1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
24 static struct kmem_cache
*extent_state_cache
;
25 static struct kmem_cache
*extent_buffer_cache
;
26 static struct bio_set
*btrfs_bioset
;
28 static inline bool extent_state_in_tree(const struct extent_state
*state
)
30 return !RB_EMPTY_NODE(&state
->rb_node
);
33 #ifdef CONFIG_BTRFS_DEBUG
34 static LIST_HEAD(buffers
);
35 static LIST_HEAD(states
);
37 static DEFINE_SPINLOCK(leak_lock
);
40 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
44 spin_lock_irqsave(&leak_lock
, flags
);
46 spin_unlock_irqrestore(&leak_lock
, flags
);
50 void btrfs_leak_debug_del(struct list_head
*entry
)
54 spin_lock_irqsave(&leak_lock
, flags
);
56 spin_unlock_irqrestore(&leak_lock
, flags
);
60 void btrfs_leak_debug_check(void)
62 struct extent_state
*state
;
63 struct extent_buffer
*eb
;
65 while (!list_empty(&states
)) {
66 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
67 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
68 state
->start
, state
->end
, state
->state
,
69 extent_state_in_tree(state
),
70 refcount_read(&state
->refs
));
71 list_del(&state
->leak_list
);
72 kmem_cache_free(extent_state_cache
, state
);
75 while (!list_empty(&buffers
)) {
76 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
77 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
78 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
79 list_del(&eb
->leak_list
);
80 kmem_cache_free(extent_buffer_cache
, eb
);
84 #define btrfs_debug_check_extent_io_range(tree, start, end) \
85 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
86 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
87 struct extent_io_tree
*tree
, u64 start
, u64 end
)
89 if (tree
->ops
&& tree
->ops
->check_extent_io_range
)
90 tree
->ops
->check_extent_io_range(tree
->private_data
, caller
,
94 #define btrfs_leak_debug_add(new, head) do {} while (0)
95 #define btrfs_leak_debug_del(entry) do {} while (0)
96 #define btrfs_leak_debug_check() do {} while (0)
97 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
100 #define BUFFER_LRU_MAX 64
105 struct rb_node rb_node
;
108 struct extent_page_data
{
110 struct extent_io_tree
*tree
;
111 get_extent_t
*get_extent
;
113 /* tells writepage not to lock the state bits for this range
114 * it still does the unlocking
116 unsigned int extent_locked
:1;
118 /* tells the submit_bio code to use REQ_SYNC */
119 unsigned int sync_io
:1;
122 static void add_extent_changeset(struct extent_state
*state
, unsigned bits
,
123 struct extent_changeset
*changeset
,
130 if (set
&& (state
->state
& bits
) == bits
)
132 if (!set
&& (state
->state
& bits
) == 0)
134 changeset
->bytes_changed
+= state
->end
- state
->start
+ 1;
135 ret
= ulist_add(&changeset
->range_changed
, state
->start
, state
->end
,
141 static noinline
void flush_write_bio(void *data
);
142 static inline struct btrfs_fs_info
*
143 tree_fs_info(struct extent_io_tree
*tree
)
146 return tree
->ops
->tree_fs_info(tree
->private_data
);
150 int __init
extent_io_init(void)
152 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
153 sizeof(struct extent_state
), 0,
154 SLAB_MEM_SPREAD
, NULL
);
155 if (!extent_state_cache
)
158 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
159 sizeof(struct extent_buffer
), 0,
160 SLAB_MEM_SPREAD
, NULL
);
161 if (!extent_buffer_cache
)
162 goto free_state_cache
;
164 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
165 offsetof(struct btrfs_io_bio
, bio
),
168 goto free_buffer_cache
;
170 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
176 bioset_free(btrfs_bioset
);
180 kmem_cache_destroy(extent_buffer_cache
);
181 extent_buffer_cache
= NULL
;
184 kmem_cache_destroy(extent_state_cache
);
185 extent_state_cache
= NULL
;
189 void extent_io_exit(void)
191 btrfs_leak_debug_check();
194 * Make sure all delayed rcu free are flushed before we
198 kmem_cache_destroy(extent_state_cache
);
199 kmem_cache_destroy(extent_buffer_cache
);
201 bioset_free(btrfs_bioset
);
204 void extent_io_tree_init(struct extent_io_tree
*tree
,
207 tree
->state
= RB_ROOT
;
209 tree
->dirty_bytes
= 0;
210 spin_lock_init(&tree
->lock
);
211 tree
->private_data
= private_data
;
214 static struct extent_state
*alloc_extent_state(gfp_t mask
)
216 struct extent_state
*state
;
219 * The given mask might be not appropriate for the slab allocator,
220 * drop the unsupported bits
222 mask
&= ~(__GFP_DMA32
|__GFP_HIGHMEM
);
223 state
= kmem_cache_alloc(extent_state_cache
, mask
);
227 state
->failrec
= NULL
;
228 RB_CLEAR_NODE(&state
->rb_node
);
229 btrfs_leak_debug_add(&state
->leak_list
, &states
);
230 refcount_set(&state
->refs
, 1);
231 init_waitqueue_head(&state
->wq
);
232 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
236 void free_extent_state(struct extent_state
*state
)
240 if (refcount_dec_and_test(&state
->refs
)) {
241 WARN_ON(extent_state_in_tree(state
));
242 btrfs_leak_debug_del(&state
->leak_list
);
243 trace_free_extent_state(state
, _RET_IP_
);
244 kmem_cache_free(extent_state_cache
, state
);
248 static struct rb_node
*tree_insert(struct rb_root
*root
,
249 struct rb_node
*search_start
,
251 struct rb_node
*node
,
252 struct rb_node
***p_in
,
253 struct rb_node
**parent_in
)
256 struct rb_node
*parent
= NULL
;
257 struct tree_entry
*entry
;
259 if (p_in
&& parent_in
) {
265 p
= search_start
? &search_start
: &root
->rb_node
;
268 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
270 if (offset
< entry
->start
)
272 else if (offset
> entry
->end
)
279 rb_link_node(node
, parent
, p
);
280 rb_insert_color(node
, root
);
284 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
285 struct rb_node
**prev_ret
,
286 struct rb_node
**next_ret
,
287 struct rb_node
***p_ret
,
288 struct rb_node
**parent_ret
)
290 struct rb_root
*root
= &tree
->state
;
291 struct rb_node
**n
= &root
->rb_node
;
292 struct rb_node
*prev
= NULL
;
293 struct rb_node
*orig_prev
= NULL
;
294 struct tree_entry
*entry
;
295 struct tree_entry
*prev_entry
= NULL
;
299 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
302 if (offset
< entry
->start
)
304 else if (offset
> entry
->end
)
317 while (prev
&& offset
> prev_entry
->end
) {
318 prev
= rb_next(prev
);
319 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
326 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
327 while (prev
&& offset
< prev_entry
->start
) {
328 prev
= rb_prev(prev
);
329 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
336 static inline struct rb_node
*
337 tree_search_for_insert(struct extent_io_tree
*tree
,
339 struct rb_node
***p_ret
,
340 struct rb_node
**parent_ret
)
342 struct rb_node
*prev
= NULL
;
345 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
351 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
354 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
357 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
358 struct extent_state
*other
)
360 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
361 tree
->ops
->merge_extent_hook(tree
->private_data
, new, other
);
365 * utility function to look for merge candidates inside a given range.
366 * Any extents with matching state are merged together into a single
367 * extent in the tree. Extents with EXTENT_IO in their state field
368 * are not merged because the end_io handlers need to be able to do
369 * operations on them without sleeping (or doing allocations/splits).
371 * This should be called with the tree lock held.
373 static void merge_state(struct extent_io_tree
*tree
,
374 struct extent_state
*state
)
376 struct extent_state
*other
;
377 struct rb_node
*other_node
;
379 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
382 other_node
= rb_prev(&state
->rb_node
);
384 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
385 if (other
->end
== state
->start
- 1 &&
386 other
->state
== state
->state
) {
387 merge_cb(tree
, state
, other
);
388 state
->start
= other
->start
;
389 rb_erase(&other
->rb_node
, &tree
->state
);
390 RB_CLEAR_NODE(&other
->rb_node
);
391 free_extent_state(other
);
394 other_node
= rb_next(&state
->rb_node
);
396 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
397 if (other
->start
== state
->end
+ 1 &&
398 other
->state
== state
->state
) {
399 merge_cb(tree
, state
, other
);
400 state
->end
= other
->end
;
401 rb_erase(&other
->rb_node
, &tree
->state
);
402 RB_CLEAR_NODE(&other
->rb_node
);
403 free_extent_state(other
);
408 static void set_state_cb(struct extent_io_tree
*tree
,
409 struct extent_state
*state
, unsigned *bits
)
411 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
412 tree
->ops
->set_bit_hook(tree
->private_data
, state
, bits
);
415 static void clear_state_cb(struct extent_io_tree
*tree
,
416 struct extent_state
*state
, unsigned *bits
)
418 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
419 tree
->ops
->clear_bit_hook(tree
->private_data
, state
, bits
);
422 static void set_state_bits(struct extent_io_tree
*tree
,
423 struct extent_state
*state
, unsigned *bits
,
424 struct extent_changeset
*changeset
);
427 * insert an extent_state struct into the tree. 'bits' are set on the
428 * struct before it is inserted.
430 * This may return -EEXIST if the extent is already there, in which case the
431 * state struct is freed.
433 * The tree lock is not taken internally. This is a utility function and
434 * probably isn't what you want to call (see set/clear_extent_bit).
436 static int insert_state(struct extent_io_tree
*tree
,
437 struct extent_state
*state
, u64 start
, u64 end
,
439 struct rb_node
**parent
,
440 unsigned *bits
, struct extent_changeset
*changeset
)
442 struct rb_node
*node
;
445 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
447 state
->start
= start
;
450 set_state_bits(tree
, state
, bits
, changeset
);
452 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
454 struct extent_state
*found
;
455 found
= rb_entry(node
, struct extent_state
, rb_node
);
456 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
457 found
->start
, found
->end
, start
, end
);
460 merge_state(tree
, state
);
464 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
467 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
468 tree
->ops
->split_extent_hook(tree
->private_data
, orig
, split
);
472 * split a given extent state struct in two, inserting the preallocated
473 * struct 'prealloc' as the newly created second half. 'split' indicates an
474 * offset inside 'orig' where it should be split.
477 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
478 * are two extent state structs in the tree:
479 * prealloc: [orig->start, split - 1]
480 * orig: [ split, orig->end ]
482 * The tree locks are not taken by this function. They need to be held
485 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
486 struct extent_state
*prealloc
, u64 split
)
488 struct rb_node
*node
;
490 split_cb(tree
, orig
, split
);
492 prealloc
->start
= orig
->start
;
493 prealloc
->end
= split
- 1;
494 prealloc
->state
= orig
->state
;
497 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
498 &prealloc
->rb_node
, NULL
, NULL
);
500 free_extent_state(prealloc
);
506 static struct extent_state
*next_state(struct extent_state
*state
)
508 struct rb_node
*next
= rb_next(&state
->rb_node
);
510 return rb_entry(next
, struct extent_state
, rb_node
);
516 * utility function to clear some bits in an extent state struct.
517 * it will optionally wake up any one waiting on this state (wake == 1).
519 * If no bits are set on the state struct after clearing things, the
520 * struct is freed and removed from the tree
522 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
523 struct extent_state
*state
,
524 unsigned *bits
, int wake
,
525 struct extent_changeset
*changeset
)
527 struct extent_state
*next
;
528 unsigned bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
530 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
531 u64 range
= state
->end
- state
->start
+ 1;
532 WARN_ON(range
> tree
->dirty_bytes
);
533 tree
->dirty_bytes
-= range
;
535 clear_state_cb(tree
, state
, bits
);
536 add_extent_changeset(state
, bits_to_clear
, changeset
, 0);
537 state
->state
&= ~bits_to_clear
;
540 if (state
->state
== 0) {
541 next
= next_state(state
);
542 if (extent_state_in_tree(state
)) {
543 rb_erase(&state
->rb_node
, &tree
->state
);
544 RB_CLEAR_NODE(&state
->rb_node
);
545 free_extent_state(state
);
550 merge_state(tree
, state
);
551 next
= next_state(state
);
556 static struct extent_state
*
557 alloc_extent_state_atomic(struct extent_state
*prealloc
)
560 prealloc
= alloc_extent_state(GFP_ATOMIC
);
565 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
567 btrfs_panic(tree_fs_info(tree
), err
,
568 "Locking error: Extent tree was modified by another thread while locked.");
572 * clear some bits on a range in the tree. This may require splitting
573 * or inserting elements in the tree, so the gfp mask is used to
574 * indicate which allocations or sleeping are allowed.
576 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
577 * the given range from the tree regardless of state (ie for truncate).
579 * the range [start, end] is inclusive.
581 * This takes the tree lock, and returns 0 on success and < 0 on error.
583 static int __clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
584 unsigned bits
, int wake
, int delete,
585 struct extent_state
**cached_state
,
586 gfp_t mask
, struct extent_changeset
*changeset
)
588 struct extent_state
*state
;
589 struct extent_state
*cached
;
590 struct extent_state
*prealloc
= NULL
;
591 struct rb_node
*node
;
596 btrfs_debug_check_extent_io_range(tree
, start
, end
);
598 if (bits
& EXTENT_DELALLOC
)
599 bits
|= EXTENT_NORESERVE
;
602 bits
|= ~EXTENT_CTLBITS
;
603 bits
|= EXTENT_FIRST_DELALLOC
;
605 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
608 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
610 * Don't care for allocation failure here because we might end
611 * up not needing the pre-allocated extent state at all, which
612 * is the case if we only have in the tree extent states that
613 * cover our input range and don't cover too any other range.
614 * If we end up needing a new extent state we allocate it later.
616 prealloc
= alloc_extent_state(mask
);
619 spin_lock(&tree
->lock
);
621 cached
= *cached_state
;
624 *cached_state
= NULL
;
628 if (cached
&& extent_state_in_tree(cached
) &&
629 cached
->start
<= start
&& cached
->end
> start
) {
631 refcount_dec(&cached
->refs
);
636 free_extent_state(cached
);
639 * this search will find the extents that end after
642 node
= tree_search(tree
, start
);
645 state
= rb_entry(node
, struct extent_state
, rb_node
);
647 if (state
->start
> end
)
649 WARN_ON(state
->end
< start
);
650 last_end
= state
->end
;
652 /* the state doesn't have the wanted bits, go ahead */
653 if (!(state
->state
& bits
)) {
654 state
= next_state(state
);
659 * | ---- desired range ---- |
661 * | ------------- state -------------- |
663 * We need to split the extent we found, and may flip
664 * bits on second half.
666 * If the extent we found extends past our range, we
667 * just split and search again. It'll get split again
668 * the next time though.
670 * If the extent we found is inside our range, we clear
671 * the desired bit on it.
674 if (state
->start
< start
) {
675 prealloc
= alloc_extent_state_atomic(prealloc
);
677 err
= split_state(tree
, state
, prealloc
, start
);
679 extent_io_tree_panic(tree
, err
);
684 if (state
->end
<= end
) {
685 state
= clear_state_bit(tree
, state
, &bits
, wake
,
692 * | ---- desired range ---- |
694 * We need to split the extent, and clear the bit
697 if (state
->start
<= end
&& state
->end
> end
) {
698 prealloc
= alloc_extent_state_atomic(prealloc
);
700 err
= split_state(tree
, state
, prealloc
, end
+ 1);
702 extent_io_tree_panic(tree
, err
);
707 clear_state_bit(tree
, prealloc
, &bits
, wake
, changeset
);
713 state
= clear_state_bit(tree
, state
, &bits
, wake
, changeset
);
715 if (last_end
== (u64
)-1)
717 start
= last_end
+ 1;
718 if (start
<= end
&& state
&& !need_resched())
724 spin_unlock(&tree
->lock
);
725 if (gfpflags_allow_blocking(mask
))
730 spin_unlock(&tree
->lock
);
732 free_extent_state(prealloc
);
738 static void wait_on_state(struct extent_io_tree
*tree
,
739 struct extent_state
*state
)
740 __releases(tree
->lock
)
741 __acquires(tree
->lock
)
744 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
745 spin_unlock(&tree
->lock
);
747 spin_lock(&tree
->lock
);
748 finish_wait(&state
->wq
, &wait
);
752 * waits for one or more bits to clear on a range in the state tree.
753 * The range [start, end] is inclusive.
