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"
18 #include "btrfs_inode.h"
20 #include "check-integrity.h"
22 #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 (unsigned long long)state
->start
,
65 (unsigned long long)state
->end
,
66 state
->state
, state
->tree
, atomic_read(&state
->refs
));
67 list_del(&state
->leak_list
);
68 kmem_cache_free(extent_state_cache
, state
);
71 while (!list_empty(&buffers
)) {
72 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
73 printk(KERN_ERR
"btrfs buffer leak start %llu len %lu "
74 "refs %d\n", (unsigned long long)eb
->start
,
75 eb
->len
, atomic_read(&eb
->refs
));
76 list_del(&eb
->leak_list
);
77 kmem_cache_free(extent_buffer_cache
, eb
);
81 #define btrfs_debug_check_extent_io_range(inode, start, end) \
82 __btrfs_debug_check_extent_io_range(__func__, (inode), (start), (end))
83 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
84 struct inode
*inode
, u64 start
, u64 end
)
86 u64 isize
= i_size_read(inode
);
88 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
89 printk_ratelimited(KERN_DEBUG
90 "btrfs: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
92 (unsigned long long)btrfs_ino(inode
),
93 (unsigned long long)isize
,
94 (unsigned long long)start
,
95 (unsigned long long)end
);
99 #define btrfs_leak_debug_add(new, head) do {} while (0)
100 #define btrfs_leak_debug_del(entry) do {} while (0)
101 #define btrfs_leak_debug_check() do {} while (0)
102 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
105 #define BUFFER_LRU_MAX 64
110 struct rb_node rb_node
;
113 struct extent_page_data
{
115 struct extent_io_tree
*tree
;
116 get_extent_t
*get_extent
;
117 unsigned long bio_flags
;
119 /* tells writepage not to lock the state bits for this range
120 * it still does the unlocking
122 unsigned int extent_locked
:1;
124 /* tells the submit_bio code to use a WRITE_SYNC */
125 unsigned int sync_io
:1;
128 static noinline
void flush_write_bio(void *data
);
129 static inline struct btrfs_fs_info
*
130 tree_fs_info(struct extent_io_tree
*tree
)
132 return btrfs_sb(tree
->mapping
->host
->i_sb
);
135 int __init
extent_io_init(void)
137 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
138 sizeof(struct extent_state
), 0,
139 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
140 if (!extent_state_cache
)
143 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
144 sizeof(struct extent_buffer
), 0,
145 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
146 if (!extent_buffer_cache
)
147 goto free_state_cache
;
149 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
150 offsetof(struct btrfs_io_bio
, bio
));
152 goto free_buffer_cache
;
156 kmem_cache_destroy(extent_buffer_cache
);
157 extent_buffer_cache
= NULL
;
160 kmem_cache_destroy(extent_state_cache
);
161 extent_state_cache
= NULL
;
165 void extent_io_exit(void)
167 btrfs_leak_debug_check();
170 * Make sure all delayed rcu free are flushed before we
174 if (extent_state_cache
)
175 kmem_cache_destroy(extent_state_cache
);
176 if (extent_buffer_cache
)
177 kmem_cache_destroy(extent_buffer_cache
);
179 bioset_free(btrfs_bioset
);
182 void extent_io_tree_init(struct extent_io_tree
*tree
,
183 struct address_space
*mapping
)
185 tree
->state
= RB_ROOT
;
186 INIT_RADIX_TREE(&tree
->buffer
, GFP_ATOMIC
);
188 tree
->dirty_bytes
= 0;
189 spin_lock_init(&tree
->lock
);
190 spin_lock_init(&tree
->buffer_lock
);
191 tree
->mapping
= mapping
;
194 static struct extent_state
*alloc_extent_state(gfp_t mask
)
196 struct extent_state
*state
;
198 state
= kmem_cache_alloc(extent_state_cache
, mask
);
204 btrfs_leak_debug_add(&state
->leak_list
, &states
);
205 atomic_set(&state
->refs
, 1);
206 init_waitqueue_head(&state
->wq
);
207 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
211 void free_extent_state(struct extent_state
*state
)
215 if (atomic_dec_and_test(&state
->refs
)) {
216 WARN_ON(state
->tree
);
217 btrfs_leak_debug_del(&state
->leak_list
);
218 trace_free_extent_state(state
, _RET_IP_
);
219 kmem_cache_free(extent_state_cache
, state
);
223 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
224 struct rb_node
*node
)
226 struct rb_node
**p
= &root
->rb_node
;
227 struct rb_node
*parent
= NULL
;
228 struct tree_entry
*entry
;
232 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
234 if (offset
< entry
->start
)
236 else if (offset
> entry
->end
)
242 rb_link_node(node
, parent
, p
);
243 rb_insert_color(node
, root
);
247 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
248 struct rb_node
**prev_ret
,
249 struct rb_node
**next_ret
)
251 struct rb_root
*root
= &tree
->state
;
252 struct rb_node
*n
= root
->rb_node
;
253 struct rb_node
*prev
= NULL
;
254 struct rb_node
*orig_prev
= NULL
;
255 struct tree_entry
*entry
;
256 struct tree_entry
*prev_entry
= NULL
;
259 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
263 if (offset
< entry
->start
)
265 else if (offset
> entry
->end
)
273 while (prev
&& offset
> prev_entry
->end
) {
274 prev
= rb_next(prev
);
275 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
282 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
283 while (prev
&& offset
< prev_entry
->start
) {
284 prev
= rb_prev(prev
);
285 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
292 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
295 struct rb_node
*prev
= NULL
;
298 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
304 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
305 struct extent_state
*other
)
307 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
308 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
313 * utility function to look for merge candidates inside a given range.
314 * Any extents with matching state are merged together into a single
315 * extent in the tree. Extents with EXTENT_IO in their state field
316 * are not merged because the end_io handlers need to be able to do
317 * operations on them without sleeping (or doing allocations/splits).
319 * This should be called with the tree lock held.
321 static void merge_state(struct extent_io_tree
*tree
,
322 struct extent_state
*state
)
324 struct extent_state
*other
;
325 struct rb_node
*other_node
;
327 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
330 other_node
= rb_prev(&state
->rb_node
);
332 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
333 if (other
->end
== state
->start
- 1 &&
334 other
->state
== state
->state
) {
335 merge_cb(tree
, state
, other
);
336 state
->start
= other
->start
;
338 rb_erase(&other
->rb_node
, &tree
->state
);
339 free_extent_state(other
);
342 other_node
= rb_next(&state
->rb_node
);
344 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
345 if (other
->start
== state
->end
+ 1 &&
346 other
->state
== state
->state
) {
347 merge_cb(tree
, state
, other
);
348 state
->end
= other
->end
;
350 rb_erase(&other
->rb_node
, &tree
->state
);
351 free_extent_state(other
);
356 static void set_state_cb(struct extent_io_tree
*tree
,
357 struct extent_state
*state
, unsigned long *bits
)
359 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
360 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
363 static void clear_state_cb(struct extent_io_tree
*tree
,
364 struct extent_state
*state
, unsigned long *bits
)
366 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
367 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
370 static void set_state_bits(struct extent_io_tree
*tree
,
371 struct extent_state
*state
, unsigned long *bits
);
374 * insert an extent_state struct into the tree. 'bits' are set on the
375 * struct before it is inserted.
377 * This may return -EEXIST if the extent is already there, in which case the
378 * state struct is freed.
380 * The tree lock is not taken internally. This is a utility function and
381 * probably isn't what you want to call (see set/clear_extent_bit).
383 static int insert_state(struct extent_io_tree
*tree
,
384 struct extent_state
*state
, u64 start
, u64 end
,
387 struct rb_node
*node
;
390 WARN(1, KERN_ERR
"btrfs end < start %llu %llu\n",
391 (unsigned long long)end
,
392 (unsigned long long)start
);
393 state
->start
= start
;
396 set_state_bits(tree
, state
, bits
);
398 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
400 struct extent_state
*found
;
401 found
= rb_entry(node
, struct extent_state
, rb_node
);
402 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
403 "%llu %llu\n", (unsigned long long)found
->start
,
404 (unsigned long long)found
->end
,
405 (unsigned long long)start
, (unsigned long long)end
);
409 merge_state(tree
, state
);
413 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
416 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
417 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
421 * split a given extent state struct in two, inserting the preallocated
422 * struct 'prealloc' as the newly created second half. 'split' indicates an
423 * offset inside 'orig' where it should be split.
426 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
427 * are two extent state structs in the tree:
428 * prealloc: [orig->start, split - 1]
429 * orig: [ split, orig->end ]
431 * The tree locks are not taken by this function. They need to be held
434 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
435 struct extent_state
*prealloc
, u64 split
)
437 struct rb_node
*node
;
439 split_cb(tree
, orig
, split
);
441 prealloc
->start
= orig
->start
;
442 prealloc
->end
= split
- 1;
443 prealloc
->state
= orig
->state
;
446 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
448 free_extent_state(prealloc
);
451 prealloc
->tree
= tree
;
455 static struct extent_state
*next_state(struct extent_state
*state
)
457 struct rb_node
*next
= rb_next(&state
->rb_node
);
459 return rb_entry(next
, struct extent_state
, rb_node
);
465 * utility function to clear some bits in an extent state struct.
466 * it will optionally wake up any one waiting on this state (wake == 1).
468 * If no bits are set on the state struct after clearing things, the
469 * struct is freed and removed from the tree
471 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
472 struct extent_state
*state
,
473 unsigned long *bits
, int wake
)
475 struct extent_state
*next
;
476 unsigned long bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
478 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
479 u64 range
= state
->end
- state
->start
+ 1;
480 WARN_ON(range
> tree
->dirty_bytes
);
481 tree
->dirty_bytes
-= range
;
483 clear_state_cb(tree
, state
, bits
);
484 state
->state
&= ~bits_to_clear
;
487 if (state
->state
== 0) {
488 next
= next_state(state
);
490 rb_erase(&state
->rb_node
, &tree
->state
);
492 free_extent_state(state
);
497 merge_state(tree
, state
);
498 next
= next_state(state
);
503 static struct extent_state
*
504 alloc_extent_state_atomic(struct extent_state
*prealloc
)
507 prealloc
= alloc_extent_state(GFP_ATOMIC
);
512 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
514 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
515 "Extent tree was modified by another "
516 "thread while locked.");
520 * clear some bits on a range in the tree. This may require splitting
521 * or inserting elements in the tree, so the gfp mask is used to
522 * indicate which allocations or sleeping are allowed.
524 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
525 * the given range from the tree regardless of state (ie for truncate).
527 * the range [start, end] is inclusive.
529 * This takes the tree lock, and returns 0 on success and < 0 on error.
531 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
532 unsigned long bits
, int wake
, int delete,
533 struct extent_state
**cached_state
,
536 struct extent_state
*state
;
537 struct extent_state
*cached
;
538 struct extent_state
*prealloc
= NULL
;
539 struct rb_node
*node
;
544 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
546 if (bits
& EXTENT_DELALLOC
)
547 bits
|= EXTENT_NORESERVE
;
550 bits
|= ~EXTENT_CTLBITS
;
551 bits
|= EXTENT_FIRST_DELALLOC
;
553 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
556 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
557 prealloc
= alloc_extent_state(mask
);
562 spin_lock(&tree
->lock
);
564 cached
= *cached_state
;
567 *cached_state
= NULL
;
571 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
572 cached
->end
> start
) {
574 atomic_dec(&cached
->refs
);
579 free_extent_state(cached
);
582 * this search will find the extents that end after
585 node
= tree_search(tree
, start
);
588 state
= rb_entry(node
, struct extent_state
, rb_node
);
590 if (state
->start
> end
)
592 WARN_ON(state
->end
< start
);
593 last_end
= state
->end
;
595 /* the state doesn't have the wanted bits, go ahead */
596 if (!(state
->state
& bits
)) {
597 state
= next_state(state
);
602 * | ---- desired range ---- |
604 * | ------------- state -------------- |
606 * We need to split the extent we found, and may flip
607 * bits on second half.
609 * If the extent we found extends past our range, we
610 * just split and search again. It'll get split again
611 * the next time though.
613 * If the extent we found is inside our range, we clear
614 * the desired bit on it.
617 if (state
->start
< start
) {
618 prealloc
= alloc_extent_state_atomic(prealloc
);
620 err
= split_state(tree
, state
, prealloc
, start
);
622 extent_io_tree_panic(tree
, err
);
627 if (state
->end
<= end
) {
628 state
= clear_state_bit(tree
, state
, &bits
, wake
);
634 * | ---- desired range ---- |
636 * We need to split the extent, and clear the bit
639 if (state
->start
<= end
&& state
->end
> end
) {
640 prealloc
= alloc_extent_state_atomic(prealloc
);
642 err
= split_state(tree
, state
, prealloc
, end
+ 1);
644 extent_io_tree_panic(tree
, err
);
649 clear_state_bit(tree
, prealloc
, &bits
, wake
);
655 state
= clear_state_bit(tree
, state
, &bits
, wake
);
657 if (last_end
== (u64
)-1)
659 start
= last_end
+ 1;
660 if (start
<= end
&& state
&& !need_resched())
665 spin_unlock(&tree
->lock
);
667 free_extent_state(prealloc
);
674 spin_unlock(&tree
->lock
);
675 if (mask
& __GFP_WAIT
)
680 static void wait_on_state(struct extent_io_tree
*tree
,
681 struct extent_state
*state
)
682 __releases(tree
->lock
)
683 __acquires(tree
->lock
)
686 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
687 spin_unlock(&tree
->lock
);
689 spin_lock(&tree
->lock
);
690 finish_wait(&state
->wq
, &wait
);
694 * waits for one or more bits to clear on a range in the state tree.
