1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/prefetch.h>
15 #include <linux/cleancache.h>
16 #include "extent_io.h"
17 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
27 static struct kmem_cache
*extent_state_cache
;
28 static struct kmem_cache
*extent_buffer_cache
;
29 static struct bio_set btrfs_bioset
;
31 static inline bool extent_state_in_tree(const struct extent_state
*state
)
33 return !RB_EMPTY_NODE(&state
->rb_node
);
36 #ifdef CONFIG_BTRFS_DEBUG
37 static LIST_HEAD(buffers
);
38 static LIST_HEAD(states
);
40 static DEFINE_SPINLOCK(leak_lock
);
43 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
47 spin_lock_irqsave(&leak_lock
, flags
);
49 spin_unlock_irqrestore(&leak_lock
, flags
);
53 void btrfs_leak_debug_del(struct list_head
*entry
)
57 spin_lock_irqsave(&leak_lock
, flags
);
59 spin_unlock_irqrestore(&leak_lock
, flags
);
63 void btrfs_leak_debug_check(void)
65 struct extent_state
*state
;
66 struct extent_buffer
*eb
;
68 while (!list_empty(&states
)) {
69 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
70 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
71 state
->start
, state
->end
, state
->state
,
72 extent_state_in_tree(state
),
73 refcount_read(&state
->refs
));
74 list_del(&state
->leak_list
);
75 kmem_cache_free(extent_state_cache
, state
);
78 while (!list_empty(&buffers
)) {
79 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
80 pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n",
81 eb
->start
, eb
->len
, atomic_read(&eb
->refs
), eb
->bflags
);
82 list_del(&eb
->leak_list
);
83 kmem_cache_free(extent_buffer_cache
, eb
);
87 #define btrfs_debug_check_extent_io_range(tree, start, end) \
88 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
89 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
90 struct extent_io_tree
*tree
, u64 start
, u64 end
)
92 struct inode
*inode
= tree
->private_data
;
95 if (!inode
|| !is_data_inode(inode
))
98 isize
= i_size_read(inode
);
99 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
100 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
101 "%s: ino %llu isize %llu odd range [%llu,%llu]",
102 caller
, btrfs_ino(BTRFS_I(inode
)), isize
, start
, end
);
106 #define btrfs_leak_debug_add(new, head) do {} while (0)
107 #define btrfs_leak_debug_del(entry) do {} while (0)
108 #define btrfs_leak_debug_check() do {} while (0)
109 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
112 #define BUFFER_LRU_MAX 64
117 struct rb_node rb_node
;
120 struct extent_page_data
{
122 struct extent_io_tree
*tree
;
123 /* tells writepage not to lock the state bits for this range
124 * it still does the unlocking
126 unsigned int extent_locked
:1;
128 /* tells the submit_bio code to use REQ_SYNC */
129 unsigned int sync_io
:1;
132 static int add_extent_changeset(struct extent_state
*state
, unsigned bits
,
133 struct extent_changeset
*changeset
,
140 if (set
&& (state
->state
& bits
) == bits
)
142 if (!set
&& (state
->state
& bits
) == 0)
144 changeset
->bytes_changed
+= state
->end
- state
->start
+ 1;
145 ret
= ulist_add(&changeset
->range_changed
, state
->start
, state
->end
,
150 static int __must_check
submit_one_bio(struct bio
*bio
, int mirror_num
,
151 unsigned long bio_flags
)
153 blk_status_t ret
= 0;
154 struct bio_vec
*bvec
= bio_last_bvec_all(bio
);
156 struct extent_io_tree
*tree
= bio
->bi_private
;
159 mp_bvec_last_segment(bvec
, &bv
);
160 start
= page_offset(bv
.bv_page
) + bv
.bv_offset
;
162 bio
->bi_private
= NULL
;
165 ret
= tree
->ops
->submit_bio_hook(tree
->private_data
, bio
,
166 mirror_num
, bio_flags
, start
);
168 btrfsic_submit_bio(bio
);
170 return blk_status_to_errno(ret
);
173 static void flush_write_bio(struct extent_page_data
*epd
)
178 ret
= submit_one_bio(epd
->bio
, 0, 0);
179 BUG_ON(ret
< 0); /* -ENOMEM */
184 int __init
extent_io_init(void)
186 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
187 sizeof(struct extent_state
), 0,
188 SLAB_MEM_SPREAD
, NULL
);
189 if (!extent_state_cache
)
192 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
193 sizeof(struct extent_buffer
), 0,
194 SLAB_MEM_SPREAD
, NULL
);
195 if (!extent_buffer_cache
)
196 goto free_state_cache
;
198 if (bioset_init(&btrfs_bioset
, BIO_POOL_SIZE
,
199 offsetof(struct btrfs_io_bio
, bio
),
201 goto free_buffer_cache
;
203 if (bioset_integrity_create(&btrfs_bioset
, BIO_POOL_SIZE
))
209 bioset_exit(&btrfs_bioset
);
212 kmem_cache_destroy(extent_buffer_cache
);
213 extent_buffer_cache
= NULL
;
216 kmem_cache_destroy(extent_state_cache
);
217 extent_state_cache
= NULL
;
221 void __cold
extent_io_exit(void)
223 btrfs_leak_debug_check();
226 * Make sure all delayed rcu free are flushed before we
230 kmem_cache_destroy(extent_state_cache
);
231 kmem_cache_destroy(extent_buffer_cache
);
232 bioset_exit(&btrfs_bioset
);
235 void extent_io_tree_init(struct extent_io_tree
*tree
,
238 tree
->state
= RB_ROOT
;
240 tree
->dirty_bytes
= 0;
241 spin_lock_init(&tree
->lock
);
242 tree
->private_data
= private_data
;
245 static struct extent_state
*alloc_extent_state(gfp_t mask
)
247 struct extent_state
*state
;
250 * The given mask might be not appropriate for the slab allocator,
251 * drop the unsupported bits
253 mask
&= ~(__GFP_DMA32
|__GFP_HIGHMEM
);
254 state
= kmem_cache_alloc(extent_state_cache
, mask
);
258 state
->failrec
= NULL
;
259 RB_CLEAR_NODE(&state
->rb_node
);
260 btrfs_leak_debug_add(&state
->leak_list
, &states
);
261 refcount_set(&state
->refs
, 1);
262 init_waitqueue_head(&state
->wq
);
263 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
267 void free_extent_state(struct extent_state
*state
)
271 if (refcount_dec_and_test(&state
->refs
)) {
272 WARN_ON(extent_state_in_tree(state
));
273 btrfs_leak_debug_del(&state
->leak_list
);
274 trace_free_extent_state(state
, _RET_IP_
);
275 kmem_cache_free(extent_state_cache
, state
);
279 static struct rb_node
*tree_insert(struct rb_root
*root
,
280 struct rb_node
*search_start
,
282 struct rb_node
*node
,
283 struct rb_node
***p_in
,
284 struct rb_node
**parent_in
)
287 struct rb_node
*parent
= NULL
;
288 struct tree_entry
*entry
;
290 if (p_in
&& parent_in
) {
296 p
= search_start
? &search_start
: &root
->rb_node
;
299 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
301 if (offset
< entry
->start
)
303 else if (offset
> entry
->end
)
310 rb_link_node(node
, parent
, p
);
311 rb_insert_color(node
, root
);
315 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
316 struct rb_node
**next_ret
,
317 struct rb_node
**prev_ret
,
318 struct rb_node
***p_ret
,
319 struct rb_node
**parent_ret
)
321 struct rb_root
*root
= &tree
->state
;
322 struct rb_node
**n
= &root
->rb_node
;
323 struct rb_node
*prev
= NULL
;
324 struct rb_node
*orig_prev
= NULL
;
325 struct tree_entry
*entry
;
326 struct tree_entry
*prev_entry
= NULL
;
330 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
333 if (offset
< entry
->start
)
335 else if (offset
> entry
->end
)
348 while (prev
&& offset
> prev_entry
->end
) {
349 prev
= rb_next(prev
);
350 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
357 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
358 while (prev
&& offset
< prev_entry
->start
) {
359 prev
= rb_prev(prev
);
360 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
367 static inline struct rb_node
*
368 tree_search_for_insert(struct extent_io_tree
*tree
,
370 struct rb_node
***p_ret
,
371 struct rb_node
**parent_ret
)
373 struct rb_node
*next
= NULL
;
376 ret
= __etree_search(tree
, offset
, &next
, NULL
, p_ret
, parent_ret
);
382 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
385 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
389 * utility function to look for merge candidates inside a given range.
390 * Any extents with matching state are merged together into a single
391 * extent in the tree. Extents with EXTENT_IO in their state field
392 * are not merged because the end_io handlers need to be able to do
393 * operations on them without sleeping (or doing allocations/splits).
395 * This should be called with the tree lock held.
397 static void merge_state(struct extent_io_tree
*tree
,
398 struct extent_state
*state
)
400 struct extent_state
*other
;
401 struct rb_node
*other_node
;
403 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
406 other_node
= rb_prev(&state
->rb_node
);
408 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
409 if (other
->end
== state
->start
- 1 &&
410 other
->state
== state
->state
) {
411 if (tree
->private_data
&&
412 is_data_inode(tree
->private_data
))
413 btrfs_merge_delalloc_extent(tree
->private_data
,
415 state
->start
= other
->start
;
416 rb_erase(&other
->rb_node
, &tree
->state
);
417 RB_CLEAR_NODE(&other
->rb_node
);
418 free_extent_state(other
);
421 other_node
= rb_next(&state
->rb_node
);
423 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
424 if (other
->start
== state
->end
+ 1 &&
425 other
->state
== state
->state
) {
426 if (tree
->private_data
&&
427 is_data_inode(tree
->private_data
))
428 btrfs_merge_delalloc_extent(tree
->private_data
,
430 state
->end
= other
->end
;
431 rb_erase(&other
->rb_node
, &tree
->state
);
432 RB_CLEAR_NODE(&other
->rb_node
);
433 free_extent_state(other
);
438 static void set_state_bits(struct extent_io_tree
*tree
,
439 struct extent_state
*state
, unsigned *bits
,
440 struct extent_changeset
*changeset
);
443 * insert an extent_state struct into the tree. 'bits' are set on the
444 * struct before it is inserted.
446 * This may return -EEXIST if the extent is already there, in which case the
447 * state struct is freed.
449 * The tree lock is not taken internally. This is a utility function and
450 * probably isn't what you want to call (see set/clear_extent_bit).
452 static int insert_state(struct extent_io_tree
*tree
,
453 struct extent_state
*state
, u64 start
, u64 end
,
455 struct rb_node
**parent
,
456 unsigned *bits
, struct extent_changeset
*changeset
)
458 struct rb_node
*node
;
461 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
463 state
->start
= start
;
466 set_state_bits(tree
, state
, bits
, changeset
);
468 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
470 struct extent_state
*found
;
471 found
= rb_entry(node
, struct extent_state
, rb_node
);
472 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
473 found
->start
, found
->end
, start
, end
);
476 merge_state(tree
, state
);
481 * split a given extent state struct in two, inserting the preallocated
482 * struct 'prealloc' as the newly created second half. 'split' indicates an
483 * offset inside 'orig' where it should be split.
486 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
487 * are two extent state structs in the tree:
488 * prealloc: [orig->start, split - 1]
489 * orig: [ split, orig->end ]
491 * The tree locks are not taken by this function. They need to be held
494 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
495 struct extent_state
*prealloc
, u64 split
)
497 struct rb_node
*node
;
499 if (tree
->private_data
&& is_data_inode(tree
->private_data
))
500 btrfs_split_delalloc_extent(tree
->private_data
, orig
, split
);
502 prealloc
->start
= orig
->start
;
503 prealloc
->end
= split
- 1;
504 prealloc
->state
= orig
->state
;
507 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
508 &prealloc
->rb_node
, NULL
, NULL
);
510 free_extent_state(prealloc
);
516 static struct extent_state
*next_state(struct extent_state
*state
)
518 struct rb_node
*next
= rb_next(&state
->rb_node
);
520 return rb_entry(next
, struct extent_state
, rb_node
);
526 * utility function to clear some bits in an extent state struct.
527 * it will optionally wake up anyone waiting on this state (wake == 1).
529 * If no bits are set on the state struct after clearing things, the
530 * struct is freed and removed from the tree
532 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
533 struct extent_state
*state
,
534 unsigned *bits
, int wake
,
535 struct extent_changeset
*changeset
)
537 struct extent_state
*next
;
538 unsigned bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
541 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
542 u64 range
= state
->end
- state
->start
+ 1;
543 WARN_ON(range
> tree
->dirty_bytes
);
544 tree
->dirty_bytes
-= range
;
547 if (tree
->private_data
&& is_data_inode(tree
->private_data
))
548 btrfs_clear_delalloc_extent(tree
->private_data
, state
, bits
);
550 ret
= add_extent_changeset(state
, bits_to_clear
, changeset
, 0);
552 state
->state
&= ~bits_to_clear
;
555 if (state
->state
== 0) {
556 next
= next_state(state
);
557 if (extent_state_in_tree(state
)) {
558 rb_erase(&state
->rb_node
, &tree
->state
);
559 RB_CLEAR_NODE(&state
->rb_node
);
560 free_extent_state(state
);
565 merge_state(tree
, state
);
566 next
= next_state(state
);
571 static struct extent_state
*
572 alloc_extent_state_atomic(struct extent_state
*prealloc
)
575 prealloc
= alloc_extent_state(GFP_ATOMIC
);
580 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
582 struct inode
*inode
= tree
->private_data
;
584 btrfs_panic(btrfs_sb(inode
->i_sb
), err
,
585 "locking error: extent tree was modified by another thread while locked");
589 * clear some bits on a range in the tree. This may require splitting
590 * or inserting elements in the tree, so the gfp mask is used to
591 * indicate which allocations or sleeping are allowed.
593 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
594 * the given range from the tree regardless of state (ie for truncate).
596 * the range [start, end] is inclusive.
598 * This takes the tree lock, and returns 0 on success and < 0 on error.
600 int __clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
601 unsigned bits
, int wake
, int delete,
602 struct extent_state
**cached_state
,
603 gfp_t mask
, struct extent_changeset
*changeset
)
605 struct extent_state
*state
;
606 struct extent_state
*cached
;
607 struct extent_state
*prealloc
= NULL
;
608 struct rb_node
*node
;
613 btrfs_debug_check_extent_io_range(tree
, start
, end
);
615 if (bits
& EXTENT_DELALLOC
)
616 bits
|= EXTENT_NORESERVE
;
619 bits
|= ~EXTENT_CTLBITS
;
621 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
624 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
626 * Don't care for allocation failure here because we might end
627 * up not needing the pre-allocated extent state at all, which
628 * is the case if we only have in the tree extent states that
629 * cover our input range and don't cover too any other range.
630 * If we end up needing a new extent state we allocate it later.
632 prealloc
= alloc_extent_state(mask
);
635 spin_lock(&tree
->lock
);
637 cached
= *cached_state
;
640 *cached_state
= NULL
;
644 if (cached
&& extent_state_in_tree(cached
) &&
645 cached
->start
<= start
&& cached
->end
> start
) {
647 refcount_dec(&cached
->refs
);
652 free_extent_state(cached
);
655 * this search will find the extents that end after
658 node
= tree_search(tree
, start
);
661 state
= rb_entry(node
, struct extent_state
, rb_node
);
663 if (state
->start
> end
)
665 WARN_ON(state
->end
< start
);
666 last_end
= state
->end
;
668 /* the state doesn't have the wanted bits, go ahead */
669 if (!(state
->state
& bits
)) {
670 state
= next_state(state
);
675 * | ---- desired range ---- |
677 * | ------------- state -------------- |
679 * We need to split the extent we found, and may flip
680 * bits on second half.
682 * If the extent we found extends past our range, we
683 * just split and search again. It'll get split again
684 * the next time though.
686 * If the extent we found is inside our range, we clear
687 * the desired bit on it.
690 if (state
->start
< start
) {
691 prealloc
= alloc_extent_state_atomic(prealloc
);
693 err
= split_state(tree
, state
, prealloc
, start
);
695 extent_io_tree_panic(tree
, err
);
700 if (state
->end
<= end
) {
701 state
= clear_state_bit(tree
, state
, &bits
, wake
,
708 * | ---- desired range ---- |
710 * We need to split the extent, and clear the bit
713 if (state
->start
<= end
&& state
->end
> end
) {
714 prealloc
= alloc_extent_state_atomic(prealloc
);
716 err
= split_state(tree
, state
, prealloc
, end
+ 1);
718 extent_io_tree_panic(tree
, err
);
723 clear_state_bit(tree
, prealloc
, &bits
, wake
, changeset
);
729 state
= clear_state_bit(tree
, state
, &bits
, wake
, changeset
);
731 if (last_end
== (u64
)-1)
733 start
= last_end
+ 1;
734 if (start
<= end
&& state
&& !need_resched())
740 spin_unlock(&tree
->lock
);
741 if (gfpflags_allow_blocking(mask
))
746 spin_unlock(&tree
->lock
);
748 free_extent_state(prealloc
);
754 static void wait_on_state(struct extent_io_tree
*tree
,
755 struct extent_state
*state
)
756 __releases(tree
->lock
)
757 __acquires(tree
->lock
)
760 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
761 spin_unlock(&tree
->lock
);
763 spin_lock(&tree
->lock
);
764 finish_wait(&state
->wq
, &wait
);
768 * waits for one or more bits to clear on a range in the state tree.