754 * The tree lock is taken by this function
756 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
759 struct extent_state
*state
;
760 struct rb_node
*node
;
762 btrfs_debug_check_extent_io_range(tree
, start
, end
);
764 spin_lock(&tree
->lock
);
768 * this search will find all the extents that end after
771 node
= tree_search(tree
, start
);
776 state
= rb_entry(node
, struct extent_state
, rb_node
);
778 if (state
->start
> end
)
781 if (state
->state
& bits
) {
782 start
= state
->start
;
783 refcount_inc(&state
->refs
);
784 wait_on_state(tree
, state
);
785 free_extent_state(state
);
788 start
= state
->end
+ 1;
793 if (!cond_resched_lock(&tree
->lock
)) {
794 node
= rb_next(node
);
799 spin_unlock(&tree
->lock
);
802 static void set_state_bits(struct extent_io_tree
*tree
,
803 struct extent_state
*state
,
804 unsigned *bits
, struct extent_changeset
*changeset
)
806 unsigned bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
808 set_state_cb(tree
, state
, bits
);
809 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
810 u64 range
= state
->end
- state
->start
+ 1;
811 tree
->dirty_bytes
+= range
;
813 add_extent_changeset(state
, bits_to_set
, changeset
, 1);
814 state
->state
|= bits_to_set
;
817 static void cache_state_if_flags(struct extent_state
*state
,
818 struct extent_state
**cached_ptr
,
821 if (cached_ptr
&& !(*cached_ptr
)) {
822 if (!flags
|| (state
->state
& flags
)) {
824 refcount_inc(&state
->refs
);
829 static void cache_state(struct extent_state
*state
,
830 struct extent_state
**cached_ptr
)
832 return cache_state_if_flags(state
, cached_ptr
,
833 EXTENT_IOBITS
| EXTENT_BOUNDARY
);
837 * set some bits on a range in the tree. This may require allocations or
838 * sleeping, so the gfp mask is used to indicate what is allowed.
840 * If any of the exclusive bits are set, this will fail with -EEXIST if some
841 * part of the range already has the desired bits set. The start of the
842 * existing range is returned in failed_start in this case.
844 * [start, end] is inclusive This takes the tree lock.
847 static int __must_check
848 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
849 unsigned bits
, unsigned exclusive_bits
,
850 u64
*failed_start
, struct extent_state
**cached_state
,
851 gfp_t mask
, struct extent_changeset
*changeset
)
853 struct extent_state
*state
;
854 struct extent_state
*prealloc
= NULL
;
855 struct rb_node
*node
;
857 struct rb_node
*parent
;
862 btrfs_debug_check_extent_io_range(tree
, start
, end
);
864 bits
|= EXTENT_FIRST_DELALLOC
;
866 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
868 * Don't care for allocation failure here because we might end
869 * up not needing the pre-allocated extent state at all, which
870 * is the case if we only have in the tree extent states that
871 * cover our input range and don't cover too any other range.
872 * If we end up needing a new extent state we allocate it later.
874 prealloc
= alloc_extent_state(mask
);
877 spin_lock(&tree
->lock
);
878 if (cached_state
&& *cached_state
) {
879 state
= *cached_state
;
880 if (state
->start
<= start
&& state
->end
> start
&&
881 extent_state_in_tree(state
)) {
882 node
= &state
->rb_node
;
887 * this search will find all the extents that end after
890 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
892 prealloc
= alloc_extent_state_atomic(prealloc
);
894 err
= insert_state(tree
, prealloc
, start
, end
,
895 &p
, &parent
, &bits
, changeset
);
897 extent_io_tree_panic(tree
, err
);
899 cache_state(prealloc
, cached_state
);
903 state
= rb_entry(node
, struct extent_state
, rb_node
);
905 last_start
= state
->start
;
906 last_end
= state
->end
;
909 * | ---- desired range ---- |
912 * Just lock what we found and keep going
914 if (state
->start
== start
&& state
->end
<= end
) {
915 if (state
->state
& exclusive_bits
) {
916 *failed_start
= state
->start
;
921 set_state_bits(tree
, state
, &bits
, changeset
);
922 cache_state(state
, cached_state
);
923 merge_state(tree
, state
);
924 if (last_end
== (u64
)-1)
926 start
= last_end
+ 1;
927 state
= next_state(state
);
928 if (start
< end
&& state
&& state
->start
== start
&&
935 * | ---- desired range ---- |
938 * | ------------- state -------------- |
940 * We need to split the extent we found, and may flip bits on
943 * If the extent we found extends past our
944 * range, we just split and search again. It'll get split
945 * again the next time though.
947 * If the extent we found is inside our range, we set the
950 if (state
->start
< start
) {
951 if (state
->state
& exclusive_bits
) {
952 *failed_start
= start
;
957 prealloc
= alloc_extent_state_atomic(prealloc
);
959 err
= split_state(tree
, state
, prealloc
, start
);
961 extent_io_tree_panic(tree
, err
);
966 if (state
->end
<= end
) {
967 set_state_bits(tree
, state
, &bits
, changeset
);
968 cache_state(state
, cached_state
);
969 merge_state(tree
, state
);
970 if (last_end
== (u64
)-1)
972 start
= last_end
+ 1;
973 state
= next_state(state
);
974 if (start
< end
&& state
&& state
->start
== start
&&
981 * | ---- desired range ---- |
982 * | state | or | state |
984 * There's a hole, we need to insert something in it and
985 * ignore the extent we found.
987 if (state
->start
> start
) {
989 if (end
< last_start
)
992 this_end
= last_start
- 1;
994 prealloc
= alloc_extent_state_atomic(prealloc
);
998 * Avoid to free 'prealloc' if it can be merged with
1001 err
= insert_state(tree
, prealloc
, start
, this_end
,
1002 NULL
, NULL
, &bits
, changeset
);
1004 extent_io_tree_panic(tree
, err
);
1006 cache_state(prealloc
, cached_state
);
1008 start
= this_end
+ 1;
1012 * | ---- desired range ---- |
1014 * We need to split the extent, and set the bit
1017 if (state
->start
<= end
&& state
->end
> end
) {
1018 if (state
->state
& exclusive_bits
) {
1019 *failed_start
= start
;
1024 prealloc
= alloc_extent_state_atomic(prealloc
);
1026 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1028 extent_io_tree_panic(tree
, err
);
1030 set_state_bits(tree
, prealloc
, &bits
, changeset
);
1031 cache_state(prealloc
, cached_state
);
1032 merge_state(tree
, prealloc
);
1040 spin_unlock(&tree
->lock
);
1041 if (gfpflags_allow_blocking(mask
))
1046 spin_unlock(&tree
->lock
);
1048 free_extent_state(prealloc
);
1054 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1055 unsigned bits
, u64
* failed_start
,
1056 struct extent_state
**cached_state
, gfp_t mask
)
1058 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1059 cached_state
, mask
, NULL
);
1064 * convert_extent_bit - convert all bits in a given range from one bit to
1066 * @tree: the io tree to search
1067 * @start: the start offset in bytes
1068 * @end: the end offset in bytes (inclusive)
1069 * @bits: the bits to set in this range
1070 * @clear_bits: the bits to clear in this range
1071 * @cached_state: state that we're going to cache
1073 * This will go through and set bits for the given range. If any states exist
1074 * already in this range they are set with the given bit and cleared of the
1075 * clear_bits. This is only meant to be used by things that are mergeable, ie
1076 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1077 * boundary bits like LOCK.
1079 * All allocations are done with GFP_NOFS.
1081 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1082 unsigned bits
, unsigned clear_bits
,
1083 struct extent_state
**cached_state
)
1085 struct extent_state
*state
;
1086 struct extent_state
*prealloc
= NULL
;
1087 struct rb_node
*node
;
1089 struct rb_node
*parent
;
1093 bool first_iteration
= true;
1095 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1100 * Best effort, don't worry if extent state allocation fails
1101 * here for the first iteration. We might have a cached state
1102 * that matches exactly the target range, in which case no
1103 * extent state allocations are needed. We'll only know this
1104 * after locking the tree.
1106 prealloc
= alloc_extent_state(GFP_NOFS
);
1107 if (!prealloc
&& !first_iteration
)
1111 spin_lock(&tree
->lock
);
1112 if (cached_state
&& *cached_state
) {
1113 state
= *cached_state
;
1114 if (state
->start
<= start
&& state
->end
> start
&&
1115 extent_state_in_tree(state
)) {
1116 node
= &state
->rb_node
;
1122 * this search will find all the extents that end after
1125 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1127 prealloc
= alloc_extent_state_atomic(prealloc
);
1132 err
= insert_state(tree
, prealloc
, start
, end
,
1133 &p
, &parent
, &bits
, NULL
);
1135 extent_io_tree_panic(tree
, err
);
1136 cache_state(prealloc
, cached_state
);
1140 state
= rb_entry(node
, struct extent_state
, rb_node
);
1142 last_start
= state
->start
;
1143 last_end
= state
->end
;
1146 * | ---- desired range ---- |
1149 * Just lock what we found and keep going
1151 if (state
->start
== start
&& state
->end
<= end
) {
1152 set_state_bits(tree
, state
, &bits
, NULL
);
1153 cache_state(state
, cached_state
);
1154 state
= clear_state_bit(tree
, state
, &clear_bits
, 0, NULL
);
1155 if (last_end
== (u64
)-1)
1157 start
= last_end
+ 1;
1158 if (start
< end
&& state
&& state
->start
== start
&&
1165 * | ---- desired range ---- |
1168 * | ------------- state -------------- |
1170 * We need to split the extent we found, and may flip bits on
1173 * If the extent we found extends past our
1174 * range, we just split and search again. It'll get split
1175 * again the next time though.
1177 * If the extent we found is inside our range, we set the
1178 * desired bit on it.
1180 if (state
->start
< start
) {
1181 prealloc
= alloc_extent_state_atomic(prealloc
);
1186 err
= split_state(tree
, state
, prealloc
, start
);
1188 extent_io_tree_panic(tree
, err
);
1192 if (state
->end
<= end
) {
1193 set_state_bits(tree
, state
, &bits
, NULL
);
1194 cache_state(state
, cached_state
);
1195 state
= clear_state_bit(tree
, state
, &clear_bits
, 0,
1197 if (last_end
== (u64
)-1)
1199 start
= last_end
+ 1;
1200 if (start
< end
&& state
&& state
->start
== start
&&
1207 * | ---- desired range ---- |
1208 * | state | or | state |
1210 * There's a hole, we need to insert something in it and
1211 * ignore the extent we found.
1213 if (state
->start
> start
) {
1215 if (end
< last_start
)
1218 this_end
= last_start
- 1;
1220 prealloc
= alloc_extent_state_atomic(prealloc
);
1227 * Avoid to free 'prealloc' if it can be merged with
1230 err
= insert_state(tree
, prealloc
, start
, this_end
,
1231 NULL
, NULL
, &bits
, NULL
);
1233 extent_io_tree_panic(tree
, err
);
1234 cache_state(prealloc
, cached_state
);
1236 start
= this_end
+ 1;
1240 * | ---- desired range ---- |
1242 * We need to split the extent, and set the bit
1245 if (state
->start
<= end
&& state
->end
> end
) {
1246 prealloc
= alloc_extent_state_atomic(prealloc
);
1252 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1254 extent_io_tree_panic(tree
, err
);
1256 set_state_bits(tree
, prealloc
, &bits
, NULL
);
1257 cache_state(prealloc
, cached_state
);
1258 clear_state_bit(tree
, prealloc
, &clear_bits
, 0, NULL
);
1266 spin_unlock(&tree
->lock
);
1268 first_iteration
= false;
1272 spin_unlock(&tree
->lock
);
1274 free_extent_state(prealloc
);
1279 /* wrappers around set/clear extent bit */
1280 int set_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1281 unsigned bits
, struct extent_changeset
*changeset
)
1284 * We don't support EXTENT_LOCKED yet, as current changeset will
1285 * record any bits changed, so for EXTENT_LOCKED case, it will
1286 * either fail with -EEXIST or changeset will record the whole
1289 BUG_ON(bits
& EXTENT_LOCKED
);
1291 return __set_extent_bit(tree
, start
, end
, bits
, 0, NULL
, NULL
, GFP_NOFS
,
1295 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1296 unsigned bits
, int wake
, int delete,
1297 struct extent_state
**cached
, gfp_t mask
)
1299 return __clear_extent_bit(tree
, start
, end
, bits
, wake
, delete,
1300 cached
, mask
, NULL
);
1303 int clear_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1304 unsigned bits
, struct extent_changeset
*changeset
)
1307 * Don't support EXTENT_LOCKED case, same reason as
1308 * set_record_extent_bits().
1310 BUG_ON(bits
& EXTENT_LOCKED
);
1312 return __clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, GFP_NOFS
,
1317 * either insert or lock state struct between start and end use mask to tell
1318 * us if waiting is desired.
1320 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1321 struct extent_state
**cached_state
)
1327 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
,
1328 EXTENT_LOCKED
, &failed_start
,
1329 cached_state
, GFP_NOFS
, NULL
);
1330 if (err
== -EEXIST
) {
1331 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1332 start
= failed_start
;
1335 WARN_ON(start
> end
);
1340 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1345 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1346 &failed_start
, NULL
, GFP_NOFS
, NULL
);
1347 if (err
== -EEXIST
) {
1348 if (failed_start
> start
)
1349 clear_extent_bit(tree
, start
, failed_start
- 1,
1350 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1356 void extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1358 unsigned long index
= start
>> PAGE_SHIFT
;
1359 unsigned long end_index
= end
>> PAGE_SHIFT
;
1362 while (index
<= end_index
) {
1363 page
= find_get_page(inode
->i_mapping
, index
);
1364 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1365 clear_page_dirty_for_io(page
);
1371 void extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1373 unsigned long index
= start
>> PAGE_SHIFT
;
1374 unsigned long end_index
= end
>> PAGE_SHIFT
;
1377 while (index
<= end_index
) {
1378 page
= find_get_page(inode
->i_mapping
, index
);
1379 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1380 __set_page_dirty_nobuffers(page
);
1381 account_page_redirty(page
);
1388 * helper function to set both pages and extents in the tree writeback
1390 static void set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1392 tree
->ops
->set_range_writeback(tree
->private_data
, start
, end
);
1395 /* find the first state struct with 'bits' set after 'start', and
1396 * return it. tree->lock must be held. NULL will returned if
1397 * nothing was found after 'start'
1399 static struct extent_state
*
1400 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1401 u64 start
, unsigned bits
)
1403 struct rb_node
*node
;
1404 struct extent_state
*state
;
1407 * this search will find all the extents that end after
1410 node
= tree_search(tree
, start
);
1415 state
= rb_entry(node
, struct extent_state
, rb_node
);
1416 if (state
->end
>= start
&& (state
->state
& bits
))
1419 node
= rb_next(node
);
1428 * find the first offset in the io tree with 'bits' set. zero is
1429 * returned if we find something, and *start_ret and *end_ret are
1430 * set to reflect the state struct that was found.
1432 * If nothing was found, 1 is returned. If found something, return 0.
