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 #ifdef CONFIG_BTRFS_DEBUG
29 static LIST_HEAD(buffers
);
30 static LIST_HEAD(states
);
32 static DEFINE_SPINLOCK(leak_lock
);
35 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
39 spin_lock_irqsave(&leak_lock
, flags
);
41 spin_unlock_irqrestore(&leak_lock
, flags
);
45 void btrfs_leak_debug_del(struct list_head
*entry
)
49 spin_lock_irqsave(&leak_lock
, flags
);
51 spin_unlock_irqrestore(&leak_lock
, flags
);
55 void btrfs_leak_debug_check(void)
57 struct extent_state
*state
;
58 struct extent_buffer
*eb
;
60 while (!list_empty(&states
)) {
61 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
62 printk(KERN_ERR
"BTRFS: state leak: start %llu end %llu "
63 "state %lu in tree %p refs %d\n",
64 state
->start
, state
->end
, state
->state
, state
->tree
,
65 atomic_read(&state
->refs
));
66 list_del(&state
->leak_list
);
67 kmem_cache_free(extent_state_cache
, state
);
70 while (!list_empty(&buffers
)) {
71 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
72 printk(KERN_ERR
"BTRFS: buffer leak start %llu len %lu "
74 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
75 list_del(&eb
->leak_list
);
76 kmem_cache_free(extent_buffer_cache
, eb
);
80 #define btrfs_debug_check_extent_io_range(tree, start, end) \
81 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
82 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
83 struct extent_io_tree
*tree
, u64 start
, u64 end
)
91 inode
= tree
->mapping
->host
;
92 isize
= i_size_read(inode
);
93 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
94 printk_ratelimited(KERN_DEBUG
95 "BTRFS: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
96 caller
, btrfs_ino(inode
), isize
, start
, end
);
100 #define btrfs_leak_debug_add(new, head) do {} while (0)
101 #define btrfs_leak_debug_del(entry) do {} while (0)
102 #define btrfs_leak_debug_check() do {} while (0)
103 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
106 #define BUFFER_LRU_MAX 64
111 struct rb_node rb_node
;
114 struct extent_page_data
{
116 struct extent_io_tree
*tree
;
117 get_extent_t
*get_extent
;
118 unsigned long bio_flags
;
120 /* tells writepage not to lock the state bits for this range
121 * it still does the unlocking
123 unsigned int extent_locked
:1;
125 /* tells the submit_bio code to use a WRITE_SYNC */
126 unsigned int sync_io
:1;
129 static noinline
void flush_write_bio(void *data
);
130 static inline struct btrfs_fs_info
*
131 tree_fs_info(struct extent_io_tree
*tree
)
135 return btrfs_sb(tree
->mapping
->host
->i_sb
);
138 int __init
extent_io_init(void)
140 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
141 sizeof(struct extent_state
), 0,
142 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
143 if (!extent_state_cache
)
146 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
147 sizeof(struct extent_buffer
), 0,
148 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
149 if (!extent_buffer_cache
)
150 goto free_state_cache
;
152 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
153 offsetof(struct btrfs_io_bio
, bio
));
155 goto free_buffer_cache
;
157 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
163 bioset_free(btrfs_bioset
);
167 kmem_cache_destroy(extent_buffer_cache
);
168 extent_buffer_cache
= NULL
;
171 kmem_cache_destroy(extent_state_cache
);
172 extent_state_cache
= NULL
;
176 void extent_io_exit(void)
178 btrfs_leak_debug_check();
181 * Make sure all delayed rcu free are flushed before we
185 if (extent_state_cache
)
186 kmem_cache_destroy(extent_state_cache
);
187 if (extent_buffer_cache
)
188 kmem_cache_destroy(extent_buffer_cache
);
190 bioset_free(btrfs_bioset
);
193 void extent_io_tree_init(struct extent_io_tree
*tree
,
194 struct address_space
*mapping
)
196 tree
->state
= RB_ROOT
;
198 tree
->dirty_bytes
= 0;
199 spin_lock_init(&tree
->lock
);
200 tree
->mapping
= mapping
;
203 static struct extent_state
*alloc_extent_state(gfp_t mask
)
205 struct extent_state
*state
;
207 state
= kmem_cache_alloc(extent_state_cache
, mask
);
213 btrfs_leak_debug_add(&state
->leak_list
, &states
);
214 atomic_set(&state
->refs
, 1);
215 init_waitqueue_head(&state
->wq
);
216 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
220 void free_extent_state(struct extent_state
*state
)
224 if (atomic_dec_and_test(&state
->refs
)) {
225 WARN_ON(state
->tree
);
226 btrfs_leak_debug_del(&state
->leak_list
);
227 trace_free_extent_state(state
, _RET_IP_
);
228 kmem_cache_free(extent_state_cache
, state
);
232 static struct rb_node
*tree_insert(struct rb_root
*root
,
233 struct rb_node
*search_start
,
235 struct rb_node
*node
,
236 struct rb_node
***p_in
,
237 struct rb_node
**parent_in
)
240 struct rb_node
*parent
= NULL
;
241 struct tree_entry
*entry
;
243 if (p_in
&& parent_in
) {
249 p
= search_start
? &search_start
: &root
->rb_node
;
252 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
254 if (offset
< entry
->start
)
256 else if (offset
> entry
->end
)
263 rb_link_node(node
, parent
, p
);
264 rb_insert_color(node
, root
);
268 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
269 struct rb_node
**prev_ret
,
270 struct rb_node
**next_ret
,
271 struct rb_node
***p_ret
,
272 struct rb_node
**parent_ret
)
274 struct rb_root
*root
= &tree
->state
;
275 struct rb_node
**n
= &root
->rb_node
;
276 struct rb_node
*prev
= NULL
;
277 struct rb_node
*orig_prev
= NULL
;
278 struct tree_entry
*entry
;
279 struct tree_entry
*prev_entry
= NULL
;
283 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
286 if (offset
< entry
->start
)
288 else if (offset
> entry
->end
)
301 while (prev
&& offset
> prev_entry
->end
) {
302 prev
= rb_next(prev
);
303 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
310 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
311 while (prev
&& offset
< prev_entry
->start
) {
312 prev
= rb_prev(prev
);
313 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
320 static inline struct rb_node
*
321 tree_search_for_insert(struct extent_io_tree
*tree
,
323 struct rb_node
***p_ret
,
324 struct rb_node
**parent_ret
)
326 struct rb_node
*prev
= NULL
;
329 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
335 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
338 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
341 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
342 struct extent_state
*other
)
344 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
345 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
350 * utility function to look for merge candidates inside a given range.
351 * Any extents with matching state are merged together into a single
352 * extent in the tree. Extents with EXTENT_IO in their state field
353 * are not merged because the end_io handlers need to be able to do
354 * operations on them without sleeping (or doing allocations/splits).
356 * This should be called with the tree lock held.
358 static void merge_state(struct extent_io_tree
*tree
,
359 struct extent_state
*state
)
361 struct extent_state
*other
;
362 struct rb_node
*other_node
;
364 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
367 other_node
= rb_prev(&state
->rb_node
);
369 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
370 if (other
->end
== state
->start
- 1 &&
371 other
->state
== state
->state
) {
372 merge_cb(tree
, state
, other
);
373 state
->start
= other
->start
;
375 rb_erase(&other
->rb_node
, &tree
->state
);
376 free_extent_state(other
);
379 other_node
= rb_next(&state
->rb_node
);
381 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
382 if (other
->start
== state
->end
+ 1 &&
383 other
->state
== state
->state
) {
384 merge_cb(tree
, state
, other
);
385 state
->end
= other
->end
;
387 rb_erase(&other
->rb_node
, &tree
->state
);
388 free_extent_state(other
);
393 static void set_state_cb(struct extent_io_tree
*tree
,
394 struct extent_state
*state
, unsigned long *bits
)
396 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
397 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
400 static void clear_state_cb(struct extent_io_tree
*tree
,
401 struct extent_state
*state
, unsigned long *bits
)
403 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
404 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
407 static void set_state_bits(struct extent_io_tree
*tree
,
408 struct extent_state
*state
, unsigned long *bits
);
411 * insert an extent_state struct into the tree. 'bits' are set on the
412 * struct before it is inserted.
414 * This may return -EEXIST if the extent is already there, in which case the
415 * state struct is freed.
417 * The tree lock is not taken internally. This is a utility function and
418 * probably isn't what you want to call (see set/clear_extent_bit).
420 static int insert_state(struct extent_io_tree
*tree
,
421 struct extent_state
*state
, u64 start
, u64 end
,
423 struct rb_node
**parent
,
426 struct rb_node
*node
;
429 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
431 state
->start
= start
;
434 set_state_bits(tree
, state
, bits
);
436 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
438 struct extent_state
*found
;
439 found
= rb_entry(node
, struct extent_state
, rb_node
);
440 printk(KERN_ERR
"BTRFS: found node %llu %llu on insert of "
442 found
->start
, found
->end
, start
, end
);
446 merge_state(tree
, state
);
450 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
453 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
454 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
458 * split a given extent state struct in two, inserting the preallocated
459 * struct 'prealloc' as the newly created second half. 'split' indicates an
460 * offset inside 'orig' where it should be split.
463 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
464 * are two extent state structs in the tree:
465 * prealloc: [orig->start, split - 1]
466 * orig: [ split, orig->end ]
468 * The tree locks are not taken by this function. They need to be held
471 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
472 struct extent_state
*prealloc
, u64 split
)
474 struct rb_node
*node
;
476 split_cb(tree
, orig
, split
);
478 prealloc
->start
= orig
->start
;
479 prealloc
->end
= split
- 1;
480 prealloc
->state
= orig
->state
;
483 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
484 &prealloc
->rb_node
, NULL
, NULL
);
486 free_extent_state(prealloc
);
489 prealloc
->tree
= tree
;
493 static struct extent_state
*next_state(struct extent_state
*state
)
495 struct rb_node
*next
= rb_next(&state
->rb_node
);
497 return rb_entry(next
, struct extent_state
, rb_node
);
503 * utility function to clear some bits in an extent state struct.
504 * it will optionally wake up any one waiting on this state (wake == 1).
506 * If no bits are set on the state struct after clearing things, the
507 * struct is freed and removed from the tree
509 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
510 struct extent_state
*state
,
511 unsigned long *bits
, int wake
)
513 struct extent_state
*next
;
514 unsigned long bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
516 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
517 u64 range
= state
->end
- state
->start
+ 1;
518 WARN_ON(range
> tree
->dirty_bytes
);
519 tree
->dirty_bytes
-= range
;
521 clear_state_cb(tree
, state
, bits
);
522 state
->state
&= ~bits_to_clear
;
525 if (state
->state
== 0) {
526 next
= next_state(state
);
528 rb_erase(&state
->rb_node
, &tree
->state
);
530 free_extent_state(state
);
535 merge_state(tree
, state
);
536 next
= next_state(state
);
541 static struct extent_state
*
542 alloc_extent_state_atomic(struct extent_state
*prealloc
)
545 prealloc
= alloc_extent_state(GFP_ATOMIC
);
550 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
552 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
553 "Extent tree was modified by another "
554 "thread while locked.");
558 * clear some bits on a range in the tree. This may require splitting
559 * or inserting elements in the tree, so the gfp mask is used to
560 * indicate which allocations or sleeping are allowed.
562 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
563 * the given range from the tree regardless of state (ie for truncate).
565 * the range [start, end] is inclusive.
567 * This takes the tree lock, and returns 0 on success and < 0 on error.
569 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
570 unsigned long bits
, int wake
, int delete,
571 struct extent_state
**cached_state
,
574 struct extent_state
*state
;
575 struct extent_state
*cached
;
576 struct extent_state
*prealloc
= NULL
;
577 struct rb_node
*node
;
582 btrfs_debug_check_extent_io_range(tree
, start
, end
);
584 if (bits
& EXTENT_DELALLOC
)
585 bits
|= EXTENT_NORESERVE
;
588 bits
|= ~EXTENT_CTLBITS
;
589 bits
|= EXTENT_FIRST_DELALLOC
;
591 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
594 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
595 prealloc
= alloc_extent_state(mask
);
600 spin_lock(&tree
->lock
);
602 cached
= *cached_state
;
605 *cached_state
= NULL
;
609 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
610 cached
->end
> start
) {
612 atomic_dec(&cached
->refs
);
617 free_extent_state(cached
);
620 * this search will find the extents that end after
623 node
= tree_search(tree
, start
);
626 state
= rb_entry(node
, struct extent_state
, rb_node
);
628 if (state
->start
> end
)
630 WARN_ON(state
->end
< start
);
631 last_end
= state
->end
;
633 /* the state doesn't have the wanted bits, go ahead */
634 if (!(state
->state
& bits
)) {
635 state
= next_state(state
);
640 * | ---- desired range ---- |
642 * | ------------- state -------------- |
644 * We need to split the extent we found, and may flip
645 * bits on second half.
647 * If the extent we found extends past our range, we
648 * just split and search again. It'll get split again
649 * the next time though.
651 * If the extent we found is inside our range, we clear
652 * the desired bit on it.
