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 state
->start
, state
->end
, state
->state
, state
->tree
,
65 atomic_read(&state
->refs
));
66 list_del(&state
->leak_list
);
67 kmem_cache_free(extent_state_cache
, state
);
70 while (!list_empty(&buffers
)) {
71 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
72 printk(KERN_ERR
"btrfs buffer leak start %llu len %lu "
74 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
75 list_del(&eb
->leak_list
);
76 kmem_cache_free(extent_buffer_cache
, eb
);
80 #define btrfs_debug_check_extent_io_range(inode, start, end) \
81 __btrfs_debug_check_extent_io_range(__func__, (inode), (start), (end))
82 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
83 struct inode
*inode
, u64 start
, u64 end
)
85 u64 isize
= i_size_read(inode
);
87 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
88 printk_ratelimited(KERN_DEBUG
89 "btrfs: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
90 caller
, btrfs_ino(inode
), isize
, start
, end
);
94 #define btrfs_leak_debug_add(new, head) do {} while (0)
95 #define btrfs_leak_debug_del(entry) do {} while (0)
96 #define btrfs_leak_debug_check() do {} while (0)
97 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
100 #define BUFFER_LRU_MAX 64
105 struct rb_node rb_node
;
108 struct extent_page_data
{
110 struct extent_io_tree
*tree
;
111 get_extent_t
*get_extent
;
112 unsigned long bio_flags
;
114 /* tells writepage not to lock the state bits for this range
115 * it still does the unlocking
117 unsigned int extent_locked
:1;
119 /* tells the submit_bio code to use a WRITE_SYNC */
120 unsigned int sync_io
:1;
123 static noinline
void flush_write_bio(void *data
);
124 static inline struct btrfs_fs_info
*
125 tree_fs_info(struct extent_io_tree
*tree
)
127 return btrfs_sb(tree
->mapping
->host
->i_sb
);
130 int __init
extent_io_init(void)
132 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
133 sizeof(struct extent_state
), 0,
134 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
135 if (!extent_state_cache
)
138 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
139 sizeof(struct extent_buffer
), 0,
140 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
141 if (!extent_buffer_cache
)
142 goto free_state_cache
;
144 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
145 offsetof(struct btrfs_io_bio
, bio
));
147 goto free_buffer_cache
;
149 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
155 bioset_free(btrfs_bioset
);
159 kmem_cache_destroy(extent_buffer_cache
);
160 extent_buffer_cache
= NULL
;
163 kmem_cache_destroy(extent_state_cache
);
164 extent_state_cache
= NULL
;
168 void extent_io_exit(void)
170 btrfs_leak_debug_check();
173 * Make sure all delayed rcu free are flushed before we
177 if (extent_state_cache
)
178 kmem_cache_destroy(extent_state_cache
);
179 if (extent_buffer_cache
)
180 kmem_cache_destroy(extent_buffer_cache
);
182 bioset_free(btrfs_bioset
);
185 void extent_io_tree_init(struct extent_io_tree
*tree
,
186 struct address_space
*mapping
)
188 tree
->state
= RB_ROOT
;
189 INIT_RADIX_TREE(&tree
->buffer
, GFP_ATOMIC
);
191 tree
->dirty_bytes
= 0;
192 spin_lock_init(&tree
->lock
);
193 spin_lock_init(&tree
->buffer_lock
);
194 tree
->mapping
= mapping
;
197 static struct extent_state
*alloc_extent_state(gfp_t mask
)
199 struct extent_state
*state
;
201 state
= kmem_cache_alloc(extent_state_cache
, mask
);
207 btrfs_leak_debug_add(&state
->leak_list
, &states
);
208 atomic_set(&state
->refs
, 1);
209 init_waitqueue_head(&state
->wq
);
210 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
214 void free_extent_state(struct extent_state
*state
)
218 if (atomic_dec_and_test(&state
->refs
)) {
219 WARN_ON(state
->tree
);
220 btrfs_leak_debug_del(&state
->leak_list
);
221 trace_free_extent_state(state
, _RET_IP_
);
222 kmem_cache_free(extent_state_cache
, state
);
226 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
227 struct rb_node
*node
)
229 struct rb_node
**p
= &root
->rb_node
;
230 struct rb_node
*parent
= NULL
;
231 struct tree_entry
*entry
;
235 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
237 if (offset
< entry
->start
)
239 else if (offset
> entry
->end
)
245 rb_link_node(node
, parent
, p
);
246 rb_insert_color(node
, root
);
250 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
251 struct rb_node
**prev_ret
,
252 struct rb_node
**next_ret
)
254 struct rb_root
*root
= &tree
->state
;
255 struct rb_node
*n
= root
->rb_node
;
256 struct rb_node
*prev
= NULL
;
257 struct rb_node
*orig_prev
= NULL
;
258 struct tree_entry
*entry
;
259 struct tree_entry
*prev_entry
= NULL
;
262 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
266 if (offset
< entry
->start
)
268 else if (offset
> entry
->end
)
276 while (prev
&& offset
> prev_entry
->end
) {
277 prev
= rb_next(prev
);
278 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
285 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
286 while (prev
&& offset
< prev_entry
->start
) {
287 prev
= rb_prev(prev
);
288 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
295 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
298 struct rb_node
*prev
= NULL
;
301 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
307 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
308 struct extent_state
*other
)
310 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
311 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
316 * utility function to look for merge candidates inside a given range.
317 * Any extents with matching state are merged together into a single
318 * extent in the tree. Extents with EXTENT_IO in their state field
319 * are not merged because the end_io handlers need to be able to do
320 * operations on them without sleeping (or doing allocations/splits).
322 * This should be called with the tree lock held.
324 static void merge_state(struct extent_io_tree
*tree
,
325 struct extent_state
*state
)
327 struct extent_state
*other
;
328 struct rb_node
*other_node
;
330 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
333 other_node
= rb_prev(&state
->rb_node
);
335 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
336 if (other
->end
== state
->start
- 1 &&
337 other
->state
== state
->state
) {
338 merge_cb(tree
, state
, other
);
339 state
->start
= other
->start
;
341 rb_erase(&other
->rb_node
, &tree
->state
);
342 free_extent_state(other
);
345 other_node
= rb_next(&state
->rb_node
);
347 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
348 if (other
->start
== state
->end
+ 1 &&
349 other
->state
== state
->state
) {
350 merge_cb(tree
, state
, other
);
351 state
->end
= other
->end
;
353 rb_erase(&other
->rb_node
, &tree
->state
);
354 free_extent_state(other
);
359 static void set_state_cb(struct extent_io_tree
*tree
,
360 struct extent_state
*state
, unsigned long *bits
)
362 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
363 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
366 static void clear_state_cb(struct extent_io_tree
*tree
,
367 struct extent_state
*state
, unsigned long *bits
)
369 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
370 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
373 static void set_state_bits(struct extent_io_tree
*tree
,
374 struct extent_state
*state
, unsigned long *bits
);
377 * insert an extent_state struct into the tree. 'bits' are set on the
378 * struct before it is inserted.
380 * This may return -EEXIST if the extent is already there, in which case the
381 * state struct is freed.
383 * The tree lock is not taken internally. This is a utility function and
384 * probably isn't what you want to call (see set/clear_extent_bit).
386 static int insert_state(struct extent_io_tree
*tree
,
387 struct extent_state
*state
, u64 start
, u64 end
,
390 struct rb_node
*node
;
393 WARN(1, KERN_ERR
"btrfs end < start %llu %llu\n",
395 state
->start
= start
;
398 set_state_bits(tree
, state
, bits
);
400 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
402 struct extent_state
*found
;
403 found
= rb_entry(node
, struct extent_state
, rb_node
);
404 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
406 found
->start
, found
->end
, start
, end
);
410 merge_state(tree
, state
);
414 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
417 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
418 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
422 * split a given extent state struct in two, inserting the preallocated
423 * struct 'prealloc' as the newly created second half. 'split' indicates an
424 * offset inside 'orig' where it should be split.
427 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
428 * are two extent state structs in the tree:
429 * prealloc: [orig->start, split - 1]
430 * orig: [ split, orig->end ]
432 * The tree locks are not taken by this function. They need to be held
435 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
436 struct extent_state
*prealloc
, u64 split
)
438 struct rb_node
*node
;
440 split_cb(tree
, orig
, split
);
442 prealloc
->start
= orig
->start
;
443 prealloc
->end
= split
- 1;
444 prealloc
->state
= orig
->state
;
447 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
449 free_extent_state(prealloc
);
452 prealloc
->tree
= tree
;
456 static struct extent_state
*next_state(struct extent_state
*state
)
458 struct rb_node
*next
= rb_next(&state
->rb_node
);
460 return rb_entry(next
, struct extent_state
, rb_node
);
466 * utility function to clear some bits in an extent state struct.
467 * it will optionally wake up any one waiting on this state (wake == 1).
469 * If no bits are set on the state struct after clearing things, the
470 * struct is freed and removed from the tree
472 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
473 struct extent_state
*state
,
474 unsigned long *bits
, int wake
)
476 struct extent_state
*next
;
477 unsigned long bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
479 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
480 u64 range
= state
->end
- state
->start
+ 1;
481 WARN_ON(range
> tree
->dirty_bytes
);
482 tree
->dirty_bytes
-= range
;
484 clear_state_cb(tree
, state
, bits
);
485 state
->state
&= ~bits_to_clear
;
488 if (state
->state
== 0) {
489 next
= next_state(state
);
491 rb_erase(&state
->rb_node
, &tree
->state
);
493 free_extent_state(state
);
498 merge_state(tree
, state
);
499 next
= next_state(state
);
504 static struct extent_state
*
505 alloc_extent_state_atomic(struct extent_state
*prealloc
)
508 prealloc
= alloc_extent_state(GFP_ATOMIC
);
513 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
515 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
516 "Extent tree was modified by another "
517 "thread while locked.");
521 * clear some bits on a range in the tree. This may require splitting
522 * or inserting elements in the tree, so the gfp mask is used to
523 * indicate which allocations or sleeping are allowed.
525 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
526 * the given range from the tree regardless of state (ie for truncate).
528 * the range [start, end] is inclusive.
530 * This takes the tree lock, and returns 0 on success and < 0 on error.
