4174 set_page_writeback(page);
4175 end_page_writeback(page);
4178 * Here we used to have a check for PageError() and then set @ret and
4179 * call end_extent_writepage().
4181 * But in fact setting @ret here will cause different error paths
4182 * between subpage and regular sectorsize.
4184 * For regular page size, we never submit current page, but only add
4185 * current page to current bio.
4186 * The bio submission can only happen in next page.
4187 * Thus if we hit the PageError() branch, @ret is already set to
4188 * non-zero value and will not get updated for regular sectorsize.
4190 * But for subpage case, it's possible we submit part of current page,
4191 * thus can get PageError() set by submitted bio of the same page,
4192 * while our @ret is still 0.
4194 * So here we unify the behavior and don't set @ret.
4195 * Error can still be properly passed to higher layer as page will
4196 * be set error, here we just don't handle the IO failure.
4198 * NOTE: This is just a hotfix for subpage.
4199 * The root fix will be properly ending ordered extent when we hit
4200 * an error during writeback.
4202 * But that needs a bigger refactoring, as we not only need to grab the
4203 * submitted OE, but also need to know exactly at which bytenr we hit
4205 * Currently the full page based __extent_writepage_io() is not
4208 if (PageError(page))
4209 end_extent_writepage(page, ret, page_start, page_end);
4210 if (epd->extent_locked) {
4212 * If epd->extent_locked, it's from extent_write_locked_range(),
4213 * the page can either be locked by lock_page() or
4214 * process_one_page().
4215 * Let btrfs_page_unlock_writer() handle both cases.
4218 btrfs_page_unlock_writer(fs_info, page, wbc->range_start,
4219 wbc->range_end + 1 - wbc->range_start);
4227 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
4229 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
4230 TASK_UNINTERRUPTIBLE);
4233 static void end_extent_buffer_writeback(struct extent_buffer *eb)
4235 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4236 smp_mb__after_atomic();
4237 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
4241 * Lock extent buffer status and pages for writeback.
4243 * May try to flush write bio if we can't get the lock.
4245 * Return 0 if the extent buffer doesn't need to be submitted.
4246 * (E.g. the extent buffer is not dirty)
4247 * Return >0 is the extent buffer is submitted to bio.
4248 * Return <0 if something went wrong, no page is locked.
4250 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
4251 struct extent_page_data *epd)
4253 struct btrfs_fs_info *fs_info = eb->fs_info;
4258 if (!btrfs_try_tree_write_lock(eb)) {
4259 submit_write_bio(epd, 0);
4261 btrfs_tree_lock(eb);
4264 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
4265 btrfs_tree_unlock(eb);
4269 submit_write_bio(epd, 0);
4273 wait_on_extent_buffer_writeback(eb);
4274 btrfs_tree_lock(eb);
4275 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
4277 btrfs_tree_unlock(eb);
4282 * We need to do this to prevent races in people who check if the eb is
4283 * under IO since we can end up having no IO bits set for a short period
4286 spin_lock(&eb->refs_lock);
4287 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4288 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4289 spin_unlock(&eb->refs_lock);
4290 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
4291 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4293 fs_info->dirty_metadata_batch);
4296 spin_unlock(&eb->refs_lock);
4299 btrfs_tree_unlock(eb);
4302 * Either we don't need to submit any tree block, or we're submitting
4304 * Subpage metadata doesn't use page locking at all, so we can skip
4307 if (!ret || fs_info->nodesize < PAGE_SIZE)
4310 num_pages = num_extent_pages(eb);
4311 for (i = 0; i < num_pages; i++) {
4312 struct page *p = eb->pages[i];
4314 if (!trylock_page(p)) {
4316 submit_write_bio(epd, 0);
4326 static void set_btree_ioerr(struct page *page, struct extent_buffer *eb)
4328 struct btrfs_fs_info *fs_info = eb->fs_info;
4330 btrfs_page_set_error(fs_info, page, eb->start, eb->len);
4331 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4335 * A read may stumble upon this buffer later, make sure that it gets an
4336 * error and knows there was an error.
4338 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4341 * We need to set the mapping with the io error as well because a write
4342 * error will flip the file system readonly, and then syncfs() will
4343 * return a 0 because we are readonly if we don't modify the err seq for
4346 mapping_set_error(page->mapping, -EIO);
4349 * If we error out, we should add back the dirty_metadata_bytes
4350 * to make it consistent.
4352 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4353 eb->len, fs_info->dirty_metadata_batch);
4356 * If writeback for a btree extent that doesn't belong to a log tree
4357 * failed, increment the counter transaction->eb_write_errors.
4358 * We do this because while the transaction is running and before it's
4359 * committing (when we call filemap_fdata[write|wait]_range against
4360 * the btree inode), we might have
4361 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
4362 * returns an error or an error happens during writeback, when we're
4363 * committing the transaction we wouldn't know about it, since the pages
4364 * can be no longer dirty nor marked anymore for writeback (if a
4365 * subsequent modification to the extent buffer didn't happen before the
4366 * transaction commit), which makes filemap_fdata[write|wait]_range not
4367 * able to find the pages tagged with SetPageError at transaction
4368 * commit time. So if this happens we must abort the transaction,
4369 * otherwise we commit a super block with btree roots that point to
4370 * btree nodes/leafs whose content on disk is invalid - either garbage
4371 * or the content of some node/leaf from a past generation that got
4372 * cowed or deleted and is no longer valid.
4374 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
4375 * not be enough - we need to distinguish between log tree extents vs
4376 * non-log tree extents, and the next filemap_fdatawait_range() call
4377 * will catch and clear such errors in the mapping - and that call might
4378 * be from a log sync and not from a transaction commit. Also, checking
4379 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
4380 * not done and would not be reliable - the eb might have been released
4381 * from memory and reading it back again means that flag would not be
4382 * set (since it's a runtime flag, not persisted on disk).
4384 * Using the flags below in the btree inode also makes us achieve the
4385 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
4386 * writeback for all dirty pages and before filemap_fdatawait_range()
4387 * is called, the writeback for all dirty pages had already finished
4388 * with errors - because we were not using AS_EIO/AS_ENOSPC,
4389 * filemap_fdatawait_range() would return success, as it could not know
4390 * that writeback errors happened (the pages were no longer tagged for
4393 switch (eb->log_index) {
4395 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
4398 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
4401 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
4404 BUG(); /* unexpected, logic error */
4409 * The endio specific version which won't touch any unsafe spinlock in endio
4412 static struct extent_buffer *find_extent_buffer_nolock(
4413 struct btrfs_fs_info *fs_info, u64 start)
4415 struct extent_buffer *eb;
4418 eb = radix_tree_lookup(&fs_info->buffer_radix,
4419 start >> fs_info->sectorsize_bits);
4420 if (eb && atomic_inc_not_zero(&eb->refs)) {
4429 * The endio function for subpage extent buffer write.
4431 * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback()
4432 * after all extent buffers in the page has finished their writeback.
4434 static void end_bio_subpage_eb_writepage(struct bio *bio)
4436 struct btrfs_fs_info *fs_info;
4437 struct bio_vec *bvec;
4438 struct bvec_iter_all iter_all;
4440 fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb);
4441 ASSERT(fs_info->nodesize < PAGE_SIZE);
4443 ASSERT(!bio_flagged(bio, BIO_CLONED));
4444 bio_for_each_segment_all(bvec, bio, iter_all) {
4445 struct page *page = bvec->bv_page;
4446 u64 bvec_start = page_offset(page) + bvec->bv_offset;
4447 u64 bvec_end = bvec_start + bvec->bv_len - 1;
4448 u64 cur_bytenr = bvec_start;
4450 ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize));
4452 /* Iterate through all extent buffers in the range */
4453 while (cur_bytenr <= bvec_end) {
4454 struct extent_buffer *eb;
4458 * Here we can't use find_extent_buffer(), as it may
4459 * try to lock eb->refs_lock, which is not safe in endio
4462 eb = find_extent_buffer_nolock(fs_info, cur_bytenr);
4465 cur_bytenr = eb->start + eb->len;
4467 ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags));
4468 done = atomic_dec_and_test(&eb->io_pages);
4471 if (bio->bi_status ||
4472 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4473 ClearPageUptodate(page);
4474 set_btree_ioerr(page, eb);
4477 btrfs_subpage_clear_writeback(fs_info, page, eb->start,
4479 end_extent_buffer_writeback(eb);
4481 * free_extent_buffer() will grab spinlock which is not
4482 * safe in endio context. Thus here we manually dec
4485 atomic_dec(&eb->refs);
4491 static void end_bio_extent_buffer_writepage(struct bio *bio)
4493 struct bio_vec *bvec;
4494 struct extent_buffer *eb;
4496 struct bvec_iter_all iter_all;
4498 ASSERT(!bio_flagged(bio, BIO_CLONED));
4499 bio_for_each_segment_all(bvec, bio, iter_all) {
4500 struct page *page = bvec->bv_page;
4502 eb = (struct extent_buffer *)page->private;
4504 done = atomic_dec_and_test(&eb->io_pages);
4506 if (bio->bi_status ||
4507 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
4508 ClearPageUptodate(page);
4509 set_btree_ioerr(page, eb);
4512 end_page_writeback(page);
4517 end_extent_buffer_writeback(eb);
4523 static void prepare_eb_write(struct extent_buffer *eb)
4526 unsigned long start;
4529 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
4530 atomic_set(&eb->io_pages, num_extent_pages(eb));
4532 /* Set btree blocks beyond nritems with 0 to avoid stale content */
4533 nritems = btrfs_header_nritems(eb);
4534 if (btrfs_header_level(eb) > 0) {
4535 end = btrfs_node_key_ptr_offset(nritems);
4536 memzero_extent_buffer(eb, end, eb->len - end);
4540 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
4542 start = btrfs_item_nr_offset(nritems);
4543 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb);
4544 memzero_extent_buffer(eb, start, end - start);
4549 * Unlike the work in write_one_eb(), we rely completely on extent locking.