695 * The range [start, end] is inclusive.
696 * The tree lock is taken by this function
698 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
701 struct extent_state
*state
;
702 struct rb_node
*node
;
704 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
706 spin_lock(&tree
->lock
);
710 * this search will find all the extents that end after
713 node
= tree_search(tree
, start
);
717 state
= rb_entry(node
, struct extent_state
, rb_node
);
719 if (state
->start
> end
)
722 if (state
->state
& bits
) {
723 start
= state
->start
;
724 atomic_inc(&state
->refs
);
725 wait_on_state(tree
, state
);
726 free_extent_state(state
);
729 start
= state
->end
+ 1;
734 cond_resched_lock(&tree
->lock
);
737 spin_unlock(&tree
->lock
);
740 static void set_state_bits(struct extent_io_tree
*tree
,
741 struct extent_state
*state
,
744 unsigned long bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
746 set_state_cb(tree
, state
, bits
);
747 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
748 u64 range
= state
->end
- state
->start
+ 1;
749 tree
->dirty_bytes
+= range
;
751 state
->state
|= bits_to_set
;
754 static void cache_state(struct extent_state
*state
,
755 struct extent_state
**cached_ptr
)
757 if (cached_ptr
&& !(*cached_ptr
)) {
758 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
760 atomic_inc(&state
->refs
);
766 * set some bits on a range in the tree. This may require allocations or
767 * sleeping, so the gfp mask is used to indicate what is allowed.
769 * If any of the exclusive bits are set, this will fail with -EEXIST if some
770 * part of the range already has the desired bits set. The start of the
771 * existing range is returned in failed_start in this case.
773 * [start, end] is inclusive This takes the tree lock.
776 static int __must_check
777 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
778 unsigned long bits
, unsigned long exclusive_bits
,
779 u64
*failed_start
, struct extent_state
**cached_state
,
782 struct extent_state
*state
;
783 struct extent_state
*prealloc
= NULL
;
784 struct rb_node
*node
;
789 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
791 bits
|= EXTENT_FIRST_DELALLOC
;
793 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
794 prealloc
= alloc_extent_state(mask
);
798 spin_lock(&tree
->lock
);
799 if (cached_state
&& *cached_state
) {
800 state
= *cached_state
;
801 if (state
->start
<= start
&& state
->end
> start
&&
803 node
= &state
->rb_node
;
808 * this search will find all the extents that end after
811 node
= tree_search(tree
, start
);
813 prealloc
= alloc_extent_state_atomic(prealloc
);
815 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
817 extent_io_tree_panic(tree
, err
);
822 state
= rb_entry(node
, struct extent_state
, rb_node
);
824 last_start
= state
->start
;
825 last_end
= state
->end
;
828 * | ---- desired range ---- |
831 * Just lock what we found and keep going
833 if (state
->start
== start
&& state
->end
<= end
) {
834 if (state
->state
& exclusive_bits
) {
835 *failed_start
= state
->start
;
840 set_state_bits(tree
, state
, &bits
);
841 cache_state(state
, cached_state
);
842 merge_state(tree
, state
);
843 if (last_end
== (u64
)-1)
845 start
= last_end
+ 1;
846 state
= next_state(state
);
847 if (start
< end
&& state
&& state
->start
== start
&&
854 * | ---- desired range ---- |
857 * | ------------- state -------------- |
859 * We need to split the extent we found, and may flip bits on
862 * If the extent we found extends past our
863 * range, we just split and search again. It'll get split
864 * again the next time though.
866 * If the extent we found is inside our range, we set the
869 if (state
->start
< start
) {
870 if (state
->state
& exclusive_bits
) {
871 *failed_start
= start
;
876 prealloc
= alloc_extent_state_atomic(prealloc
);
878 err
= split_state(tree
, state
, prealloc
, start
);
880 extent_io_tree_panic(tree
, err
);
885 if (state
->end
<= end
) {
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 ---- |
901 * | state | or | state |
903 * There's a hole, we need to insert something in it and
904 * ignore the extent we found.
906 if (state
->start
> start
) {
908 if (end
< last_start
)
911 this_end
= last_start
- 1;
913 prealloc
= alloc_extent_state_atomic(prealloc
);
917 * Avoid to free 'prealloc' if it can be merged with
920 err
= insert_state(tree
, prealloc
, start
, this_end
,
923 extent_io_tree_panic(tree
, err
);
925 cache_state(prealloc
, cached_state
);
927 start
= this_end
+ 1;
931 * | ---- desired range ---- |
933 * We need to split the extent, and set the bit
936 if (state
->start
<= end
&& state
->end
> end
) {
937 if (state
->state
& exclusive_bits
) {
938 *failed_start
= start
;
943 prealloc
= alloc_extent_state_atomic(prealloc
);
945 err
= split_state(tree
, state
, prealloc
, end
+ 1);
947 extent_io_tree_panic(tree
, err
);
949 set_state_bits(tree
, prealloc
, &bits
);
950 cache_state(prealloc
, cached_state
);
951 merge_state(tree
, prealloc
);
959 spin_unlock(&tree
->lock
);
961 free_extent_state(prealloc
);
968 spin_unlock(&tree
->lock
);
969 if (mask
& __GFP_WAIT
)
974 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
975 unsigned long bits
, u64
* failed_start
,
976 struct extent_state
**cached_state
, gfp_t mask
)
978 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
984 * convert_extent_bit - convert all bits in a given range from one bit to
986 * @tree: the io tree to search
987 * @start: the start offset in bytes
988 * @end: the end offset in bytes (inclusive)
989 * @bits: the bits to set in this range
990 * @clear_bits: the bits to clear in this range
991 * @cached_state: state that we're going to cache
992 * @mask: the allocation mask
994 * This will go through and set bits for the given range. If any states exist
995 * already in this range they are set with the given bit and cleared of the
996 * clear_bits. This is only meant to be used by things that are mergeable, ie
997 * converting from say DELALLOC to DIRTY. This is not meant to be used with
998 * boundary bits like LOCK.
1000 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1001 unsigned long bits
, unsigned long clear_bits
,
1002 struct extent_state
**cached_state
, gfp_t mask
)
1004 struct extent_state
*state
;
1005 struct extent_state
*prealloc
= NULL
;
1006 struct rb_node
*node
;
1011 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
1014 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
1015 prealloc
= alloc_extent_state(mask
);
1020 spin_lock(&tree
->lock
);
1021 if (cached_state
&& *cached_state
) {
1022 state
= *cached_state
;
1023 if (state
->start
<= start
&& state
->end
> start
&&
1025 node
= &state
->rb_node
;
1031 * this search will find all the extents that end after
1034 node
= tree_search(tree
, start
);
1036 prealloc
= alloc_extent_state_atomic(prealloc
);
1041 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
1044 extent_io_tree_panic(tree
, err
);
1047 state
= rb_entry(node
, struct extent_state
, rb_node
);
1049 last_start
= state
->start
;
1050 last_end
= state
->end
;
1053 * | ---- desired range ---- |
1056 * Just lock what we found and keep going
1058 if (state
->start
== start
&& state
->end
<= end
) {
1059 set_state_bits(tree
, state
, &bits
);
1060 cache_state(state
, cached_state
);
1061 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1062 if (last_end
== (u64
)-1)
1064 start
= last_end
+ 1;
1065 if (start
< end
&& state
&& state
->start
== start
&&
1072 * | ---- desired range ---- |
1075 * | ------------- state -------------- |
1077 * We need to split the extent we found, and may flip bits on
1080 * If the extent we found extends past our
1081 * range, we just split and search again. It'll get split
1082 * again the next time though.
1084 * If the extent we found is inside our range, we set the
1085 * desired bit on it.
1087 if (state
->start
< start
) {
1088 prealloc
= alloc_extent_state_atomic(prealloc
);
1093 err
= split_state(tree
, state
, prealloc
, start
);
1095 extent_io_tree_panic(tree
, err
);
1099 if (state
->end
<= end
) {
1100 set_state_bits(tree
, state
, &bits
);
1101 cache_state(state
, cached_state
);
1102 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1103 if (last_end
== (u64
)-1)
1105 start
= last_end
+ 1;
1106 if (start
< end
&& state
&& state
->start
== start
&&
1113 * | ---- desired range ---- |
1114 * | state | or | state |
1116 * There's a hole, we need to insert something in it and
1117 * ignore the extent we found.
1119 if (state
->start
> start
) {
1121 if (end
< last_start
)
1124 this_end
= last_start
- 1;
1126 prealloc
= alloc_extent_state_atomic(prealloc
);
1133 * Avoid to free 'prealloc' if it can be merged with
1136 err
= insert_state(tree
, prealloc
, start
, this_end
,
1139 extent_io_tree_panic(tree
, err
);
1140 cache_state(prealloc
, cached_state
);
1142 start
= this_end
+ 1;
1146 * | ---- desired range ---- |
1148 * We need to split the extent, and set the bit
1151 if (state
->start
<= end
&& state
->end
> end
) {
1152 prealloc
= alloc_extent_state_atomic(prealloc
);
1158 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1160 extent_io_tree_panic(tree
, err
);
1162 set_state_bits(tree
, prealloc
, &bits
);
1163 cache_state(prealloc
, cached_state
);
1164 clear_state_bit(tree
, prealloc
, &clear_bits
, 0);
1172 spin_unlock(&tree
->lock
);
1174 free_extent_state(prealloc
);
1181 spin_unlock(&tree
->lock
);
1182 if (mask
& __GFP_WAIT
)
1187 /* wrappers around set/clear extent bit */
1188 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1191 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, NULL
,
1195 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1196 unsigned long bits
, gfp_t mask
)
1198 return set_extent_bit(tree
, start
, end
, bits
, NULL
,
1202 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1203 unsigned long bits
, gfp_t mask
)
1205 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
1208 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1209 struct extent_state
**cached_state
, gfp_t mask
)
1211 return set_extent_bit(tree
, start
, end
,
1212 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1213 NULL
, cached_state
, mask
);
1216 int set_extent_defrag(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1217 struct extent_state
**cached_state
, gfp_t mask
)
1219 return set_extent_bit(tree
, start
, end
,
1220 EXTENT_DELALLOC
| EXTENT_UPTODATE
| EXTENT_DEFRAG
,
1221 NULL
, cached_state
, mask
);
1224 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1227 return clear_extent_bit(tree
, start
, end
,
1228 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1229 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1232 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1235 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, NULL
,
1239 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1240 struct extent_state
**cached_state
, gfp_t mask
)
1242 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, NULL
,
1243 cached_state
, mask
);
1246 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1247 struct extent_state
**cached_state
, gfp_t mask
)
1249 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1250 cached_state
, mask
);
1254 * either insert or lock state struct between start and end use mask to tell
1255 * us if waiting is desired.
1257 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1258 unsigned long bits
, struct extent_state
**cached_state
)
1263 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1264 EXTENT_LOCKED
, &failed_start
,
1265 cached_state
, GFP_NOFS
);
1266 if (err
== -EEXIST
) {
1267 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1268 start
= failed_start
;
1271 WARN_ON(start
> end
);
1276 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1278 return lock_extent_bits(tree
, start
, end
, 0, NULL
);
1281 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1286 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1287 &failed_start
, NULL
, GFP_NOFS
);
1288 if (err
== -EEXIST
) {
1289 if (failed_start
> start
)
1290 clear_extent_bit(tree
, start
, failed_start
- 1,
1291 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1297 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1298 struct extent_state
**cached
, gfp_t mask
)
1300 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1304 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1306 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1310 int extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1312 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1313 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1316 while (index
<= end_index
) {
1317 page
= find_get_page(inode
->i_mapping
, index
);
1318 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1319 clear_page_dirty_for_io(page
);
1320 page_cache_release(page
);
1326 int extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1328 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1329 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1332 while (index
<= end_index
) {
1333 page
= find_get_page(inode
->i_mapping
, index
);
1334 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1335 account_page_redirty(page
);
1336 __set_page_dirty_nobuffers(page
);
1337 page_cache_release(page
);
1344 * helper function to set both pages and extents in the tree writeback
1346 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1348 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1349 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1352 while (index
<= end_index
) {
1353 page
= find_get_page(tree
->mapping
, index
);
1354 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1355 set_page_writeback(page
);
1356 page_cache_release(page
);
1362 /* find the first state struct with 'bits' set after 'start', and
1363 * return it. tree->lock must be held. NULL will returned if
1364 * nothing was found after 'start'
1366 static struct extent_state
*
1367 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1368 u64 start
, unsigned long bits
)
1370 struct rb_node
*node
;
1371 struct extent_state
*state
;
1374 * this search will find all the extents that end after
1377 node
= tree_search(tree
, start
);
1382 state
= rb_entry(node
, struct extent_state
, rb_node
);
1383 if (state
->end
>= start
&& (state
->state
& bits
))
1386 node
= rb_next(node
);
1395 * find the first offset in the io tree with 'bits' set. zero is
1396 * returned if we find something, and *start_ret and *end_ret are
1397 * set to reflect the state struct that was found.