1434 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1435 u64
*start_ret
, u64
*end_ret
, unsigned bits
,
1436 struct extent_state
**cached_state
)
1438 struct extent_state
*state
;
1442 spin_lock(&tree
->lock
);
1443 if (cached_state
&& *cached_state
) {
1444 state
= *cached_state
;
1445 if (state
->end
== start
- 1 && extent_state_in_tree(state
)) {
1446 n
= rb_next(&state
->rb_node
);
1448 state
= rb_entry(n
, struct extent_state
,
1450 if (state
->state
& bits
)
1454 free_extent_state(*cached_state
);
1455 *cached_state
= NULL
;
1458 free_extent_state(*cached_state
);
1459 *cached_state
= NULL
;
1462 state
= find_first_extent_bit_state(tree
, start
, bits
);
1465 cache_state_if_flags(state
, cached_state
, 0);
1466 *start_ret
= state
->start
;
1467 *end_ret
= state
->end
;
1471 spin_unlock(&tree
->lock
);
1476 * find a contiguous range of bytes in the file marked as delalloc, not
1477 * more than 'max_bytes'. start and end are used to return the range,
1479 * 1 is returned if we find something, 0 if nothing was in the tree
1481 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1482 u64
*start
, u64
*end
, u64 max_bytes
,
1483 struct extent_state
**cached_state
)
1485 struct rb_node
*node
;
1486 struct extent_state
*state
;
1487 u64 cur_start
= *start
;
1489 u64 total_bytes
= 0;
1491 spin_lock(&tree
->lock
);
1494 * this search will find all the extents that end after
1497 node
= tree_search(tree
, cur_start
);
1505 state
= rb_entry(node
, struct extent_state
, rb_node
);
1506 if (found
&& (state
->start
!= cur_start
||
1507 (state
->state
& EXTENT_BOUNDARY
))) {
1510 if (!(state
->state
& EXTENT_DELALLOC
)) {
1516 *start
= state
->start
;
1517 *cached_state
= state
;
1518 refcount_inc(&state
->refs
);
1522 cur_start
= state
->end
+ 1;
1523 node
= rb_next(node
);
1524 total_bytes
+= state
->end
- state
->start
+ 1;
1525 if (total_bytes
>= max_bytes
)
1531 spin_unlock(&tree
->lock
);
1535 static int __process_pages_contig(struct address_space
*mapping
,
1536 struct page
*locked_page
,
1537 pgoff_t start_index
, pgoff_t end_index
,
1538 unsigned long page_ops
, pgoff_t
*index_ret
);
1540 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1541 struct page
*locked_page
,
1544 unsigned long index
= start
>> PAGE_SHIFT
;
1545 unsigned long end_index
= end
>> PAGE_SHIFT
;
1547 ASSERT(locked_page
);
1548 if (index
== locked_page
->index
&& end_index
== index
)
1551 __process_pages_contig(inode
->i_mapping
, locked_page
, index
, end_index
,
1555 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1556 struct page
*locked_page
,
1560 unsigned long index
= delalloc_start
>> PAGE_SHIFT
;
1561 unsigned long index_ret
= index
;
1562 unsigned long end_index
= delalloc_end
>> PAGE_SHIFT
;
1565 ASSERT(locked_page
);
1566 if (index
== locked_page
->index
&& index
== end_index
)
1569 ret
= __process_pages_contig(inode
->i_mapping
, locked_page
, index
,
1570 end_index
, PAGE_LOCK
, &index_ret
);
1572 __unlock_for_delalloc(inode
, locked_page
, delalloc_start
,
1573 (u64
)index_ret
<< PAGE_SHIFT
);
1578 * find a contiguous range of bytes in the file marked as delalloc, not
1579 * more than 'max_bytes'. start and end are used to return the range,
1581 * 1 is returned if we find something, 0 if nothing was in the tree
1583 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1584 struct extent_io_tree
*tree
,
1585 struct page
*locked_page
, u64
*start
,
1586 u64
*end
, u64 max_bytes
)
1591 struct extent_state
*cached_state
= NULL
;
1596 /* step one, find a bunch of delalloc bytes starting at start */
1597 delalloc_start
= *start
;
1599 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1600 max_bytes
, &cached_state
);
1601 if (!found
|| delalloc_end
<= *start
) {
1602 *start
= delalloc_start
;
1603 *end
= delalloc_end
;
1604 free_extent_state(cached_state
);
1609 * start comes from the offset of locked_page. We have to lock
1610 * pages in order, so we can't process delalloc bytes before
1613 if (delalloc_start
< *start
)
1614 delalloc_start
= *start
;
1617 * make sure to limit the number of pages we try to lock down
1619 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1620 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1622 /* step two, lock all the pages after the page that has start */
1623 ret
= lock_delalloc_pages(inode
, locked_page
,
1624 delalloc_start
, delalloc_end
);
1625 if (ret
== -EAGAIN
) {
1626 /* some of the pages are gone, lets avoid looping by
1627 * shortening the size of the delalloc range we're searching
1629 free_extent_state(cached_state
);
1630 cached_state
= NULL
;
1632 max_bytes
= PAGE_SIZE
;
1640 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1642 /* step three, lock the state bits for the whole range */
1643 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, &cached_state
);
1645 /* then test to make sure it is all still delalloc */
1646 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1647 EXTENT_DELALLOC
, 1, cached_state
);
1649 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1650 &cached_state
, GFP_NOFS
);
1651 __unlock_for_delalloc(inode
, locked_page
,
1652 delalloc_start
, delalloc_end
);
1656 free_extent_state(cached_state
);
1657 *start
= delalloc_start
;
1658 *end
= delalloc_end
;
1663 static int __process_pages_contig(struct address_space
*mapping
,
1664 struct page
*locked_page
,
1665 pgoff_t start_index
, pgoff_t end_index
,
1666 unsigned long page_ops
, pgoff_t
*index_ret
)
1668 unsigned long nr_pages
= end_index
- start_index
+ 1;
1669 unsigned long pages_locked
= 0;
1670 pgoff_t index
= start_index
;
1671 struct page
*pages
[16];
1676 if (page_ops
& PAGE_LOCK
) {
1677 ASSERT(page_ops
== PAGE_LOCK
);
1678 ASSERT(index_ret
&& *index_ret
== start_index
);
1681 if ((page_ops
& PAGE_SET_ERROR
) && nr_pages
> 0)
1682 mapping_set_error(mapping
, -EIO
);
1684 while (nr_pages
> 0) {
1685 ret
= find_get_pages_contig(mapping
, index
,
1686 min_t(unsigned long,
1687 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1690 * Only if we're going to lock these pages,
1691 * can we find nothing at @index.
1693 ASSERT(page_ops
& PAGE_LOCK
);
1698 for (i
= 0; i
< ret
; i
++) {
1699 if (page_ops
& PAGE_SET_PRIVATE2
)
1700 SetPagePrivate2(pages
[i
]);
1702 if (pages
[i
] == locked_page
) {
1707 if (page_ops
& PAGE_CLEAR_DIRTY
)
1708 clear_page_dirty_for_io(pages
[i
]);
1709 if (page_ops
& PAGE_SET_WRITEBACK
)
1710 set_page_writeback(pages
[i
]);
1711 if (page_ops
& PAGE_SET_ERROR
)
1712 SetPageError(pages
[i
]);
1713 if (page_ops
& PAGE_END_WRITEBACK
)
1714 end_page_writeback(pages
[i
]);
1715 if (page_ops
& PAGE_UNLOCK
)
1716 unlock_page(pages
[i
]);
1717 if (page_ops
& PAGE_LOCK
) {
1718 lock_page(pages
[i
]);
1719 if (!PageDirty(pages
[i
]) ||
1720 pages
[i
]->mapping
!= mapping
) {
1721 unlock_page(pages
[i
]);
1735 if (err
&& index_ret
)
1736 *index_ret
= start_index
+ pages_locked
- 1;
1740 void extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1741 u64 delalloc_end
, struct page
*locked_page
,
1742 unsigned clear_bits
,
1743 unsigned long page_ops
)
1745 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, clear_bits
, 1, 0,
1748 __process_pages_contig(inode
->i_mapping
, locked_page
,
1749 start
>> PAGE_SHIFT
, end
>> PAGE_SHIFT
,
1754 * count the number of bytes in the tree that have a given bit(s)
1755 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1756 * cached. The total number found is returned.
1758 u64
count_range_bits(struct extent_io_tree
*tree
,
1759 u64
*start
, u64 search_end
, u64 max_bytes
,
1760 unsigned bits
, int contig
)
1762 struct rb_node
*node
;
1763 struct extent_state
*state
;
1764 u64 cur_start
= *start
;
1765 u64 total_bytes
= 0;
1769 if (WARN_ON(search_end
<= cur_start
))
1772 spin_lock(&tree
->lock
);
1773 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1774 total_bytes
= tree
->dirty_bytes
;
1778 * this search will find all the extents that end after
1781 node
= tree_search(tree
, cur_start
);
1786 state
= rb_entry(node
, struct extent_state
, rb_node
);
1787 if (state
->start
> search_end
)
1789 if (contig
&& found
&& state
->start
> last
+ 1)
1791 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1792 total_bytes
+= min(search_end
, state
->end
) + 1 -
1793 max(cur_start
, state
->start
);
1794 if (total_bytes
>= max_bytes
)
1797 *start
= max(cur_start
, state
->start
);
1801 } else if (contig
&& found
) {
1804 node
= rb_next(node
);
1809 spin_unlock(&tree
->lock
);
1814 * set the private field for a given byte offset in the tree. If there isn't
1815 * an extent_state there already, this does nothing.
1817 static noinline
int set_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1818 struct io_failure_record
*failrec
)
1820 struct rb_node
*node
;
1821 struct extent_state
*state
;
1824 spin_lock(&tree
->lock
);
1826 * this search will find all the extents that end after
1829 node
= tree_search(tree
, start
);
1834 state
= rb_entry(node
, struct extent_state
, rb_node
);
1835 if (state
->start
!= start
) {
1839 state
->failrec
= failrec
;
1841 spin_unlock(&tree
->lock
);
1845 static noinline
int get_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1846 struct io_failure_record
**failrec
)
1848 struct rb_node
*node
;
1849 struct extent_state
*state
;
1852 spin_lock(&tree
->lock
);
1854 * this search will find all the extents that end after
1857 node
= tree_search(tree
, start
);
1862 state
= rb_entry(node
, struct extent_state
, rb_node
);
1863 if (state
->start
!= start
) {
1867 *failrec
= state
->failrec
;
1869 spin_unlock(&tree
->lock
);
1874 * searches a range in the state tree for a given mask.
1875 * If 'filled' == 1, this returns 1 only if every extent in the tree
1876 * has the bits set. Otherwise, 1 is returned if any bit in the
1877 * range is found set.
1879 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1880 unsigned bits
, int filled
, struct extent_state
*cached
)
1882 struct extent_state
*state
= NULL
;
1883 struct rb_node
*node
;
1886 spin_lock(&tree
->lock
);
1887 if (cached
&& extent_state_in_tree(cached
) && cached
->start
<= start
&&
1888 cached
->end
> start
)
1889 node
= &cached
->rb_node
;
1891 node
= tree_search(tree
, start
);
1892 while (node
&& start
<= end
) {
1893 state
= rb_entry(node
, struct extent_state
, rb_node
);
1895 if (filled
&& state
->start
> start
) {
1900 if (state
->start
> end
)
1903 if (state
->state
& bits
) {
1907 } else if (filled
) {
1912 if (state
->end
== (u64
)-1)
1915 start
= state
->end
+ 1;
1918 node
= rb_next(node
);
1925 spin_unlock(&tree
->lock
);
1930 * helper function to set a given page up to date if all the
1931 * extents in the tree for that page are up to date
1933 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1935 u64 start
= page_offset(page
);
1936 u64 end
= start
+ PAGE_SIZE
- 1;
1937 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1938 SetPageUptodate(page
);
1941 int free_io_failure(struct extent_io_tree
*failure_tree
,
1942 struct extent_io_tree
*io_tree
,
1943 struct io_failure_record
*rec
)
1948 set_state_failrec(failure_tree
, rec
->start
, NULL
);
1949 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1950 rec
->start
+ rec
->len
- 1,
1951 EXTENT_LOCKED
| EXTENT_DIRTY
);
1955 ret
= clear_extent_bits(io_tree
, rec
->start
,
1956 rec
->start
+ rec
->len
- 1,
1966 * this bypasses the standard btrfs submit functions deliberately, as
1967 * the standard behavior is to write all copies in a raid setup. here we only
1968 * want to write the one bad copy. so we do the mapping for ourselves and issue
1969 * submit_bio directly.
1970 * to avoid any synchronization issues, wait for the data after writing, which
1971 * actually prevents the read that triggered the error from finishing.
1972 * currently, there can be no more than two copies of every data bit. thus,
1973 * exactly one rewrite is required.
1975 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 ino
, u64 start
,
1976 u64 length
, u64 logical
, struct page
*page
,
1977 unsigned int pg_offset
, int mirror_num
)
1980 struct btrfs_device
*dev
;
1983 struct btrfs_bio
*bbio
= NULL
;
1986 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
1987 BUG_ON(!mirror_num
);
1989 bio
= btrfs_io_bio_alloc(1);
1990 bio
->bi_iter
.bi_size
= 0;
1991 map_length
= length
;
1994 * Avoid races with device replace and make sure our bbio has devices
1995 * associated to its stripes that don't go away while we are doing the
1996 * read repair operation.
1998 btrfs_bio_counter_inc_blocked(fs_info
);
1999 if (btrfs_is_parity_mirror(fs_info
, logical
, length
)) {
2001 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2002 * to update all raid stripes, but here we just want to correct
2003 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2004 * stripe's dev and sector.
2006 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_READ
, logical
,
2007 &map_length
, &bbio
, 0);
2009 btrfs_bio_counter_dec(fs_info
);
2013 ASSERT(bbio
->mirror_num
== 1);
2015 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_WRITE
, logical
,
2016 &map_length
, &bbio
, mirror_num
);
2018 btrfs_bio_counter_dec(fs_info
);
2022 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2025 sector
= bbio
->stripes
[bbio
->mirror_num
- 1].physical
>> 9;
2026 bio
->bi_iter
.bi_sector
= sector
;
2027 dev
= bbio
->stripes
[bbio
->mirror_num
- 1].dev
;
2028 btrfs_put_bbio(bbio
);
2029 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2030 btrfs_bio_counter_dec(fs_info
);
2034 bio_set_dev(bio
, dev
->bdev
);
2035 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
;
2036 bio_add_page(bio
, page
, length
, pg_offset
);
2038 if (btrfsic_submit_bio_wait(bio
)) {
2039 /* try to remap that extent elsewhere? */
2040 btrfs_bio_counter_dec(fs_info
);
2042 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2046 btrfs_info_rl_in_rcu(fs_info
,
2047 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2049 rcu_str_deref(dev
->name
), sector
);
2050 btrfs_bio_counter_dec(fs_info
);
2055 int repair_eb_io_failure(struct btrfs_fs_info
*fs_info
,
2056 struct extent_buffer
*eb
, int mirror_num
)
2058 u64 start
= eb
->start
;
2059 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2062 if (sb_rdonly(fs_info
->sb
))
2065 for (i
= 0; i
< num_pages
; i
++) {
2066 struct page
*p
= eb
->pages
[i
];
2068 ret
= repair_io_failure(fs_info
, 0, start
, PAGE_SIZE
, start
, p
,
2069 start
- page_offset(p
), mirror_num
);
2079 * each time an IO finishes, we do a fast check in the IO failure tree
2080 * to see if we need to process or clean up an io_failure_record
2082 int clean_io_failure(struct btrfs_fs_info
*fs_info
,
2083 struct extent_io_tree
*failure_tree
,
2084 struct extent_io_tree
*io_tree
, u64 start
,
2085 struct page
*page
, u64 ino
, unsigned int pg_offset
)
2088 struct io_failure_record
*failrec
;
2089 struct extent_state
*state
;
2094 ret
= count_range_bits(failure_tree
, &private, (u64
)-1, 1,
2099 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2103 BUG_ON(!failrec
->this_mirror
);
2105 if (failrec
->in_validation
) {
2106 /* there was no real error, just free the record */
2107 btrfs_debug(fs_info
,
2108 "clean_io_failure: freeing dummy error at %llu",
2112 if (sb_rdonly(fs_info
->sb
))
2115 spin_lock(&io_tree
->lock
);
2116 state
= find_first_extent_bit_state(io_tree
,
2119 spin_unlock(&io_tree
->lock
);
2121 if (state
&& state
->start
<= failrec
->start
&&
2122 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2123 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2125 if (num_copies
> 1) {
2126 repair_io_failure(fs_info
, ino
, start
, failrec
->len
,
2127 failrec
->logical
, page
, pg_offset
,
2128 failrec
->failed_mirror
);
2133 free_io_failure(failure_tree
, io_tree
, failrec
);
2139 * Can be called when
2140 * - hold extent lock
2141 * - under ordered extent
2142 * - the inode is freeing
2144 void btrfs_free_io_failure_record(struct btrfs_inode
*inode
, u64 start
, u64 end
)
2146 struct extent_io_tree
*failure_tree
= &inode
->io_failure_tree
;
2147 struct io_failure_record
*failrec
;
2148 struct extent_state
*state
, *next
;
2150 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2153 spin_lock(&failure_tree
->lock
);
2154 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2156 if (state
->start
> end
)
2159 ASSERT(state
->end
<= end
);
2161 next
= next_state(state
);
2163 failrec
= state
->failrec
;
2164 free_extent_state(state
);
2169 spin_unlock(&failure_tree
->lock
);
2172 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2173 struct io_failure_record
**failrec_ret
)
2175 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2176 struct io_failure_record
*failrec
;
2177 struct extent_map
*em
;
2178 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2179 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2180 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2184 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2186 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2190 failrec
->start
= start
;
2191 failrec
->len
= end
- start
+ 1;
2192 failrec
->this_mirror
= 0;
2193 failrec
->bio_flags
= 0;
2194 failrec
->in_validation
= 0;
2196 read_lock(&em_tree
->lock
);
2197 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2199 read_unlock(&em_tree
->lock
);
2204 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2205 free_extent_map(em
);
2208 read_unlock(&em_tree
->lock
);
2214 logical
= start
- em
->start
;
2215 logical
= em
->block_start
+ logical
;
2216 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2217 logical
= em
->block_start
;
2218 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2219 extent_set_compress_type(&failrec
->bio_flags
,
2223 btrfs_debug(fs_info
,
2224 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2225 logical
, start
, failrec
->len
);
2227 failrec
->logical
= logical
;
2228 free_extent_map(em
);
2230 /* set the bits in the private failure tree */
2231 ret
= set_extent_bits(failure_tree
, start
, end
,
2232 EXTENT_LOCKED
| EXTENT_DIRTY
);
2234 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2235 /* set the bits in the inode's tree */
2237 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2243 btrfs_debug(fs_info
,
2244 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2245 failrec
->logical
, failrec
->start
, failrec
->len
,
2246 failrec
->in_validation
);
2248 * when data can be on disk more than twice, add to failrec here
2249 * (e.g. with a list for failed_mirror) to make
2250 * clean_io_failure() clean all those errors at once.