4550 * Page locking is only utilized at minimum to keep the VMM code happy.
4552 static int write_one_subpage_eb(struct extent_buffer *eb,
4553 struct writeback_control *wbc,
4554 struct extent_page_data *epd)
4556 struct btrfs_fs_info *fs_info = eb->fs_info;
4557 struct page *page = eb->pages[0];
4558 blk_opf_t write_flags = wbc_to_write_flags(wbc);
4559 bool no_dirty_ebs = false;
4562 prepare_eb_write(eb);
4564 /* clear_page_dirty_for_io() in subpage helper needs page locked */
4566 btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len);
4568 /* Check if this is the last dirty bit to update nr_written */
4569 no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page,
4570 eb->start, eb->len);
4572 clear_page_dirty_for_io(page);
4574 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4575 &epd->bio_ctrl, page, eb->start, eb->len,
4576 eb->start - page_offset(page),
4577 end_bio_subpage_eb_writepage, 0, false);
4579 btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len);
4580 set_btree_ioerr(page, eb);
4583 if (atomic_dec_and_test(&eb->io_pages))
4584 end_extent_buffer_writeback(eb);
4589 * Submission finished without problem, if no range of the page is
4590 * dirty anymore, we have submitted a page. Update nr_written in wbc.
4597 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
4598 struct writeback_control *wbc,
4599 struct extent_page_data *epd)
4601 u64 disk_bytenr = eb->start;
4603 blk_opf_t write_flags = wbc_to_write_flags(wbc);
4606 prepare_eb_write(eb);
4608 num_pages = num_extent_pages(eb);
4609 for (i = 0; i < num_pages; i++) {
4610 struct page *p = eb->pages[i];
4612 clear_page_dirty_for_io(p);
4613 set_page_writeback(p);
4614 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
4615 &epd->bio_ctrl, p, disk_bytenr,
4617 end_bio_extent_buffer_writepage,
4620 set_btree_ioerr(p, eb);
4621 if (PageWriteback(p))
4622 end_page_writeback(p);
4623 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
4624 end_extent_buffer_writeback(eb);
4628 disk_bytenr += PAGE_SIZE;
4633 if (unlikely(ret)) {
4634 for (; i < num_pages; i++) {
4635 struct page *p = eb->pages[i];
4636 clear_page_dirty_for_io(p);
4645 * Submit one subpage btree page.
4647 * The main difference to submit_eb_page() is:
4649 * For subpage, we don't rely on page locking at all.
4652 * We only flush bio if we may be unable to fit current extent buffers into
4655 * Return >=0 for the number of submitted extent buffers.
4656 * Return <0 for fatal error.
4658 static int submit_eb_subpage(struct page *page,
4659 struct writeback_control *wbc,
4660 struct extent_page_data *epd)
4662 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4664 u64 page_start = page_offset(page);
4666 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
4669 /* Lock and write each dirty extent buffers in the range */
4670 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
4671 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
4672 struct extent_buffer *eb;
4673 unsigned long flags;
4677 * Take private lock to ensure the subpage won't be detached
4680 spin_lock(&page->mapping->private_lock);
4681 if (!PagePrivate(page)) {
4682 spin_unlock(&page->mapping->private_lock);
4685 spin_lock_irqsave(&subpage->lock, flags);
4686 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
4687 subpage->bitmaps)) {
4688 spin_unlock_irqrestore(&subpage->lock, flags);
4689 spin_unlock(&page->mapping->private_lock);
4694 start = page_start + bit_start * fs_info->sectorsize;
4695 bit_start += sectors_per_node;
4698 * Here we just want to grab the eb without touching extra
4699 * spin locks, so call find_extent_buffer_nolock().
4701 eb = find_extent_buffer_nolock(fs_info, start);
4702 spin_unlock_irqrestore(&subpage->lock, flags);
4703 spin_unlock(&page->mapping->private_lock);
4706 * The eb has already reached 0 refs thus find_extent_buffer()
4707 * doesn't return it. We don't need to write back such eb
4713 ret = lock_extent_buffer_for_io(eb, epd);
4715 free_extent_buffer(eb);
4719 free_extent_buffer(eb);
4722 ret = write_one_subpage_eb(eb, wbc, epd);
4723 free_extent_buffer(eb);
4731 /* We hit error, end bio for the submitted extent buffers */
4732 submit_write_bio(epd, ret);
4737 * Submit all page(s) of one extent buffer.
4739 * @page: the page of one extent buffer
4740 * @eb_context: to determine if we need to submit this page, if current page
4741 * belongs to this eb, we don't need to submit
4743 * The caller should pass each page in their bytenr order, and here we use
4744 * @eb_context to determine if we have submitted pages of one extent buffer.
4746 * If we have, we just skip until we hit a new page that doesn't belong to
4747 * current @eb_context.
4749 * If not, we submit all the page(s) of the extent buffer.
4751 * Return >0 if we have submitted the extent buffer successfully.
4752 * Return 0 if we don't need to submit the page, as it's already submitted by
4754 * Return <0 for fatal error.
4756 static int submit_eb_page(struct page *page, struct writeback_control *wbc,
4757 struct extent_page_data *epd,
4758 struct extent_buffer **eb_context)
4760 struct address_space *mapping = page->mapping;
4761 struct btrfs_block_group *cache = NULL;
4762 struct extent_buffer *eb;
4765 if (!PagePrivate(page))
4768 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
4769 return submit_eb_subpage(page, wbc, epd);
4771 spin_lock(&mapping->private_lock);
4772 if (!PagePrivate(page)) {
4773 spin_unlock(&mapping->private_lock);
4777 eb = (struct extent_buffer *)page->private;
4780 * Shouldn't happen and normally this would be a BUG_ON but no point
4781 * crashing the machine for something we can survive anyway.
4784 spin_unlock(&mapping->private_lock);
4788 if (eb == *eb_context) {
4789 spin_unlock(&mapping->private_lock);
4792 ret = atomic_inc_not_zero(&eb->refs);
4793 spin_unlock(&mapping->private_lock);
4797 if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) {
4799 * If for_sync, this hole will be filled with
4800 * trasnsaction commit.
4802 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4806 free_extent_buffer(eb);
4812 ret = lock_extent_buffer_for_io(eb, epd);
4814 btrfs_revert_meta_write_pointer(cache, eb);
4816 btrfs_put_block_group(cache);
4817 free_extent_buffer(eb);
4822 * Implies write in zoned mode. Mark the last eb in a block group.
4824 btrfs_schedule_zone_finish_bg(cache, eb);
4825 btrfs_put_block_group(cache);
4827 ret = write_one_eb(eb, wbc, epd);
4828 free_extent_buffer(eb);
4834 int btree_write_cache_pages(struct address_space *mapping,
4835 struct writeback_control *wbc)
4837 struct extent_buffer *eb_context = NULL;
4838 struct extent_page_data epd = {
4841 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4843 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
4846 int nr_to_write_done = 0;
4847 struct pagevec pvec;
4850 pgoff_t end; /* Inclusive */
4854 pagevec_init(&pvec);
4855 if (wbc->range_cyclic) {
4856 index = mapping->writeback_index; /* Start from prev offset */
4859 * Start from the beginning does not need to cycle over the
4860 * range, mark it as scanned.
4862 scanned = (index == 0);
4864 index = wbc->range_start >> PAGE_SHIFT;
4865 end = wbc->range_end >> PAGE_SHIFT;
4868 if (wbc->sync_mode == WB_SYNC_ALL)
4869 tag = PAGECACHE_TAG_TOWRITE;
4871 tag = PAGECACHE_TAG_DIRTY;
4872 btrfs_zoned_meta_io_lock(fs_info);
4874 if (wbc->sync_mode == WB_SYNC_ALL)
4875 tag_pages_for_writeback(mapping, index, end);
4876 while (!done && !nr_to_write_done && (index <= end) &&
4877 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
4881 for (i = 0; i < nr_pages; i++) {
4882 struct page *page = pvec.pages[i];
4884 ret = submit_eb_page(page, wbc, &epd, &eb_context);
4893 * the filesystem may choose to bump up nr_to_write.
4894 * We have to make sure to honor the new nr_to_write
4897 nr_to_write_done = wbc->nr_to_write <= 0;
4899 pagevec_release(&pvec);
4902 if (!scanned && !done) {
4904 * We hit the last page and there is more work to be done: wrap
4905 * back to the start of the file
4912 * If something went wrong, don't allow any metadata write bio to be
4915 * This would prevent use-after-free if we had dirty pages not
4916 * cleaned up, which can still happen by fuzzed images.
4919 * Allowing existing tree block to be allocated for other trees.
4921 * - Log tree operations
4922 * Exiting tree blocks get allocated to log tree, bumps its
4923 * generation, then get cleaned in tree re-balance.
4924 * Such tree block will not be written back, since it's clean,
4925 * thus no WRITTEN flag set.
4926 * And after log writes back, this tree block is not traced by
4927 * any dirty extent_io_tree.
4929 * - Offending tree block gets re-dirtied from its original owner
4930 * Since it has bumped generation, no WRITTEN flag, it can be
4931 * reused without COWing. This tree block will not be traced
4932 * by btrfs_transaction::dirty_pages.
4934 * Now such dirty tree block will not be cleaned by any dirty
4935 * extent io tree. Thus we don't want to submit such wild eb
4936 * if the fs already has error.