1399 * If nothing was found, 1 is returned. If found something, return 0.
1401 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1402 u64
*start_ret
, u64
*end_ret
, unsigned long bits
,
1403 struct extent_state
**cached_state
)
1405 struct extent_state
*state
;
1409 spin_lock(&tree
->lock
);
1410 if (cached_state
&& *cached_state
) {
1411 state
= *cached_state
;
1412 if (state
->end
== start
- 1 && state
->tree
) {
1413 n
= rb_next(&state
->rb_node
);
1415 state
= rb_entry(n
, struct extent_state
,
1417 if (state
->state
& bits
)
1421 free_extent_state(*cached_state
);
1422 *cached_state
= NULL
;
1425 free_extent_state(*cached_state
);
1426 *cached_state
= NULL
;
1429 state
= find_first_extent_bit_state(tree
, start
, bits
);
1432 cache_state(state
, cached_state
);
1433 *start_ret
= state
->start
;
1434 *end_ret
= state
->end
;
1438 spin_unlock(&tree
->lock
);
1443 * find a contiguous range of bytes in the file marked as delalloc, not
1444 * more than 'max_bytes'. start and end are used to return the range,
1446 * 1 is returned if we find something, 0 if nothing was in the tree
1448 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1449 u64
*start
, u64
*end
, u64 max_bytes
,
1450 struct extent_state
**cached_state
)
1452 struct rb_node
*node
;
1453 struct extent_state
*state
;
1454 u64 cur_start
= *start
;
1456 u64 total_bytes
= 0;
1458 spin_lock(&tree
->lock
);
1461 * this search will find all the extents that end after
1464 node
= tree_search(tree
, cur_start
);
1472 state
= rb_entry(node
, struct extent_state
, rb_node
);
1473 if (found
&& (state
->start
!= cur_start
||
1474 (state
->state
& EXTENT_BOUNDARY
))) {
1477 if (!(state
->state
& EXTENT_DELALLOC
)) {
1483 *start
= state
->start
;
1484 *cached_state
= state
;
1485 atomic_inc(&state
->refs
);
1489 cur_start
= state
->end
+ 1;
1490 node
= rb_next(node
);
1493 total_bytes
+= state
->end
- state
->start
+ 1;
1494 if (total_bytes
>= max_bytes
)
1498 spin_unlock(&tree
->lock
);
1502 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1503 struct page
*locked_page
,
1507 struct page
*pages
[16];
1508 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1509 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1510 unsigned long nr_pages
= end_index
- index
+ 1;
1513 if (index
== locked_page
->index
&& end_index
== index
)
1516 while (nr_pages
> 0) {
1517 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1518 min_t(unsigned long, nr_pages
,
1519 ARRAY_SIZE(pages
)), pages
);
1520 for (i
= 0; i
< ret
; i
++) {
1521 if (pages
[i
] != locked_page
)
1522 unlock_page(pages
[i
]);
1523 page_cache_release(pages
[i
]);
1531 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1532 struct page
*locked_page
,
1536 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1537 unsigned long start_index
= index
;
1538 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1539 unsigned long pages_locked
= 0;
1540 struct page
*pages
[16];
1541 unsigned long nrpages
;
1545 /* the caller is responsible for locking the start index */
1546 if (index
== locked_page
->index
&& index
== end_index
)
1549 /* skip the page at the start index */
1550 nrpages
= end_index
- index
+ 1;
1551 while (nrpages
> 0) {
1552 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1553 min_t(unsigned long,
1554 nrpages
, ARRAY_SIZE(pages
)), pages
);
1559 /* now we have an array of pages, lock them all */
1560 for (i
= 0; i
< ret
; i
++) {
1562 * the caller is taking responsibility for
1565 if (pages
[i
] != locked_page
) {
1566 lock_page(pages
[i
]);
1567 if (!PageDirty(pages
[i
]) ||
1568 pages
[i
]->mapping
!= inode
->i_mapping
) {
1570 unlock_page(pages
[i
]);
1571 page_cache_release(pages
[i
]);
1575 page_cache_release(pages
[i
]);
1584 if (ret
&& pages_locked
) {
1585 __unlock_for_delalloc(inode
, locked_page
,
1587 ((u64
)(start_index
+ pages_locked
- 1)) <<
1594 * find a contiguous range of bytes in the file marked as delalloc, not
1595 * more than 'max_bytes'. start and end are used to return the range,
1597 * 1 is returned if we find something, 0 if nothing was in the tree
1599 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1600 struct extent_io_tree
*tree
,
1601 struct page
*locked_page
,
1602 u64
*start
, u64
*end
,
1608 struct extent_state
*cached_state
= NULL
;
1613 /* step one, find a bunch of delalloc bytes starting at start */
1614 delalloc_start
= *start
;
1616 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1617 max_bytes
, &cached_state
);
1618 if (!found
|| delalloc_end
<= *start
) {
1619 *start
= delalloc_start
;
1620 *end
= delalloc_end
;
1621 free_extent_state(cached_state
);
1626 * start comes from the offset of locked_page. We have to lock
1627 * pages in order, so we can't process delalloc bytes before
1630 if (delalloc_start
< *start
)
1631 delalloc_start
= *start
;
1634 * make sure to limit the number of pages we try to lock down
1637 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
)
1638 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1640 /* step two, lock all the pages after the page that has start */
1641 ret
= lock_delalloc_pages(inode
, locked_page
,
1642 delalloc_start
, delalloc_end
);
1643 if (ret
== -EAGAIN
) {
1644 /* some of the pages are gone, lets avoid looping by
1645 * shortening the size of the delalloc range we're searching
1647 free_extent_state(cached_state
);
1649 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1650 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1658 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1660 /* step three, lock the state bits for the whole range */
1661 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1663 /* then test to make sure it is all still delalloc */
1664 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1665 EXTENT_DELALLOC
, 1, cached_state
);
1667 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1668 &cached_state
, GFP_NOFS
);
1669 __unlock_for_delalloc(inode
, locked_page
,
1670 delalloc_start
, delalloc_end
);
1674 free_extent_state(cached_state
);
1675 *start
= delalloc_start
;
1676 *end
= delalloc_end
;
1681 int extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1682 struct page
*locked_page
,
1683 unsigned long clear_bits
,
1684 unsigned long page_ops
)
1686 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1688 struct page
*pages
[16];
1689 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1690 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1691 unsigned long nr_pages
= end_index
- index
+ 1;
1694 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1698 while (nr_pages
> 0) {
1699 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1700 min_t(unsigned long,
1701 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1702 for (i
= 0; i
< ret
; i
++) {
1704 if (page_ops
& PAGE_SET_PRIVATE2
)
1705 SetPagePrivate2(pages
[i
]);
1707 if (pages
[i
] == locked_page
) {
1708 page_cache_release(pages
[i
]);
1711 if (page_ops
& PAGE_CLEAR_DIRTY
)
1712 clear_page_dirty_for_io(pages
[i
]);
1713 if (page_ops
& PAGE_SET_WRITEBACK
)
1714 set_page_writeback(pages
[i
]);
1715 if (page_ops
& PAGE_END_WRITEBACK
)
1716 end_page_writeback(pages
[i
]);
1717 if (page_ops
& PAGE_UNLOCK
)
1718 unlock_page(pages
[i
]);
1719 page_cache_release(pages
[i
]);
1729 * count the number of bytes in the tree that have a given bit(s)
1730 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1731 * cached. The total number found is returned.
1733 u64
count_range_bits(struct extent_io_tree
*tree
,
1734 u64
*start
, u64 search_end
, u64 max_bytes
,
1735 unsigned long bits
, int contig
)
1737 struct rb_node
*node
;
1738 struct extent_state
*state
;
1739 u64 cur_start
= *start
;
1740 u64 total_bytes
= 0;
1744 if (search_end
<= cur_start
) {
1749 spin_lock(&tree
->lock
);
1750 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1751 total_bytes
= tree
->dirty_bytes
;
1755 * this search will find all the extents that end after
1758 node
= tree_search(tree
, cur_start
);
1763 state
= rb_entry(node
, struct extent_state
, rb_node
);
1764 if (state
->start
> search_end
)
1766 if (contig
&& found
&& state
->start
> last
+ 1)
1768 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1769 total_bytes
+= min(search_end
, state
->end
) + 1 -
1770 max(cur_start
, state
->start
);
1771 if (total_bytes
>= max_bytes
)
1774 *start
= max(cur_start
, state
->start
);
1778 } else if (contig
&& found
) {
1781 node
= rb_next(node
);
1786 spin_unlock(&tree
->lock
);
1791 * set the private field for a given byte offset in the tree. If there isn't
1792 * an extent_state there already, this does nothing.
1794 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1796 struct rb_node
*node
;
1797 struct extent_state
*state
;
1800 spin_lock(&tree
->lock
);
1802 * this search will find all the extents that end after
1805 node
= tree_search(tree
, start
);
1810 state
= rb_entry(node
, struct extent_state
, rb_node
);
1811 if (state
->start
!= start
) {
1815 state
->private = private;
1817 spin_unlock(&tree
->lock
);
1821 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1823 struct rb_node
*node
;
1824 struct extent_state
*state
;
1827 spin_lock(&tree
->lock
);
1829 * this search will find all the extents that end after
1832 node
= tree_search(tree
, start
);
1837 state
= rb_entry(node
, struct extent_state
, rb_node
);
1838 if (state
->start
!= start
) {
1842 *private = state
->private;
1844 spin_unlock(&tree
->lock
);
1849 * searches a range in the state tree for a given mask.
1850 * If 'filled' == 1, this returns 1 only if every extent in the tree
1851 * has the bits set. Otherwise, 1 is returned if any bit in the
1852 * range is found set.
1854 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1855 unsigned long bits
, int filled
, struct extent_state
*cached
)
1857 struct extent_state
*state
= NULL
;
1858 struct rb_node
*node
;
1861 spin_lock(&tree
->lock
);
1862 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1863 cached
->end
> start
)
1864 node
= &cached
->rb_node
;
1866 node
= tree_search(tree
, start
);
1867 while (node
&& start
<= end
) {
1868 state
= rb_entry(node
, struct extent_state
, rb_node
);
1870 if (filled
&& state
->start
> start
) {
1875 if (state
->start
> end
)
1878 if (state
->state
& bits
) {
1882 } else if (filled
) {
1887 if (state
->end
== (u64
)-1)
1890 start
= state
->end
+ 1;
1893 node
= rb_next(node
);
1900 spin_unlock(&tree
->lock
);
1905 * helper function to set a given page up to date if all the
1906 * extents in the tree for that page are up to date
1908 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1910 u64 start
= page_offset(page
);
1911 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1912 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1913 SetPageUptodate(page
);
1917 * When IO fails, either with EIO or csum verification fails, we
1918 * try other mirrors that might have a good copy of the data. This
1919 * io_failure_record is used to record state as we go through all the
1920 * mirrors. If another mirror has good data, the page is set up to date
1921 * and things continue. If a good mirror can't be found, the original
1922 * bio end_io callback is called to indicate things have failed.
1924 struct io_failure_record
{
1929 unsigned long bio_flags
;
1935 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
1940 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1942 set_state_private(failure_tree
, rec
->start
, 0);
1943 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1944 rec
->start
+ rec
->len
- 1,
1945 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1949 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1950 rec
->start
+ rec
->len
- 1,
1951 EXTENT_DAMAGED
, GFP_NOFS
);
1959 static void repair_io_failure_callback(struct bio
*bio
, int err
)
1961 complete(bio
->bi_private
);
1965 * this bypasses the standard btrfs submit functions deliberately, as
1966 * the standard behavior is to write all copies in a raid setup. here we only
1967 * want to write the one bad copy. so we do the mapping for ourselves and issue
1968 * submit_bio directly.
1969 * to avoid any synchronization issues, wait for the data after writing, which
1970 * actually prevents the read that triggered the error from finishing.
1971 * currently, there can be no more than two copies of every data bit. thus,
1972 * exactly one rewrite is required.