2254 *failrec_ret
= failrec
;
2259 bool btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2260 struct io_failure_record
*failrec
, int failed_mirror
)
2262 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2265 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
2266 if (num_copies
== 1) {
2268 * we only have a single copy of the data, so don't bother with
2269 * all the retry and error correction code that follows. no
2270 * matter what the error is, it is very likely to persist.
2272 btrfs_debug(fs_info
,
2273 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2274 num_copies
, failrec
->this_mirror
, failed_mirror
);
2279 * there are two premises:
2280 * a) deliver good data to the caller
2281 * b) correct the bad sectors on disk
2283 if (failed_bio
->bi_vcnt
> 1) {
2285 * to fulfill b), we need to know the exact failing sectors, as
2286 * we don't want to rewrite any more than the failed ones. thus,
2287 * we need separate read requests for the failed bio
2289 * if the following BUG_ON triggers, our validation request got
2290 * merged. we need separate requests for our algorithm to work.
2292 BUG_ON(failrec
->in_validation
);
2293 failrec
->in_validation
= 1;
2294 failrec
->this_mirror
= failed_mirror
;
2297 * we're ready to fulfill a) and b) alongside. get a good copy
2298 * of the failed sector and if we succeed, we have setup
2299 * everything for repair_io_failure to do the rest for us.
2301 if (failrec
->in_validation
) {
2302 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2303 failrec
->in_validation
= 0;
2304 failrec
->this_mirror
= 0;
2306 failrec
->failed_mirror
= failed_mirror
;
2307 failrec
->this_mirror
++;
2308 if (failrec
->this_mirror
== failed_mirror
)
2309 failrec
->this_mirror
++;
2312 if (failrec
->this_mirror
> num_copies
) {
2313 btrfs_debug(fs_info
,
2314 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2315 num_copies
, failrec
->this_mirror
, failed_mirror
);
2323 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2324 struct io_failure_record
*failrec
,
2325 struct page
*page
, int pg_offset
, int icsum
,
2326 bio_end_io_t
*endio_func
, void *data
)
2328 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2330 struct btrfs_io_bio
*btrfs_failed_bio
;
2331 struct btrfs_io_bio
*btrfs_bio
;
2333 bio
= btrfs_io_bio_alloc(1);
2334 bio
->bi_end_io
= endio_func
;
2335 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2336 bio_set_dev(bio
, fs_info
->fs_devices
->latest_bdev
);
2337 bio
->bi_iter
.bi_size
= 0;
2338 bio
->bi_private
= data
;
2340 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2341 if (btrfs_failed_bio
->csum
) {
2342 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2344 btrfs_bio
= btrfs_io_bio(bio
);
2345 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2347 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2351 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2357 * this is a generic handler for readpage errors (default
2358 * readpage_io_failed_hook). if other copies exist, read those and write back
2359 * good data to the failed position. does not investigate in remapping the
2360 * failed extent elsewhere, hoping the device will be smart enough to do this as
2364 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2365 struct page
*page
, u64 start
, u64 end
,
2368 struct io_failure_record
*failrec
;
2369 struct inode
*inode
= page
->mapping
->host
;
2370 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2371 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2374 blk_status_t status
;
2377 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
2379 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2383 if (!btrfs_check_repairable(inode
, failed_bio
, failrec
,
2385 free_io_failure(failure_tree
, tree
, failrec
);
2389 if (failed_bio
->bi_vcnt
> 1)
2390 read_mode
|= REQ_FAILFAST_DEV
;
2392 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2393 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2394 start
- page_offset(page
),
2395 (int)phy_offset
, failed_bio
->bi_end_io
,
2397 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
2399 btrfs_debug(btrfs_sb(inode
->i_sb
),
2400 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2401 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2403 status
= tree
->ops
->submit_bio_hook(tree
->private_data
, bio
, failrec
->this_mirror
,
2404 failrec
->bio_flags
, 0);
2406 free_io_failure(failure_tree
, tree
, failrec
);
2408 ret
= blk_status_to_errno(status
);
2414 /* lots and lots of room for performance fixes in the end_bio funcs */
2416 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2418 int uptodate
= (err
== 0);
2419 struct extent_io_tree
*tree
;
2422 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2424 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2425 tree
->ops
->writepage_end_io_hook(page
, start
, end
, NULL
,
2429 ClearPageUptodate(page
);
2431 ret
= err
< 0 ? err
: -EIO
;
2432 mapping_set_error(page
->mapping
, ret
);
2437 * after a writepage IO is done, we need to:
2438 * clear the uptodate bits on error
2439 * clear the writeback bits in the extent tree for this IO
2440 * end_page_writeback if the page has no more pending IO
2442 * Scheduling is not allowed, so the extent state tree is expected
2443 * to have one and only one object corresponding to this IO.
2445 static void end_bio_extent_writepage(struct bio
*bio
)
2447 int error
= blk_status_to_errno(bio
->bi_status
);
2448 struct bio_vec
*bvec
;
2453 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
2454 bio_for_each_segment_all(bvec
, bio
, i
) {
2455 struct page
*page
= bvec
->bv_page
;
2456 struct inode
*inode
= page
->mapping
->host
;
2457 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2459 /* We always issue full-page reads, but if some block
2460 * in a page fails to read, blk_update_request() will
2461 * advance bv_offset and adjust bv_len to compensate.
2462 * Print a warning for nonzero offsets, and an error
2463 * if they don't add up to a full page. */
2464 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2465 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2467 "partial page write in btrfs with offset %u and length %u",
2468 bvec
->bv_offset
, bvec
->bv_len
);
2471 "incomplete page write in btrfs with offset %u and length %u",
2472 bvec
->bv_offset
, bvec
->bv_len
);
2475 start
= page_offset(page
);
2476 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2478 end_extent_writepage(page
, error
, start
, end
);
2479 end_page_writeback(page
);
2486 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2489 struct extent_state
*cached
= NULL
;
2490 u64 end
= start
+ len
- 1;
2492 if (uptodate
&& tree
->track_uptodate
)
2493 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2494 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2498 * after a readpage IO is done, we need to:
2499 * clear the uptodate bits on error
2500 * set the uptodate bits if things worked
2501 * set the page up to date if all extents in the tree are uptodate
2502 * clear the lock bit in the extent tree
2503 * unlock the page if there are no other extents locked for it
2505 * Scheduling is not allowed, so the extent state tree is expected
2506 * to have one and only one object corresponding to this IO.
2508 static void end_bio_extent_readpage(struct bio
*bio
)
2510 struct bio_vec
*bvec
;
2511 int uptodate
= !bio
->bi_status
;
2512 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2513 struct extent_io_tree
*tree
, *failure_tree
;
2518 u64 extent_start
= 0;
2524 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
2525 bio_for_each_segment_all(bvec
, bio
, i
) {
2526 struct page
*page
= bvec
->bv_page
;
2527 struct inode
*inode
= page
->mapping
->host
;
2528 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2530 btrfs_debug(fs_info
,
2531 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2532 (u64
)bio
->bi_iter
.bi_sector
, bio
->bi_status
,
2533 io_bio
->mirror_num
);
2534 tree
= &BTRFS_I(inode
)->io_tree
;
2535 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2537 /* We always issue full-page reads, but if some block
2538 * in a page fails to read, blk_update_request() will
2539 * advance bv_offset and adjust bv_len to compensate.
2540 * Print a warning for nonzero offsets, and an error
2541 * if they don't add up to a full page. */
2542 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2543 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2545 "partial page read in btrfs with offset %u and length %u",
2546 bvec
->bv_offset
, bvec
->bv_len
);
2549 "incomplete page read in btrfs with offset %u and length %u",
2550 bvec
->bv_offset
, bvec
->bv_len
);
2553 start
= page_offset(page
);
2554 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2557 mirror
= io_bio
->mirror_num
;
2558 if (likely(uptodate
&& tree
->ops
)) {
2559 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2565 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
,
2566 failure_tree
, tree
, start
,
2568 btrfs_ino(BTRFS_I(inode
)), 0);
2571 if (likely(uptodate
))
2575 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2576 if (ret
== -EAGAIN
) {
2578 * Data inode's readpage_io_failed_hook() always
2581 * The generic bio_readpage_error handles errors
2582 * the following way: If possible, new read
2583 * requests are created and submitted and will
2584 * end up in end_bio_extent_readpage as well (if
2585 * we're lucky, not in the !uptodate case). In
2586 * that case it returns 0 and we just go on with
2587 * the next page in our bio. If it can't handle
2588 * the error it will return -EIO and we remain
2589 * responsible for that page.
2591 ret
= bio_readpage_error(bio
, offset
, page
,
2592 start
, end
, mirror
);
2594 uptodate
= !bio
->bi_status
;
2601 * metadata's readpage_io_failed_hook() always returns
2602 * -EIO and fixes nothing. -EIO is also returned if
2603 * data inode error could not be fixed.
2605 ASSERT(ret
== -EIO
);
2608 if (likely(uptodate
)) {
2609 loff_t i_size
= i_size_read(inode
);
2610 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2613 /* Zero out the end if this page straddles i_size */
2614 off
= i_size
& (PAGE_SIZE
-1);
2615 if (page
->index
== end_index
&& off
)
2616 zero_user_segment(page
, off
, PAGE_SIZE
);
2617 SetPageUptodate(page
);
2619 ClearPageUptodate(page
);
2625 if (unlikely(!uptodate
)) {
2627 endio_readpage_release_extent(tree
,
2633 endio_readpage_release_extent(tree
, start
,
2634 end
- start
+ 1, 0);
2635 } else if (!extent_len
) {
2636 extent_start
= start
;
2637 extent_len
= end
+ 1 - start
;
2638 } else if (extent_start
+ extent_len
== start
) {
2639 extent_len
+= end
+ 1 - start
;
2641 endio_readpage_release_extent(tree
, extent_start
,
2642 extent_len
, uptodate
);
2643 extent_start
= start
;
2644 extent_len
= end
+ 1 - start
;
2649 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2652 io_bio
->end_io(io_bio
, blk_status_to_errno(bio
->bi_status
));
2657 * Initialize the members up to but not including 'bio'. Use after allocating a
2658 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2659 * 'bio' because use of __GFP_ZERO is not supported.
2661 static inline void btrfs_io_bio_init(struct btrfs_io_bio
*btrfs_bio
)
2663 memset(btrfs_bio
, 0, offsetof(struct btrfs_io_bio
, bio
));
2667 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2668 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2669 * for the appropriate container_of magic
2671 struct bio
*btrfs_bio_alloc(struct block_device
*bdev
, u64 first_byte
)
2675 bio
= bio_alloc_bioset(GFP_NOFS
, BIO_MAX_PAGES
, btrfs_bioset
);
2676 bio_set_dev(bio
, bdev
);
2677 bio
->bi_iter
.bi_sector
= first_byte
>> 9;
2678 btrfs_io_bio_init(btrfs_io_bio(bio
));
2682 struct bio
*btrfs_bio_clone(struct bio
*bio
)
2684 struct btrfs_io_bio
*btrfs_bio
;
2687 /* Bio allocation backed by a bioset does not fail */
2688 new = bio_clone_fast(bio
, GFP_NOFS
, btrfs_bioset
);
2689 btrfs_bio
= btrfs_io_bio(new);
2690 btrfs_io_bio_init(btrfs_bio
);
2691 btrfs_bio
->iter
= bio
->bi_iter
;
2695 struct bio
*btrfs_io_bio_alloc(unsigned int nr_iovecs
)
2699 /* Bio allocation backed by a bioset does not fail */
2700 bio
= bio_alloc_bioset(GFP_NOFS
, nr_iovecs
, btrfs_bioset
);
2701 btrfs_io_bio_init(btrfs_io_bio(bio
));
2705 struct bio
*btrfs_bio_clone_partial(struct bio
*orig
, int offset
, int size
)
2708 struct btrfs_io_bio
*btrfs_bio
;
2710 /* this will never fail when it's backed by a bioset */
2711 bio
= bio_clone_fast(orig
, GFP_NOFS
, btrfs_bioset
);
2714 btrfs_bio
= btrfs_io_bio(bio
);
2715 btrfs_io_bio_init(btrfs_bio
);
2717 bio_trim(bio
, offset
>> 9, size
>> 9);
2718 btrfs_bio
->iter
= bio
->bi_iter
;
2722 static int __must_check
submit_one_bio(struct bio
*bio
, int mirror_num
,
2723 unsigned long bio_flags
)
2725 blk_status_t ret
= 0;
2726 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2727 struct page
*page
= bvec
->bv_page
;
2728 struct extent_io_tree
*tree
= bio
->bi_private
;
2731 start
= page_offset(page
) + bvec
->bv_offset
;
2733 bio
->bi_private
= NULL
;
2737 ret
= tree
->ops
->submit_bio_hook(tree
->private_data
, bio
,
2738 mirror_num
, bio_flags
, start
);
2740 btrfsic_submit_bio(bio
);
2743 return blk_status_to_errno(ret
);
2746 static int merge_bio(struct extent_io_tree
*tree
, struct page
*page
,
2747 unsigned long offset
, size_t size
, struct bio
*bio
,
2748 unsigned long bio_flags
)
2752 ret
= tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
,
2759 * @opf: bio REQ_OP_* and REQ_* flags as one value
2761 static int submit_extent_page(unsigned int opf
, struct extent_io_tree
*tree
,
2762 struct writeback_control
*wbc
,
2763 struct page
*page
, u64 offset
,
2764 size_t size
, unsigned long pg_offset
,
2765 struct block_device
*bdev
,
2766 struct bio
**bio_ret
,
2767 bio_end_io_t end_io_func
,
2769 unsigned long prev_bio_flags
,
2770 unsigned long bio_flags
,
2771 bool force_bio_submit
)
2776 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2777 size_t page_size
= min_t(size_t, size
, PAGE_SIZE
);
2778 sector_t sector
= offset
>> 9;
2780 if (bio_ret
&& *bio_ret
) {
2783 contig
= bio
->bi_iter
.bi_sector
== sector
;
2785 contig
= bio_end_sector(bio
) == sector
;
2787 if (prev_bio_flags
!= bio_flags
|| !contig
||
2789 merge_bio(tree
, page
, pg_offset
, page_size
, bio
, bio_flags
) ||
2790 bio_add_page(bio
, page
, page_size
, pg_offset
) < page_size
) {
2791 ret
= submit_one_bio(bio
, mirror_num
, prev_bio_flags
);
2799 wbc_account_io(wbc
, page
, page_size
);
2804 bio
= btrfs_bio_alloc(bdev
, offset
);
2805 bio_add_page(bio
, page
, page_size
, pg_offset
);
2806 bio
->bi_end_io
= end_io_func
;
2807 bio
->bi_private
= tree
;
2808 bio
->bi_write_hint
= page
->mapping
->host
->i_write_hint
;
2811 wbc_init_bio(wbc
, bio
);
2812 wbc_account_io(wbc
, page
, page_size
);
2818 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
2823 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2826 if (!PagePrivate(page
)) {
2827 SetPagePrivate(page
);
2829 set_page_private(page
, (unsigned long)eb
);
2831 WARN_ON(page
->private != (unsigned long)eb
);
2835 void set_page_extent_mapped(struct page
*page
)
2837 if (!PagePrivate(page
)) {
2838 SetPagePrivate(page
);
2840 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2844 static struct extent_map
*
2845 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2846 u64 start
, u64 len
, get_extent_t
*get_extent
,
2847 struct extent_map
**em_cached
)
2849 struct extent_map
*em
;
2851 if (em_cached
&& *em_cached
) {
2853 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2854 start
< extent_map_end(em
)) {
2855 refcount_inc(&em
->refs
);
2859 free_extent_map(em
);
2863 em
= get_extent(BTRFS_I(inode
), page
, pg_offset
, start
, len
, 0);
2864 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2866 refcount_inc(&em
->refs
);
2872 * basic readpage implementation. Locked extent state structs are inserted
2873 * into the tree that are removed when the IO is done (by the end_io
2875 * XXX JDM: This needs looking at to ensure proper page locking
2876 * return 0 on success, otherwise return error
2878 static int __do_readpage(struct extent_io_tree
*tree
,
2880 get_extent_t
*get_extent
,
2881 struct extent_map
**em_cached
,
2882 struct bio
**bio
, int mirror_num
,
2883 unsigned long *bio_flags
, unsigned int read_flags
,
2886 struct inode
*inode
= page
->mapping
->host
;
2887 u64 start
= page_offset(page
);
2888 u64 page_end
= start
+ PAGE_SIZE
- 1;
2892 u64 last_byte
= i_size_read(inode
);
2895 struct extent_map
*em
;
2896 struct block_device
*bdev
;
2899 size_t pg_offset
= 0;
2901 size_t disk_io_size
;
2902 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2903 unsigned long this_bio_flag
= 0;
2905 set_page_extent_mapped(page
);
2908 if (!PageUptodate(page
)) {
2909 if (cleancache_get_page(page
) == 0) {
2910 BUG_ON(blocksize
!= PAGE_SIZE
);
2911 unlock_extent(tree
, start
, end
);
2916 if (page
->index
== last_byte
>> PAGE_SHIFT
) {
2918 size_t zero_offset
= last_byte
& (PAGE_SIZE
- 1);
2921 iosize
= PAGE_SIZE
- zero_offset
;
2922 userpage
= kmap_atomic(page
);
2923 memset(userpage
+ zero_offset
, 0, iosize
);
2924 flush_dcache_page(page
);
2925 kunmap_atomic(userpage
);
2928 while (cur
<= end
) {
2929 bool force_bio_submit
= false;
2932 if (cur
>= last_byte
) {
2934 struct extent_state
*cached
= NULL
;
2936 iosize
= PAGE_SIZE
- pg_offset
;
2937 userpage
= kmap_atomic(page
);
2938 memset(userpage
+ pg_offset
, 0, iosize
);
2939 flush_dcache_page(page
);
2940 kunmap_atomic(userpage
);
2941 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2943 unlock_extent_cached(tree
, cur
,
2948 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2949 end
- cur
+ 1, get_extent
, em_cached
);
2950 if (IS_ERR_OR_NULL(em
)) {
2952 unlock_extent(tree
, cur
, end
);
2955 extent_offset
= cur
- em
->start
;
2956 BUG_ON(extent_map_end(em
) <= cur
);
2959 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2960 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2961 extent_set_compress_type(&this_bio_flag
,
2965 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2966 cur_end
= min(extent_map_end(em
) - 1, end
);
2967 iosize
= ALIGN(iosize
, blocksize
);
2968 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2969 disk_io_size
= em
->block_len
;
2970 offset
= em
->block_start
;
2972 offset
= em
->block_start
+ extent_offset
;
2973 disk_io_size
= iosize
;
2976 block_start
= em
->block_start
;
2977 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2978 block_start
= EXTENT_MAP_HOLE
;
2981 * If we have a file range that points to a compressed extent
2982 * and it's followed by a consecutive file range that points to
2983 * to the same compressed extent (possibly with a different
2984 * offset and/or length, so it either points to the whole extent
2985 * or only part of it), we must make sure we do not submit a
2986 * single bio to populate the pages for the 2 ranges because
2987 * this makes the compressed extent read zero out the pages
2988 * belonging to the 2nd range. Imagine the following scenario:
2991 * [0 - 8K] [8K - 24K]
2994 * points to extent X, points to extent X,
2995 * offset 4K, length of 8K offset 0, length 16K
2997 * [extent X, compressed length = 4K uncompressed length = 16K]
2999 * If the bio to read the compressed extent covers both ranges,
3000 * it will decompress extent X into the pages belonging to the
3001 * first range and then it will stop, zeroing out the remaining
3002 * pages that belong to the other range that points to extent X.