532 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
533 unsigned long bits
, int wake
, int delete,
534 struct extent_state
**cached_state
,
537 struct extent_state
*state
;
538 struct extent_state
*cached
;
539 struct extent_state
*prealloc
= NULL
;
540 struct rb_node
*node
;
545 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
547 if (bits
& EXTENT_DELALLOC
)
548 bits
|= EXTENT_NORESERVE
;
551 bits
|= ~EXTENT_CTLBITS
;
552 bits
|= EXTENT_FIRST_DELALLOC
;
554 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
557 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
558 prealloc
= alloc_extent_state(mask
);
563 spin_lock(&tree
->lock
);
565 cached
= *cached_state
;
568 *cached_state
= NULL
;
572 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
573 cached
->end
> start
) {
575 atomic_dec(&cached
->refs
);
580 free_extent_state(cached
);
583 * this search will find the extents that end after
586 node
= tree_search(tree
, start
);
589 state
= rb_entry(node
, struct extent_state
, rb_node
);
591 if (state
->start
> end
)
593 WARN_ON(state
->end
< start
);
594 last_end
= state
->end
;
596 /* the state doesn't have the wanted bits, go ahead */
597 if (!(state
->state
& bits
)) {
598 state
= next_state(state
);
603 * | ---- desired range ---- |
605 * | ------------- state -------------- |
607 * We need to split the extent we found, and may flip
608 * bits on second half.
610 * If the extent we found extends past our range, we
611 * just split and search again. It'll get split again
612 * the next time though.
614 * If the extent we found is inside our range, we clear
615 * the desired bit on it.
618 if (state
->start
< start
) {
619 prealloc
= alloc_extent_state_atomic(prealloc
);
621 err
= split_state(tree
, state
, prealloc
, start
);
623 extent_io_tree_panic(tree
, err
);
628 if (state
->end
<= end
) {
629 state
= clear_state_bit(tree
, state
, &bits
, wake
);
635 * | ---- desired range ---- |
637 * We need to split the extent, and clear the bit
640 if (state
->start
<= end
&& state
->end
> end
) {
641 prealloc
= alloc_extent_state_atomic(prealloc
);
643 err
= split_state(tree
, state
, prealloc
, end
+ 1);
645 extent_io_tree_panic(tree
, err
);
650 clear_state_bit(tree
, prealloc
, &bits
, wake
);
656 state
= clear_state_bit(tree
, state
, &bits
, wake
);
658 if (last_end
== (u64
)-1)
660 start
= last_end
+ 1;
661 if (start
<= end
&& state
&& !need_resched())
666 spin_unlock(&tree
->lock
);
668 free_extent_state(prealloc
);
675 spin_unlock(&tree
->lock
);
676 if (mask
& __GFP_WAIT
)
681 static void wait_on_state(struct extent_io_tree
*tree
,
682 struct extent_state
*state
)
683 __releases(tree
->lock
)
684 __acquires(tree
->lock
)
687 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
688 spin_unlock(&tree
->lock
);
690 spin_lock(&tree
->lock
);
691 finish_wait(&state
->wq
, &wait
);
695 * waits for one or more bits to clear on a range in the state tree.
696 * The range [start, end] is inclusive.
697 * The tree lock is taken by this function
699 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
702 struct extent_state
*state
;
703 struct rb_node
*node
;
705 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
707 spin_lock(&tree
->lock
);
711 * this search will find all the extents that end after
714 node
= tree_search(tree
, start
);
718 state
= rb_entry(node
, struct extent_state
, rb_node
);
720 if (state
->start
> end
)
723 if (state
->state
& bits
) {
724 start
= state
->start
;
725 atomic_inc(&state
->refs
);
726 wait_on_state(tree
, state
);
727 free_extent_state(state
);
730 start
= state
->end
+ 1;
735 cond_resched_lock(&tree
->lock
);
738 spin_unlock(&tree
->lock
);
741 static void set_state_bits(struct extent_io_tree
*tree
,
742 struct extent_state
*state
,
745 unsigned long bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
747 set_state_cb(tree
, state
, bits
);
748 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
749 u64 range
= state
->end
- state
->start
+ 1;
750 tree
->dirty_bytes
+= range
;
752 state
->state
|= bits_to_set
;
755 static void cache_state(struct extent_state
*state
,
756 struct extent_state
**cached_ptr
)
758 if (cached_ptr
&& !(*cached_ptr
)) {
759 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
761 atomic_inc(&state
->refs
);
767 * set some bits on a range in the tree. This may require allocations or
768 * sleeping, so the gfp mask is used to indicate what is allowed.
770 * If any of the exclusive bits are set, this will fail with -EEXIST if some
771 * part of the range already has the desired bits set. The start of the
772 * existing range is returned in failed_start in this case.
774 * [start, end] is inclusive This takes the tree lock.
777 static int __must_check
778 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
779 unsigned long bits
, unsigned long exclusive_bits
,
780 u64
*failed_start
, struct extent_state
**cached_state
,
783 struct extent_state
*state
;
784 struct extent_state
*prealloc
= NULL
;
785 struct rb_node
*node
;
790 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
792 bits
|= EXTENT_FIRST_DELALLOC
;
794 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
795 prealloc
= alloc_extent_state(mask
);
799 spin_lock(&tree
->lock
);
800 if (cached_state
&& *cached_state
) {
801 state
= *cached_state
;
802 if (state
->start
<= start
&& state
->end
> start
&&
804 node
= &state
->rb_node
;
809 * this search will find all the extents that end after
812 node
= tree_search(tree
, start
);
814 prealloc
= alloc_extent_state_atomic(prealloc
);
816 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
818 extent_io_tree_panic(tree
, err
);
823 state
= rb_entry(node
, struct extent_state
, rb_node
);
825 last_start
= state
->start
;
826 last_end
= state
->end
;
829 * | ---- desired range ---- |
832 * Just lock what we found and keep going
834 if (state
->start
== start
&& state
->end
<= end
) {
835 if (state
->state
& exclusive_bits
) {
836 *failed_start
= state
->start
;
841 set_state_bits(tree
, state
, &bits
);
842 cache_state(state
, cached_state
);
843 merge_state(tree
, state
);
844 if (last_end
== (u64
)-1)
846 start
= last_end
+ 1;
847 state
= next_state(state
);
848 if (start
< end
&& state
&& state
->start
== start
&&
855 * | ---- desired range ---- |
858 * | ------------- state -------------- |
860 * We need to split the extent we found, and may flip bits on
863 * If the extent we found extends past our
864 * range, we just split and search again. It'll get split
865 * again the next time though.
867 * If the extent we found is inside our range, we set the
870 if (state
->start
< start
) {
871 if (state
->state
& exclusive_bits
) {
872 *failed_start
= start
;
877 prealloc
= alloc_extent_state_atomic(prealloc
);
879 err
= split_state(tree
, state
, prealloc
, start
);
881 extent_io_tree_panic(tree
, err
);
886 if (state
->end
<= end
) {
887 set_state_bits(tree
, state
, &bits
);
888 cache_state(state
, cached_state
);
889 merge_state(tree
, state
);
890 if (last_end
== (u64
)-1)
892 start
= last_end
+ 1;
893 state
= next_state(state
);
894 if (start
< end
&& state
&& state
->start
== start
&&
901 * | ---- desired range ---- |
902 * | state | or | state |
904 * There's a hole, we need to insert something in it and
905 * ignore the extent we found.
907 if (state
->start
> start
) {
909 if (end
< last_start
)
912 this_end
= last_start
- 1;
914 prealloc
= alloc_extent_state_atomic(prealloc
);
918 * Avoid to free 'prealloc' if it can be merged with
921 err
= insert_state(tree
, prealloc
, start
, this_end
,
924 extent_io_tree_panic(tree
, err
);
926 cache_state(prealloc
, cached_state
);
928 start
= this_end
+ 1;
932 * | ---- desired range ---- |
934 * We need to split the extent, and set the bit
937 if (state
->start
<= end
&& state
->end
> end
) {
938 if (state
->state
& exclusive_bits
) {
939 *failed_start
= start
;
944 prealloc
= alloc_extent_state_atomic(prealloc
);
946 err
= split_state(tree
, state
, prealloc
, end
+ 1);
948 extent_io_tree_panic(tree
, err
);
950 set_state_bits(tree
, prealloc
, &bits
);
951 cache_state(prealloc
, cached_state
);
952 merge_state(tree
, prealloc
);
960 spin_unlock(&tree
->lock
);
962 free_extent_state(prealloc
);
969 spin_unlock(&tree
->lock
);
970 if (mask
& __GFP_WAIT
)
975 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
976 unsigned long bits
, u64
* failed_start
,
977 struct extent_state
**cached_state
, gfp_t mask
)
979 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
985 * convert_extent_bit - convert all bits in a given range from one bit to
987 * @tree: the io tree to search
988 * @start: the start offset in bytes
989 * @end: the end offset in bytes (inclusive)
990 * @bits: the bits to set in this range
991 * @clear_bits: the bits to clear in this range
992 * @cached_state: state that we're going to cache
993 * @mask: the allocation mask
995 * This will go through and set bits for the given range. If any states exist
996 * already in this range they are set with the given bit and cleared of the
997 * clear_bits. This is only meant to be used by things that are mergeable, ie
998 * converting from say DELALLOC to DIRTY. This is not meant to be used with
999 * boundary bits like LOCK.
1001 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1002 unsigned long bits
, unsigned long clear_bits
,
1003 struct extent_state
**cached_state
, gfp_t mask
)
1005 struct extent_state
*state
;
1006 struct extent_state
*prealloc
= NULL
;
1007 struct rb_node
*node
;
1012 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
1015 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
1016 prealloc
= alloc_extent_state(mask
);
1021 spin_lock(&tree
->lock
);
1022 if (cached_state
&& *cached_state
) {
1023 state
= *cached_state
;
1024 if (state
->start
<= start
&& state
->end
> start
&&
1026 node
= &state
->rb_node
;
1032 * this search will find all the extents that end after
1035 node
= tree_search(tree
, start
);
1037 prealloc
= alloc_extent_state_atomic(prealloc
);
1042 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
1045 extent_io_tree_panic(tree
, err
);
1048 state
= rb_entry(node
, struct extent_state
, rb_node
);
1050 last_start
= state
->start
;
1051 last_end
= state
->end
;
1054 * | ---- desired range ---- |
1057 * Just lock what we found and keep going
1059 if (state
->start
== start
&& state
->end
<= end
) {
1060 set_state_bits(tree
, state
, &bits
);
1061 cache_state(state
, cached_state
);
1062 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1063 if (last_end
== (u64
)-1)
1065 start
= last_end
+ 1;
1066 if (start
< end
&& state
&& state
->start
== start
&&
1073 * | ---- desired range ---- |
1076 * | ------------- state -------------- |
1078 * We need to split the extent we found, and may flip bits on
1081 * If the extent we found extends past our
1082 * range, we just split and search again. It'll get split
1083 * again the next time though.