769 * The range [start, end] is inclusive.
770 * The tree lock is taken by this function
772 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
775 struct extent_state
*state
;
776 struct rb_node
*node
;
778 btrfs_debug_check_extent_io_range(tree
, start
, end
);
780 spin_lock(&tree
->lock
);
784 * this search will find all the extents that end after
787 node
= tree_search(tree
, start
);
792 state
= rb_entry(node
, struct extent_state
, rb_node
);
794 if (state
->start
> end
)
797 if (state
->state
& bits
) {
798 start
= state
->start
;
799 refcount_inc(&state
->refs
);
800 wait_on_state(tree
, state
);
801 free_extent_state(state
);
804 start
= state
->end
+ 1;
809 if (!cond_resched_lock(&tree
->lock
)) {
810 node
= rb_next(node
);
815 spin_unlock(&tree
->lock
);
818 static void set_state_bits(struct extent_io_tree
*tree
,
819 struct extent_state
*state
,
820 unsigned *bits
, struct extent_changeset
*changeset
)
822 unsigned bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
825 if (tree
->private_data
&& is_data_inode(tree
->private_data
))
826 btrfs_set_delalloc_extent(tree
->private_data
, state
, bits
);
828 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
829 u64 range
= state
->end
- state
->start
+ 1;
830 tree
->dirty_bytes
+= range
;
832 ret
= add_extent_changeset(state
, bits_to_set
, changeset
, 1);
834 state
->state
|= bits_to_set
;
837 static void cache_state_if_flags(struct extent_state
*state
,
838 struct extent_state
**cached_ptr
,
841 if (cached_ptr
&& !(*cached_ptr
)) {
842 if (!flags
|| (state
->state
& flags
)) {
844 refcount_inc(&state
->refs
);
849 static void cache_state(struct extent_state
*state
,
850 struct extent_state
**cached_ptr
)
852 return cache_state_if_flags(state
, cached_ptr
,
853 EXTENT_IOBITS
| EXTENT_BOUNDARY
);
857 * set some bits on a range in the tree. This may require allocations or
858 * sleeping, so the gfp mask is used to indicate what is allowed.
860 * If any of the exclusive bits are set, this will fail with -EEXIST if some
861 * part of the range already has the desired bits set. The start of the
862 * existing range is returned in failed_start in this case.
864 * [start, end] is inclusive This takes the tree lock.
867 static int __must_check
868 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
869 unsigned bits
, unsigned exclusive_bits
,
870 u64
*failed_start
, struct extent_state
**cached_state
,
871 gfp_t mask
, struct extent_changeset
*changeset
)
873 struct extent_state
*state
;
874 struct extent_state
*prealloc
= NULL
;
875 struct rb_node
*node
;
877 struct rb_node
*parent
;
882 btrfs_debug_check_extent_io_range(tree
, start
, end
);
885 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
887 * Don't care for allocation failure here because we might end
888 * up not needing the pre-allocated extent state at all, which
889 * is the case if we only have in the tree extent states that
890 * cover our input range and don't cover too any other range.
891 * If we end up needing a new extent state we allocate it later.
893 prealloc
= alloc_extent_state(mask
);
896 spin_lock(&tree
->lock
);
897 if (cached_state
&& *cached_state
) {
898 state
= *cached_state
;
899 if (state
->start
<= start
&& state
->end
> start
&&
900 extent_state_in_tree(state
)) {
901 node
= &state
->rb_node
;
906 * this search will find all the extents that end after
909 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
911 prealloc
= alloc_extent_state_atomic(prealloc
);
913 err
= insert_state(tree
, prealloc
, start
, end
,
914 &p
, &parent
, &bits
, changeset
);
916 extent_io_tree_panic(tree
, err
);
918 cache_state(prealloc
, cached_state
);
922 state
= rb_entry(node
, struct extent_state
, rb_node
);
924 last_start
= state
->start
;
925 last_end
= state
->end
;
928 * | ---- desired range ---- |
931 * Just lock what we found and keep going
933 if (state
->start
== start
&& state
->end
<= end
) {
934 if (state
->state
& exclusive_bits
) {
935 *failed_start
= state
->start
;
940 set_state_bits(tree
, state
, &bits
, changeset
);
941 cache_state(state
, cached_state
);
942 merge_state(tree
, state
);
943 if (last_end
== (u64
)-1)
945 start
= last_end
+ 1;
946 state
= next_state(state
);
947 if (start
< end
&& state
&& state
->start
== start
&&
954 * | ---- desired range ---- |
957 * | ------------- state -------------- |
959 * We need to split the extent we found, and may flip bits on
962 * If the extent we found extends past our
963 * range, we just split and search again. It'll get split
964 * again the next time though.
966 * If the extent we found is inside our range, we set the
969 if (state
->start
< start
) {
970 if (state
->state
& exclusive_bits
) {
971 *failed_start
= start
;
976 prealloc
= alloc_extent_state_atomic(prealloc
);
978 err
= split_state(tree
, state
, prealloc
, start
);
980 extent_io_tree_panic(tree
, err
);
985 if (state
->end
<= end
) {
986 set_state_bits(tree
, state
, &bits
, changeset
);
987 cache_state(state
, cached_state
);
988 merge_state(tree
, state
);
989 if (last_end
== (u64
)-1)
991 start
= last_end
+ 1;
992 state
= next_state(state
);
993 if (start
< end
&& state
&& state
->start
== start
&&
1000 * | ---- desired range ---- |
1001 * | state | or | state |
1003 * There's a hole, we need to insert something in it and
1004 * ignore the extent we found.
1006 if (state
->start
> start
) {
1008 if (end
< last_start
)
1011 this_end
= last_start
- 1;
1013 prealloc
= alloc_extent_state_atomic(prealloc
);
1017 * Avoid to free 'prealloc' if it can be merged with
1020 err
= insert_state(tree
, prealloc
, start
, this_end
,
1021 NULL
, NULL
, &bits
, changeset
);
1023 extent_io_tree_panic(tree
, err
);
1025 cache_state(prealloc
, cached_state
);
1027 start
= this_end
+ 1;
1031 * | ---- desired range ---- |
1033 * We need to split the extent, and set the bit
1036 if (state
->start
<= end
&& state
->end
> end
) {
1037 if (state
->state
& exclusive_bits
) {
1038 *failed_start
= start
;
1043 prealloc
= alloc_extent_state_atomic(prealloc
);
1045 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1047 extent_io_tree_panic(tree
, err
);
1049 set_state_bits(tree
, prealloc
, &bits
, changeset
);
1050 cache_state(prealloc
, cached_state
);
1051 merge_state(tree
, prealloc
);
1059 spin_unlock(&tree
->lock
);
1060 if (gfpflags_allow_blocking(mask
))
1065 spin_unlock(&tree
->lock
);
1067 free_extent_state(prealloc
);
1073 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1074 unsigned bits
, u64
* failed_start
,
1075 struct extent_state
**cached_state
, gfp_t mask
)
1077 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1078 cached_state
, mask
, NULL
);
1083 * convert_extent_bit - convert all bits in a given range from one bit to
1085 * @tree: the io tree to search
1086 * @start: the start offset in bytes
1087 * @end: the end offset in bytes (inclusive)
1088 * @bits: the bits to set in this range
1089 * @clear_bits: the bits to clear in this range
1090 * @cached_state: state that we're going to cache
1092 * This will go through and set bits for the given range. If any states exist
1093 * already in this range they are set with the given bit and cleared of the
1094 * clear_bits. This is only meant to be used by things that are mergeable, ie
1095 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1096 * boundary bits like LOCK.
1098 * All allocations are done with GFP_NOFS.
1100 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1101 unsigned bits
, unsigned clear_bits
,
1102 struct extent_state
**cached_state
)
1104 struct extent_state
*state
;
1105 struct extent_state
*prealloc
= NULL
;
1106 struct rb_node
*node
;
1108 struct rb_node
*parent
;
1112 bool first_iteration
= true;
1114 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1119 * Best effort, don't worry if extent state allocation fails
1120 * here for the first iteration. We might have a cached state
1121 * that matches exactly the target range, in which case no
1122 * extent state allocations are needed. We'll only know this
1123 * after locking the tree.
1125 prealloc
= alloc_extent_state(GFP_NOFS
);
1126 if (!prealloc
&& !first_iteration
)
1130 spin_lock(&tree
->lock
);
1131 if (cached_state
&& *cached_state
) {
1132 state
= *cached_state
;
1133 if (state
->start
<= start
&& state
->end
> start
&&
1134 extent_state_in_tree(state
)) {
1135 node
= &state
->rb_node
;
1141 * this search will find all the extents that end after
1144 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1146 prealloc
= alloc_extent_state_atomic(prealloc
);
1151 err
= insert_state(tree
, prealloc
, start
, end
,
1152 &p
, &parent
, &bits
, NULL
);
1154 extent_io_tree_panic(tree
, err
);
1155 cache_state(prealloc
, cached_state
);
1159 state
= rb_entry(node
, struct extent_state
, rb_node
);
1161 last_start
= state
->start
;
1162 last_end
= state
->end
;
1165 * | ---- desired range ---- |
1168 * Just lock what we found and keep going
1170 if (state
->start
== start
&& state
->end
<= end
) {
1171 set_state_bits(tree
, state
, &bits
, NULL
);
1172 cache_state(state
, cached_state
);
1173 state
= clear_state_bit(tree
, state
, &clear_bits
, 0, NULL
);
1174 if (last_end
== (u64
)-1)
1176 start
= last_end
+ 1;
1177 if (start
< end
&& state
&& state
->start
== start
&&
1184 * | ---- desired range ---- |
1187 * | ------------- state -------------- |
1189 * We need to split the extent we found, and may flip bits on
1192 * If the extent we found extends past our
1193 * range, we just split and search again. It'll get split
1194 * again the next time though.
1196 * If the extent we found is inside our range, we set the
1197 * desired bit on it.
1199 if (state
->start
< start
) {
1200 prealloc
= alloc_extent_state_atomic(prealloc
);
1205 err
= split_state(tree
, state
, prealloc
, start
);
1207 extent_io_tree_panic(tree
, err
);
1211 if (state
->end
<= end
) {
1212 set_state_bits(tree
, state
, &bits
, NULL
);
1213 cache_state(state
, cached_state
);
1214 state
= clear_state_bit(tree
, state
, &clear_bits
, 0,
1216 if (last_end
== (u64
)-1)
1218 start
= last_end
+ 1;
1219 if (start
< end
&& state
&& state
->start
== start
&&
1226 * | ---- desired range ---- |
1227 * | state | or | state |
1229 * There's a hole, we need to insert something in it and
1230 * ignore the extent we found.
1232 if (state
->start
> start
) {
1234 if (end
< last_start
)
1237 this_end
= last_start
- 1;
1239 prealloc
= alloc_extent_state_atomic(prealloc
);
1246 * Avoid to free 'prealloc' if it can be merged with
1249 err
= insert_state(tree
, prealloc
, start
, this_end
,
1250 NULL
, NULL
, &bits
, NULL
);
1252 extent_io_tree_panic(tree
, err
);
1253 cache_state(prealloc
, cached_state
);
1255 start
= this_end
+ 1;
1259 * | ---- desired range ---- |
1261 * We need to split the extent, and set the bit
1264 if (state
->start
<= end
&& state
->end
> end
) {
1265 prealloc
= alloc_extent_state_atomic(prealloc
);
1271 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1273 extent_io_tree_panic(tree
, err
);
1275 set_state_bits(tree
, prealloc
, &bits
, NULL
);
1276 cache_state(prealloc
, cached_state
);
1277 clear_state_bit(tree
, prealloc
, &clear_bits
, 0, NULL
);
1285 spin_unlock(&tree
->lock
);
1287 first_iteration
= false;
1291 spin_unlock(&tree
->lock
);
1293 free_extent_state(prealloc
);
1298 /* wrappers around set/clear extent bit */
1299 int set_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1300 unsigned bits
, struct extent_changeset
*changeset
)
1303 * We don't support EXTENT_LOCKED yet, as current changeset will
1304 * record any bits changed, so for EXTENT_LOCKED case, it will
1305 * either fail with -EEXIST or changeset will record the whole
1308 BUG_ON(bits
& EXTENT_LOCKED
);
1310 return __set_extent_bit(tree
, start
, end
, bits
, 0, NULL
, NULL
, GFP_NOFS
,
1314 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1315 unsigned bits
, int wake
, int delete,
1316 struct extent_state
**cached
)
1318 return __clear_extent_bit(tree
, start
, end
, bits
, wake
, delete,
1319 cached
, GFP_NOFS
, NULL
);
1322 int clear_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1323 unsigned bits
, struct extent_changeset
*changeset
)
1326 * Don't support EXTENT_LOCKED case, same reason as
1327 * set_record_extent_bits().
1329 BUG_ON(bits
& EXTENT_LOCKED
);
1331 return __clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, GFP_NOFS
,
1336 * either insert or lock state struct between start and end use mask to tell
1337 * us if waiting is desired.
1339 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1340 struct extent_state
**cached_state
)
1346 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
,
1347 EXTENT_LOCKED
, &failed_start
,
1348 cached_state
, GFP_NOFS
, NULL
);
1349 if (err
== -EEXIST
) {
1350 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1351 start
= failed_start
;
1354 WARN_ON(start
> end
);
1359 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1364 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1365 &failed_start
, NULL
, GFP_NOFS
, NULL
);
1366 if (err
== -EEXIST
) {
1367 if (failed_start
> start
)
1368 clear_extent_bit(tree
, start
, failed_start
- 1,
1369 EXTENT_LOCKED
, 1, 0, NULL
);
1375 void extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1377 unsigned long index
= start
>> PAGE_SHIFT
;
1378 unsigned long end_index
= end
>> PAGE_SHIFT
;
1381 while (index
<= end_index
) {
1382 page
= find_get_page(inode
->i_mapping
, index
);
1383 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1384 clear_page_dirty_for_io(page
);
1390 void extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1392 unsigned long index
= start
>> PAGE_SHIFT
;
1393 unsigned long end_index
= end
>> PAGE_SHIFT
;
1396 while (index
<= end_index
) {
1397 page
= find_get_page(inode
->i_mapping
, index
);
1398 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1399 __set_page_dirty_nobuffers(page
);
1400 account_page_redirty(page
);
1406 /* find the first state struct with 'bits' set after 'start', and
1407 * return it. tree->lock must be held. NULL will returned if
1408 * nothing was found after 'start'
1410 static struct extent_state
*
1411 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1412 u64 start
, unsigned bits
)
1414 struct rb_node
*node
;
1415 struct extent_state
*state
;
1418 * this search will find all the extents that end after
1421 node
= tree_search(tree
, start
);
1426 state
= rb_entry(node
, struct extent_state
, rb_node
);
1427 if (state
->end
>= start
&& (state
->state
& bits
))
1430 node
= rb_next(node
);
1439 * find the first offset in the io tree with 'bits' set. zero is
1440 * returned if we find something, and *start_ret and *end_ret are
1441 * set to reflect the state struct that was found.
1443 * If nothing was found, 1 is returned. If found something, return 0.