4938 * We can get ret > 0 from submit_extent_page() indicating how many ebs
4939 * were submitted. Reset it to 0 to avoid false alerts for the caller.
4943 if (!ret && BTRFS_FS_ERROR(fs_info))
4945 submit_write_bio(&epd, ret);
4947 btrfs_zoned_meta_io_unlock(fs_info);
4952 * Walk the list of dirty pages of the given address space and write all of them.
4954 * @mapping: address space structure to write
4955 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4956 * @epd: holds context for the write, namely the bio
4958 * If a page is already under I/O, write_cache_pages() skips it, even
4959 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4960 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4961 * and msync() need to guarantee that all the data which was dirty at the time
4962 * the call was made get new I/O started against them. If wbc->sync_mode is
4963 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4964 * existing IO to complete.
4966 static int extent_write_cache_pages(struct address_space *mapping,
4967 struct writeback_control *wbc,
4968 struct extent_page_data *epd)
4970 struct inode *inode = mapping->host;
4973 int nr_to_write_done = 0;
4974 struct pagevec pvec;
4977 pgoff_t end; /* Inclusive */
4979 int range_whole = 0;
4984 * We have to hold onto the inode so that ordered extents can do their
4985 * work when the IO finishes. The alternative to this is failing to add
4986 * an ordered extent if the igrab() fails there and that is a huge pain
4987 * to deal with, so instead just hold onto the inode throughout the
4988 * writepages operation. If it fails here we are freeing up the inode
4989 * anyway and we'd rather not waste our time writing out stuff that is
4990 * going to be truncated anyway.
4995 pagevec_init(&pvec);
4996 if (wbc->range_cyclic) {
4997 index = mapping->writeback_index; /* Start from prev offset */
5000 * Start from the beginning does not need to cycle over the
5001 * range, mark it as scanned.
5003 scanned = (index == 0);
5005 index = wbc->range_start >> PAGE_SHIFT;
5006 end = wbc->range_end >> PAGE_SHIFT;
5007 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
5013 * We do the tagged writepage as long as the snapshot flush bit is set
5014 * and we are the first one who do the filemap_flush() on this inode.
5016 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
5017 * not race in and drop the bit.
5019 if (range_whole && wbc->nr_to_write == LONG_MAX &&
5020 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
5021 &BTRFS_I(inode)->runtime_flags))
5022 wbc->tagged_writepages = 1;
5024 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
5025 tag = PAGECACHE_TAG_TOWRITE;
5027 tag = PAGECACHE_TAG_DIRTY;
5029 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
5030 tag_pages_for_writeback(mapping, index, end);
5032 while (!done && !nr_to_write_done && (index <= end) &&
5033 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
5034 &index, end, tag))) {
5037 for (i = 0; i < nr_pages; i++) {
5038 struct page *page = pvec.pages[i];
5040 done_index = page->index + 1;
5042 * At this point we hold neither the i_pages lock nor
5043 * the page lock: the page may be truncated or
5044 * invalidated (changing page->mapping to NULL),
5045 * or even swizzled back from swapper_space to
5046 * tmpfs file mapping
5048 if (!trylock_page(page)) {
5049 submit_write_bio(epd, 0);
5053 if (unlikely(page->mapping != mapping)) {
5058 if (wbc->sync_mode != WB_SYNC_NONE) {
5059 if (PageWriteback(page))
5060 submit_write_bio(epd, 0);
5061 wait_on_page_writeback(page);
5064 if (PageWriteback(page) ||
5065 !clear_page_dirty_for_io(page)) {
5070 ret = __extent_writepage(page, wbc, epd);
5077 * the filesystem may choose to bump up nr_to_write.
5078 * We have to make sure to honor the new nr_to_write
5081 nr_to_write_done = wbc->nr_to_write <= 0;
5083 pagevec_release(&pvec);
5086 if (!scanned && !done) {
5088 * We hit the last page and there is more work to be done: wrap
5089 * back to the start of the file
5095 * If we're looping we could run into a page that is locked by a
5096 * writer and that writer could be waiting on writeback for a
5097 * page in our current bio, and thus deadlock, so flush the
5100 submit_write_bio(epd, 0);
5104 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
5105 mapping->writeback_index = done_index;
5107 btrfs_add_delayed_iput(inode);
5112 * Submit the pages in the range to bio for call sites which delalloc range has
5113 * already been ran (aka, ordered extent inserted) and all pages are still
5116 int extent_write_locked_range(struct inode *inode, u64 start, u64 end)
5118 bool found_error = false;
5119 int first_error = 0;
5121 struct address_space *mapping = inode->i_mapping;
5124 unsigned long nr_pages;
5125 const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize;
5126 struct extent_page_data epd = {
5131 struct writeback_control wbc_writepages = {
5132 .sync_mode = WB_SYNC_ALL,
5133 .range_start = start,
5134 .range_end = end + 1,
5135 /* We're called from an async helper function */
5136 .punt_to_cgroup = 1,
5137 .no_cgroup_owner = 1,
5140 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
5141 nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >>
5143 wbc_writepages.nr_to_write = nr_pages * 2;
5145 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
5146 while (cur <= end) {
5147 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
5149 page = find_get_page(mapping, cur >> PAGE_SHIFT);
5151 * All pages in the range are locked since
5152 * btrfs_run_delalloc_range(), thus there is no way to clear
5153 * the page dirty flag.
5155 ASSERT(PageLocked(page));
5156 ASSERT(PageDirty(page));
5157 clear_page_dirty_for_io(page);
5158 ret = __extent_writepage(page, &wbc_writepages, &epd);
5168 submit_write_bio(&epd, found_error ? ret : 0);
5170 wbc_detach_inode(&wbc_writepages);
5176 int extent_writepages(struct address_space *mapping,
5177 struct writeback_control *wbc)
5179 struct inode *inode = mapping->host;
5181 struct extent_page_data epd = {
5184 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
5188 * Allow only a single thread to do the reloc work in zoned mode to
5189 * protect the write pointer updates.
5191 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
5192 ret = extent_write_cache_pages(mapping, wbc, &epd);
5193 submit_write_bio(&epd, ret);
5194 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
5198 void extent_readahead(struct readahead_control *rac)
5200 struct btrfs_bio_ctrl bio_ctrl = { 0 };
5201 struct page *pagepool[16];
5202 struct extent_map *em_cached = NULL;
5203 u64 prev_em_start = (u64)-1;
5206 while ((nr = readahead_page_batch(rac, pagepool))) {
5207 u64 contig_start = readahead_pos(rac);
5208 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
5210 contiguous_readpages(pagepool, nr, contig_start, contig_end,
5211 &em_cached, &bio_ctrl, &prev_em_start);
5215 free_extent_map(em_cached);
5216 submit_one_bio(&bio_ctrl);
5220 * basic invalidate_folio code, this waits on any locked or writeback
5221 * ranges corresponding to the folio, and then deletes any extent state
5222 * records from the tree
5224 int extent_invalidate_folio(struct extent_io_tree *tree,
5225 struct folio *folio, size_t offset)
5227 struct extent_state *cached_state = NULL;
5228 u64 start = folio_pos(folio);
5229 u64 end = start + folio_size(folio) - 1;
5230 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
5232 /* This function is only called for the btree inode */
5233 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
5235 start += ALIGN(offset, blocksize);
5239 lock_extent_bits(tree, start, end, &cached_state);
5240 folio_wait_writeback(folio);
5243 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
5244 * so here we only need to unlock the extent range to free any
5245 * existing extent state.
5247 unlock_extent_cached(tree, start, end, &cached_state);
5252 * a helper for release_folio, this tests for areas of the page that
5253 * are locked or under IO and drops the related state bits if it is safe
5256 static int try_release_extent_state(struct extent_io_tree *tree,
5257 struct page *page, gfp_t mask)
5259 u64 start = page_offset(page);
5260 u64 end = start + PAGE_SIZE - 1;
5263 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
5267 * At this point we can safely clear everything except the
5268 * locked bit, the nodatasum bit and the delalloc new bit.
5269 * The delalloc new bit will be cleared by ordered extent
5272 ret = __clear_extent_bit(tree, start, end,
5273 ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW),
5274 0, 0, NULL, mask, NULL);
5276 /* if clear_extent_bit failed for enomem reasons,
5277 * we can't allow the release to continue.
5288 * a helper for release_folio. As long as there are no locked extents
5289 * in the range corresponding to the page, both state records and extent
5290 * map records are removed
5292 int try_release_extent_mapping(struct page *page, gfp_t mask)
5294 struct extent_map *em;
5295 u64 start = page_offset(page);
5296 u64 end = start + PAGE_SIZE - 1;
5297 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
5298 struct extent_io_tree *tree = &btrfs_inode->io_tree;
5299 struct extent_map_tree *map = &btrfs_inode->extent_tree;
5301 if (gfpflags_allow_blocking(mask) &&
5302 page->mapping->host->i_size > SZ_16M) {
5304 while (start <= end) {
5305 struct btrfs_fs_info *fs_info;
5308 len = end - start + 1;
5309 write_lock(&map->lock);
5310 em = lookup_extent_mapping(map, start, len);
5312 write_unlock(&map->lock);
5315 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
5316 em->start != start) {
5317 write_unlock(&map->lock);
5318 free_extent_map(em);
5321 if (test_range_bit(tree, em->start,
5322 extent_map_end(em) - 1,
5323 EXTENT_LOCKED, 0, NULL))
5326 * If it's not in the list of modified extents, used
5327 * by a fast fsync, we can remove it. If it's being
5328 * logged we can safely remove it since fsync took an
5329 * extra reference on the em.