1085 * If the extent we found is inside our range, we set the
1086 * desired bit on it.
1088 if (state
->start
< start
) {
1089 prealloc
= alloc_extent_state_atomic(prealloc
);
1094 err
= split_state(tree
, state
, prealloc
, start
);
1096 extent_io_tree_panic(tree
, err
);
1100 if (state
->end
<= end
) {
1101 set_state_bits(tree
, state
, &bits
);
1102 cache_state(state
, cached_state
);
1103 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1104 if (last_end
== (u64
)-1)
1106 start
= last_end
+ 1;
1107 if (start
< end
&& state
&& state
->start
== start
&&
1114 * | ---- desired range ---- |
1115 * | state | or | state |
1117 * There's a hole, we need to insert something in it and
1118 * ignore the extent we found.
1120 if (state
->start
> start
) {
1122 if (end
< last_start
)
1125 this_end
= last_start
- 1;
1127 prealloc
= alloc_extent_state_atomic(prealloc
);
1134 * Avoid to free 'prealloc' if it can be merged with
1137 err
= insert_state(tree
, prealloc
, start
, this_end
,
1140 extent_io_tree_panic(tree
, err
);
1141 cache_state(prealloc
, cached_state
);
1143 start
= this_end
+ 1;
1147 * | ---- desired range ---- |
1149 * We need to split the extent, and set the bit
1152 if (state
->start
<= end
&& state
->end
> end
) {
1153 prealloc
= alloc_extent_state_atomic(prealloc
);
1159 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1161 extent_io_tree_panic(tree
, err
);
1163 set_state_bits(tree
, prealloc
, &bits
);
1164 cache_state(prealloc
, cached_state
);
1165 clear_state_bit(tree
, prealloc
, &clear_bits
, 0);
1173 spin_unlock(&tree
->lock
);
1175 free_extent_state(prealloc
);
1182 spin_unlock(&tree
->lock
);
1183 if (mask
& __GFP_WAIT
)
1188 /* wrappers around set/clear extent bit */
1189 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1192 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, NULL
,
1196 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1197 unsigned long bits
, gfp_t mask
)
1199 return set_extent_bit(tree
, start
, end
, bits
, NULL
,
1203 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1204 unsigned long bits
, gfp_t mask
)
1206 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
1209 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1210 struct extent_state
**cached_state
, gfp_t mask
)
1212 return set_extent_bit(tree
, start
, end
,
1213 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1214 NULL
, cached_state
, mask
);
1217 int set_extent_defrag(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1218 struct extent_state
**cached_state
, gfp_t mask
)
1220 return set_extent_bit(tree
, start
, end
,
1221 EXTENT_DELALLOC
| EXTENT_UPTODATE
| EXTENT_DEFRAG
,
1222 NULL
, cached_state
, mask
);
1225 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1228 return clear_extent_bit(tree
, start
, end
,
1229 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1230 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1233 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1236 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, NULL
,
1240 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1241 struct extent_state
**cached_state
, gfp_t mask
)
1243 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, NULL
,
1244 cached_state
, mask
);
1247 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1248 struct extent_state
**cached_state
, gfp_t mask
)
1250 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1251 cached_state
, mask
);
1255 * either insert or lock state struct between start and end use mask to tell
1256 * us if waiting is desired.
1258 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1259 unsigned long bits
, struct extent_state
**cached_state
)
1264 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1265 EXTENT_LOCKED
, &failed_start
,
1266 cached_state
, GFP_NOFS
);
1267 if (err
== -EEXIST
) {
1268 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1269 start
= failed_start
;
1272 WARN_ON(start
> end
);
1277 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1279 return lock_extent_bits(tree
, start
, end
, 0, NULL
);
1282 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1287 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1288 &failed_start
, NULL
, GFP_NOFS
);
1289 if (err
== -EEXIST
) {
1290 if (failed_start
> start
)
1291 clear_extent_bit(tree
, start
, failed_start
- 1,
1292 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1298 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1299 struct extent_state
**cached
, gfp_t mask
)
1301 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1305 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1307 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1311 int extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1313 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1314 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1317 while (index
<= end_index
) {
1318 page
= find_get_page(inode
->i_mapping
, index
);
1319 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1320 clear_page_dirty_for_io(page
);
1321 page_cache_release(page
);
1327 int extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1329 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1330 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1333 while (index
<= end_index
) {
1334 page
= find_get_page(inode
->i_mapping
, index
);
1335 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1336 account_page_redirty(page
);
1337 __set_page_dirty_nobuffers(page
);
1338 page_cache_release(page
);
1345 * helper function to set both pages and extents in the tree writeback
1347 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1349 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1350 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1353 while (index
<= end_index
) {
1354 page
= find_get_page(tree
->mapping
, index
);
1355 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1356 set_page_writeback(page
);
1357 page_cache_release(page
);
1363 /* find the first state struct with 'bits' set after 'start', and
1364 * return it. tree->lock must be held. NULL will returned if
1365 * nothing was found after 'start'
1367 static struct extent_state
*
1368 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1369 u64 start
, unsigned long bits
)
1371 struct rb_node
*node
;
1372 struct extent_state
*state
;
1375 * this search will find all the extents that end after
1378 node
= tree_search(tree
, start
);
1383 state
= rb_entry(node
, struct extent_state
, rb_node
);
1384 if (state
->end
>= start
&& (state
->state
& bits
))
1387 node
= rb_next(node
);
1396 * find the first offset in the io tree with 'bits' set. zero is
1397 * returned if we find something, and *start_ret and *end_ret are
1398 * set to reflect the state struct that was found.
1400 * If nothing was found, 1 is returned. If found something, return 0.
1402 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1403 u64
*start_ret
, u64
*end_ret
, unsigned long bits
,
1404 struct extent_state
**cached_state
)
1406 struct extent_state
*state
;
1410 spin_lock(&tree
->lock
);
1411 if (cached_state
&& *cached_state
) {
1412 state
= *cached_state
;
1413 if (state
->end
== start
- 1 && state
->tree
) {
1414 n
= rb_next(&state
->rb_node
);
1416 state
= rb_entry(n
, struct extent_state
,
1418 if (state
->state
& bits
)
1422 free_extent_state(*cached_state
);
1423 *cached_state
= NULL
;
1426 free_extent_state(*cached_state
);
1427 *cached_state
= NULL
;
1430 state
= find_first_extent_bit_state(tree
, start
, bits
);
1433 cache_state(state
, cached_state
);
1434 *start_ret
= state
->start
;
1435 *end_ret
= state
->end
;
1439 spin_unlock(&tree
->lock
);
1444 * find a contiguous range of bytes in the file marked as delalloc, not
1445 * more than 'max_bytes'. start and end are used to return the range,
1447 * 1 is returned if we find something, 0 if nothing was in the tree
1449 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1450 u64
*start
, u64
*end
, u64 max_bytes
,
1451 struct extent_state
**cached_state
)
1453 struct rb_node
*node
;
1454 struct extent_state
*state
;
1455 u64 cur_start
= *start
;
1457 u64 total_bytes
= 0;
1459 spin_lock(&tree
->lock
);
1462 * this search will find all the extents that end after
1465 node
= tree_search(tree
, cur_start
);
1473 state
= rb_entry(node
, struct extent_state
, rb_node
);
1474 if (found
&& (state
->start
!= cur_start
||
1475 (state
->state
& EXTENT_BOUNDARY
))) {
1478 if (!(state
->state
& EXTENT_DELALLOC
)) {
1484 *start
= state
->start
;
1485 *cached_state
= state
;
1486 atomic_inc(&state
->refs
);
1490 cur_start
= state
->end
+ 1;
1491 node
= rb_next(node
);
1492 total_bytes
+= state
->end
- state
->start
+ 1;
1493 if (total_bytes
>= max_bytes
)
1499 spin_unlock(&tree
->lock
);
1503 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1504 struct page
*locked_page
,
1508 struct page
*pages
[16];
1509 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1510 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1511 unsigned long nr_pages
= end_index
- index
+ 1;
1514 if (index
== locked_page
->index
&& end_index
== index
)
1517 while (nr_pages
> 0) {
1518 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1519 min_t(unsigned long, nr_pages
,
1520 ARRAY_SIZE(pages
)), pages
);
1521 for (i
= 0; i
< ret
; i
++) {
1522 if (pages
[i
] != locked_page
)
1523 unlock_page(pages
[i
]);
1524 page_cache_release(pages
[i
]);
1532 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1533 struct page
*locked_page
,
1537 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1538 unsigned long start_index
= index
;
1539 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1540 unsigned long pages_locked
= 0;
1541 struct page
*pages
[16];
1542 unsigned long nrpages
;
1546 /* the caller is responsible for locking the start index */
1547 if (index
== locked_page
->index
&& index
== end_index
)
1550 /* skip the page at the start index */
1551 nrpages
= end_index
- index
+ 1;
1552 while (nrpages
> 0) {
1553 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1554 min_t(unsigned long,
1555 nrpages
, ARRAY_SIZE(pages
)), pages
);
1560 /* now we have an array of pages, lock them all */
1561 for (i
= 0; i
< ret
; i
++) {
1563 * the caller is taking responsibility for
1566 if (pages
[i
] != locked_page
) {
1567 lock_page(pages
[i
]);
1568 if (!PageDirty(pages
[i
]) ||
1569 pages
[i
]->mapping
!= inode
->i_mapping
) {
1571 unlock_page(pages
[i
]);
1572 page_cache_release(pages
[i
]);
1576 page_cache_release(pages
[i
]);
1585 if (ret
&& pages_locked
) {
1586 __unlock_for_delalloc(inode
, locked_page
,
1588 ((u64
)(start_index
+ pages_locked
- 1)) <<
1595 * find a contiguous range of bytes in the file marked as delalloc, not
1596 * more than 'max_bytes'. start and end are used to return the range,
1598 * 1 is returned if we find something, 0 if nothing was in the tree
1600 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1601 struct extent_io_tree
*tree
,
1602 struct page
*locked_page
,
1603 u64
*start
, u64
*end
,
1609 struct extent_state
*cached_state
= NULL
;
1614 /* step one, find a bunch of delalloc bytes starting at start */
1615 delalloc_start
= *start
;
1617 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1618 max_bytes
, &cached_state
);
1619 if (!found
|| delalloc_end
<= *start
) {
1620 *start
= delalloc_start
;
1621 *end
= delalloc_end
;
1622 free_extent_state(cached_state
);
1627 * start comes from the offset of locked_page. We have to lock
1628 * pages in order, so we can't process delalloc bytes before
1631 if (delalloc_start
< *start
)
1632 delalloc_start
= *start
;
1635 * make sure to limit the number of pages we try to lock down
1637 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1638 delalloc_end
= delalloc_start
+ max_bytes
- 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 max_bytes
= PAGE_CACHE_SIZE
;
1657 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1659 /* step three, lock the state bits for the whole range */
1660 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1662 /* then test to make sure it is all still delalloc */
1663 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1664 EXTENT_DELALLOC
, 1, cached_state
);
1666 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1667 &cached_state
, GFP_NOFS
);
1668 __unlock_for_delalloc(inode
, locked_page
,
1669 delalloc_start
, delalloc_end
);
1673 free_extent_state(cached_state
);
1674 *start
= delalloc_start
;
1675 *end
= delalloc_end
;
1680 int extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1681 struct page
*locked_page
,
1682 unsigned long clear_bits
,
1683 unsigned long page_ops
)
1685 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1687 struct page
*pages
[16];
1688 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1689 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1690 unsigned long nr_pages
= end_index
- index
+ 1;
1693 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1697 while (nr_pages
> 0) {
1698 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1699 min_t(unsigned long,
1700 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1701 for (i
= 0; i
< ret
; i
++) {
1703 if (page_ops
& PAGE_SET_PRIVATE2
)
1704 SetPagePrivate2(pages
[i
]);
1706 if (pages
[i
] == locked_page
) {
1707 page_cache_release(pages
[i
]);
1710 if (page_ops
& PAGE_CLEAR_DIRTY
)
1711 clear_page_dirty_for_io(pages
[i
]);
1712 if (page_ops
& PAGE_SET_WRITEBACK
)
1713 set_page_writeback(pages
[i
]);
1714 if (page_ops
& PAGE_END_WRITEBACK
)
1715 end_page_writeback(pages
[i
]);
1716 if (page_ops
& PAGE_UNLOCK
)
1717 unlock_page(pages
[i
]);
1718 page_cache_release(pages
[i
]);
1728 * count the number of bytes in the tree that have a given bit(s)
1729 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1730 * cached. The total number found is returned.