1974 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 start
,
1975 u64 length
, u64 logical
, struct page
*page
,
1979 struct btrfs_device
*dev
;
1980 DECLARE_COMPLETION_ONSTACK(compl);
1983 struct btrfs_bio
*bbio
= NULL
;
1984 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
1987 BUG_ON(!mirror_num
);
1989 /* we can't repair anything in raid56 yet */
1990 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
1993 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
1996 bio
->bi_private
= &compl;
1997 bio
->bi_end_io
= repair_io_failure_callback
;
1999 map_length
= length
;
2001 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2002 &map_length
, &bbio
, mirror_num
);
2007 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2008 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2009 bio
->bi_sector
= sector
;
2010 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2012 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2016 bio
->bi_bdev
= dev
->bdev
;
2017 bio_add_page(bio
, page
, length
, start
- page_offset(page
));
2018 btrfsic_submit_bio(WRITE_SYNC
, bio
);
2019 wait_for_completion(&compl);
2021 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
2022 /* try to remap that extent elsewhere? */
2024 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2028 printk_ratelimited_in_rcu(KERN_INFO
"btrfs read error corrected: ino %lu off %llu "
2029 "(dev %s sector %llu)\n", page
->mapping
->host
->i_ino
,
2030 start
, rcu_str_deref(dev
->name
), sector
);
2036 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2039 u64 start
= eb
->start
;
2040 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2043 for (i
= 0; i
< num_pages
; i
++) {
2044 struct page
*p
= extent_buffer_page(eb
, i
);
2045 ret
= repair_io_failure(root
->fs_info
, start
, PAGE_CACHE_SIZE
,
2046 start
, p
, mirror_num
);
2049 start
+= PAGE_CACHE_SIZE
;
2056 * each time an IO finishes, we do a fast check in the IO failure tree
2057 * to see if we need to process or clean up an io_failure_record
2059 static int clean_io_failure(u64 start
, struct page
*page
)
2062 u64 private_failure
;
2063 struct io_failure_record
*failrec
;
2064 struct btrfs_fs_info
*fs_info
;
2065 struct extent_state
*state
;
2069 struct inode
*inode
= page
->mapping
->host
;
2072 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2073 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2077 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2082 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2083 BUG_ON(!failrec
->this_mirror
);
2085 if (failrec
->in_validation
) {
2086 /* there was no real error, just free the record */
2087 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2093 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2094 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2097 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2099 if (state
&& state
->start
<= failrec
->start
&&
2100 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2101 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2102 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2104 if (num_copies
> 1) {
2105 ret
= repair_io_failure(fs_info
, start
, failrec
->len
,
2106 failrec
->logical
, page
,
2107 failrec
->failed_mirror
);
2115 ret
= free_io_failure(inode
, failrec
, did_repair
);
2121 * this is a generic handler for readpage errors (default
2122 * readpage_io_failed_hook). if other copies exist, read those and write back
2123 * good data to the failed position. does not investigate in remapping the
2124 * failed extent elsewhere, hoping the device will be smart enough to do this as
2128 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2129 struct page
*page
, u64 start
, u64 end
,
2132 struct io_failure_record
*failrec
= NULL
;
2134 struct extent_map
*em
;
2135 struct inode
*inode
= page
->mapping
->host
;
2136 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2137 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2138 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2140 struct btrfs_io_bio
*btrfs_failed_bio
;
2141 struct btrfs_io_bio
*btrfs_bio
;
2147 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2149 ret
= get_state_private(failure_tree
, start
, &private);
2151 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2154 failrec
->start
= start
;
2155 failrec
->len
= end
- start
+ 1;
2156 failrec
->this_mirror
= 0;
2157 failrec
->bio_flags
= 0;
2158 failrec
->in_validation
= 0;
2160 read_lock(&em_tree
->lock
);
2161 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2163 read_unlock(&em_tree
->lock
);
2168 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
2169 free_extent_map(em
);
2172 read_unlock(&em_tree
->lock
);
2178 logical
= start
- em
->start
;
2179 logical
= em
->block_start
+ logical
;
2180 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2181 logical
= em
->block_start
;
2182 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2183 extent_set_compress_type(&failrec
->bio_flags
,
2186 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2187 "len=%llu\n", logical
, start
, failrec
->len
);
2188 failrec
->logical
= logical
;
2189 free_extent_map(em
);
2191 /* set the bits in the private failure tree */
2192 ret
= set_extent_bits(failure_tree
, start
, end
,
2193 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2195 ret
= set_state_private(failure_tree
, start
,
2196 (u64
)(unsigned long)failrec
);
2197 /* set the bits in the inode's tree */
2199 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2206 failrec
= (struct io_failure_record
*)(unsigned long)private;
2207 pr_debug("bio_readpage_error: (found) logical=%llu, "
2208 "start=%llu, len=%llu, validation=%d\n",
2209 failrec
->logical
, failrec
->start
, failrec
->len
,
2210 failrec
->in_validation
);
2212 * when data can be on disk more than twice, add to failrec here
2213 * (e.g. with a list for failed_mirror) to make
2214 * clean_io_failure() clean all those errors at once.
2217 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2218 failrec
->logical
, failrec
->len
);
2219 if (num_copies
== 1) {
2221 * we only have a single copy of the data, so don't bother with
2222 * all the retry and error correction code that follows. no
2223 * matter what the error is, it is very likely to persist.
2225 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2226 num_copies
, failrec
->this_mirror
, failed_mirror
);
2227 free_io_failure(inode
, failrec
, 0);
2232 * there are two premises:
2233 * a) deliver good data to the caller
2234 * b) correct the bad sectors on disk
2236 if (failed_bio
->bi_vcnt
> 1) {
2238 * to fulfill b), we need to know the exact failing sectors, as
2239 * we don't want to rewrite any more than the failed ones. thus,
2240 * we need separate read requests for the failed bio
2242 * if the following BUG_ON triggers, our validation request got
2243 * merged. we need separate requests for our algorithm to work.
2245 BUG_ON(failrec
->in_validation
);
2246 failrec
->in_validation
= 1;
2247 failrec
->this_mirror
= failed_mirror
;
2248 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2251 * we're ready to fulfill a) and b) alongside. get a good copy
2252 * of the failed sector and if we succeed, we have setup
2253 * everything for repair_io_failure to do the rest for us.
2255 if (failrec
->in_validation
) {
2256 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2257 failrec
->in_validation
= 0;
2258 failrec
->this_mirror
= 0;
2260 failrec
->failed_mirror
= failed_mirror
;
2261 failrec
->this_mirror
++;
2262 if (failrec
->this_mirror
== failed_mirror
)
2263 failrec
->this_mirror
++;
2264 read_mode
= READ_SYNC
;
2267 if (failrec
->this_mirror
> num_copies
) {
2268 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2269 num_copies
, failrec
->this_mirror
, failed_mirror
);
2270 free_io_failure(inode
, failrec
, 0);
2274 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2276 free_io_failure(inode
, failrec
, 0);
2279 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2280 bio
->bi_sector
= failrec
->logical
>> 9;
2281 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2284 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2285 if (btrfs_failed_bio
->csum
) {
2286 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2287 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2289 btrfs_bio
= btrfs_io_bio(bio
);
2290 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2291 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2292 phy_offset
*= csum_size
;
2293 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ phy_offset
,
2297 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2299 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2300 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2301 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2303 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2304 failrec
->this_mirror
,
2305 failrec
->bio_flags
, 0);
2309 /* lots and lots of room for performance fixes in the end_bio funcs */
2311 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2313 int uptodate
= (err
== 0);
2314 struct extent_io_tree
*tree
;
2317 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2319 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2320 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2321 end
, NULL
, uptodate
);
2327 ClearPageUptodate(page
);
2334 * after a writepage IO is done, we need to:
2335 * clear the uptodate bits on error
2336 * clear the writeback bits in the extent tree for this IO
2337 * end_page_writeback if the page has no more pending IO
2339 * Scheduling is not allowed, so the extent state tree is expected
2340 * to have one and only one object corresponding to this IO.
2342 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2344 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2345 struct extent_io_tree
*tree
;
2350 struct page
*page
= bvec
->bv_page
;
2351 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2353 /* We always issue full-page reads, but if some block
2354 * in a page fails to read, blk_update_request() will
2355 * advance bv_offset and adjust bv_len to compensate.
2356 * Print a warning for nonzero offsets, and an error
2357 * if they don't add up to a full page. */
2358 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2359 printk("%s page write in btrfs with offset %u and length %u\n",
2360 bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
2361 ? KERN_ERR
"partial" : KERN_INFO
"incomplete",
2362 bvec
->bv_offset
, bvec
->bv_len
);
2364 start
= page_offset(page
);
2365 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2367 if (--bvec
>= bio
->bi_io_vec
)
2368 prefetchw(&bvec
->bv_page
->flags
);
2370 if (end_extent_writepage(page
, err
, start
, end
))
2373 end_page_writeback(page
);
2374 } while (bvec
>= bio
->bi_io_vec
);
2380 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2383 struct extent_state
*cached
= NULL
;
2384 u64 end
= start
+ len
- 1;
2386 if (uptodate
&& tree
->track_uptodate
)
2387 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2388 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2392 * after a readpage IO is done, we need to:
2393 * clear the uptodate bits on error
2394 * set the uptodate bits if things worked
2395 * set the page up to date if all extents in the tree are uptodate
2396 * clear the lock bit in the extent tree
2397 * unlock the page if there are no other extents locked for it
2399 * Scheduling is not allowed, so the extent state tree is expected
2400 * to have one and only one object corresponding to this IO.
2402 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2404 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2405 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2406 struct bio_vec
*bvec
= bio
->bi_io_vec
;
2407 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2408 struct extent_io_tree
*tree
;
2413 u64 extent_start
= 0;
2422 struct page
*page
= bvec
->bv_page
;
2423 struct inode
*inode
= page
->mapping
->host
;
2425 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2426 "mirror=%lu\n", (u64
)bio
->bi_sector
, err
,
2427 io_bio
->mirror_num
);
2428 tree
= &BTRFS_I(inode
)->io_tree
;
2430 /* We always issue full-page reads, but if some block
2431 * in a page fails to read, blk_update_request() will
2432 * advance bv_offset and adjust bv_len to compensate.
2433 * Print a warning for nonzero offsets, and an error
2434 * if they don't add up to a full page. */
2435 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2436 printk("%s page read in btrfs with offset %u and length %u\n",
2437 bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
2438 ? KERN_ERR
"partial" : KERN_INFO
"incomplete",
2439 bvec
->bv_offset
, bvec
->bv_len
);
2441 start
= page_offset(page
);
2442 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2445 if (++bvec
<= bvec_end
)
2446 prefetchw(&bvec
->bv_page
->flags
);
2448 mirror
= io_bio
->mirror_num
;
2449 if (likely(uptodate
&& tree
->ops
&&
2450 tree
->ops
->readpage_end_io_hook
)) {
2451 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2457 clean_io_failure(start
, page
);
2460 if (likely(uptodate
))
2463 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2464 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2466 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2470 * The generic bio_readpage_error handles errors the
2471 * following way: If possible, new read requests are
2472 * created and submitted and will end up in
2473 * end_bio_extent_readpage as well (if we're lucky, not
2474 * in the !uptodate case). In that case it returns 0 and
2475 * we just go on with the next page in our bio. If it
2476 * can't handle the error it will return -EIO and we
2477 * remain responsible for that page.