1445 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1446 u64
*start_ret
, u64
*end_ret
, unsigned bits
,
1447 struct extent_state
**cached_state
)
1449 struct extent_state
*state
;
1452 spin_lock(&tree
->lock
);
1453 if (cached_state
&& *cached_state
) {
1454 state
= *cached_state
;
1455 if (state
->end
== start
- 1 && extent_state_in_tree(state
)) {
1456 while ((state
= next_state(state
)) != NULL
) {
1457 if (state
->state
& bits
)
1460 free_extent_state(*cached_state
);
1461 *cached_state
= NULL
;
1464 free_extent_state(*cached_state
);
1465 *cached_state
= NULL
;
1468 state
= find_first_extent_bit_state(tree
, start
, bits
);
1471 cache_state_if_flags(state
, cached_state
, 0);
1472 *start_ret
= state
->start
;
1473 *end_ret
= state
->end
;
1477 spin_unlock(&tree
->lock
);
1482 * find a contiguous range of bytes in the file marked as delalloc, not
1483 * more than 'max_bytes'. start and end are used to return the range,
1485 * true is returned if we find something, false if nothing was in the tree
1487 static noinline
bool find_delalloc_range(struct extent_io_tree
*tree
,
1488 u64
*start
, u64
*end
, u64 max_bytes
,
1489 struct extent_state
**cached_state
)
1491 struct rb_node
*node
;
1492 struct extent_state
*state
;
1493 u64 cur_start
= *start
;
1495 u64 total_bytes
= 0;
1497 spin_lock(&tree
->lock
);
1500 * this search will find all the extents that end after
1503 node
= tree_search(tree
, cur_start
);
1510 state
= rb_entry(node
, struct extent_state
, rb_node
);
1511 if (found
&& (state
->start
!= cur_start
||
1512 (state
->state
& EXTENT_BOUNDARY
))) {
1515 if (!(state
->state
& EXTENT_DELALLOC
)) {
1521 *start
= state
->start
;
1522 *cached_state
= state
;
1523 refcount_inc(&state
->refs
);
1527 cur_start
= state
->end
+ 1;
1528 node
= rb_next(node
);
1529 total_bytes
+= state
->end
- state
->start
+ 1;
1530 if (total_bytes
>= max_bytes
)
1536 spin_unlock(&tree
->lock
);
1540 static int __process_pages_contig(struct address_space
*mapping
,
1541 struct page
*locked_page
,
1542 pgoff_t start_index
, pgoff_t end_index
,
1543 unsigned long page_ops
, pgoff_t
*index_ret
);
1545 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1546 struct page
*locked_page
,
1549 unsigned long index
= start
>> PAGE_SHIFT
;
1550 unsigned long end_index
= end
>> PAGE_SHIFT
;
1552 ASSERT(locked_page
);
1553 if (index
== locked_page
->index
&& end_index
== index
)
1556 __process_pages_contig(inode
->i_mapping
, locked_page
, index
, end_index
,
1560 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1561 struct page
*locked_page
,
1565 unsigned long index
= delalloc_start
>> PAGE_SHIFT
;
1566 unsigned long index_ret
= index
;
1567 unsigned long end_index
= delalloc_end
>> PAGE_SHIFT
;
1570 ASSERT(locked_page
);
1571 if (index
== locked_page
->index
&& index
== end_index
)
1574 ret
= __process_pages_contig(inode
->i_mapping
, locked_page
, index
,
1575 end_index
, PAGE_LOCK
, &index_ret
);
1577 __unlock_for_delalloc(inode
, locked_page
, delalloc_start
,
1578 (u64
)index_ret
<< PAGE_SHIFT
);
1583 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1584 * more than @max_bytes. @Start and @end are used to return the range,
1586 * Return: true if we find something
1587 * false if nothing was in the tree
1590 noinline_for_stack
bool find_lock_delalloc_range(struct inode
*inode
,
1591 struct extent_io_tree
*tree
,
1592 struct page
*locked_page
, u64
*start
,
1595 u64 max_bytes
= BTRFS_MAX_EXTENT_SIZE
;
1599 struct extent_state
*cached_state
= NULL
;
1604 /* step one, find a bunch of delalloc bytes starting at start */
1605 delalloc_start
= *start
;
1607 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1608 max_bytes
, &cached_state
);
1609 if (!found
|| delalloc_end
<= *start
) {
1610 *start
= delalloc_start
;
1611 *end
= delalloc_end
;
1612 free_extent_state(cached_state
);
1617 * start comes from the offset of locked_page. We have to lock
1618 * pages in order, so we can't process delalloc bytes before
1621 if (delalloc_start
< *start
)
1622 delalloc_start
= *start
;
1625 * make sure to limit the number of pages we try to lock down
1627 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1628 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1630 /* step two, lock all the pages after the page that has start */
1631 ret
= lock_delalloc_pages(inode
, locked_page
,
1632 delalloc_start
, delalloc_end
);
1633 ASSERT(!ret
|| ret
== -EAGAIN
);
1634 if (ret
== -EAGAIN
) {
1635 /* some of the pages are gone, lets avoid looping by
1636 * shortening the size of the delalloc range we're searching
1638 free_extent_state(cached_state
);
1639 cached_state
= NULL
;
1641 max_bytes
= PAGE_SIZE
;
1650 /* step three, lock the state bits for the whole range */
1651 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, &cached_state
);
1653 /* then test to make sure it is all still delalloc */
1654 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1655 EXTENT_DELALLOC
, 1, cached_state
);
1657 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1659 __unlock_for_delalloc(inode
, locked_page
,
1660 delalloc_start
, delalloc_end
);
1664 free_extent_state(cached_state
);
1665 *start
= delalloc_start
;
1666 *end
= delalloc_end
;
1671 static int __process_pages_contig(struct address_space
*mapping
,
1672 struct page
*locked_page
,
1673 pgoff_t start_index
, pgoff_t end_index
,
1674 unsigned long page_ops
, pgoff_t
*index_ret
)
1676 unsigned long nr_pages
= end_index
- start_index
+ 1;
1677 unsigned long pages_locked
= 0;
1678 pgoff_t index
= start_index
;
1679 struct page
*pages
[16];
1684 if (page_ops
& PAGE_LOCK
) {
1685 ASSERT(page_ops
== PAGE_LOCK
);
1686 ASSERT(index_ret
&& *index_ret
== start_index
);
1689 if ((page_ops
& PAGE_SET_ERROR
) && nr_pages
> 0)
1690 mapping_set_error(mapping
, -EIO
);
1692 while (nr_pages
> 0) {
1693 ret
= find_get_pages_contig(mapping
, index
,
1694 min_t(unsigned long,
1695 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1698 * Only if we're going to lock these pages,
1699 * can we find nothing at @index.
1701 ASSERT(page_ops
& PAGE_LOCK
);
1706 for (i
= 0; i
< ret
; i
++) {
1707 if (page_ops
& PAGE_SET_PRIVATE2
)
1708 SetPagePrivate2(pages
[i
]);
1710 if (pages
[i
] == locked_page
) {
1715 if (page_ops
& PAGE_CLEAR_DIRTY
)
1716 clear_page_dirty_for_io(pages
[i
]);
1717 if (page_ops
& PAGE_SET_WRITEBACK
)
1718 set_page_writeback(pages
[i
]);
1719 if (page_ops
& PAGE_SET_ERROR
)
1720 SetPageError(pages
[i
]);
1721 if (page_ops
& PAGE_END_WRITEBACK
)
1722 end_page_writeback(pages
[i
]);
1723 if (page_ops
& PAGE_UNLOCK
)
1724 unlock_page(pages
[i
]);
1725 if (page_ops
& PAGE_LOCK
) {
1726 lock_page(pages
[i
]);
1727 if (!PageDirty(pages
[i
]) ||
1728 pages
[i
]->mapping
!= mapping
) {
1729 unlock_page(pages
[i
]);
1743 if (err
&& index_ret
)
1744 *index_ret
= start_index
+ pages_locked
- 1;
1748 void extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1749 u64 delalloc_end
, struct page
*locked_page
,
1750 unsigned clear_bits
,
1751 unsigned long page_ops
)
1753 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, clear_bits
, 1, 0,
1756 __process_pages_contig(inode
->i_mapping
, locked_page
,
1757 start
>> PAGE_SHIFT
, end
>> PAGE_SHIFT
,
1762 * count the number of bytes in the tree that have a given bit(s)
1763 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1764 * cached. The total number found is returned.
1766 u64
count_range_bits(struct extent_io_tree
*tree
,
1767 u64
*start
, u64 search_end
, u64 max_bytes
,
1768 unsigned bits
, int contig
)
1770 struct rb_node
*node
;
1771 struct extent_state
*state
;
1772 u64 cur_start
= *start
;
1773 u64 total_bytes
= 0;
1777 if (WARN_ON(search_end
<= cur_start
))
1780 spin_lock(&tree
->lock
);
1781 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1782 total_bytes
= tree
->dirty_bytes
;
1786 * this search will find all the extents that end after
1789 node
= tree_search(tree
, cur_start
);
1794 state
= rb_entry(node
, struct extent_state
, rb_node
);
1795 if (state
->start
> search_end
)
1797 if (contig
&& found
&& state
->start
> last
+ 1)
1799 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1800 total_bytes
+= min(search_end
, state
->end
) + 1 -
1801 max(cur_start
, state
->start
);
1802 if (total_bytes
>= max_bytes
)
1805 *start
= max(cur_start
, state
->start
);
1809 } else if (contig
&& found
) {
1812 node
= rb_next(node
);
1817 spin_unlock(&tree
->lock
);
1822 * set the private field for a given byte offset in the tree. If there isn't
1823 * an extent_state there already, this does nothing.
1825 static noinline
int set_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1826 struct io_failure_record
*failrec
)
1828 struct rb_node
*node
;
1829 struct extent_state
*state
;
1832 spin_lock(&tree
->lock
);
1834 * this search will find all the extents that end after
1837 node
= tree_search(tree
, start
);
1842 state
= rb_entry(node
, struct extent_state
, rb_node
);
1843 if (state
->start
!= start
) {
1847 state
->failrec
= failrec
;
1849 spin_unlock(&tree
->lock
);
1853 static noinline
int get_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1854 struct io_failure_record
**failrec
)
1856 struct rb_node
*node
;
1857 struct extent_state
*state
;
1860 spin_lock(&tree
->lock
);
1862 * this search will find all the extents that end after
1865 node
= tree_search(tree
, start
);
1870 state
= rb_entry(node
, struct extent_state
, rb_node
);
1871 if (state
->start
!= start
) {
1875 *failrec
= state
->failrec
;
1877 spin_unlock(&tree
->lock
);
1882 * searches a range in the state tree for a given mask.
1883 * If 'filled' == 1, this returns 1 only if every extent in the tree
1884 * has the bits set. Otherwise, 1 is returned if any bit in the
1885 * range is found set.
1887 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1888 unsigned bits
, int filled
, struct extent_state
*cached
)
1890 struct extent_state
*state
= NULL
;
1891 struct rb_node
*node
;
1894 spin_lock(&tree
->lock
);
1895 if (cached
&& extent_state_in_tree(cached
) && cached
->start
<= start
&&
1896 cached
->end
> start
)
1897 node
= &cached
->rb_node
;
1899 node
= tree_search(tree
, start
);
1900 while (node
&& start
<= end
) {
1901 state
= rb_entry(node
, struct extent_state
, rb_node
);
1903 if (filled
&& state
->start
> start
) {
1908 if (state
->start
> end
)
1911 if (state
->state
& bits
) {
1915 } else if (filled
) {
1920 if (state
->end
== (u64
)-1)
1923 start
= state
->end
+ 1;
1926 node
= rb_next(node
);
1933 spin_unlock(&tree
->lock
);
1938 * helper function to set a given page up to date if all the
1939 * extents in the tree for that page are up to date
1941 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1943 u64 start
= page_offset(page
);
1944 u64 end
= start
+ PAGE_SIZE
- 1;
1945 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1946 SetPageUptodate(page
);
1949 int free_io_failure(struct extent_io_tree
*failure_tree
,
1950 struct extent_io_tree
*io_tree
,
1951 struct io_failure_record
*rec
)
1956 set_state_failrec(failure_tree
, rec
->start
, NULL
);
1957 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1958 rec
->start
+ rec
->len
- 1,
1959 EXTENT_LOCKED
| EXTENT_DIRTY
);
1963 ret
= clear_extent_bits(io_tree
, rec
->start
,
1964 rec
->start
+ rec
->len
- 1,
1974 * this bypasses the standard btrfs submit functions deliberately, as
1975 * the standard behavior is to write all copies in a raid setup. here we only
1976 * want to write the one bad copy. so we do the mapping for ourselves and issue
1977 * submit_bio directly.
1978 * to avoid any synchronization issues, wait for the data after writing, which
1979 * actually prevents the read that triggered the error from finishing.
1980 * currently, there can be no more than two copies of every data bit. thus,
1981 * exactly one rewrite is required.
1983 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 ino
, u64 start
,
1984 u64 length
, u64 logical
, struct page
*page
,
1985 unsigned int pg_offset
, int mirror_num
)
1988 struct btrfs_device
*dev
;
1991 struct btrfs_bio
*bbio
= NULL
;
1994 ASSERT(!(fs_info
->sb
->s_flags
& SB_RDONLY
));
1995 BUG_ON(!mirror_num
);
1997 bio
= btrfs_io_bio_alloc(1);
1998 bio
->bi_iter
.bi_size
= 0;
1999 map_length
= length
;
2002 * Avoid races with device replace and make sure our bbio has devices
2003 * associated to its stripes that don't go away while we are doing the
2004 * read repair operation.
2006 btrfs_bio_counter_inc_blocked(fs_info
);
2007 if (btrfs_is_parity_mirror(fs_info
, logical
, length
)) {
2009 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2010 * to update all raid stripes, but here we just want to correct
2011 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2012 * stripe's dev and sector.
2014 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_READ
, logical
,
2015 &map_length
, &bbio
, 0);
2017 btrfs_bio_counter_dec(fs_info
);
2021 ASSERT(bbio
->mirror_num
== 1);
2023 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_WRITE
, logical
,
2024 &map_length
, &bbio
, mirror_num
);
2026 btrfs_bio_counter_dec(fs_info
);
2030 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2033 sector
= bbio
->stripes
[bbio
->mirror_num
- 1].physical
>> 9;
2034 bio
->bi_iter
.bi_sector
= sector
;
2035 dev
= bbio
->stripes
[bbio
->mirror_num
- 1].dev
;
2036 btrfs_put_bbio(bbio
);
2037 if (!dev
|| !dev
->bdev
||
2038 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
)) {
2039 btrfs_bio_counter_dec(fs_info
);
2043 bio_set_dev(bio
, dev
->bdev
);
2044 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
;
2045 bio_add_page(bio
, page
, length
, pg_offset
);
2047 if (btrfsic_submit_bio_wait(bio
)) {
2048 /* try to remap that extent elsewhere? */
2049 btrfs_bio_counter_dec(fs_info
);
2051 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2055 btrfs_info_rl_in_rcu(fs_info
,
2056 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2058 rcu_str_deref(dev
->name
), sector
);
2059 btrfs_bio_counter_dec(fs_info
);
2064 int repair_eb_io_failure(struct btrfs_fs_info
*fs_info
,
2065 struct extent_buffer
*eb
, int mirror_num
)
2067 u64 start
= eb
->start
;
2068 int i
, num_pages
= num_extent_pages(eb
);
2071 if (sb_rdonly(fs_info
->sb
))
2074 for (i
= 0; i
< num_pages
; i
++) {
2075 struct page
*p
= eb
->pages
[i
];
2077 ret
= repair_io_failure(fs_info
, 0, start
, PAGE_SIZE
, start
, p
,
2078 start
- page_offset(p
), mirror_num
);
2088 * each time an IO finishes, we do a fast check in the IO failure tree
2089 * to see if we need to process or clean up an io_failure_record
2091 int clean_io_failure(struct btrfs_fs_info
*fs_info
,
2092 struct extent_io_tree
*failure_tree
,
2093 struct extent_io_tree
*io_tree
, u64 start
,
2094 struct page
*page
, u64 ino
, unsigned int pg_offset
)
2097 struct io_failure_record
*failrec
;
2098 struct extent_state
*state
;
2103 ret
= count_range_bits(failure_tree
, &private, (u64
)-1, 1,
2108 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2112 BUG_ON(!failrec
->this_mirror
);
2114 if (failrec
->in_validation
) {
2115 /* there was no real error, just free the record */
2116 btrfs_debug(fs_info
,
2117 "clean_io_failure: freeing dummy error at %llu",
2121 if (sb_rdonly(fs_info
->sb
))
2124 spin_lock(&io_tree
->lock
);
2125 state
= find_first_extent_bit_state(io_tree
,
2128 spin_unlock(&io_tree
->lock
);
2130 if (state
&& state
->start
<= failrec
->start
&&
2131 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2132 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2134 if (num_copies
> 1) {
2135 repair_io_failure(fs_info
, ino
, start
, failrec
->len
,
2136 failrec
->logical
, page
, pg_offset
,
2137 failrec
->failed_mirror
);
2142 free_io_failure(failure_tree
, io_tree
, failrec
);
2148 * Can be called when
2149 * - hold extent lock
2150 * - under ordered extent
2151 * - the inode is freeing
2153 void btrfs_free_io_failure_record(struct btrfs_inode
*inode
, u64 start
, u64 end
)
2155 struct extent_io_tree
*failure_tree
= &inode
->io_failure_tree
;
2156 struct io_failure_record
*failrec
;
2157 struct extent_state
*state
, *next
;
2159 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2162 spin_lock(&failure_tree
->lock
);
2163 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2165 if (state
->start
> end
)
2168 ASSERT(state
->end
<= end
);
2170 next
= next_state(state
);
2172 failrec
= state
->failrec
;
2173 free_extent_state(state
);
2178 spin_unlock(&failure_tree
->lock
);
2181 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2182 struct io_failure_record
**failrec_ret
)
2184 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2185 struct io_failure_record
*failrec
;
2186 struct extent_map
*em
;
2187 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2188 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2189 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2193 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2195 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2199 failrec
->start
= start
;
2200 failrec
->len
= end
- start
+ 1;
2201 failrec
->this_mirror
= 0;
2202 failrec
->bio_flags
= 0;
2203 failrec
->in_validation
= 0;
2205 read_lock(&em_tree
->lock
);
2206 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2208 read_unlock(&em_tree
->lock
);
2213 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2214 free_extent_map(em
);
2217 read_unlock(&em_tree
->lock
);
2223 logical
= start
- em
->start
;
2224 logical
= em
->block_start
+ logical
;
2225 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2226 logical
= em
->block_start
;
2227 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2228 extent_set_compress_type(&failrec
->bio_flags
,
2232 btrfs_debug(fs_info
,
2233 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2234 logical
, start
, failrec
->len
);
2236 failrec
->logical
= logical
;
2237 free_extent_map(em
);
2239 /* set the bits in the private failure tree */
2240 ret
= set_extent_bits(failure_tree
, start
, end
,
2241 EXTENT_LOCKED
| EXTENT_DIRTY
);
2243 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2244 /* set the bits in the inode's tree */
2246 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2252 btrfs_debug(fs_info
,
2253 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2254 failrec
->logical
, failrec
->start
, failrec
->len
,
2255 failrec
->in_validation
);
2257 * when data can be on disk more than twice, add to failrec here
2258 * (e.g. with a list for failed_mirror) to make
2259 * clean_io_failure() clean all those errors at once.