3003 * So here we make sure we submit 2 bios, one for the first
3004 * range and another one for the third range. Both will target
3005 * the same physical extent from disk, but we can't currently
3006 * make the compressed bio endio callback populate the pages
3007 * for both ranges because each compressed bio is tightly
3008 * coupled with a single extent map, and each range can have
3009 * an extent map with a different offset value relative to the
3010 * uncompressed data of our extent and different lengths. This
3011 * is a corner case so we prioritize correctness over
3012 * non-optimal behavior (submitting 2 bios for the same extent).
3014 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3015 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3016 *prev_em_start
!= em
->orig_start
)
3017 force_bio_submit
= true;
3020 *prev_em_start
= em
->orig_start
;
3022 free_extent_map(em
);
3025 /* we've found a hole, just zero and go on */
3026 if (block_start
== EXTENT_MAP_HOLE
) {
3028 struct extent_state
*cached
= NULL
;
3030 userpage
= kmap_atomic(page
);
3031 memset(userpage
+ pg_offset
, 0, iosize
);
3032 flush_dcache_page(page
);
3033 kunmap_atomic(userpage
);
3035 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3037 unlock_extent_cached(tree
, cur
,
3041 pg_offset
+= iosize
;
3044 /* the get_extent function already copied into the page */
3045 if (test_range_bit(tree
, cur
, cur_end
,
3046 EXTENT_UPTODATE
, 1, NULL
)) {
3047 check_page_uptodate(tree
, page
);
3048 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3050 pg_offset
+= iosize
;
3053 /* we have an inline extent but it didn't get marked up
3054 * to date. Error out
3056 if (block_start
== EXTENT_MAP_INLINE
) {
3058 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3060 pg_offset
+= iosize
;
3064 ret
= submit_extent_page(REQ_OP_READ
| read_flags
, tree
, NULL
,
3065 page
, offset
, disk_io_size
,
3066 pg_offset
, bdev
, bio
,
3067 end_bio_extent_readpage
, mirror_num
,
3073 *bio_flags
= this_bio_flag
;
3076 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3080 pg_offset
+= iosize
;
3084 if (!PageError(page
))
3085 SetPageUptodate(page
);
3091 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3092 struct page
*pages
[], int nr_pages
,
3094 get_extent_t
*get_extent
,
3095 struct extent_map
**em_cached
,
3096 struct bio
**bio
, int mirror_num
,
3097 unsigned long *bio_flags
,
3100 struct inode
*inode
;
3101 struct btrfs_ordered_extent
*ordered
;
3104 inode
= pages
[0]->mapping
->host
;
3106 lock_extent(tree
, start
, end
);
3107 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
3111 unlock_extent(tree
, start
, end
);
3112 btrfs_start_ordered_extent(inode
, ordered
, 1);
3113 btrfs_put_ordered_extent(ordered
);
3116 for (index
= 0; index
< nr_pages
; index
++) {
3117 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3118 mirror_num
, bio_flags
, 0, prev_em_start
);
3119 put_page(pages
[index
]);
3123 static void __extent_readpages(struct extent_io_tree
*tree
,
3124 struct page
*pages
[],
3125 int nr_pages
, get_extent_t
*get_extent
,
3126 struct extent_map
**em_cached
,
3127 struct bio
**bio
, int mirror_num
,
3128 unsigned long *bio_flags
,
3135 int first_index
= 0;
3137 for (index
= 0; index
< nr_pages
; index
++) {
3138 page_start
= page_offset(pages
[index
]);
3141 end
= start
+ PAGE_SIZE
- 1;
3142 first_index
= index
;
3143 } else if (end
+ 1 == page_start
) {
3146 __do_contiguous_readpages(tree
, &pages
[first_index
],
3147 index
- first_index
, start
,
3148 end
, get_extent
, em_cached
,
3149 bio
, mirror_num
, bio_flags
,
3152 end
= start
+ PAGE_SIZE
- 1;
3153 first_index
= index
;
3158 __do_contiguous_readpages(tree
, &pages
[first_index
],
3159 index
- first_index
, start
,
3160 end
, get_extent
, em_cached
, bio
,
3161 mirror_num
, bio_flags
,
3165 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3167 get_extent_t
*get_extent
,
3168 struct bio
**bio
, int mirror_num
,
3169 unsigned long *bio_flags
,
3170 unsigned int read_flags
)
3172 struct inode
*inode
= page
->mapping
->host
;
3173 struct btrfs_ordered_extent
*ordered
;
3174 u64 start
= page_offset(page
);
3175 u64 end
= start
+ PAGE_SIZE
- 1;
3179 lock_extent(tree
, start
, end
);
3180 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
3184 unlock_extent(tree
, start
, end
);
3185 btrfs_start_ordered_extent(inode
, ordered
, 1);
3186 btrfs_put_ordered_extent(ordered
);
3189 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3190 bio_flags
, read_flags
, NULL
);
3194 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3195 get_extent_t
*get_extent
, int mirror_num
)
3197 struct bio
*bio
= NULL
;
3198 unsigned long bio_flags
= 0;
3201 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3204 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
3208 static void update_nr_written(struct writeback_control
*wbc
,
3209 unsigned long nr_written
)
3211 wbc
->nr_to_write
-= nr_written
;
3215 * helper for __extent_writepage, doing all of the delayed allocation setup.
3217 * This returns 1 if our fill_delalloc function did all the work required
3218 * to write the page (copy into inline extent). In this case the IO has
3219 * been started and the page is already unlocked.
3221 * This returns 0 if all went well (page still locked)
3222 * This returns < 0 if there were errors (page still locked)
3224 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3225 struct page
*page
, struct writeback_control
*wbc
,
3226 struct extent_page_data
*epd
,
3228 unsigned long *nr_written
)
3230 struct extent_io_tree
*tree
= epd
->tree
;
3231 u64 page_end
= delalloc_start
+ PAGE_SIZE
- 1;
3233 u64 delalloc_to_write
= 0;
3234 u64 delalloc_end
= 0;
3236 int page_started
= 0;
3238 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3241 while (delalloc_end
< page_end
) {
3242 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3246 BTRFS_MAX_EXTENT_SIZE
);
3247 if (nr_delalloc
== 0) {
3248 delalloc_start
= delalloc_end
+ 1;
3251 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3256 /* File system has been set read-only */
3259 /* fill_delalloc should be return < 0 for error
3260 * but just in case, we use > 0 here meaning the
3261 * IO is started, so we don't want to return > 0
3262 * unless things are going well.
3264 ret
= ret
< 0 ? ret
: -EIO
;
3268 * delalloc_end is already one less than the total length, so
3269 * we don't subtract one from PAGE_SIZE
3271 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3272 PAGE_SIZE
) >> PAGE_SHIFT
;
3273 delalloc_start
= delalloc_end
+ 1;
3275 if (wbc
->nr_to_write
< delalloc_to_write
) {
3278 if (delalloc_to_write
< thresh
* 2)
3279 thresh
= delalloc_to_write
;
3280 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3284 /* did the fill delalloc function already unlock and start
3289 * we've unlocked the page, so we can't update
3290 * the mapping's writeback index, just update
3293 wbc
->nr_to_write
-= *nr_written
;
3304 * helper for __extent_writepage. This calls the writepage start hooks,
3305 * and does the loop to map the page into extents and bios.
3307 * We return 1 if the IO is started and the page is unlocked,
3308 * 0 if all went well (page still locked)
3309 * < 0 if there were errors (page still locked)
3311 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3313 struct writeback_control
*wbc
,
3314 struct extent_page_data
*epd
,
3316 unsigned long nr_written
,
3317 unsigned int write_flags
, int *nr_ret
)
3319 struct extent_io_tree
*tree
= epd
->tree
;
3320 u64 start
= page_offset(page
);
3321 u64 page_end
= start
+ PAGE_SIZE
- 1;
3327 struct extent_map
*em
;
3328 struct block_device
*bdev
;
3329 size_t pg_offset
= 0;
3335 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3336 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3339 /* Fixup worker will requeue */
3341 wbc
->pages_skipped
++;
3343 redirty_page_for_writepage(wbc
, page
);
3345 update_nr_written(wbc
, nr_written
);
3352 * we don't want to touch the inode after unlocking the page,
3353 * so we update the mapping writeback index now
3355 update_nr_written(wbc
, nr_written
+ 1);
3358 if (i_size
<= start
) {
3359 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3360 tree
->ops
->writepage_end_io_hook(page
, start
,
3365 blocksize
= inode
->i_sb
->s_blocksize
;
3367 while (cur
<= end
) {
3371 if (cur
>= i_size
) {
3372 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3373 tree
->ops
->writepage_end_io_hook(page
, cur
,
3377 em
= epd
->get_extent(BTRFS_I(inode
), page
, pg_offset
, cur
,
3379 if (IS_ERR_OR_NULL(em
)) {
3381 ret
= PTR_ERR_OR_ZERO(em
);
3385 extent_offset
= cur
- em
->start
;
3386 em_end
= extent_map_end(em
);
3387 BUG_ON(em_end
<= cur
);
3389 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3390 iosize
= ALIGN(iosize
, blocksize
);
3391 offset
= em
->block_start
+ extent_offset
;
3393 block_start
= em
->block_start
;
3394 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3395 free_extent_map(em
);
3399 * compressed and inline extents are written through other
3402 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3403 block_start
== EXTENT_MAP_INLINE
) {
3405 * end_io notification does not happen here for
3406 * compressed extents
3408 if (!compressed
&& tree
->ops
&&
3409 tree
->ops
->writepage_end_io_hook
)
3410 tree
->ops
->writepage_end_io_hook(page
, cur
,
3413 else if (compressed
) {
3414 /* we don't want to end_page_writeback on
3415 * a compressed extent. this happens
3422 pg_offset
+= iosize
;
3426 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3427 if (!PageWriteback(page
)) {
3428 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3429 "page %lu not writeback, cur %llu end %llu",
3430 page
->index
, cur
, end
);
3433 ret
= submit_extent_page(REQ_OP_WRITE
| write_flags
, tree
, wbc
,
3434 page
, offset
, iosize
, pg_offset
,
3436 end_bio_extent_writepage
,
3440 if (PageWriteback(page
))
3441 end_page_writeback(page
);
3445 pg_offset
+= iosize
;
3454 * the writepage semantics are similar to regular writepage. extent
3455 * records are inserted to lock ranges in the tree, and as dirty areas
3456 * are found, they are marked writeback. Then the lock bits are removed
3457 * and the end_io handler clears the writeback ranges
3459 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3462 struct inode
*inode
= page
->mapping
->host
;
3463 struct extent_page_data
*epd
= data
;
3464 u64 start
= page_offset(page
);
3465 u64 page_end
= start
+ PAGE_SIZE
- 1;
3468 size_t pg_offset
= 0;
3469 loff_t i_size
= i_size_read(inode
);
3470 unsigned long end_index
= i_size
>> PAGE_SHIFT
;
3471 unsigned int write_flags
= 0;
3472 unsigned long nr_written
= 0;
3474 write_flags
= wbc_to_write_flags(wbc
);
3476 trace___extent_writepage(page
, inode
, wbc
);
3478 WARN_ON(!PageLocked(page
));
3480 ClearPageError(page
);
3482 pg_offset
= i_size
& (PAGE_SIZE
- 1);
3483 if (page
->index
> end_index
||
3484 (page
->index
== end_index
&& !pg_offset
)) {
3485 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3490 if (page
->index
== end_index
) {
3493 userpage
= kmap_atomic(page
);
3494 memset(userpage
+ pg_offset
, 0,
3495 PAGE_SIZE
- pg_offset
);
3496 kunmap_atomic(userpage
);
3497 flush_dcache_page(page
);
3502 set_page_extent_mapped(page
);
3504 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3510 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3511 i_size
, nr_written
, write_flags
, &nr
);
3517 /* make sure the mapping tag for page dirty gets cleared */
3518 set_page_writeback(page
);
3519 end_page_writeback(page
);
3521 if (PageError(page
)) {
3522 ret
= ret
< 0 ? ret
: -EIO
;
3523 end_extent_writepage(page
, ret
, start
, page_end
);
3532 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3534 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3535 TASK_UNINTERRUPTIBLE
);
3538 static noinline_for_stack
int
3539 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3540 struct btrfs_fs_info
*fs_info
,
3541 struct extent_page_data
*epd
)
3543 unsigned long i
, num_pages
;
3547 if (!btrfs_try_tree_write_lock(eb
)) {
3549 flush_write_bio(epd
);
3550 btrfs_tree_lock(eb
);
3553 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3554 btrfs_tree_unlock(eb
);
3558 flush_write_bio(epd
);
3562 wait_on_extent_buffer_writeback(eb
);
3563 btrfs_tree_lock(eb
);
3564 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3566 btrfs_tree_unlock(eb
);
3571 * We need to do this to prevent races in people who check if the eb is
3572 * under IO since we can end up having no IO bits set for a short period
3575 spin_lock(&eb
->refs_lock
);
3576 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3577 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3578 spin_unlock(&eb
->refs_lock
);
3579 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3580 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
3582 fs_info
->dirty_metadata_batch
);
3585 spin_unlock(&eb
->refs_lock
);
3588 btrfs_tree_unlock(eb
);
3593 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3594 for (i
= 0; i
< num_pages
; i
++) {
3595 struct page
*p
= eb
->pages
[i
];
3597 if (!trylock_page(p
)) {
3599 flush_write_bio(epd
);
3609 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3611 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3612 smp_mb__after_atomic();
3613 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3616 static void set_btree_ioerr(struct page
*page
)
3618 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3621 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3625 * If writeback for a btree extent that doesn't belong to a log tree
3626 * failed, increment the counter transaction->eb_write_errors.