2479 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2483 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2490 if (likely(uptodate
)) {
2491 loff_t i_size
= i_size_read(inode
);
2492 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2495 /* Zero out the end if this page straddles i_size */
2496 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2497 if (page
->index
== end_index
&& offset
)
2498 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2499 SetPageUptodate(page
);
2501 ClearPageUptodate(page
);
2507 if (unlikely(!uptodate
)) {
2509 endio_readpage_release_extent(tree
,
2515 endio_readpage_release_extent(tree
, start
,
2516 end
- start
+ 1, 0);
2517 } else if (!extent_len
) {
2518 extent_start
= start
;
2519 extent_len
= end
+ 1 - start
;
2520 } else if (extent_start
+ extent_len
== start
) {
2521 extent_len
+= end
+ 1 - start
;
2523 endio_readpage_release_extent(tree
, extent_start
,
2524 extent_len
, uptodate
);
2525 extent_start
= start
;
2526 extent_len
= end
+ 1 - start
;
2528 } while (bvec
<= bvec_end
);
2531 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2534 io_bio
->end_io(io_bio
, err
);
2539 * this allocates from the btrfs_bioset. We're returning a bio right now
2540 * but you can call btrfs_io_bio for the appropriate container_of magic
2543 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2546 struct btrfs_io_bio
*btrfs_bio
;
2549 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2551 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2552 while (!bio
&& (nr_vecs
/= 2)) {
2553 bio
= bio_alloc_bioset(gfp_flags
,
2554 nr_vecs
, btrfs_bioset
);
2560 bio
->bi_bdev
= bdev
;
2561 bio
->bi_sector
= first_sector
;
2562 btrfs_bio
= btrfs_io_bio(bio
);
2563 btrfs_bio
->csum
= NULL
;
2564 btrfs_bio
->csum_allocated
= NULL
;
2565 btrfs_bio
->end_io
= NULL
;
2570 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2572 return bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2576 /* this also allocates from the btrfs_bioset */
2577 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2579 struct btrfs_io_bio
*btrfs_bio
;
2582 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2584 btrfs_bio
= btrfs_io_bio(bio
);
2585 btrfs_bio
->csum
= NULL
;
2586 btrfs_bio
->csum_allocated
= NULL
;
2587 btrfs_bio
->end_io
= NULL
;
2593 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2594 int mirror_num
, unsigned long bio_flags
)
2597 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2598 struct page
*page
= bvec
->bv_page
;
2599 struct extent_io_tree
*tree
= bio
->bi_private
;
2602 start
= page_offset(page
) + bvec
->bv_offset
;
2604 bio
->bi_private
= NULL
;
2608 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2609 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2610 mirror_num
, bio_flags
, start
);
2612 btrfsic_submit_bio(rw
, bio
);
2614 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2620 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2621 unsigned long offset
, size_t size
, struct bio
*bio
,
2622 unsigned long bio_flags
)
2625 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2626 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2633 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2634 struct page
*page
, sector_t sector
,
2635 size_t size
, unsigned long offset
,
2636 struct block_device
*bdev
,
2637 struct bio
**bio_ret
,
2638 unsigned long max_pages
,
2639 bio_end_io_t end_io_func
,
2641 unsigned long prev_bio_flags
,
2642 unsigned long bio_flags
)
2648 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2649 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2650 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2652 if (bio_ret
&& *bio_ret
) {
2655 contig
= bio
->bi_sector
== sector
;
2657 contig
= bio_end_sector(bio
) == sector
;
2659 if (prev_bio_flags
!= bio_flags
|| !contig
||
2660 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2661 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2662 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2671 if (this_compressed
)
2674 nr
= bio_get_nr_vecs(bdev
);
2676 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2680 bio_add_page(bio
, page
, page_size
, offset
);
2681 bio
->bi_end_io
= end_io_func
;
2682 bio
->bi_private
= tree
;
2687 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2692 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2695 if (!PagePrivate(page
)) {
2696 SetPagePrivate(page
);
2697 page_cache_get(page
);
2698 set_page_private(page
, (unsigned long)eb
);
2700 WARN_ON(page
->private != (unsigned long)eb
);
2704 void set_page_extent_mapped(struct page
*page
)
2706 if (!PagePrivate(page
)) {
2707 SetPagePrivate(page
);
2708 page_cache_get(page
);
2709 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2713 static struct extent_map
*
2714 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2715 u64 start
, u64 len
, get_extent_t
*get_extent
,
2716 struct extent_map
**em_cached
)
2718 struct extent_map
*em
;
2720 if (em_cached
&& *em_cached
) {
2722 if (em
->in_tree
&& start
>= em
->start
&&
2723 start
< extent_map_end(em
)) {
2724 atomic_inc(&em
->refs
);
2728 free_extent_map(em
);
2732 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2733 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2735 atomic_inc(&em
->refs
);
2741 * basic readpage implementation. Locked extent state structs are inserted
2742 * into the tree that are removed when the IO is done (by the end_io
2744 * XXX JDM: This needs looking at to ensure proper page locking
2746 static int __do_readpage(struct extent_io_tree
*tree
,
2748 get_extent_t
*get_extent
,
2749 struct extent_map
**em_cached
,
2750 struct bio
**bio
, int mirror_num
,
2751 unsigned long *bio_flags
, int rw
)
2753 struct inode
*inode
= page
->mapping
->host
;
2754 u64 start
= page_offset(page
);
2755 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2759 u64 last_byte
= i_size_read(inode
);
2763 struct extent_map
*em
;
2764 struct block_device
*bdev
;
2767 int parent_locked
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2768 size_t pg_offset
= 0;
2770 size_t disk_io_size
;
2771 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2772 unsigned long this_bio_flag
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2774 set_page_extent_mapped(page
);
2777 if (!PageUptodate(page
)) {
2778 if (cleancache_get_page(page
) == 0) {
2779 BUG_ON(blocksize
!= PAGE_SIZE
);
2780 unlock_extent(tree
, start
, end
);
2785 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2787 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2790 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2791 userpage
= kmap_atomic(page
);
2792 memset(userpage
+ zero_offset
, 0, iosize
);
2793 flush_dcache_page(page
);
2794 kunmap_atomic(userpage
);
2797 while (cur
<= end
) {
2798 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2800 if (cur
>= last_byte
) {
2802 struct extent_state
*cached
= NULL
;
2804 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2805 userpage
= kmap_atomic(page
);
2806 memset(userpage
+ pg_offset
, 0, iosize
);
2807 flush_dcache_page(page
);
2808 kunmap_atomic(userpage
);
2809 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2812 unlock_extent_cached(tree
, cur
,
2817 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2818 end
- cur
+ 1, get_extent
, em_cached
);
2819 if (IS_ERR_OR_NULL(em
)) {
2822 unlock_extent(tree
, cur
, end
);
2825 extent_offset
= cur
- em
->start
;
2826 BUG_ON(extent_map_end(em
) <= cur
);
2829 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2830 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2831 extent_set_compress_type(&this_bio_flag
,
2835 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2836 cur_end
= min(extent_map_end(em
) - 1, end
);
2837 iosize
= ALIGN(iosize
, blocksize
);
2838 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2839 disk_io_size
= em
->block_len
;
2840 sector
= em
->block_start
>> 9;
2842 sector
= (em
->block_start
+ extent_offset
) >> 9;
2843 disk_io_size
= iosize
;
2846 block_start
= em
->block_start
;
2847 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2848 block_start
= EXTENT_MAP_HOLE
;
2849 free_extent_map(em
);
2852 /* we've found a hole, just zero and go on */
2853 if (block_start
== EXTENT_MAP_HOLE
) {
2855 struct extent_state
*cached
= NULL
;
2857 userpage
= kmap_atomic(page
);
2858 memset(userpage
+ pg_offset
, 0, iosize
);
2859 flush_dcache_page(page
);
2860 kunmap_atomic(userpage
);
2862 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2864 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2867 pg_offset
+= iosize
;
2870 /* the get_extent function already copied into the page */
2871 if (test_range_bit(tree
, cur
, cur_end
,
2872 EXTENT_UPTODATE
, 1, NULL
)) {
2873 check_page_uptodate(tree
, page
);
2875 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2877 pg_offset
+= iosize
;
2880 /* we have an inline extent but it didn't get marked up
2881 * to date. Error out
2883 if (block_start
== EXTENT_MAP_INLINE
) {
2886 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2888 pg_offset
+= iosize
;
2893 ret
= submit_extent_page(rw
, tree
, page
,
2894 sector
, disk_io_size
, pg_offset
,
2896 end_bio_extent_readpage
, mirror_num
,
2901 *bio_flags
= this_bio_flag
;
2905 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2908 pg_offset
+= iosize
;
2912 if (!PageError(page
))
2913 SetPageUptodate(page
);
2919 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
2920 struct page
*pages
[], int nr_pages
,
2922 get_extent_t
*get_extent
,
2923 struct extent_map
**em_cached
,
2924 struct bio
**bio
, int mirror_num
,
2925 unsigned long *bio_flags
, int rw
)
2927 struct inode
*inode
;
2928 struct btrfs_ordered_extent
*ordered
;
2931 inode
= pages
[0]->mapping
->host
;
2933 lock_extent(tree
, start
, end
);
2934 ordered
= btrfs_lookup_ordered_range(inode
, start
,
2938 unlock_extent(tree
, start
, end
);
2939 btrfs_start_ordered_extent(inode
, ordered
, 1);
2940 btrfs_put_ordered_extent(ordered
);
2943 for (index
= 0; index
< nr_pages
; index
++) {
2944 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
2945 mirror_num
, bio_flags
, rw
);
2946 page_cache_release(pages
[index
]);
2950 static void __extent_readpages(struct extent_io_tree
*tree
,
2951 struct page
*pages
[],
2952 int nr_pages
, get_extent_t
*get_extent
,
2953 struct extent_map
**em_cached
,
2954 struct bio
**bio
, int mirror_num
,
2955 unsigned long *bio_flags
, int rw
)
2963 for (index
= 0; index
< nr_pages
; index
++) {
2964 page_start
= page_offset(pages
[index
]);
2967 end
= start
+ PAGE_CACHE_SIZE
- 1;
2968 first_index
= index
;
2969 } else if (end
+ 1 == page_start
) {
2970 end
+= PAGE_CACHE_SIZE
;
2972 __do_contiguous_readpages(tree
, &pages
[first_index
],
2973 index
- first_index
, start
,
2974 end
, get_extent
, em_cached
,
2975 bio
, mirror_num
, bio_flags
,
2978 end
= start
+ PAGE_CACHE_SIZE
- 1;
2979 first_index
= index
;
2984 __do_contiguous_readpages(tree
, &pages
[first_index
],
2985 index
- first_index
, start
,
2986 end
, get_extent
, em_cached
, bio
,
2987 mirror_num
, bio_flags
, rw
);
2990 static int __extent_read_full_page(struct extent_io_tree
*tree
,
2992 get_extent_t
*get_extent
,
2993 struct bio
**bio
, int mirror_num
,
2994 unsigned long *bio_flags
, int rw
)
2996 struct inode
*inode
= page
->mapping
->host
;
2997 struct btrfs_ordered_extent
*ordered
;
2998 u64 start
= page_offset(page
);
2999 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3003 lock_extent(tree
, start
, end
);
3004 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
3007 unlock_extent(tree
, start
, end
);
3008 btrfs_start_ordered_extent(inode
, ordered
, 1);
3009 btrfs_put_ordered_extent(ordered
);
3012 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3017 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3018 get_extent_t
*get_extent
, int mirror_num
)
3020 struct bio
*bio
= NULL
;
3021 unsigned long bio_flags
= 0;
3024 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3027 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3031 int extent_read_full_page_nolock(struct extent_io_tree
*tree
, struct page
*page
,
3032 get_extent_t
*get_extent
, int mirror_num
)
3034 struct bio
*bio
= NULL
;
3035 unsigned long bio_flags
= EXTENT_BIO_PARENT_LOCKED
;
3038 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, &bio
, mirror_num
,
3041 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3045 static noinline
void update_nr_written(struct page
*page
,
3046 struct writeback_control
*wbc
,
3047 unsigned long nr_written
)
3049 wbc
->nr_to_write
-= nr_written
;
3050 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
3051 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
3052 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
3056 * the writepage semantics are similar to regular writepage. extent
3057 * records are inserted to lock ranges in the tree, and as dirty areas
3058 * are found, they are marked writeback. Then the lock bits are removed
3059 * and the end_io handler clears the writeback ranges
3061 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3064 struct inode
*inode
= page
->mapping
->host
;
3065 struct extent_page_data
*epd
= data
;
3066 struct extent_io_tree
*tree
= epd
->tree
;
3067 u64 start
= page_offset(page
);
3069 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3073 u64 last_byte
= i_size_read(inode
);
3077 struct extent_state
*cached_state
= NULL
;
3078 struct extent_map
*em
;
3079 struct block_device
*bdev
;
3082 size_t pg_offset
= 0;
3084 loff_t i_size
= i_size_read(inode
);
3085 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
3091 unsigned long nr_written
= 0;
3092 bool fill_delalloc
= true;
3094 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3095 write_flags
= WRITE_SYNC
;
3097 write_flags
= WRITE
;
3099 trace___extent_writepage(page
, inode
, wbc
);
3101 WARN_ON(!PageLocked(page
));
3103 ClearPageError(page
);
3105 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
3106 if (page
->index
> end_index
||
3107 (page
->index
== end_index
&& !pg_offset
)) {
3108 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
3113 if (page
->index
== end_index
) {
3116 userpage
= kmap_atomic(page
);
3117 memset(userpage
+ pg_offset
, 0,
3118 PAGE_CACHE_SIZE
- pg_offset
);
3119 kunmap_atomic(userpage
);
3120 flush_dcache_page(page
);
3124 set_page_extent_mapped(page
);
3126 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
3127 fill_delalloc
= false;
3129 delalloc_start
= start
;
3132 if (!epd
->extent_locked
&& fill_delalloc
) {
3133 u64 delalloc_to_write
= 0;
3135 * make sure the wbc mapping index is at least updated
3138 update_nr_written(page
, wbc
, 0);
3140 while (delalloc_end
< page_end
) {
3141 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3146 if (nr_delalloc
== 0) {
3147 delalloc_start
= delalloc_end
+ 1;
3150 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3155 /* File system has been set read-only */
3161 * delalloc_end is already one less than the total
3162 * length, so we don't subtract one from
3165 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3168 delalloc_start
= delalloc_end
+ 1;
3170 if (wbc
->nr_to_write
< delalloc_to_write
) {
3173 if (delalloc_to_write
< thresh
* 2)
3174 thresh
= delalloc_to_write
;
3175 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3179 /* did the fill delalloc function already unlock and start
3185 * we've unlocked the page, so we can't update
3186 * the mapping's writeback index, just update
3189 wbc
->nr_to_write
-= nr_written
;
3193 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3194 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3197 /* Fixup worker will requeue */
3199 wbc
->pages_skipped
++;
3201 redirty_page_for_writepage(wbc
, page
);
3202 update_nr_written(page
, wbc
, nr_written
);
3210 * we don't want to touch the inode after unlocking the page,
3211 * so we update the mapping writeback index now
3213 update_nr_written(page
, wbc
, nr_written
+ 1);
3216 if (last_byte
<= start
) {
3217 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3218 tree
->ops
->writepage_end_io_hook(page
, start
,
3223 blocksize
