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
);
1697 max_bytes
= PAGE_CACHE_SIZE
;
1705 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1707 /* step three, lock the state bits for the whole range */
1708 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1710 /* then test to make sure it is all still delalloc */
1711 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1712 EXTENT_DELALLOC
, 1, cached_state
);
1714 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1715 &cached_state
, GFP_NOFS
);
1716 __unlock_for_delalloc(inode
, locked_page
,
1717 delalloc_start
, delalloc_end
);
1721 free_extent_state(cached_state
);
1722 *start
= delalloc_start
;
1723 *end
= delalloc_end
;
1728 int extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1729 struct page
*locked_page
,
1730 unsigned long clear_bits
,
1731 unsigned long page_ops
)
1733 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1735 struct page
*pages
[16];
1736 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1737 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1738 unsigned long nr_pages
= end_index
- index
+ 1;
1741 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1745 while (nr_pages
> 0) {
1746 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1747 min_t(unsigned long,
1748 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1749 for (i
= 0; i
< ret
; i
++) {
1751 if (page_ops
& PAGE_SET_PRIVATE2
)
1752 SetPagePrivate2(pages
[i
]);
1754 if (pages
[i
] == locked_page
) {
1755 page_cache_release(pages
[i
]);
1758 if (page_ops
& PAGE_CLEAR_DIRTY
)
1759 clear_page_dirty_for_io(pages
[i
]);
1760 if (page_ops
& PAGE_SET_WRITEBACK
)
1761 set_page_writeback(pages
[i
]);
1762 if (page_ops
& PAGE_END_WRITEBACK
)
1763 end_page_writeback(pages
[i
]);
1764 if (page_ops
& PAGE_UNLOCK
)
1765 unlock_page(pages
[i
]);
1766 page_cache_release(pages
[i
]);
1776 * count the number of bytes in the tree that have a given bit(s)
1777 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1778 * cached. The total number found is returned.
1780 u64
count_range_bits(struct extent_io_tree
*tree
,
1781 u64
*start
, u64 search_end
, u64 max_bytes
,
1782 unsigned long bits
, int contig
)
1784 struct rb_node
*node
;
1785 struct extent_state
*state
;
1786 u64 cur_start
= *start
;
1787 u64 total_bytes
= 0;
1791 if (WARN_ON(search_end
<= cur_start
))
1794 spin_lock(&tree
->lock
);
1795 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1796 total_bytes
= tree
->dirty_bytes
;
1800 * this search will find all the extents that end after
1803 node
= tree_search(tree
, cur_start
);
1808 state
= rb_entry(node
, struct extent_state
, rb_node
);
1809 if (state
->start
> search_end
)
1811 if (contig
&& found
&& state
->start
> last
+ 1)
1813 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1814 total_bytes
+= min(search_end
, state
->end
) + 1 -
1815 max(cur_start
, state
->start
);
1816 if (total_bytes
>= max_bytes
)
1819 *start
= max(cur_start
, state
->start
);
1823 } else if (contig
&& found
) {
1826 node
= rb_next(node
);
1831 spin_unlock(&tree
->lock
);
1836 * set the private field for a given byte offset in the tree. If there isn't
1837 * an extent_state there already, this does nothing.
1839 static int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1841 struct rb_node
*node
;
1842 struct extent_state
*state
;
1845 spin_lock(&tree
->lock
);
1847 * this search will find all the extents that end after
1850 node
= tree_search(tree
, start
);
1855 state
= rb_entry(node
, struct extent_state
, rb_node
);
1856 if (state
->start
!= start
) {
1860 state
->private = private;
1862 spin_unlock(&tree
->lock
);
1866 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1868 struct rb_node
*node
;
1869 struct extent_state
*state
;
1872 spin_lock(&tree
->lock
);
1874 * this search will find all the extents that end after
1877 node
= tree_search(tree
, start
);
1882 state
= rb_entry(node
, struct extent_state
, rb_node
);
1883 if (state
->start
!= start
) {
1887 *private = state
->private;
1889 spin_unlock(&tree
->lock
);
1894 * searches a range in the state tree for a given mask.
1895 * If 'filled' == 1, this returns 1 only if every extent in the tree
1896 * has the bits set. Otherwise, 1 is returned if any bit in the
1897 * range is found set.
1899 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1900 unsigned long bits
, int filled
, struct extent_state
*cached
)
1902 struct extent_state
*state
= NULL
;
1903 struct rb_node
*node
;
1906 spin_lock(&tree
->lock
);
1907 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1908 cached
->end
> start
)
1909 node
= &cached
->rb_node
;
1911 node
= tree_search(tree
, start
);
1912 while (node
&& start
<= end
) {
1913 state
= rb_entry(node
, struct extent_state
, rb_node
);
1915 if (filled
&& state
->start
> start
) {
1920 if (state
->start
> end
)
1923 if (state
->state
& bits
) {
1927 } else if (filled
) {
1932 if (state
->end
== (u64
)-1)
1935 start
= state
->end
+ 1;
1938 node
= rb_next(node
);
1945 spin_unlock(&tree
->lock
);
1950 * helper function to set a given page up to date if all the
1951 * extents in the tree for that page are up to date
1953 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1955 u64 start
= page_offset(page
);
1956 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1957 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1958 SetPageUptodate(page
);
1962 * When IO fails, either with EIO or csum verification fails, we
1963 * try other mirrors that might have a good copy of the data. This
1964 * io_failure_record is used to record state as we go through all the
1965 * mirrors. If another mirror has good data, the page is set up to date
1966 * and things continue. If a good mirror can't be found, the original
1967 * bio end_io callback is called to indicate things have failed.
1969 struct io_failure_record
{
1974 unsigned long bio_flags
;
1980 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
1985 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1987 set_state_private(failure_tree
, rec
->start
, 0);
1988 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1989 rec
->start
+ rec
->len
- 1,
1990 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1994 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1995 rec
->start
+ rec
->len
- 1,
1996 EXTENT_DAMAGED
, GFP_NOFS
);
2005 * this bypasses the standard btrfs submit functions deliberately, as
2006 * the standard behavior is to write all copies in a raid setup. here we only
2007 * want to write the one bad copy. so we do the mapping for ourselves and issue
2008 * submit_bio directly.
2009 * to avoid any synchronization issues, wait for the data after writing, which
2010 * actually prevents the read that triggered the error from finishing.
2011 * currently, there can be no more than two copies of every data bit. thus,
2012 * exactly one rewrite is required.
2014 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 start
,
2015 u64 length
, u64 logical
, struct page
*page
,
2019 struct btrfs_device
*dev
;
2022 struct btrfs_bio
*bbio
= NULL
;
2023 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2026 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2027 BUG_ON(!mirror_num
);
2029 /* we can't repair anything in raid56 yet */
2030 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2033 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2036 bio
->bi_iter
.bi_size
= 0;
2037 map_length
= length
;
2039 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2040 &map_length
, &bbio
, mirror_num
);
2045 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2046 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2047 bio
->bi_iter
.bi_sector
= sector
;
2048 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2050 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2054 bio
->bi_bdev
= dev
->bdev
;
2055 bio_add_page(bio
, page
, length
, start
- page_offset(page
));
2057 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
)) {
2058 /* try to remap that extent elsewhere? */
2060 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2064 printk_ratelimited_in_rcu(KERN_INFO
2065 "BTRFS: read error corrected: ino %lu off %llu "
2066 "(dev %s sector %llu)\n", page
->mapping
->host
->i_ino
,
2067 start
, rcu_str_deref(dev
->name
), sector
);
2073 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2076 u64 start
= eb
->start
;
2077 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2080 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2083 for (i
= 0; i
< num_pages
; i
++) {
2084 struct page
*p
= extent_buffer_page(eb
, i
);
2085 ret
= repair_io_failure(root
->fs_info
, start
, PAGE_CACHE_SIZE
,
2086 start
, p
, mirror_num
);
2089 start
+= PAGE_CACHE_SIZE
;
2096 * each time an IO finishes, we do a fast check in the IO failure tree
2097 * to see if we need to process or clean up an io_failure_record
2099 static int clean_io_failure(u64 start
, struct page
*page
)
2102 u64 private_failure
;
2103 struct io_failure_record
*failrec
;
2104 struct inode
*inode
= page
->mapping
->host
;
2105 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2106 struct extent_state
*state
;
2112 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2113 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2117 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2122 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2123 BUG_ON(!failrec
->this_mirror
);
2125 if (failrec
->in_validation
) {
2126 /* there was no real error, just free the record */
2127 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2132 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2135 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2136 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2139 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2141 if (state
&& state
->start
<= failrec
->start
&&
2142 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2143 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2145 if (num_copies
> 1) {
2146 ret
= repair_io_failure(fs_info
, start
, failrec
->len
,
2147 failrec
->logical
, page
,
2148 failrec
->failed_mirror
);
2156 ret
= free_io_failure(inode
, failrec
, did_repair
);
2162 * this is a generic handler for readpage errors (default
2163 * readpage_io_failed_hook). if other copies exist, read those and write back
2164 * good data to the failed position. does not investigate in remapping the
2165 * failed extent elsewhere, hoping the device will be smart enough to do this as
2169 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2170 struct page
*page
, u64 start
, u64 end
,
2173 struct io_failure_record
*failrec
= NULL
;
2175 struct extent_map
*em
;
2176 struct inode
*inode
= page
->mapping
->host
;
2177 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2178 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2179 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2181 struct btrfs_io_bio
*btrfs_failed_bio
;
2182 struct btrfs_io_bio
*btrfs_bio
;
2188 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2190 ret
= get_state_private(failure_tree
, start
, &private);
2192 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2195 failrec
->start
= start
;
2196 failrec
->len
= end
- start
+ 1;
2197 failrec
->this_mirror
= 0;
2198 failrec
->bio_flags
= 0;
2199 failrec
->in_validation
= 0;
2201 read_lock(&em_tree
->lock
);
2202 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2204 read_unlock(&em_tree
->lock
);
2209 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2210 free_extent_map(em
);
2213 read_unlock(&em_tree
->lock
);
2219 logical
= start
- em
->start
;
2220 logical
= em
->block_start
+ logical
;
2221 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2222 logical
= em
->block_start
;
2223 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2224 extent_set_compress_type(&failrec
->bio_flags
,
2227 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2228 "len=%llu\n", logical
, start
, failrec
->len
);
2229 failrec
->logical
= logical
;
2230 free_extent_map(em
);
2232 /* set the bits in the private failure tree */
2233 ret
= set_extent_bits(failure_tree
, start
, end
,
2234 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2236 ret
= set_state_private(failure_tree
, start
,
2237 (u64
)(unsigned long)failrec
);
2238 /* set the bits in the inode's tree */
2240 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2247 failrec
= (struct io_failure_record
*)(unsigned long)private;
2248 pr_debug("bio_readpage_error: (found) logical=%llu, "
2249 "start=%llu, len=%llu, validation=%d\n",
2250 failrec
->logical
, failrec
->start
, failrec
->len
,
2251 failrec
->in_validation
);
2253 * when data can be on disk more than twice, add to failrec here
2254 * (e.g. with a list for failed_mirror) to make
2255 * clean_io_failure() clean all those errors at once.