3627 * We do this because while the transaction is running and before it's
3628 * committing (when we call filemap_fdata[write|wait]_range against
3629 * the btree inode), we might have
3630 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3631 * returns an error or an error happens during writeback, when we're
3632 * committing the transaction we wouldn't know about it, since the pages
3633 * can be no longer dirty nor marked anymore for writeback (if a
3634 * subsequent modification to the extent buffer didn't happen before the
3635 * transaction commit), which makes filemap_fdata[write|wait]_range not
3636 * able to find the pages tagged with SetPageError at transaction
3637 * commit time. So if this happens we must abort the transaction,
3638 * otherwise we commit a super block with btree roots that point to
3639 * btree nodes/leafs whose content on disk is invalid - either garbage
3640 * or the content of some node/leaf from a past generation that got
3641 * cowed or deleted and is no longer valid.
3643 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3644 * not be enough - we need to distinguish between log tree extents vs
3645 * non-log tree extents, and the next filemap_fdatawait_range() call
3646 * will catch and clear such errors in the mapping - and that call might
3647 * be from a log sync and not from a transaction commit. Also, checking
3648 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3649 * not done and would not be reliable - the eb might have been released
3650 * from memory and reading it back again means that flag would not be
3651 * set (since it's a runtime flag, not persisted on disk).
3653 * Using the flags below in the btree inode also makes us achieve the
3654 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3655 * writeback for all dirty pages and before filemap_fdatawait_range()
3656 * is called, the writeback for all dirty pages had already finished
3657 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3658 * filemap_fdatawait_range() would return success, as it could not know
3659 * that writeback errors happened (the pages were no longer tagged for
3662 switch (eb
->log_index
) {
3664 set_bit(BTRFS_FS_BTREE_ERR
, &eb
->fs_info
->flags
);
3667 set_bit(BTRFS_FS_LOG1_ERR
, &eb
->fs_info
->flags
);
3670 set_bit(BTRFS_FS_LOG2_ERR
, &eb
->fs_info
->flags
);
3673 BUG(); /* unexpected, logic error */
3677 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3679 struct bio_vec
*bvec
;
3680 struct extent_buffer
*eb
;
3683 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
3684 bio_for_each_segment_all(bvec
, bio
, i
) {
3685 struct page
*page
= bvec
->bv_page
;
3687 eb
= (struct extent_buffer
*)page
->private;
3689 done
= atomic_dec_and_test(&eb
->io_pages
);
3691 if (bio
->bi_status
||
3692 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3693 ClearPageUptodate(page
);
3694 set_btree_ioerr(page
);
3697 end_page_writeback(page
);
3702 end_extent_buffer_writeback(eb
);
3708 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3709 struct btrfs_fs_info
*fs_info
,
3710 struct writeback_control
*wbc
,
3711 struct extent_page_data
*epd
)
3713 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3714 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3715 u64 offset
= eb
->start
;
3717 unsigned long i
, num_pages
;
3718 unsigned long start
, end
;
3719 unsigned int write_flags
= wbc_to_write_flags(wbc
) | REQ_META
;
3722 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3723 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3724 atomic_set(&eb
->io_pages
, num_pages
);
3726 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3727 nritems
= btrfs_header_nritems(eb
);
3728 if (btrfs_header_level(eb
) > 0) {
3729 end
= btrfs_node_key_ptr_offset(nritems
);
3731 memzero_extent_buffer(eb
, end
, eb
->len
- end
);
3735 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3737 start
= btrfs_item_nr_offset(nritems
);
3738 end
= BTRFS_LEAF_DATA_OFFSET
+ leaf_data_end(fs_info
, eb
);
3739 memzero_extent_buffer(eb
, start
, end
- start
);
3742 for (i
= 0; i
< num_pages
; i
++) {
3743 struct page
*p
= eb
->pages
[i
];
3745 clear_page_dirty_for_io(p
);
3746 set_page_writeback(p
);
3747 ret
= submit_extent_page(REQ_OP_WRITE
| write_flags
, tree
, wbc
,
3748 p
, offset
, PAGE_SIZE
, 0, bdev
,
3750 end_bio_extent_buffer_writepage
,
3754 if (PageWriteback(p
))
3755 end_page_writeback(p
);
3756 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3757 end_extent_buffer_writeback(eb
);
3761 offset
+= PAGE_SIZE
;
3762 update_nr_written(wbc
, 1);
3766 if (unlikely(ret
)) {
3767 for (; i
< num_pages
; i
++) {
3768 struct page
*p
= eb
->pages
[i
];
3769 clear_page_dirty_for_io(p
);
3777 int btree_write_cache_pages(struct address_space
*mapping
,
3778 struct writeback_control
*wbc
)
3780 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3781 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3782 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3783 struct extent_page_data epd
= {
3787 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3791 int nr_to_write_done
= 0;
3792 struct pagevec pvec
;
3795 pgoff_t end
; /* Inclusive */
3799 pagevec_init(&pvec
, 0);
3800 if (wbc
->range_cyclic
) {
3801 index
= mapping
->writeback_index
; /* Start from prev offset */
3804 index
= wbc
->range_start
>> PAGE_SHIFT
;
3805 end
= wbc
->range_end
>> PAGE_SHIFT
;
3808 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3809 tag
= PAGECACHE_TAG_TOWRITE
;
3811 tag
= PAGECACHE_TAG_DIRTY
;
3813 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3814 tag_pages_for_writeback(mapping
, index
, end
);
3815 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3816 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3817 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3821 for (i
= 0; i
< nr_pages
; i
++) {
3822 struct page
*page
= pvec
.pages
[i
];
3824 if (!PagePrivate(page
))
3827 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3832 spin_lock(&mapping
->private_lock
);
3833 if (!PagePrivate(page
)) {
3834 spin_unlock(&mapping
->private_lock
);
3838 eb
= (struct extent_buffer
*)page
->private;
3841 * Shouldn't happen and normally this would be a BUG_ON
3842 * but no sense in crashing the users box for something
3843 * we can survive anyway.
3846 spin_unlock(&mapping
->private_lock
);
3850 if (eb
== prev_eb
) {
3851 spin_unlock(&mapping
->private_lock
);
3855 ret
= atomic_inc_not_zero(&eb
->refs
);
3856 spin_unlock(&mapping
->private_lock
);
3861 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3863 free_extent_buffer(eb
);
3867 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3870 free_extent_buffer(eb
);
3873 free_extent_buffer(eb
);
3876 * the filesystem may choose to bump up nr_to_write.
3877 * We have to make sure to honor the new nr_to_write
3880 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3882 pagevec_release(&pvec
);
3885 if (!scanned
&& !done
) {
3887 * We hit the last page and there is more work to be done: wrap
3888 * back to the start of the file
3894 flush_write_bio(&epd
);
3899 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3900 * @mapping: address space structure to write
3901 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3902 * @writepage: function called for each page
3903 * @data: data passed to writepage function
3905 * If a page is already under I/O, write_cache_pages() skips it, even
3906 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3907 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3908 * and msync() need to guarantee that all the data which was dirty at the time
3909 * the call was made get new I/O started against them. If wbc->sync_mode is
3910 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3911 * existing IO to complete.
3913 static int extent_write_cache_pages(struct address_space
*mapping
,
3914 struct writeback_control
*wbc
,
3915 writepage_t writepage
, void *data
,
3916 void (*flush_fn
)(void *))
3918 struct inode
*inode
= mapping
->host
;
3921 int nr_to_write_done
= 0;
3922 struct pagevec pvec
;
3925 pgoff_t end
; /* Inclusive */
3927 int range_whole
= 0;
3932 * We have to hold onto the inode so that ordered extents can do their
3933 * work when the IO finishes. The alternative to this is failing to add
3934 * an ordered extent if the igrab() fails there and that is a huge pain
3935 * to deal with, so instead just hold onto the inode throughout the
3936 * writepages operation. If it fails here we are freeing up the inode
3937 * anyway and we'd rather not waste our time writing out stuff that is
3938 * going to be truncated anyway.
3943 pagevec_init(&pvec
, 0);
3944 if (wbc
->range_cyclic
) {
3945 index
= mapping
->writeback_index
; /* Start from prev offset */
3948 index
= wbc
->range_start
>> PAGE_SHIFT
;
3949 end
= wbc
->range_end
>> PAGE_SHIFT
;
3950 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3954 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3955 tag
= PAGECACHE_TAG_TOWRITE
;
3957 tag
= PAGECACHE_TAG_DIRTY
;
3959 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3960 tag_pages_for_writeback(mapping
, index
, end
);
3962 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3963 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3964 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3968 for (i
= 0; i
< nr_pages
; i
++) {
3969 struct page
*page
= pvec
.pages
[i
];
3971 done_index
= page
->index
;
3973 * At this point we hold neither mapping->tree_lock nor
3974 * lock on the page itself: the page may be truncated or
3975 * invalidated (changing page->mapping to NULL), or even
3976 * swizzled back from swapper_space to tmpfs file
3979 if (!trylock_page(page
)) {
3984 if (unlikely(page
->mapping
!= mapping
)) {
3989 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3995 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3996 if (PageWriteback(page
))
3998 wait_on_page_writeback(page
);
4001 if (PageWriteback(page
) ||
4002 !clear_page_dirty_for_io(page
)) {
4007 ret
= (*writepage
)(page
, wbc
, data
);
4009 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4015 * done_index is set past this page,
4016 * so media errors will not choke
4017 * background writeout for the entire
4018 * file. This has consequences for
4019 * range_cyclic semantics (ie. it may
4020 * not be suitable for data integrity
4023 done_index
= page
->index
+ 1;
4029 * the filesystem may choose to bump up nr_to_write.
4030 * We have to make sure to honor the new nr_to_write
4033 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4035 pagevec_release(&pvec
);
4038 if (!scanned
&& !done
) {
4040 * We hit the last page and there is more work to be done: wrap
4041 * back to the start of the file
4048 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4049 mapping
->writeback_index
= done_index
;
4051 btrfs_add_delayed_iput(inode
);
4055 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4060 ret
= submit_one_bio(epd
->bio
, 0, 0);
4061 BUG_ON(ret
< 0); /* -ENOMEM */
4066 static noinline
void flush_write_bio(void *data
)
4068 struct extent_page_data
*epd
= data
;
4069 flush_epd_write_bio(epd
);
4072 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4073 get_extent_t
*get_extent
,
4074 struct writeback_control
*wbc
)
4077 struct extent_page_data epd
= {
4080 .get_extent
= get_extent
,
4082 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4085 ret
= __extent_writepage(page
, wbc
, &epd
);
4087 flush_epd_write_bio(&epd
);
4091 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4092 u64 start
, u64 end
, get_extent_t
*get_extent
,
4096 struct address_space
*mapping
= inode
->i_mapping
;
4098 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4101 struct extent_page_data epd
= {
4104 .get_extent
= get_extent
,
4106 .sync_io
= mode
== WB_SYNC_ALL
,
4108 struct writeback_control wbc_writepages
= {
4110 .nr_to_write
= nr_pages
* 2,
4111 .range_start
= start
,
4112 .range_end
= end
+ 1,
4115 while (start
<= end
) {
4116 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4117 if (clear_page_dirty_for_io(page
))
4118 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4120 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4121 tree
->ops
->writepage_end_io_hook(page
, start
,
4122 start
+ PAGE_SIZE
- 1,
4130 flush_epd_write_bio(&epd
);
4134 int extent_writepages(struct extent_io_tree
*tree
,
4135 struct address_space
*mapping
,
4136 get_extent_t
*get_extent
,
4137 struct writeback_control
*wbc
)
4140 struct extent_page_data epd
= {
4143 .get_extent
= get_extent
,
4145 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4148 ret
= extent_write_cache_pages(mapping
, wbc
, __extent_writepage
, &epd
,
4150 flush_epd_write_bio(&epd
);
4154 int extent_readpages(struct extent_io_tree
*tree
,
4155 struct address_space
*mapping
,
4156 struct list_head
*pages
, unsigned nr_pages
,
4157 get_extent_t get_extent
)
4159 struct bio
*bio
= NULL
;
4161 unsigned long bio_flags
= 0;
4162 struct page
*pagepool
[16];
4164 struct extent_map
*em_cached
= NULL
;
4166 u64 prev_em_start
= (u64
)-1;
4168 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4169 page
= list_entry(pages
->prev
, struct page
, lru
);
4171 prefetchw(&page
->flags
);
4172 list_del(&page
->lru
);
4173 if (add_to_page_cache_lru(page
, mapping
,
4175 readahead_gfp_mask(mapping
))) {
4180 pagepool
[nr
++] = page
;
4181 if (nr
< ARRAY_SIZE(pagepool
))
4183 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4184 &bio
, 0, &bio_flags
, &prev_em_start
);
4188 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4189 &bio
, 0, &bio_flags
, &prev_em_start
);
4192 free_extent_map(em_cached
);
4194 BUG_ON(!list_empty(pages
));
4196 return submit_one_bio(bio
, 0, bio_flags
);
4201 * basic invalidatepage code, this waits on any locked or writeback
4202 * ranges corresponding to the page, and then deletes any extent state
4203 * records from the tree
4205 int extent_invalidatepage(struct extent_io_tree
*tree
,
4206 struct page
*page
, unsigned long offset
)
4208 struct extent_state
*cached_state
= NULL
;
4209 u64 start
= page_offset(page
);
4210 u64 end
= start
+ PAGE_SIZE
- 1;
4211 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4213 start
+= ALIGN(offset
, blocksize
);
4217 lock_extent_bits(tree
, start
, end
, &cached_state
);
4218 wait_on_page_writeback(page
);
4219 clear_extent_bit(tree
, start
, end
,
4220 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4221 EXTENT_DO_ACCOUNTING
,
4222 1, 1, &cached_state
, GFP_NOFS
);
4227 * a helper for releasepage, this tests for areas of the page that
4228 * are locked or under IO and drops the related state bits if it is safe
4231 static int try_release_extent_state(struct extent_map_tree
*map
,
4232 struct extent_io_tree
*tree
,
4233 struct page
*page
, gfp_t mask
)
4235 u64 start
= page_offset(page
);
4236 u64 end
= start
+ PAGE_SIZE
- 1;
4239 if (test_range_bit(tree
, start
, end
,
4240 EXTENT_IOBITS
, 0, NULL
))
4244 * at this point we can safely clear everything except the
4245 * locked bit and the nodatasum bit
4247 ret
= clear_extent_bit(tree
, start
, end
,
4248 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4251 /* if clear_extent_bit failed for enomem reasons,
4252 * we can't allow the release to continue.
4263 * a helper for releasepage. As long as there are no locked extents
4264 * in the range corresponding to the page, both state records and extent
4265 * map records are removed
4267 int try_release_extent_mapping(struct extent_map_tree
*map
,
4268 struct extent_io_tree
*tree
, struct page
*page
,
4271 struct extent_map
*em
;
4272 u64 start
= page_offset(page
);
4273 u64 end
= start
+ PAGE_SIZE
- 1;
4275 if (gfpflags_allow_blocking(mask
) &&
4276 page
->mapping
->host
->i_size
> SZ_16M
) {
4278 while (start
<= end
) {
4279 len
= end
- start
+ 1;
4280 write_lock(&map
->lock
);
4281 em
= lookup_extent_mapping(map
, start
, len
);
4283 write_unlock(&map
->lock
);
4286 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4287 em
->start
!= start
) {
4288 write_unlock(&map
->lock
);
4289 free_extent_map(em
);
4292 if (!test_range_bit(tree
, em
->start
,
4293 extent_map_end(em
) - 1,
4294 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4296 remove_extent_mapping(map
, em
);
4297 /* once for the rb tree */
4298 free_extent_map(em
);
4300 start
= extent_map_end(em
);
4301 write_unlock(&map
->lock
);
4304 free_extent_map(em
);
4307 return try_release_extent_state(map
, tree
, page
, mask
);
4311 * helper function for fiemap, which doesn't want to see any holes.
4312 * This maps until we find something past 'last'
4314 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4317 get_extent_t
*get_extent
)
4319 u64 sectorsize
= btrfs_inode_sectorsize(inode
);
4320 struct extent_map
*em
;
4327 len
= last
- offset
;
4330 len
= ALIGN(len
, sectorsize
);
4331 em
= get_extent(BTRFS_I(inode
), NULL
, 0, offset
, len
, 0);
4332 if (IS_ERR_OR_NULL(em
))
4335 /* if this isn't a hole return it */
4336 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4337 em
->block_start
!= EXTENT_MAP_HOLE
) {
4341 /* this is a hole, advance to the next extent */
4342 offset
= extent_map_end(em
);
4343 free_extent_map(em
);
4351 * To cache previous fiemap extent
4353 * Will be used for merging fiemap extent
4355 struct fiemap_cache
{
4364 * Helper to submit fiemap extent.
4366 * Will try to merge current fiemap extent specified by @offset, @phys,
4367 * @len and @flags with cached one.
4368 * And only when we fails to merge, cached one will be submitted as
4371 * Return value is the same as fiemap_fill_next_extent().