= inode
->i_sb
->s_blocksize
;
3225 while (cur
<= end
) {
3226 if (cur
>= last_byte
) {
3227 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3228 tree
->ops
->writepage_end_io_hook(page
, cur
,
3232 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3234 if (IS_ERR_OR_NULL(em
)) {
3239 extent_offset
= cur
- em
->start
;
3240 BUG_ON(extent_map_end(em
) <= cur
);
3242 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
3243 iosize
= ALIGN(iosize
, blocksize
);
3244 sector
= (em
->block_start
+ extent_offset
) >> 9;
3246 block_start
= em
->block_start
;
3247 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3248 free_extent_map(em
);
3252 * compressed and inline extents are written through other
3255 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3256 block_start
== EXTENT_MAP_INLINE
) {
3258 * end_io notification does not happen here for
3259 * compressed extents
3261 if (!compressed
&& tree
->ops
&&
3262 tree
->ops
->writepage_end_io_hook
)
3263 tree
->ops
->writepage_end_io_hook(page
, cur
,
3266 else if (compressed
) {
3267 /* we don't want to end_page_writeback on
3268 * a compressed extent. this happens
3275 pg_offset
+= iosize
;
3278 /* leave this out until we have a page_mkwrite call */
3279 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
3280 EXTENT_DIRTY
, 0, NULL
)) {
3282 pg_offset
+= iosize
;
3286 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3287 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3295 unsigned long max_nr
= end_index
+ 1;
3297 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3298 if (!PageWriteback(page
)) {
3299 printk(KERN_ERR
"btrfs warning page %lu not "
3300 "writeback, cur %llu end %llu\n",
3301 page
->index
, (unsigned long long)cur
,
3302 (unsigned long long)end
);
3305 ret
= submit_extent_page(write_flags
, tree
, page
,
3306 sector
, iosize
, pg_offset
,
3307 bdev
, &epd
->bio
, max_nr
,
3308 end_bio_extent_writepage
,
3314 pg_offset
+= iosize
;
3319 /* make sure the mapping tag for page dirty gets cleared */
3320 set_page_writeback(page
);
3321 end_page_writeback(page
);
3327 /* drop our reference on any cached states */
3328 free_extent_state(cached_state
);
3332 static int eb_wait(void *word
)
3338 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3340 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3341 TASK_UNINTERRUPTIBLE
);
3344 static int lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3345 struct btrfs_fs_info
*fs_info
,
3346 struct extent_page_data
*epd
)
3348 unsigned long i
, num_pages
;
3352 if (!btrfs_try_tree_write_lock(eb
)) {
3354 flush_write_bio(epd
);
3355 btrfs_tree_lock(eb
);
3358 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3359 btrfs_tree_unlock(eb
);
3363 flush_write_bio(epd
);
3367 wait_on_extent_buffer_writeback(eb
);
3368 btrfs_tree_lock(eb
);
3369 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3371 btrfs_tree_unlock(eb
);
3376 * We need to do this to prevent races in people who check if the eb is
3377 * under IO since we can end up having no IO bits set for a short period
3380 spin_lock(&eb
->refs_lock
);
3381 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3382 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3383 spin_unlock(&eb
->refs_lock
);
3384 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3385 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3387 fs_info
->dirty_metadata_batch
);
3390 spin_unlock(&eb
->refs_lock
);
3393 btrfs_tree_unlock(eb
);
3398 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3399 for (i
= 0; i
< num_pages
; i
++) {
3400 struct page
*p
= extent_buffer_page(eb
, i
);
3402 if (!trylock_page(p
)) {
3404 flush_write_bio(epd
);
3414 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3416 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3417 smp_mb__after_clear_bit();
3418 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3421 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3423 int uptodate
= err
== 0;
3424 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
3425 struct extent_buffer
*eb
;
3429 struct page
*page
= bvec
->bv_page
;
3432 eb
= (struct extent_buffer
*)page
->private;
3434 done
= atomic_dec_and_test(&eb
->io_pages
);
3436 if (!uptodate
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3437 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3438 ClearPageUptodate(page
);
3442 end_page_writeback(page
);
3447 end_extent_buffer_writeback(eb
);
3448 } while (bvec
>= bio
->bi_io_vec
);
3454 static int write_one_eb(struct extent_buffer
*eb
,
3455 struct btrfs_fs_info
*fs_info
,
3456 struct writeback_control
*wbc
,
3457 struct extent_page_data
*epd
)
3459 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3460 u64 offset
= eb
->start
;
3461 unsigned long i
, num_pages
;
3462 unsigned long bio_flags
= 0;
3463 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3466 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3467 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3468 atomic_set(&eb
->io_pages
, num_pages
);
3469 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3470 bio_flags
= EXTENT_BIO_TREE_LOG
;
3472 for (i
= 0; i
< num_pages
; i
++) {
3473 struct page
*p
= extent_buffer_page(eb
, i
);
3475 clear_page_dirty_for_io(p
);
3476 set_page_writeback(p
);
3477 ret
= submit_extent_page(rw
, eb
->tree
, p
, offset
>> 9,
3478 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3479 -1, end_bio_extent_buffer_writepage
,
3480 0, epd
->bio_flags
, bio_flags
);
3481 epd
->bio_flags
= bio_flags
;
3483 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3485 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3486 end_extent_buffer_writeback(eb
);
3490 offset
+= PAGE_CACHE_SIZE
;
3491 update_nr_written(p
, wbc
, 1);
3495 if (unlikely(ret
)) {
3496 for (; i
< num_pages
; i
++) {
3497 struct page
*p
= extent_buffer_page(eb
, i
);
3505 int btree_write_cache_pages(struct address_space
*mapping
,
3506 struct writeback_control
*wbc
)
3508 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3509 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3510 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3511 struct extent_page_data epd
= {
3515 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3520 int nr_to_write_done
= 0;
3521 struct pagevec pvec
;
3524 pgoff_t end
; /* Inclusive */
3528 pagevec_init(&pvec
, 0);
3529 if (wbc
->range_cyclic
) {
3530 index
= mapping
->writeback_index
; /* Start from prev offset */
3533 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3534 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3537 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3538 tag
= PAGECACHE_TAG_TOWRITE
;
3540 tag
= PAGECACHE_TAG_DIRTY
;
3542 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3543 tag_pages_for_writeback(mapping
, index
, end
);
3544 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3545 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3546 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3550 for (i
= 0; i
< nr_pages
; i
++) {
3551 struct page
*page
= pvec
.pages
[i
];
3553 if (!PagePrivate(page
))
3556 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3561 spin_lock(&mapping
->private_lock
);
3562 if (!PagePrivate(page
)) {
3563 spin_unlock(&mapping
->private_lock
);
3567 eb
= (struct extent_buffer
*)page
->private;
3570 * Shouldn't happen and normally this would be a BUG_ON
3571 * but no sense in crashing the users box for something
3572 * we can survive anyway.
3575 spin_unlock(&mapping
->private_lock
);
3580 if (eb
== prev_eb
) {
3581 spin_unlock(&mapping
->private_lock
);
3585 ret
= atomic_inc_not_zero(&eb
->refs
);
3586 spin_unlock(&mapping
->private_lock
);
3591 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3593 free_extent_buffer(eb
);
3597 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3600 free_extent_buffer(eb
);
3603 free_extent_buffer(eb
);
3606 * the filesystem may choose to bump up nr_to_write.
3607 * We have to make sure to honor the new nr_to_write
3610 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3612 pagevec_release(&pvec
);
3615 if (!scanned
&& !done
) {
3617 * We hit the last page and there is more work to be done: wrap
3618 * back to the start of the file
3624 flush_write_bio(&epd
);
3629 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3630 * @mapping: address space structure to write
3631 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3632 * @writepage: function called for each page
3633 * @data: data passed to writepage function
3635 * If a page is already under I/O, write_cache_pages() skips it, even
3636 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3637 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3638 * and msync() need to guarantee that all the data which was dirty at the time
3639 * the call was made get new I/O started against them. If wbc->sync_mode is
3640 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3641 * existing IO to complete.
3643 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3644 struct address_space
*mapping
,
3645 struct writeback_control
*wbc
,
3646 writepage_t writepage
, void *data
,
3647 void (*flush_fn
)(void *))
3649 struct inode
*inode
= mapping
->host
;
3652 int nr_to_write_done
= 0;
3653 struct pagevec pvec
;
3656 pgoff_t end
; /* Inclusive */
3661 * We have to hold onto the inode so that ordered extents can do their
3662 * work when the IO finishes. The alternative to this is failing to add
3663 * an ordered extent if the igrab() fails there and that is a huge pain
3664 * to deal with, so instead just hold onto the inode throughout the
3665 * writepages operation. If it fails here we are freeing up the inode
3666 * anyway and we'd rather not waste our time writing out stuff that is
3667 * going to be truncated anyway.
3672 pagevec_init(&pvec
, 0);
3673 if (wbc
->range_cyclic
) {
3674 index
= mapping
->writeback_index
; /* Start from prev offset */
3677 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3678 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3681 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3682 tag
= PAGECACHE_TAG_TOWRITE
;
3684 tag
= PAGECACHE_TAG_DIRTY
;
3686 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3687 tag_pages_for_writeback(mapping
, index
, end
);
3688 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3689 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3690 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3694 for (i
= 0; i
< nr_pages
; i
++) {
3695 struct page
*page
= pvec
.pages
[i
];
3698 * At this point we hold neither mapping->tree_lock nor
3699 * lock on the page itself: the page may be truncated or
3700 * invalidated (changing page->mapping to NULL), or even
3701 * swizzled back from swapper_space to tmpfs file
3704 if (!trylock_page(page
)) {
3709 if (unlikely(page
->mapping
!= mapping
)) {
3714 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3720 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3721 if (PageWriteback(page
))
3723 wait_on_page_writeback(page
);
3726 if (PageWriteback(page
) ||
3727 !clear_page_dirty_for_io(page
)) {
3732 ret
= (*writepage
)(page
, wbc
, data
);
3734 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3742 * the filesystem may choose to bump up nr_to_write.
3743 * We have to make sure to honor the new nr_to_write
3746 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3748 pagevec_release(&pvec
);
3751 if (!scanned
&& !done
) {
3753 * We hit the last page and there is more work to be done: wrap
3754 * back to the start of the file
3760 btrfs_add_delayed_iput(inode
);
3764 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3773 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
3774 BUG_ON(ret
< 0); /* -ENOMEM */
3779 static noinline
void flush_write_bio(void *data
)
3781 struct extent_page_data
*epd
= data
;
3782 flush_epd_write_bio(epd
);
3785 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3786 get_extent_t
*get_extent
,
3787 struct writeback_control
*wbc
)
3790 struct extent_page_data epd
= {
3793 .get_extent
= get_extent
,
3795 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3799 ret
= __extent_writepage(page
, wbc
, &epd
);
3801 flush_epd_write_bio(&epd
);
3805 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3806 u64 start
, u64 end
, get_extent_t
*get_extent
,
3810 struct address_space
*mapping
= inode
->i_mapping
;
3812 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3815 struct extent_page_data epd
= {
3818 .get_extent
= get_extent
,
3820 .sync_io
= mode
== WB_SYNC_ALL
,
3823 struct writeback_control wbc_writepages
= {
3825 .nr_to_write
= nr_pages
* 2,
3826 .range_start
= start
,
3827 .range_end
= end
+ 1,
3830 while (start
<= end
) {
3831 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3832 if (clear_page_dirty_for_io(page
))
3833 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3835 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3836 tree
->ops
->writepage_end_io_hook(page
, start
,
3837 start
+ PAGE_CACHE_SIZE
- 1,
3841 page_cache_release(page
);
3842 start
+= PAGE_CACHE_SIZE
;
3845 flush_epd_write_bio(&epd
);
3849 int extent_writepages(struct extent_io_tree
*tree
,
3850 struct address_space
*mapping
,
3851 get_extent_t
*get_extent
,
3852 struct writeback_control
*wbc
)
3855 struct extent_page_data epd
= {
3858 .get_extent
= get_extent
,
3860 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3864 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3865 __extent_writepage
, &epd
,
3867 flush_epd_write_bio(&epd
);
3871 int extent_readpages(struct extent_io_tree
*tree
,
3872 struct address_space
*mapping
,
3873 struct list_head
*pages
, unsigned nr_pages
,
3874 get_extent_t get_extent
)
3876 struct bio
*bio
= NULL
;
3878 unsigned long bio_flags
= 0;
3879 struct page
*pagepool
[16];
3881 struct extent_map
*em_cached
= NULL
;
3884 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3885 page
= list_entry(pages
->prev
, struct page
, lru
);
3887 prefetchw(&page
->flags
);
3888 list_del(&page
->lru
);
3889 if (add_to_page_cache_lru(page
, mapping
,
3890 page
->index
, GFP_NOFS
)) {
3891 page_cache_release(page
);
3895 pagepool
[nr
++] = page
;
3896 if (nr
< ARRAY_SIZE(pagepool
))
3898 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3899 &bio
, 0, &bio_flags
, READ
);
3903 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3904 &bio
, 0, &bio_flags
, READ
);
3907 free_extent_map(em_cached
);
3909 BUG_ON(!list_empty(pages
));
3911 return submit_one_bio(READ
, bio
, 0, bio_flags
);
3916 * basic invalidatepage code, this waits on any locked or writeback
3917 * ranges corresponding to the page, and then deletes any extent state
3918 * records from the tree
3920 int extent_invalidatepage(struct extent_io_tree
*tree
,
3921 struct page
*page
, unsigned long offset
)
3923 struct extent_state
*cached_state
= NULL
;
3924 u64 start
= page_offset(page
);
3925 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3926 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
3928 start
+= ALIGN(offset
, blocksize
);
3932 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
3933 wait_on_page_writeback(page
);
3934 clear_extent_bit(tree
, start
, end
,
3935 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3936 EXTENT_DO_ACCOUNTING
,
3937 1, 1, &cached_state
, GFP_NOFS
);
3942 * a helper for releasepage, this tests for areas of the page that
3943 * are locked or under IO and drops the related state bits if it is safe
3946 static int try_release_extent_state(struct extent_map_tree
*map
,
3947 struct extent_io_tree
*tree
,
3948 struct page
*page
, gfp_t mask
)
3950 u64 start
= page_offset(page
);
3951 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3954 if (test_range_bit(tree
, start
, end
,
3955 EXTENT_IOBITS
, 0, NULL
))
3958 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
3961 * at this point we can safely clear everything except the
3962 * locked bit and the nodatasum bit
3964 ret
= clear_extent_bit(tree
, start
, end
,
3965 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
3968 /* if clear_extent_bit failed for enomem reasons,
3969 * we can't allow the release to continue.