1732 u64
count_range_bits(struct extent_io_tree
*tree
,
1733 u64
*start
, u64 search_end
, u64 max_bytes
,
1734 unsigned long bits
, int contig
)
1736 struct rb_node
*node
;
1737 struct extent_state
*state
;
1738 u64 cur_start
= *start
;
1739 u64 total_bytes
= 0;
1743 if (search_end
<= cur_start
) {
1748 spin_lock(&tree
->lock
);
1749 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1750 total_bytes
= tree
->dirty_bytes
;
1754 * this search will find all the extents that end after
1757 node
= tree_search(tree
, cur_start
);
1762 state
= rb_entry(node
, struct extent_state
, rb_node
);
1763 if (state
->start
> search_end
)
1765 if (contig
&& found
&& state
->start
> last
+ 1)
1767 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1768 total_bytes
+= min(search_end
, state
->end
) + 1 -
1769 max(cur_start
, state
->start
);
1770 if (total_bytes
>= max_bytes
)
1773 *start
= max(cur_start
, state
->start
);
1777 } else if (contig
&& found
) {
1780 node
= rb_next(node
);
1785 spin_unlock(&tree
->lock
);
1790 * set the private field for a given byte offset in the tree. If there isn't
1791 * an extent_state there already, this does nothing.
1793 static int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1795 struct rb_node
*node
;
1796 struct extent_state
*state
;
1799 spin_lock(&tree
->lock
);
1801 * this search will find all the extents that end after
1804 node
= tree_search(tree
, start
);
1809 state
= rb_entry(node
, struct extent_state
, rb_node
);
1810 if (state
->start
!= start
) {
1814 state
->private = private;
1816 spin_unlock(&tree
->lock
);
1820 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1822 struct rb_node
*node
;
1823 struct extent_state
*state
;
1826 spin_lock(&tree
->lock
);
1828 * this search will find all the extents that end after
1831 node
= tree_search(tree
, start
);
1836 state
= rb_entry(node
, struct extent_state
, rb_node
);
1837 if (state
->start
!= start
) {
1841 *private = state
->private;
1843 spin_unlock(&tree
->lock
);
1848 * searches a range in the state tree for a given mask.
1849 * If 'filled' == 1, this returns 1 only if every extent in the tree
1850 * has the bits set. Otherwise, 1 is returned if any bit in the
1851 * range is found set.
1853 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1854 unsigned long bits
, int filled
, struct extent_state
*cached
)
1856 struct extent_state
*state
= NULL
;
1857 struct rb_node
*node
;
1860 spin_lock(&tree
->lock
);
1861 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1862 cached
->end
> start
)
1863 node
= &cached
->rb_node
;
1865 node
= tree_search(tree
, start
);
1866 while (node
&& start
<= end
) {
1867 state
= rb_entry(node
, struct extent_state
, rb_node
);
1869 if (filled
&& state
->start
> start
) {
1874 if (state
->start
> end
)
1877 if (state
->state
& bits
) {
1881 } else if (filled
) {
1886 if (state
->end
== (u64
)-1)
1889 start
= state
->end
+ 1;
1892 node
= rb_next(node
);
1899 spin_unlock(&tree
->lock
);
1904 * helper function to set a given page up to date if all the
1905 * extents in the tree for that page are up to date
1907 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1909 u64 start
= page_offset(page
);
1910 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1911 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1912 SetPageUptodate(page
);
1916 * When IO fails, either with EIO or csum verification fails, we
1917 * try other mirrors that might have a good copy of the data. This
1918 * io_failure_record is used to record state as we go through all the
1919 * mirrors. If another mirror has good data, the page is set up to date
1920 * and things continue. If a good mirror can't be found, the original
1921 * bio end_io callback is called to indicate things have failed.
1923 struct io_failure_record
{
1928 unsigned long bio_flags
;
1934 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
1939 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1941 set_state_private(failure_tree
, rec
->start
, 0);
1942 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1943 rec
->start
+ rec
->len
- 1,
1944 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1948 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1949 rec
->start
+ rec
->len
- 1,
1950 EXTENT_DAMAGED
, GFP_NOFS
);
1958 static void repair_io_failure_callback(struct bio
*bio
, int err
)
1960 complete(bio
->bi_private
);
1964 * this bypasses the standard btrfs submit functions deliberately, as
1965 * the standard behavior is to write all copies in a raid setup. here we only
1966 * want to write the one bad copy. so we do the mapping for ourselves and issue
1967 * submit_bio directly.
1968 * to avoid any synchronization issues, wait for the data after writing, which
1969 * actually prevents the read that triggered the error from finishing.
1970 * currently, there can be no more than two copies of every data bit. thus,
1971 * exactly one rewrite is required.
1973 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 start
,
1974 u64 length
, u64 logical
, struct page
*page
,
1978 struct btrfs_device
*dev
;
1979 DECLARE_COMPLETION_ONSTACK(compl);
1982 struct btrfs_bio
*bbio
= NULL
;
1983 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
1986 BUG_ON(!mirror_num
);
1988 /* we can't repair anything in raid56 yet */
1989 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
1992 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
1995 bio
->bi_private
= &compl;
1996 bio
->bi_end_io
= repair_io_failure_callback
;
1998 map_length
= length
;
2000 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2001 &map_length
, &bbio
, mirror_num
);
2006 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2007 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2008 bio
->bi_sector
= sector
;
2009 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2011 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2015 bio
->bi_bdev
= dev
->bdev
;
2016 bio_add_page(bio
, page
, length
, start
- page_offset(page
));
2017 btrfsic_submit_bio(WRITE_SYNC
, bio
);
2018 wait_for_completion(&compl);
2020 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
2021 /* try to remap that extent elsewhere? */
2023 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2027 printk_ratelimited_in_rcu(KERN_INFO
"btrfs read error corrected: ino %lu off %llu "
2028 "(dev %s sector %llu)\n", page
->mapping
->host
->i_ino
,
2029 start
, rcu_str_deref(dev
->name
), sector
);
2035 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2038 u64 start
= eb
->start
;
2039 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2042 for (i
= 0; i
< num_pages
; i
++) {
2043 struct page
*p
= extent_buffer_page(eb
, i
);
2044 ret
= repair_io_failure(root
->fs_info
, start
, PAGE_CACHE_SIZE
,
2045 start
, p
, mirror_num
);
2048 start
+= PAGE_CACHE_SIZE
;
2055 * each time an IO finishes, we do a fast check in the IO failure tree
2056 * to see if we need to process or clean up an io_failure_record
2058 static int clean_io_failure(u64 start
, struct page
*page
)
2061 u64 private_failure
;
2062 struct io_failure_record
*failrec
;
2063 struct btrfs_fs_info
*fs_info
;
2064 struct extent_state
*state
;
2068 struct inode
*inode
= page
->mapping
->host
;
2071 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2072 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2076 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2081 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2082 BUG_ON(!failrec
->this_mirror
);
2084 if (failrec
->in_validation
) {
2085 /* there was no real error, just free the record */
2086 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2092 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2093 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2096 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2098 if (state
&& state
->start
<= failrec
->start
&&
2099 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2100 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2101 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2103 if (num_copies
> 1) {
2104 ret
= repair_io_failure(fs_info
, start
, failrec
->len
,
2105 failrec
->logical
, page
,
2106 failrec
->failed_mirror
);
2114 ret
= free_io_failure(inode
, failrec
, did_repair
);
2120 * this is a generic handler for readpage errors (default
2121 * readpage_io_failed_hook). if other copies exist, read those and write back
2122 * good data to the failed position. does not investigate in remapping the
2123 * failed extent elsewhere, hoping the device will be smart enough to do this as
2127 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2128 struct page
*page
, u64 start
, u64 end
,
2131 struct io_failure_record
*failrec
= NULL
;
2133 struct extent_map
*em
;
2134 struct inode
*inode
= page
->mapping
->host
;
2135 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2136 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2137 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2139 struct btrfs_io_bio
*btrfs_failed_bio
;
2140 struct btrfs_io_bio
*btrfs_bio
;
2146 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2148 ret
= get_state_private(failure_tree
, start
, &private);
2150 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2153 failrec
->start
= start
;
2154 failrec
->len
= end
- start
+ 1;
2155 failrec
->this_mirror
= 0;
2156 failrec
->bio_flags
= 0;
2157 failrec
->in_validation
= 0;
2159 read_lock(&em_tree
->lock
);
2160 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2162 read_unlock(&em_tree
->lock
);
2167 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
2168 free_extent_map(em
);
2171 read_unlock(&em_tree
->lock
);
2177 logical
= start
- em
->start
;
2178 logical
= em
->block_start
+ logical
;
2179 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2180 logical
= em
->block_start
;
2181 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2182 extent_set_compress_type(&failrec
->bio_flags
,
2185 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2186 "len=%llu\n", logical
, start
, failrec
->len
);
2187 failrec
->logical
= logical
;
2188 free_extent_map(em
);
2190 /* set the bits in the private failure tree */
2191 ret
= set_extent_bits(failure_tree
, start
, end
,
2192 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2194 ret
= set_state_private(failure_tree
, start
,
2195 (u64
)(unsigned long)failrec
);
2196 /* set the bits in the inode's tree */
2198 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2205 failrec
= (struct io_failure_record
*)(unsigned long)private;
2206 pr_debug("bio_readpage_error: (found) logical=%llu, "
2207 "start=%llu, len=%llu, validation=%d\n",
2208 failrec
->logical
, failrec
->start
, failrec
->len
,
2209 failrec
->in_validation
);
2211 * when data can be on disk more than twice, add to failrec here
2212 * (e.g. with a list for failed_mirror) to make
2213 * clean_io_failure() clean all those errors at once.