2263 *failrec_ret
= failrec
;
2268 bool btrfs_check_repairable(struct inode
*inode
, unsigned failed_bio_pages
,
2269 struct io_failure_record
*failrec
, int failed_mirror
)
2271 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2274 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
2275 if (num_copies
== 1) {
2277 * we only have a single copy of the data, so don't bother with
2278 * all the retry and error correction code that follows. no
2279 * matter what the error is, it is very likely to persist.
2281 btrfs_debug(fs_info
,
2282 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2283 num_copies
, failrec
->this_mirror
, failed_mirror
);
2288 * there are two premises:
2289 * a) deliver good data to the caller
2290 * b) correct the bad sectors on disk
2292 if (failed_bio_pages
> 1) {
2294 * to fulfill b), we need to know the exact failing sectors, as
2295 * we don't want to rewrite any more than the failed ones. thus,
2296 * we need separate read requests for the failed bio
2298 * if the following BUG_ON triggers, our validation request got
2299 * merged. we need separate requests for our algorithm to work.
2301 BUG_ON(failrec
->in_validation
);
2302 failrec
->in_validation
= 1;
2303 failrec
->this_mirror
= failed_mirror
;
2306 * we're ready to fulfill a) and b) alongside. get a good copy
2307 * of the failed sector and if we succeed, we have setup
2308 * everything for repair_io_failure to do the rest for us.
2310 if (failrec
->in_validation
) {
2311 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2312 failrec
->in_validation
= 0;
2313 failrec
->this_mirror
= 0;
2315 failrec
->failed_mirror
= failed_mirror
;
2316 failrec
->this_mirror
++;
2317 if (failrec
->this_mirror
== failed_mirror
)
2318 failrec
->this_mirror
++;
2321 if (failrec
->this_mirror
> num_copies
) {
2322 btrfs_debug(fs_info
,
2323 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2324 num_copies
, failrec
->this_mirror
, failed_mirror
);
2332 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2333 struct io_failure_record
*failrec
,
2334 struct page
*page
, int pg_offset
, int icsum
,
2335 bio_end_io_t
*endio_func
, void *data
)
2337 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2339 struct btrfs_io_bio
*btrfs_failed_bio
;
2340 struct btrfs_io_bio
*btrfs_bio
;
2342 bio
= btrfs_io_bio_alloc(1);
2343 bio
->bi_end_io
= endio_func
;
2344 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2345 bio_set_dev(bio
, fs_info
->fs_devices
->latest_bdev
);
2346 bio
->bi_iter
.bi_size
= 0;
2347 bio
->bi_private
= data
;
2349 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2350 if (btrfs_failed_bio
->csum
) {
2351 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2353 btrfs_bio
= btrfs_io_bio(bio
);
2354 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2356 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2360 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2366 * This is a generic handler for readpage errors. If other copies exist, read
2367 * those and write back good data to the failed position. Does not investigate
2368 * in remapping the failed extent elsewhere, hoping the device will be smart
2369 * enough to do this as needed
2371 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2372 struct page
*page
, u64 start
, u64 end
,
2375 struct io_failure_record
*failrec
;
2376 struct inode
*inode
= page
->mapping
->host
;
2377 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2378 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2381 blk_status_t status
;
2383 unsigned failed_bio_pages
= failed_bio
->bi_iter
.bi_size
>> PAGE_SHIFT
;
2385 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
2387 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2391 if (!btrfs_check_repairable(inode
, failed_bio_pages
, failrec
,
2393 free_io_failure(failure_tree
, tree
, failrec
);
2397 if (failed_bio_pages
> 1)
2398 read_mode
|= REQ_FAILFAST_DEV
;
2400 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2401 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2402 start
- page_offset(page
),
2403 (int)phy_offset
, failed_bio
->bi_end_io
,
2405 bio
->bi_opf
= REQ_OP_READ
| read_mode
;
2407 btrfs_debug(btrfs_sb(inode
->i_sb
),
2408 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2409 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2411 status
= tree
->ops
->submit_bio_hook(tree
->private_data
, bio
, failrec
->this_mirror
,
2412 failrec
->bio_flags
, 0);
2414 free_io_failure(failure_tree
, tree
, failrec
);
2416 ret
= blk_status_to_errno(status
);
2422 /* lots and lots of room for performance fixes in the end_bio funcs */
2424 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2426 int uptodate
= (err
== 0);
2429 btrfs_writepage_endio_finish_ordered(page
, start
, end
, uptodate
);
2432 ClearPageUptodate(page
);
2434 ret
= err
< 0 ? err
: -EIO
;
2435 mapping_set_error(page
->mapping
, ret
);
2440 * after a writepage IO is done, we need to:
2441 * clear the uptodate bits on error
2442 * clear the writeback bits in the extent tree for this IO
2443 * end_page_writeback if the page has no more pending IO
2445 * Scheduling is not allowed, so the extent state tree is expected
2446 * to have one and only one object corresponding to this IO.
2448 static void end_bio_extent_writepage(struct bio
*bio
)
2450 int error
= blk_status_to_errno(bio
->bi_status
);
2451 struct bio_vec
*bvec
;
2455 struct bvec_iter_all iter_all
;
2457 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
2458 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
2459 struct page
*page
= bvec
->bv_page
;
2460 struct inode
*inode
= page
->mapping
->host
;
2461 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2463 /* We always issue full-page reads, but if some block
2464 * in a page fails to read, blk_update_request() will
2465 * advance bv_offset and adjust bv_len to compensate.
2466 * Print a warning for nonzero offsets, and an error
2467 * if they don't add up to a full page. */
2468 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2469 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2471 "partial page write in btrfs with offset %u and length %u",
2472 bvec
->bv_offset
, bvec
->bv_len
);
2475 "incomplete page write in btrfs with offset %u and length %u",
2476 bvec
->bv_offset
, bvec
->bv_len
);
2479 start
= page_offset(page
);
2480 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2482 end_extent_writepage(page
, error
, start
, end
);
2483 end_page_writeback(page
);
2490 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2493 struct extent_state
*cached
= NULL
;
2494 u64 end
= start
+ len
- 1;
2496 if (uptodate
&& tree
->track_uptodate
)
2497 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2498 unlock_extent_cached_atomic(tree
, start
, end
, &cached
);
2502 * after a readpage IO is done, we need to:
2503 * clear the uptodate bits on error
2504 * set the uptodate bits if things worked
2505 * set the page up to date if all extents in the tree are uptodate
2506 * clear the lock bit in the extent tree
2507 * unlock the page if there are no other extents locked for it
2509 * Scheduling is not allowed, so the extent state tree is expected
2510 * to have one and only one object corresponding to this IO.
2512 static void end_bio_extent_readpage(struct bio
*bio
)
2514 struct bio_vec
*bvec
;
2515 int uptodate
= !bio
->bi_status
;
2516 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2517 struct extent_io_tree
*tree
, *failure_tree
;
2522 u64 extent_start
= 0;
2527 struct bvec_iter_all iter_all
;
2529 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
2530 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
2531 struct page
*page
= bvec
->bv_page
;
2532 struct inode
*inode
= page
->mapping
->host
;
2533 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2534 bool data_inode
= btrfs_ino(BTRFS_I(inode
))
2535 != BTRFS_BTREE_INODE_OBJECTID
;
2537 btrfs_debug(fs_info
,
2538 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2539 (u64
)bio
->bi_iter
.bi_sector
, bio
->bi_status
,
2540 io_bio
->mirror_num
);
2541 tree
= &BTRFS_I(inode
)->io_tree
;
2542 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2544 /* We always issue full-page reads, but if some block
2545 * in a page fails to read, blk_update_request() will
2546 * advance bv_offset and adjust bv_len to compensate.
2547 * Print a warning for nonzero offsets, and an error
2548 * if they don't add up to a full page. */
2549 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2550 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2552 "partial page read in btrfs with offset %u and length %u",
2553 bvec
->bv_offset
, bvec
->bv_len
);
2556 "incomplete page read in btrfs with offset %u and length %u",
2557 bvec
->bv_offset
, bvec
->bv_len
);
2560 start
= page_offset(page
);
2561 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2564 mirror
= io_bio
->mirror_num
;
2565 if (likely(uptodate
)) {
2566 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2572 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
,
2573 failure_tree
, tree
, start
,
2575 btrfs_ino(BTRFS_I(inode
)), 0);
2578 if (likely(uptodate
))
2584 * The generic bio_readpage_error handles errors the
2585 * following way: If possible, new read requests are
2586 * created and submitted and will end up in
2587 * end_bio_extent_readpage as well (if we're lucky,
2588 * not in the !uptodate case). In that case it returns
2589 * 0 and we just go on with the next page in our bio.
2590 * If it can't handle the error it will return -EIO and
2591 * we remain responsible for that page.
2593 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2596 uptodate
= !bio
->bi_status
;
2601 struct extent_buffer
*eb
;
2603 eb
= (struct extent_buffer
*)page
->private;
2604 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
2605 eb
->read_mirror
= mirror
;
2606 atomic_dec(&eb
->io_pages
);
2607 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
,
2609 btree_readahead_hook(eb
, -EIO
);
2614 if (likely(uptodate
)) {
2615 loff_t i_size
= i_size_read(inode
);
2616 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2619 /* Zero out the end if this page straddles i_size */
2620 off
= offset_in_page(i_size
);
2621 if (page
->index
== end_index
&& off
)
2622 zero_user_segment(page
, off
, PAGE_SIZE
);
2623 SetPageUptodate(page
);
2625 ClearPageUptodate(page
);
2631 if (unlikely(!uptodate
)) {
2633 endio_readpage_release_extent(tree
,
2639 endio_readpage_release_extent(tree
, start
,
2640 end
- start
+ 1, 0);
2641 } else if (!extent_len
) {
2642 extent_start
= start
;
2643 extent_len
= end
+ 1 - start
;
2644 } else if (extent_start
+ extent_len
== start
) {
2645 extent_len
+= end
+ 1 - start
;
2647 endio_readpage_release_extent(tree
, extent_start
,
2648 extent_len
, uptodate
);
2649 extent_start
= start
;
2650 extent_len
= end
+ 1 - start
;
2655 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2657 btrfs_io_bio_free_csum(io_bio
);
2662 * Initialize the members up to but not including 'bio'. Use after allocating a
2663 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2664 * 'bio' because use of __GFP_ZERO is not supported.
2666 static inline void btrfs_io_bio_init(struct btrfs_io_bio
*btrfs_bio
)
2668 memset(btrfs_bio
, 0, offsetof(struct btrfs_io_bio
, bio
));
2672 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2673 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2674 * for the appropriate container_of magic
2676 struct bio
*btrfs_bio_alloc(struct block_device
*bdev
, u64 first_byte
)
2680 bio
= bio_alloc_bioset(GFP_NOFS
, BIO_MAX_PAGES
, &btrfs_bioset
);
2681 bio_set_dev(bio
, bdev
);
2682 bio
->bi_iter
.bi_sector
= first_byte
>> 9;
2683 btrfs_io_bio_init(btrfs_io_bio(bio
));
2687 struct bio
*btrfs_bio_clone(struct bio
*bio
)
2689 struct btrfs_io_bio
*btrfs_bio
;
2692 /* Bio allocation backed by a bioset does not fail */
2693 new = bio_clone_fast(bio
, GFP_NOFS
, &btrfs_bioset
);
2694 btrfs_bio
= btrfs_io_bio(new);
2695 btrfs_io_bio_init(btrfs_bio
);
2696 btrfs_bio
->iter
= bio
->bi_iter
;
2700 struct bio
*btrfs_io_bio_alloc(unsigned int nr_iovecs
)
2704 /* Bio allocation backed by a bioset does not fail */
2705 bio
= bio_alloc_bioset(GFP_NOFS
, nr_iovecs
, &btrfs_bioset
);
2706 btrfs_io_bio_init(btrfs_io_bio(bio
));
2710 struct bio
*btrfs_bio_clone_partial(struct bio
*orig
, int offset
, int size
)
2713 struct btrfs_io_bio
*btrfs_bio
;
2715 /* this will never fail when it's backed by a bioset */
2716 bio
= bio_clone_fast(orig
, GFP_NOFS
, &btrfs_bioset
);
2719 btrfs_bio
= btrfs_io_bio(bio
);
2720 btrfs_io_bio_init(btrfs_bio
);
2722 bio_trim(bio
, offset
>> 9, size
>> 9);
2723 btrfs_bio
->iter
= bio
->bi_iter
;
2728 * @opf: bio REQ_OP_* and REQ_* flags as one value
2729 * @tree: tree so we can call our merge_bio hook
2730 * @wbc: optional writeback control for io accounting
2731 * @page: page to add to the bio
2732 * @pg_offset: offset of the new bio or to check whether we are adding
2733 * a contiguous page to the previous one
2734 * @size: portion of page that we want to write
2735 * @offset: starting offset in the page
2736 * @bdev: attach newly created bios to this bdev
2737 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2738 * @end_io_func: end_io callback for new bio
2739 * @mirror_num: desired mirror to read/write
2740 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2741 * @bio_flags: flags of the current bio to see if we can merge them
2743 static int submit_extent_page(unsigned int opf
, struct extent_io_tree
*tree
,
2744 struct writeback_control
*wbc
,
2745 struct page
*page
, u64 offset
,
2746 size_t size
, unsigned long pg_offset
,
2747 struct block_device
*bdev
,
2748 struct bio
**bio_ret
,
2749 bio_end_io_t end_io_func
,
2751 unsigned long prev_bio_flags
,
2752 unsigned long bio_flags
,
2753 bool force_bio_submit
)
2757 size_t page_size
= min_t(size_t, size
, PAGE_SIZE
);
2758 sector_t sector
= offset
>> 9;
2764 bool can_merge
= true;
2767 if (prev_bio_flags
& EXTENT_BIO_COMPRESSED
)
2768 contig
= bio
->bi_iter
.bi_sector
== sector
;
2770 contig
= bio_end_sector(bio
) == sector
;
2773 if (btrfs_bio_fits_in_stripe(page
, page_size
, bio
, bio_flags
))
2776 if (prev_bio_flags
!= bio_flags
|| !contig
|| !can_merge
||
2778 bio_add_page(bio
, page
, page_size
, pg_offset
) < page_size
) {
2779 ret
= submit_one_bio(bio
, mirror_num
, prev_bio_flags
);
2787 wbc_account_io(wbc
, page
, page_size
);
2792 bio
= btrfs_bio_alloc(bdev
, offset
);
2793 bio_add_page(bio
, page
, page_size
, pg_offset
);
2794 bio
->bi_end_io
= end_io_func
;
2795 bio
->bi_private
= tree
;
2796 bio
->bi_write_hint
= page
->mapping
->host
->i_write_hint
;
2799 wbc_init_bio(wbc
, bio
);
2800 wbc_account_io(wbc
, page
, page_size
);
2808 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2811 if (!PagePrivate(page
)) {
2812 SetPagePrivate(page
);
2814 set_page_private(page
, (unsigned long)eb
);
2816 WARN_ON(page
->private != (unsigned long)eb
);
2820 void set_page_extent_mapped(struct page
*page
)
2822 if (!PagePrivate(page
)) {
2823 SetPagePrivate(page
);
2825 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2829 static struct extent_map
*
2830 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2831 u64 start
, u64 len
, get_extent_t
*get_extent
,
2832 struct extent_map
**em_cached
)
2834 struct extent_map
*em
;
2836 if (em_cached
&& *em_cached
) {
2838 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2839 start
< extent_map_end(em
)) {
2840 refcount_inc(&em
->refs
);
2844 free_extent_map(em
);
2848 em
= get_extent(BTRFS_I(inode
), page
, pg_offset
, start
, len
, 0);
2849 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2851 refcount_inc(&em
->refs
);
2857 * basic readpage implementation. Locked extent state structs are inserted
2858 * into the tree that are removed when the IO is done (by the end_io
2860 * XXX JDM: This needs looking at to ensure proper page locking
2861 * return 0 on success, otherwise return error
2863 static int __do_readpage(struct extent_io_tree
*tree
,
2865 get_extent_t
*get_extent
,
2866 struct extent_map
**em_cached
,
2867 struct bio
**bio
, int mirror_num
,
2868 unsigned long *bio_flags
, unsigned int read_flags
,
2871 struct inode
*inode
= page
->mapping
->host
;
2872 u64 start
= page_offset(page
);
2873 const u64 end
= start
+ PAGE_SIZE
- 1;
2876 u64 last_byte
= i_size_read(inode
);
2879 struct extent_map
*em
;
2880 struct block_device
*bdev
;
2883 size_t pg_offset
= 0;
2885 size_t disk_io_size
;
2886 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2887 unsigned long this_bio_flag
= 0;
2889 set_page_extent_mapped(page
);
2891 if (!PageUptodate(page
)) {
2892 if (cleancache_get_page(page
) == 0) {
2893 BUG_ON(blocksize
!= PAGE_SIZE
);
2894 unlock_extent(tree
, start
, end
);
2899 if (page
->index
== last_byte
>> PAGE_SHIFT
) {
2901 size_t zero_offset
= offset_in_page(last_byte
);
2904 iosize
= PAGE_SIZE
- zero_offset
;
2905 userpage
= kmap_atomic(page
);
2906 memset(userpage
+ zero_offset
, 0, iosize
);
2907 flush_dcache_page(page
);
2908 kunmap_atomic(userpage
);
2911 while (cur
<= end
) {
2912 bool force_bio_submit
= false;
2915 if (cur
>= last_byte
) {
2917 struct extent_state
*cached
= NULL
;
2919 iosize
= PAGE_SIZE
- pg_offset
;
2920 userpage
= kmap_atomic(page
);
2921 memset(userpage
+ pg_offset
, 0, iosize
);
2922 flush_dcache_page(page
);
2923 kunmap_atomic(userpage
);
2924 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2926 unlock_extent_cached(tree
, cur
,
2927 cur
+ iosize
- 1, &cached
);
2930 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2931 end
- cur
+ 1, get_extent
, em_cached
);
2932 if (IS_ERR_OR_NULL(em
)) {
2934 unlock_extent(tree
, cur
, end
);
2937 extent_offset
= cur
- em
->start
;
2938 BUG_ON(extent_map_end(em
) <= cur
);
2941 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2942 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2943 extent_set_compress_type(&this_bio_flag
,
2947 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2948 cur_end
= min(extent_map_end(em
) - 1, end
);
2949 iosize
= ALIGN(iosize
, blocksize
);
2950 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2951 disk_io_size
= em
->block_len
;
2952 offset
= em
->block_start
;
2954 offset
= em
->block_start
+ extent_offset
;
2955 disk_io_size
= iosize
;
2958 block_start
= em
->block_start
;
2959 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2960 block_start
= EXTENT_MAP_HOLE
;
2963 * If we have a file range that points to a compressed extent
2964 * and it's followed by a consecutive file range that points to
2965 * to the same compressed extent (possibly with a different
2966 * offset and/or length, so it either points to the whole extent
2967 * or only part of it), we must make sure we do not submit a
2968 * single bio to populate the pages for the 2 ranges because
2969 * this makes the compressed extent read zero out the pages
2970 * belonging to the 2nd range. Imagine the following scenario:
2973 * [0 - 8K] [8K - 24K]
2976 * points to extent X, points to extent X,
2977 * offset 4K, length of 8K offset 0, length 16K
2979 * [extent X, compressed length = 4K uncompressed length = 16K]
2981 * If the bio to read the compressed extent covers both ranges,
2982 * it will decompress extent X into the pages belonging to the
2983 * first range and then it will stop, zeroing out the remaining
2984 * pages that belong to the other range that points to extent X.