655 if (state
->start
< start
) {
656 prealloc
= alloc_extent_state_atomic(prealloc
);
658 err
= split_state(tree
, state
, prealloc
, start
);
660 extent_io_tree_panic(tree
, err
);
665 if (state
->end
<= end
) {
666 state
= clear_state_bit(tree
, state
, &bits
, wake
);
672 * | ---- desired range ---- |
674 * We need to split the extent, and clear the bit
677 if (state
->start
<= end
&& state
->end
> end
) {
678 prealloc
= alloc_extent_state_atomic(prealloc
);
680 err
= split_state(tree
, state
, prealloc
, end
+ 1);
682 extent_io_tree_panic(tree
, err
);
687 clear_state_bit(tree
, prealloc
, &bits
, wake
);
693 state
= clear_state_bit(tree
, state
, &bits
, wake
);
695 if (last_end
== (u64
)-1)
697 start
= last_end
+ 1;
698 if (start
<= end
&& state
&& !need_resched())
703 spin_unlock(&tree
->lock
);
705 free_extent_state(prealloc
);
712 spin_unlock(&tree
->lock
);
713 if (mask
& __GFP_WAIT
)
718 static void wait_on_state(struct extent_io_tree
*tree
,
719 struct extent_state
*state
)
720 __releases(tree
->lock
)
721 __acquires(tree
->lock
)
724 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
725 spin_unlock(&tree
->lock
);
727 spin_lock(&tree
->lock
);
728 finish_wait(&state
->wq
, &wait
);
732 * waits for one or more bits to clear on a range in the state tree.
733 * The range [start, end] is inclusive.
734 * The tree lock is taken by this function
736 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
739 struct extent_state
*state
;
740 struct rb_node
*node
;
742 btrfs_debug_check_extent_io_range(tree
, start
, end
);
744 spin_lock(&tree
->lock
);
748 * this search will find all the extents that end after
751 node
= tree_search(tree
, start
);
756 state
= rb_entry(node
, struct extent_state
, rb_node
);
758 if (state
->start
> end
)
761 if (state
->state
& bits
) {
762 start
= state
->start
;
763 atomic_inc(&state
->refs
);
764 wait_on_state(tree
, state
);
765 free_extent_state(state
);
768 start
= state
->end
+ 1;
773 if (!cond_resched_lock(&tree
->lock
)) {
774 node
= rb_next(node
);
779 spin_unlock(&tree
->lock
);
782 static void set_state_bits(struct extent_io_tree
*tree
,
783 struct extent_state
*state
,
786 unsigned long bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
788 set_state_cb(tree
, state
, bits
);
789 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
790 u64 range
= state
->end
- state
->start
+ 1;
791 tree
->dirty_bytes
+= range
;
793 state
->state
|= bits_to_set
;
796 static void cache_state(struct extent_state
*state
,
797 struct extent_state
**cached_ptr
)
799 if (cached_ptr
&& !(*cached_ptr
)) {
800 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
802 atomic_inc(&state
->refs
);
808 * set some bits on a range in the tree. This may require allocations or
809 * sleeping, so the gfp mask is used to indicate what is allowed.
811 * If any of the exclusive bits are set, this will fail with -EEXIST if some
812 * part of the range already has the desired bits set. The start of the
813 * existing range is returned in failed_start in this case.
815 * [start, end] is inclusive This takes the tree lock.
818 static int __must_check
819 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
820 unsigned long bits
, unsigned long exclusive_bits
,
821 u64
*failed_start
, struct extent_state
**cached_state
,
824 struct extent_state
*state
;
825 struct extent_state
*prealloc
= NULL
;
826 struct rb_node
*node
;
828 struct rb_node
*parent
;
833 btrfs_debug_check_extent_io_range(tree
, start
, end
);
835 bits
|= EXTENT_FIRST_DELALLOC
;
837 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
838 prealloc
= alloc_extent_state(mask
);
842 spin_lock(&tree
->lock
);
843 if (cached_state
&& *cached_state
) {
844 state
= *cached_state
;
845 if (state
->start
<= start
&& state
->end
> start
&&
847 node
= &state
->rb_node
;
852 * this search will find all the extents that end after
855 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
857 prealloc
= alloc_extent_state_atomic(prealloc
);
859 err
= insert_state(tree
, prealloc
, start
, end
,
862 extent_io_tree_panic(tree
, err
);
864 cache_state(prealloc
, cached_state
);
868 state
= rb_entry(node
, struct extent_state
, rb_node
);
870 last_start
= state
->start
;
871 last_end
= state
->end
;
874 * | ---- desired range ---- |
877 * Just lock what we found and keep going
879 if (state
->start
== start
&& state
->end
<= end
) {
880 if (state
->state
& exclusive_bits
) {
881 *failed_start
= state
->start
;
886 set_state_bits(tree
, state
, &bits
);
887 cache_state(state
, cached_state
);
888 merge_state(tree
, state
);
889 if (last_end
== (u64
)-1)
891 start
= last_end
+ 1;
892 state
= next_state(state
);
893 if (start
< end
&& state
&& state
->start
== start
&&
900 * | ---- desired range ---- |
903 * | ------------- state -------------- |
905 * We need to split the extent we found, and may flip bits on
908 * If the extent we found extends past our
909 * range, we just split and search again. It'll get split
910 * again the next time though.
912 * If the extent we found is inside our range, we set the
915 if (state
->start
< start
) {
916 if (state
->state
& exclusive_bits
) {
917 *failed_start
= start
;
922 prealloc
= alloc_extent_state_atomic(prealloc
);
924 err
= split_state(tree
, state
, prealloc
, start
);
926 extent_io_tree_panic(tree
, err
);
931 if (state
->end
<= end
) {
932 set_state_bits(tree
, state
, &bits
);
933 cache_state(state
, cached_state
);
934 merge_state(tree
, state
);
935 if (last_end
== (u64
)-1)
937 start
= last_end
+ 1;
938 state
= next_state(state
);
939 if (start
< end
&& state
&& state
->start
== start
&&
946 * | ---- desired range ---- |
947 * | state | or | state |
949 * There's a hole, we need to insert something in it and
950 * ignore the extent we found.
952 if (state
->start
> start
) {
954 if (end
< last_start
)
957 this_end
= last_start
- 1;
959 prealloc
= alloc_extent_state_atomic(prealloc
);
963 * Avoid to free 'prealloc' if it can be merged with
966 err
= insert_state(tree
, prealloc
, start
, this_end
,
969 extent_io_tree_panic(tree
, err
);
971 cache_state(prealloc
, cached_state
);
973 start
= this_end
+ 1;
977 * | ---- desired range ---- |
979 * We need to split the extent, and set the bit
982 if (state
->start
<= end
&& state
->end
> end
) {
983 if (state
->state
& exclusive_bits
) {
984 *failed_start
= start
;
989 prealloc
= alloc_extent_state_atomic(prealloc
);
991 err
= split_state(tree
, state
, prealloc
, end
+ 1);
993 extent_io_tree_panic(tree
, err
);
995 set_state_bits(tree
, prealloc
, &bits
);
996 cache_state(prealloc
, cached_state
);
997 merge_state(tree
, prealloc
);
1005 spin_unlock(&tree
->lock
);
1007 free_extent_state(prealloc
);
1014 spin_unlock(&tree
->lock
);
1015 if (mask
& __GFP_WAIT
)
1020 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1021 unsigned long bits
, u64
* failed_start
,
1022 struct extent_state
**cached_state
, gfp_t mask
)
1024 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1025 cached_state
, mask
);
1030 * convert_extent_bit - convert all bits in a given range from one bit to
1032 * @tree: the io tree to search
1033 * @start: the start offset in bytes
1034 * @end: the end offset in bytes (inclusive)
1035 * @bits: the bits to set in this range
1036 * @clear_bits: the bits to clear in this range
1037 * @cached_state: state that we're going to cache
1038 * @mask: the allocation mask
1040 * This will go through and set bits for the given range. If any states exist
1041 * already in this range they are set with the given bit and cleared of the
1042 * clear_bits. This is only meant to be used by things that are mergeable, ie
1043 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1044 * boundary bits like LOCK.
1046 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1047 unsigned long bits
, unsigned long clear_bits
,
1048 struct extent_state
**cached_state
, gfp_t mask
)
1050 struct extent_state
*state
;
1051 struct extent_state
*prealloc
= NULL
;
1052 struct rb_node
*node
;
1054 struct rb_node
*parent
;
1059 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1062 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
1063 prealloc
= alloc_extent_state(mask
);
1068 spin_lock(&tree
->lock
);
1069 if (cached_state
&& *cached_state
) {
1070 state
= *cached_state
;
1071 if (state
->start
<= start
&& state
->end
> start
&&
1073 node
= &state
->rb_node
;
1079 * this search will find all the extents that end after
1082 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1084 prealloc
= alloc_extent_state_atomic(prealloc
);
1089 err
= insert_state(tree
, prealloc
, start
, end
,
1090 &p
, &parent
, &bits
);
1092 extent_io_tree_panic(tree
, err
);
1093 cache_state(prealloc
, cached_state
);
1097 state
= rb_entry(node
, struct extent_state
, rb_node
);
1099 last_start
= state
->start
;
1100 last_end
= state
->end
;
1103 * | ---- desired range ---- |
1106 * Just lock what we found and keep going
1108 if (state
->start
== start
&& state
->end
<= end
) {
1109 set_state_bits(tree
, state
, &bits
);
1110 cache_state(state
, cached_state
);
1111 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1112 if (last_end
== (u64
)-1)
1114 start
= last_end
+ 1;
1115 if (start
< end
&& state
&& state
->start
== start
&&
1122 * | ---- desired range ---- |
1125 * | ------------- state -------------- |
1127 * We need to split the extent we found, and may flip bits on
1130 * If the extent we found extends past our
1131 * range, we just split and search again. It'll get split
1132 * again the next time though.
1134 * If the extent we found is inside our range, we set the
1135 * desired bit on it.
1137 if (state
->start
< start
) {
1138 prealloc
= alloc_extent_state_atomic(prealloc
);
1143 err
= split_state(tree
, state
, prealloc
, start
);
1145 extent_io_tree_panic(tree
, err
);
1149 if (state
->end
<= end
) {
1150 set_state_bits(tree
, state
, &bits
);
1151 cache_state(state
, cached_state
);
1152 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1153 if (last_end
== (u64
)-1)
1155 start
= last_end
+ 1;
1156 if (start
< end
&& state
&& state
->start
== start
&&
1163 * | ---- desired range ---- |
1164 * | state | or | state |
1166 * There's a hole, we need to insert something in it and
1167 * ignore the extent we found.
1169 if (state
->start
> start
) {
1171 if (end
< last_start
)
1174 this_end
= last_start
- 1;
1176 prealloc
= alloc_extent_state_atomic(prealloc
);
1183 * Avoid to free 'prealloc' if it can be merged with
1186 err
= insert_state(tree
, prealloc
, start
, this_end
,
1189 extent_io_tree_panic(tree
, err
);
1190 cache_state(prealloc
, cached_state
);
1192 start
= this_end
+ 1;
1196 * | ---- desired range ---- |
1198 * We need to split the extent, and set the bit
1201 if (state
->start
<= end
&& state
->end
> end
) {
1202 prealloc
= alloc_extent_state_atomic(prealloc
);
1208 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1210 extent_io_tree_panic(tree
, err
);
1212 set_state_bits(tree
, prealloc
, &bits
);
1213 cache_state(prealloc
, cached_state
);
1214 clear_state_bit(tree
, prealloc
, &clear_bits
, 0);
1222 spin_unlock(&tree
->lock
);
1224 free_extent_state(prealloc
);
1231 spin_unlock(&tree
->lock
);
1232 if (mask
& __GFP_WAIT
)
1237 /* wrappers around set/clear extent bit */
1238 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1241 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, NULL
,
1245 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1246 unsigned long bits
, gfp_t mask
)
1248 return set_extent_bit(tree
, start
, end
, bits
, NULL
,
1252 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1253 unsigned long bits
, gfp_t mask
)
1255 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
1258 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1259 struct extent_state
**cached_state
, gfp_t mask
)
1261 return set_extent_bit(tree
, start
, end
,
1262 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1263 NULL
, cached_state
, mask
);
1266 int set_extent_defrag(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1267 struct extent_state
**cached_state
, gfp_t mask
)
1269 return set_extent_bit(tree
, start
, end
,
1270 EXTENT_DELALLOC
| EXTENT_UPTODATE
| EXTENT_DEFRAG
,
1271 NULL
, cached_state
, mask
);
1274 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1277 return clear_extent_bit(tree
, start
, end
,
1278 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1279 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1282 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1285 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, NULL
,
1289 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1290 struct extent_state
**cached_state
, gfp_t mask
)
1292 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, NULL
,
1293 cached_state
, mask
);
1296 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1297 struct extent_state
**cached_state
, gfp_t mask
)
1299 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1300 cached_state
, mask
);
1304 * either insert or lock state struct between start and end use mask to tell
1305 * us if waiting is desired.