5331 if (list_empty(&em->list) ||
5332 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
5335 * If it's in the list of modified extents, remove it
5336 * only if its generation is older then the current one,
5337 * in which case we don't need it for a fast fsync.
5338 * Otherwise don't remove it, we could be racing with an
5339 * ongoing fast fsync that could miss the new extent.
5341 fs_info = btrfs_inode->root->fs_info;
5342 spin_lock(&fs_info->trans_lock);
5343 cur_gen = fs_info->generation;
5344 spin_unlock(&fs_info->trans_lock);
5345 if (em->generation >= cur_gen)
5349 * We only remove extent maps that are not in the list of
5350 * modified extents or that are in the list but with a
5351 * generation lower then the current generation, so there
5352 * is no need to set the full fsync flag on the inode (it
5353 * hurts the fsync performance for workloads with a data
5354 * size that exceeds or is close to the system's memory).
5356 remove_extent_mapping(map, em);
5357 /* once for the rb tree */
5358 free_extent_map(em);
5360 start = extent_map_end(em);
5361 write_unlock(&map->lock);
5364 free_extent_map(em);
5366 cond_resched(); /* Allow large-extent preemption. */
5369 return try_release_extent_state(tree, page, mask);
5373 * helper function for fiemap, which doesn't want to see any holes.
5374 * This maps until we find something past 'last'
5376 static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode,
5377 u64 offset, u64 last)
5379 u64 sectorsize = btrfs_inode_sectorsize(inode);
5380 struct extent_map *em;
5387 len = last - offset;
5390 len = ALIGN(len, sectorsize);
5391 em = btrfs_get_extent_fiemap(inode, offset, len);
5395 /* if this isn't a hole return it */
5396 if (em->block_start != EXTENT_MAP_HOLE)
5399 /* this is a hole, advance to the next extent */
5400 offset = extent_map_end(em);
5401 free_extent_map(em);
5409 * To cache previous fiemap extent
5411 * Will be used for merging fiemap extent
5413 struct fiemap_cache {
5422 * Helper to submit fiemap extent.
5424 * Will try to merge current fiemap extent specified by @offset, @phys,
5425 * @len and @flags with cached one.
5426 * And only when we fails to merge, cached one will be submitted as
5429 * Return value is the same as fiemap_fill_next_extent().
5431 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
5432 struct fiemap_cache *cache,
5433 u64 offset, u64 phys, u64 len, u32 flags)
5441 * Sanity check, extent_fiemap() should have ensured that new
5442 * fiemap extent won't overlap with cached one.
5445 * NOTE: Physical address can overlap, due to compression
5447 if (cache->offset + cache->len > offset) {
5453 * Only merges fiemap extents if
5454 * 1) Their logical addresses are continuous
5456 * 2) Their physical addresses are continuous
5457 * So truly compressed (physical size smaller than logical size)
5458 * extents won't get merged with each other
5460 * 3) Share same flags except FIEMAP_EXTENT_LAST
5461 * So regular extent won't get merged with prealloc extent
5463 if (cache->offset + cache->len == offset &&
5464 cache->phys + cache->len == phys &&
5465 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
5466 (flags & ~FIEMAP_EXTENT_LAST)) {
5468 cache->flags |= flags;
5469 goto try_submit_last;
5472 /* Not mergeable, need to submit cached one */
5473 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5474 cache->len, cache->flags);
5475 cache->cached = false;
5479 cache->cached = true;
5480 cache->offset = offset;
5483 cache->flags = flags;
5485 if (cache->flags & FIEMAP_EXTENT_LAST) {
5486 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
5487 cache->phys, cache->len, cache->flags);
5488 cache->cached = false;
5494 * Emit last fiemap cache
5496 * The last fiemap cache may still be cached in the following case:
5498 * |<- Fiemap range ->|
5499 * |<------------ First extent ----------->|
5501 * In this case, the first extent range will be cached but not emitted.
5502 * So we must emit it before ending extent_fiemap().
5504 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
5505 struct fiemap_cache *cache)
5512 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
5513 cache->len, cache->flags);
5514 cache->cached = false;
5520 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
5525 u64 max = start + len;
5529 u64 last_for_get_extent = 0;
5531 u64 isize = i_size_read(&inode->vfs_inode);
5532 struct btrfs_key found_key;
5533 struct extent_map *em = NULL;
5534 struct extent_state *cached_state = NULL;
5535 struct btrfs_path *path;
5536 struct btrfs_root *root = inode->root;
5537 struct fiemap_cache cache = { 0 };
5538 struct ulist *roots;
5539 struct ulist *tmp_ulist;
5548 path = btrfs_alloc_path();
5552 roots = ulist_alloc(GFP_KERNEL);
5553 tmp_ulist = ulist_alloc(GFP_KERNEL);
5554 if (!roots || !tmp_ulist) {
5556 goto out_free_ulist;
5560 * We can't initialize that to 'start' as this could miss extents due
5561 * to extent item merging
5564 start = round_down(start, btrfs_inode_sectorsize(inode));
5565 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
5568 * lookup the last file extent. We're not using i_size here
5569 * because there might be preallocation past i_size
5571 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
5574 goto out_free_ulist;
5582 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5583 found_type = found_key.type;
5585 /* No extents, but there might be delalloc bits */
5586 if (found_key.objectid != btrfs_ino(inode) ||
5587 found_type != BTRFS_EXTENT_DATA_KEY) {
5588 /* have to trust i_size as the end */
5590 last_for_get_extent = isize;
5593 * remember the start of the last extent. There are a
5594 * bunch of different factors that go into the length of the
5595 * extent, so its much less complex to remember where it started
5597 last = found_key.offset;
5598 last_for_get_extent = last + 1;
5600 btrfs_release_path(path);
5603 * we might have some extents allocated but more delalloc past those
5604 * extents. so, we trust isize unless the start of the last extent is
5609 last_for_get_extent = isize;
5612 lock_extent_bits(&inode->io_tree, start, start + len - 1,
5615 em = get_extent_skip_holes(inode, start, last_for_get_extent);
5624 u64 offset_in_extent = 0;
5626 /* break if the extent we found is outside the range */
5627 if (em->start >= max || extent_map_end(em) < off)
5631 * get_extent may return an extent that starts before our
5632 * requested range. We have to make sure the ranges
5633 * we return to fiemap always move forward and don't
5634 * overlap, so adjust the offsets here
5636 em_start = max(em->start, off);
5639 * record the offset from the start of the extent
5640 * for adjusting the disk offset below. Only do this if the
5641 * extent isn't compressed since our in ram offset may be past
5642 * what we have actually allocated on disk.
5644 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5645 offset_in_extent = em_start - em->start;
5646 em_end = extent_map_end(em);
5647 em_len = em_end - em_start;
5649 if (em->block_start < EXTENT_MAP_LAST_BYTE)
5650 disko = em->block_start + offset_in_extent;
5655 * bump off for our next call to get_extent
5657 off = extent_map_end(em);
5661 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
5663 flags |= FIEMAP_EXTENT_LAST;
5664 } else if (em->block_start == EXTENT_MAP_INLINE) {
5665 flags |= (FIEMAP_EXTENT_DATA_INLINE |
5666 FIEMAP_EXTENT_NOT_ALIGNED);
5667 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
5668 flags |= (FIEMAP_EXTENT_DELALLOC |
5669 FIEMAP_EXTENT_UNKNOWN);
5670 } else if (fieinfo->fi_extents_max) {
5671 u64 bytenr = em->block_start -
5672 (em->start - em->orig_start);
5675 * As btrfs supports shared space, this information
5676 * can be exported to userspace tools via
5677 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
5678 * then we're just getting a count and we can skip the
5681 ret = btrfs_check_shared(root, btrfs_ino(inode),
5682 bytenr, roots, tmp_ulist);
5686 flags |= FIEMAP_EXTENT_SHARED;
5689 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
5690 flags |= FIEMAP_EXTENT_ENCODED;
5691 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5692 flags |= FIEMAP_EXTENT_UNWRITTEN;
5694 free_extent_map(em);
5696 if ((em_start >= last) || em_len == (u64)-1 ||
5697 (last == (u64)-1 && isize <= em_end)) {
5698 flags |= FIEMAP_EXTENT_LAST;
5702 /* now scan forward to see if this is really the last extent. */
5703 em = get_extent_skip_holes(inode, off, last_for_get_extent);
5709 flags |= FIEMAP_EXTENT_LAST;
5712 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
5722 ret = emit_last_fiemap_cache(fieinfo, &cache);
5723 free_extent_map(em);
5725 unlock_extent_cached(&inode->io_tree, start, start + len - 1,
5729 btrfs_free_path(path);
5731 ulist_free(tmp_ulist);
5735 static void __free_extent_buffer(struct extent_buffer *eb)
5737 kmem_cache_free(extent_buffer_cache, eb);
5740 int extent_buffer_under_io(const struct extent_buffer *eb)
5742 return (atomic_read(&eb->io_pages) ||
5743 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
5744 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5747 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
5749 struct btrfs_subpage *subpage;
5751 lockdep_assert_held(&page->mapping->private_lock);
5753 if (PagePrivate(page)) {
5754 subpage = (struct btrfs_subpage *)page->private;
5755 if (atomic_read(&subpage->eb_refs))
5758 * Even there is no eb refs here, we may still have
5759 * end_page_read() call relying on page::private.
5761 if (atomic_read(&subpage->readers))
5767 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
5769 struct btrfs_fs_info *fs_info = eb->fs_info;
5770 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5773 * For mapped eb, we're going to change the page private, which should
5774 * be done under the private_lock.