4373 static int emit_fiemap_extent(struct fiemap_extent_info
*fieinfo
,
4374 struct fiemap_cache
*cache
,
4375 u64 offset
, u64 phys
, u64 len
, u32 flags
)
4383 * Sanity check, extent_fiemap() should have ensured that new
4384 * fiemap extent won't overlap with cahced one.
4387 * NOTE: Physical address can overlap, due to compression
4389 if (cache
->offset
+ cache
->len
> offset
) {
4395 * Only merges fiemap extents if
4396 * 1) Their logical addresses are continuous
4398 * 2) Their physical addresses are continuous
4399 * So truly compressed (physical size smaller than logical size)
4400 * extents won't get merged with each other
4402 * 3) Share same flags except FIEMAP_EXTENT_LAST
4403 * So regular extent won't get merged with prealloc extent
4405 if (cache
->offset
+ cache
->len
== offset
&&
4406 cache
->phys
+ cache
->len
== phys
&&
4407 (cache
->flags
& ~FIEMAP_EXTENT_LAST
) ==
4408 (flags
& ~FIEMAP_EXTENT_LAST
)) {
4410 cache
->flags
|= flags
;
4411 goto try_submit_last
;
4414 /* Not mergeable, need to submit cached one */
4415 ret
= fiemap_fill_next_extent(fieinfo
, cache
->offset
, cache
->phys
,
4416 cache
->len
, cache
->flags
);
4417 cache
->cached
= false;
4421 cache
->cached
= true;
4422 cache
->offset
= offset
;
4425 cache
->flags
= flags
;
4427 if (cache
->flags
& FIEMAP_EXTENT_LAST
) {
4428 ret
= fiemap_fill_next_extent(fieinfo
, cache
->offset
,
4429 cache
->phys
, cache
->len
, cache
->flags
);
4430 cache
->cached
= false;
4436 * Emit last fiemap cache
4438 * The last fiemap cache may still be cached in the following case:
4440 * |<- Fiemap range ->|
4441 * |<------------ First extent ----------->|
4443 * In this case, the first extent range will be cached but not emitted.
4444 * So we must emit it before ending extent_fiemap().
4446 static int emit_last_fiemap_cache(struct btrfs_fs_info
*fs_info
,
4447 struct fiemap_extent_info
*fieinfo
,
4448 struct fiemap_cache
*cache
)
4455 ret
= fiemap_fill_next_extent(fieinfo
, cache
->offset
, cache
->phys
,
4456 cache
->len
, cache
->flags
);
4457 cache
->cached
= false;
4463 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4464 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4468 u64 max
= start
+ len
;
4472 u64 last_for_get_extent
= 0;
4474 u64 isize
= i_size_read(inode
);
4475 struct btrfs_key found_key
;
4476 struct extent_map
*em
= NULL
;
4477 struct extent_state
*cached_state
= NULL
;
4478 struct btrfs_path
*path
;
4479 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4480 struct fiemap_cache cache
= { 0 };
4489 path
= btrfs_alloc_path();
4492 path
->leave_spinning
= 1;
4494 start
= round_down(start
, btrfs_inode_sectorsize(inode
));
4495 len
= round_up(max
, btrfs_inode_sectorsize(inode
)) - start
;
4498 * lookup the last file extent. We're not using i_size here
4499 * because there might be preallocation past i_size
4501 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
4502 btrfs_ino(BTRFS_I(inode
)), -1, 0);
4504 btrfs_free_path(path
);
4513 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4514 found_type
= found_key
.type
;
4516 /* No extents, but there might be delalloc bits */
4517 if (found_key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
4518 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4519 /* have to trust i_size as the end */
4521 last_for_get_extent
= isize
;
4524 * remember the start of the last extent. There are a
4525 * bunch of different factors that go into the length of the
4526 * extent, so its much less complex to remember where it started
4528 last
= found_key
.offset
;
4529 last_for_get_extent
= last
+ 1;
4531 btrfs_release_path(path
);
4534 * we might have some extents allocated but more delalloc past those
4535 * extents. so, we trust isize unless the start of the last extent is
4540 last_for_get_extent
= isize
;
4543 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4546 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4556 u64 offset_in_extent
= 0;
4558 /* break if the extent we found is outside the range */
4559 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4563 * get_extent may return an extent that starts before our
4564 * requested range. We have to make sure the ranges
4565 * we return to fiemap always move forward and don't
4566 * overlap, so adjust the offsets here
4568 em_start
= max(em
->start
, off
);
4571 * record the offset from the start of the extent
4572 * for adjusting the disk offset below. Only do this if the
4573 * extent isn't compressed since our in ram offset may be past
4574 * what we have actually allocated on disk.
4576 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4577 offset_in_extent
= em_start
- em
->start
;
4578 em_end
= extent_map_end(em
);
4579 em_len
= em_end
- em_start
;
4584 * bump off for our next call to get_extent
4586 off
= extent_map_end(em
);
4590 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4592 flags
|= FIEMAP_EXTENT_LAST
;
4593 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4594 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4595 FIEMAP_EXTENT_NOT_ALIGNED
);
4596 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4597 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4598 FIEMAP_EXTENT_UNKNOWN
);
4599 } else if (fieinfo
->fi_extents_max
) {
4600 u64 bytenr
= em
->block_start
-
4601 (em
->start
- em
->orig_start
);
4603 disko
= em
->block_start
+ offset_in_extent
;
4606 * As btrfs supports shared space, this information
4607 * can be exported to userspace tools via
4608 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4609 * then we're just getting a count and we can skip the
4612 ret
= btrfs_check_shared(root
,
4613 btrfs_ino(BTRFS_I(inode
)),
4618 flags
|= FIEMAP_EXTENT_SHARED
;
4621 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4622 flags
|= FIEMAP_EXTENT_ENCODED
;
4623 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4624 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4626 free_extent_map(em
);
4628 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4629 (last
== (u64
)-1 && isize
<= em_end
)) {
4630 flags
|= FIEMAP_EXTENT_LAST
;
4634 /* now scan forward to see if this is really the last extent. */
4635 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4642 flags
|= FIEMAP_EXTENT_LAST
;
4645 ret
= emit_fiemap_extent(fieinfo
, &cache
, em_start
, disko
,
4655 ret
= emit_last_fiemap_cache(root
->fs_info
, fieinfo
, &cache
);
4656 free_extent_map(em
);
4658 btrfs_free_path(path
);
4659 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4660 &cached_state
, GFP_NOFS
);
4664 static void __free_extent_buffer(struct extent_buffer
*eb
)
4666 btrfs_leak_debug_del(&eb
->leak_list
);
4667 kmem_cache_free(extent_buffer_cache
, eb
);
4670 int extent_buffer_under_io(struct extent_buffer
*eb
)
4672 return (atomic_read(&eb
->io_pages
) ||
4673 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4674 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4678 * Helper for releasing extent buffer page.
4680 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4682 unsigned long index
;
4684 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4686 BUG_ON(extent_buffer_under_io(eb
));
4688 index
= num_extent_pages(eb
->start
, eb
->len
);
4694 page
= eb
->pages
[index
];
4698 spin_lock(&page
->mapping
->private_lock
);
4700 * We do this since we'll remove the pages after we've
4701 * removed the eb from the radix tree, so we could race
4702 * and have this page now attached to the new eb. So
4703 * only clear page_private if it's still connected to
4706 if (PagePrivate(page
) &&
4707 page
->private == (unsigned long)eb
) {
4708 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4709 BUG_ON(PageDirty(page
));
4710 BUG_ON(PageWriteback(page
));
4712 * We need to make sure we haven't be attached
4715 ClearPagePrivate(page
);
4716 set_page_private(page
, 0);
4717 /* One for the page private */
4722 spin_unlock(&page
->mapping
->private_lock
);
4724 /* One for when we allocated the page */
4726 } while (index
!= 0);
4730 * Helper for releasing the extent buffer.
4732 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4734 btrfs_release_extent_buffer_page(eb
);
4735 __free_extent_buffer(eb
);
4738 static struct extent_buffer
*
4739 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4742 struct extent_buffer
*eb
= NULL
;
4744 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4747 eb
->fs_info
= fs_info
;
4749 rwlock_init(&eb
->lock
);
4750 atomic_set(&eb
->write_locks
, 0);
4751 atomic_set(&eb
->read_locks
, 0);
4752 atomic_set(&eb
->blocking_readers
, 0);
4753 atomic_set(&eb
->blocking_writers
, 0);
4754 atomic_set(&eb
->spinning_readers
, 0);
4755 atomic_set(&eb
->spinning_writers
, 0);
4756 eb
->lock_nested
= 0;
4757 init_waitqueue_head(&eb
->write_lock_wq
);
4758 init_waitqueue_head(&eb
->read_lock_wq
);
4760 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4762 spin_lock_init(&eb
->refs_lock
);
4763 atomic_set(&eb
->refs
, 1);
4764 atomic_set(&eb
->io_pages
, 0);
4767 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4769 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4770 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4771 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4776 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4780 struct extent_buffer
*new;
4781 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4783 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4787 for (i
= 0; i
< num_pages
; i
++) {
4788 p
= alloc_page(GFP_NOFS
);
4790 btrfs_release_extent_buffer(new);
4793 attach_extent_buffer_page(new, p
);
4794 WARN_ON(PageDirty(p
));
4797 copy_page(page_address(p
), page_address(src
->pages
[i
]));
4800 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4801 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4806 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4807 u64 start
, unsigned long len
)
4809 struct extent_buffer
*eb
;
4810 unsigned long num_pages
;
4813 num_pages
= num_extent_pages(start
, len
);
4815 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4819 for (i
= 0; i
< num_pages
; i
++) {
4820 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4824 set_extent_buffer_uptodate(eb
);
4825 btrfs_set_header_nritems(eb
, 0);
4826 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4831 __free_page(eb
->pages
[i
- 1]);
4832 __free_extent_buffer(eb
);
4836 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4839 return __alloc_dummy_extent_buffer(fs_info
, start
, fs_info
->nodesize
);
4842 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4845 /* the ref bit is tricky. We have to make sure it is set
4846 * if we have the buffer dirty. Otherwise the
4847 * code to free a buffer can end up dropping a dirty
4850 * Once the ref bit is set, it won't go away while the
4851 * buffer is dirty or in writeback, and it also won't
4852 * go away while we have the reference count on the
4855 * We can't just set the ref bit without bumping the
4856 * ref on the eb because free_extent_buffer might
4857 * see the ref bit and try to clear it. If this happens
4858 * free_extent_buffer might end up dropping our original
4859 * ref by mistake and freeing the page before we are able
4860 * to add one more ref.
4862 * So bump the ref count first, then set the bit. If someone
4863 * beat us to it, drop the ref we added.
4865 refs
= atomic_read(&eb
->refs
);
4866 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4869 spin_lock(&eb
->refs_lock
);
4870 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4871 atomic_inc(&eb
->refs
);
4872 spin_unlock(&eb
->refs_lock
);
4875 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4876 struct page
*accessed
)
4878 unsigned long num_pages
, i
;
4880 check_buffer_tree_ref(eb
);
4882 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4883 for (i
= 0; i
< num_pages
; i
++) {
4884 struct page
*p
= eb
->pages
[i
];
4887 mark_page_accessed(p
);
4891 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4894 struct extent_buffer
*eb
;
4897 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4898 start
>> PAGE_SHIFT
);
4899 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4902 * Lock our eb's refs_lock to avoid races with
4903 * free_extent_buffer. When we get our eb it might be flagged
4904 * with EXTENT_BUFFER_STALE and another task running
4905 * free_extent_buffer might have seen that flag set,
4906 * eb->refs == 2, that the buffer isn't under IO (dirty and
4907 * writeback flags not set) and it's still in the tree (flag
4908 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4909 * of decrementing the extent buffer's reference count twice.
4910 * So here we could race and increment the eb's reference count,
4911 * clear its stale flag, mark it as dirty and drop our reference
4912 * before the other task finishes executing free_extent_buffer,
4913 * which would later result in an attempt to free an extent
4914 * buffer that is dirty.
4916 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4917 spin_lock(&eb
->refs_lock
);
4918 spin_unlock(&eb
->refs_lock
);
4920 mark_extent_buffer_accessed(eb
, NULL
);
4928 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4929 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4932 struct extent_buffer
*eb
, *exists
= NULL
;
4935 eb
= find_extent_buffer(fs_info
, start
);
4938 eb
= alloc_dummy_extent_buffer(fs_info
, start
);
4941 eb
->fs_info
= fs_info
;
4943 ret
= radix_tree_preload(GFP_NOFS
);
4946 spin_lock(&fs_info
->buffer_lock
);
4947 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4948 start
>> PAGE_SHIFT
, eb
);
4949 spin_unlock(&fs_info
->buffer_lock
);
4950 radix_tree_preload_end();
4951 if (ret
== -EEXIST
) {
4952 exists
= find_extent_buffer(fs_info
, start
);
4958 check_buffer_tree_ref(eb
);
4959 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4962 * We will free dummy extent buffer's if they come into
4963 * free_extent_buffer with a ref count of 2, but if we are using this we
4964 * want the buffers to stay in memory until we're done with them, so
4965 * bump the ref count again.
4967 atomic_inc(&eb
->refs
);
4970 btrfs_release_extent_buffer(eb
);
4975 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4978 unsigned long len
= fs_info
->nodesize
;
4979 unsigned long num_pages
= num_extent_pages(start
, len
);
4981 unsigned long index
= start
>> PAGE_SHIFT
;
4982 struct extent_buffer
*eb
;
4983 struct extent_buffer
*exists
= NULL
;
4985 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4989 if (!IS_ALIGNED(start
, fs_info
->sectorsize
)) {
4990 btrfs_err(fs_info
, "bad tree block start %llu", start
);
4991 return ERR_PTR(-EINVAL
);
4994 eb
= find_extent_buffer(fs_info
, start
);
4998 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
5000 return ERR_PTR(-ENOMEM
);
5002 for (i
= 0; i
< num_pages
; i
++, index
++) {
5003 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
5005 exists
= ERR_PTR(-ENOMEM
);
5009 spin_lock(&mapping
->private_lock
);
5010 if (PagePrivate(p
)) {
5012 * We could have already allocated an eb for this page
5013 * and attached one so lets see if we can get a ref on
5014 * the existing eb, and if we can we know it's good and
5015 * we can just return that one, else we know we can just
5016 * overwrite page->private.
5018 exists
= (struct extent_buffer
*)p
->private;
5019 if (atomic_inc_not_zero(&exists
->refs
)) {
5020 spin_unlock(&mapping
->private_lock
);
5023 mark_extent_buffer_accessed(exists
, p
);
5029 * Do this so attach doesn't complain and we need to
5030 * drop the ref the old guy had.
5032 ClearPagePrivate(p
);
5033 WARN_ON(PageDirty(p
));
5036 attach_extent_buffer_page(eb
, p
);
5037 spin_unlock(&mapping
->private_lock
);
5038 WARN_ON(PageDirty(p
));
5040 if (!PageUptodate(p
))
5044 * see below about how we avoid a nasty race with release page
5045 * and why we unlock later
5049 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5051 ret
= radix_tree_preload(GFP_NOFS
);
5053 exists
= ERR_PTR(ret
);
5057 spin_lock(&fs_info
->buffer_lock
);
5058 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
5059 start
>> PAGE_SHIFT
, eb
);
5060 spin_unlock(&fs_info
->buffer_lock
);
5061 radix_tree_preload_end();
5062 if (ret
== -EEXIST
) {
5063 exists
= find_extent_buffer(fs_info
, start
);
5069 /* add one reference for the tree */
5070 check_buffer_tree_ref(eb
);
5071 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
5074 * there is a race where release page may have
5075 * tried to find this extent buffer in the radix
5076 * but failed. It will tell the VM it is safe to
5077 * reclaim the, and it will clear the page private bit.
5078 * We must make sure to set the page private bit properly
5079 * after the extent buffer is in the radix tree so
5080 * it doesn't get lost
5082 SetPageChecked(eb
->pages
[0]);
5083 for (i
= 1; i
< num_pages
; i
++) {
5085 ClearPageChecked(p
);
5088 unlock_page(eb
->pages
[0]);
5092 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
5093 for (i
= 0; i
< num_pages
; i
++) {
5095 unlock_page(eb
->pages
[i
]);
5098 btrfs_release_extent_buffer(eb
);
5102 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
5104 struct extent_buffer
*eb
=
5105 container_of(head
, struct extent_buffer
, rcu_head
);
5107 __free_extent_buffer(eb
);
5110 /* Expects to have eb->eb_lock already held */
5111 static int release_extent_buffer(struct extent_buffer
*eb
)
5113 WARN_ON(atomic_read(&eb
->refs
) == 0);
5114 if (atomic_dec_and_test(&eb
->refs
)) {
5115 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5116 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5118 spin_unlock(&eb
->refs_lock
);
5120 spin_lock(&fs_info
->buffer_lock
);
5121 radix_tree_delete(&fs_info
->buffer_radix
,
5122 eb
->start
>> PAGE_SHIFT
);
5123 spin_unlock(&fs_info
->buffer_lock
);
5125 spin_unlock(&eb
->refs_lock
);
5128 /* Should be safe to release our pages at this point */
5129 btrfs_release_extent_buffer_page(eb
);
5130 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5131 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5132 __free_extent_buffer(eb
);
5136 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5139 spin_unlock(&eb
->refs_lock
);
5144 void free_extent_buffer(struct extent_buffer
*eb
)
5152 refs
= atomic_read(&eb
->refs
);
5155 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5160 spin_lock(&eb
->refs_lock
);
5161 if (atomic_read(&eb
->refs
) == 2 &&
5162 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5163 atomic_dec(&eb
->refs
);
5165 if (atomic_read(&eb
->refs
) == 2 &&
5166 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5167 !extent_buffer_under_io(eb
) &&
5168 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5169 atomic_dec(&eb
->refs
);
5172 * I know this is terrible, but it's temporary until we stop tracking
5173 * the uptodate bits and such for the extent buffers.