1307 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1308 unsigned long bits
, struct extent_state
**cached_state
)
1313 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1314 EXTENT_LOCKED
, &failed_start
,
1315 cached_state
, GFP_NOFS
);
1316 if (err
== -EEXIST
) {
1317 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1318 start
= failed_start
;
1321 WARN_ON(start
> end
);
1326 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1328 return lock_extent_bits(tree
, start
, end
, 0, NULL
);
1331 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1336 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1337 &failed_start
, NULL
, GFP_NOFS
);
1338 if (err
== -EEXIST
) {
1339 if (failed_start
> start
)
1340 clear_extent_bit(tree
, start
, failed_start
- 1,
1341 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1347 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1348 struct extent_state
**cached
, gfp_t mask
)
1350 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1354 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1356 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1360 int extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1362 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1363 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1366 while (index
<= end_index
) {
1367 page
= find_get_page(inode
->i_mapping
, index
);
1368 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1369 clear_page_dirty_for_io(page
);
1370 page_cache_release(page
);
1376 int extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1378 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1379 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1382 while (index
<= end_index
) {
1383 page
= find_get_page(inode
->i_mapping
, index
);
1384 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1385 account_page_redirty(page
);
1386 __set_page_dirty_nobuffers(page
);
1387 page_cache_release(page
);
1394 * helper function to set both pages and extents in the tree writeback
1396 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1398 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1399 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1402 while (index
<= end_index
) {
1403 page
= find_get_page(tree
->mapping
, index
);
1404 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1405 set_page_writeback(page
);
1406 page_cache_release(page
);
1412 /* find the first state struct with 'bits' set after 'start', and
1413 * return it. tree->lock must be held. NULL will returned if
1414 * nothing was found after 'start'
1416 static struct extent_state
*
1417 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1418 u64 start
, unsigned long bits
)
1420 struct rb_node
*node
;
1421 struct extent_state
*state
;
1424 * this search will find all the extents that end after
1427 node
= tree_search(tree
, start
);
1432 state
= rb_entry(node
, struct extent_state
, rb_node
);
1433 if (state
->end
>= start
&& (state
->state
& bits
))
1436 node
= rb_next(node
);
1445 * find the first offset in the io tree with 'bits' set. zero is
1446 * returned if we find something, and *start_ret and *end_ret are
1447 * set to reflect the state struct that was found.
1449 * If nothing was found, 1 is returned. If found something, return 0.
1451 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1452 u64
*start_ret
, u64
*end_ret
, unsigned long bits
,
1453 struct extent_state
**cached_state
)
1455 struct extent_state
*state
;
1459 spin_lock(&tree
->lock
);
1460 if (cached_state
&& *cached_state
) {
1461 state
= *cached_state
;
1462 if (state
->end
== start
- 1 && state
->tree
) {
1463 n
= rb_next(&state
->rb_node
);
1465 state
= rb_entry(n
, struct extent_state
,
1467 if (state
->state
& bits
)
1471 free_extent_state(*cached_state
);
1472 *cached_state
= NULL
;
1475 free_extent_state(*cached_state
);
1476 *cached_state
= NULL
;
1479 state
= find_first_extent_bit_state(tree
, start
, bits
);
1482 cache_state(state
, cached_state
);
1483 *start_ret
= state
->start
;
1484 *end_ret
= state
->end
;
1488 spin_unlock(&tree
->lock
);
1493 * find a contiguous range of bytes in the file marked as delalloc, not
1494 * more than 'max_bytes'. start and end are used to return the range,
1496 * 1 is returned if we find something, 0 if nothing was in the tree
1498 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1499 u64
*start
, u64
*end
, u64 max_bytes
,
1500 struct extent_state
**cached_state
)
1502 struct rb_node
*node
;
1503 struct extent_state
*state
;
1504 u64 cur_start
= *start
;
1506 u64 total_bytes
= 0;
1508 spin_lock(&tree
->lock
);
1511 * this search will find all the extents that end after
1514 node
= tree_search(tree
, cur_start
);
1522 state
= rb_entry(node
, struct extent_state
, rb_node
);
1523 if (found
&& (state
->start
!= cur_start
||
1524 (state
->state
& EXTENT_BOUNDARY
))) {
1527 if (!(state
->state
& EXTENT_DELALLOC
)) {
1533 *start
= state
->start
;
1534 *cached_state
= state
;
1535 atomic_inc(&state
->refs
);
1539 cur_start
= state
->end
+ 1;
1540 node
= rb_next(node
);
1541 total_bytes
+= state
->end
- state
->start
+ 1;
1542 if (total_bytes
>= max_bytes
)
1548 spin_unlock(&tree
->lock
);
1552 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1553 struct page
*locked_page
,
1557 struct page
*pages
[16];
1558 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1559 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1560 unsigned long nr_pages
= end_index
- index
+ 1;
1563 if (index
== locked_page
->index
&& end_index
== index
)
1566 while (nr_pages
> 0) {
1567 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1568 min_t(unsigned long, nr_pages
,
1569 ARRAY_SIZE(pages
)), pages
);
1570 for (i
= 0; i
< ret
; i
++) {
1571 if (pages
[i
] != locked_page
)
1572 unlock_page(pages
[i
]);
1573 page_cache_release(pages
[i
]);
1581 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1582 struct page
*locked_page
,
1586 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1587 unsigned long start_index
= index
;
1588 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1589 unsigned long pages_locked
= 0;
1590 struct page
*pages
[16];
1591 unsigned long nrpages
;
1595 /* the caller is responsible for locking the start index */
1596 if (index
== locked_page
->index
&& index
== end_index
)
1599 /* skip the page at the start index */
1600 nrpages
= end_index
- index
+ 1;
1601 while (nrpages
> 0) {
1602 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1603 min_t(unsigned long,
1604 nrpages
, ARRAY_SIZE(pages
)), pages
);
1609 /* now we have an array of pages, lock them all */
1610 for (i
= 0; i
< ret
; i
++) {
1612 * the caller is taking responsibility for
1615 if (pages
[i
] != locked_page
) {
1616 lock_page(pages
[i
]);
1617 if (!PageDirty(pages
[i
]) ||
1618 pages
[i
]->mapping
!= inode
->i_mapping
) {
1620 unlock_page(pages
[i
]);
1621 page_cache_release(pages
[i
]);
1625 page_cache_release(pages
[i
]);
1634 if (ret
&& pages_locked
) {
1635 __unlock_for_delalloc(inode
, locked_page
,
1637 ((u64
)(start_index
+ pages_locked
- 1)) <<
1644 * find a contiguous range of bytes in the file marked as delalloc, not
1645 * more than 'max_bytes'. start and end are used to return the range,
1647 * 1 is returned if we find something, 0 if nothing was in the tree
1649 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1650 struct extent_io_tree
*tree
,
1651 struct page
*locked_page
, u64
*start
,
1652 u64
*end
, u64 max_bytes
)
1657 struct extent_state
*cached_state
= NULL
;
1662 /* step one, find a bunch of delalloc bytes starting at start */
1663 delalloc_start
= *start
;
1665 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1666 max_bytes
, &cached_state
);
1667 if (!found
|| delalloc_end
<= *start
) {
1668 *start
= delalloc_start
;
1669 *end
= delalloc_end
;
1670 free_extent_state(cached_state
);
1675 * start comes from the offset of locked_page. We have to lock
1676 * pages in order, so we can't process delalloc bytes before
1679 if (delalloc_start
< *start
)
1680 delalloc_start
= *start
;
1683 * make sure to limit the number of pages we try to lock down
1685 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1686 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1688 /* step two, lock all the pages after the page that has start */
1689 ret
= lock_delalloc_pages(inode
, locked_page
,
1690 delalloc_start
, delalloc_end
);
1691 if (ret
== -EAGAIN
) {
1692 /* some of the pages are gone, lets avoid looping by
1693 * shortening the size of the delalloc range we're searching
1695 free_extent_state(cached_state
);
1696 cached_state
= NULL
;
1698 max_bytes
= PAGE_CACHE_SIZE
;
1706 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1708 /* step three, lock the state bits for the whole range */
1709 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1711 /* then test to make sure it is all still delalloc */
1712 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1713 EXTENT_DELALLOC
, 1, cached_state
);
1715 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1716 &cached_state
, GFP_NOFS
);
1717 __unlock_for_delalloc(inode
, locked_page
,
1718 delalloc_start
, delalloc_end
);
1722 free_extent_state(cached_state
);
1723 *start
= delalloc_start
;
1724 *end
= delalloc_end
;
1729 int extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1730 struct page
*locked_page
,
1731 unsigned long clear_bits
,
1732 unsigned long page_ops
)
1734 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1736 struct page
*pages
[16];
1737 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1738 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1739 unsigned long nr_pages
= end_index
- index
+ 1;
1742 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1746 while (nr_pages
> 0) {
1747 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1748 min_t(unsigned long,
1749 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1750 for (i
= 0; i
< ret
; i
++) {
1752 if (page_ops
& PAGE_SET_PRIVATE2
)
1753 SetPagePrivate2(pages
[i
]);
1755 if (pages
[i
] == locked_page
) {
1756 page_cache_release(pages
[i
]);
1759 if (page_ops
& PAGE_CLEAR_DIRTY
)
1760 clear_page_dirty_for_io(pages
[i
]);
1761 if (page_ops
& PAGE_SET_WRITEBACK
)
1762 set_page_writeback(pages
[i
]);
1763 if (page_ops
& PAGE_END_WRITEBACK
)
1764 end_page_writeback(pages
[i
]);
1765 if (page_ops
& PAGE_UNLOCK
)
1766 unlock_page(pages
[i
]);
1767 page_cache_release(pages
[i
]);
1777 * count the number of bytes in the tree that have a given bit(s)
1778 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1779 * cached. The total number found is returned.
1781 u64
count_range_bits(struct extent_io_tree
*tree
,
1782 u64
*start
, u64 search_end
, u64 max_bytes
,
1783 unsigned long bits
, int contig
)
1785 struct rb_node
*node
;
1786 struct extent_state
*state
;
1787 u64 cur_start
= *start
;
1788 u64 total_bytes
= 0;
1792 if (WARN_ON(search_end
<= cur_start
))
1795 spin_lock(&tree
->lock
);
1796 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1797 total_bytes
= tree
->dirty_bytes
;
1801 * this search will find all the extents that end after
1804 node
= tree_search(tree
, cur_start
);
1809 state
= rb_entry(node
, struct extent_state
, rb_node
);
1810 if (state
->start
> search_end
)
1812 if (contig
&& found
&& state
->start
> last
+ 1)
1814 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1815 total_bytes
+= min(search_end
, state
->end
) + 1 -
1816 max(cur_start
, state
->start
);
1817 if (total_bytes
>= max_bytes
)
1820 *start
= max(cur_start
, state
->start
);
1824 } else if (contig
&& found
) {
1827 node
= rb_next(node
);
1832 spin_unlock(&tree
->lock
);
1837 * set the private field for a given byte offset in the tree. If there isn't
1838 * an extent_state there already, this does nothing.
1840 static int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1842 struct rb_node
*node
;
1843 struct extent_state
*state
;
1846 spin_lock(&tree
->lock
);
1848 * this search will find all the extents that end after
1851 node
= tree_search(tree
, start
);
1856 state
= rb_entry(node
, struct extent_state
, rb_node
);
1857 if (state
->start
!= start
) {
1861 state
->private = private;
1863 spin_unlock(&tree
->lock
);
1867 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1869 struct rb_node
*node
;
1870 struct extent_state
*state
;
1873 spin_lock(&tree
->lock
);
1875 * this search will find all the extents that end after
1878 node
= tree_search(tree
, start
);
1883 state
= rb_entry(node
, struct extent_state
, rb_node
);
1884 if (state
->start
!= start
) {
1888 *private = state
->private;
1890 spin_unlock(&tree
->lock
);
1895 * searches a range in the state tree for a given mask.
1896 * If 'filled' == 1, this returns 1 only if every extent in the tree
1897 * has the bits set. Otherwise, 1 is returned if any bit in the
1898 * range is found set.
1900 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1901 unsigned long bits
, int filled
, struct extent_state
*cached
)
1903 struct extent_state
*state
= NULL
;
1904 struct rb_node
*node
;
1907 spin_lock(&tree
->lock
);
1908 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1909 cached
->end
> start
)
1910 node
= &cached
->rb_node
;
1912 node
= tree_search(tree
, start
);
1913 while (node
&& start
<= end
) {
1914 state
= rb_entry(node
, struct extent_state
, rb_node
);
1916 if (filled
&& state
->start
> start
) {
1921 if (state
->start
> end
)
1924 if (state
->state
& bits
) {
1928 } else if (filled
) {
1933 if (state
->end
== (u64
)-1)
1936 start
= state
->end
+ 1;
1939 node
= rb_next(node
);
1946 spin_unlock(&tree
->lock
);
1951 * helper function to set a given page up to date if all the
1952 * extents in the tree for that page are up to date
1954 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1956 u64 start
= page_offset(page
);
1957 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1958 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1959 SetPageUptodate(page
);
1963 * When IO fails, either with EIO or csum verification fails, we
1964 * try other mirrors that might have a good copy of the data. This
1965 * io_failure_record is used to record state as we go through all the
1966 * mirrors. If another mirror has good data, the page is set up to date
1967 * and things continue. If a good mirror can't be found, the original
1968 * bio end_io callback is called to indicate things have failed.
1970 struct io_failure_record
{
1975 unsigned long bio_flags
;
1981 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
1986 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1988 set_state_private(failure_tree
, rec
->start
, 0);
1989 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1990 rec
->start
+ rec
->len
- 1,
1991 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1995 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1996 rec
->start
+ rec
->len
- 1,
1997 EXTENT_DAMAGED
, GFP_NOFS
);
2006 * this bypasses the standard btrfs submit functions deliberately, as
2007 * the standard behavior is to write all copies in a raid setup. here we only
2008 * want to write the one bad copy. so we do the mapping for ourselves and issue
2009 * submit_bio directly.
2010 * to avoid any synchronization issues, wait for the data after writing, which
2011 * actually prevents the read that triggered the error from finishing.
2012 * currently, there can be no more than two copies of every data bit. thus,
2013 * exactly one rewrite is required.