2216 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2217 failrec
->logical
, failrec
->len
);
2218 if (num_copies
== 1) {
2220 * we only have a single copy of the data, so don't bother with
2221 * all the retry and error correction code that follows. no
2222 * matter what the error is, it is very likely to persist.
2224 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2225 num_copies
, failrec
->this_mirror
, failed_mirror
);
2226 free_io_failure(inode
, failrec
, 0);
2231 * there are two premises:
2232 * a) deliver good data to the caller
2233 * b) correct the bad sectors on disk
2235 if (failed_bio
->bi_vcnt
> 1) {
2237 * to fulfill b), we need to know the exact failing sectors, as
2238 * we don't want to rewrite any more than the failed ones. thus,
2239 * we need separate read requests for the failed bio
2241 * if the following BUG_ON triggers, our validation request got
2242 * merged. we need separate requests for our algorithm to work.
2244 BUG_ON(failrec
->in_validation
);
2245 failrec
->in_validation
= 1;
2246 failrec
->this_mirror
= failed_mirror
;
2247 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2250 * we're ready to fulfill a) and b) alongside. get a good copy
2251 * of the failed sector and if we succeed, we have setup
2252 * everything for repair_io_failure to do the rest for us.
2254 if (failrec
->in_validation
) {
2255 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2256 failrec
->in_validation
= 0;
2257 failrec
->this_mirror
= 0;
2259 failrec
->failed_mirror
= failed_mirror
;
2260 failrec
->this_mirror
++;
2261 if (failrec
->this_mirror
== failed_mirror
)
2262 failrec
->this_mirror
++;
2263 read_mode
= READ_SYNC
;
2266 if (failrec
->this_mirror
> num_copies
) {
2267 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2268 num_copies
, failrec
->this_mirror
, failed_mirror
);
2269 free_io_failure(inode
, failrec
, 0);
2273 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2275 free_io_failure(inode
, failrec
, 0);
2278 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2279 bio
->bi_sector
= failrec
->logical
>> 9;
2280 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2283 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2284 if (btrfs_failed_bio
->csum
) {
2285 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2286 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2288 btrfs_bio
= btrfs_io_bio(bio
);
2289 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2290 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2291 phy_offset
*= csum_size
;
2292 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ phy_offset
,
2296 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2298 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2299 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2300 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2302 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2303 failrec
->this_mirror
,
2304 failrec
->bio_flags
, 0);
2308 /* lots and lots of room for performance fixes in the end_bio funcs */
2310 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2312 int uptodate
= (err
== 0);
2313 struct extent_io_tree
*tree
;
2316 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2318 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2319 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2320 end
, NULL
, uptodate
);
2326 ClearPageUptodate(page
);
2333 * after a writepage IO is done, we need to:
2334 * clear the uptodate bits on error
2335 * clear the writeback bits in the extent tree for this IO
2336 * end_page_writeback if the page has no more pending IO
2338 * Scheduling is not allowed, so the extent state tree is expected
2339 * to have one and only one object corresponding to this IO.
2341 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2343 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2344 struct extent_io_tree
*tree
;
2349 struct page
*page
= bvec
->bv_page
;
2350 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2352 /* We always issue full-page reads, but if some block
2353 * in a page fails to read, blk_update_request() will
2354 * advance bv_offset and adjust bv_len to compensate.
2355 * Print a warning for nonzero offsets, and an error
2356 * if they don't add up to a full page. */
2357 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2358 printk("%s page write in btrfs with offset %u and length %u\n",
2359 bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
2360 ? KERN_ERR
"partial" : KERN_INFO
"incomplete",
2361 bvec
->bv_offset
, bvec
->bv_len
);
2363 start
= page_offset(page
);
2364 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2366 if (--bvec
>= bio
->bi_io_vec
)
2367 prefetchw(&bvec
->bv_page
->flags
);
2369 if (end_extent_writepage(page
, err
, start
, end
))
2372 end_page_writeback(page
);
2373 } while (bvec
>= bio
->bi_io_vec
);
2379 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2382 struct extent_state
*cached
= NULL
;
2383 u64 end
= start
+ len
- 1;
2385 if (uptodate
&& tree
->track_uptodate
)
2386 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2387 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2391 * after a readpage IO is done, we need to:
2392 * clear the uptodate bits on error
2393 * set the uptodate bits if things worked
2394 * set the page up to date if all extents in the tree are uptodate
2395 * clear the lock bit in the extent tree
2396 * unlock the page if there are no other extents locked for it
2398 * Scheduling is not allowed, so the extent state tree is expected
2399 * to have one and only one object corresponding to this IO.
2401 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2403 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2404 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2405 struct bio_vec
*bvec
= bio
->bi_io_vec
;
2406 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2407 struct extent_io_tree
*tree
;
2412 u64 extent_start
= 0;
2421 struct page
*page
= bvec
->bv_page
;
2422 struct inode
*inode
= page
->mapping
->host
;
2424 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2425 "mirror=%lu\n", (u64
)bio
->bi_sector
, err
,
2426 io_bio
->mirror_num
);
2427 tree
= &BTRFS_I(inode
)->io_tree
;
2429 /* We always issue full-page reads, but if some block
2430 * in a page fails to read, blk_update_request() will
2431 * advance bv_offset and adjust bv_len to compensate.
2432 * Print a warning for nonzero offsets, and an error
2433 * if they don't add up to a full page. */
2434 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2435 printk("%s page read in btrfs with offset %u and length %u\n",
2436 bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
2437 ? KERN_ERR
"partial" : KERN_INFO
"incomplete",
2438 bvec
->bv_offset
, bvec
->bv_len
);
2440 start
= page_offset(page
);
2441 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2444 if (++bvec
<= bvec_end
)
2445 prefetchw(&bvec
->bv_page
->flags
);
2447 mirror
= io_bio
->mirror_num
;
2448 if (likely(uptodate
&& tree
->ops
&&
2449 tree
->ops
->readpage_end_io_hook
)) {
2450 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2456 clean_io_failure(start
, page
);
2459 if (likely(uptodate
))
2462 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2463 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2465 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2469 * The generic bio_readpage_error handles errors the
2470 * following way: If possible, new read requests are
2471 * created and submitted and will end up in
2472 * end_bio_extent_readpage as well (if we're lucky, not
2473 * in the !uptodate case). In that case it returns 0 and
2474 * we just go on with the next page in our bio. If it
2475 * can't handle the error it will return -EIO and we
2476 * remain responsible for that page.