2985 * So here we make sure we submit 2 bios, one for the first
2986 * range and another one for the third range. Both will target
2987 * the same physical extent from disk, but we can't currently
2988 * make the compressed bio endio callback populate the pages
2989 * for both ranges because each compressed bio is tightly
2990 * coupled with a single extent map, and each range can have
2991 * an extent map with a different offset value relative to the
2992 * uncompressed data of our extent and different lengths. This
2993 * is a corner case so we prioritize correctness over
2994 * non-optimal behavior (submitting 2 bios for the same extent).
2996 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
2997 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
2998 *prev_em_start
!= em
->start
)
2999 force_bio_submit
= true;
3002 *prev_em_start
= em
->start
;
3004 free_extent_map(em
);
3007 /* we've found a hole, just zero and go on */
3008 if (block_start
== EXTENT_MAP_HOLE
) {
3010 struct extent_state
*cached
= NULL
;
3012 userpage
= kmap_atomic(page
);
3013 memset(userpage
+ pg_offset
, 0, iosize
);
3014 flush_dcache_page(page
);
3015 kunmap_atomic(userpage
);
3017 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3019 unlock_extent_cached(tree
, cur
,
3020 cur
+ iosize
- 1, &cached
);
3022 pg_offset
+= iosize
;
3025 /* the get_extent function already copied into the page */
3026 if (test_range_bit(tree
, cur
, cur_end
,
3027 EXTENT_UPTODATE
, 1, NULL
)) {
3028 check_page_uptodate(tree
, page
);
3029 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3031 pg_offset
+= iosize
;
3034 /* we have an inline extent but it didn't get marked up
3035 * to date. Error out
3037 if (block_start
== EXTENT_MAP_INLINE
) {
3039 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3041 pg_offset
+= iosize
;
3045 ret
= submit_extent_page(REQ_OP_READ
| read_flags
, tree
, NULL
,
3046 page
, offset
, disk_io_size
,
3047 pg_offset
, bdev
, bio
,
3048 end_bio_extent_readpage
, mirror_num
,
3054 *bio_flags
= this_bio_flag
;
3057 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3061 pg_offset
+= iosize
;
3065 if (!PageError(page
))
3066 SetPageUptodate(page
);
3072 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3073 struct page
*pages
[], int nr_pages
,
3075 struct extent_map
**em_cached
,
3077 unsigned long *bio_flags
,
3080 struct inode
*inode
;
3081 struct btrfs_ordered_extent
*ordered
;
3084 inode
= pages
[0]->mapping
->host
;
3086 lock_extent(tree
, start
, end
);
3087 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
3091 unlock_extent(tree
, start
, end
);
3092 btrfs_start_ordered_extent(inode
, ordered
, 1);
3093 btrfs_put_ordered_extent(ordered
);
3096 for (index
= 0; index
< nr_pages
; index
++) {
3097 __do_readpage(tree
, pages
[index
], btrfs_get_extent
, em_cached
,
3098 bio
, 0, bio_flags
, REQ_RAHEAD
, prev_em_start
);
3099 put_page(pages
[index
]);
3103 static void __extent_readpages(struct extent_io_tree
*tree
,
3104 struct page
*pages
[],
3106 struct extent_map
**em_cached
,
3107 struct bio
**bio
, unsigned long *bio_flags
,
3114 int first_index
= 0;
3116 for (index
= 0; index
< nr_pages
; index
++) {
3117 page_start
= page_offset(pages
[index
]);
3120 end
= start
+ PAGE_SIZE
- 1;
3121 first_index
= index
;
3122 } else if (end
+ 1 == page_start
) {
3125 __do_contiguous_readpages(tree
, &pages
[first_index
],
3126 index
- first_index
, start
,
3131 end
= start
+ PAGE_SIZE
- 1;
3132 first_index
= index
;
3137 __do_contiguous_readpages(tree
, &pages
[first_index
],
3138 index
- first_index
, start
,
3139 end
, em_cached
, bio
,
3140 bio_flags
, prev_em_start
);
3143 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3145 get_extent_t
*get_extent
,
3146 struct bio
**bio
, int mirror_num
,
3147 unsigned long *bio_flags
,
3148 unsigned int read_flags
)
3150 struct inode
*inode
= page
->mapping
->host
;
3151 struct btrfs_ordered_extent
*ordered
;
3152 u64 start
= page_offset(page
);
3153 u64 end
= start
+ PAGE_SIZE
- 1;
3157 lock_extent(tree
, start
, end
);
3158 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
3162 unlock_extent(tree
, start
, end
);
3163 btrfs_start_ordered_extent(inode
, ordered
, 1);
3164 btrfs_put_ordered_extent(ordered
);
3167 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3168 bio_flags
, read_flags
, NULL
);
3172 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3173 get_extent_t
*get_extent
, int mirror_num
)
3175 struct bio
*bio
= NULL
;
3176 unsigned long bio_flags
= 0;
3179 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3182 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
3186 static void update_nr_written(struct writeback_control
*wbc
,
3187 unsigned long nr_written
)
3189 wbc
->nr_to_write
-= nr_written
;
3193 * helper for __extent_writepage, doing all of the delayed allocation setup.
3195 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3196 * to write the page (copy into inline extent). In this case the IO has
3197 * been started and the page is already unlocked.
3199 * This returns 0 if all went well (page still locked)
3200 * This returns < 0 if there were errors (page still locked)
3202 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3203 struct page
*page
, struct writeback_control
*wbc
,
3204 u64 delalloc_start
, unsigned long *nr_written
)
3206 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
3207 u64 page_end
= delalloc_start
+ PAGE_SIZE
- 1;
3209 u64 delalloc_to_write
= 0;
3210 u64 delalloc_end
= 0;
3212 int page_started
= 0;
3215 while (delalloc_end
< page_end
) {
3216 found
= find_lock_delalloc_range(inode
, tree
,
3221 delalloc_start
= delalloc_end
+ 1;
3224 ret
= btrfs_run_delalloc_range(inode
, page
, delalloc_start
,
3225 delalloc_end
, &page_started
, nr_written
, wbc
);
3226 /* File system has been set read-only */
3230 * btrfs_run_delalloc_range should return < 0 for error
3231 * but just in case, we use > 0 here meaning the IO is
3232 * started, so we don't want to return > 0 unless
3233 * things are going well.
3235 ret
= ret
< 0 ? ret
: -EIO
;
3239 * delalloc_end is already one less than the total length, so
3240 * we don't subtract one from PAGE_SIZE
3242 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3243 PAGE_SIZE
) >> PAGE_SHIFT
;
3244 delalloc_start
= delalloc_end
+ 1;
3246 if (wbc
->nr_to_write
< delalloc_to_write
) {
3249 if (delalloc_to_write
< thresh
* 2)
3250 thresh
= delalloc_to_write
;
3251 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3255 /* did the fill delalloc function already unlock and start
3260 * we've unlocked the page, so we can't update
3261 * the mapping's writeback index, just update
3264 wbc
->nr_to_write
-= *nr_written
;
3275 * helper for __extent_writepage. This calls the writepage start hooks,
3276 * and does the loop to map the page into extents and bios.
3278 * We return 1 if the IO is started and the page is unlocked,
3279 * 0 if all went well (page still locked)
3280 * < 0 if there were errors (page still locked)
3282 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3284 struct writeback_control
*wbc
,
3285 struct extent_page_data
*epd
,
3287 unsigned long nr_written
,
3288 unsigned int write_flags
, int *nr_ret
)
3290 struct extent_io_tree
*tree
= epd
->tree
;
3291 u64 start
= page_offset(page
);
3292 u64 page_end
= start
+ PAGE_SIZE
- 1;
3298 struct extent_map
*em
;
3299 struct block_device
*bdev
;
3300 size_t pg_offset
= 0;
3306 ret
= btrfs_writepage_cow_fixup(page
, start
, page_end
);
3308 /* Fixup worker will requeue */
3310 wbc
->pages_skipped
++;
3312 redirty_page_for_writepage(wbc
, page
);
3314 update_nr_written(wbc
, nr_written
);
3320 * we don't want to touch the inode after unlocking the page,
3321 * so we update the mapping writeback index now
3323 update_nr_written(wbc
, nr_written
+ 1);
3326 if (i_size
<= start
) {
3327 btrfs_writepage_endio_finish_ordered(page
, start
, page_end
, 1);
3331 blocksize
= inode
->i_sb
->s_blocksize
;
3333 while (cur
<= end
) {
3337 if (cur
>= i_size
) {
3338 btrfs_writepage_endio_finish_ordered(page
, cur
,
3342 em
= btrfs_get_extent(BTRFS_I(inode
), page
, pg_offset
, cur
,
3344 if (IS_ERR_OR_NULL(em
)) {
3346 ret
= PTR_ERR_OR_ZERO(em
);
3350 extent_offset
= cur
- em
->start
;
3351 em_end
= extent_map_end(em
);
3352 BUG_ON(em_end
<= cur
);
3354 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3355 iosize
= ALIGN(iosize
, blocksize
);
3356 offset
= em
->block_start
+ extent_offset
;
3358 block_start
= em
->block_start
;
3359 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3360 free_extent_map(em
);
3364 * compressed and inline extents are written through other
3367 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3368 block_start
== EXTENT_MAP_INLINE
) {
3370 * end_io notification does not happen here for
3371 * compressed extents
3374 btrfs_writepage_endio_finish_ordered(page
, cur
,
3377 else if (compressed
) {
3378 /* we don't want to end_page_writeback on
3379 * a compressed extent. this happens
3386 pg_offset
+= iosize
;
3390 btrfs_set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3391 if (!PageWriteback(page
)) {
3392 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3393 "page %lu not writeback, cur %llu end %llu",
3394 page
->index
, cur
, end
);
3397 ret
= submit_extent_page(REQ_OP_WRITE
| write_flags
, tree
, wbc
,
3398 page
, offset
, iosize
, pg_offset
,
3400 end_bio_extent_writepage
,
3404 if (PageWriteback(page
))
3405 end_page_writeback(page
);
3409 pg_offset
+= iosize
;
3418 * the writepage semantics are similar to regular writepage. extent
3419 * records are inserted to lock ranges in the tree, and as dirty areas
3420 * are found, they are marked writeback. Then the lock bits are removed
3421 * and the end_io handler clears the writeback ranges
3423 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3424 struct extent_page_data
*epd
)
3426 struct inode
*inode
= page
->mapping
->host
;
3427 u64 start
= page_offset(page
);
3428 u64 page_end
= start
+ PAGE_SIZE
- 1;
3431 size_t pg_offset
= 0;
3432 loff_t i_size
= i_size_read(inode
);
3433 unsigned long end_index
= i_size
>> PAGE_SHIFT
;
3434 unsigned int write_flags
= 0;
3435 unsigned long nr_written
= 0;
3437 write_flags
= wbc_to_write_flags(wbc
);
3439 trace___extent_writepage(page
, inode
, wbc
);
3441 WARN_ON(!PageLocked(page
));
3443 ClearPageError(page
);
3445 pg_offset
= offset_in_page(i_size
);
3446 if (page
->index
> end_index
||
3447 (page
->index
== end_index
&& !pg_offset
)) {
3448 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3453 if (page
->index
== end_index
) {
3456 userpage
= kmap_atomic(page
);
3457 memset(userpage
+ pg_offset
, 0,
3458 PAGE_SIZE
- pg_offset
);
3459 kunmap_atomic(userpage
);
3460 flush_dcache_page(page
);
3465 set_page_extent_mapped(page
);
3467 if (!epd
->extent_locked
) {
3468 ret
= writepage_delalloc(inode
, page
, wbc
, start
, &nr_written
);
3475 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3476 i_size
, nr_written
, write_flags
, &nr
);
3482 /* make sure the mapping tag for page dirty gets cleared */
3483 set_page_writeback(page
);
3484 end_page_writeback(page
);
3486 if (PageError(page
)) {
3487 ret
= ret
< 0 ? ret
: -EIO
;
3488 end_extent_writepage(page
, ret
, start
, page_end
);
3497 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3499 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3500 TASK_UNINTERRUPTIBLE
);
3503 static noinline_for_stack
int
3504 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3505 struct btrfs_fs_info
*fs_info
,
3506 struct extent_page_data
*epd
)
3512 if (!btrfs_try_tree_write_lock(eb
)) {
3514 flush_write_bio(epd
);
3515 btrfs_tree_lock(eb
);
3518 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3519 btrfs_tree_unlock(eb
);
3523 flush_write_bio(epd
);
3527 wait_on_extent_buffer_writeback(eb
);
3528 btrfs_tree_lock(eb
);
3529 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3531 btrfs_tree_unlock(eb
);
3536 * We need to do this to prevent races in people who check if the eb is
3537 * under IO since we can end up having no IO bits set for a short period
3540 spin_lock(&eb
->refs_lock
);
3541 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3542 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3543 spin_unlock(&eb
->refs_lock
);
3544 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3545 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
3547 fs_info
->dirty_metadata_batch
);
3550 spin_unlock(&eb
->refs_lock
);
3553 btrfs_tree_unlock(eb
);
3558 num_pages
= num_extent_pages(eb
);
3559 for (i
= 0; i
< num_pages
; i
++) {
3560 struct page
*p
= eb
->pages
[i
];
3562 if (!trylock_page(p
)) {
3564 flush_write_bio(epd
);
3574 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3576 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3577 smp_mb__after_atomic();
3578 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3581 static void set_btree_ioerr(struct page
*page
)
3583 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3586 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3590 * If writeback for a btree extent that doesn't belong to a log tree
3591 * failed, increment the counter transaction->eb_write_errors.