2258 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2259 failrec
->logical
, failrec
->len
);
2260 if (num_copies
== 1) {
2262 * we only have a single copy of the data, so don't bother with
2263 * all the retry and error correction code that follows. no
2264 * matter what the error is, it is very likely to persist.
2266 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2267 num_copies
, failrec
->this_mirror
, failed_mirror
);
2268 free_io_failure(inode
, failrec
, 0);
2273 * there are two premises:
2274 * a) deliver good data to the caller
2275 * b) correct the bad sectors on disk
2277 if (failed_bio
->bi_vcnt
> 1) {
2279 * to fulfill b), we need to know the exact failing sectors, as
2280 * we don't want to rewrite any more than the failed ones. thus,
2281 * we need separate read requests for the failed bio
2283 * if the following BUG_ON triggers, our validation request got
2284 * merged. we need separate requests for our algorithm to work.
2286 BUG_ON(failrec
->in_validation
);
2287 failrec
->in_validation
= 1;
2288 failrec
->this_mirror
= failed_mirror
;
2289 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2292 * we're ready to fulfill a) and b) alongside. get a good copy
2293 * of the failed sector and if we succeed, we have setup
2294 * everything for repair_io_failure to do the rest for us.
2296 if (failrec
->in_validation
) {
2297 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2298 failrec
->in_validation
= 0;
2299 failrec
->this_mirror
= 0;
2301 failrec
->failed_mirror
= failed_mirror
;
2302 failrec
->this_mirror
++;
2303 if (failrec
->this_mirror
== failed_mirror
)
2304 failrec
->this_mirror
++;
2305 read_mode
= READ_SYNC
;
2308 if (failrec
->this_mirror
> num_copies
) {
2309 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2310 num_copies
, failrec
->this_mirror
, failed_mirror
);
2311 free_io_failure(inode
, failrec
, 0);
2315 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2317 free_io_failure(inode
, failrec
, 0);
2320 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2321 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2322 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2323 bio
->bi_iter
.bi_size
= 0;
2325 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2326 if (btrfs_failed_bio
->csum
) {
2327 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2328 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2330 btrfs_bio
= btrfs_io_bio(bio
);
2331 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2332 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2333 phy_offset
*= csum_size
;
2334 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ phy_offset
,
2338 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2340 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2341 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2342 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2344 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2345 failrec
->this_mirror
,
2346 failrec
->bio_flags
, 0);
2350 /* lots and lots of room for performance fixes in the end_bio funcs */
2352 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2354 int uptodate
= (err
== 0);
2355 struct extent_io_tree
*tree
;
2358 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2360 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2361 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2362 end
, NULL
, uptodate
);
2368 ClearPageUptodate(page
);
2375 * after a writepage IO is done, we need to:
2376 * clear the uptodate bits on error
2377 * clear the writeback bits in the extent tree for this IO
2378 * end_page_writeback if the page has no more pending IO
2380 * Scheduling is not allowed, so the extent state tree is expected
2381 * to have one and only one object corresponding to this IO.
2383 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2385 struct bio_vec
*bvec
;
2390 bio_for_each_segment_all(bvec
, bio
, i
) {
2391 struct page
*page
= bvec
->bv_page
;
2393 /* We always issue full-page reads, but if some block
2394 * in a page fails to read, blk_update_request() will
2395 * advance bv_offset and adjust bv_len to compensate.
2396 * Print a warning for nonzero offsets, and an error
2397 * if they don't add up to a full page. */
2398 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2399 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2400 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2401 "partial page write in btrfs with offset %u and length %u",
2402 bvec
->bv_offset
, bvec
->bv_len
);
2404 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2405 "incomplete page write in btrfs with offset %u and "
2407 bvec
->bv_offset
, bvec
->bv_len
);
2410 start
= page_offset(page
);
2411 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2413 if (end_extent_writepage(page
, err
, start
, end
))
2416 end_page_writeback(page
);
2423 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2426 struct extent_state
*cached
= NULL
;
2427 u64 end
= start
+ len
- 1;
2429 if (uptodate
&& tree
->track_uptodate
)
2430 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2431 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2435 * after a readpage IO is done, we need to:
2436 * clear the uptodate bits on error
2437 * set the uptodate bits if things worked
2438 * set the page up to date if all extents in the tree are uptodate
2439 * clear the lock bit in the extent tree
2440 * unlock the page if there are no other extents locked for it
2442 * Scheduling is not allowed, so the extent state tree is expected
2443 * to have one and only one object corresponding to this IO.
2445 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2447 struct bio_vec
*bvec
;
2448 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2449 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2450 struct extent_io_tree
*tree
;
2455 u64 extent_start
= 0;
2464 bio_for_each_segment_all(bvec
, bio
, i
) {
2465 struct page
*page
= bvec
->bv_page
;
2466 struct inode
*inode
= page
->mapping
->host
;
2468 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2469 "mirror=%lu\n", (u64
)bio
->bi_iter
.bi_sector
, err
,
2470 io_bio
->mirror_num
);
2471 tree
= &BTRFS_I(inode
)->io_tree
;
2473 /* We always issue full-page reads, but if some block
2474 * in a page fails to read, blk_update_request() will
2475 * advance bv_offset and adjust bv_len to compensate.
2476 * Print a warning for nonzero offsets, and an error
2477 * if they don't add up to a full page. */
2478 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2479 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2480 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2481 "partial page read in btrfs with offset %u and length %u",
2482 bvec
->bv_offset
, bvec
->bv_len
);
2484 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2485 "incomplete page read in btrfs with offset %u and "
2487 bvec
->bv_offset
, bvec
->bv_len
);
2490 start
= page_offset(page
);
2491 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2494 mirror
= io_bio
->mirror_num
;
2495 if (likely(uptodate
&& tree
->ops
&&
2496 tree
->ops
->readpage_end_io_hook
)) {
2497 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2503 clean_io_failure(start
, page
);
2506 if (likely(uptodate
))
2509 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2510 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2512 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2516 * The generic bio_readpage_error handles errors the
2517 * following way: If possible, new read requests are
2518 * created and submitted and will end up in
2519 * end_bio_extent_readpage as well (if we're lucky, not
2520 * in the !uptodate case). In that case it returns 0 and
2521 * we just go on with the next page in our bio. If it
2522 * can't handle the error it will return -EIO and we
2523 * remain responsible for that page.