2478 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2482 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2489 if (likely(uptodate
)) {
2490 loff_t i_size
= i_size_read(inode
);
2491 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2494 /* Zero out the end if this page straddles i_size */
2495 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2496 if (page
->index
== end_index
&& offset
)
2497 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2498 SetPageUptodate(page
);
2500 ClearPageUptodate(page
);
2506 if (unlikely(!uptodate
)) {
2508 endio_readpage_release_extent(tree
,
2514 endio_readpage_release_extent(tree
, start
,
2515 end
- start
+ 1, 0);
2516 } else if (!extent_len
) {
2517 extent_start
= start
;
2518 extent_len
= end
+ 1 - start
;
2519 } else if (extent_start
+ extent_len
== start
) {
2520 extent_len
+= end
+ 1 - start
;
2522 endio_readpage_release_extent(tree
, extent_start
,
2523 extent_len
, uptodate
);
2524 extent_start
= start
;
2525 extent_len
= end
+ 1 - start
;
2527 } while (bvec
<= bvec_end
);
2530 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2533 io_bio
->end_io(io_bio
, err
);
2538 * this allocates from the btrfs_bioset. We're returning a bio right now
2539 * but you can call btrfs_io_bio for the appropriate container_of magic
2542 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2545 struct btrfs_io_bio
*btrfs_bio
;
2548 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2550 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2551 while (!bio
&& (nr_vecs
/= 2)) {
2552 bio
= bio_alloc_bioset(gfp_flags
,
2553 nr_vecs
, btrfs_bioset
);
2559 bio
->bi_bdev
= bdev
;
2560 bio
->bi_sector
= first_sector
;
2561 btrfs_bio
= btrfs_io_bio(bio
);
2562 btrfs_bio
->csum
= NULL
;
2563 btrfs_bio
->csum_allocated
= NULL
;
2564 btrfs_bio
->end_io
= NULL
;
2569 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2571 return bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2575 /* this also allocates from the btrfs_bioset */
2576 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2578 struct btrfs_io_bio
*btrfs_bio
;
2581 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2583 btrfs_bio
= btrfs_io_bio(bio
);
2584 btrfs_bio
->csum
= NULL
;
2585 btrfs_bio
->csum_allocated
= NULL
;
2586 btrfs_bio
->end_io
= NULL
;
2592 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2593 int mirror_num
, unsigned long bio_flags
)
2596 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2597 struct page
*page
= bvec
->bv_page
;
2598 struct extent_io_tree
*tree
= bio
->bi_private
;
2601 start
= page_offset(page
) + bvec
->bv_offset
;
2603 bio
->bi_private
= NULL
;
2607 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2608 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2609 mirror_num
, bio_flags
, start
);
2611 btrfsic_submit_bio(rw
, bio
);
2613 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2619 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2620 unsigned long offset
, size_t size
, struct bio
*bio
,
2621 unsigned long bio_flags
)
2624 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2625 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2632 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2633 struct page
*page
, sector_t sector
,
2634 size_t size
, unsigned long offset
,
2635 struct block_device
*bdev
,
2636 struct bio
**bio_ret
,
2637 unsigned long max_pages
,
2638 bio_end_io_t end_io_func
,
2640 unsigned long prev_bio_flags
,
2641 unsigned long bio_flags
)
2647 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2648 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2649 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2651 if (bio_ret
&& *bio_ret
) {
2654 contig
= bio
->bi_sector
== sector
;
2656 contig
= bio_end_sector(bio
) == sector
;
2658 if (prev_bio_flags
!= bio_flags
|| !contig
||
2659 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2660 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2661 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2670 if (this_compressed
)
2673 nr
= bio_get_nr_vecs(bdev
);
2675 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2679 bio_add_page(bio
, page
, page_size
, offset
);
2680 bio
->bi_end_io
= end_io_func
;
2681 bio
->bi_private
= tree
;
2686 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2691 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2694 if (!PagePrivate(page
)) {
2695 SetPagePrivate(page
);
2696 page_cache_get(page
);
2697 set_page_private(page
, (unsigned long)eb
);
2699 WARN_ON(page
->private != (unsigned long)eb
);
2703 void set_page_extent_mapped(struct page
*page
)
2705 if (!PagePrivate(page
)) {
2706 SetPagePrivate(page
);
2707 page_cache_get(page
);
2708 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2712 static struct extent_map
*
2713 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2714 u64 start
, u64 len
, get_extent_t
*get_extent
,
2715 struct extent_map
**em_cached
)
2717 struct extent_map
*em
;
2719 if (em_cached
&& *em_cached
) {
2721 if (em
->in_tree
&& start
>= em
->start
&&
2722 start
< extent_map_end(em
)) {
2723 atomic_inc(&em
->refs
);
2727 free_extent_map(em
);
2731 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2732 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2734 atomic_inc(&em
->refs
);
2740 * basic readpage implementation. Locked extent state structs are inserted
2741 * into the tree that are removed when the IO is done (by the end_io
2743 * XXX JDM: This needs looking at to ensure proper page locking
2745 static int __do_readpage(struct extent_io_tree
*tree
,
2747 get_extent_t
*get_extent
,
2748 struct extent_map
**em_cached
,
2749 struct bio
**bio
, int mirror_num
,
2750 unsigned long *bio_flags
, int rw
)
2752 struct inode
*inode
= page
->mapping
->host
;
2753 u64 start
= page_offset(page
);
2754 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2758 u64 last_byte
= i_size_read(inode
);
2762 struct extent_map
*em
;
2763 struct block_device
*bdev
;
2766 int parent_locked
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2767 size_t pg_offset
= 0;
2769 size_t disk_io_size
;
2770 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2771 unsigned long this_bio_flag
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2773 set_page_extent_mapped(page
);
2776 if (!PageUptodate(page
)) {
2777 if (cleancache_get_page(page
) == 0) {
2778 BUG_ON(blocksize
!= PAGE_SIZE
);
2779 unlock_extent(tree
, start
, end
);
2784 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2786 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2789 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2790 userpage
= kmap_atomic(page
);
2791 memset(userpage
+ zero_offset
, 0, iosize
);
2792 flush_dcache_page(page
);
2793 kunmap_atomic(userpage
);
2796 while (cur
<= end
) {
2797 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2799 if (cur
>= last_byte
) {
2801 struct extent_state
*cached
= NULL
;
2803 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2804 userpage
= kmap_atomic(page
);
2805 memset(userpage
+ pg_offset
, 0, iosize
);
2806 flush_dcache_page(page
);
2807 kunmap_atomic(userpage
);
2808 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2811 unlock_extent_cached(tree
, cur
,
2816 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2817 end
- cur
+ 1, get_extent
, em_cached
);
2818 if (IS_ERR_OR_NULL(em
)) {
2821 unlock_extent(tree
, cur
, end
);
2824 extent_offset
= cur
- em
->start
;
2825 BUG_ON(extent_map_end(em
) <= cur
);
2828 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2829 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2830 extent_set_compress_type(&this_bio_flag
,
2834 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2835 cur_end
= min(extent_map_end(em
) - 1, end
);
2836 iosize
= ALIGN(iosize
, blocksize
);
2837 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2838 disk_io_size
= em
->block_len
;
2839 sector
= em
->block_start
>> 9;
2841 sector
= (em
->block_start
+ extent_offset
) >> 9;
2842 disk_io_size
= iosize
;
2845 block_start
= em
->block_start
;
2846 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2847 block_start
= EXTENT_MAP_HOLE
;
2848 free_extent_map(em
);
2851 /* we've found a hole, just zero and go on */
2852 if (block_start
== EXTENT_MAP_HOLE
) {
2854 struct extent_state
*cached
= NULL
;
2856 userpage
= kmap_atomic(page
);
2857 memset(userpage
+ pg_offset
, 0, iosize
);
2858 flush_dcache_page(page
);
2859 kunmap_atomic(userpage
);
2861 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2863 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2866 pg_offset
+= iosize
;
2869 /* the get_extent function already copied into the page */
2870 if (test_range_bit(tree
, cur
, cur_end
,
2871 EXTENT_UPTODATE
, 1, NULL
)) {
2872 check_page_uptodate(tree
, page
);
2874 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2876 pg_offset
+= iosize
;
2879 /* we have an inline extent but it didn't get marked up
2880 * to date. Error out
2882 if (block_start
== EXTENT_MAP_INLINE
) {
2885 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2887 pg_offset
+= iosize
;
2892 ret
= submit_extent_page(rw
, tree
, page
,
2893 sector
, disk_io_size
, pg_offset
,
2895 end_bio_extent_readpage
, mirror_num
,
2900 *bio_flags
= this_bio_flag
;
2904 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2907 pg_offset
+= iosize
;
2911 if (!PageError(page
))
2912 SetPageUptodate(page
);
2918 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
2919 struct page
*pages
[], int nr_pages
,
2921 get_extent_t
*get_extent
,
2922 struct extent_map
**em_cached
,
2923 struct bio
**bio
, int mirror_num
,
2924 unsigned long *bio_flags
, int rw
)
2926 struct inode
*inode
;
2927 struct btrfs_ordered_extent
*ordered
;
2930 inode
= pages
[0]->mapping
->host
;
2932 lock_extent(tree
, start
, end
);
2933 ordered
= btrfs_lookup_ordered_range(inode
, start
,
2937 unlock_extent(tree
, start
, end
);
2938 btrfs_start_ordered_extent(inode
, ordered
, 1);
2939 btrfs_put_ordered_extent(ordered
);
2942 for (index
= 0; index
< nr_pages
; index
++) {
2943 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
2944 mirror_num
, bio_flags
, rw
);
2945 page_cache_release(pages
[index
]);
2949 static void __extent_readpages(struct extent_io_tree
*tree
,
2950 struct page
*pages
[],
2951 int nr_pages
, get_extent_t
*get_extent
,
2952 struct extent_map
**em_cached
,
2953 struct bio
**bio
, int mirror_num
,
2954 unsigned long *bio_flags
, int rw
)
2960 int first_index
= 0;
2962 for (index
= 0; index
< nr_pages
; index
++) {
2963 page_start
= page_offset(pages
[index
]);
2966 end
= start
+ PAGE_CACHE_SIZE
- 1;
2967 first_index
= index
;
2968 } else if (end
+ 1 == page_start
) {
2969 end
+= PAGE_CACHE_SIZE
;
2971 __do_contiguous_readpages(tree
, &pages
[first_index
],
2972 index
- first_index
, start
,
2973 end
, get_extent
, em_cached
,
2974 bio
, mirror_num
, bio_flags
,
2977 end
= start
+ PAGE_CACHE_SIZE
- 1;
2978 first_index
= index
;
2983 __do_contiguous_readpages(tree
, &pages
[first_index
],
2984 index
- first_index
, start
,
2985 end
, get_extent
, em_cached
, bio
,
2986 mirror_num
, bio_flags
, rw
);
2989 static int __extent_read_full_page(struct extent_io_tree
*tree
,
2991 get_extent_t
*get_extent
,
2992 struct bio
**bio
, int mirror_num
,
2993 unsigned long *bio_flags
, int rw
)
2995 struct inode
*inode
= page
->mapping
->host
;
2996 struct btrfs_ordered_extent
*ordered
;
2997 u64 start
= page_offset(page
);
2998 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3002 lock_extent(tree
, start
, end
);
3003 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
3006 unlock_extent(tree
, start
, end
);
3007 btrfs_start_ordered_extent(inode
, ordered
, 1);
3008 btrfs_put_ordered_extent(ordered
);
3011 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3016 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3017 get_extent_t
*get_extent
, int mirror_num
)
3019 struct bio
*bio
= NULL
;
3020 unsigned long bio_flags
= 0;
3023 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3026 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3030 int extent_read_full_page_nolock(struct extent_io_tree
*tree
, struct page
*page
,
3031 get_extent_t
*get_extent
, int mirror_num
)
3033 struct bio
*bio
= NULL
;
3034 unsigned long bio_flags
= EXTENT_BIO_PARENT_LOCKED
;
3037 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, &bio
, mirror_num
,
3040 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3044 static noinline
void update_nr_written(struct page
*page
,
3045 struct writeback_control