2015 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 start
,
2016 u64 length
, u64 logical
, struct page
*page
,
2020 struct btrfs_device
*dev
;
2023 struct btrfs_bio
*bbio
= NULL
;
2024 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2027 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2028 BUG_ON(!mirror_num
);
2030 /* we can't repair anything in raid56 yet */
2031 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2034 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2037 bio
->bi_iter
.bi_size
= 0;
2038 map_length
= length
;
2040 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2041 &map_length
, &bbio
, mirror_num
);
2046 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2047 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2048 bio
->bi_iter
.bi_sector
= sector
;
2049 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2051 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2055 bio
->bi_bdev
= dev
->bdev
;
2056 bio_add_page(bio
, page
, length
, start
- page_offset(page
));
2058 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
)) {
2059 /* try to remap that extent elsewhere? */
2061 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2065 printk_ratelimited_in_rcu(KERN_INFO
2066 "BTRFS: read error corrected: ino %lu off %llu "
2067 "(dev %s sector %llu)\n", page
->mapping
->host
->i_ino
,
2068 start
, rcu_str_deref(dev
->name
), sector
);
2074 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2077 u64 start
= eb
->start
;
2078 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2081 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2084 for (i
= 0; i
< num_pages
; i
++) {
2085 struct page
*p
= extent_buffer_page(eb
, i
);
2086 ret
= repair_io_failure(root
->fs_info
, start
, PAGE_CACHE_SIZE
,
2087 start
, p
, mirror_num
);
2090 start
+= PAGE_CACHE_SIZE
;
2097 * each time an IO finishes, we do a fast check in the IO failure tree
2098 * to see if we need to process or clean up an io_failure_record
2100 static int clean_io_failure(u64 start
, struct page
*page
)
2103 u64 private_failure
;
2104 struct io_failure_record
*failrec
;
2105 struct inode
*inode
= page
->mapping
->host
;
2106 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2107 struct extent_state
*state
;
2113 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2114 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2118 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2123 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2124 BUG_ON(!failrec
->this_mirror
);
2126 if (failrec
->in_validation
) {
2127 /* there was no real error, just free the record */
2128 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2133 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2136 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2137 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2140 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2142 if (state
&& state
->start
<= failrec
->start
&&
2143 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2144 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2146 if (num_copies
> 1) {
2147 ret
= repair_io_failure(fs_info
, start
, failrec
->len
,
2148 failrec
->logical
, page
,
2149 failrec
->failed_mirror
);
2157 ret
= free_io_failure(inode
, failrec
, did_repair
);
2163 * this is a generic handler for readpage errors (default
2164 * readpage_io_failed_hook). if other copies exist, read those and write back
2165 * good data to the failed position. does not investigate in remapping the
2166 * failed extent elsewhere, hoping the device will be smart enough to do this as
2170 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2171 struct page
*page
, u64 start
, u64 end
,
2174 struct io_failure_record
*failrec
= NULL
;
2176 struct extent_map
*em
;
2177 struct inode
*inode
= page
->mapping
->host
;
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
;
2182 struct btrfs_io_bio
*btrfs_failed_bio
;
2183 struct btrfs_io_bio
*btrfs_bio
;
2189 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2191 ret
= get_state_private(failure_tree
, start
, &private);
2193 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2196 failrec
->start
= start
;
2197 failrec
->len
= end
- start
+ 1;
2198 failrec
->this_mirror
= 0;
2199 failrec
->bio_flags
= 0;
2200 failrec
->in_validation
= 0;
2202 read_lock(&em_tree
->lock
);
2203 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2205 read_unlock(&em_tree
->lock
);
2210 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2211 free_extent_map(em
);
2214 read_unlock(&em_tree
->lock
);
2220 logical
= start
- em
->start
;
2221 logical
= em
->block_start
+ logical
;
2222 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2223 logical
= em
->block_start
;
2224 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2225 extent_set_compress_type(&failrec
->bio_flags
,
2228 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2229 "len=%llu\n", logical
, start
, failrec
->len
);
2230 failrec
->logical
= logical
;
2231 free_extent_map(em
);
2233 /* set the bits in the private failure tree */
2234 ret
= set_extent_bits(failure_tree
, start
, end
,
2235 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2237 ret
= set_state_private(failure_tree
, start
,
2238 (u64
)(unsigned long)failrec
);
2239 /* set the bits in the inode's tree */
2241 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2248 failrec
= (struct io_failure_record
*)(unsigned long)private;
2249 pr_debug("bio_readpage_error: (found) logical=%llu, "
2250 "start=%llu, len=%llu, validation=%d\n",
2251 failrec
->logical
, failrec
->start
, failrec
->len
,
2252 failrec
->in_validation
);
2254 * when data can be on disk more than twice, add to failrec here
2255 * (e.g. with a list for failed_mirror) to make
2256 * clean_io_failure() clean all those errors at once.
2259 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2260 failrec
->logical
, failrec
->len
);
2261 if (num_copies
== 1) {
2263 * we only have a single copy of the data, so don't bother with
2264 * all the retry and error correction code that follows. no
2265 * matter what the error is, it is very likely to persist.
2267 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2268 num_copies
, failrec
->this_mirror
, failed_mirror
);
2269 free_io_failure(inode
, failrec
, 0);
2274 * there are two premises:
2275 * a) deliver good data to the caller
2276 * b) correct the bad sectors on disk
2278 if (failed_bio
->bi_vcnt
> 1) {
2280 * to fulfill b), we need to know the exact failing sectors, as
2281 * we don't want to rewrite any more than the failed ones. thus,
2282 * we need separate read requests for the failed bio
2284 * if the following BUG_ON triggers, our validation request got
2285 * merged. we need separate requests for our algorithm to work.
2287 BUG_ON(failrec
->in_validation
);
2288 failrec
->in_validation
= 1;
2289 failrec
->this_mirror
= failed_mirror
;
2290 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2293 * we're ready to fulfill a) and b) alongside. get a good copy
2294 * of the failed sector and if we succeed, we have setup
2295 * everything for repair_io_failure to do the rest for us.
2297 if (failrec
->in_validation
) {
2298 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2299 failrec
->in_validation
= 0;
2300 failrec
->this_mirror
= 0;
2302 failrec
->failed_mirror
= failed_mirror
;
2303 failrec
->this_mirror
++;
2304 if (failrec
->this_mirror
== failed_mirror
)
2305 failrec
->this_mirror
++;
2306 read_mode
= READ_SYNC
;
2309 if (failrec
->this_mirror
> num_copies
) {
2310 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2311 num_copies
, failrec
->this_mirror
, failed_mirror
);
2312 free_io_failure(inode
, failrec
, 0);
2316 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2318 free_io_failure(inode
, failrec
, 0);
2321 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2322 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2323 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2324 bio
->bi_iter
.bi_size
= 0;
2326 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2327 if (btrfs_failed_bio
->csum
) {
2328 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2329 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2331 btrfs_bio
= btrfs_io_bio(bio
);
2332 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2333 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2334 phy_offset
*= csum_size
;
2335 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ phy_offset
,
2339 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2341 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2342 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2343 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2345 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2346 failrec
->this_mirror
,
2347 failrec
->bio_flags
, 0);
2351 /* lots and lots of room for performance fixes in the end_bio funcs */
2353 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2355 int uptodate
= (err
== 0);
2356 struct extent_io_tree
*tree
;
2359 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2361 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2362 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2363 end
, NULL
, uptodate
);
2369 ClearPageUptodate(page
);
2371 ret
= ret
< 0 ? ret
: -EIO
;
2372 mapping_set_error(page
->mapping
, ret
);
2378 * after a writepage IO is done, we need to:
2379 * clear the uptodate bits on error
2380 * clear the writeback bits in the extent tree for this IO
2381 * end_page_writeback if the page has no more pending IO
2383 * Scheduling is not allowed, so the extent state tree is expected
2384 * to have one and only one object corresponding to this IO.
2386 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2388 struct bio_vec
*bvec
;
2393 bio_for_each_segment_all(bvec
, bio
, i
) {
2394 struct page
*page
= bvec
->bv_page
;
2396 /* We always issue full-page reads, but if some block
2397 * in a page fails to read, blk_update_request() will
2398 * advance bv_offset and adjust bv_len to compensate.
2399 * Print a warning for nonzero offsets, and an error
2400 * if they don't add up to a full page. */
2401 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2402 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2403 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2404 "partial page write in btrfs with offset %u and length %u",
2405 bvec
->bv_offset
, bvec
->bv_len
);
2407 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2408 "incomplete page write in btrfs with offset %u and "
2410 bvec
->bv_offset
, bvec
->bv_len
);
2413 start
= page_offset(page
);
2414 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2416 if (end_extent_writepage(page
, err
, start
, end
))
2419 end_page_writeback(page
);
2426 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2429 struct extent_state
*cached
= NULL
;
2430 u64 end
= start
+ len
- 1;
2432 if (uptodate
&& tree
->track_uptodate
)
2433 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2434 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2438 * after a readpage IO is done, we need to:
2439 * clear the uptodate bits on error
2440 * set the uptodate bits if things worked
2441 * set the page up to date if all extents in the tree are uptodate
2442 * clear the lock bit in the extent tree
2443 * unlock the page if there are no other extents locked for it
2445 * Scheduling is not allowed, so the extent state tree is expected
2446 * to have one and only one object corresponding to this IO.
2448 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2450 struct bio_vec
*bvec
;
2451 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2452 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2453 struct extent_io_tree
*tree
;
2458 u64 extent_start
= 0;
2467 bio_for_each_segment_all(bvec
, bio
, i
) {
2468 struct page
*page
= bvec
->bv_page
;
2469 struct inode
*inode
= page
->mapping
->host
;
2471 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2472 "mirror=%lu\n", (u64
)bio
->bi_iter
.bi_sector
, err
,
2473 io_bio
->mirror_num
);
2474 tree
= &BTRFS_I(inode
)->io_tree
;
2476 /* We always issue full-page reads, but if some block
2477 * in a page fails to read, blk_update_request() will
2478 * advance bv_offset and adjust bv_len to compensate.
2479 * Print a warning for nonzero offsets, and an error
2480 * if they don't add up to a full page. */
2481 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2482 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2483 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2484 "partial page read in btrfs with offset %u and length %u",
2485 bvec
->bv_offset
, bvec
->bv_len
);
2487 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2488 "incomplete page read in btrfs with offset %u and "
2490 bvec
->bv_offset
, bvec
->bv_len
);
2493 start
= page_offset(page
);
2494 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2497 mirror
= io_bio
->mirror_num
;
2498 if (likely(uptodate
&& tree
->ops
&&
2499 tree
->ops
->readpage_end_io_hook
)) {
2500 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2506 clean_io_failure(start
, page
);
2509 if (likely(uptodate
))
2512 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2513 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2515 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2519 * The generic bio_readpage_error handles errors the
2520 * following way: If possible, new read requests are
2521 * created and submitted and will end up in
2522 * end_bio_extent_readpage as well (if we're lucky, not
2523 * in the !uptodate case). In that case it returns 0 and
2524 * we just go on with the next page in our bio. If it
2525 * can't handle the error it will return -EIO and we
2526 * remain responsible for that page.