2525 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2529 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2536 if (likely(uptodate
)) {
2537 loff_t i_size
= i_size_read(inode
);
2538 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2541 /* Zero out the end if this page straddles i_size */
2542 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2543 if (page
->index
== end_index
&& offset
)
2544 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2545 SetPageUptodate(page
);
2547 ClearPageUptodate(page
);
2553 if (unlikely(!uptodate
)) {
2555 endio_readpage_release_extent(tree
,
2561 endio_readpage_release_extent(tree
, start
,
2562 end
- start
+ 1, 0);
2563 } else if (!extent_len
) {
2564 extent_start
= start
;
2565 extent_len
= end
+ 1 - start
;
2566 } else if (extent_start
+ extent_len
== start
) {
2567 extent_len
+= end
+ 1 - start
;
2569 endio_readpage_release_extent(tree
, extent_start
,
2570 extent_len
, uptodate
);
2571 extent_start
= start
;
2572 extent_len
= end
+ 1 - start
;
2577 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2580 io_bio
->end_io(io_bio
, err
);
2585 * this allocates from the btrfs_bioset. We're returning a bio right now
2586 * but you can call btrfs_io_bio for the appropriate container_of magic
2589 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2592 struct btrfs_io_bio
*btrfs_bio
;
2595 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2597 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2598 while (!bio
&& (nr_vecs
/= 2)) {
2599 bio
= bio_alloc_bioset(gfp_flags
,
2600 nr_vecs
, btrfs_bioset
);
2605 bio
->bi_bdev
= bdev
;
2606 bio
->bi_iter
.bi_sector
= first_sector
;
2607 btrfs_bio
= btrfs_io_bio(bio
);
2608 btrfs_bio
->csum
= NULL
;
2609 btrfs_bio
->csum_allocated
= NULL
;
2610 btrfs_bio
->end_io
= NULL
;
2615 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2617 return bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2621 /* this also allocates from the btrfs_bioset */
2622 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2624 struct btrfs_io_bio
*btrfs_bio
;
2627 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2629 btrfs_bio
= btrfs_io_bio(bio
);
2630 btrfs_bio
->csum
= NULL
;
2631 btrfs_bio
->csum_allocated
= NULL
;
2632 btrfs_bio
->end_io
= NULL
;
2638 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2639 int mirror_num
, unsigned long bio_flags
)
2642 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2643 struct page
*page
= bvec
->bv_page
;
2644 struct extent_io_tree
*tree
= bio
->bi_private
;
2647 start
= page_offset(page
) + bvec
->bv_offset
;
2649 bio
->bi_private
= NULL
;
2653 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2654 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2655 mirror_num
, bio_flags
, start
);
2657 btrfsic_submit_bio(rw
, bio
);
2659 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2665 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2666 unsigned long offset
, size_t size
, struct bio
*bio
,
2667 unsigned long bio_flags
)
2670 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2671 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2678 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2679 struct page
*page
, sector_t sector
,
2680 size_t size
, unsigned long offset
,
2681 struct block_device
*bdev
,
2682 struct bio
**bio_ret
,
2683 unsigned long max_pages
,
2684 bio_end_io_t end_io_func
,
2686 unsigned long prev_bio_flags
,
2687 unsigned long bio_flags
)
2693 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2694 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2695 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2697 if (bio_ret
&& *bio_ret
) {
2700 contig
= bio
->bi_iter
.bi_sector
== sector
;
2702 contig
= bio_end_sector(bio
) == sector
;
2704 if (prev_bio_flags
!= bio_flags
|| !contig
||
2705 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2706 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2707 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2716 if (this_compressed
)
2719 nr
= bio_get_nr_vecs(bdev
);
2721 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2725 bio_add_page(bio
, page
, page_size
, offset
);
2726 bio
->bi_end_io
= end_io_func
;
2727 bio
->bi_private
= tree
;
2732 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2737 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2740 if (!PagePrivate(page
)) {
2741 SetPagePrivate(page
);
2742 page_cache_get(page
);
2743 set_page_private(page
, (unsigned long)eb
);
2745 WARN_ON(page
->private != (unsigned long)eb
);
2749 void set_page_extent_mapped(struct page
*page
)
2751 if (!PagePrivate(page
)) {
2752 SetPagePrivate(page
);
2753 page_cache_get(page
);
2754 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2758 static struct extent_map
*
2759 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2760 u64 start
, u64 len
, get_extent_t
*get_extent
,
2761 struct extent_map
**em_cached
)
2763 struct extent_map
*em
;
2765 if (em_cached
&& *em_cached
) {
2767 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2768 start
< extent_map_end(em
)) {
2769 atomic_inc(&em
->refs
);
2773 free_extent_map(em
);
2777 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2778 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2780 atomic_inc(&em
->refs
);
2786 * basic readpage implementation. Locked extent state structs are inserted
2787 * into the tree that are removed when the IO is done (by the end_io
2789 * XXX JDM: This needs looking at to ensure proper page locking
2791 static int __do_readpage(struct extent_io_tree
*tree
,
2793 get_extent_t
*get_extent
,
2794 struct extent_map
**em_cached
,
2795 struct bio
**bio
, int mirror_num
,
2796 unsigned long *bio_flags
, int rw
)
2798 struct inode
*inode
= page
->mapping
->host
;
2799 u64 start
= page_offset(page
);
2800 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2804 u64 last_byte
= i_size_read(inode
);
2808 struct extent_map
*em
;
2809 struct block_device
*bdev
;
2812 int parent_locked
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2813 size_t pg_offset
= 0;
2815 size_t disk_io_size
;
2816 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2817 unsigned long this_bio_flag
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2819 set_page_extent_mapped(page
);
2822 if (!PageUptodate(page
)) {
2823 if (cleancache_get_page(page
) == 0) {
2824 BUG_ON(blocksize
!= PAGE_SIZE
);
2825 unlock_extent(tree
, start
, end
);
2830 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2832 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2835 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2836 userpage
= kmap_atomic(page
);
2837 memset(userpage
+ zero_offset
, 0, iosize
);
2838 flush_dcache_page(page
);
2839 kunmap_atomic(userpage
);
2842 while (cur
<= end
) {
2843 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2845 if (cur
>= last_byte
) {
2847 struct extent_state
*cached
= NULL
;
2849 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2850 userpage
= kmap_atomic(page
);
2851 memset(userpage
+ pg_offset
, 0, iosize
);
2852 flush_dcache_page(page
);
2853 kunmap_atomic(userpage
);
2854 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2857 unlock_extent_cached(tree
, cur
,
2862 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2863 end
- cur
+ 1, get_extent
, em_cached
);
2864 if (IS_ERR_OR_NULL(em
)) {
2867 unlock_extent(tree
, cur
, end
);
2870 extent_offset
= cur
- em
->start
;
2871 BUG_ON(extent_map_end(em
) <= cur
);
2874 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2875 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2876 extent_set_compress_type(&this_bio_flag
,
2880 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2881 cur_end
= min(extent_map_end(em
) - 1, end
);
2882 iosize
= ALIGN(iosize
, blocksize
);
2883 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2884 disk_io_size
= em
->block_len
;
2885 sector
= em
->block_start
>> 9;
2887 sector
= (em
->block_start
+ extent_offset
) >> 9;
2888 disk_io_size
= iosize
;
2891 block_start
= em
->block_start
;
2892 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2893 block_start
= EXTENT_MAP_HOLE
;
2894 free_extent_map(em
);
2897 /* we've found a hole, just zero and go on */
2898 if (block_start
== EXTENT_MAP_HOLE
) {
2900 struct extent_state
*cached
= NULL
;
2902 userpage
= kmap_atomic(page
);
2903 memset(userpage
+ pg_offset
, 0, iosize
);
2904 flush_dcache_page(page
);
2905 kunmap_atomic(userpage
);
2907 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2909 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2912 pg_offset
+= iosize
;
2915 /* the get_extent function already copied into the page */
2916 if (test_range_bit(tree
, cur
, cur_end
,
2917 EXTENT_UPTODATE
, 1, NULL
)) {
2918 check_page_uptodate(tree
, page
);
2920 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2922 pg_offset
+= iosize
;
2925 /* we have an inline extent but it didn't get marked up
2926 * to date. Error out
2928 if (block_start
== EXTENT_MAP_INLINE
) {
2931 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2933 pg_offset
+= iosize
;
2938 ret
= submit_extent_page(rw
, tree
, page
,
2939 sector
, disk_io_size
, pg_offset
,
2941 end_bio_extent_readpage
, mirror_num
,
2946 *bio_flags
= this_bio_flag
;
2950 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2953 pg_offset
+= iosize
;
2957 if (!PageError(page
))
2958 SetPageUptodate(page
);
2964 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
2965 struct page
*pages
[], int nr_pages
,
2967 get_extent_t
*get_extent
,
2968 struct extent_map
**em_cached
,
2969 struct bio
**bio
, int mirror_num
,
2970 unsigned long *bio_flags
, int rw
)
2972 struct inode
*inode
;
2973 struct btrfs_ordered_extent
*ordered
;
2976 inode
= pages
[0]->mapping
->host
;
2978 lock_extent(tree
, start
, end
);
2979 ordered
= btrfs_lookup_ordered_range(inode
, start
,
2983 unlock_extent(tree
, start
, end
);
2984 btrfs_start_ordered_extent(inode
, ordered
, 1);
2985 btrfs_put_ordered_extent(ordered
);
2988 for (index
= 0; index
< nr_pages
; index
++) {
2989 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
2990 mirror_num
, bio_flags
, rw
);
2991 page_cache_release(pages
[index
]);
2995 static void __extent_readpages(struct extent_io_tree
*tree
,
2996 struct page
*pages
[],
2997 int nr_pages
, get_extent_t
*get_extent
,
2998 struct extent_map
**em_cached
,
2999 struct bio
**bio
, int mirror_num
,
3000 unsigned long *bio_flags
, int rw
)
3006 int first_index
= 0;
3008 for (index
= 0; index
< nr_pages
; index
++) {
3009 page_start
= page_offset(pages
[index
]);
3012 end
= start
+ PAGE_CACHE_SIZE
- 1;
3013 first_index
= index
;
3014 } else if (end
+ 1 == page_start
) {
3015 end
+= PAGE_CACHE_SIZE
;
3017 __do_contiguous_readpages(tree