3592 * We do this because while the transaction is running and before it's
3593 * committing (when we call filemap_fdata[write|wait]_range against
3594 * the btree inode), we might have
3595 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3596 * returns an error or an error happens during writeback, when we're
3597 * committing the transaction we wouldn't know about it, since the pages
3598 * can be no longer dirty nor marked anymore for writeback (if a
3599 * subsequent modification to the extent buffer didn't happen before the
3600 * transaction commit), which makes filemap_fdata[write|wait]_range not
3601 * able to find the pages tagged with SetPageError at transaction
3602 * commit time. So if this happens we must abort the transaction,
3603 * otherwise we commit a super block with btree roots that point to
3604 * btree nodes/leafs whose content on disk is invalid - either garbage
3605 * or the content of some node/leaf from a past generation that got
3606 * cowed or deleted and is no longer valid.
3608 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3609 * not be enough - we need to distinguish between log tree extents vs
3610 * non-log tree extents, and the next filemap_fdatawait_range() call
3611 * will catch and clear such errors in the mapping - and that call might
3612 * be from a log sync and not from a transaction commit. Also, checking
3613 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3614 * not done and would not be reliable - the eb might have been released
3615 * from memory and reading it back again means that flag would not be
3616 * set (since it's a runtime flag, not persisted on disk).
3618 * Using the flags below in the btree inode also makes us achieve the
3619 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3620 * writeback for all dirty pages and before filemap_fdatawait_range()
3621 * is called, the writeback for all dirty pages had already finished
3622 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3623 * filemap_fdatawait_range() would return success, as it could not know
3624 * that writeback errors happened (the pages were no longer tagged for
3627 switch (eb
->log_index
) {
3629 set_bit(BTRFS_FS_BTREE_ERR
, &eb
->fs_info
->flags
);
3632 set_bit(BTRFS_FS_LOG1_ERR
, &eb
->fs_info
->flags
);
3635 set_bit(BTRFS_FS_LOG2_ERR
, &eb
->fs_info
->flags
);
3638 BUG(); /* unexpected, logic error */
3642 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3644 struct bio_vec
*bvec
;
3645 struct extent_buffer
*eb
;
3647 struct bvec_iter_all iter_all
;
3649 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
3650 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
3651 struct page
*page
= bvec
->bv_page
;
3653 eb
= (struct extent_buffer
*)page
->private;
3655 done
= atomic_dec_and_test(&eb
->io_pages
);
3657 if (bio
->bi_status
||
3658 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3659 ClearPageUptodate(page
);
3660 set_btree_ioerr(page
);
3663 end_page_writeback(page
);
3668 end_extent_buffer_writeback(eb
);
3674 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3675 struct btrfs_fs_info
*fs_info
,
3676 struct writeback_control
*wbc
,
3677 struct extent_page_data
*epd
)
3679 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3680 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3681 u64 offset
= eb
->start
;
3684 unsigned long start
, end
;
3685 unsigned int write_flags
= wbc_to_write_flags(wbc
) | REQ_META
;
3688 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3689 num_pages
= num_extent_pages(eb
);
3690 atomic_set(&eb
->io_pages
, num_pages
);
3692 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3693 nritems
= btrfs_header_nritems(eb
);
3694 if (btrfs_header_level(eb
) > 0) {
3695 end
= btrfs_node_key_ptr_offset(nritems
);
3697 memzero_extent_buffer(eb
, end
, eb
->len
- end
);
3701 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3703 start
= btrfs_item_nr_offset(nritems
);
3704 end
= BTRFS_LEAF_DATA_OFFSET
+ leaf_data_end(fs_info
, eb
);
3705 memzero_extent_buffer(eb
, start
, end
- start
);
3708 for (i
= 0; i
< num_pages
; i
++) {
3709 struct page
*p
= eb
->pages
[i
];
3711 clear_page_dirty_for_io(p
);
3712 set_page_writeback(p
);
3713 ret
= submit_extent_page(REQ_OP_WRITE
| write_flags
, tree
, wbc
,
3714 p
, offset
, PAGE_SIZE
, 0, bdev
,
3716 end_bio_extent_buffer_writepage
,
3720 if (PageWriteback(p
))
3721 end_page_writeback(p
);
3722 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3723 end_extent_buffer_writeback(eb
);
3727 offset
+= PAGE_SIZE
;
3728 update_nr_written(wbc
, 1);
3732 if (unlikely(ret
)) {
3733 for (; i
< num_pages
; i
++) {
3734 struct page
*p
= eb
->pages
[i
];
3735 clear_page_dirty_for_io(p
);
3743 int btree_write_cache_pages(struct address_space
*mapping
,
3744 struct writeback_control
*wbc
)
3746 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3747 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3748 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3749 struct extent_page_data epd
= {
3753 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3757 int nr_to_write_done
= 0;
3758 struct pagevec pvec
;
3761 pgoff_t end
; /* Inclusive */
3765 pagevec_init(&pvec
);
3766 if (wbc
->range_cyclic
) {
3767 index
= mapping
->writeback_index
; /* Start from prev offset */
3770 index
= wbc
->range_start
>> PAGE_SHIFT
;
3771 end
= wbc
->range_end
>> PAGE_SHIFT
;
3774 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3775 tag
= PAGECACHE_TAG_TOWRITE
;
3777 tag
= PAGECACHE_TAG_DIRTY
;
3779 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3780 tag_pages_for_writeback(mapping
, index
, end
);
3781 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3782 (nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
, end
,
3787 for (i
= 0; i
< nr_pages
; i
++) {
3788 struct page
*page
= pvec
.pages
[i
];
3790 if (!PagePrivate(page
))
3793 spin_lock(&mapping
->private_lock
);
3794 if (!PagePrivate(page
)) {
3795 spin_unlock(&mapping
->private_lock
);
3799 eb
= (struct extent_buffer
*)page
->private;
3802 * Shouldn't happen and normally this would be a BUG_ON
3803 * but no sense in crashing the users box for something
3804 * we can survive anyway.
3807 spin_unlock(&mapping
->private_lock
);
3811 if (eb
== prev_eb
) {
3812 spin_unlock(&mapping
->private_lock
);
3816 ret
= atomic_inc_not_zero(&eb
->refs
);
3817 spin_unlock(&mapping
->private_lock
);
3822 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3824 free_extent_buffer(eb
);
3828 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3831 free_extent_buffer(eb
);
3834 free_extent_buffer(eb
);
3837 * the filesystem may choose to bump up nr_to_write.
3838 * We have to make sure to honor the new nr_to_write
3841 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3843 pagevec_release(&pvec
);
3846 if (!scanned
&& !done
) {
3848 * We hit the last page and there is more work to be done: wrap
3849 * back to the start of the file
3855 flush_write_bio(&epd
);
3860 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3861 * @mapping: address space structure to write
3862 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3863 * @data: data passed to __extent_writepage function
3865 * If a page is already under I/O, write_cache_pages() skips it, even
3866 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3867 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3868 * and msync() need to guarantee that all the data which was dirty at the time
3869 * the call was made get new I/O started against them. If wbc->sync_mode is
3870 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3871 * existing IO to complete.
3873 static int extent_write_cache_pages(struct address_space
*mapping
,
3874 struct writeback_control
*wbc
,
3875 struct extent_page_data
*epd
)
3877 struct inode
*inode
= mapping
->host
;
3880 int nr_to_write_done
= 0;
3881 struct pagevec pvec
;
3884 pgoff_t end
; /* Inclusive */
3886 int range_whole
= 0;
3891 * We have to hold onto the inode so that ordered extents can do their
3892 * work when the IO finishes. The alternative to this is failing to add
3893 * an ordered extent if the igrab() fails there and that is a huge pain
3894 * to deal with, so instead just hold onto the inode throughout the
3895 * writepages operation. If it fails here we are freeing up the inode
3896 * anyway and we'd rather not waste our time writing out stuff that is
3897 * going to be truncated anyway.
3902 pagevec_init(&pvec
);
3903 if (wbc
->range_cyclic
) {
3904 index
= mapping
->writeback_index
; /* Start from prev offset */
3907 index
= wbc
->range_start
>> PAGE_SHIFT
;
3908 end
= wbc
->range_end
>> PAGE_SHIFT
;
3909 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3915 * We do the tagged writepage as long as the snapshot flush bit is set
3916 * and we are the first one who do the filemap_flush() on this inode.
3918 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
3919 * not race in and drop the bit.
3921 if (range_whole
&& wbc
->nr_to_write
== LONG_MAX
&&
3922 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH
,
3923 &BTRFS_I(inode
)->runtime_flags
))
3924 wbc
->tagged_writepages
= 1;
3926 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
3927 tag
= PAGECACHE_TAG_TOWRITE
;
3929 tag
= PAGECACHE_TAG_DIRTY
;
3931 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
3932 tag_pages_for_writeback(mapping
, index
, end
);
3934 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3935 (nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
,
3936 &index
, end
, tag
))) {
3940 for (i
= 0; i
< nr_pages
; i
++) {
3941 struct page
*page
= pvec
.pages
[i
];
3943 done_index
= page
->index
;
3945 * At this point we hold neither the i_pages lock nor
3946 * the page lock: the page may be truncated or
3947 * invalidated (changing page->mapping to NULL),
3948 * or even swizzled back from swapper_space to
3949 * tmpfs file mapping
3951 if (!trylock_page(page
)) {
3952 flush_write_bio(epd
);
3956 if (unlikely(page
->mapping
!= mapping
)) {
3961 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3962 if (PageWriteback(page
))
3963 flush_write_bio(epd
);
3964 wait_on_page_writeback(page
);
3967 if (PageWriteback(page
) ||
3968 !clear_page_dirty_for_io(page
)) {
3973 ret
= __extent_writepage(page
, wbc
, epd
);
3975 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3981 * done_index is set past this page,
3982 * so media errors will not choke
3983 * background writeout for the entire
3984 * file. This has consequences for
3985 * range_cyclic semantics (ie. it may
3986 * not be suitable for data integrity
3989 done_index
= page
->index
+ 1;
3995 * the filesystem may choose to bump up nr_to_write.
3996 * We have to make sure to honor the new nr_to_write
3999 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4001 pagevec_release(&pvec
);
4004 if (!scanned
&& !done
) {
4006 * We hit the last page and there is more work to be done: wrap
4007 * back to the start of the file
4014 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4015 mapping
->writeback_index
= done_index
;
4017 btrfs_add_delayed_iput(inode
);
4021 int extent_write_full_page(struct page
*page
, struct writeback_control
*wbc
)
4024 struct extent_page_data epd
= {
4026 .tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
,
4028 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4031 ret
= __extent_writepage(page
, wbc
, &epd
);
4033 flush_write_bio(&epd
);
4037 int extent_write_locked_range(struct inode
*inode
, u64 start
, u64 end
,
4041 struct address_space
*mapping
= inode
->i_mapping
;
4042 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
4044 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4047 struct extent_page_data epd
= {
4051 .sync_io
= mode
== WB_SYNC_ALL
,
4053 struct writeback_control wbc_writepages
= {
4055 .nr_to_write
= nr_pages
* 2,
4056 .range_start
= start
,
4057 .range_end
= end
+ 1,
4060 while (start
<= end
) {
4061 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4062 if (clear_page_dirty_for_io(page
))
4063 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4065 btrfs_writepage_endio_finish_ordered(page
, start
,
4066 start
+ PAGE_SIZE
- 1, 1);
4073 flush_write_bio(&epd
);
4077 int extent_writepages(struct address_space
*mapping
,
4078 struct writeback_control
*wbc
)
4081 struct extent_page_data epd
= {
4083 .tree
= &BTRFS_I(mapping
->host
)->io_tree
,
4085 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4088 ret
= extent_write_cache_pages(mapping
, wbc
, &epd
);
4089 flush_write_bio(&epd
);
4093 int extent_readpages(struct address_space
*mapping
, struct list_head
*pages
,
4096 struct bio
*bio
= NULL
;
4097 unsigned long bio_flags
= 0;
4098 struct page
*pagepool
[16];
4099 struct extent_map
*em_cached
= NULL
;
4100 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4102 u64 prev_em_start
= (u64
)-1;
4104 while (!list_empty(pages
)) {
4105 for (nr
= 0; nr
< ARRAY_SIZE(pagepool
) && !list_empty(pages
);) {
4106 struct page
*page
= lru_to_page(pages
);
4108 prefetchw(&page
->flags
);
4109 list_del(&page
->lru
);
4110 if (add_to_page_cache_lru(page
, mapping
, page
->index
,
4111 readahead_gfp_mask(mapping
))) {
4116 pagepool
[nr
++] = page
;
4119 __extent_readpages(tree
, pagepool
, nr
, &em_cached
, &bio
,
4120 &bio_flags
, &prev_em_start
);
4124 free_extent_map(em_cached
);
4127 return submit_one_bio(bio
, 0, bio_flags
);
4132 * basic invalidatepage code, this waits on any locked or writeback
4133 * ranges corresponding to the page, and then deletes any extent state
4134 * records from the tree
4136 int extent_invalidatepage(struct extent_io_tree
*tree
,
4137 struct page
*page
, unsigned long offset
)
4139 struct extent_state
*cached_state
= NULL
;
4140 u64 start
= page_offset(page
);
4141 u64 end
= start
+ PAGE_SIZE
- 1;
4142 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4144 start
+= ALIGN(offset
, blocksize
);
4148 lock_extent_bits(tree
, start
, end
, &cached_state
);
4149 wait_on_page_writeback(page
);
4150 clear_extent_bit(tree
, start
, end
,
4151 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4152 EXTENT_DO_ACCOUNTING
,
4153 1, 1, &cached_state
);
4158 * a helper for releasepage, this tests for areas of the page that
4159 * are locked or under IO and drops the related state bits if it is safe
4162 static int try_release_extent_state(struct extent_io_tree
*tree
,
4163 struct page
*page
, gfp_t mask
)
4165 u64 start
= page_offset(page
);
4166 u64 end
= start
+ PAGE_SIZE
- 1;
4169 if (test_range_bit(tree
, start
, end
,
4170 EXTENT_IOBITS
, 0, NULL
))
4174 * at this point we can safely clear everything except the
4175 * locked bit and the nodatasum bit
4177 ret
= __clear_extent_bit(tree
, start
, end
,
4178 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4179 0, 0, NULL
, mask
, NULL
);
4181 /* if clear_extent_bit failed for enomem reasons,
4182 * we can't allow the release to continue.
4193 * a helper for releasepage. As long as there are no locked extents
4194 * in the range corresponding to the page, both state records and extent
4195 * map records are removed
4197 int try_release_extent_mapping(struct page
*page
, gfp_t mask
)
4199 struct extent_map
*em
;
4200 u64 start
= page_offset(page
);
4201 u64 end
= start
+ PAGE_SIZE
- 1;
4202 struct btrfs_inode
*btrfs_inode
= BTRFS_I(page
->mapping
->host
);
4203 struct extent_io_tree
*tree
= &btrfs_inode
->io_tree
;
4204 struct extent_map_tree
*map
= &btrfs_inode
->extent_tree
;
4206 if (gfpflags_allow_blocking(mask
) &&
4207 page
->mapping
->host
->i_size
> SZ_16M
) {
4209 while (start
<= end
) {
4210 len
= end
- start
+ 1;
4211 write_lock(&map
->lock
);
4212 em
= lookup_extent_mapping(map
, start
, len
);
4214 write_unlock(&map
->lock
);
4217 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4218 em
->start
!= start
) {
4219 write_unlock(&map
->lock
);
4220 free_extent_map(em
);
4223 if (!test_range_bit(tree
, em
->start
,
4224 extent_map_end(em
) - 1,
4225 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4227 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4228 &btrfs_inode
->runtime_flags
);
4229 remove_extent_mapping(map
, em
);
4230 /* once for the rb tree */
4231 free_extent_map(em
);
4233 start
= extent_map_end(em
);
4234 write_unlock(&map
->lock
);
4237 free_extent_map(em
);
4240 return try_release_extent_state(tree
, page
, mask
);
4244 * helper function for fiemap, which doesn't want to see any holes.
4245 * This maps until we find something past 'last'
4247 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4248 u64 offset
, u64 last
)
4250 u64 sectorsize
= btrfs_inode_sectorsize(inode
);
4251 struct extent_map
*em
;
4258 len
= last
- offset
;
4261 len
= ALIGN(len
, sectorsize
);
4262 em
= btrfs_get_extent_fiemap(BTRFS_I(inode
), offset
, len
);
4263 if (IS_ERR_OR_NULL(em
))
4266 /* if this isn't a hole return it */
4267 if (em
->block_start
!= EXTENT_MAP_HOLE
)
4270 /* this is a hole, advance to the next extent */
4271 offset
= extent_map_end(em
);
4272 free_extent_map(em
);
4280 * To cache previous fiemap extent
4282 * Will be used for merging fiemap extent
4284 struct fiemap_cache
{
4293 * Helper to submit fiemap extent.
4295 * Will try to merge current fiemap extent specified by @offset, @phys,
4296 * @len and @flags with cached one.
4297 * And only when we fails to merge, cached one will be submitted as
4300 * Return value is the same as fiemap_fill_next_extent().