3980 * a helper for releasepage. As long as there are no locked extents
3981 * in the range corresponding to the page, both state records and extent
3982 * map records are removed
3984 int try_release_extent_mapping(struct extent_map_tree
*map
,
3985 struct extent_io_tree
*tree
, struct page
*page
,
3988 struct extent_map
*em
;
3989 u64 start
= page_offset(page
);
3990 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3992 if ((mask
& __GFP_WAIT
) &&
3993 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
3995 while (start
<= end
) {
3996 len
= end
- start
+ 1;
3997 write_lock(&map
->lock
);
3998 em
= lookup_extent_mapping(map
, start
, len
);
4000 write_unlock(&map
->lock
);
4003 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4004 em
->start
!= start
) {
4005 write_unlock(&map
->lock
);
4006 free_extent_map(em
);
4009 if (!test_range_bit(tree
, em
->start
,
4010 extent_map_end(em
) - 1,
4011 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4013 remove_extent_mapping(map
, em
);
4014 /* once for the rb tree */
4015 free_extent_map(em
);
4017 start
= extent_map_end(em
);
4018 write_unlock(&map
->lock
);
4021 free_extent_map(em
);
4024 return try_release_extent_state(map
, tree
, page
, mask
);
4028 * helper function for fiemap, which doesn't want to see any holes.
4029 * This maps until we find something past 'last'
4031 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4034 get_extent_t
*get_extent
)
4036 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4037 struct extent_map
*em
;
4044 len
= last
- offset
;
4047 len
= ALIGN(len
, sectorsize
);
4048 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4049 if (IS_ERR_OR_NULL(em
))
4052 /* if this isn't a hole return it */
4053 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4054 em
->block_start
!= EXTENT_MAP_HOLE
) {
4058 /* this is a hole, advance to the next extent */
4059 offset
= extent_map_end(em
);
4060 free_extent_map(em
);
4067 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4068 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4072 u64 max
= start
+ len
;
4076 u64 last_for_get_extent
= 0;
4078 u64 isize
= i_size_read(inode
);
4079 struct btrfs_key found_key
;
4080 struct extent_map
*em
= NULL
;
4081 struct extent_state
*cached_state
= NULL
;
4082 struct btrfs_path
*path
;
4083 struct btrfs_file_extent_item
*item
;
4088 unsigned long emflags
;
4093 path
= btrfs_alloc_path();
4096 path
->leave_spinning
= 1;
4098 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
4099 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
4102 * lookup the last file extent. We're not using i_size here
4103 * because there might be preallocation past i_size
4105 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
4106 path
, btrfs_ino(inode
), -1, 0);
4108 btrfs_free_path(path
);
4113 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4114 struct btrfs_file_extent_item
);
4115 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4116 found_type
= btrfs_key_type(&found_key
);
4118 /* No extents, but there might be delalloc bits */
4119 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4120 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4121 /* have to trust i_size as the end */
4123 last_for_get_extent
= isize
;
4126 * remember the start of the last extent. There are a
4127 * bunch of different factors that go into the length of the
4128 * extent, so its much less complex to remember where it started
4130 last
= found_key
.offset
;
4131 last_for_get_extent
= last
+ 1;
4133 btrfs_free_path(path
);
4136 * we might have some extents allocated but more delalloc past those
4137 * extents. so, we trust isize unless the start of the last extent is
4142 last_for_get_extent
= isize
;
4145 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1, 0,
4148 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4158 u64 offset_in_extent
= 0;
4160 /* break if the extent we found is outside the range */
4161 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4165 * get_extent may return an extent that starts before our
4166 * requested range. We have to make sure the ranges
4167 * we return to fiemap always move forward and don't
4168 * overlap, so adjust the offsets here
4170 em_start
= max(em
->start
, off
);
4173 * record the offset from the start of the extent
4174 * for adjusting the disk offset below. Only do this if the
4175 * extent isn't compressed since our in ram offset may be past
4176 * what we have actually allocated on disk.
4178 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4179 offset_in_extent
= em_start
- em
->start
;
4180 em_end
= extent_map_end(em
);
4181 em_len
= em_end
- em_start
;
4182 emflags
= em
->flags
;
4187 * bump off for our next call to get_extent
4189 off
= extent_map_end(em
);
4193 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4195 flags
|= FIEMAP_EXTENT_LAST
;
4196 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4197 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4198 FIEMAP_EXTENT_NOT_ALIGNED
);
4199 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4200 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4201 FIEMAP_EXTENT_UNKNOWN
);
4203 disko
= em
->block_start
+ offset_in_extent
;
4205 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4206 flags
|= FIEMAP_EXTENT_ENCODED
;
4208 free_extent_map(em
);
4210 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4211 (last
== (u64
)-1 && isize
<= em_end
)) {
4212 flags
|= FIEMAP_EXTENT_LAST
;
4216 /* now scan forward to see if this is really the last extent. */
4217 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4224 flags
|= FIEMAP_EXTENT_LAST
;
4227 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4233 free_extent_map(em
);
4235 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4236 &cached_state
, GFP_NOFS
);
4240 static void __free_extent_buffer(struct extent_buffer
*eb
)
4242 btrfs_leak_debug_del(&eb
->leak_list
);
4243 kmem_cache_free(extent_buffer_cache
, eb
);
4246 static int extent_buffer_under_io(struct extent_buffer
*eb
)
4248 return (atomic_read(&eb
->io_pages
) ||
4249 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4250 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4254 * Helper for releasing extent buffer page.
4256 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4257 unsigned long start_idx
)
4259 unsigned long index
;
4260 unsigned long num_pages
;
4262 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4264 BUG_ON(extent_buffer_under_io(eb
));
4266 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4267 index
= start_idx
+ num_pages
;
4268 if (start_idx
>= index
)
4273 page
= extent_buffer_page(eb
, index
);
4274 if (page
&& mapped
) {
4275 spin_lock(&page
->mapping
->private_lock
);
4277 * We do this since we'll remove the pages after we've
4278 * removed the eb from the radix tree, so we could race
4279 * and have this page now attached to the new eb. So
4280 * only clear page_private if it's still connected to
4283 if (PagePrivate(page
) &&
4284 page
->private == (unsigned long)eb
) {
4285 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4286 BUG_ON(PageDirty(page
));
4287 BUG_ON(PageWriteback(page
));
4289 * We need to make sure we haven't be attached
4292 ClearPagePrivate(page
);
4293 set_page_private(page
, 0);
4294 /* One for the page private */
4295 page_cache_release(page
);
4297 spin_unlock(&page
->mapping
->private_lock
);
4301 /* One for when we alloced the page */
4302 page_cache_release(page
);
4304 } while (index
!= start_idx
);
4308 * Helper for releasing the extent buffer.
4310 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4312 btrfs_release_extent_buffer_page(eb
, 0);
4313 __free_extent_buffer(eb
);
4316 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
4321 struct extent_buffer
*eb
= NULL
;
4323 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
4330 rwlock_init(&eb
->lock
);
4331 atomic_set(&eb
->write_locks
, 0);
4332 atomic_set(&eb
->read_locks
, 0);
4333 atomic_set(&eb
->blocking_readers
, 0);
4334 atomic_set(&eb
->blocking_writers
, 0);
4335 atomic_set(&eb
->spinning_readers
, 0);
4336 atomic_set(&eb
->spinning_writers
, 0);
4337 eb
->lock_nested
= 0;
4338 init_waitqueue_head(&eb
->write_lock_wq
);
4339 init_waitqueue_head(&eb
->read_lock_wq
);
4341 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4343 spin_lock_init(&eb
->refs_lock
);
4344 atomic_set(&eb
->refs
, 1);
4345 atomic_set(&eb
->io_pages
, 0);
4348 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4350 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4351 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4352 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4357 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4361 struct extent_buffer
*new;
4362 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4364 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_NOFS
);
4368 for (i
= 0; i
< num_pages
; i
++) {
4369 p
= alloc_page(GFP_NOFS
);
4371 btrfs_release_extent_buffer(new);
4374 attach_extent_buffer_page(new, p
);
4375 WARN_ON(PageDirty(p
));
4380 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4381 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4382 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4387 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
4389 struct extent_buffer
*eb
;
4390 unsigned long num_pages
= num_extent_pages(0, len
);
4393 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_NOFS
);
4397 for (i
= 0; i
< num_pages
; i
++) {
4398 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4402 set_extent_buffer_uptodate(eb
);
4403 btrfs_set_header_nritems(eb
, 0);
4404 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4409 __free_page(eb
->pages
[i
- 1]);
4410 __free_extent_buffer(eb
);
4414 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4417 /* the ref bit is tricky. We have to make sure it is set
4418 * if we have the buffer dirty. Otherwise the
4419 * code to free a buffer can end up dropping a dirty
4422 * Once the ref bit is set, it won't go away while the
4423 * buffer is dirty or in writeback, and it also won't
4424 * go away while we have the reference count on the
4427 * We can't just set the ref bit without bumping the
4428 * ref on the eb because free_extent_buffer might
4429 * see the ref bit and try to clear it. If this happens
4430 * free_extent_buffer might end up dropping our original
4431 * ref by mistake and freeing the page before we are able
4432 * to add one more ref.
4434 * So bump the ref count first, then set the bit. If someone
4435 * beat us to it, drop the ref we added.
4437 refs
= atomic_read(&eb
->refs
);
4438 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4441 spin_lock(&eb
->refs_lock
);
4442 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4443 atomic_inc(&eb
->refs
);
4444 spin_unlock(&eb
->refs_lock
);
4447 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
)
4449 unsigned long num_pages
, i
;
4451 check_buffer_tree_ref(eb
);
4453 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4454 for (i
= 0; i
< num_pages
; i
++) {
4455 struct page
*p
= extent_buffer_page(eb
, i
);
4456 mark_page_accessed(p
);
4460 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
4461 u64 start
, unsigned long len
)
4463 unsigned long num_pages
= num_extent_pages(start
, len
);
4465 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4466 struct extent_buffer
*eb
;
4467 struct extent_buffer
*exists
= NULL
;
4469 struct address_space
*mapping
= tree
->mapping
;
4474 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4475 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4477 mark_extent_buffer_accessed(eb
);
4482 eb
= __alloc_extent_buffer(tree
, start
, len
, GFP_NOFS
);
4486 for (i
= 0; i
< num_pages
; i
++, index
++) {
4487 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4491 spin_lock(&mapping
->private_lock
);
4492 if (PagePrivate(p
)) {
4494 * We could have already allocated an eb for this page
4495 * and attached one so lets see if we can get a ref on
4496 * the existing eb, and if we can we know it's good and
4497 * we can just return that one, else we know we can just
4498 * overwrite page->private.
4500 exists
= (struct extent_buffer
*)p
->private;
4501 if (atomic_inc_not_zero(&exists
->refs
)) {
4502 spin_unlock(&mapping
->private_lock
);
4504 page_cache_release(p
);
4505 mark_extent_buffer_accessed(exists
);
4510 * Do this so attach doesn't complain and we need to
4511 * drop the ref the old guy had.