2528 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2532 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2539 if (likely(uptodate
)) {
2540 loff_t i_size
= i_size_read(inode
);
2541 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2544 /* Zero out the end if this page straddles i_size */
2545 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2546 if (page
->index
== end_index
&& offset
)
2547 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2548 SetPageUptodate(page
);
2550 ClearPageUptodate(page
);
2556 if (unlikely(!uptodate
)) {
2558 endio_readpage_release_extent(tree
,
2564 endio_readpage_release_extent(tree
, start
,
2565 end
- start
+ 1, 0);
2566 } else if (!extent_len
) {
2567 extent_start
= start
;
2568 extent_len
= end
+ 1 - start
;
2569 } else if (extent_start
+ extent_len
== start
) {
2570 extent_len
+= end
+ 1 - start
;
2572 endio_readpage_release_extent(tree
, extent_start
,
2573 extent_len
, uptodate
);
2574 extent_start
= start
;
2575 extent_len
= end
+ 1 - start
;
2580 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2583 io_bio
->end_io(io_bio
, err
);
2588 * this allocates from the btrfs_bioset. We're returning a bio right now
2589 * but you can call btrfs_io_bio for the appropriate container_of magic
2592 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2595 struct btrfs_io_bio
*btrfs_bio
;
2598 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2600 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2601 while (!bio
&& (nr_vecs
/= 2)) {
2602 bio
= bio_alloc_bioset(gfp_flags
,
2603 nr_vecs
, btrfs_bioset
);
2608 bio
->bi_bdev
= bdev
;
2609 bio
->bi_iter
.bi_sector
= first_sector
;
2610 btrfs_bio
= btrfs_io_bio(bio
);
2611 btrfs_bio
->csum
= NULL
;
2612 btrfs_bio
->csum_allocated
= NULL
;
2613 btrfs_bio
->end_io
= NULL
;
2618 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2620 return bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2624 /* this also allocates from the btrfs_bioset */
2625 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2627 struct btrfs_io_bio
*btrfs_bio
;
2630 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2632 btrfs_bio
= btrfs_io_bio(bio
);
2633 btrfs_bio
->csum
= NULL
;
2634 btrfs_bio
->csum_allocated
= NULL
;
2635 btrfs_bio
->end_io
= NULL
;
2641 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2642 int mirror_num
, unsigned long bio_flags
)
2645 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2646 struct page
*page
= bvec
->bv_page
;
2647 struct extent_io_tree
*tree
= bio
->bi_private
;
2650 start
= page_offset(page
) + bvec
->bv_offset
;
2652 bio
->bi_private
= NULL
;
2656 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2657 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2658 mirror_num
, bio_flags
, start
);
2660 btrfsic_submit_bio(rw
, bio
);
2662 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2668 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2669 unsigned long offset
, size_t size
, struct bio
*bio
,
2670 unsigned long bio_flags
)
2673 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2674 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2681 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2682 struct page
*page
, sector_t sector
,
2683 size_t size
, unsigned long offset
,
2684 struct block_device
*bdev
,
2685 struct bio
**bio_ret
,
2686 unsigned long max_pages
,
2687 bio_end_io_t end_io_func
,
2689 unsigned long prev_bio_flags
,
2690 unsigned long bio_flags
)
2696 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2697 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2698 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2700 if (bio_ret
&& *bio_ret
) {
2703 contig
= bio
->bi_iter
.bi_sector
== sector
;
2705 contig
= bio_end_sector(bio
) == sector
;
2707 if (prev_bio_flags
!= bio_flags
|| !contig
||
2708 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2709 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2710 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2719 if (this_compressed
)
2722 nr
= bio_get_nr_vecs(bdev
);
2724 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2728 bio_add_page(bio
, page
, page_size
, offset
);
2729 bio
->bi_end_io
= end_io_func
;
2730 bio
->bi_private
= tree
;
2735 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2740 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2743 if (!PagePrivate(page
)) {
2744 SetPagePrivate(page
);
2745 page_cache_get(page
);
2746 set_page_private(page
, (unsigned long)eb
);
2748 WARN_ON(page
->private != (unsigned long)eb
);
2752 void set_page_extent_mapped(struct page
*page
)
2754 if (!PagePrivate(page
)) {
2755 SetPagePrivate(page
);
2756 page_cache_get(page
);
2757 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2761 static struct extent_map
*
2762 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2763 u64 start
, u64 len
, get_extent_t
*get_extent
,
2764 struct extent_map
**em_cached
)
2766 struct extent_map
*em
;
2768 if (em_cached
&& *em_cached
) {
2770 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2771 start
< extent_map_end(em
)) {
2772 atomic_inc(&em
->refs
);
2776 free_extent_map(em
);
2780 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2781 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2783 atomic_inc(&em
->refs
);
2789 * basic readpage implementation. Locked extent state structs are inserted
2790 * into the tree that are removed when the IO is done (by the end_io
2792 * XXX JDM: This needs looking at to ensure proper page locking
2794 static int __do_readpage(struct extent_io_tree
*tree
,
2796 get_extent_t
*get_extent
,
2797 struct extent_map
**em_cached
,
2798 struct bio
**bio
, int mirror_num
,
2799 unsigned long *bio_flags
, int rw
)
2801 struct inode
*inode
= page
->mapping
->host
;
2802 u64 start
= page_offset(page
);
2803 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2807 u64 last_byte
= i_size_read(inode
);
2811 struct extent_map
*em
;
2812 struct block_device
*bdev
;
2815 int parent_locked
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2816 size_t pg_offset
= 0;
2818 size_t disk_io_size
;
2819 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2820 unsigned long this_bio_flag
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2822 set_page_extent_mapped(page
);
2825 if (!PageUptodate(page
)) {
2826 if (cleancache_get_page(page
) == 0) {
2827 BUG_ON(blocksize
!= PAGE_SIZE
);
2828 unlock_extent(tree
, start
, end
);
2833 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2835 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2838 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2839 userpage
= kmap_atomic(page
);
2840 memset(userpage
+ zero_offset
, 0, iosize
);
2841 flush_dcache_page(page
);
2842 kunmap_atomic(userpage
);
2845 while (cur
<= end
) {
2846 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2848 if (cur
>= last_byte
) {
2850 struct extent_state
*cached
= NULL
;
2852 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2853 userpage
= kmap_atomic(page
);
2854 memset(userpage
+ pg_offset
, 0, iosize
);
2855 flush_dcache_page(page
);
2856 kunmap_atomic(userpage
);
2857 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2860 unlock_extent_cached(tree
, cur
,
2865 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2866 end
- cur
+ 1, get_extent
, em_cached
);
2867 if (IS_ERR_OR_NULL(em
)) {
2870 unlock_extent(tree
, cur
, end
);
2873 extent_offset
= cur
- em
->start
;
2874 BUG_ON(extent_map_end(em
) <= cur
);
2877 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2878 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2879 extent_set_compress_type(&this_bio_flag
,
2883 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2884 cur_end
= min(extent_map_end(em
) - 1, end
);
2885 iosize
= ALIGN(iosize
, blocksize
);
2886 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2887 disk_io_size
= em
->block_len
;
2888 sector
= em
->block_start
>> 9;
2890 sector
= (em
->block_start
+ extent_offset
) >> 9;
2891 disk_io_size
= iosize
;
2894 block_start
= em
->block_start
;
2895 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2896 block_start
= EXTENT_MAP_HOLE
;
2897 free_extent_map(em
);
2900 /* we've found a hole, just zero and go on */
2901 if (block_start
== EXTENT_MAP_HOLE
) {
2903 struct extent_state
*cached
= NULL
;
2905 userpage
= kmap_atomic(page
);
2906 memset(userpage
+ pg_offset
, 0, iosize
);
2907 flush_dcache_page(page
);
2908 kunmap_atomic(userpage
);
2910 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2912 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2915 pg_offset
+= iosize
;
2918 /* the get_extent function already copied into the page */
2919 if (test_range_bit(tree
, cur
, cur_end
,
2920 EXTENT_UPTODATE
, 1, NULL
)) {
2921 check_page_uptodate(tree
, page
);
2923 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2925 pg_offset
+= iosize
;
2928 /* we have an inline extent but it didn't get marked up
2929 * to date. Error out
2931 if (block_start
== EXTENT_MAP_INLINE
) {
2934 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2936 pg_offset
+= iosize
;
2941 ret
= submit_extent_page(rw
, tree
, page
,
2942 sector
, disk_io_size
, pg_offset
,
2944 end_bio_extent_readpage
, mirror_num
,
2949 *bio_flags
= this_bio_flag
;
2953 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2956 pg_offset
+= iosize
;
2960 if (!PageError(page
))
2961 SetPageUptodate(page
);
2967 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
2968 struct page
*pages
[], int nr_pages
,
2970 get_extent_t
*get_extent
,
2971 struct extent_map
**em_cached
,
2972 struct bio
**bio
, int mirror_num
,
2973 unsigned long *bio_flags
, int rw
)
2975 struct inode
*inode
;
2976 struct btrfs_ordered_extent
*ordered
;
2979 inode
= pages
[0]->mapping
->host
;
2981 lock_extent(tree
, start
, end
);
2982 ordered
= btrfs_lookup_ordered_range(inode
, start
,
2986 unlock_extent(tree
, start
, end
);
2987 btrfs_start_ordered_extent(inode
, ordered
, 1);
2988 btrfs_put_ordered_extent(ordered
);
2991 for (index
= 0; index
< nr_pages
; index
++) {
2992 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
2993 mirror_num
, bio_flags
, rw
);
2994 page_cache_release(pages
[index
]);
2998 static void __extent_readpages(struct extent_io_tree
*tree
,
2999 struct page
*pages
[],
3000 int nr_pages
, get_extent_t
*get_extent
,
3001 struct extent_map
**em_cached
,
3002 struct bio
**bio
, int mirror_num
,
3003 unsigned long *bio_flags
, int rw
)
3009 int first_index
= 0;
3011 for (index
= 0; index
< nr_pages
; index
++) {
3012 page_start
= page_offset(pages
[index
]);
3015 end
= start
+ PAGE_CACHE_SIZE
- 1;
3016 first_index
= index
;
3017 } else if (end
+ 1 == page_start
) {
3018 end
+= PAGE_CACHE_SIZE
;
3020 __do_contiguous_readpages(tree
, &pages
[first_index
],
3021 index
- first_index
, start
,
3022 end
, get_extent
, em_cached
,
3023 bio
, mirror_num
, bio_flags
,
3026 end
= start
+ PAGE_CACHE_SIZE
- 1;
3027 first_index
= index
;
3032 __do_contiguous_readpages(tree
, &pages
[first_index
],
3033 index
- first_index
, start
,
3034 end
, get_extent
, em_cached
, bio
,
3035 mirror_num
, bio_flags
, rw
);
3038 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3040 get_extent_t
*get_extent
,
3041 struct bio
**bio
, int mirror_num
,
3042 unsigned long *bio_flags
, int rw
)
3044 struct inode
*inode
= page
->mapping
->host
;
3045 struct btrfs_ordered_extent
*ordered
;
3046 u64 start
= page_offset(page
);
3047 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3051 lock_extent(tree
, start
, end
);
3052 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
3055 unlock_extent(tree
, start
, end
);
3056 btrfs_start_ordered_extent(inode
, ordered
, 1);
3057 btrfs_put_ordered_extent(ordered
);
3060 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3065 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3066 get_extent_t
*get_extent
, int mirror_num
)
3068 struct bio
*bio
= NULL
;
3069 unsigned long bio_flags
= 0;
3072 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3075 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3079 int extent_read_full_page_nolock(struct extent_io_tree
*tree
, struct page
*page
,
3080 get_extent_t
*get_extent
, int mirror_num
)
3082 struct bio
*bio
= NULL
;
3083 unsigned long bio_flags
= EXTENT_BIO_PARENT_LOCKED
;
3086 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, &bio
, mirror_num
,
3089 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3093 static noinline
void update_nr_written(struct page
*page
,
3094 struct writeback_control
*wbc
,
3095 unsigned long nr_written
)
3097 wbc
->nr_to_write
-= nr_written
;
3098 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
3099 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
3100 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
3104 * helper for __extent_writepage, doing all of the delayed allocation setup.
3106 * This returns 1 if our fill_delalloc function did all the work required
3107 * to write the page (copy into inline extent). In this case the IO has
3108 * been started and the page is already unlocked.
3110 * This returns 0 if all went well (page still locked)
3111 * This returns < 0 if there were errors (page still locked)
3113 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3114 struct page
*page
, struct writeback_control
*wbc
,
3115 struct extent_page_data
*epd
,
3117 unsigned long *nr_written
)
3119 struct extent_io_tree
*tree
= epd
->tree
;
3120 u64 page_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
3122 u64 delalloc_to_write
= 0;
3123 u64 delalloc_end
= 0;
3125 int page_started
= 0;
3127 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3130 while (delalloc_end
< page_end
) {
3131 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3136 if (nr_delalloc
== 0) {
3137 delalloc_start
= delalloc_end
+ 1;
3140 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3145 /* File system has been set read-only */
3148 /* fill_delalloc should be return < 0 for error
3149 * but just in case, we use > 0 here meaning the
3150 * IO is started, so we don't want to return > 0
3151 * unless things are going well.
3153 ret
= ret
< 0 ? ret
: -EIO
;
3157 * delalloc_end is already one less than the total
3158 * length, so we don't subtract one from
3161 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3164 delalloc_start
= delalloc_end
+ 1;
3166 if (wbc
->nr_to_write
< delalloc_to_write
) {
3169 if (delalloc_to_write
< thresh
* 2)
3170 thresh
= delalloc_to_write
;
3171 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3175 /* did the fill delalloc function already unlock and start
3180 * we've unlocked the page, so we can't update
3181 * the mapping's writeback index, just update
3184 wbc
->nr_to_write
-= *nr_written
;
3195 * helper for __extent_writepage. This calls the writepage start hooks,
3196 * and does the loop to map the page into extents and bios.