, &pages
[first_index
],
3018 index
- first_index
, start
,
3019 end
, get_extent
, em_cached
,
3020 bio
, mirror_num
, bio_flags
,
3023 end
= start
+ PAGE_CACHE_SIZE
- 1;
3024 first_index
= index
;
3029 __do_contiguous_readpages(tree
, &pages
[first_index
],
3030 index
- first_index
, start
,
3031 end
, get_extent
, em_cached
, bio
,
3032 mirror_num
, bio_flags
, rw
);
3035 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3037 get_extent_t
*get_extent
,
3038 struct bio
**bio
, int mirror_num
,
3039 unsigned long *bio_flags
, int rw
)
3041 struct inode
*inode
= page
->mapping
->host
;
3042 struct btrfs_ordered_extent
*ordered
;
3043 u64 start
= page_offset(page
);
3044 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3048 lock_extent(tree
, start
, end
);
3049 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
3052 unlock_extent(tree
, start
, end
);
3053 btrfs_start_ordered_extent(inode
, ordered
, 1);
3054 btrfs_put_ordered_extent(ordered
);
3057 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3062 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3063 get_extent_t
*get_extent
, int mirror_num
)
3065 struct bio
*bio
= NULL
;
3066 unsigned long bio_flags
= 0;
3069 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3072 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3076 int extent_read_full_page_nolock(struct extent_io_tree
*tree
, struct page
*page
,
3077 get_extent_t
*get_extent
, int mirror_num
)
3079 struct bio
*bio
= NULL
;
3080 unsigned long bio_flags
= EXTENT_BIO_PARENT_LOCKED
;
3083 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, &bio
, mirror_num
,
3086 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3090 static noinline
void update_nr_written(struct page
*page
,
3091 struct writeback_control
*wbc
,
3092 unsigned long nr_written
)
3094 wbc
->nr_to_write
-= nr_written
;
3095 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
3096 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
3097 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
3101 * the writepage semantics are similar to regular writepage. extent
3102 * records are inserted to lock ranges in the tree, and as dirty areas
3103 * are found, they are marked writeback. Then the lock bits are removed
3104 * and the end_io handler clears the writeback ranges
3106 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3109 struct inode
*inode
= page
->mapping
->host
;
3110 struct extent_page_data
*epd
= data
;
3111 struct extent_io_tree
*tree
= epd
->tree
;
3112 u64 start
= page_offset(page
);
3114 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3118 u64 last_byte
= i_size_read(inode
);
3122 struct extent_state
*cached_state
= NULL
;
3123 struct extent_map
*em
;
3124 struct block_device
*bdev
;
3127 size_t pg_offset
= 0;
3129 loff_t i_size
= i_size_read(inode
);
3130 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
3136 unsigned long nr_written
= 0;
3137 bool fill_delalloc
= true;
3139 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3140 write_flags
= WRITE_SYNC
;
3142 write_flags
= WRITE
;
3144 trace___extent_writepage(page
, inode
, wbc
);
3146 WARN_ON(!PageLocked(page
));
3148 ClearPageError(page
);
3150 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
3151 if (page
->index
> end_index
||
3152 (page
->index
== end_index
&& !pg_offset
)) {
3153 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
3158 if (page
->index
== end_index
) {
3161 userpage
= kmap_atomic(page
);
3162 memset(userpage
+ pg_offset
, 0,
3163 PAGE_CACHE_SIZE
- pg_offset
);
3164 kunmap_atomic(userpage
);
3165 flush_dcache_page(page
);
3169 set_page_extent_mapped(page
);
3171 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
3172 fill_delalloc
= false;
3174 delalloc_start
= start
;
3177 if (!epd
->extent_locked
&& fill_delalloc
) {
3178 u64 delalloc_to_write
= 0;
3180 * make sure the wbc mapping index is at least updated
3183 update_nr_written(page
, wbc
, 0);
3185 while (delalloc_end
< page_end
) {
3186 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3191 if (nr_delalloc
== 0) {
3192 delalloc_start
= delalloc_end
+ 1;
3195 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3200 /* File system has been set read-only */
3206 * delalloc_end is already one less than the total
3207 * length, so we don't subtract one from
3210 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3213 delalloc_start
= delalloc_end
+ 1;
3215 if (wbc
->nr_to_write
< delalloc_to_write
) {
3218 if (delalloc_to_write
< thresh
* 2)
3219 thresh
= delalloc_to_write
;
3220 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3224 /* did the fill delalloc function already unlock and start
3230 * we've unlocked the page, so we can't update
3231 * the mapping's writeback index, just update
3234 wbc
->nr_to_write
-= nr_written
;
3238 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3239 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3242 /* Fixup worker will requeue */
3244 wbc
->pages_skipped
++;
3246 redirty_page_for_writepage(wbc
, page
);
3247 update_nr_written(page
, wbc
, nr_written
);
3255 * we don't want to touch the inode after unlocking the page,
3256 * so we update the mapping writeback index now
3258 update_nr_written(page
, wbc
, nr_written
+ 1);
3261 if (last_byte
<= start
) {
3262 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3263 tree
->ops
->writepage_end_io_hook(page
, start
,
3268 blocksize
= inode
->i_sb
->s_blocksize
;
3270 while (cur
<= end
) {
3271 if (cur
>= last_byte
) {
3272 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3273 tree
->ops
->writepage_end_io_hook(page
, cur
,
3277 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3279 if (IS_ERR_OR_NULL(em
)) {
3284 extent_offset
= cur
- em
->start
;
3285 BUG_ON(extent_map_end(em
) <= cur
);
3287 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
3288 iosize
= ALIGN(iosize
, blocksize
);
3289 sector
= (em
->block_start
+ extent_offset
) >> 9;
3291 block_start
= em
->block_start
;
3292 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3293 free_extent_map(em
);
3297 * compressed and inline extents are written through other
3300 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3301 block_start
== EXTENT_MAP_INLINE
) {
3303 * end_io notification does not happen here for
3304 * compressed extents
3306 if (!compressed
&& tree
->ops
&&
3307 tree
->ops
->writepage_end_io_hook
)
3308 tree
->ops
->writepage_end_io_hook(page
, cur
,
3311 else if (compressed
) {
3312 /* we don't want to end_page_writeback on
3313 * a compressed extent. this happens
3320 pg_offset
+= iosize
;
3323 /* leave this out until we have a page_mkwrite call */
3324 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
3325 EXTENT_DIRTY
, 0, NULL
)) {
3327 pg_offset
+= iosize
;
3331 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3332 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3340 unsigned long max_nr
= end_index
+ 1;
3342 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3343 if (!PageWriteback(page
)) {
3344 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3345 "page %lu not writeback, cur %llu end %llu",
3346 page
->index
, cur
, end
);
3349 ret
= submit_extent_page(write_flags
, tree
, page
,
3350 sector
, iosize
, pg_offset
,
3351 bdev
, &epd
->bio
, max_nr
,
3352 end_bio_extent_writepage
,
3358 pg_offset
+= iosize
;
3363 /* make sure the mapping tag for page dirty gets cleared */
3364 set_page_writeback(page
);
3365 end_page_writeback(page
);
3371 /* drop our reference on any cached states */
3372 free_extent_state(cached_state
);
3376 static int eb_wait(void *word
)
3382 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3384 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3385 TASK_UNINTERRUPTIBLE
);
3388 static int lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3389 struct btrfs_fs_info
*fs_info
,
3390 struct extent_page_data
*epd
)
3392 unsigned long i
, num_pages
;
3396 if (!btrfs_try_tree_write_lock(eb
)) {
3398 flush_write_bio(epd
);
3399 btrfs_tree_lock(eb
);
3402 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3403 btrfs_tree_unlock(eb
);
3407 flush_write_bio(epd
);
3411 wait_on_extent_buffer_writeback(eb
);
3412 btrfs_tree_lock(eb
);
3413 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3415 btrfs_tree_unlock(eb
);
3420 * We need to do this to prevent races in people who check if the eb is
3421 * under IO since we can end up having no IO bits set for a short period
3424 spin_lock(&eb
->refs_lock
);
3425 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3426 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3427 spin_unlock(&eb
->refs_lock
);
3428 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3429 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3431 fs_info
->dirty_metadata_batch
);
3434 spin_unlock(&eb
->refs_lock
);
3437 btrfs_tree_unlock(eb
);
3442 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3443 for (i
= 0; i
< num_pages
; i
++) {
3444 struct page
*p
= extent_buffer_page(eb
, i
);
3446 if (!trylock_page(p
)) {
3448 flush_write_bio(epd
);
3458 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3460 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3461 smp_mb__after_atomic();
3462 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3465 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3467 struct bio_vec
*bvec
;
3468 struct extent_buffer
*eb
;
3471 bio_for_each_segment_all(bvec
, bio
, i
) {
3472 struct page
*page
= bvec
->bv_page
;
3474 eb
= (struct extent_buffer
*)page
->private;
3476 done
= atomic_dec_and_test(&eb
->io_pages
);
3478 if (err
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3479 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3480 ClearPageUptodate(page
);
3484 end_page_writeback(page
);
3489 end_extent_buffer_writeback(eb
);
3495 static int write_one_eb(struct extent_buffer
*eb
,
3496 struct btrfs_fs_info
*fs_info
,
3497 struct writeback_control
*wbc
,
3498 struct extent_page_data
*epd
)
3500 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3501 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3502 u64 offset
= eb
->start
;
3503 unsigned long i
, num_pages
;
3504 unsigned long bio_flags
= 0;
3505 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3508 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3509 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3510 atomic_set(&eb
->io_pages
, num_pages
);
3511 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3512 bio_flags
= EXTENT_BIO_TREE_LOG
;
3514 for (i
= 0; i
< num_pages
; i
++) {
3515 struct page
*p
= extent_buffer_page(eb
, i
);
3517 clear_page_dirty_for_io(p
);
3518 set_page_writeback(p
);
3519 ret
= submit_extent_page(rw
, tree
, p
, offset
>> 9,