*wbc
,
3046 unsigned long nr_written
)
3048 wbc
->nr_to_write
-= nr_written
;
3049 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
3050 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
3051 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
3055 * the writepage semantics are similar to regular writepage. extent
3056 * records are inserted to lock ranges in the tree, and as dirty areas
3057 * are found, they are marked writeback. Then the lock bits are removed
3058 * and the end_io handler clears the writeback ranges
3060 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3063 struct inode
*inode
= page
->mapping
->host
;
3064 struct extent_page_data
*epd
= data
;
3065 struct extent_io_tree
*tree
= epd
->tree
;
3066 u64 start
= page_offset(page
);
3068 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3072 u64 last_byte
= i_size_read(inode
);
3076 struct extent_state
*cached_state
= NULL
;
3077 struct extent_map
*em
;
3078 struct block_device
*bdev
;
3081 size_t pg_offset
= 0;
3083 loff_t i_size
= i_size_read(inode
);
3084 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
3090 unsigned long nr_written
= 0;
3091 bool fill_delalloc
= true;
3093 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3094 write_flags
= WRITE_SYNC
;
3096 write_flags
= WRITE
;
3098 trace___extent_writepage(page
, inode
, wbc
);
3100 WARN_ON(!PageLocked(page
));
3102 ClearPageError(page
);
3104 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
3105 if (page
->index
> end_index
||
3106 (page
->index
== end_index
&& !pg_offset
)) {
3107 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
3112 if (page
->index
== end_index
) {
3115 userpage
= kmap_atomic(page
);
3116 memset(userpage
+ pg_offset
, 0,
3117 PAGE_CACHE_SIZE
- pg_offset
);
3118 kunmap_atomic(userpage
);
3119 flush_dcache_page(page
);
3123 set_page_extent_mapped(page
);
3125 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
3126 fill_delalloc
= false;
3128 delalloc_start
= start
;
3131 if (!epd
->extent_locked
&& fill_delalloc
) {
3132 u64 delalloc_to_write
= 0;
3134 * make sure the wbc mapping index is at least updated
3137 update_nr_written(page
, wbc
, 0);
3139 while (delalloc_end
< page_end
) {
3140 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3145 if (nr_delalloc
== 0) {
3146 delalloc_start
= delalloc_end
+ 1;
3149 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3154 /* File system has been set read-only */
3160 * delalloc_end is already one less than the total
3161 * length, so we don't subtract one from
3164 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3167 delalloc_start
= delalloc_end
+ 1;
3169 if (wbc
->nr_to_write
< delalloc_to_write
) {
3172 if (delalloc_to_write
< thresh
* 2)
3173 thresh
= delalloc_to_write
;
3174 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3178 /* did the fill delalloc function already unlock and start
3184 * we've unlocked the page, so we can't update
3185 * the mapping's writeback index, just update
3188 wbc
->nr_to_write
-= nr_written
;
3192 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3193 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3196 /* Fixup worker will requeue */
3198 wbc
->pages_skipped
++;
3200 redirty_page_for_writepage(wbc
, page
);
3201 update_nr_written(page
, wbc
, nr_written
);
3209 * we don't want to touch the inode after unlocking the page,
3210 * so we update the mapping writeback index now
3212 update_nr_written(page
, wbc
, nr_written
+ 1);
3215 if (last_byte
<= start
) {
3216 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3217 tree
->ops
->writepage_end_io_hook(page
, start
,
3222 blocksize
= inode
->i_sb
->s_blocksize
;
3224 while (cur
<= end
) {
3225 if (cur
>= last_byte
) {
3226 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3227 tree
->ops
->writepage_end_io_hook(page
, cur
,
3231 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3233 if (IS_ERR_OR_NULL(em
)) {
3238 extent_offset
= cur
- em
->start
;
3239 BUG_ON(extent_map_end(em
) <= cur
);
3241 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
3242 iosize
= ALIGN(iosize
, blocksize
);
3243 sector
= (em
->block_start
+ extent_offset
) >> 9;
3245 block_start
= em
->block_start
;
3246 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3247 free_extent_map(em
);
3251 * compressed and inline extents are written through other
3254 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3255 block_start
== EXTENT_MAP_INLINE
) {
3257 * end_io notification does not happen here for
3258 * compressed extents
3260 if (!compressed
&& tree
->ops
&&
3261 tree
->ops
->writepage_end_io_hook
)
3262 tree
->ops
->writepage_end_io_hook(page
, cur
,
3265 else if (compressed
) {
3266 /* we don't want to end_page_writeback on
3267 * a compressed extent. this happens
3274 pg_offset
+= iosize
;
3277 /* leave this out until we have a page_mkwrite call */
3278 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
3279 EXTENT_DIRTY
, 0, NULL
)) {
3281 pg_offset
+= iosize
;
3285 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3286 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3294 unsigned long max_nr
= end_index
+ 1;
3296 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3297 if (!PageWriteback(page
)) {
3298 printk(KERN_ERR
"btrfs warning page %lu not "
3299 "writeback, cur %llu end %llu\n",
3300 page
->index
, cur
, end
);
3303 ret
= submit_extent_page(write_flags
, tree
, page
,
3304 sector
, iosize
, pg_offset
,
3305 bdev
, &epd
->bio
, max_nr
,
3306 end_bio_extent_writepage
,
3312 pg_offset
+= iosize
;
3317 /* make sure the mapping tag for page dirty gets cleared */
3318 set_page_writeback(page
);
3319 end_page_writeback(page
);
3325 /* drop our reference on any cached states */
3326 free_extent_state(cached_state
);
3330 static int eb_wait(void *word
)
3336 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3338 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3339 TASK_UNINTERRUPTIBLE
);
3342 static int lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3343 struct btrfs_fs_info
*fs_info
,
3344 struct extent_page_data
*epd
)
3346 unsigned long i
, num_pages
;
3350 if (!btrfs_try_tree_write_lock(eb
)) {
3352 flush_write_bio(epd
);
3353 btrfs_tree_lock(eb
);
3356 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3357 btrfs_tree_unlock(eb
);
3361 flush_write_bio(epd
);
3365 wait_on_extent_buffer_writeback(eb
);
3366 btrfs_tree_lock(eb
);
3367 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3369 btrfs_tree_unlock(eb
);
3374 * We need to do this to prevent races in people who check if the eb is
3375 * under IO since we can end up having no IO bits set for a short period
3378 spin_lock(&eb
->refs_lock
);
3379 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3380 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3381 spin_unlock(&eb
->refs_lock
);
3382 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3383 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3385 fs_info
->dirty_metadata_batch
);
3388 spin_unlock(&eb
->refs_lock
);
3391 btrfs_tree_unlock(eb
);
3396 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3397 for (i
= 0; i
< num_pages
; i
++) {
3398 struct page
*p
= extent_buffer_page(eb
, i
);
3400 if (!trylock_page(p
)) {
3402 flush_write_bio(epd
);
3412 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3414 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3415 smp_mb__after_clear_bit();
3416 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3419 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3421 int uptodate
= err
== 0;
3422 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
3423 struct extent_buffer
*eb
;
3427 struct page
*page
= bvec
->bv_page
;
3430 eb
= (struct extent_buffer
*)page
->private;
3432 done
= atomic_dec_and_test(&eb
->io_pages
);
3434 if (!uptodate
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3435 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3436 ClearPageUptodate(page
);
3440 end_page_writeback(page
);
3445 end_extent_buffer_writeback(eb
);
3446 } while (bvec
>= bio
->bi_io_vec
);
3452 static int write_one_eb(struct extent_buffer
*eb
,
3453 struct btrfs_fs_info
*fs_info
,
3454 struct writeback_control
*wbc
,
3455 struct extent_page_data
*epd
)
3457 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3458 u64 offset
= eb
->start
;
3459 unsigned long i
, num_pages
;
3460 unsigned long bio_flags
= 0;
3461 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3464 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3465 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3466 atomic_set(&eb
->io_pages
, num_pages
);
3467 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3468 bio_flags
= EXTENT_BIO_TREE_LOG
;
3470 for (i
= 0; i
< num_pages
; i
++) {
3471 struct page
*p
= extent_buffer_page(eb
, i
);
3473 clear_page_dirty_for_io(p
);
3474 set_page_writeback(p
);
3475 ret
= submit_extent_page(rw
, eb
->tree
, p
, offset
>> 9,
3476 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3477 -1, end_bio_extent_buffer_writepage
,
3478 0, epd
->bio_flags
, bio_flags
);
3479 epd
->bio_flags
= bio_flags
;
3481 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3483 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3484 end_extent_buffer_writeback(eb
);
3488 offset
+= PAGE_CACHE_SIZE
;
3489 update_nr_written(p
, wbc
, 1);
3493 if (unlikely(ret
)) {
3494 for (; i
< num_pages
; i
++) {
3495 struct page
*p
= extent_buffer_page(eb
, i
);
3503 int btree_write_cache_pages(struct address_space
*mapping
,
3504 struct writeback_control
*wbc
)
3506 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3507 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3508 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3509 struct extent_page_data epd
= {
3513 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3518 int nr_to_write_done
= 0;
3519 struct pagevec pvec
;
3522 pgoff_t end
; /* Inclusive */
3526 pagevec_init(&pvec
, 0);
3527 if (wbc
->range_cyclic
) {
3528 index
= mapping
->writeback_index
; /* Start from prev offset */
3531 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3532 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3535 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3536 tag
= PAGECACHE_TAG_TOWRITE
;
3538 tag
= PAGECACHE_TAG_DIRTY
;
3540 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3541 tag_pages_for_writeback(mapping
, index
, end
);
3542 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3543 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3544 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3548 for (i
= 0; i
< nr_pages
; i
++) {
3549 struct page
*page
= pvec
.pages
[i
];
3551 if (!PagePrivate(page
))
3554 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3559 spin_lock(&mapping
->private_lock
);
3560 if (!PagePrivate(page
)) {
3561 spin_unlock(&mapping
->private_lock
);
3565 eb
= (struct extent_buffer
*)page
->private;
3568 * Shouldn't happen and normally this would be a BUG_ON
3569 * but no sense in crashing the users box for something
3570 * we can survive anyway.
3573 spin_unlock(&mapping
->private_lock
);
3578 if (eb
== prev_eb
) {
3579 spin_unlock(&mapping
->private_lock
);
3583 ret
= atomic_inc_not_zero(&eb
->refs
);
3584 spin_unlock(&mapping
->private_lock
);
3589 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3591 free_extent_buffer(eb
);
3595 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3598 free_extent_buffer(eb
);
3601 free_extent_buffer(eb
);
3604 * the filesystem may choose to bump up nr_to_write.
3605 * We have to make sure to honor the new nr_to_write
3608 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3610 pagevec_release(&pvec
);
3613 if (!scanned
&& !done
) {
3615 * We hit the last page and there is more work to be done: wrap
3616 * back to the start of the file
3622 flush_write_bio(&epd
);
3627 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3628 * @mapping: address space structure to write
3629 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3630 * @writepage: function called for each page
3631 * @data: data passed to writepage function
3633 * If a page is already under I/O, write_cache_pages() skips it, even
3634 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3635 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3636 * and msync() need to guarantee that all the data which was dirty at the time
3637 * the call was made get new I/O started against them. If wbc->sync_mode is
3638 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3639 * existing IO to complete.