3198 * We return 1 if the IO is started and the page is unlocked,
3199 * 0 if all went well (page still locked)
3200 * < 0 if there were errors (page still locked)
3202 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3204 struct writeback_control
*wbc
,
3205 struct extent_page_data
*epd
,
3207 unsigned long nr_written
,
3208 int write_flags
, int *nr_ret
)
3210 struct extent_io_tree
*tree
= epd
->tree
;
3211 u64 start
= page_offset(page
);
3212 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3219 struct extent_state
*cached_state
= NULL
;
3220 struct extent_map
*em
;
3221 struct block_device
*bdev
;
3222 size_t pg_offset
= 0;
3228 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3229 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3232 /* Fixup worker will requeue */
3234 wbc
->pages_skipped
++;
3236 redirty_page_for_writepage(wbc
, page
);
3238 update_nr_written(page
, wbc
, nr_written
);
3246 * we don't want to touch the inode after unlocking the page,
3247 * so we update the mapping writeback index now
3249 update_nr_written(page
, wbc
, nr_written
+ 1);
3252 if (i_size
<= start
) {
3253 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3254 tree
->ops
->writepage_end_io_hook(page
, start
,
3259 blocksize
= inode
->i_sb
->s_blocksize
;
3261 while (cur
<= end
) {
3263 if (cur
>= i_size
) {
3264 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3265 tree
->ops
->writepage_end_io_hook(page
, cur
,
3269 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3271 if (IS_ERR_OR_NULL(em
)) {
3273 ret
= PTR_ERR_OR_ZERO(em
);
3277 extent_offset
= cur
- em
->start
;
3278 em_end
= extent_map_end(em
);
3279 BUG_ON(em_end
<= cur
);
3281 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3282 iosize
= ALIGN(iosize
, blocksize
);
3283 sector
= (em
->block_start
+ extent_offset
) >> 9;
3285 block_start
= em
->block_start
;
3286 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3287 free_extent_map(em
);
3291 * compressed and inline extents are written through other
3294 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3295 block_start
== EXTENT_MAP_INLINE
) {
3297 * end_io notification does not happen here for
3298 * compressed extents
3300 if (!compressed
&& tree
->ops
&&
3301 tree
->ops
->writepage_end_io_hook
)
3302 tree
->ops
->writepage_end_io_hook(page
, cur
,
3305 else if (compressed
) {
3306 /* we don't want to end_page_writeback on
3307 * a compressed extent. this happens
3314 pg_offset
+= iosize
;
3318 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3319 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3327 unsigned long max_nr
= (i_size
>> PAGE_CACHE_SHIFT
) + 1;
3329 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3330 if (!PageWriteback(page
)) {
3331 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3332 "page %lu not writeback, cur %llu end %llu",
3333 page
->index
, cur
, end
);
3336 ret
= submit_extent_page(write_flags
, tree
, page
,
3337 sector
, iosize
, pg_offset
,
3338 bdev
, &epd
->bio
, max_nr
,
3339 end_bio_extent_writepage
,
3345 pg_offset
+= iosize
;
3353 /* drop our reference on any cached states */
3354 free_extent_state(cached_state
);
3359 * the writepage semantics are similar to regular writepage. extent
3360 * records are inserted to lock ranges in the tree, and as dirty areas
3361 * are found, they are marked writeback. Then the lock bits are removed
3362 * and the end_io handler clears the writeback ranges
3364 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3367 struct inode
*inode
= page
->mapping
->host
;
3368 struct extent_page_data
*epd
= data
;
3369 u64 start
= page_offset(page
);
3370 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3373 size_t pg_offset
= 0;
3374 loff_t i_size
= i_size_read(inode
);
3375 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
3377 unsigned long nr_written
= 0;
3379 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3380 write_flags
= WRITE_SYNC
;
3382 write_flags
= WRITE
;
3384 trace___extent_writepage(page
, inode
, wbc
);
3386 WARN_ON(!PageLocked(page
));
3388 ClearPageError(page
);
3390 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
3391 if (page
->index
> end_index
||
3392 (page
->index
== end_index
&& !pg_offset
)) {
3393 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
3398 if (page
->index
== end_index
) {
3401 userpage
= kmap_atomic(page
);
3402 memset(userpage
+ pg_offset
, 0,
3403 PAGE_CACHE_SIZE
- pg_offset
);
3404 kunmap_atomic(userpage
);
3405 flush_dcache_page(page
);
3410 set_page_extent_mapped(page
);
3412 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3418 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3419 i_size
, nr_written
, write_flags
, &nr
);
3425 /* make sure the mapping tag for page dirty gets cleared */
3426 set_page_writeback(page
);
3427 end_page_writeback(page
);
3429 if (PageError(page
)) {
3430 ret
= ret
< 0 ? ret
: -EIO
;
3431 end_extent_writepage(page
, ret
, start
, page_end
);
3440 static int eb_wait(void *word
)
3446 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3448 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3449 TASK_UNINTERRUPTIBLE
);
3452 static noinline_for_stack
int
3453 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3454 struct btrfs_fs_info
*fs_info
,
3455 struct extent_page_data
*epd
)
3457 unsigned long i
, num_pages
;
3461 if (!btrfs_try_tree_write_lock(eb
)) {
3463 flush_write_bio(epd
);
3464 btrfs_tree_lock(eb
);
3467 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3468 btrfs_tree_unlock(eb
);
3472 flush_write_bio(epd
);
3476 wait_on_extent_buffer_writeback(eb
);
3477 btrfs_tree_lock(eb
);
3478 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3480 btrfs_tree_unlock(eb
);
3485 * We need to do this to prevent races in people who check if the eb is
3486 * under IO since we can end up having no IO bits set for a short period
3489 spin_lock(&eb
->refs_lock
);
3490 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3491 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3492 spin_unlock(&eb
->refs_lock
);
3493 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3494 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3496 fs_info
->dirty_metadata_batch
);
3499 spin_unlock(&eb
->refs_lock
);
3502 btrfs_tree_unlock(eb
);
3507 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3508 for (i
= 0; i
< num_pages
; i
++) {
3509 struct page
*p
= extent_buffer_page(eb
, i
);
3511 if (!trylock_page(p
)) {
3513 flush_write_bio(epd
);
3523 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3525 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3526 smp_mb__after_clear_bit();
3527 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3530 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3532 struct bio_vec
*bvec
;
3533 struct extent_buffer
*eb
;
3536 bio_for_each_segment_all(bvec
, bio
, i
) {
3537 struct page
*page
= bvec
->bv_page
;
3539 eb
= (struct extent_buffer
*)page
->private;
3541 done
= atomic_dec_and_test(&eb
->io_pages
);
3543 if (err
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3544 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3545 ClearPageUptodate(page
);
3549 end_page_writeback(page
);
3554 end_extent_buffer_writeback(eb
);
3560 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3561 struct btrfs_fs_info
*fs_info
,
3562 struct writeback_control
*wbc
,
3563 struct extent_page_data
*epd
)
3565 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3566 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3567 u64 offset
= eb
->start
;
3568 unsigned long i
, num_pages
;
3569 unsigned long bio_flags
= 0;
3570 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3573 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3574 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3575 atomic_set(&eb
->io_pages
, num_pages
);
3576 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3577 bio_flags
= EXTENT_BIO_TREE_LOG
;
3579 for (i
= 0; i
< num_pages
; i
++) {
3580 struct page
*p
= extent_buffer_page(eb
, i
);
3582 clear_page_dirty_for_io(p
);
3583 set_page_writeback(p
);
3584 ret
= submit_extent_page(rw
, tree
, p
, offset
>> 9,
3585 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3586 -1, end_bio_extent_buffer_writepage
,
3587 0, epd
->bio_flags
, bio_flags
);
3588 epd
->bio_flags
= bio_flags
;
3590 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3592 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3593 end_extent_buffer_writeback(eb
);
3597 offset
+= PAGE_CACHE_SIZE
;
3598 update_nr_written(p
, wbc
, 1);
3602 if (unlikely(ret
)) {
3603 for (; i
< num_pages
; i
++) {
3604 struct page
*p
= extent_buffer_page(eb
, i
);
3612 int btree_write_cache_pages(struct address_space
*mapping
,
3613 struct writeback_control
*wbc
)
3615 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3616 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3617 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3618 struct extent_page_data epd
= {
3622 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3627 int nr_to_write_done
= 0;
3628 struct pagevec pvec
;
3631 pgoff_t end
; /* Inclusive */
3635 pagevec_init(&pvec
, 0);
3636 if (wbc
->range_cyclic
) {
3637 index
= mapping
->writeback_index
; /* Start from prev offset */
3640 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3641 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3644 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3645 tag
= PAGECACHE_TAG_TOWRITE
;
3647 tag
= PAGECACHE_TAG_DIRTY
;
3649 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3650 tag_pages_for_writeback(mapping
, index
, end
);
3651 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3652 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3653 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3657 for (i
= 0; i
< nr_pages
; i
++) {
3658 struct page
*page
= pvec
.pages
[i
];
3660 if (!PagePrivate(page
))
3663 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3668 spin_lock(&mapping
->private_lock
);
3669 if (!PagePrivate(page
)) {
3670 spin_unlock(&mapping
->private_lock
);
3674 eb
= (struct extent_buffer
*)page
->private;
3677 * Shouldn't happen and normally this would be a BUG_ON
3678 * but no sense in crashing the users box for something
3679 * we can survive anyway.
3682 spin_unlock(&mapping
->private_lock
);
3686 if (eb
== prev_eb
) {
3687 spin_unlock(&mapping
->private_lock
);
3691 ret
= atomic_inc_not_zero(&eb
->refs
);
3692 spin_unlock(&mapping
->private_lock
);
3697 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3699 free_extent_buffer(eb
);
3703 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3706 free_extent_buffer(eb
);
3709 free_extent_buffer(eb
);
3712 * the filesystem may choose to bump up nr_to_write.
3713 * We have to make sure to honor the new nr_to_write
3716 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3718 pagevec_release(&pvec
);
3721 if (!scanned
&& !done
) {
3723 * We hit the last page and there is more work to be done: wrap
3724 * back to the start of the file
3730 flush_write_bio(&epd
);
3735 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3736 * @mapping: address space structure to write
3737 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3738 * @writepage: function called for each page
3739 * @data: data passed to writepage function
3741 * If a page is already under I/O, write_cache_pages() skips it, even
3742 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3743 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3744 * and msync() need to guarantee that all the data which was dirty at the time
3745 * the call was made get new I/O started against them. If wbc->sync_mode is
3746 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3747 * existing IO to complete.
3749 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3750 struct address_space
*mapping
,
3751 struct writeback_control
*wbc
,
3752 writepage_t writepage
, void *data
,
3753 void (*flush_fn
)(void *))
3755 struct inode
*inode
= mapping
->host
;
3759 int nr_to_write_done
= 0;
3760 struct pagevec pvec
;
3763 pgoff_t end
; /* Inclusive */
3768 * We have to hold onto the inode so that ordered extents can do their
3769 * work when the IO finishes. The alternative to this is failing to add
3770 * an ordered extent if the igrab() fails there and that is a huge pain
3771 * to deal with, so instead just hold onto the inode throughout the
3772 * writepages operation. If it fails here we are freeing up the inode
3773 * anyway and we'd rather not waste our time writing out stuff that is
3774 * going to be truncated anyway.
3779 pagevec_init(&pvec
, 0);
3780 if (wbc
->range_cyclic
) {
3781 index
= mapping
->writeback_index
; /* Start from prev offset */
3784 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3785 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3788 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3789 tag
= PAGECACHE_TAG_TOWRITE
;
3791 tag
= PAGECACHE_TAG_DIRTY
;
3793 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3794 tag_pages_for_writeback(mapping
, index
, end
);
3795 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3796 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3797 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3801 for (i
= 0; i
< nr_pages
; i
++) {
3802 struct page
*page
= pvec
.pages
[i
];
3805 * At this point we hold neither mapping->tree_lock nor
3806 * lock on the page itself: the page may be truncated or
3807 * invalidated (changing page->mapping to NULL), or even
3808 * swizzled back from swapper_space to tmpfs file
3811 if (!trylock_page(page
)) {
3816 if (unlikely(page
->mapping
!= mapping
)) {
3821 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3827 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3828 if (PageWriteback(page
))
3830 wait_on_page_writeback(page
);
3833 if (PageWriteback(page
) ||
3834 !clear_page_dirty_for_io(page
)) {
3839 ret
= (*writepage
)(page
, wbc
, data
);
3841 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3845 if (!err
&& ret
< 0)
3849 * the filesystem may choose to bump up nr_to_write.
3850 * We have to make sure to honor the new nr_to_write
3853 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3855 pagevec_release(&pvec
);
3858 if (!scanned
&& !done
&& !err
) {
3860 * We hit the last page and there is more work to be done: wrap
3861 * back to the start of the file
3867 btrfs_add_delayed_iput(inode
);
3871 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3880 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
3881 BUG_ON(ret
< 0); /* -ENOMEM */
3886 static noinline
void flush_write_bio(void *data
)
3888 struct extent_page_data
*epd
= data
;
3889 flush_epd_write_bio(epd
);
3892 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3893 get_extent_t
*get_extent
,
3894 struct writeback_control
*wbc
)
3897 struct extent_page_data epd
= {
3900 .get_extent
= get_extent
,
3902 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3906 ret
= __extent_writepage(page
, wbc
, &epd
);
3908 flush_epd_write_bio(&epd
);
3912 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3913 u64 start
, u64 end
, get_extent_t
*get_extent
,
3917 struct address_space
*mapping
= inode
->i_mapping
;
3919 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3922 struct extent_page_data epd
= {
3925 .get_extent
= get_extent
,
3927 .sync_io
= mode
== WB_SYNC_ALL
,
3930 struct writeback_control wbc_writepages
= {
3932 .nr_to_write
= nr_pages
* 2,
3933 .range_start
= start
,
3934 .range_end
= end
+ 1,
3937 while (start
<= end
) {
3938 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3939 if (clear_page_dirty_for_io(page
))
3940 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3942 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3943 tree
->ops
->writepage_end_io_hook(page
, start
,
3944 start
+ PAGE_CACHE_SIZE
- 1,
3948 page_cache_release(page
);
3949 start
+= PAGE_CACHE_SIZE
;
3952 flush_epd_write_bio(&epd
);
3956 int extent_writepages(struct extent_io_tree
*tree
,
3957 struct address_space
*mapping
,
3958 get_extent_t
*get_extent
,
3959 struct writeback_control
*wbc
)
3962 struct extent_page_data epd
= {
3965 .get_extent
= get_extent
,
3967 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3971 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3972 __extent_writepage
, &epd
,
3974 flush_epd_write_bio(&epd
);
3978 int extent_readpages(struct extent_io_tree
*tree
,
3979 struct address_space
*mapping
,
3980 struct list_head
*pages
, unsigned nr_pages
,
3981 get_extent_t get_extent
)
3983 struct bio
*bio
= NULL
;
3985 unsigned long bio_flags
= 0;
3986 struct page
*pagepool
[16];
3988 struct extent_map
*em_cached
= NULL
;
3991 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3992 page
= list_entry(pages
->prev
, struct page
, lru
);
3994 prefetchw(&page
->flags
);
3995 list_del(&page
->lru
);
3996 if (add_to_page_cache_lru(page
, mapping
,
3997 page
->index
, GFP_NOFS
)) {
3998 page_cache_release(page
);
4002 pagepool
[nr
++] = page
;
4003 if (nr
< ARRAY_SIZE(pagepool
))
4005 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4006 &bio
, 0, &bio_flags
, READ
);
4010 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4011 &bio
, 0, &bio_flags
, READ
);
4014 free_extent_map(em_cached
);
4016 BUG_ON(!list_empty(pages
));
4018 return submit_one_bio(READ
, bio
, 0, bio_flags
);
4023 * basic invalidatepage code, this waits on any locked or writeback
4024 * ranges corresponding to the page, and then deletes any extent state
4025 * records from the tree
4027 int extent_invalidatepage(struct extent_io_tree
*tree
,
4028 struct page
*page
, unsigned long offset
)
4030 struct extent_state
*cached_state
= NULL
;
4031 u64 start
= page_offset(page
);
4032 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4033 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4035 start
+= ALIGN(offset
, blocksize
);
4039 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
4040 wait_on_page_writeback(page
);
4041 clear_extent_bit(tree
, start
, end
,
4042 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4043 EXTENT_DO_ACCOUNTING
,
4044 1, 1, &cached_state
, GFP_NOFS
);
4049 * a helper for releasepage, this tests for areas of the page that
4050 * are locked or under IO and drops the related state bits if it is safe
4053 static int try_release_extent_state(struct extent_map_tree
*map
,
4054 struct extent_io_tree
*tree
,
4055 struct page
*page
, gfp_t mask
)
4057 u64 start
= page_offset(page
);
4058 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4061 if (test_range_bit(tree
, start
, end
,
4062 EXTENT_IOBITS
, 0, NULL
))
4065 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4068 * at this point we can safely clear everything except the
4069 * locked bit and the nodatasum bit
4071 ret
= clear_extent_bit(tree
, start
, end
,
4072 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4075 /* if clear_extent_bit failed for enomem reasons,
4076 * we can't allow the release to continue.