3520 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3521 -1, end_bio_extent_buffer_writepage
,
3522 0, epd
->bio_flags
, bio_flags
);
3523 epd
->bio_flags
= bio_flags
;
3525 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3527 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3528 end_extent_buffer_writeback(eb
);
3532 offset
+= PAGE_CACHE_SIZE
;
3533 update_nr_written(p
, wbc
, 1);
3537 if (unlikely(ret
)) {
3538 for (; i
< num_pages
; i
++) {
3539 struct page
*p
= extent_buffer_page(eb
, i
);
3547 int btree_write_cache_pages(struct address_space
*mapping
,
3548 struct writeback_control
*wbc
)
3550 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3551 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3552 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3553 struct extent_page_data epd
= {
3557 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3562 int nr_to_write_done
= 0;
3563 struct pagevec pvec
;
3566 pgoff_t end
; /* Inclusive */
3570 pagevec_init(&pvec
, 0);
3571 if (wbc
->range_cyclic
) {
3572 index
= mapping
->writeback_index
; /* Start from prev offset */
3575 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3576 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3579 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3580 tag
= PAGECACHE_TAG_TOWRITE
;
3582 tag
= PAGECACHE_TAG_DIRTY
;
3584 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3585 tag_pages_for_writeback(mapping
, index
, end
);
3586 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3587 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3588 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3592 for (i
= 0; i
< nr_pages
; i
++) {
3593 struct page
*page
= pvec
.pages
[i
];
3595 if (!PagePrivate(page
))
3598 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3603 spin_lock(&mapping
->private_lock
);
3604 if (!PagePrivate(page
)) {
3605 spin_unlock(&mapping
->private_lock
);
3609 eb
= (struct extent_buffer
*)page
->private;
3612 * Shouldn't happen and normally this would be a BUG_ON
3613 * but no sense in crashing the users box for something
3614 * we can survive anyway.
3617 spin_unlock(&mapping
->private_lock
);
3621 if (eb
== prev_eb
) {
3622 spin_unlock(&mapping
->private_lock
);
3626 ret
= atomic_inc_not_zero(&eb
->refs
);
3627 spin_unlock(&mapping
->private_lock
);
3632 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3634 free_extent_buffer(eb
);
3638 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3641 free_extent_buffer(eb
);
3644 free_extent_buffer(eb
);
3647 * the filesystem may choose to bump up nr_to_write.
3648 * We have to make sure to honor the new nr_to_write
3651 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3653 pagevec_release(&pvec
);
3656 if (!scanned
&& !done
) {
3658 * We hit the last page and there is more work to be done: wrap
3659 * back to the start of the file
3665 flush_write_bio(&epd
);
3670 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3671 * @mapping: address space structure to write
3672 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3673 * @writepage: function called for each page
3674 * @data: data passed to writepage function
3676 * If a page is already under I/O, write_cache_pages() skips it, even
3677 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3678 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3679 * and msync() need to guarantee that all the data which was dirty at the time
3680 * the call was made get new I/O started against them. If wbc->sync_mode is
3681 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3682 * existing IO to complete.
3684 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3685 struct address_space
*mapping
,
3686 struct writeback_control
*wbc
,
3687 writepage_t writepage
, void *data
,
3688 void (*flush_fn
)(void *))
3690 struct inode
*inode
= mapping
->host
;
3693 int nr_to_write_done
= 0;
3694 struct pagevec pvec
;
3697 pgoff_t end
; /* Inclusive */
3702 * We have to hold onto the inode so that ordered extents can do their
3703 * work when the IO finishes. The alternative to this is failing to add
3704 * an ordered extent if the igrab() fails there and that is a huge pain
3705 * to deal with, so instead just hold onto the inode throughout the
3706 * writepages operation. If it fails here we are freeing up the inode
3707 * anyway and we'd rather not waste our time writing out stuff that is
3708 * going to be truncated anyway.
3713 pagevec_init(&pvec
, 0);
3714 if (wbc
->range_cyclic
) {
3715 index
= mapping
->writeback_index
; /* Start from prev offset */
3718 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3719 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3722 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3723 tag
= PAGECACHE_TAG_TOWRITE
;
3725 tag
= PAGECACHE_TAG_DIRTY
;
3727 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3728 tag_pages_for_writeback(mapping
, index
, end
);
3729 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3730 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3731 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3735 for (i
= 0; i
< nr_pages
; i
++) {
3736 struct page
*page
= pvec
.pages
[i
];
3739 * At this point we hold neither mapping->tree_lock nor
3740 * lock on the page itself: the page may be truncated or
3741 * invalidated (changing page->mapping to NULL), or even
3742 * swizzled back from swapper_space to tmpfs file
3745 if (!trylock_page(page
)) {
3750 if (unlikely(page
->mapping
!= mapping
)) {
3755 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3761 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3762 if (PageWriteback(page
))
3764 wait_on_page_writeback(page
);
3767 if (PageWriteback(page
) ||
3768 !clear_page_dirty_for_io(page
)) {
3773 ret
= (*writepage
)(page
, wbc
, data
);
3775 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3783 * the filesystem may choose to bump up nr_to_write.
3784 * We have to make sure to honor the new nr_to_write
3787 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3789 pagevec_release(&pvec
);
3792 if (!scanned
&& !done
) {
3794 * We hit the last page and there is more work to be done: wrap
3795 * back to the start of the file
3801 btrfs_add_delayed_iput(inode
);
3805 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3814 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
3815 BUG_ON(ret
< 0); /* -ENOMEM */
3820 static noinline
void flush_write_bio(void *data
)
3822 struct extent_page_data
*epd
= data
;
3823 flush_epd_write_bio(epd
);
3826 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3827 get_extent_t
*get_extent
,
3828 struct writeback_control
*wbc
)
3831 struct extent_page_data epd
= {
3834 .get_extent
= get_extent
,
3836 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3840 ret
= __extent_writepage(page
, wbc
, &epd
);
3842 flush_epd_write_bio(&epd
);
3846 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3847 u64 start
, u64 end
, get_extent_t
*get_extent
,
3851 struct address_space
*mapping
= inode
->i_mapping
;
3853 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3856 struct extent_page_data epd
= {
3859 .get_extent
= get_extent
,
3861 .sync_io
= mode
== WB_SYNC_ALL
,
3864 struct writeback_control wbc_writepages
= {
3866 .nr_to_write
= nr_pages
* 2,
3867 .range_start
= start
,
3868 .range_end
= end
+ 1,
3871 while (start
<= end
) {
3872 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3873 if (clear_page_dirty_for_io(page
))
3874 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3876 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3877 tree
->ops
->writepage_end_io_hook(page
, start
,
3878 start
+ PAGE_CACHE_SIZE
- 1,
3882 page_cache_release(page
);
3883 start
+= PAGE_CACHE_SIZE
;
3886 flush_epd_write_bio(&epd
);
3890 int extent_writepages(struct extent_io_tree
*tree
,
3891 struct address_space
*mapping
,
3892 get_extent_t
*get_extent
,
3893 struct writeback_control
*wbc
)
3896 struct extent_page_data epd
= {
3899 .get_extent
= get_extent
,
3901 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3905 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3906 __extent_writepage
, &epd
,
3908 flush_epd_write_bio(&epd
);
3912 int extent_readpages(struct extent_io_tree
*tree
,
3913 struct address_space
*mapping
,
3914 struct list_head
*pages
, unsigned nr_pages
,
3915 get_extent_t get_extent
)
3917 struct bio
*bio
= NULL
;
3919 unsigned long bio_flags
= 0;
3920 struct page
*pagepool
[16];
3922 struct extent_map
*em_cached
= NULL
;
3925 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3926 page
= list_entry(pages
->prev
, struct page
, lru
);
3928 prefetchw(&page
->flags
);
3929 list_del(&page
->lru
);
3930 if (add_to_page_cache_lru(page
, mapping
,
3931 page
->index
, GFP_NOFS
)) {
3932 page_cache_release(page
);
3936 pagepool
[nr
++] = page
;
3937 if (nr
< ARRAY_SIZE(pagepool
))
3939 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3940 &bio
, 0, &bio_flags
, READ
);
3944 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3945 &bio
, 0, &bio_flags
, READ
);
3948 free_extent_map(em_cached
);
3950 BUG_ON(!list_empty(pages
));
3952 return submit_one_bio(READ
, bio
, 0, bio_flags
);
3957 * basic invalidatepage code, this waits on any locked or writeback
3958 * ranges corresponding to the page, and then deletes any extent state
3959 * records from the tree
3961 int extent_invalidatepage(struct extent_io_tree
*tree
,
3962 struct page
*page
, unsigned long offset
)
3964 struct extent_state
*cached_state
= NULL
;
3965 u64 start
= page_offset(page
);
3966 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3967 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
3969 start
+= ALIGN(offset
, blocksize
);
3973 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
3974 wait_on_page_writeback(page
);
3975 clear_extent_bit(tree
, start
, end
,
3976 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3977 EXTENT_DO_ACCOUNTING
,
3978 1, 1, &cached_state
, GFP_NOFS
);
3983 * a helper for releasepage, this tests for areas of the page that
3984 * are locked or under IO and drops the related state bits if it is safe
3987 static int try_release_extent_state(struct extent_map_tree
*map
,
3988 struct extent_io_tree
*tree
,
3989 struct page
*page
, gfp_t mask
)
3991 u64 start
= page_offset(page
);
3992 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3995 if (test_range_bit(tree
, start
, end
,
3996 EXTENT_IOBITS
, 0, NULL
))
3999 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4002 * at this point we can safely clear everything except the
4003 * locked bit and the nodatasum bit
4005 ret
= clear_extent_bit(tree
, start
, end
,
4006 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4009 /* if clear_extent_bit failed for enomem reasons,
4010 * we can't allow the release to continue.