4302 static int emit_fiemap_extent(struct fiemap_extent_info
*fieinfo
,
4303 struct fiemap_cache
*cache
,
4304 u64 offset
, u64 phys
, u64 len
, u32 flags
)
4312 * Sanity check, extent_fiemap() should have ensured that new
4313 * fiemap extent won't overlap with cached one.
4316 * NOTE: Physical address can overlap, due to compression
4318 if (cache
->offset
+ cache
->len
> offset
) {
4324 * Only merges fiemap extents if
4325 * 1) Their logical addresses are continuous
4327 * 2) Their physical addresses are continuous
4328 * So truly compressed (physical size smaller than logical size)
4329 * extents won't get merged with each other
4331 * 3) Share same flags except FIEMAP_EXTENT_LAST
4332 * So regular extent won't get merged with prealloc extent
4334 if (cache
->offset
+ cache
->len
== offset
&&
4335 cache
->phys
+ cache
->len
== phys
&&
4336 (cache
->flags
& ~FIEMAP_EXTENT_LAST
) ==
4337 (flags
& ~FIEMAP_EXTENT_LAST
)) {
4339 cache
->flags
|= flags
;
4340 goto try_submit_last
;
4343 /* Not mergeable, need to submit cached one */
4344 ret
= fiemap_fill_next_extent(fieinfo
, cache
->offset
, cache
->phys
,
4345 cache
->len
, cache
->flags
);
4346 cache
->cached
= false;
4350 cache
->cached
= true;
4351 cache
->offset
= offset
;
4354 cache
->flags
= flags
;
4356 if (cache
->flags
& FIEMAP_EXTENT_LAST
) {
4357 ret
= fiemap_fill_next_extent(fieinfo
, cache
->offset
,
4358 cache
->phys
, cache
->len
, cache
->flags
);
4359 cache
->cached
= false;
4365 * Emit last fiemap cache
4367 * The last fiemap cache may still be cached in the following case:
4369 * |<- Fiemap range ->|
4370 * |<------------ First extent ----------->|
4372 * In this case, the first extent range will be cached but not emitted.
4373 * So we must emit it before ending extent_fiemap().
4375 static int emit_last_fiemap_cache(struct btrfs_fs_info
*fs_info
,
4376 struct fiemap_extent_info
*fieinfo
,
4377 struct fiemap_cache
*cache
)
4384 ret
= fiemap_fill_next_extent(fieinfo
, cache
->offset
, cache
->phys
,
4385 cache
->len
, cache
->flags
);
4386 cache
->cached
= false;
4392 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4393 __u64 start
, __u64 len
)
4397 u64 max
= start
+ len
;
4401 u64 last_for_get_extent
= 0;
4403 u64 isize
= i_size_read(inode
);
4404 struct btrfs_key found_key
;
4405 struct extent_map
*em
= NULL
;
4406 struct extent_state
*cached_state
= NULL
;
4407 struct btrfs_path
*path
;
4408 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4409 struct fiemap_cache cache
= { 0 };
4418 path
= btrfs_alloc_path();
4421 path
->leave_spinning
= 1;
4423 start
= round_down(start
, btrfs_inode_sectorsize(inode
));
4424 len
= round_up(max
, btrfs_inode_sectorsize(inode
)) - start
;
4427 * lookup the last file extent. We're not using i_size here
4428 * because there might be preallocation past i_size
4430 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
4431 btrfs_ino(BTRFS_I(inode
)), -1, 0);
4433 btrfs_free_path(path
);
4442 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4443 found_type
= found_key
.type
;
4445 /* No extents, but there might be delalloc bits */
4446 if (found_key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
4447 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4448 /* have to trust i_size as the end */
4450 last_for_get_extent
= isize
;
4453 * remember the start of the last extent. There are a
4454 * bunch of different factors that go into the length of the
4455 * extent, so its much less complex to remember where it started
4457 last
= found_key
.offset
;
4458 last_for_get_extent
= last
+ 1;
4460 btrfs_release_path(path
);
4463 * we might have some extents allocated but more delalloc past those
4464 * extents. so, we trust isize unless the start of the last extent is
4469 last_for_get_extent
= isize
;
4472 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4475 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
);
4484 u64 offset_in_extent
= 0;
4486 /* break if the extent we found is outside the range */
4487 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4491 * get_extent may return an extent that starts before our
4492 * requested range. We have to make sure the ranges
4493 * we return to fiemap always move forward and don't
4494 * overlap, so adjust the offsets here
4496 em_start
= max(em
->start
, off
);
4499 * record the offset from the start of the extent
4500 * for adjusting the disk offset below. Only do this if the
4501 * extent isn't compressed since our in ram offset may be past
4502 * what we have actually allocated on disk.
4504 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4505 offset_in_extent
= em_start
- em
->start
;
4506 em_end
= extent_map_end(em
);
4507 em_len
= em_end
- em_start
;
4509 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
)
4510 disko
= em
->block_start
+ offset_in_extent
;
4515 * bump off for our next call to get_extent
4517 off
= extent_map_end(em
);
4521 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4523 flags
|= FIEMAP_EXTENT_LAST
;
4524 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4525 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4526 FIEMAP_EXTENT_NOT_ALIGNED
);
4527 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4528 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4529 FIEMAP_EXTENT_UNKNOWN
);
4530 } else if (fieinfo
->fi_extents_max
) {
4531 u64 bytenr
= em
->block_start
-
4532 (em
->start
- em
->orig_start
);
4535 * As btrfs supports shared space, this information
4536 * can be exported to userspace tools via
4537 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4538 * then we're just getting a count and we can skip the
4541 ret
= btrfs_check_shared(root
,
4542 btrfs_ino(BTRFS_I(inode
)),
4547 flags
|= FIEMAP_EXTENT_SHARED
;
4550 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4551 flags
|= FIEMAP_EXTENT_ENCODED
;
4552 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4553 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4555 free_extent_map(em
);
4557 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4558 (last
== (u64
)-1 && isize
<= em_end
)) {
4559 flags
|= FIEMAP_EXTENT_LAST
;
4563 /* now scan forward to see if this is really the last extent. */
4564 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
);
4570 flags
|= FIEMAP_EXTENT_LAST
;
4573 ret
= emit_fiemap_extent(fieinfo
, &cache
, em_start
, disko
,
4583 ret
= emit_last_fiemap_cache(root
->fs_info
, fieinfo
, &cache
);
4584 free_extent_map(em
);
4586 btrfs_free_path(path
);
4587 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4592 static void __free_extent_buffer(struct extent_buffer
*eb
)
4594 btrfs_leak_debug_del(&eb
->leak_list
);
4595 kmem_cache_free(extent_buffer_cache
, eb
);
4598 int extent_buffer_under_io(struct extent_buffer
*eb
)
4600 return (atomic_read(&eb
->io_pages
) ||
4601 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4602 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4606 * Release all pages attached to the extent buffer.
4608 static void btrfs_release_extent_buffer_pages(struct extent_buffer
*eb
)
4612 int mapped
= !test_bit(EXTENT_BUFFER_UNMAPPED
, &eb
->bflags
);
4614 BUG_ON(extent_buffer_under_io(eb
));
4616 num_pages
= num_extent_pages(eb
);
4617 for (i
= 0; i
< num_pages
; i
++) {
4618 struct page
*page
= eb
->pages
[i
];
4623 spin_lock(&page
->mapping
->private_lock
);
4625 * We do this since we'll remove the pages after we've
4626 * removed the eb from the radix tree, so we could race
4627 * and have this page now attached to the new eb. So
4628 * only clear page_private if it's still connected to
4631 if (PagePrivate(page
) &&
4632 page
->private == (unsigned long)eb
) {
4633 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4634 BUG_ON(PageDirty(page
));
4635 BUG_ON(PageWriteback(page
));
4637 * We need to make sure we haven't be attached
4640 ClearPagePrivate(page
);
4641 set_page_private(page
, 0);
4642 /* One for the page private */
4647 spin_unlock(&page
->mapping
->private_lock
);
4649 /* One for when we allocated the page */
4655 * Helper for releasing the extent buffer.
4657 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4659 btrfs_release_extent_buffer_pages(eb
);
4660 __free_extent_buffer(eb
);
4663 static struct extent_buffer
*
4664 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4667 struct extent_buffer
*eb
= NULL
;
4669 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4672 eb
->fs_info
= fs_info
;
4674 rwlock_init(&eb
->lock
);
4675 atomic_set(&eb
->write_locks
, 0);
4676 atomic_set(&eb
->read_locks
, 0);
4677 atomic_set(&eb
->blocking_readers
, 0);
4678 atomic_set(&eb
->blocking_writers
, 0);
4679 atomic_set(&eb
->spinning_readers
, 0);
4680 atomic_set(&eb
->spinning_writers
, 0);
4681 eb
->lock_nested
= 0;
4682 init_waitqueue_head(&eb
->write_lock_wq
);
4683 init_waitqueue_head(&eb
->read_lock_wq
);
4685 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4687 spin_lock_init(&eb
->refs_lock
);
4688 atomic_set(&eb
->refs
, 1);
4689 atomic_set(&eb
->io_pages
, 0);
4692 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4694 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4695 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4696 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4701 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4705 struct extent_buffer
*new;
4706 int num_pages
= num_extent_pages(src
);
4708 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4712 for (i
= 0; i
< num_pages
; i
++) {
4713 p
= alloc_page(GFP_NOFS
);
4715 btrfs_release_extent_buffer(new);
4718 attach_extent_buffer_page(new, p
);
4719 WARN_ON(PageDirty(p
));
4722 copy_page(page_address(p
), page_address(src
->pages
[i
]));
4725 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4726 set_bit(EXTENT_BUFFER_UNMAPPED
, &new->bflags
);
4731 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4732 u64 start
, unsigned long len
)
4734 struct extent_buffer
*eb
;
4738 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4742 num_pages
= num_extent_pages(eb
);
4743 for (i
= 0; i
< num_pages
; i
++) {
4744 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4748 set_extent_buffer_uptodate(eb
);
4749 btrfs_set_header_nritems(eb
, 0);
4750 set_bit(EXTENT_BUFFER_UNMAPPED
, &eb
->bflags
);
4755 __free_page(eb
->pages
[i
- 1]);
4756 __free_extent_buffer(eb
);
4760 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4763 return __alloc_dummy_extent_buffer(fs_info
, start
, fs_info
->nodesize
);
4766 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4769 /* the ref bit is tricky. We have to make sure it is set
4770 * if we have the buffer dirty. Otherwise the
4771 * code to free a buffer can end up dropping a dirty
4774 * Once the ref bit is set, it won't go away while the
4775 * buffer is dirty or in writeback, and it also won't
4776 * go away while we have the reference count on the
4779 * We can't just set the ref bit without bumping the
4780 * ref on the eb because free_extent_buffer might
4781 * see the ref bit and try to clear it. If this happens
4782 * free_extent_buffer might end up dropping our original
4783 * ref by mistake and freeing the page before we are able
4784 * to add one more ref.
4786 * So bump the ref count first, then set the bit. If someone
4787 * beat us to it, drop the ref we added.
4789 refs
= atomic_read(&eb
->refs
);
4790 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4793 spin_lock(&eb
->refs_lock
);
4794 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4795 atomic_inc(&eb
->refs
);
4796 spin_unlock(&eb
->refs_lock
);
4799 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4800 struct page
*accessed
)
4804 check_buffer_tree_ref(eb
);
4806 num_pages
= num_extent_pages(eb
);
4807 for (i
= 0; i
< num_pages
; i
++) {
4808 struct page
*p
= eb
->pages
[i
];
4811 mark_page_accessed(p
);
4815 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4818 struct extent_buffer
*eb
;
4821 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4822 start
>> PAGE_SHIFT
);
4823 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4826 * Lock our eb's refs_lock to avoid races with
4827 * free_extent_buffer. When we get our eb it might be flagged
4828 * with EXTENT_BUFFER_STALE and another task running
4829 * free_extent_buffer might have seen that flag set,
4830 * eb->refs == 2, that the buffer isn't under IO (dirty and
4831 * writeback flags not set) and it's still in the tree (flag
4832 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4833 * of decrementing the extent buffer's reference count twice.
4834 * So here we could race and increment the eb's reference count,
4835 * clear its stale flag, mark it as dirty and drop our reference
4836 * before the other task finishes executing free_extent_buffer,
4837 * which would later result in an attempt to free an extent
4838 * buffer that is dirty.
4840 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4841 spin_lock(&eb
->refs_lock
);
4842 spin_unlock(&eb
->refs_lock
);
4844 mark_extent_buffer_accessed(eb
, NULL
);
4852 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4853 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4856 struct extent_buffer
*eb
, *exists
= NULL
;
4859 eb
= find_extent_buffer(fs_info
, start
);
4862 eb
= alloc_dummy_extent_buffer(fs_info
, start
);
4865 eb
->fs_info
= fs_info
;
4867 ret
= radix_tree_preload(GFP_NOFS
);
4870 spin_lock(&fs_info
->buffer_lock
);
4871 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4872 start
>> PAGE_SHIFT
, eb
);
4873 spin_unlock(&fs_info
->buffer_lock
);
4874 radix_tree_preload_end();
4875 if (ret
== -EEXIST
) {
4876 exists
= find_extent_buffer(fs_info
, start
);
4882 check_buffer_tree_ref(eb
);
4883 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4887 btrfs_release_extent_buffer(eb
);
4892 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4895 unsigned long len
= fs_info
->nodesize
;
4898 unsigned long index
= start
>> PAGE_SHIFT
;
4899 struct extent_buffer
*eb
;
4900 struct extent_buffer
*exists
= NULL
;
4902 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4906 if (!IS_ALIGNED(start
, fs_info
->sectorsize
)) {
4907 btrfs_err(fs_info
, "bad tree block start %llu", start
);
4908 return ERR_PTR(-EINVAL
);
4911 eb
= find_extent_buffer(fs_info
, start
);
4915 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4917 return ERR_PTR(-ENOMEM
);
4919 num_pages
= num_extent_pages(eb
);
4920 for (i
= 0; i
< num_pages
; i
++, index
++) {
4921 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4923 exists
= ERR_PTR(-ENOMEM
);
4927 spin_lock(&mapping
->private_lock
);
4928 if (PagePrivate(p
)) {
4930 * We could have already allocated an eb for this page
4931 * and attached one so lets see if we can get a ref on
4932 * the existing eb, and if we can we know it's good and
4933 * we can just return that one, else we know we can just
4934 * overwrite page->private.
4936 exists
= (struct extent_buffer
*)p
->private;
4937 if (atomic_inc_not_zero(&exists
->refs
)) {
4938 spin_unlock(&mapping
->private_lock
);
4941 mark_extent_buffer_accessed(exists
, p
);
4947 * Do this so attach doesn't complain and we need to
4948 * drop the ref the old guy had.
4950 ClearPagePrivate(p
);
4951 WARN_ON(PageDirty(p
));
4954 attach_extent_buffer_page(eb
, p
);
4955 spin_unlock(&mapping
->private_lock
);
4956 WARN_ON(PageDirty(p
));
4958 if (!PageUptodate(p
))
4962 * We can't unlock the pages just yet since the extent buffer
4963 * hasn't been properly inserted in the radix tree, this
4964 * opens a race with btree_releasepage which can free a page
4965 * while we are still filling in all pages for the buffer and
4970 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4972 ret
= radix_tree_preload(GFP_NOFS
);
4974 exists
= ERR_PTR(ret
);
4978 spin_lock(&fs_info
->buffer_lock
);
4979 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4980 start
>> PAGE_SHIFT
, eb
);
4981 spin_unlock(&fs_info
->buffer_lock
);
4982 radix_tree_preload_end();
4983 if (ret
== -EEXIST
) {
4984 exists
= find_extent_buffer(fs_info
, start
);
4990 /* add one reference for the tree */
4991 check_buffer_tree_ref(eb
);
4992 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4995 * Now it's safe to unlock the pages because any calls to
4996 * btree_releasepage will correctly detect that a page belongs to a
4997 * live buffer and won't free them prematurely.