4087 * a helper for releasepage. As long as there are no locked extents
4088 * in the range corresponding to the page, both state records and extent
4089 * map records are removed
4091 int try_release_extent_mapping(struct extent_map_tree
*map
,
4092 struct extent_io_tree
*tree
, struct page
*page
,
4095 struct extent_map
*em
;
4096 u64 start
= page_offset(page
);
4097 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4099 if ((mask
& __GFP_WAIT
) &&
4100 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
4102 while (start
<= end
) {
4103 len
= end
- start
+ 1;
4104 write_lock(&map
->lock
);
4105 em
= lookup_extent_mapping(map
, start
, len
);
4107 write_unlock(&map
->lock
);
4110 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4111 em
->start
!= start
) {
4112 write_unlock(&map
->lock
);
4113 free_extent_map(em
);
4116 if (!test_range_bit(tree
, em
->start
,
4117 extent_map_end(em
) - 1,
4118 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4120 remove_extent_mapping(map
, em
);
4121 /* once for the rb tree */
4122 free_extent_map(em
);
4124 start
= extent_map_end(em
);
4125 write_unlock(&map
->lock
);
4128 free_extent_map(em
);
4131 return try_release_extent_state(map
, tree
, page
, mask
);
4135 * helper function for fiemap, which doesn't want to see any holes.
4136 * This maps until we find something past 'last'
4138 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4141 get_extent_t
*get_extent
)
4143 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4144 struct extent_map
*em
;
4151 len
= last
- offset
;
4154 len
= ALIGN(len
, sectorsize
);
4155 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4156 if (IS_ERR_OR_NULL(em
))
4159 /* if this isn't a hole return it */
4160 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4161 em
->block_start
!= EXTENT_MAP_HOLE
) {
4165 /* this is a hole, advance to the next extent */
4166 offset
= extent_map_end(em
);
4167 free_extent_map(em
);
4174 static noinline
int count_ext_ref(u64 inum
, u64 offset
, u64 root_id
, void *ctx
)
4176 unsigned long cnt
= *((unsigned long *)ctx
);
4179 *((unsigned long *)ctx
) = cnt
;
4181 /* Now we're sure that the extent is shared. */
4187 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4188 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4192 u64 max
= start
+ len
;
4196 u64 last_for_get_extent
= 0;
4198 u64 isize
= i_size_read(inode
);
4199 struct btrfs_key found_key
;
4200 struct extent_map
*em
= NULL
;
4201 struct extent_state
*cached_state
= NULL
;
4202 struct btrfs_path
*path
;
4211 path
= btrfs_alloc_path();
4214 path
->leave_spinning
= 1;
4216 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
4217 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
4220 * lookup the last file extent. We're not using i_size here
4221 * because there might be preallocation past i_size
4223 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
4224 path
, btrfs_ino(inode
), -1, 0);
4226 btrfs_free_path(path
);
4231 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4232 found_type
= btrfs_key_type(&found_key
);
4234 /* No extents, but there might be delalloc bits */
4235 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4236 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4237 /* have to trust i_size as the end */
4239 last_for_get_extent
= isize
;
4242 * remember the start of the last extent. There are a
4243 * bunch of different factors that go into the length of the
4244 * extent, so its much less complex to remember where it started
4246 last
= found_key
.offset
;
4247 last_for_get_extent
= last
+ 1;
4249 btrfs_release_path(path
);
4252 * we might have some extents allocated but more delalloc past those
4253 * extents. so, we trust isize unless the start of the last extent is
4258 last_for_get_extent
= isize
;
4261 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1, 0,
4264 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4274 u64 offset_in_extent
= 0;
4276 /* break if the extent we found is outside the range */
4277 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4281 * get_extent may return an extent that starts before our
4282 * requested range. We have to make sure the ranges
4283 * we return to fiemap always move forward and don't
4284 * overlap, so adjust the offsets here
4286 em_start
= max(em
->start
, off
);
4289 * record the offset from the start of the extent
4290 * for adjusting the disk offset below. Only do this if the
4291 * extent isn't compressed since our in ram offset may be past
4292 * what we have actually allocated on disk.
4294 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4295 offset_in_extent
= em_start
- em
->start
;
4296 em_end
= extent_map_end(em
);
4297 em_len
= em_end
- em_start
;
4302 * bump off for our next call to get_extent
4304 off
= extent_map_end(em
);
4308 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4310 flags
|= FIEMAP_EXTENT_LAST
;
4311 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4312 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4313 FIEMAP_EXTENT_NOT_ALIGNED
);
4314 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4315 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4316 FIEMAP_EXTENT_UNKNOWN
);
4318 unsigned long ref_cnt
= 0;
4320 disko
= em
->block_start
+ offset_in_extent
;
4323 * As btrfs supports shared space, this information
4324 * can be exported to userspace tools via
4325 * flag FIEMAP_EXTENT_SHARED.
4327 ret
= iterate_inodes_from_logical(
4329 BTRFS_I(inode
)->root
->fs_info
,
4330 path
, count_ext_ref
, &ref_cnt
);
4331 if (ret
< 0 && ret
!= -ENOENT
)
4335 flags
|= FIEMAP_EXTENT_SHARED
;
4337 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4338 flags
|= FIEMAP_EXTENT_ENCODED
;
4340 free_extent_map(em
);
4342 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4343 (last
== (u64
)-1 && isize
<= em_end
)) {
4344 flags
|= FIEMAP_EXTENT_LAST
;
4348 /* now scan forward to see if this is really the last extent. */
4349 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4356 flags
|= FIEMAP_EXTENT_LAST
;
4359 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4365 free_extent_map(em
);
4367 btrfs_free_path(path
);
4368 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4369 &cached_state
, GFP_NOFS
);
4373 static void __free_extent_buffer(struct extent_buffer
*eb
)
4375 btrfs_leak_debug_del(&eb
->leak_list
);
4376 kmem_cache_free(extent_buffer_cache
, eb
);
4379 int extent_buffer_under_io(struct extent_buffer
*eb
)
4381 return (atomic_read(&eb
->io_pages
) ||
4382 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4383 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4387 * Helper for releasing extent buffer page.
4389 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4390 unsigned long start_idx
)
4392 unsigned long index
;
4393 unsigned long num_pages
;
4395 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4397 BUG_ON(extent_buffer_under_io(eb
));
4399 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4400 index
= start_idx
+ num_pages
;
4401 if (start_idx
>= index
)
4406 page
= extent_buffer_page(eb
, index
);
4407 if (page
&& mapped
) {
4408 spin_lock(&page
->mapping
->private_lock
);
4410 * We do this since we'll remove the pages after we've
4411 * removed the eb from the radix tree, so we could race
4412 * and have this page now attached to the new eb. So
4413 * only clear page_private if it's still connected to
4416 if (PagePrivate(page
) &&
4417 page
->private == (unsigned long)eb
) {
4418 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4419 BUG_ON(PageDirty(page
));
4420 BUG_ON(PageWriteback(page
));
4422 * We need to make sure we haven't be attached
4425 ClearPagePrivate(page
);
4426 set_page_private(page
, 0);
4427 /* One for the page private */
4428 page_cache_release(page
);
4430 spin_unlock(&page
->mapping
->private_lock
);
4434 /* One for when we alloced the page */
4435 page_cache_release(page
);
4437 } while (index
!= start_idx
);
4441 * Helper for releasing the extent buffer.
4443 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4445 btrfs_release_extent_buffer_page(eb
, 0);
4446 __free_extent_buffer(eb
);
4449 static struct extent_buffer
*
4450 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4451 unsigned long len
, gfp_t mask
)
4453 struct extent_buffer
*eb
= NULL
;
4455 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
4460 eb
->fs_info
= fs_info
;
4462 rwlock_init(&eb
->lock
);
4463 atomic_set(&eb
->write_locks
, 0);
4464 atomic_set(&eb
->read_locks
, 0);
4465 atomic_set(&eb
->blocking_readers
, 0);
4466 atomic_set(&eb
->blocking_writers
, 0);
4467 atomic_set(&eb
->spinning_readers
, 0);
4468 atomic_set(&eb
->spinning_writers
, 0);
4469 eb
->lock_nested
= 0;
4470 init_waitqueue_head(&eb
->write_lock_wq
);
4471 init_waitqueue_head(&eb
->read_lock_wq
);
4473 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4475 spin_lock_init(&eb
->refs_lock
);
4476 atomic_set(&eb
->refs
, 1);
4477 atomic_set(&eb
->io_pages
, 0);
4480 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4482 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4483 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4484 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4489 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4493 struct extent_buffer
*new;
4494 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4496 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_NOFS
);
4500 for (i
= 0; i
< num_pages
; i
++) {
4501 p
= alloc_page(GFP_NOFS
);
4503 btrfs_release_extent_buffer(new);
4506 attach_extent_buffer_page(new, p
);
4507 WARN_ON(PageDirty(p
));
4512 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4513 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4514 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4519 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
4521 struct extent_buffer
*eb
;
4522 unsigned long num_pages
= num_extent_pages(0, len
);
4525 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_NOFS
);
4529 for (i
= 0; i
< num_pages
; i
++) {
4530 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4534 set_extent_buffer_uptodate(eb
);
4535 btrfs_set_header_nritems(eb
, 0);
4536 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4541 __free_page(eb
->pages
[i
- 1]);
4542 __free_extent_buffer(eb
);
4546 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4549 /* the ref bit is tricky. We have to make sure it is set
4550 * if we have the buffer dirty. Otherwise the
4551 * code to free a buffer can end up dropping a dirty
4554 * Once the ref bit is set, it won't go away while the
4555 * buffer is dirty or in writeback, and it also won't
4556 * go away while we have the reference count on the
4559 * We can't just set the ref bit without bumping the
4560 * ref on the eb because free_extent_buffer might
4561 * see the ref bit and try to clear it. If this happens
4562 * free_extent_buffer might end up dropping our original
4563 * ref by mistake and freeing the page before we are able
4564 * to add one more ref.
4566 * So bump the ref count first, then set the bit. If someone
4567 * beat us to it, drop the ref we added.
4569 refs
= atomic_read(&eb
->refs
);
4570 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4573 spin_lock(&eb
->refs_lock
);
4574 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4575 atomic_inc(&eb
->refs
);
4576 spin_unlock(&eb
->refs_lock
);
4579 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
)
4581 unsigned long num_pages
, i
;
4583 check_buffer_tree_ref(eb
);
4585 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4586 for (i
= 0; i
< num_pages
; i
++) {
4587 struct page
*p
= extent_buffer_page(eb
, i
);
4588 mark_page_accessed(p
);
4592 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4595 struct extent_buffer
*eb
;
4598 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4599 start
>> PAGE_CACHE_SHIFT
);
4600 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4602 mark_extent_buffer_accessed(eb
);
4610 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4611 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4612 u64 start
, unsigned long len
)
4614 struct extent_buffer
*eb
, *exists
= NULL
;
4617 eb
= find_extent_buffer(fs_info
, start
);
4620 eb
= alloc_dummy_extent_buffer(start
, len
);
4623 eb
->fs_info
= fs_info
;
4625 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4628 spin_lock(&fs_info
->buffer_lock
);
4629 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4630 start
>> PAGE_CACHE_SHIFT
, eb
);
4631 spin_unlock(&fs_info
->buffer_lock
);
4632 radix_tree_preload_end();
4633 if (ret
== -EEXIST
) {
4634 exists
= find_extent_buffer(fs_info
, start
);
4640 check_buffer_tree_ref(eb
);
4641 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4644 * We will free dummy extent buffer's if they come into
4645 * free_extent_buffer with a ref count of 2, but if we are using this we
4646 * want the buffers to stay in memory until we're done with them, so
4647 * bump the ref count again.
4649 atomic_inc(&eb
->refs
);
4652 btrfs_release_extent_buffer(eb
);
4657 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4658 u64 start
, unsigned long len
)
4660 unsigned long num_pages
= num_extent_pages(start
, len
);
4662 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4663 struct extent_buffer
*eb
;
4664 struct extent_buffer
*exists
= NULL
;
4666 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4670 eb
= find_extent_buffer(fs_info
, start
);
4674 eb
= __alloc_extent_buffer(fs_info
, start
, len
, GFP_NOFS
);
4678 for (i
= 0; i
< num_pages
; i
++, index
++) {
4679 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4683 spin_lock(&mapping
->private_lock
);
4684 if (PagePrivate(p
)) {
4686 * We could have already allocated an eb for this page
4687 * and attached one so lets see if we can get a ref on
4688 * the existing eb, and if we can we know it's good and
4689 * we can just return that one, else we know we can just
4690 * overwrite page->private.