5777 spin_lock(&page->mapping->private_lock);
5779 if (!PagePrivate(page)) {
5781 spin_unlock(&page->mapping->private_lock);
5785 if (fs_info->nodesize >= PAGE_SIZE) {
5787 * We do this since we'll remove the pages after we've
5788 * removed the eb from the radix tree, so we could race
5789 * and have this page now attached to the new eb. So
5790 * only clear page_private if it's still connected to
5793 if (PagePrivate(page) &&
5794 page->private == (unsigned long)eb) {
5795 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
5796 BUG_ON(PageDirty(page));
5797 BUG_ON(PageWriteback(page));
5799 * We need to make sure we haven't be attached
5802 detach_page_private(page);
5805 spin_unlock(&page->mapping->private_lock);
5810 * For subpage, we can have dummy eb with page private. In this case,
5811 * we can directly detach the private as such page is only attached to
5812 * one dummy eb, no sharing.
5815 btrfs_detach_subpage(fs_info, page);
5819 btrfs_page_dec_eb_refs(fs_info, page);
5822 * We can only detach the page private if there are no other ebs in the
5823 * page range and no unfinished IO.
5825 if (!page_range_has_eb(fs_info, page))
5826 btrfs_detach_subpage(fs_info, page);
5828 spin_unlock(&page->mapping->private_lock);
5831 /* Release all pages attached to the extent buffer */
5832 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
5837 ASSERT(!extent_buffer_under_io(eb));
5839 num_pages = num_extent_pages(eb);
5840 for (i = 0; i < num_pages; i++) {
5841 struct page *page = eb->pages[i];
5846 detach_extent_buffer_page(eb, page);
5848 /* One for when we allocated the page */
5854 * Helper for releasing the extent buffer.
5856 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
5858 btrfs_release_extent_buffer_pages(eb);
5859 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
5860 __free_extent_buffer(eb);
5863 static struct extent_buffer *
5864 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
5867 struct extent_buffer *eb = NULL;
5869 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
5872 eb->fs_info = fs_info;
5874 init_rwsem(&eb->lock);
5876 btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list,
5877 &fs_info->allocated_ebs);
5878 INIT_LIST_HEAD(&eb->release_list);
5880 spin_lock_init(&eb->refs_lock);
5881 atomic_set(&eb->refs, 1);
5882 atomic_set(&eb->io_pages, 0);
5884 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
5889 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
5892 struct extent_buffer *new;
5893 int num_pages = num_extent_pages(src);
5896 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
5901 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
5902 * btrfs_release_extent_buffer() have different behavior for
5903 * UNMAPPED subpage extent buffer.
5905 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
5907 memset(new->pages, 0, sizeof(*new->pages) * num_pages);
5908 ret = btrfs_alloc_page_array(num_pages, new->pages);
5910 btrfs_release_extent_buffer(new);
5914 for (i = 0; i < num_pages; i++) {
5916 struct page *p = new->pages[i];
5918 ret = attach_extent_buffer_page(new, p, NULL);
5920 btrfs_release_extent_buffer(new);
5923 WARN_ON(PageDirty(p));
5924 copy_page(page_address(p), page_address(src->pages[i]));
5926 set_extent_buffer_uptodate(new);
5931 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5932 u64 start, unsigned long len)
5934 struct extent_buffer *eb;
5939 eb = __alloc_extent_buffer(fs_info, start, len);
5943 num_pages = num_extent_pages(eb);
5944 ret = btrfs_alloc_page_array(num_pages, eb->pages);
5948 for (i = 0; i < num_pages; i++) {
5949 struct page *p = eb->pages[i];
5951 ret = attach_extent_buffer_page(eb, p, NULL);
5956 set_extent_buffer_uptodate(eb);
5957 btrfs_set_header_nritems(eb, 0);
5958 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
5962 for (i = 0; i < num_pages; i++) {
5964 detach_extent_buffer_page(eb, eb->pages[i]);
5965 __free_page(eb->pages[i]);
5968 __free_extent_buffer(eb);
5972 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
5975 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
5978 static void check_buffer_tree_ref(struct extent_buffer *eb)
5982 * The TREE_REF bit is first set when the extent_buffer is added
5983 * to the radix tree. It is also reset, if unset, when a new reference
5984 * is created by find_extent_buffer.
5986 * It is only cleared in two cases: freeing the last non-tree
5987 * reference to the extent_buffer when its STALE bit is set or
5988 * calling release_folio when the tree reference is the only reference.
5990 * In both cases, care is taken to ensure that the extent_buffer's
5991 * pages are not under io. However, release_folio can be concurrently
5992 * called with creating new references, which is prone to race
5993 * conditions between the calls to check_buffer_tree_ref in those
5994 * codepaths and clearing TREE_REF in try_release_extent_buffer.
5996 * The actual lifetime of the extent_buffer in the radix tree is
5997 * adequately protected by the refcount, but the TREE_REF bit and
5998 * its corresponding reference are not. To protect against this
5999 * class of races, we call check_buffer_tree_ref from the codepaths
6000 * which trigger io after they set eb->io_pages. Note that once io is
6001 * initiated, TREE_REF can no longer be cleared, so that is the
6002 * moment at which any such race is best fixed.
6004 refs = atomic_read(&eb->refs);
6005 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6008 spin_lock(&eb->refs_lock);
6009 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6010 atomic_inc(&eb->refs);
6011 spin_unlock(&eb->refs_lock);
6014 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
6015 struct page *accessed)
6019 check_buffer_tree_ref(eb);
6021 num_pages = num_extent_pages(eb);
6022 for (i = 0; i < num_pages; i++) {
6023 struct page *p = eb->pages[i];
6026 mark_page_accessed(p);
6030 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
6033 struct extent_buffer *eb;
6035 eb = find_extent_buffer_nolock(fs_info, start);
6039 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
6040 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
6041 * another task running free_extent_buffer() might have seen that flag
6042 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
6043 * writeback flags not set) and it's still in the tree (flag
6044 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
6045 * decrementing the extent buffer's reference count twice. So here we
6046 * could race and increment the eb's reference count, clear its stale
6047 * flag, mark it as dirty and drop our reference before the other task
6048 * finishes executing free_extent_buffer, which would later result in
6049 * an attempt to free an extent buffer that is dirty.
6051 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
6052 spin_lock(&eb->refs_lock);
6053 spin_unlock(&eb->refs_lock);
6055 mark_extent_buffer_accessed(eb, NULL);
6059 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6060 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
6063 struct extent_buffer *eb, *exists = NULL;
6066 eb = find_extent_buffer(fs_info, start);
6069 eb = alloc_dummy_extent_buffer(fs_info, start);
6071 return ERR_PTR(-ENOMEM);
6072 eb->fs_info = fs_info;
6074 ret = radix_tree_preload(GFP_NOFS);
6076 exists = ERR_PTR(ret);
6079 spin_lock(&fs_info->buffer_lock);
6080 ret = radix_tree_insert(&fs_info->buffer_radix,
6081 start >> fs_info->sectorsize_bits, eb);
6082 spin_unlock(&fs_info->buffer_lock);
6083 radix_tree_preload_end();
6084 if (ret == -EEXIST) {
6085 exists = find_extent_buffer(fs_info, start);
6091 check_buffer_tree_ref(eb);
6092 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
6096 btrfs_release_extent_buffer(eb);
6101 static struct extent_buffer *grab_extent_buffer(
6102 struct btrfs_fs_info *fs_info, struct page *page)
6104 struct extent_buffer *exists;
6107 * For subpage case, we completely rely on radix tree to ensure we
6108 * don't try to insert two ebs for the same bytenr. So here we always
6109 * return NULL and just continue.
6111 if (fs_info->nodesize < PAGE_SIZE)
6114 /* Page not yet attached to an extent buffer */
6115 if (!PagePrivate(page))
6119 * We could have already allocated an eb for this page and attached one
6120 * so lets see if we can get a ref on the existing eb, and if we can we
6121 * know it's good and we can just return that one, else we know we can
6122 * just overwrite page->private.
6124 exists = (struct extent_buffer *)page->private;
6125 if (atomic_inc_not_zero(&exists->refs))
6128 WARN_ON(PageDirty(page));
6129 detach_page_private(page);
6133 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
6135 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
6136 btrfs_err(fs_info, "bad tree block start %llu", start);
6140 if (fs_info->nodesize < PAGE_SIZE &&
6141 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
6143 "tree block crosses page boundary, start %llu nodesize %u",
6144 start, fs_info->nodesize);
6147 if (fs_info->nodesize >= PAGE_SIZE &&
6148 !PAGE_ALIGNED(start)) {
6150 "tree block is not page aligned, start %llu nodesize %u",
6151 start, fs_info->nodesize);
6157 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
6158 u64 start, u64 owner_root, int level)
6160 unsigned long len = fs_info->nodesize;
6163 unsigned long index = start >> PAGE_SHIFT;
6164 struct extent_buffer *eb;
6165 struct extent_buffer *exists = NULL;
6167 struct address_space *mapping = fs_info->btree_inode->i_mapping;
6168 u64 lockdep_owner = owner_root;
6172 if (check_eb_alignment(fs_info, start))
6173 return ERR_PTR(-EINVAL);
6175 #if BITS_PER_LONG == 32
6176 if (start >= MAX_LFS_FILESIZE) {
6177 btrfs_err_rl(fs_info,
6178 "extent buffer %llu is beyond 32bit page cache limit", start);
6179 btrfs_err_32bit_limit(fs_info);
6180 return ERR_PTR(-EOVERFLOW);
6182 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6183 btrfs_warn_32bit_limit(fs_info);
6186 eb = find_extent_buffer(fs_info, start);
6190 eb = __alloc_extent_buffer(fs_info, start, len);
6192 return ERR_PTR(-ENOMEM);
6195 * The reloc trees are just snapshots, so we need them to appear to be
6196 * just like any other fs tree WRT lockdep.