4021 * a helper for releasepage. As long as there are no locked extents
4022 * in the range corresponding to the page, both state records and extent
4023 * map records are removed
4025 int try_release_extent_mapping(struct extent_map_tree
*map
,
4026 struct extent_io_tree
*tree
, struct page
*page
,
4029 struct extent_map
*em
;
4030 u64 start
= page_offset(page
);
4031 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4033 if ((mask
& __GFP_WAIT
) &&
4034 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
4036 while (start
<= end
) {
4037 len
= end
- start
+ 1;
4038 write_lock(&map
->lock
);
4039 em
= lookup_extent_mapping(map
, start
, len
);
4041 write_unlock(&map
->lock
);
4044 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4045 em
->start
!= start
) {
4046 write_unlock(&map
->lock
);
4047 free_extent_map(em
);
4050 if (!test_range_bit(tree
, em
->start
,
4051 extent_map_end(em
) - 1,
4052 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4054 remove_extent_mapping(map
, em
);
4055 /* once for the rb tree */
4056 free_extent_map(em
);
4058 start
= extent_map_end(em
);
4059 write_unlock(&map
->lock
);
4062 free_extent_map(em
);
4065 return try_release_extent_state(map
, tree
, page
, mask
);
4069 * helper function for fiemap, which doesn't want to see any holes.
4070 * This maps until we find something past 'last'
4072 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4075 get_extent_t
*get_extent
)
4077 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4078 struct extent_map
*em
;
4085 len
= last
- offset
;
4088 len
= ALIGN(len
, sectorsize
);
4089 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4090 if (IS_ERR_OR_NULL(em
))
4093 /* if this isn't a hole return it */
4094 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4095 em
->block_start
!= EXTENT_MAP_HOLE
) {
4099 /* this is a hole, advance to the next extent */
4100 offset
= extent_map_end(em
);
4101 free_extent_map(em
);
4108 static noinline
int count_ext_ref(u64 inum
, u64 offset
, u64 root_id
, void *ctx
)
4110 unsigned long cnt
= *((unsigned long *)ctx
);
4113 *((unsigned long *)ctx
) = cnt
;
4115 /* Now we're sure that the extent is shared. */
4121 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4122 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4126 u64 max
= start
+ len
;
4130 u64 last_for_get_extent
= 0;
4132 u64 isize
= i_size_read(inode
);
4133 struct btrfs_key found_key
;
4134 struct extent_map
*em
= NULL
;
4135 struct extent_state
*cached_state
= NULL
;
4136 struct btrfs_path
*path
;
4145 path
= btrfs_alloc_path();
4148 path
->leave_spinning
= 1;
4150 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
4151 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
4154 * lookup the last file extent. We're not using i_size here
4155 * because there might be preallocation past i_size
4157 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
4158 path
, btrfs_ino(inode
), -1, 0);
4160 btrfs_free_path(path
);
4165 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4166 found_type
= btrfs_key_type(&found_key
);
4168 /* No extents, but there might be delalloc bits */
4169 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4170 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4171 /* have to trust i_size as the end */
4173 last_for_get_extent
= isize
;
4176 * remember the start of the last extent. There are a
4177 * bunch of different factors that go into the length of the
4178 * extent, so its much less complex to remember where it started
4180 last
= found_key
.offset
;
4181 last_for_get_extent
= last
+ 1;
4183 btrfs_release_path(path
);
4186 * we might have some extents allocated but more delalloc past those
4187 * extents. so, we trust isize unless the start of the last extent is
4192 last_for_get_extent
= isize
;
4195 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1, 0,
4198 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4208 u64 offset_in_extent
= 0;
4210 /* break if the extent we found is outside the range */
4211 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4215 * get_extent may return an extent that starts before our
4216 * requested range. We have to make sure the ranges
4217 * we return to fiemap always move forward and don't
4218 * overlap, so adjust the offsets here
4220 em_start
= max(em
->start
, off
);
4223 * record the offset from the start of the extent
4224 * for adjusting the disk offset below. Only do this if the
4225 * extent isn't compressed since our in ram offset may be past
4226 * what we have actually allocated on disk.
4228 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4229 offset_in_extent
= em_start
- em
->start
;
4230 em_end
= extent_map_end(em
);
4231 em_len
= em_end
- em_start
;
4236 * bump off for our next call to get_extent
4238 off
= extent_map_end(em
);
4242 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4244 flags
|= FIEMAP_EXTENT_LAST
;
4245 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4246 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4247 FIEMAP_EXTENT_NOT_ALIGNED
);
4248 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4249 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4250 FIEMAP_EXTENT_UNKNOWN
);
4252 unsigned long ref_cnt
= 0;
4254 disko
= em
->block_start
+ offset_in_extent
;
4257 * As btrfs supports shared space, this information
4258 * can be exported to userspace tools via
4259 * flag FIEMAP_EXTENT_SHARED.
4261 ret
= iterate_inodes_from_logical(
4263 BTRFS_I(inode
)->root
->fs_info
,
4264 path
, count_ext_ref
, &ref_cnt
);
4265 if (ret
< 0 && ret
!= -ENOENT
)
4269 flags
|= FIEMAP_EXTENT_SHARED
;
4271 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4272 flags
|= FIEMAP_EXTENT_ENCODED
;
4274 free_extent_map(em
);
4276 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4277 (last
== (u64
)-1 && isize
<= em_end
)) {
4278 flags
|= FIEMAP_EXTENT_LAST
;
4282 /* now scan forward to see if this is really the last extent. */
4283 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4290 flags
|= FIEMAP_EXTENT_LAST
;
4293 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4299 free_extent_map(em
);
4301 btrfs_free_path(path
);
4302 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4303 &cached_state
, GFP_NOFS
);
4307 static void __free_extent_buffer(struct extent_buffer
*eb
)
4309 btrfs_leak_debug_del(&eb
->leak_list
);
4310 kmem_cache_free(extent_buffer_cache
, eb
);
4313 int extent_buffer_under_io(struct extent_buffer
*eb
)
4315 return (atomic_read(&eb
->io_pages
) ||
4316 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4317 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4321 * Helper for releasing extent buffer page.
4323 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4324 unsigned long start_idx
)
4326 unsigned long index
;
4327 unsigned long num_pages
;
4329 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4331 BUG_ON(extent_buffer_under_io(eb
));
4333 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4334 index
= start_idx
+ num_pages
;
4335 if (start_idx
>= index
)
4340 page
= extent_buffer_page(eb
, index
);
4341 if (page
&& mapped
) {
4342 spin_lock(&page
->mapping
->private_lock
);
4344 * We do this since we'll remove the pages after we've
4345 * removed the eb from the radix tree, so we could race
4346 * and have this page now attached to the new eb. So
4347 * only clear page_private if it's still connected to
4350 if (PagePrivate(page
) &&
4351 page
->private == (unsigned long)eb
) {
4352 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4353 BUG_ON(PageDirty(page
));
4354 BUG_ON(PageWriteback(page
));
4356 * We need to make sure we haven't be attached
4359 ClearPagePrivate(page
);
4360 set_page_private(page
, 0);
4361 /* One for the page private */
4362 page_cache_release(page
);
4364 spin_unlock(&page
->mapping
->private_lock
);
4368 /* One for when we alloced the page */
4369 page_cache_release(page
);
4371 } while (index
!= start_idx
);
4375 * Helper for releasing the extent buffer.