4692 exists
= (struct extent_buffer
*)p
->private;
4693 if (atomic_inc_not_zero(&exists
->refs
)) {
4694 spin_unlock(&mapping
->private_lock
);
4696 page_cache_release(p
);
4697 mark_extent_buffer_accessed(exists
);
4702 * Do this so attach doesn't complain and we need to
4703 * drop the ref the old guy had.
4705 ClearPagePrivate(p
);
4706 WARN_ON(PageDirty(p
));
4707 page_cache_release(p
);
4709 attach_extent_buffer_page(eb
, p
);
4710 spin_unlock(&mapping
->private_lock
);
4711 WARN_ON(PageDirty(p
));
4712 mark_page_accessed(p
);
4714 if (!PageUptodate(p
))
4718 * see below about how we avoid a nasty race with release page
4719 * and why we unlock later
4723 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4725 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4729 spin_lock(&fs_info
->buffer_lock
);
4730 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4731 start
>> PAGE_CACHE_SHIFT
, eb
);
4732 spin_unlock(&fs_info
->buffer_lock
);
4733 radix_tree_preload_end();
4734 if (ret
== -EEXIST
) {
4735 exists
= find_extent_buffer(fs_info
, start
);
4741 /* add one reference for the tree */
4742 check_buffer_tree_ref(eb
);
4743 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4746 * there is a race where release page may have
4747 * tried to find this extent buffer in the radix
4748 * but failed. It will tell the VM it is safe to
4749 * reclaim the, and it will clear the page private bit.
4750 * We must make sure to set the page private bit properly
4751 * after the extent buffer is in the radix tree so
4752 * it doesn't get lost
4754 SetPageChecked(eb
->pages
[0]);
4755 for (i
= 1; i
< num_pages
; i
++) {
4756 p
= extent_buffer_page(eb
, i
);
4757 ClearPageChecked(p
);
4760 unlock_page(eb
->pages
[0]);
4764 for (i
= 0; i
< num_pages
; i
++) {
4766 unlock_page(eb
->pages
[i
]);
4769 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4770 btrfs_release_extent_buffer(eb
);
4774 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4776 struct extent_buffer
*eb
=
4777 container_of(head
, struct extent_buffer
, rcu_head
);
4779 __free_extent_buffer(eb
);
4782 /* Expects to have eb->eb_lock already held */
4783 static int release_extent_buffer(struct extent_buffer
*eb
)
4785 WARN_ON(atomic_read(&eb
->refs
) == 0);
4786 if (atomic_dec_and_test(&eb
->refs
)) {
4787 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
4788 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
4790 spin_unlock(&eb
->refs_lock
);
4792 spin_lock(&fs_info
->buffer_lock
);
4793 radix_tree_delete(&fs_info
->buffer_radix
,
4794 eb
->start
>> PAGE_CACHE_SHIFT
);
4795 spin_unlock(&fs_info
->buffer_lock
);
4797 spin_unlock(&eb
->refs_lock
);
4800 /* Should be safe to release our pages at this point */
4801 btrfs_release_extent_buffer_page(eb
, 0);
4802 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4805 spin_unlock(&eb
->refs_lock
);
4810 void free_extent_buffer(struct extent_buffer
*eb
)
4818 refs
= atomic_read(&eb
->refs
);
4821 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
4826 spin_lock(&eb
->refs_lock
);
4827 if (atomic_read(&eb
->refs
) == 2 &&
4828 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4829 atomic_dec(&eb
->refs
);
4831 if (atomic_read(&eb
->refs
) == 2 &&
4832 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4833 !extent_buffer_under_io(eb
) &&
4834 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4835 atomic_dec(&eb
->refs
);
4838 * I know this is terrible, but it's temporary until we stop tracking
4839 * the uptodate bits and such for the extent buffers.
4841 release_extent_buffer(eb
);
4844 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4849 spin_lock(&eb
->refs_lock
);
4850 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4852 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4853 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4854 atomic_dec(&eb
->refs
);
4855 release_extent_buffer(eb
);
4858 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4861 unsigned long num_pages
;
4864 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4866 for (i
= 0; i
< num_pages
; i
++) {
4867 page
= extent_buffer_page(eb
, i
);
4868 if (!PageDirty(page
))
4872 WARN_ON(!PagePrivate(page
));
4874 clear_page_dirty_for_io(page
);
4875 spin_lock_irq(&page
->mapping
->tree_lock
);
4876 if (!PageDirty(page
)) {
4877 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4879 PAGECACHE_TAG_DIRTY
);
4881 spin_unlock_irq(&page
->mapping
->tree_lock
);
4882 ClearPageError(page
);
4885 WARN_ON(atomic_read(&eb
->refs
) == 0);
4888 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4891 unsigned long num_pages
;
4894 check_buffer_tree_ref(eb
);
4896 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4898 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4899 WARN_ON(atomic_read(&eb
->refs
) == 0);
4900 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4902 for (i
= 0; i
< num_pages
; i
++)
4903 set_page_dirty(extent_buffer_page(eb
, i
));
4907 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4911 unsigned long num_pages
;
4913 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4914 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4915 for (i
= 0; i
< num_pages
; i
++) {
4916 page
= extent_buffer_page(eb
, i
);
4918 ClearPageUptodate(page
);
4923 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4927 unsigned long num_pages
;
4929 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4930 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4931 for (i
= 0; i
< num_pages
; i
++) {
4932 page
= extent_buffer_page(eb
, i
);
4933 SetPageUptodate(page
);
4938 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4940 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4943 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4944 struct extent_buffer
*eb
, u64 start
, int wait
,
4945 get_extent_t
*get_extent
, int mirror_num
)
4948 unsigned long start_i
;
4952 int locked_pages
= 0;
4953 int all_uptodate
= 1;
4954 unsigned long num_pages
;
4955 unsigned long num_reads
= 0;
4956 struct bio
*bio
= NULL
;
4957 unsigned long bio_flags
= 0;
4959 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4963 WARN_ON(start
< eb
->start
);
4964 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4965 (eb
->start
>> PAGE_CACHE_SHIFT
);
4970 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4971 for (i
= start_i
; i
< num_pages
; i
++) {
4972 page
= extent_buffer_page(eb
, i
);
4973 if (wait
== WAIT_NONE
) {
4974 if (!trylock_page(page
))
4980 if (!PageUptodate(page
)) {
4987 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4991 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4992 eb
->read_mirror
= 0;
4993 atomic_set(&eb
->io_pages
, num_reads
);
4994 for (i
= start_i
; i
< num_pages
; i
++) {
4995 page
= extent_buffer_page(eb
, i
);
4996 if (!PageUptodate(page
)) {
4997 ClearPageError(page
);
4998 err
= __extent_read_full_page(tree
, page
,
5000 mirror_num
, &bio_flags
,
5010 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
5016 if (ret
|| wait
!= WAIT_COMPLETE
)
5019 for (i
= start_i
; i
< num_pages
; i
++) {
5020 page
= extent_buffer_page(eb
, i
);
5021 wait_on_page_locked(page
);
5022 if (!PageUptodate(page
))
5030 while (locked_pages
> 0) {
5031 page
= extent_buffer_page(eb
, i
);
5039 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5040 unsigned long start
,
5047 char *dst
= (char *)dstv
;
5048 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5049 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5051 WARN_ON(start
> eb
->len
);
5052 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5054 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5057 page
= extent_buffer_page(eb
, i
);
5059 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5060 kaddr
= page_address(page
);
5061 memcpy(dst
, kaddr
+ offset
, cur
);
5070 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5071 unsigned long min_len
, char **map
,
5072 unsigned long *map_start
,
5073 unsigned long *map_len
)
5075 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
5078 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5079 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5080 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5087 offset
= start_offset
;
5091 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
5094 if (start
+ min_len
> eb
->len
) {
5095 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
5097 eb
->start
, eb
->len
, start
, min_len
);
5101 p
= extent_buffer_page(eb
, i
);
5102 kaddr
= page_address(p
);
5103 *map
= kaddr
+ offset
;
5104 *map_len
= PAGE_CACHE_SIZE
- offset
;
5108 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5109 unsigned long start
,
5116 char *ptr
= (char *)ptrv
;
5117 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5118 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5121 WARN_ON(start
> eb
->len
);
5122 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5124 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5127 page
= extent_buffer_page(eb
, i
);
5129 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5131 kaddr
= page_address(page
);
5132 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5144 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5145 unsigned long start
, unsigned long len
)
5151 char *src
= (char *)srcv
;
5152 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5153 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5155 WARN_ON(start
> eb
->len
);
5156 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5158 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5161 page
= extent_buffer_page(eb
, i
);
5162 WARN_ON(!PageUptodate(page
));
5164 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5165 kaddr
= page_address(page
);
5166 memcpy(kaddr
+ offset
, src
, cur
);
5175 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5176 unsigned long start
, unsigned long len
)
5182 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5183 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5185 WARN_ON(start
> eb
->len
);
5186 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5188 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5191 page
= extent_buffer_page(eb
, i
);
5192 WARN_ON(!PageUptodate(page
));
5194 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5195 kaddr
= page_address(page
);
5196 memset(kaddr
+ offset
, c
, cur
);
5204 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5205 unsigned long dst_offset
, unsigned long src_offset
,
5208 u64 dst_len
= dst
->len
;
5213 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5214 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5216 WARN_ON(src
->len
!= dst_len
);
5218 offset
= (start_offset
+ dst_offset
) &
5219 (PAGE_CACHE_SIZE
- 1);
5222 page
= extent_buffer_page(dst
, i
);
5223 WARN_ON(!PageUptodate(page
));
5225 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
5227 kaddr
= page_address(page
);
5228 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5237 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5239 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5240 return distance
< len
;
5243 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5244 unsigned long dst_off
, unsigned long src_off
,
5247 char *dst_kaddr
= page_address(dst_page
);
5249 int must_memmove
= 0;
5251 if (dst_page
!= src_page
) {
5252 src_kaddr
= page_address(src_page
);
5254 src_kaddr
= dst_kaddr
;
5255 if (areas_overlap(src_off
, dst_off
, len
))
5260 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5262 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5265 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5266 unsigned long src_offset
, unsigned long len
)
5269 size_t dst_off_in_page
;
5270 size_t src_off_in_page
;
5271 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5272 unsigned long dst_i
;
5273 unsigned long src_i
;
5275 if (src_offset
+ len
> dst
->len
) {
5276 printk(KERN_ERR
"BTRFS: memmove bogus src_offset %lu move "
5277 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
5280 if (dst_offset
+ len
> dst
->len
) {
5281 printk(KERN_ERR
"BTRFS: memmove bogus dst_offset %lu move "
5282 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
5287 dst_off_in_page
= (start_offset
+ dst_offset
) &
5288 (PAGE_CACHE_SIZE
- 1);
5289 src_off_in_page
= (start_offset
+ src_offset
) &
5290 (PAGE_CACHE_SIZE
- 1);
5292 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5293 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5295 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5297 cur
= min_t(unsigned long, cur
,
5298 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5300 copy_pages(extent_buffer_page(dst
, dst_i
),
5301 extent_buffer_page(dst
, src_i
),
5302 dst_off_in_page
, src_off_in_page
, cur
);
5310 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5311 unsigned long src_offset
, unsigned long len
)
5314 size_t dst_off_in_page
;
5315 size_t src_off_in_page
;
5316 unsigned long dst_end
= dst_offset
+ len
- 1;
5317 unsigned long src_end
= src_offset
+ len
- 1;
5318 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5319 unsigned long dst_i
;
5320 unsigned long src_i
;
5322 if (src_offset
+ len
> dst
->len
) {
5323 printk(KERN_ERR
"BTRFS: memmove bogus src_offset %lu move "
5324 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
5327 if (dst_offset
+ len
> dst
->len
) {
5328 printk(KERN_ERR
"BTRFS: memmove bogus dst_offset %lu move "
5329 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
5332 if (dst_offset
< src_offset
) {
5333 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5337 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5338 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5340 dst_off_in_page
= (start_offset
+ dst_end
) &
5341 (PAGE_CACHE_SIZE
- 1);
5342 src_off_in_page
= (start_offset
+ src_end
) &
5343 (PAGE_CACHE_SIZE
- 1);
5345 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5346 cur
= min(cur
, dst_off_in_page
+ 1);
5347 copy_pages(extent_buffer_page(dst
, dst_i
),
5348 extent_buffer_page(dst
, src_i
),
5349 dst_off_in_page
- cur
+ 1,
5350 src_off_in_page
- cur
+ 1, cur
);
5358 int try_release_extent_buffer(struct page
*page
)
5360 struct extent_buffer
*eb
;
5363 * We need to make sure noboody is attaching this page to an eb right
5366 spin_lock(&page
->mapping
->private_lock
);
5367 if (!PagePrivate(page
)) {
5368 spin_unlock(&page
->mapping
->private_lock
);
5372 eb
= (struct extent_buffer
*)page
->private;
5376 * This is a little awful but should be ok, we need to make sure that
5377 * the eb doesn't disappear out from under us while we're looking at
5380 spin_lock(&eb
->refs_lock
);
5381 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5382 spin_unlock(&eb
->refs_lock
);
5383 spin_unlock(&page
->mapping
->private_lock
);
5386 spin_unlock(&page
->mapping
->private_lock
);
5389 * If tree ref isn't set then we know the ref on this eb is a real ref,
5390 * so just return, this page will likely be freed soon anyway.
5392 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5393 spin_unlock(&eb
->refs_lock
);
5397 return release_extent_buffer(eb
);