6198 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
6199 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
6201 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
6203 num_pages = num_extent_pages(eb);
6204 for (i = 0; i < num_pages; i++, index++) {
6205 struct btrfs_subpage *prealloc = NULL;
6207 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
6209 exists = ERR_PTR(-ENOMEM);
6214 * Preallocate page->private for subpage case, so that we won't
6215 * allocate memory with private_lock hold. The memory will be
6216 * freed by attach_extent_buffer_page() or freed manually if
6219 * Although we have ensured one subpage eb can only have one
6220 * page, but it may change in the future for 16K page size
6221 * support, so we still preallocate the memory in the loop.
6223 if (fs_info->nodesize < PAGE_SIZE) {
6224 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
6225 if (IS_ERR(prealloc)) {
6226 ret = PTR_ERR(prealloc);
6229 exists = ERR_PTR(ret);
6234 spin_lock(&mapping->private_lock);
6235 exists = grab_extent_buffer(fs_info, p);
6237 spin_unlock(&mapping->private_lock);
6240 mark_extent_buffer_accessed(exists, p);
6241 btrfs_free_subpage(prealloc);
6244 /* Should not fail, as we have preallocated the memory */
6245 ret = attach_extent_buffer_page(eb, p, prealloc);
6248 * To inform we have extra eb under allocation, so that
6249 * detach_extent_buffer_page() won't release the page private
6250 * when the eb hasn't yet been inserted into radix tree.
6252 * The ref will be decreased when the eb released the page, in
6253 * detach_extent_buffer_page().
6254 * Thus needs no special handling in error path.
6256 btrfs_page_inc_eb_refs(fs_info, p);
6257 spin_unlock(&mapping->private_lock);
6259 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
6261 if (!PageUptodate(p))
6265 * We can't unlock the pages just yet since the extent buffer
6266 * hasn't been properly inserted in the radix tree, this
6267 * opens a race with btree_release_folio which can free a page
6268 * while we are still filling in all pages for the buffer and
6273 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6275 ret = radix_tree_preload(GFP_NOFS);
6277 exists = ERR_PTR(ret);
6281 spin_lock(&fs_info->buffer_lock);
6282 ret = radix_tree_insert(&fs_info->buffer_radix,
6283 start >> fs_info->sectorsize_bits, eb);
6284 spin_unlock(&fs_info->buffer_lock);
6285 radix_tree_preload_end();
6286 if (ret == -EEXIST) {
6287 exists = find_extent_buffer(fs_info, start);
6293 /* add one reference for the tree */
6294 check_buffer_tree_ref(eb);
6295 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
6298 * Now it's safe to unlock the pages because any calls to
6299 * btree_release_folio will correctly detect that a page belongs to a
6300 * live buffer and won't free them prematurely.
6302 for (i = 0; i < num_pages; i++)
6303 unlock_page(eb->pages[i]);
6307 WARN_ON(!atomic_dec_and_test(&eb->refs));
6308 for (i = 0; i < num_pages; i++) {
6310 unlock_page(eb->pages[i]);
6313 btrfs_release_extent_buffer(eb);
6317 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
6319 struct extent_buffer *eb =
6320 container_of(head, struct extent_buffer, rcu_head);
6322 __free_extent_buffer(eb);
6325 static int release_extent_buffer(struct extent_buffer *eb)
6326 __releases(&eb->refs_lock)
6328 lockdep_assert_held(&eb->refs_lock);
6330 WARN_ON(atomic_read(&eb->refs) == 0);
6331 if (atomic_dec_and_test(&eb->refs)) {
6332 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
6333 struct btrfs_fs_info *fs_info = eb->fs_info;
6335 spin_unlock(&eb->refs_lock);
6337 spin_lock(&fs_info->buffer_lock);
6338 radix_tree_delete(&fs_info->buffer_radix,
6339 eb->start >> fs_info->sectorsize_bits);
6340 spin_unlock(&fs_info->buffer_lock);
6342 spin_unlock(&eb->refs_lock);
6345 btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list);
6346 /* Should be safe to release our pages at this point */
6347 btrfs_release_extent_buffer_pages(eb);
6348 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6349 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
6350 __free_extent_buffer(eb);
6354 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
6357 spin_unlock(&eb->refs_lock);
6362 void free_extent_buffer(struct extent_buffer *eb)
6370 refs = atomic_read(&eb->refs);
6371 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
6372 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
6375 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
6380 spin_lock(&eb->refs_lock);
6381 if (atomic_read(&eb->refs) == 2 &&
6382 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
6383 !extent_buffer_under_io(eb) &&
6384 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6385 atomic_dec(&eb->refs);
6388 * I know this is terrible, but it's temporary until we stop tracking
6389 * the uptodate bits and such for the extent buffers.
6391 release_extent_buffer(eb);
6394 void free_extent_buffer_stale(struct extent_buffer *eb)
6399 spin_lock(&eb->refs_lock);
6400 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
6402 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
6403 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
6404 atomic_dec(&eb->refs);
6405 release_extent_buffer(eb);
6408 static void btree_clear_page_dirty(struct page *page)
6410 ASSERT(PageDirty(page));
6411 ASSERT(PageLocked(page));
6412 clear_page_dirty_for_io(page);
6413 xa_lock_irq(&page->mapping->i_pages);
6414 if (!PageDirty(page))
6415 __xa_clear_mark(&page->mapping->i_pages,
6416 page_index(page), PAGECACHE_TAG_DIRTY);
6417 xa_unlock_irq(&page->mapping->i_pages);
6420 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
6422 struct btrfs_fs_info *fs_info = eb->fs_info;
6423 struct page *page = eb->pages[0];
6426 /* btree_clear_page_dirty() needs page locked */
6428 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
6431 btree_clear_page_dirty(page);
6433 WARN_ON(atomic_read(&eb->refs) == 0);
6436 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
6442 if (eb->fs_info->nodesize < PAGE_SIZE)
6443 return clear_subpage_extent_buffer_dirty(eb);
6445 num_pages = num_extent_pages(eb);
6447 for (i = 0; i < num_pages; i++) {
6448 page = eb->pages[i];
6449 if (!PageDirty(page))
6452 btree_clear_page_dirty(page);
6453 ClearPageError(page);
6456 WARN_ON(atomic_read(&eb->refs) == 0);
6459 bool set_extent_buffer_dirty(struct extent_buffer *eb)
6465 check_buffer_tree_ref(eb);
6467 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
6469 num_pages = num_extent_pages(eb);
6470 WARN_ON(atomic_read(&eb->refs) == 0);
6471 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
6474 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
6477 * For subpage case, we can have other extent buffers in the
6478 * same page, and in clear_subpage_extent_buffer_dirty() we
6479 * have to clear page dirty without subpage lock held.
6480 * This can cause race where our page gets dirty cleared after
6483 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
6484 * its page for other reasons, we can use page lock to prevent
6488 lock_page(eb->pages[0]);
6489 for (i = 0; i < num_pages; i++)
6490 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
6491 eb->start, eb->len);
6493 unlock_page(eb->pages[0]);
6495 #ifdef CONFIG_BTRFS_DEBUG
6496 for (i = 0; i < num_pages; i++)
6497 ASSERT(PageDirty(eb->pages[i]));
6503 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
6505 struct btrfs_fs_info *fs_info = eb->fs_info;
6510 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6511 num_pages = num_extent_pages(eb);
6512 for (i = 0; i < num_pages; i++) {
6513 page = eb->pages[i];
6518 * This is special handling for metadata subpage, as regular
6519 * btrfs_is_subpage() can not handle cloned/dummy metadata.
6521 if (fs_info->nodesize >= PAGE_SIZE)
6522 ClearPageUptodate(page);
6524 btrfs_subpage_clear_uptodate(fs_info, page, eb->start,
6529 void set_extent_buffer_uptodate(struct extent_buffer *eb)
6531 struct btrfs_fs_info *fs_info = eb->fs_info;
6536 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6537 num_pages = num_extent_pages(eb);
6538 for (i = 0; i < num_pages; i++) {
6539 page = eb->pages[i];
6542 * This is special handling for metadata subpage, as regular
6543 * btrfs_is_subpage() can not handle cloned/dummy metadata.
6545 if (fs_info->nodesize >= PAGE_SIZE)
6546 SetPageUptodate(page);
6548 btrfs_subpage_set_uptodate(fs_info, page, eb->start,
6553 static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait,
6556 struct btrfs_fs_info *fs_info = eb->fs_info;
6557 struct extent_io_tree *io_tree;
6558 struct page *page = eb->pages[0];
6559 struct btrfs_bio_ctrl bio_ctrl = {
6560 .mirror_num = mirror_num,
6564 ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags));
6565 ASSERT(PagePrivate(page));
6566 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
6568 if (wait == WAIT_NONE) {
6569 if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1))
6572 ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6578 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) ||
6579 PageUptodate(page) ||
6580 btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) {
6581 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6582 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1);
6586 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6587 eb->read_mirror = 0;
6588 atomic_set(&eb->io_pages, 1);
6589 check_buffer_tree_ref(eb);
6590 btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len);
6592 btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len);
6593 ret = submit_extent_page(REQ_OP_READ, NULL, &bio_ctrl,
6594 page, eb->start, eb->len,
6595 eb->start - page_offset(page),
6596 end_bio_extent_readpage, 0, true);
6599 * In the endio function, if we hit something wrong we will
6600 * increase the io_pages, so here we need to decrease it for
6603 atomic_dec(&eb->io_pages);
6605 submit_one_bio(&bio_ctrl);
6606 if (ret || wait != WAIT_COMPLETE)
6609 wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED);
6610 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6615 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
6621 int locked_pages = 0;
6622 int all_uptodate = 1;
6624 unsigned long num_reads = 0;
6625 struct btrfs_bio_ctrl bio_ctrl = {
6626 .mirror_num = mirror_num,
6629 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
6633 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
6634 * operation, which could potentially still be in flight. In this case
6635 * we simply want to return an error.