4377 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4379 btrfs_release_extent_buffer_page(eb
, 0);
4380 __free_extent_buffer(eb
);
4383 static struct extent_buffer
*
4384 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4385 unsigned long len
, gfp_t mask
)
4387 struct extent_buffer
*eb
= NULL
;
4389 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
4394 eb
->fs_info
= fs_info
;
4396 rwlock_init(&eb
->lock
);
4397 atomic_set(&eb
->write_locks
, 0);
4398 atomic_set(&eb
->read_locks
, 0);
4399 atomic_set(&eb
->blocking_readers
, 0);
4400 atomic_set(&eb
->blocking_writers
, 0);
4401 atomic_set(&eb
->spinning_readers
, 0);
4402 atomic_set(&eb
->spinning_writers
, 0);
4403 eb
->lock_nested
= 0;
4404 init_waitqueue_head(&eb
->write_lock_wq
);
4405 init_waitqueue_head(&eb
->read_lock_wq
);
4407 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4409 spin_lock_init(&eb
->refs_lock
);
4410 atomic_set(&eb
->refs
, 1);
4411 atomic_set(&eb
->io_pages
, 0);
4414 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4416 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4417 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4418 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4423 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4427 struct extent_buffer
*new;
4428 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4430 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_NOFS
);
4434 for (i
= 0; i
< num_pages
; i
++) {
4435 p
= alloc_page(GFP_NOFS
);
4437 btrfs_release_extent_buffer(new);
4440 attach_extent_buffer_page(new, p
);
4441 WARN_ON(PageDirty(p
));
4446 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4447 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4448 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4453 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
4455 struct extent_buffer
*eb
;
4456 unsigned long num_pages
= num_extent_pages(0, len
);
4459 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_NOFS
);
4463 for (i
= 0; i
< num_pages
; i
++) {
4464 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4468 set_extent_buffer_uptodate(eb
);
4469 btrfs_set_header_nritems(eb
, 0);
4470 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4475 __free_page(eb
->pages
[i
- 1]);
4476 __free_extent_buffer(eb
);
4480 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4483 /* the ref bit is tricky. We have to make sure it is set
4484 * if we have the buffer dirty. Otherwise the
4485 * code to free a buffer can end up dropping a dirty
4488 * Once the ref bit is set, it won't go away while the
4489 * buffer is dirty or in writeback, and it also won't
4490 * go away while we have the reference count on the
4493 * We can't just set the ref bit without bumping the
4494 * ref on the eb because free_extent_buffer might
4495 * see the ref bit and try to clear it. If this happens
4496 * free_extent_buffer might end up dropping our original
4497 * ref by mistake and freeing the page before we are able
4498 * to add one more ref.
4500 * So bump the ref count first, then set the bit. If someone
4501 * beat us to it, drop the ref we added.
4503 refs
= atomic_read(&eb
->refs
);
4504 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4507 spin_lock(&eb
->refs_lock
);
4508 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4509 atomic_inc(&eb
->refs
);
4510 spin_unlock(&eb
->refs_lock
);
4513 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4514 struct page
*accessed
)
4516 unsigned long num_pages
, i
;
4518 check_buffer_tree_ref(eb
);
4520 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4521 for (i
= 0; i
< num_pages
; i
++) {
4522 struct page
*p
= extent_buffer_page(eb
, i
);
4524 mark_page_accessed(p
);
4528 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4531 struct extent_buffer
*eb
;
4534 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4535 start
>> PAGE_CACHE_SHIFT
);
4536 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4538 mark_extent_buffer_accessed(eb
, NULL
);
4546 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4547 u64 start
, unsigned long len
)
4549 unsigned long num_pages
= num_extent_pages(start
, len
);
4551 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4552 struct extent_buffer
*eb
;
4553 struct extent_buffer
*exists
= NULL
;
4555 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4559 eb
= find_extent_buffer(fs_info
, start
);
4563 eb
= __alloc_extent_buffer(fs_info
, start
, len
, GFP_NOFS
);
4567 for (i
= 0; i
< num_pages
; i
++, index
++) {
4568 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4572 spin_lock(&mapping
->private_lock
);
4573 if (PagePrivate(p
)) {
4575 * We could have already allocated an eb for this page
4576 * and attached one so lets see if we can get a ref on
4577 * the existing eb, and if we can we know it's good and
4578 * we can just return that one, else we know we can just
4579 * overwrite page->private.
4581 exists
= (struct extent_buffer
*)p
->private;
4582 if (atomic_inc_not_zero(&exists
->refs
)) {
4583 spin_unlock(&mapping
->private_lock
);
4585 page_cache_release(p
);
4586 mark_extent_buffer_accessed(exists
, p
);
4591 * Do this so attach doesn't complain and we need to
4592 * drop the ref the old guy had.
4594 ClearPagePrivate(p
);
4595 WARN_ON(PageDirty(p
));
4596 page_cache_release(p
);
4598 attach_extent_buffer_page(eb
, p
);
4599 spin_unlock(&mapping
->private_lock
);
4600 WARN_ON(PageDirty(p
));
4602 if (!PageUptodate(p
))
4606 * see below about how we avoid a nasty race with release page
4607 * and why we unlock later
4611 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4613 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4617 spin_lock(&fs_info
->buffer_lock
);
4618 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4619 start
>> PAGE_CACHE_SHIFT
, eb
);
4620 spin_unlock(&fs_info
->buffer_lock
);
4621 radix_tree_preload_end();
4622 if (ret
== -EEXIST
) {
4623 exists
= find_extent_buffer(fs_info
, start
);
4629 /* add one reference for the tree */
4630 check_buffer_tree_ref(eb
);
4631 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4634 * there is a race where release page may have
4635 * tried to find this extent buffer in the radix
4636 * but failed. It will tell the VM it is safe to
4637 * reclaim the, and it will clear the page private bit.
4638 * We must make sure to set the page private bit properly
4639 * after the extent buffer is in the radix tree so
4640 * it doesn't get lost
4642 SetPageChecked(eb
->pages
[0]);
4643 for (i
= 1; i
< num_pages
; i
++) {
4644 p
= extent_buffer_page(eb
, i
);
4645 ClearPageChecked(p
);
4648 unlock_page(eb
->pages
[0]);
4652 for (i
= 0; i
< num_pages
; i
++) {
4654 unlock_page(eb
->pages
[i
]);
4657 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4658 btrfs_release_extent_buffer(eb
);
4662 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4664 struct extent_buffer
*eb
=
4665 container_of(head
, struct extent_buffer
, rcu_head
);
4667 __free_extent_buffer(eb
);
4670 /* Expects to have eb->eb_lock already held */
4671 static int release_extent_buffer(struct extent_buffer
*eb
)
4673 WARN_ON(atomic_read(&eb
->refs
) == 0);
4674 if (atomic_dec_and_test(&eb
->refs
)) {
4675 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
4676 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
4678 spin_unlock(&eb
->refs_lock
);
4680 spin_lock(&fs_info
->buffer_lock
);
4681 radix_tree_delete(&fs_info
->buffer_radix
,
4682 eb
->start
>> PAGE_CACHE_SHIFT
);
4683 spin_unlock(&fs_info
->buffer_lock
);
4685 spin_unlock(&eb
->refs_lock
);
4688 /* Should be safe to release our pages at this point */
4689 btrfs_release_extent_buffer_page(eb
, 0);
4690 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4693 spin_unlock(&eb
->refs_lock
);
4698 void free_extent_buffer(struct extent_buffer
*eb
)
4706 refs
= atomic_read(&eb
->refs
);
4709 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
4714 spin_lock(&eb
->refs_lock
);
4715 if (atomic_read(&eb
->refs
) == 2 &&
4716 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4717 atomic_dec(&eb
->refs
);
4719 if (atomic_read(&eb
->refs
) == 2 &&
4720 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4721 !extent_buffer_under_io(eb
) &&
4722 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4723 atomic_dec(&eb
->refs
);
4726 * I know this is terrible, but it's temporary until we stop tracking
4727 * the uptodate bits and such for the extent buffers.