5175 release_extent_buffer(eb
);
5178 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5183 spin_lock(&eb
->refs_lock
);
5184 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5186 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5187 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5188 atomic_dec(&eb
->refs
);
5189 release_extent_buffer(eb
);
5192 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5195 unsigned long num_pages
;
5198 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5200 for (i
= 0; i
< num_pages
; i
++) {
5201 page
= eb
->pages
[i
];
5202 if (!PageDirty(page
))
5206 WARN_ON(!PagePrivate(page
));
5208 clear_page_dirty_for_io(page
);
5209 spin_lock_irq(&page
->mapping
->tree_lock
);
5210 if (!PageDirty(page
)) {
5211 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5213 PAGECACHE_TAG_DIRTY
);
5215 spin_unlock_irq(&page
->mapping
->tree_lock
);
5216 ClearPageError(page
);
5219 WARN_ON(atomic_read(&eb
->refs
) == 0);
5222 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5225 unsigned long num_pages
;
5228 check_buffer_tree_ref(eb
);
5230 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5232 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5233 WARN_ON(atomic_read(&eb
->refs
) == 0);
5234 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5236 for (i
= 0; i
< num_pages
; i
++)
5237 set_page_dirty(eb
->pages
[i
]);
5241 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5245 unsigned long num_pages
;
5247 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5248 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5249 for (i
= 0; i
< num_pages
; i
++) {
5250 page
= eb
->pages
[i
];
5252 ClearPageUptodate(page
);
5256 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5260 unsigned long num_pages
;
5262 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5263 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5264 for (i
= 0; i
< num_pages
; i
++) {
5265 page
= eb
->pages
[i
];
5266 SetPageUptodate(page
);
5270 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5272 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5275 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5276 struct extent_buffer
*eb
, int wait
,
5277 get_extent_t
*get_extent
, int mirror_num
)
5283 int locked_pages
= 0;
5284 int all_uptodate
= 1;
5285 unsigned long num_pages
;
5286 unsigned long num_reads
= 0;
5287 struct bio
*bio
= NULL
;
5288 unsigned long bio_flags
= 0;
5290 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5293 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5294 for (i
= 0; i
< num_pages
; i
++) {
5295 page
= eb
->pages
[i
];
5296 if (wait
== WAIT_NONE
) {
5297 if (!trylock_page(page
))
5305 * We need to firstly lock all pages to make sure that
5306 * the uptodate bit of our pages won't be affected by
5307 * clear_extent_buffer_uptodate().
5309 for (i
= 0; i
< num_pages
; i
++) {
5310 page
= eb
->pages
[i
];
5311 if (!PageUptodate(page
)) {
5318 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5322 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5323 eb
->read_mirror
= 0;
5324 atomic_set(&eb
->io_pages
, num_reads
);
5325 for (i
= 0; i
< num_pages
; i
++) {
5326 page
= eb
->pages
[i
];
5328 if (!PageUptodate(page
)) {
5330 atomic_dec(&eb
->io_pages
);
5335 ClearPageError(page
);
5336 err
= __extent_read_full_page(tree
, page
,
5338 mirror_num
, &bio_flags
,
5343 * We use &bio in above __extent_read_full_page,
5344 * so we ensure that if it returns error, the
5345 * current page fails to add itself to bio and
5346 * it's been unlocked.
5348 * We must dec io_pages by ourselves.
5350 atomic_dec(&eb
->io_pages
);
5358 err
= submit_one_bio(bio
, mirror_num
, bio_flags
);
5363 if (ret
|| wait
!= WAIT_COMPLETE
)
5366 for (i
= 0; i
< num_pages
; i
++) {
5367 page
= eb
->pages
[i
];
5368 wait_on_page_locked(page
);
5369 if (!PageUptodate(page
))
5376 while (locked_pages
> 0) {
5378 page
= eb
->pages
[locked_pages
];
5384 void read_extent_buffer(const struct extent_buffer
*eb
, void *dstv
,
5385 unsigned long start
, unsigned long len
)
5391 char *dst
= (char *)dstv
;
5392 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5393 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5395 if (start
+ len
> eb
->len
) {
5396 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5397 eb
->start
, eb
->len
, start
, len
);
5398 memset(dst
, 0, len
);
5402 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5405 page
= eb
->pages
[i
];
5407 cur
= min(len
, (PAGE_SIZE
- offset
));
5408 kaddr
= page_address(page
);
5409 memcpy(dst
, kaddr
+ offset
, cur
);
5418 int read_extent_buffer_to_user(const struct extent_buffer
*eb
,
5420 unsigned long start
, unsigned long len
)
5426 char __user
*dst
= (char __user
*)dstv
;
5427 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5428 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5431 WARN_ON(start
> eb
->len
);
5432 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5434 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5437 page
= eb
->pages
[i
];
5439 cur
= min(len
, (PAGE_SIZE
- offset
));
5440 kaddr
= page_address(page
);
5441 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5456 * return 0 if the item is found within a page.
5457 * return 1 if the item spans two pages.
5458 * return -EINVAL otherwise.
5460 int map_private_extent_buffer(const struct extent_buffer
*eb
,
5461 unsigned long start
, unsigned long min_len
,
5462 char **map
, unsigned long *map_start
,
5463 unsigned long *map_len
)
5465 size_t offset
= start
& (PAGE_SIZE
- 1);
5468 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5469 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5470 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5473 if (start
+ min_len
> eb
->len
) {
5474 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5475 eb
->start
, eb
->len
, start
, min_len
);
5483 offset
= start_offset
;
5487 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5491 kaddr
= page_address(p
);
5492 *map
= kaddr
+ offset
;
5493 *map_len
= PAGE_SIZE
- offset
;
5497 int memcmp_extent_buffer(const struct extent_buffer
*eb
, const void *ptrv
,
5498 unsigned long start
, unsigned long len
)
5504 char *ptr
= (char *)ptrv
;
5505 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5506 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5509 WARN_ON(start
> eb
->len
);
5510 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5512 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5515 page
= eb
->pages
[i
];
5517 cur
= min(len
, (PAGE_SIZE
- offset
));
5519 kaddr
= page_address(page
);
5520 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5532 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer
*eb
,
5537 WARN_ON(!PageUptodate(eb
->pages
[0]));
5538 kaddr
= page_address(eb
->pages
[0]);
5539 memcpy(kaddr
+ offsetof(struct btrfs_header
, chunk_tree_uuid
), srcv
,
5543 void write_extent_buffer_fsid(struct extent_buffer
*eb
, const void *srcv
)
5547 WARN_ON(!PageUptodate(eb
->pages
[0]));
5548 kaddr
= page_address(eb
->pages
[0]);
5549 memcpy(kaddr
+ offsetof(struct btrfs_header
, fsid
), srcv
,
5553 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5554 unsigned long start
, unsigned long len
)
5560 char *src
= (char *)srcv
;
5561 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5562 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5564 WARN_ON(start
> eb
->len
);
5565 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5567 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5570 page
= eb
->pages
[i
];
5571 WARN_ON(!PageUptodate(page
));
5573 cur
= min(len
, PAGE_SIZE
- offset
);
5574 kaddr
= page_address(page
);
5575 memcpy(kaddr
+ offset
, src
, cur
);
5584 void memzero_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5591 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5592 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5594 WARN_ON(start
> eb
->len
);
5595 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5597 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5600 page
= eb
->pages
[i
];
5601 WARN_ON(!PageUptodate(page
));
5603 cur
= min(len
, PAGE_SIZE
- offset
);
5604 kaddr
= page_address(page
);
5605 memset(kaddr
+ offset
, 0, cur
);
5613 void copy_extent_buffer_full(struct extent_buffer
*dst
,
5614 struct extent_buffer
*src
)
5619 ASSERT(dst
->len
== src
->len
);
5621 num_pages
= num_extent_pages(dst
->start
, dst
->len
);
5622 for (i
= 0; i
< num_pages
; i
++)
5623 copy_page(page_address(dst
->pages
[i
]),
5624 page_address(src
->pages
[i
]));
5627 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5628 unsigned long dst_offset
, unsigned long src_offset
,
5631 u64 dst_len
= dst
->len
;
5636 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5637 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5639 WARN_ON(src
->len
!= dst_len
);
5641 offset
= (start_offset
+ dst_offset
) &
5645 page
= dst
->pages
[i
];
5646 WARN_ON(!PageUptodate(page
));
5648 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5650 kaddr
= page_address(page
);
5651 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5660 void le_bitmap_set(u8
*map
, unsigned int start
, int len
)
5662 u8
*p
= map
+ BIT_BYTE(start
);
5663 const unsigned int size
= start
+ len
;
5664 int bits_to_set
= BITS_PER_BYTE
- (start
% BITS_PER_BYTE
);
5665 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(start
);
5667 while (len
- bits_to_set
>= 0) {
5670 bits_to_set
= BITS_PER_BYTE
;
5675 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5680 void le_bitmap_clear(u8
*map
, unsigned int start
, int len
)
5682 u8
*p
= map
+ BIT_BYTE(start
);
5683 const unsigned int size
= start
+ len
;
5684 int bits_to_clear
= BITS_PER_BYTE
- (start
% BITS_PER_BYTE
);
5685 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(start
);
5687 while (len
- bits_to_clear
>= 0) {
5688 *p
&= ~mask_to_clear
;
5689 len
-= bits_to_clear
;
5690 bits_to_clear
= BITS_PER_BYTE
;
5695 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5696 *p
&= ~mask_to_clear
;
5701 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5703 * @eb: the extent buffer
5704 * @start: offset of the bitmap item in the extent buffer
5706 * @page_index: return index of the page in the extent buffer that contains the
5708 * @page_offset: return offset into the page given by page_index
5710 * This helper hides the ugliness of finding the byte in an extent buffer which
5711 * contains a given bit.
5713 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5714 unsigned long start
, unsigned long nr
,
5715 unsigned long *page_index
,
5716 size_t *page_offset
)
5718 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5719 size_t byte_offset
= BIT_BYTE(nr
);
5723 * The byte we want is the offset of the extent buffer + the offset of
5724 * the bitmap item in the extent buffer + the offset of the byte in the
5727 offset
= start_offset
+ start
+ byte_offset
;
5729 *page_index
= offset
>> PAGE_SHIFT
;
5730 *page_offset
= offset
& (PAGE_SIZE
- 1);
5734 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5735 * @eb: the extent buffer
5736 * @start: offset of the bitmap item in the extent buffer
5737 * @nr: bit number to test
5739 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5747 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5748 page
= eb
->pages
[i
];
5749 WARN_ON(!PageUptodate(page
));
5750 kaddr
= page_address(page
);
5751 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5755 * extent_buffer_bitmap_set - set an area of a bitmap
5756 * @eb: the extent buffer
5757 * @start: offset of the bitmap item in the extent buffer
5758 * @pos: bit number of the first bit
5759 * @len: number of bits to set
5761 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5762 unsigned long pos
, unsigned long len
)
5768 const unsigned int size
= pos
+ len
;
5769 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5770 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5772 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5773 page
= eb
->pages
[i
];
5774 WARN_ON(!PageUptodate(page
));
5775 kaddr
= page_address(page
);
5777 while (len
>= bits_to_set
) {
5778 kaddr
[offset
] |= mask_to_set
;
5780 bits_to_set
= BITS_PER_BYTE
;
5782 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5784 page
= eb
->pages
[++i
];
5785 WARN_ON(!PageUptodate(page
));
5786 kaddr
= page_address(page
);
5790 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5791 kaddr
[offset
] |= mask_to_set
;
5797 * extent_buffer_bitmap_clear - clear an area of a bitmap
5798 * @eb: the extent buffer
5799 * @start: offset of the bitmap item in the extent buffer
5800 * @pos: bit number of the first bit
5801 * @len: number of bits to clear
5803 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5804 unsigned long pos
, unsigned long len
)
5810 const unsigned int size
= pos
+ len
;
5811 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5812 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5814 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5815 page
= eb
->pages
[i
];
5816 WARN_ON(!PageUptodate(page
));
5817 kaddr
= page_address(page
);
5819 while (len
>= bits_to_clear
) {
5820 kaddr
[offset
] &= ~mask_to_clear
;
5821 len
-= bits_to_clear
;
5822 bits_to_clear
= BITS_PER_BYTE
;
5824 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5826 page
= eb
->pages
[++i
];
5827 WARN_ON(!PageUptodate(page
));
5828 kaddr
= page_address(page
);
5832 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5833 kaddr
[offset
] &= ~mask_to_clear
;
5837 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5839 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5840 return distance
< len
;
5843 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5844 unsigned long dst_off
, unsigned long src_off
,
5847 char *dst_kaddr
= page_address(dst_page
);
5849 int must_memmove
= 0;
5851 if (dst_page
!= src_page
) {
5852 src_kaddr
= page_address(src_page
);
5854 src_kaddr
= dst_kaddr
;
5855 if (areas_overlap(src_off
, dst_off
, len
))
5860 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5862 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5865 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5866 unsigned long src_offset
, unsigned long len
)
5868 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5870 size_t dst_off_in_page
;
5871 size_t src_off_in_page
;
5872 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5873 unsigned long dst_i
;
5874 unsigned long src_i
;
5876 if (src_offset
+ len
> dst
->len
) {
5878 "memmove bogus src_offset %lu move len %lu dst len %lu",
5879 src_offset
, len
, dst
->len
);
5882 if (dst_offset
+ len
> dst
->len
) {
5884 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5885 dst_offset
, len
, dst
->len
);
5890 dst_off_in_page
= (start_offset
+ dst_offset
) &
5892 src_off_in_page
= (start_offset
+ src_offset
) &
5895 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5896 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5898 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5900 cur
= min_t(unsigned long, cur
,
5901 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5903 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5904 dst_off_in_page
, src_off_in_page
, cur
);
5912 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5913 unsigned long src_offset
, unsigned long len
)
5915 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5917 size_t dst_off_in_page
;
5918 size_t src_off_in_page
;
5919 unsigned long dst_end
= dst_offset
+ len
- 1;
5920 unsigned long src_end
= src_offset
+ len
- 1;
5921 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5922 unsigned long dst_i
;
5923 unsigned long src_i
;
5925 if (src_offset
+ len
> dst
->len
) {
5927 "memmove bogus src_offset %lu move len %lu len %lu",
5928 src_offset
, len
, dst
->len
);
5931 if (dst_offset
+ len
> dst
->len
) {
5933 "memmove bogus dst_offset %lu move len %lu len %lu",
5934 dst_offset
, len
, dst
->len
);
5937 if (dst_offset
< src_offset
) {
5938 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5942 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5943 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5945 dst_off_in_page
= (start_offset
+ dst_end
) &
5947 src_off_in_page
= (start_offset
+ src_end
) &
5950 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5951 cur
= min(cur
, dst_off_in_page
+ 1);
5952 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5953 dst_off_in_page
- cur
+ 1,
5954 src_off_in_page
- cur
+ 1, cur
);
5962 int try_release_extent_buffer(struct page
*page
)
5964 struct extent_buffer
*eb
;
5967 * We need to make sure nobody is attaching this page to an eb right
5970 spin_lock(&page
->mapping
->private_lock
);
5971 if (!PagePrivate(page
)) {
5972 spin_unlock(&page
->mapping
->private_lock
);
5976 eb
= (struct extent_buffer
*)page
->private;
5980 * This is a little awful but should be ok, we need to make sure that
5981 * the eb doesn't disappear out from under us while we're looking at
5984 spin_lock(&eb
->refs_lock
);
5985 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5986 spin_unlock(&eb
->refs_lock
);
5987 spin_unlock(&page
->mapping
->private_lock
);
5990 spin_unlock(&page
->mapping
->private_lock
);
5993 * If tree ref isn't set then we know the ref on this eb is a real ref,
5994 * so just return, this page will likely be freed soon anyway.
5996 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5997 spin_unlock(&eb
->refs_lock
);
6001 return release_extent_buffer(eb
);