6637 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
6640 if (eb->fs_info->nodesize < PAGE_SIZE)
6641 return read_extent_buffer_subpage(eb, wait, mirror_num);
6643 num_pages = num_extent_pages(eb);
6644 for (i = 0; i < num_pages; i++) {
6645 page = eb->pages[i];
6646 if (wait == WAIT_NONE) {
6648 * WAIT_NONE is only utilized by readahead. If we can't
6649 * acquire the lock atomically it means either the eb
6650 * is being read out or under modification.
6651 * Either way the eb will be or has been cached,
6652 * readahead can exit safely.
6654 if (!trylock_page(page))
6662 * We need to firstly lock all pages to make sure that
6663 * the uptodate bit of our pages won't be affected by
6664 * clear_extent_buffer_uptodate().
6666 for (i = 0; i < num_pages; i++) {
6667 page = eb->pages[i];
6668 if (!PageUptodate(page)) {
6675 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
6679 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
6680 eb->read_mirror = 0;
6681 atomic_set(&eb->io_pages, num_reads);
6683 * It is possible for release_folio to clear the TREE_REF bit before we
6684 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
6686 check_buffer_tree_ref(eb);
6687 for (i = 0; i < num_pages; i++) {
6688 page = eb->pages[i];
6690 if (!PageUptodate(page)) {
6692 atomic_dec(&eb->io_pages);
6697 ClearPageError(page);
6698 err = submit_extent_page(REQ_OP_READ, NULL,
6699 &bio_ctrl, page, page_offset(page),
6700 PAGE_SIZE, 0, end_bio_extent_readpage,
6704 * We failed to submit the bio so it's the
6705 * caller's responsibility to perform cleanup
6706 * i.e unlock page/set error bit.
6711 atomic_dec(&eb->io_pages);
6718 submit_one_bio(&bio_ctrl);
6720 if (ret || wait != WAIT_COMPLETE)
6723 for (i = 0; i < num_pages; i++) {
6724 page = eb->pages[i];
6725 wait_on_page_locked(page);
6726 if (!PageUptodate(page))
6733 while (locked_pages > 0) {
6735 page = eb->pages[locked_pages];
6741 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
6744 btrfs_warn(eb->fs_info,
6745 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
6746 eb->start, eb->len, start, len);
6747 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6753 * Check if the [start, start + len) range is valid before reading/writing
6755 * NOTE: @start and @len are offset inside the eb, not logical address.
6757 * Caller should not touch the dst/src memory if this function returns error.
6759 static inline int check_eb_range(const struct extent_buffer *eb,
6760 unsigned long start, unsigned long len)
6762 unsigned long offset;
6764 /* start, start + len should not go beyond eb->len nor overflow */
6765 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
6766 return report_eb_range(eb, start, len);
6771 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
6772 unsigned long start, unsigned long len)
6778 char *dst = (char *)dstv;
6779 unsigned long i = get_eb_page_index(start);
6781 if (check_eb_range(eb, start, len))
6784 offset = get_eb_offset_in_page(eb, start);
6787 page = eb->pages[i];
6789 cur = min(len, (PAGE_SIZE - offset));
6790 kaddr = page_address(page);
6791 memcpy(dst, kaddr + offset, cur);
6800 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
6802 unsigned long start, unsigned long len)
6808 char __user *dst = (char __user *)dstv;
6809 unsigned long i = get_eb_page_index(start);
6812 WARN_ON(start > eb->len);
6813 WARN_ON(start + len > eb->start + eb->len);
6815 offset = get_eb_offset_in_page(eb, start);
6818 page = eb->pages[i];
6820 cur = min(len, (PAGE_SIZE - offset));
6821 kaddr = page_address(page);
6822 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
6836 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
6837 unsigned long start, unsigned long len)
6843 char *ptr = (char *)ptrv;
6844 unsigned long i = get_eb_page_index(start);
6847 if (check_eb_range(eb, start, len))
6850 offset = get_eb_offset_in_page(eb, start);
6853 page = eb->pages[i];
6855 cur = min(len, (PAGE_SIZE - offset));
6857 kaddr = page_address(page);
6858 ret = memcmp(ptr, kaddr + offset, cur);
6871 * Check that the extent buffer is uptodate.
6873 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
6874 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
6876 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
6879 struct btrfs_fs_info *fs_info = eb->fs_info;
6882 * If we are using the commit root we could potentially clear a page
6883 * Uptodate while we're using the extent buffer that we've previously
6884 * looked up. We don't want to complain in this case, as the page was
6885 * valid before, we just didn't write it out. Instead we want to catch
6886 * the case where we didn't actually read the block properly, which
6887 * would have !PageUptodate && !PageError, as we clear PageError before
6890 if (fs_info->nodesize < PAGE_SIZE) {
6891 bool uptodate, error;
6893 uptodate = btrfs_subpage_test_uptodate(fs_info, page,
6894 eb->start, eb->len);
6895 error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len);
6896 WARN_ON(!uptodate && !error);
6898 WARN_ON(!PageUptodate(page) && !PageError(page));
6902 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
6907 assert_eb_page_uptodate(eb, eb->pages[0]);
6908 kaddr = page_address(eb->pages[0]) +
6909 get_eb_offset_in_page(eb, offsetof(struct btrfs_header,
6911 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6914 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
6918 assert_eb_page_uptodate(eb, eb->pages[0]);
6919 kaddr = page_address(eb->pages[0]) +
6920 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid));
6921 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
6924 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
6925 unsigned long start, unsigned long len)
6931 char *src = (char *)srcv;
6932 unsigned long i = get_eb_page_index(start);
6934 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
6936 if (check_eb_range(eb, start, len))
6939 offset = get_eb_offset_in_page(eb, start);
6942 page = eb->pages[i];
6943 assert_eb_page_uptodate(eb, page);
6945 cur = min(len, PAGE_SIZE - offset);
6946 kaddr = page_address(page);
6947 memcpy(kaddr + offset, src, cur);
6956 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
6963 unsigned long i = get_eb_page_index(start);
6965 if (check_eb_range(eb, start, len))
6968 offset = get_eb_offset_in_page(eb, start);
6971 page = eb->pages[i];
6972 assert_eb_page_uptodate(eb, page);
6974 cur = min(len, PAGE_SIZE - offset);
6975 kaddr = page_address(page);
6976 memset(kaddr + offset, 0, cur);
6984 void copy_extent_buffer_full(const struct extent_buffer *dst,
6985 const struct extent_buffer *src)
6990 ASSERT(dst->len == src->len);
6992 if (dst->fs_info->nodesize >= PAGE_SIZE) {
6993 num_pages = num_extent_pages(dst);
6994 for (i = 0; i < num_pages; i++)
6995 copy_page(page_address(dst->pages[i]),
6996 page_address(src->pages[i]));
6998 size_t src_offset = get_eb_offset_in_page(src, 0);
6999 size_t dst_offset = get_eb_offset_in_page(dst, 0);
7001 ASSERT(src->fs_info->nodesize < PAGE_SIZE);
7002 memcpy(page_address(dst->pages[0]) + dst_offset,
7003 page_address(src->pages[0]) + src_offset,
7008 void copy_extent_buffer(const struct extent_buffer *dst,
7009 const struct extent_buffer *src,
7010 unsigned long dst_offset, unsigned long src_offset,
7013 u64 dst_len = dst->len;
7018 unsigned long i = get_eb_page_index(dst_offset);
7020 if (check_eb_range(dst, dst_offset, len) ||
7021 check_eb_range(src, src_offset, len))
7024 WARN_ON(src->len != dst_len);
7026 offset = get_eb_offset_in_page(dst, dst_offset);
7029 page = dst->pages[i];
7030 assert_eb_page_uptodate(dst, page);
7032 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
7034 kaddr = page_address(page);
7035 read_extent_buffer(src, kaddr + offset, src_offset, cur);
7045 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
7047 * @eb: the extent buffer
7048 * @start: offset of the bitmap item in the extent buffer
7050 * @page_index: return index of the page in the extent buffer that contains the
7052 * @page_offset: return offset into the page given by page_index
7054 * This helper hides the ugliness of finding the byte in an extent buffer which
7055 * contains a given bit.