4729 release_extent_buffer(eb
);
4732 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4737 spin_lock(&eb
->refs_lock
);
4738 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4740 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4741 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4742 atomic_dec(&eb
->refs
);
4743 release_extent_buffer(eb
);
4746 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4749 unsigned long num_pages
;
4752 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4754 for (i
= 0; i
< num_pages
; i
++) {
4755 page
= extent_buffer_page(eb
, i
);
4756 if (!PageDirty(page
))
4760 WARN_ON(!PagePrivate(page
));
4762 clear_page_dirty_for_io(page
);
4763 spin_lock_irq(&page
->mapping
->tree_lock
);
4764 if (!PageDirty(page
)) {
4765 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4767 PAGECACHE_TAG_DIRTY
);
4769 spin_unlock_irq(&page
->mapping
->tree_lock
);
4770 ClearPageError(page
);
4773 WARN_ON(atomic_read(&eb
->refs
) == 0);
4776 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4779 unsigned long num_pages
;
4782 check_buffer_tree_ref(eb
);
4784 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4786 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4787 WARN_ON(atomic_read(&eb
->refs
) == 0);
4788 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4790 for (i
= 0; i
< num_pages
; i
++)
4791 set_page_dirty(extent_buffer_page(eb
, i
));
4795 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4799 unsigned long num_pages
;
4801 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4802 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4803 for (i
= 0; i
< num_pages
; i
++) {
4804 page
= extent_buffer_page(eb
, i
);
4806 ClearPageUptodate(page
);
4811 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4815 unsigned long num_pages
;
4817 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4818 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4819 for (i
= 0; i
< num_pages
; i
++) {
4820 page
= extent_buffer_page(eb
, i
);
4821 SetPageUptodate(page
);
4826 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4828 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4831 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4832 struct extent_buffer
*eb
, u64 start
, int wait
,
4833 get_extent_t
*get_extent
, int mirror_num
)
4836 unsigned long start_i
;
4840 int locked_pages
= 0;
4841 int all_uptodate
= 1;
4842 unsigned long num_pages
;
4843 unsigned long num_reads
= 0;
4844 struct bio
*bio
= NULL
;
4845 unsigned long bio_flags
= 0;
4847 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4851 WARN_ON(start
< eb
->start
);
4852 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4853 (eb
->start
>> PAGE_CACHE_SHIFT
);
4858 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4859 for (i
= start_i
; i
< num_pages
; i
++) {
4860 page
= extent_buffer_page(eb
, i
);
4861 if (wait
== WAIT_NONE
) {
4862 if (!trylock_page(page
))
4868 if (!PageUptodate(page
)) {
4875 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4879 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4880 eb
->read_mirror
= 0;
4881 atomic_set(&eb
->io_pages
, num_reads
);
4882 for (i
= start_i
; i
< num_pages
; i
++) {
4883 page
= extent_buffer_page(eb
, i
);
4884 if (!PageUptodate(page
)) {
4885 ClearPageError(page
);
4886 err
= __extent_read_full_page(tree
, page
,
4888 mirror_num
, &bio_flags
,
4898 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
4904 if (ret
|| wait
!= WAIT_COMPLETE
)
4907 for (i
= start_i
; i
< num_pages
; i
++) {
4908 page
= extent_buffer_page(eb
, i
);
4909 wait_on_page_locked(page
);
4910 if (!PageUptodate(page
))
4918 while (locked_pages
> 0) {
4919 page
= extent_buffer_page(eb
, i
);
4927 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
4928 unsigned long start
,
4935 char *dst
= (char *)dstv
;
4936 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4937 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4939 WARN_ON(start
> eb
->len
);
4940 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4942 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
4945 page
= extent_buffer_page(eb
, i
);
4947 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4948 kaddr
= page_address(page
);
4949 memcpy(dst
, kaddr
+ offset
, cur
);
4958 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
4959 unsigned long min_len
, char **map
,
4960 unsigned long *map_start
,
4961 unsigned long *map_len
)
4963 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
4966 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4967 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4968 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
4975 offset
= start_offset
;
4979 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
4982 if (start
+ min_len
> eb
->len
) {
4983 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
4985 eb
->start
, eb
->len
, start
, min_len
);
4989 p
= extent_buffer_page(eb
, i
);
4990 kaddr
= page_address(p
);
4991 *map
= kaddr
+ offset
;
4992 *map_len
= PAGE_CACHE_SIZE
- offset
;
4996 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
4997 unsigned long start
,
5004 char *ptr
= (char *)ptrv
;
5005 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5006 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5009 WARN_ON(start
> eb
->len
);
5010 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5012 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5015 page
= extent_buffer_page(eb
, i
);
5017 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5019 kaddr
= page_address(page
);
5020 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5032 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5033 unsigned long start
, unsigned long len
)
5039 char *src
= (char *)srcv
;
5040 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5041 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5043 WARN_ON(start
> eb
->len
);
5044 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5046 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5049 page
= extent_buffer_page(eb
, i
);
5050 WARN_ON(!PageUptodate(page
));
5052 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5053 kaddr
= page_address(page
);
5054 memcpy(kaddr
+ offset
, src
, cur
);
5063 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5064 unsigned long start
, unsigned long len
)
5070 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5071 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5073 WARN_ON(start
> eb
->len
);
5074 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5076 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5079 page
= extent_buffer_page(eb
, i
);
5080 WARN_ON(!PageUptodate(page
));
5082 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5083 kaddr
= page_address(page
);
5084 memset(kaddr
+ offset
, c
, cur
);
5092 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5093 unsigned long dst_offset
, unsigned long src_offset
,
5096 u64 dst_len
= dst
->len
;
5101 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5102 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5104 WARN_ON(src
->len
!= dst_len
);
5106 offset
= (start_offset
+ dst_offset
) &
5107 (PAGE_CACHE_SIZE
- 1);
5110 page
= extent_buffer_page(dst
, i
);
5111 WARN_ON(!PageUptodate(page
));
5113 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
5115 kaddr
= page_address(page
);
5116 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5125 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5127 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5128 return distance
< len
;
5131 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5132 unsigned long dst_off
, unsigned long src_off
,
5135 char *dst_kaddr
= page_address(dst_page
);
5137 int must_memmove
= 0;
5139 if (dst_page
!= src_page
) {
5140 src_kaddr
= page_address(src_page
);
5142 src_kaddr
= dst_kaddr
;
5143 if (areas_overlap(src_off
, dst_off
, len
))
5148 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5150 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5153 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5154 unsigned long src_offset
, unsigned long len
)
5157 size_t dst_off_in_page
;
5158 size_t src_off_in_page
;
5159 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5160 unsigned long dst_i
;
5161 unsigned long src_i
;
5163 if (src_offset
+ len
> dst
->len
) {
5164 printk(KERN_ERR
"BTRFS: memmove bogus src_offset %lu move "
5165 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
5168 if (dst_offset
+ len
> dst
->len
) {
5169 printk(KERN_ERR
"BTRFS: memmove bogus dst_offset %lu move "
5170 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
5175 dst_off_in_page
= (start_offset
+ dst_offset
) &
5176 (PAGE_CACHE_SIZE
- 1);
5177 src_off_in_page
= (start_offset
+ src_offset
) &
5178 (PAGE_CACHE_SIZE
- 1);
5180 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5181 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5183 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5185 cur
= min_t(unsigned long, cur
,
5186 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5188 copy_pages(extent_buffer_page(dst
, dst_i
),
5189 extent_buffer_page(dst
, src_i
),
5190 dst_off_in_page
, src_off_in_page
, cur
);
5198 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5199 unsigned long src_offset
, unsigned long len
)
5202 size_t dst_off_in_page
;
5203 size_t src_off_in_page
;
5204 unsigned long dst_end
= dst_offset
+ len
- 1;
5205 unsigned long src_end
= src_offset
+ len
- 1;
5206 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5207 unsigned long dst_i
;
5208 unsigned long src_i
;
5210 if (src_offset
+ len
> dst
->len
) {
5211 printk(KERN_ERR
"BTRFS: memmove bogus src_offset %lu move "
5212 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
5215 if (dst_offset
+ len
> dst
->len
) {
5216 printk(KERN_ERR
"BTRFS: memmove bogus dst_offset %lu move "
5217 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
5220 if (dst_offset
< src_offset
) {
5221 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5225 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5226 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5228 dst_off_in_page
= (start_offset
+ dst_end
) &
5229 (PAGE_CACHE_SIZE
- 1);
5230 src_off_in_page
= (start_offset
+ src_end
) &
5231 (PAGE_CACHE_SIZE
- 1);
5233 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5234 cur
= min(cur
, dst_off_in_page
+ 1);
5235 copy_pages(extent_buffer_page(dst
, dst_i
),
5236 extent_buffer_page(dst
, src_i
),
5237 dst_off_in_page
- cur
+ 1,
5238 src_off_in_page
- cur
+ 1, cur
);
5246 int try_release_extent_buffer(struct page
*page
)
5248 struct extent_buffer
*eb
;
5251 * We need to make sure noboody is attaching this page to an eb right
5254 spin_lock(&page
->mapping
->private_lock
);
5255 if (!PagePrivate(page
)) {
5256 spin_unlock(&page
->mapping
->private_lock
);
5260 eb
= (struct extent_buffer
*)page
->private;
5264 * This is a little awful but should be ok, we need to make sure that
5265 * the eb doesn't disappear out from under us while we're looking at
5268 spin_lock(&eb
->refs_lock
);
5269 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5270 spin_unlock(&eb
->refs_lock
);
5271 spin_unlock(&page
->mapping
->private_lock
);
5274 spin_unlock(&page
->mapping
->private_lock
);
5277 * If tree ref isn't set then we know the ref on this eb is a real ref,
5278 * so just return, this page will likely be freed soon anyway.
5280 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5281 spin_unlock(&eb
->refs_lock
);
5285 return release_extent_buffer(eb
);