4999 for (i
= 0; i
< num_pages
; i
++)
5000 unlock_page(eb
->pages
[i
]);
5004 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
5005 for (i
= 0; i
< num_pages
; i
++) {
5007 unlock_page(eb
->pages
[i
]);
5010 btrfs_release_extent_buffer(eb
);
5014 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
5016 struct extent_buffer
*eb
=
5017 container_of(head
, struct extent_buffer
, rcu_head
);
5019 __free_extent_buffer(eb
);
5022 static int release_extent_buffer(struct extent_buffer
*eb
)
5024 lockdep_assert_held(&eb
->refs_lock
);
5026 WARN_ON(atomic_read(&eb
->refs
) == 0);
5027 if (atomic_dec_and_test(&eb
->refs
)) {
5028 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5029 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5031 spin_unlock(&eb
->refs_lock
);
5033 spin_lock(&fs_info
->buffer_lock
);
5034 radix_tree_delete(&fs_info
->buffer_radix
,
5035 eb
->start
>> PAGE_SHIFT
);
5036 spin_unlock(&fs_info
->buffer_lock
);
5038 spin_unlock(&eb
->refs_lock
);
5041 /* Should be safe to release our pages at this point */
5042 btrfs_release_extent_buffer_pages(eb
);
5043 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5044 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED
, &eb
->bflags
))) {
5045 __free_extent_buffer(eb
);
5049 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5052 spin_unlock(&eb
->refs_lock
);
5057 void free_extent_buffer(struct extent_buffer
*eb
)
5065 refs
= atomic_read(&eb
->refs
);
5066 if ((!test_bit(EXTENT_BUFFER_UNMAPPED
, &eb
->bflags
) && refs
<= 3)
5067 || (test_bit(EXTENT_BUFFER_UNMAPPED
, &eb
->bflags
) &&
5070 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5075 spin_lock(&eb
->refs_lock
);
5076 if (atomic_read(&eb
->refs
) == 2 &&
5077 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5078 !extent_buffer_under_io(eb
) &&
5079 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5080 atomic_dec(&eb
->refs
);
5083 * I know this is terrible, but it's temporary until we stop tracking
5084 * the uptodate bits and such for the extent buffers.
5086 release_extent_buffer(eb
);
5089 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5094 spin_lock(&eb
->refs_lock
);
5095 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5097 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5098 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5099 atomic_dec(&eb
->refs
);
5100 release_extent_buffer(eb
);
5103 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5109 num_pages
= num_extent_pages(eb
);
5111 for (i
= 0; i
< num_pages
; i
++) {
5112 page
= eb
->pages
[i
];
5113 if (!PageDirty(page
))
5117 WARN_ON(!PagePrivate(page
));
5119 clear_page_dirty_for_io(page
);
5120 xa_lock_irq(&page
->mapping
->i_pages
);
5121 if (!PageDirty(page
))
5122 __xa_clear_mark(&page
->mapping
->i_pages
,
5123 page_index(page
), PAGECACHE_TAG_DIRTY
);
5124 xa_unlock_irq(&page
->mapping
->i_pages
);
5125 ClearPageError(page
);
5128 WARN_ON(atomic_read(&eb
->refs
) == 0);
5131 bool set_extent_buffer_dirty(struct extent_buffer
*eb
)
5137 check_buffer_tree_ref(eb
);
5139 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5141 num_pages
= num_extent_pages(eb
);
5142 WARN_ON(atomic_read(&eb
->refs
) == 0);
5143 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5146 for (i
= 0; i
< num_pages
; i
++)
5147 set_page_dirty(eb
->pages
[i
]);
5149 #ifdef CONFIG_BTRFS_DEBUG
5150 for (i
= 0; i
< num_pages
; i
++)
5151 ASSERT(PageDirty(eb
->pages
[i
]));
5157 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5163 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5164 num_pages
= num_extent_pages(eb
);
5165 for (i
= 0; i
< num_pages
; i
++) {
5166 page
= eb
->pages
[i
];
5168 ClearPageUptodate(page
);
5172 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5178 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5179 num_pages
= num_extent_pages(eb
);
5180 for (i
= 0; i
< num_pages
; i
++) {
5181 page
= eb
->pages
[i
];
5182 SetPageUptodate(page
);
5186 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5187 struct extent_buffer
*eb
, int wait
, int mirror_num
)
5193 int locked_pages
= 0;
5194 int all_uptodate
= 1;
5196 unsigned long num_reads
= 0;
5197 struct bio
*bio
= NULL
;
5198 unsigned long bio_flags
= 0;
5200 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5203 num_pages
= num_extent_pages(eb
);
5204 for (i
= 0; i
< num_pages
; i
++) {
5205 page
= eb
->pages
[i
];
5206 if (wait
== WAIT_NONE
) {
5207 if (!trylock_page(page
))
5215 * We need to firstly lock all pages to make sure that
5216 * the uptodate bit of our pages won't be affected by
5217 * clear_extent_buffer_uptodate().
5219 for (i
= 0; i
< num_pages
; i
++) {
5220 page
= eb
->pages
[i
];
5221 if (!PageUptodate(page
)) {
5228 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5232 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5233 eb
->read_mirror
= 0;
5234 atomic_set(&eb
->io_pages
, num_reads
);
5235 for (i
= 0; i
< num_pages
; i
++) {
5236 page
= eb
->pages
[i
];
5238 if (!PageUptodate(page
)) {
5240 atomic_dec(&eb
->io_pages
);
5245 ClearPageError(page
);
5246 err
= __extent_read_full_page(tree
, page
,
5247 btree_get_extent
, &bio
,
5248 mirror_num
, &bio_flags
,
5253 * We use &bio in above __extent_read_full_page,
5254 * so we ensure that if it returns error, the
5255 * current page fails to add itself to bio and
5256 * it's been unlocked.
5258 * We must dec io_pages by ourselves.
5260 atomic_dec(&eb
->io_pages
);
5268 err
= submit_one_bio(bio
, mirror_num
, bio_flags
);
5273 if (ret
|| wait
!= WAIT_COMPLETE
)
5276 for (i
= 0; i
< num_pages
; i
++) {
5277 page
= eb
->pages
[i
];
5278 wait_on_page_locked(page
);
5279 if (!PageUptodate(page
))
5286 while (locked_pages
> 0) {
5288 page
= eb
->pages
[locked_pages
];
5294 void read_extent_buffer(const struct extent_buffer
*eb
, void *dstv
,
5295 unsigned long start
, unsigned long len
)
5301 char *dst
= (char *)dstv
;
5302 size_t start_offset
= offset_in_page(eb
->start
);
5303 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5305 if (start
+ len
> eb
->len
) {
5306 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5307 eb
->start
, eb
->len
, start
, len
);
5308 memset(dst
, 0, len
);
5312 offset
= offset_in_page(start_offset
+ start
);
5315 page
= eb
->pages
[i
];
5317 cur
= min(len
, (PAGE_SIZE
- offset
));
5318 kaddr
= page_address(page
);
5319 memcpy(dst
, kaddr
+ offset
, cur
);
5328 int read_extent_buffer_to_user(const struct extent_buffer
*eb
,
5330 unsigned long start
, unsigned long len
)
5336 char __user
*dst
= (char __user
*)dstv
;
5337 size_t start_offset
= offset_in_page(eb
->start
);
5338 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5341 WARN_ON(start
> eb
->len
);
5342 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5344 offset
= offset_in_page(start_offset
+ start
);
5347 page
= eb
->pages
[i
];
5349 cur
= min(len
, (PAGE_SIZE
- offset
));
5350 kaddr
= page_address(page
);
5351 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5366 * return 0 if the item is found within a page.
5367 * return 1 if the item spans two pages.
5368 * return -EINVAL otherwise.
5370 int map_private_extent_buffer(const struct extent_buffer
*eb
,
5371 unsigned long start
, unsigned long min_len
,
5372 char **map
, unsigned long *map_start
,
5373 unsigned long *map_len
)
5378 size_t start_offset
= offset_in_page(eb
->start
);
5379 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5380 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5383 if (start
+ min_len
> eb
->len
) {
5384 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5385 eb
->start
, eb
->len
, start
, min_len
);
5393 offset
= start_offset
;
5397 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5401 kaddr
= page_address(p
);
5402 *map
= kaddr
+ offset
;
5403 *map_len
= PAGE_SIZE
- offset
;
5407 int memcmp_extent_buffer(const struct extent_buffer
*eb
, const void *ptrv
,
5408 unsigned long start
, unsigned long len
)
5414 char *ptr
= (char *)ptrv
;
5415 size_t start_offset
= offset_in_page(eb
->start
);
5416 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5419 WARN_ON(start
> eb
->len
);
5420 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5422 offset
= offset_in_page(start_offset
+ start
);
5425 page
= eb
->pages
[i
];
5427 cur
= min(len
, (PAGE_SIZE
- offset
));
5429 kaddr
= page_address(page
);
5430 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5442 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer
*eb
,
5447 WARN_ON(!PageUptodate(eb
->pages
[0]));
5448 kaddr
= page_address(eb
->pages
[0]);
5449 memcpy(kaddr
+ offsetof(struct btrfs_header
, chunk_tree_uuid
), srcv
,
5453 void write_extent_buffer_fsid(struct extent_buffer
*eb
, const void *srcv
)
5457 WARN_ON(!PageUptodate(eb
->pages
[0]));
5458 kaddr
= page_address(eb
->pages
[0]);
5459 memcpy(kaddr
+ offsetof(struct btrfs_header
, fsid
), srcv
,
5463 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5464 unsigned long start
, unsigned long len
)
5470 char *src
= (char *)srcv
;
5471 size_t start_offset
= offset_in_page(eb
->start
);
5472 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5474 WARN_ON(start
> eb
->len
);
5475 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5477 offset
= offset_in_page(start_offset
+ start
);
5480 page
= eb
->pages
[i
];
5481 WARN_ON(!PageUptodate(page
));
5483 cur
= min(len
, PAGE_SIZE
- offset
);
5484 kaddr
= page_address(page
);
5485 memcpy(kaddr
+ offset
, src
, cur
);
5494 void memzero_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5501 size_t start_offset
= offset_in_page(eb
->start
);
5502 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5504 WARN_ON(start
> eb
->len
);
5505 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5507 offset
= offset_in_page(start_offset
+ start
);
5510 page
= eb
->pages
[i
];
5511 WARN_ON(!PageUptodate(page
));
5513 cur
= min(len
, PAGE_SIZE
- offset
);
5514 kaddr
= page_address(page
);
5515 memset(kaddr
+ offset
, 0, cur
);
5523 void copy_extent_buffer_full(struct extent_buffer
*dst
,
5524 struct extent_buffer
*src
)
5529 ASSERT(dst
->len
== src
->len
);
5531 num_pages
= num_extent_pages(dst
);
5532 for (i
= 0; i
< num_pages
; i
++)
5533 copy_page(page_address(dst
->pages
[i
]),
5534 page_address(src
->pages
[i
]));
5537 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5538 unsigned long dst_offset
, unsigned long src_offset
,
5541 u64 dst_len
= dst
->len
;
5546 size_t start_offset
= offset_in_page(dst
->start
);
5547 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5549 WARN_ON(src
->len
!= dst_len
);
5551 offset
= offset_in_page(start_offset
+ dst_offset
);
5554 page
= dst
->pages
[i
];
5555 WARN_ON(!PageUptodate(page
));
5557 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5559 kaddr
= page_address(page
);
5560 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5570 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5572 * @eb: the extent buffer
5573 * @start: offset of the bitmap item in the extent buffer
5575 * @page_index: return index of the page in the extent buffer that contains the
5577 * @page_offset: return offset into the page given by page_index
5579 * This helper hides the ugliness of finding the byte in an extent buffer which
5580 * contains a given bit.
5582 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5583 unsigned long start
, unsigned long nr
,
5584 unsigned long *page_index
,
5585 size_t *page_offset
)
5587 size_t start_offset
= offset_in_page(eb
->start
);
5588 size_t byte_offset
= BIT_BYTE(nr
);
5592 * The byte we want is the offset of the extent buffer + the offset of
5593 * the bitmap item in the extent buffer + the offset of the byte in the
5596 offset
= start_offset
+ start
+ byte_offset
;
5598 *page_index
= offset
>> PAGE_SHIFT
;
5599 *page_offset
= offset_in_page(offset
);
5603 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5604 * @eb: the extent buffer
5605 * @start: offset of the bitmap item in the extent buffer
5606 * @nr: bit number to test
5608 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5616 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5617 page
= eb
->pages
[i
];
5618 WARN_ON(!PageUptodate(page
));
5619 kaddr
= page_address(page
);
5620 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5624 * extent_buffer_bitmap_set - set an area of a bitmap
5625 * @eb: the extent buffer
5626 * @start: offset of the bitmap item in the extent buffer
5627 * @pos: bit number of the first bit
5628 * @len: number of bits to set
5630 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5631 unsigned long pos
, unsigned long len
)
5637 const unsigned int size
= pos
+ len
;
5638 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5639 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5641 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5642 page
= eb
->pages
[i
];
5643 WARN_ON(!PageUptodate(page
));
5644 kaddr
= page_address(page
);
5646 while (len
>= bits_to_set
) {
5647 kaddr
[offset
] |= mask_to_set
;
5649 bits_to_set
= BITS_PER_BYTE
;
5651 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5653 page
= eb
->pages
[++i
];
5654 WARN_ON(!PageUptodate(page
));
5655 kaddr
= page_address(page
);
5659 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5660 kaddr
[offset
] |= mask_to_set
;
5666 * extent_buffer_bitmap_clear - clear an area of a bitmap
5667 * @eb: the extent buffer
5668 * @start: offset of the bitmap item in the extent buffer
5669 * @pos: bit number of the first bit
5670 * @len: number of bits to clear
5672 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5673 unsigned long pos
, unsigned long len
)
5679 const unsigned int size
= pos
+ len
;
5680 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5681 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5683 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5684 page
= eb
->pages
[i
];
5685 WARN_ON(!PageUptodate(page
));
5686 kaddr
= page_address(page
);
5688 while (len
>= bits_to_clear
) {
5689 kaddr
[offset
] &= ~mask_to_clear
;
5690 len
-= bits_to_clear
;
5691 bits_to_clear
= BITS_PER_BYTE
;
5693 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5695 page
= eb
->pages
[++i
];
5696 WARN_ON(!PageUptodate(page
));
5697 kaddr
= page_address(page
);
5701 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5702 kaddr
[offset
] &= ~mask_to_clear
;
5706 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5708 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5709 return distance
< len
;
5712 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5713 unsigned long dst_off
, unsigned long src_off
,
5716 char *dst_kaddr
= page_address(dst_page
);
5718 int must_memmove
= 0;
5720 if (dst_page
!= src_page
) {
5721 src_kaddr
= page_address(src_page
);
5723 src_kaddr
= dst_kaddr
;
5724 if (areas_overlap(src_off
, dst_off
, len
))
5729 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5731 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5734 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5735 unsigned long src_offset
, unsigned long len
)
5737 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5739 size_t dst_off_in_page
;
5740 size_t src_off_in_page
;
5741 size_t start_offset
= offset_in_page(dst
->start
);
5742 unsigned long dst_i
;
5743 unsigned long src_i
;
5745 if (src_offset
+ len
> dst
->len
) {
5747 "memmove bogus src_offset %lu move len %lu dst len %lu",
5748 src_offset
, len
, dst
->len
);
5751 if (dst_offset
+ len
> dst
->len
) {
5753 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5754 dst_offset
, len
, dst
->len
);
5759 dst_off_in_page
= offset_in_page(start_offset
+ dst_offset
);
5760 src_off_in_page
= offset_in_page(start_offset
+ src_offset
);
5762 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5763 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5765 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5767 cur
= min_t(unsigned long, cur
,
5768 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5770 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5771 dst_off_in_page
, src_off_in_page
, cur
);
5779 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5780 unsigned long src_offset
, unsigned long len
)
5782 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5784 size_t dst_off_in_page
;
5785 size_t src_off_in_page
;
5786 unsigned long dst_end
= dst_offset
+ len
- 1;
5787 unsigned long src_end
= src_offset
+ len
- 1;
5788 size_t start_offset
= offset_in_page(dst
->start
);
5789 unsigned long dst_i
;
5790 unsigned long src_i
;
5792 if (src_offset
+ len
> dst
->len
) {
5794 "memmove bogus src_offset %lu move len %lu len %lu",
5795 src_offset
, len
, dst
->len
);
5798 if (dst_offset
+ len
> dst
->len
) {
5800 "memmove bogus dst_offset %lu move len %lu len %lu",
5801 dst_offset
, len
, dst
->len
);
5804 if (dst_offset
< src_offset
) {
5805 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5809 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5810 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5812 dst_off_in_page
= offset_in_page(start_offset
+ dst_end
);
5813 src_off_in_page
= offset_in_page(start_offset
+ src_end
);
5815 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5816 cur
= min(cur
, dst_off_in_page
+ 1);
5817 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5818 dst_off_in_page
- cur
+ 1,
5819 src_off_in_page
- cur
+ 1, cur
);
5827 int try_release_extent_buffer(struct page
*page
)
5829 struct extent_buffer
*eb
;
5832 * We need to make sure nobody is attaching this page to an eb right
5835 spin_lock(&page
->mapping
->private_lock
);
5836 if (!PagePrivate(page
)) {
5837 spin_unlock(&page
->mapping
->private_lock
);
5841 eb
= (struct extent_buffer
*)page
->private;
5845 * This is a little awful but should be ok, we need to make sure that
5846 * the eb doesn't disappear out from under us while we're looking at
5849 spin_lock(&eb
->refs_lock
);
5850 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5851 spin_unlock(&eb
->refs_lock
);
5852 spin_unlock(&page
->mapping
->private_lock
);
5855 spin_unlock(&page
->mapping
->private_lock
);
5858 * If tree ref isn't set then we know the ref on this eb is a real ref,
5859 * so just return, this page will likely be freed soon anyway.
5861 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5862 spin_unlock(&eb
->refs_lock
);
5866 return release_extent_buffer(eb
);