7057 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
7058 unsigned long start, unsigned long nr,
7059 unsigned long *page_index,
7060 size_t *page_offset)
7062 size_t byte_offset = BIT_BYTE(nr);
7066 * The byte we want is the offset of the extent buffer + the offset of
7067 * the bitmap item in the extent buffer + the offset of the byte in the
7070 offset = start + offset_in_page(eb->start) + byte_offset;
7072 *page_index = offset >> PAGE_SHIFT;
7073 *page_offset = offset_in_page(offset);
7077 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
7078 * @eb: the extent buffer
7079 * @start: offset of the bitmap item in the extent buffer
7080 * @nr: bit number to test
7082 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
7090 eb_bitmap_offset(eb, start, nr, &i, &offset);
7091 page = eb->pages[i];
7092 assert_eb_page_uptodate(eb, page);
7093 kaddr = page_address(page);
7094 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
7098 * extent_buffer_bitmap_set - set an area of a bitmap
7099 * @eb: the extent buffer
7100 * @start: offset of the bitmap item in the extent buffer
7101 * @pos: bit number of the first bit
7102 * @len: number of bits to set
7104 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
7105 unsigned long pos, unsigned long len)
7111 const unsigned int size = pos + len;
7112 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
7113 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
7115 eb_bitmap_offset(eb, start, pos, &i, &offset);
7116 page = eb->pages[i];
7117 assert_eb_page_uptodate(eb, page);
7118 kaddr = page_address(page);
7120 while (len >= bits_to_set) {
7121 kaddr[offset] |= mask_to_set;
7123 bits_to_set = BITS_PER_BYTE;
7125 if (++offset >= PAGE_SIZE && len > 0) {
7127 page = eb->pages[++i];
7128 assert_eb_page_uptodate(eb, page);
7129 kaddr = page_address(page);
7133 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
7134 kaddr[offset] |= mask_to_set;
7140 * extent_buffer_bitmap_clear - clear an area of a bitmap
7141 * @eb: the extent buffer
7142 * @start: offset of the bitmap item in the extent buffer
7143 * @pos: bit number of the first bit
7144 * @len: number of bits to clear
7146 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
7147 unsigned long start, unsigned long pos,
7154 const unsigned int size = pos + len;
7155 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
7156 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
7158 eb_bitmap_offset(eb, start, pos, &i, &offset);
7159 page = eb->pages[i];
7160 assert_eb_page_uptodate(eb, page);
7161 kaddr = page_address(page);
7163 while (len >= bits_to_clear) {
7164 kaddr[offset] &= ~mask_to_clear;
7165 len -= bits_to_clear;
7166 bits_to_clear = BITS_PER_BYTE;
7168 if (++offset >= PAGE_SIZE && len > 0) {
7170 page = eb->pages[++i];
7171 assert_eb_page_uptodate(eb, page);
7172 kaddr = page_address(page);
7176 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
7177 kaddr[offset] &= ~mask_to_clear;
7181 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
7183 unsigned long distance = (src > dst) ? src - dst : dst - src;
7184 return distance < len;
7187 static void copy_pages(struct page *dst_page, struct page *src_page,
7188 unsigned long dst_off, unsigned long src_off,
7191 char *dst_kaddr = page_address(dst_page);
7193 int must_memmove = 0;
7195 if (dst_page != src_page) {
7196 src_kaddr = page_address(src_page);
7198 src_kaddr = dst_kaddr;
7199 if (areas_overlap(src_off, dst_off, len))
7204 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
7206 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
7209 void memcpy_extent_buffer(const struct extent_buffer *dst,
7210 unsigned long dst_offset, unsigned long src_offset,
7214 size_t dst_off_in_page;
7215 size_t src_off_in_page;
7216 unsigned long dst_i;
7217 unsigned long src_i;
7219 if (check_eb_range(dst, dst_offset, len) ||
7220 check_eb_range(dst, src_offset, len))
7224 dst_off_in_page = get_eb_offset_in_page(dst, dst_offset);
7225 src_off_in_page = get_eb_offset_in_page(dst, src_offset);
7227 dst_i = get_eb_page_index(dst_offset);
7228 src_i = get_eb_page_index(src_offset);
7230 cur = min(len, (unsigned long)(PAGE_SIZE -
7232 cur = min_t(unsigned long, cur,
7233 (unsigned long)(PAGE_SIZE - dst_off_in_page));
7235 copy_pages(dst->pages[dst_i], dst->pages[src_i],
7236 dst_off_in_page, src_off_in_page, cur);
7244 void memmove_extent_buffer(const struct extent_buffer *dst,
7245 unsigned long dst_offset, unsigned long src_offset,
7249 size_t dst_off_in_page;
7250 size_t src_off_in_page;
7251 unsigned long dst_end = dst_offset + len - 1;
7252 unsigned long src_end = src_offset + len - 1;
7253 unsigned long dst_i;
7254 unsigned long src_i;
7256 if (check_eb_range(dst, dst_offset, len) ||
7257 check_eb_range(dst, src_offset, len))
7259 if (dst_offset < src_offset) {
7260 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
7264 dst_i = get_eb_page_index(dst_end);
7265 src_i = get_eb_page_index(src_end);
7267 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
7268 src_off_in_page = get_eb_offset_in_page(dst, src_end);
7270 cur = min_t(unsigned long, len, src_off_in_page + 1);
7271 cur = min(cur, dst_off_in_page + 1);
7272 copy_pages(dst->pages[dst_i], dst->pages[src_i],
7273 dst_off_in_page - cur + 1,
7274 src_off_in_page - cur + 1, cur);
7282 #define GANG_LOOKUP_SIZE 16
7283 static struct extent_buffer *get_next_extent_buffer(
7284 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
7286 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
7287 struct extent_buffer *found = NULL;
7288 u64 page_start = page_offset(page);
7289 u64 cur = page_start;
7291 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
7292 lockdep_assert_held(&fs_info->buffer_lock);
7294 while (cur < page_start + PAGE_SIZE) {
7298 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
7299 (void **)gang, cur >> fs_info->sectorsize_bits,
7300 min_t(unsigned int, GANG_LOOKUP_SIZE,
7301 PAGE_SIZE / fs_info->nodesize));
7304 for (i = 0; i < ret; i++) {
7305 /* Already beyond page end */
7306 if (gang[i]->start >= page_start + PAGE_SIZE)
7309 if (gang[i]->start >= bytenr) {
7314 cur = gang[ret - 1]->start + gang[ret - 1]->len;
7320 static int try_release_subpage_extent_buffer(struct page *page)
7322 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7323 u64 cur = page_offset(page);
7324 const u64 end = page_offset(page) + PAGE_SIZE;
7328 struct extent_buffer *eb = NULL;
7331 * Unlike try_release_extent_buffer() which uses page->private
7332 * to grab buffer, for subpage case we rely on radix tree, thus
7333 * we need to ensure radix tree consistency.
7335 * We also want an atomic snapshot of the radix tree, thus go
7336 * with spinlock rather than RCU.
7338 spin_lock(&fs_info->buffer_lock);
7339 eb = get_next_extent_buffer(fs_info, page, cur);
7341 /* No more eb in the page range after or at cur */
7342 spin_unlock(&fs_info->buffer_lock);
7345 cur = eb->start + eb->len;
7348 * The same as try_release_extent_buffer(), to ensure the eb
7349 * won't disappear out from under us.
7351 spin_lock(&eb->refs_lock);
7352 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7353 spin_unlock(&eb->refs_lock);
7354 spin_unlock(&fs_info->buffer_lock);
7357 spin_unlock(&fs_info->buffer_lock);
7360 * If tree ref isn't set then we know the ref on this eb is a
7361 * real ref, so just return, this eb will likely be freed soon
7364 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7365 spin_unlock(&eb->refs_lock);
7370 * Here we don't care about the return value, we will always
7371 * check the page private at the end. And
7372 * release_extent_buffer() will release the refs_lock.
7374 release_extent_buffer(eb);
7377 * Finally to check if we have cleared page private, as if we have
7378 * released all ebs in the page, the page private should be cleared now.
7380 spin_lock(&page->mapping->private_lock);
7381 if (!PagePrivate(page))
7385 spin_unlock(&page->mapping->private_lock);
7390 int try_release_extent_buffer(struct page *page)
7392 struct extent_buffer *eb;
7394 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
7395 return try_release_subpage_extent_buffer(page);
7398 * We need to make sure nobody is changing page->private, as we rely on
7399 * page->private as the pointer to extent buffer.
7401 spin_lock(&page->mapping->private_lock);
7402 if (!PagePrivate(page)) {
7403 spin_unlock(&page->mapping->private_lock);
7407 eb = (struct extent_buffer *)page->private;
7411 * This is a little awful but should be ok, we need to make sure that
7412 * the eb doesn't disappear out from under us while we're looking at
7415 spin_lock(&eb->refs_lock);
7416 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
7417 spin_unlock(&eb->refs_lock);
7418 spin_unlock(&page->mapping->private_lock);
7421 spin_unlock(&page->mapping->private_lock);
7424 * If tree ref isn't set then we know the ref on this eb is a real ref,
7425 * so just return, this page will likely be freed soon anyway.
7427 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
7428 spin_unlock(&eb->refs_lock);
7432 return release_extent_buffer(eb);
7436 * btrfs_readahead_tree_block - attempt to readahead a child block
7437 * @fs_info: the fs_info
7438 * @bytenr: bytenr to read
7439 * @owner_root: objectid of the root that owns this eb
7440 * @gen: generation for the uptodate check, can be 0
7441 * @level: level for the eb
7443 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
7444 * normal uptodate check of the eb, without checking the generation. If we have
7445 * to read the block we will not block on anything.
7447 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
7448 u64 bytenr, u64 owner_root, u64 gen, int level)
7450 struct extent_buffer *eb;
7453 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
7457 if (btrfs_buffer_uptodate(eb, gen, 1)) {
7458 free_extent_buffer(eb);
7462 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0);
7464 free_extent_buffer_stale(eb);
7466 free_extent_buffer(eb);
7470 * btrfs_readahead_node_child - readahead a node's child block
7471 * @node: parent node we're reading from
7472 * @slot: slot in the parent node for the child we want to read
7474 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
7475 * the slot in the node provided.
7477 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
7479 btrfs_readahead_tree_block(node->fs_info,
7480 btrfs_node_blockptr(node, slot),
7481 btrfs_header_owner(node),
7482 btrfs_node_ptr_generation(node, slot),
7483 btrfs_header_level(node) - 1);