4513 ClearPagePrivate(p
);
4514 WARN_ON(PageDirty(p
));
4515 page_cache_release(p
);
4517 attach_extent_buffer_page(eb
, p
);
4518 spin_unlock(&mapping
->private_lock
);
4519 WARN_ON(PageDirty(p
));
4520 mark_page_accessed(p
);
4522 if (!PageUptodate(p
))
4526 * see below about how we avoid a nasty race with release page
4527 * and why we unlock later
4531 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4533 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4537 spin_lock(&tree
->buffer_lock
);
4538 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
4539 if (ret
== -EEXIST
) {
4540 exists
= radix_tree_lookup(&tree
->buffer
,
4541 start
>> PAGE_CACHE_SHIFT
);
4542 if (!atomic_inc_not_zero(&exists
->refs
)) {
4543 spin_unlock(&tree
->buffer_lock
);
4544 radix_tree_preload_end();
4548 spin_unlock(&tree
->buffer_lock
);
4549 radix_tree_preload_end();
4550 mark_extent_buffer_accessed(exists
);
4553 /* add one reference for the tree */
4554 check_buffer_tree_ref(eb
);
4555 spin_unlock(&tree
->buffer_lock
);
4556 radix_tree_preload_end();
4559 * there is a race where release page may have
4560 * tried to find this extent buffer in the radix
4561 * but failed. It will tell the VM it is safe to
4562 * reclaim the, and it will clear the page private bit.
4563 * We must make sure to set the page private bit properly
4564 * after the extent buffer is in the radix tree so
4565 * it doesn't get lost
4567 SetPageChecked(eb
->pages
[0]);
4568 for (i
= 1; i
< num_pages
; i
++) {
4569 p
= extent_buffer_page(eb
, i
);
4570 ClearPageChecked(p
);
4573 unlock_page(eb
->pages
[0]);
4577 for (i
= 0; i
< num_pages
; i
++) {
4579 unlock_page(eb
->pages
[i
]);
4582 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4583 btrfs_release_extent_buffer(eb
);
4587 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
4588 u64 start
, unsigned long len
)
4590 struct extent_buffer
*eb
;
4593 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4594 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4596 mark_extent_buffer_accessed(eb
);
4604 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4606 struct extent_buffer
*eb
=
4607 container_of(head
, struct extent_buffer
, rcu_head
);
4609 __free_extent_buffer(eb
);
4612 /* Expects to have eb->eb_lock already held */
4613 static int release_extent_buffer(struct extent_buffer
*eb
)
4615 WARN_ON(atomic_read(&eb
->refs
) == 0);
4616 if (atomic_dec_and_test(&eb
->refs
)) {
4617 if (test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
)) {
4618 spin_unlock(&eb
->refs_lock
);
4620 struct extent_io_tree
*tree
= eb
->tree
;
4622 spin_unlock(&eb
->refs_lock
);
4624 spin_lock(&tree
->buffer_lock
);
4625 radix_tree_delete(&tree
->buffer
,
4626 eb
->start
>> PAGE_CACHE_SHIFT
);
4627 spin_unlock(&tree
->buffer_lock
);
4630 /* Should be safe to release our pages at this point */
4631 btrfs_release_extent_buffer_page(eb
, 0);
4632 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4635 spin_unlock(&eb
->refs_lock
);
4640 void free_extent_buffer(struct extent_buffer
*eb
)
4648 refs
= atomic_read(&eb
->refs
);
4651 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
4656 spin_lock(&eb
->refs_lock
);
4657 if (atomic_read(&eb
->refs
) == 2 &&
4658 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4659 atomic_dec(&eb
->refs
);
4661 if (atomic_read(&eb
->refs
) == 2 &&
4662 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4663 !extent_buffer_under_io(eb
) &&
4664 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4665 atomic_dec(&eb
->refs
);
4668 * I know this is terrible, but it's temporary until we stop tracking
4669 * the uptodate bits and such for the extent buffers.
4671 release_extent_buffer(eb
);
4674 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4679 spin_lock(&eb
->refs_lock
);
4680 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4682 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4683 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4684 atomic_dec(&eb
->refs
);
4685 release_extent_buffer(eb
);
4688 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4691 unsigned long num_pages
;
4694 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4696 for (i
= 0; i
< num_pages
; i
++) {
4697 page
= extent_buffer_page(eb
, i
);
4698 if (!PageDirty(page
))
4702 WARN_ON(!PagePrivate(page
));
4704 clear_page_dirty_for_io(page
);
4705 spin_lock_irq(&page
->mapping
->tree_lock
);
4706 if (!PageDirty(page
)) {
4707 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4709 PAGECACHE_TAG_DIRTY
);
4711 spin_unlock_irq(&page
->mapping
->tree_lock
);
4712 ClearPageError(page
);
4715 WARN_ON(atomic_read(&eb
->refs
) == 0);
4718 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4721 unsigned long num_pages
;
4724 check_buffer_tree_ref(eb
);
4726 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4728 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4729 WARN_ON(atomic_read(&eb
->refs
) == 0);
4730 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4732 for (i
= 0; i
< num_pages
; i
++)
4733 set_page_dirty(extent_buffer_page(eb
, i
));
4737 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4741 unsigned long num_pages
;
4743 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4744 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4745 for (i
= 0; i
< num_pages
; i
++) {
4746 page
= extent_buffer_page(eb
, i
);
4748 ClearPageUptodate(page
);
4753 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4757 unsigned long num_pages
;
4759 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4760 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4761 for (i
= 0; i
< num_pages
; i
++) {
4762 page
= extent_buffer_page(eb
, i
);
4763 SetPageUptodate(page
);
4768 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4770 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4773 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4774 struct extent_buffer
*eb
, u64 start
, int wait
,
4775 get_extent_t
*get_extent
, int mirror_num
)
4778 unsigned long start_i
;
4782 int locked_pages
= 0;
4783 int all_uptodate
= 1;
4784 unsigned long num_pages
;
4785 unsigned long num_reads
= 0;
4786 struct bio
*bio
= NULL
;
4787 unsigned long bio_flags
= 0;
4789 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4793 WARN_ON(start
< eb
->start
);
4794 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4795 (eb
->start
>> PAGE_CACHE_SHIFT
);
4800 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4801 for (i
= start_i
; i
< num_pages
; i
++) {
4802 page
= extent_buffer_page(eb
, i
);
4803 if (wait
== WAIT_NONE
) {
4804 if (!trylock_page(page
))
4810 if (!PageUptodate(page
)) {
4817 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4821 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4822 eb
->read_mirror
= 0;
4823 atomic_set(&eb
->io_pages
, num_reads
);
4824 for (i
= start_i
; i
< num_pages
; i
++) {
4825 page
= extent_buffer_page(eb
, i
);
4826 if (!PageUptodate(page
)) {
4827 ClearPageError(page
);
4828 err
= __extent_read_full_page(tree
, page
,
4830 mirror_num
, &bio_flags
,
4840 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
4846 if (ret
|| wait
!= WAIT_COMPLETE
)
4849 for (i
= start_i
; i
< num_pages
; i
++) {
4850 page
= extent_buffer_page(eb
, i
);
4851 wait_on_page_locked(page
);
4852 if (!PageUptodate(page
))
4860 while (locked_pages
> 0) {
4861 page
= extent_buffer_page(eb
, i
);
4869 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
4870 unsigned long start
,
4877 char *dst
= (char *)dstv
;
4878 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4879 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4881 WARN_ON(start
> eb
->len
);
4882 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4884 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4887 page
= extent_buffer_page(eb
, i
);
4889 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4890 kaddr
= page_address(page
);
4891 memcpy(dst
, kaddr
+ offset
, cur
);
4900 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
4901 unsigned long min_len
, char **map
,
4902 unsigned long *map_start
,
4903 unsigned long *map_len
)
4905 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
4908 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4909 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4910 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
4917 offset
= start_offset
;
4921 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
4924 if (start
+ min_len
> eb
->len
) {
4925 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
4926 "wanted %lu %lu\n", (unsigned long long)eb
->start
,
4927 eb
->len
, start
, min_len
);
4931 p
= extent_buffer_page(eb
, i
);
4932 kaddr
= page_address(p
);
4933 *map
= kaddr
+ offset
;
4934 *map_len
= PAGE_CACHE_SIZE
- offset
;
4938 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
4939 unsigned long start
,
4946 char *ptr
= (char *)ptrv
;
4947 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4948 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4951 WARN_ON(start
> eb
->len
);
4952 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4954 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4957 page
= extent_buffer_page(eb
, i
);
4959 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4961 kaddr
= page_address(page
);
4962 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
4974 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
4975 unsigned long start
, unsigned long len
)
4981 char *src
= (char *)srcv
;
4982 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4983 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4985 WARN_ON(start
> eb
->len
);
4986 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4988 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4991 page
= extent_buffer_page(eb
, i
);
4992 WARN_ON(!PageUptodate(page
));
4994 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4995 kaddr
= page_address(page
);
4996 memcpy(kaddr
+ offset
, src
, cur
);
5005 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5006 unsigned long start
, unsigned long len
)
5012 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5013 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5015 WARN_ON(start
> eb
->len
);
5016 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5018 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
5021 page
= extent_buffer_page(eb
, i
);
5022 WARN_ON(!PageUptodate(page
));
5024 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5025 kaddr
= page_address(page
);
5026 memset(kaddr
+ offset
, c
, cur
);
5034 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5035 unsigned long dst_offset
, unsigned long src_offset
,
5038 u64 dst_len
= dst
->len
;
5043 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5044 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5046 WARN_ON(src
->len
!= dst_len
);
5048 offset
= (start_offset
+ dst_offset
) &
5049 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5052 page
= extent_buffer_page(dst
, i
);
5053 WARN_ON(!PageUptodate(page
));
5055 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
5057 kaddr
= page_address(page
);
5058 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5067 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
5068 unsigned long dst_off
, unsigned long src_off
,
5071 char *dst_kaddr
= page_address(dst_page
);
5072 if (dst_page
== src_page
) {
5073 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
5075 char *src_kaddr
= page_address(src_page
);
5076 char *p
= dst_kaddr
+ dst_off
+ len
;
5077 char *s
= src_kaddr
+ src_off
+ len
;
5084 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5086 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5087 return distance
< len
;
5090 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5091 unsigned long dst_off
, unsigned long src_off
,
5094 char *dst_kaddr
= page_address(dst_page
);
5096 int must_memmove
= 0;
5098 if (dst_page
!= src_page
) {
5099 src_kaddr
= page_address(src_page
);
5101 src_kaddr
= dst_kaddr
;
5102 if (areas_overlap(src_off
, dst_off
, len
))
5107 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5109 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5112 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5113 unsigned long src_offset
, unsigned long len
)
5116 size_t dst_off_in_page
;
5117 size_t src_off_in_page
;
5118 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5119 unsigned long dst_i
;
5120 unsigned long src_i
;
5122 if (src_offset
+ len
> dst
->len
) {
5123 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
5124 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
5127 if (dst_offset
+ len
> dst
->len
) {
5128 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
5129 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
5134 dst_off_in_page
= (start_offset
+ dst_offset
) &
5135 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5136 src_off_in_page
= (start_offset
+ src_offset
) &
5137 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5139 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5140 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5142 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5144 cur
= min_t(unsigned long, cur
,
5145 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5147 copy_pages(extent_buffer_page(dst
, dst_i
),
5148 extent_buffer_page(dst
, src_i
),
5149 dst_off_in_page
, src_off_in_page
, cur
);
5157 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5158 unsigned long src_offset
, unsigned long len
)
5161 size_t dst_off_in_page
;
5162 size_t src_off_in_page
;
5163 unsigned long dst_end
= dst_offset
+ len
- 1;
5164 unsigned long src_end
= src_offset
+ len
- 1;
5165 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5166 unsigned long dst_i
;
5167 unsigned long src_i
;
5169 if (src_offset
+ len
> dst
->len
) {
5170 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
5171 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
5174 if (dst_offset
+ len
> dst
->len
) {
5175 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
5176 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
5179 if (dst_offset
< src_offset
) {
5180 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5184 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5185 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5187 dst_off_in_page
= (start_offset
+ dst_end
) &
5188 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5189 src_off_in_page
= (start_offset
+ src_end
) &
5190 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5192 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5193 cur
= min(cur
, dst_off_in_page
+ 1);
5194 move_pages(extent_buffer_page(dst
, dst_i
),
5195 extent_buffer_page(dst
, src_i
),
5196 dst_off_in_page
- cur
+ 1,
5197 src_off_in_page
- cur
+ 1, cur
);
5205 int try_release_extent_buffer(struct page
*page
)
5207 struct extent_buffer
*eb
;
5210 * We need to make sure noboody is attaching this page to an eb right
5213 spin_lock(&page
->mapping
->private_lock
);
5214 if (!PagePrivate(page
)) {
5215 spin_unlock(&page
->mapping
->private_lock
);
5219 eb
= (struct extent_buffer
*)page
->private;
5223 * This is a little awful but should be ok, we need to make sure that
5224 * the eb doesn't disappear out from under us while we're looking at
5227 spin_lock(&eb
->refs_lock
);
5228 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5229 spin_unlock(&eb
->refs_lock
);
5230 spin_unlock(&page
->mapping
->private_lock
);
5233 spin_unlock(&page
->mapping
->private_lock
);
5236 * If tree ref isn't set then we know the ref on this eb is a real ref,
5237 * so just return, this page will likely be freed soon anyway.
5239 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5240 spin_unlock(&eb
->refs_lock
);
5244 return release_extent_buffer(eb
);