3641 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3642 struct address_space
*mapping
,
3643 struct writeback_control
*wbc
,
3644 writepage_t writepage
, void *data
,
3645 void (*flush_fn
)(void *))
3647 struct inode
*inode
= mapping
->host
;
3650 int nr_to_write_done
= 0;
3651 struct pagevec pvec
;
3654 pgoff_t end
; /* Inclusive */
3659 * We have to hold onto the inode so that ordered extents can do their
3660 * work when the IO finishes. The alternative to this is failing to add
3661 * an ordered extent if the igrab() fails there and that is a huge pain
3662 * to deal with, so instead just hold onto the inode throughout the
3663 * writepages operation. If it fails here we are freeing up the inode
3664 * anyway and we'd rather not waste our time writing out stuff that is
3665 * going to be truncated anyway.
3670 pagevec_init(&pvec
, 0);
3671 if (wbc
->range_cyclic
) {
3672 index
= mapping
->writeback_index
; /* Start from prev offset */
3675 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3676 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3679 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3680 tag
= PAGECACHE_TAG_TOWRITE
;
3682 tag
= PAGECACHE_TAG_DIRTY
;
3684 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3685 tag_pages_for_writeback(mapping
, index
, end
);
3686 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3687 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3688 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3692 for (i
= 0; i
< nr_pages
; i
++) {
3693 struct page
*page
= pvec
.pages
[i
];
3696 * At this point we hold neither mapping->tree_lock nor
3697 * lock on the page itself: the page may be truncated or
3698 * invalidated (changing page->mapping to NULL), or even
3699 * swizzled back from swapper_space to tmpfs file
3702 if (!trylock_page(page
)) {
3707 if (unlikely(page
->mapping
!= mapping
)) {
3712 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3718 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3719 if (PageWriteback(page
))
3721 wait_on_page_writeback(page
);
3724 if (PageWriteback(page
) ||
3725 !clear_page_dirty_for_io(page
)) {
3730 ret
= (*writepage
)(page
, wbc
, data
);
3732 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3740 * the filesystem may choose to bump up nr_to_write.
3741 * We have to make sure to honor the new nr_to_write
3744 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3746 pagevec_release(&pvec
);
3749 if (!scanned
&& !done
) {
3751 * We hit the last page and there is more work to be done: wrap
3752 * back to the start of the file
3758 btrfs_add_delayed_iput(inode
);
3762 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3771 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
3772 BUG_ON(ret
< 0); /* -ENOMEM */
3777 static noinline
void flush_write_bio(void *data
)
3779 struct extent_page_data
*epd
= data
;
3780 flush_epd_write_bio(epd
);
3783 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3784 get_extent_t
*get_extent
,
3785 struct writeback_control
*wbc
)
3788 struct extent_page_data epd
= {
3791 .get_extent
= get_extent
,
3793 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3797 ret
= __extent_writepage(page
, wbc
, &epd
);
3799 flush_epd_write_bio(&epd
);
3803 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3804 u64 start
, u64 end
, get_extent_t
*get_extent
,
3808 struct address_space
*mapping
= inode
->i_mapping
;
3810 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3813 struct extent_page_data epd
= {
3816 .get_extent
= get_extent
,
3818 .sync_io
= mode
== WB_SYNC_ALL
,
3821 struct writeback_control wbc_writepages
= {
3823 .nr_to_write
= nr_pages
* 2,
3824 .range_start
= start
,
3825 .range_end
= end
+ 1,
3828 while (start
<= end
) {
3829 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3830 if (clear_page_dirty_for_io(page
))
3831 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3833 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3834 tree
->ops
->writepage_end_io_hook(page
, start
,
3835 start
+ PAGE_CACHE_SIZE
- 1,
3839 page_cache_release(page
);
3840 start
+= PAGE_CACHE_SIZE
;
3843 flush_epd_write_bio(&epd
);
3847 int extent_writepages(struct extent_io_tree
*tree
,
3848 struct address_space
*mapping
,
3849 get_extent_t
*get_extent
,
3850 struct writeback_control
*wbc
)
3853 struct extent_page_data epd
= {
3856 .get_extent
= get_extent
,
3858 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3862 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3863 __extent_writepage
, &epd
,
3865 flush_epd_write_bio(&epd
);
3869 int extent_readpages(struct extent_io_tree
*tree
,
3870 struct address_space
*mapping
,
3871 struct list_head
*pages
, unsigned nr_pages
,
3872 get_extent_t get_extent
)
3874 struct bio
*bio
= NULL
;
3876 unsigned long bio_flags
= 0;
3877 struct page
*pagepool
[16];
3879 struct extent_map
*em_cached
= NULL
;
3882 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3883 page
= list_entry(pages
->prev
, struct page
, lru
);
3885 prefetchw(&page
->flags
);
3886 list_del(&page
->lru
);
3887 if (add_to_page_cache_lru(page
, mapping
,
3888 page
->index
, GFP_NOFS
)) {
3889 page_cache_release(page
);
3893 pagepool
[nr
++] = page
;
3894 if (nr
< ARRAY_SIZE(pagepool
))
3896 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3897 &bio
, 0, &bio_flags
, READ
);
3901 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3902 &bio
, 0, &bio_flags
, READ
);
3905 free_extent_map(em_cached
);
3907 BUG_ON(!list_empty(pages
));
3909 return submit_one_bio(READ
, bio
, 0, bio_flags
);
3914 * basic invalidatepage code, this waits on any locked or writeback
3915 * ranges corresponding to the page, and then deletes any extent state
3916 * records from the tree
3918 int extent_invalidatepage(struct extent_io_tree
*tree
,
3919 struct page
*page
, unsigned long offset
)
3921 struct extent_state
*cached_state
= NULL
;
3922 u64 start
= page_offset(page
);
3923 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3924 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
3926 start
+= ALIGN(offset
, blocksize
);
3930 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
3931 wait_on_page_writeback(page
);
3932 clear_extent_bit(tree
, start
, end
,
3933 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3934 EXTENT_DO_ACCOUNTING
,
3935 1, 1, &cached_state
, GFP_NOFS
);
3940 * a helper for releasepage, this tests for areas of the page that
3941 * are locked or under IO and drops the related state bits if it is safe
3944 static int try_release_extent_state(struct extent_map_tree
*map
,
3945 struct extent_io_tree
*tree
,
3946 struct page
*page
, gfp_t mask
)
3948 u64 start
= page_offset(page
);
3949 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3952 if (test_range_bit(tree
, start
, end
,
3953 EXTENT_IOBITS
, 0, NULL
))
3956 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
3959 * at this point we can safely clear everything except the
3960 * locked bit and the nodatasum bit
3962 ret
= clear_extent_bit(tree
, start
, end
,
3963 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
3966 /* if clear_extent_bit failed for enomem reasons,
3967 * we can't allow the release to continue.
3978 * a helper for releasepage. As long as there are no locked extents
3979 * in the range corresponding to the page, both state records and extent
3980 * map records are removed
3982 int try_release_extent_mapping(struct extent_map_tree
*map
,
3983 struct extent_io_tree
*tree
, struct page
*page
,
3986 struct extent_map
*em
;
3987 u64 start
= page_offset(page
);
3988 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3990 if ((mask
& __GFP_WAIT
) &&
3991 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
3993 while (start
<= end
) {
3994 len
= end
- start
+ 1;
3995 write_lock(&map
->lock
);
3996 em
= lookup_extent_mapping(map
, start
, len
);
3998 write_unlock(&map
->lock
);
4001 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4002 em
->start
!= start
) {
4003 write_unlock(&map
->lock
);
4004 free_extent_map(em
);
4007 if (!test_range_bit(tree
, em
->start
,
4008 extent_map_end(em
) - 1,
4009 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4011 remove_extent_mapping(map
, em
);
4012 /* once for the rb tree */
4013 free_extent_map(em
);
4015 start
= extent_map_end(em
);
4016 write_unlock(&map
->lock
);
4019 free_extent_map(em
);
4022 return try_release_extent_state(map
, tree
, page
, mask
);
4026 * helper function for fiemap, which doesn't want to see any holes.
4027 * This maps until we find something past 'last'
4029 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4032 get_extent_t
*get_extent
)
4034 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4035 struct extent_map
*em
;
4042 len
= last
- offset
;
4045 len
= ALIGN(len
, sectorsize
);
4046 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4047 if (IS_ERR_OR_NULL(em
))
4050 /* if this isn't a hole return it */
4051 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4052 em
->block_start
!= EXTENT_MAP_HOLE
) {
4056 /* this is a hole, advance to the next extent */
4057 offset
= extent_map_end(em
);
4058 free_extent_map(em
);
4065 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4066 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4070 u64 max
= start
+ len
;
4074 u64 last_for_get_extent
= 0;
4076 u64 isize
= i_size_read(inode
);
4077 struct btrfs_key found_key
;
4078 struct extent_map
*em
= NULL
;
4079 struct extent_state
*cached_state
= NULL
;
4080 struct btrfs_path
*path
;
4081 struct btrfs_file_extent_item
*item
;
4086 unsigned long emflags
;
4091 path
= btrfs_alloc_path();
4094 path
->leave_spinning
= 1;
4096 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
4097 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
4100 * lookup the last file extent. We're not using i_size here
4101 * because there might be preallocation past i_size
4103 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
4104 path
, btrfs_ino(inode
), -1, 0);
4106 btrfs_free_path(path
);
4111 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4112 struct btrfs_file_extent_item
);
4113 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4114 found_type
= btrfs_key_type(&found_key
);
4116 /* No extents, but there might be delalloc bits */
4117 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4118 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4119 /* have to trust i_size as the end */
4121 last_for_get_extent
= isize
;
4124 * remember the start of the last extent. There are a
4125 * bunch of different factors that go into the length of the
4126 * extent, so its much less complex to remember where it started
4128 last
= found_key
.offset
;
4129 last_for_get_extent
= last
+ 1;
4131 btrfs_free_path(path
);
4134 * we might have some extents allocated but more delalloc past those
4135 * extents. so, we trust isize unless the start of the last extent is
4140 last_for_get_extent
= isize
;
4143 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1, 0,
4146 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4156 u64 offset_in_extent
= 0;
4158 /* break if the extent we found is outside the range */
4159 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4163 * get_extent may return an extent that starts before our
4164 * requested range. We have to make sure the ranges
4165 * we return to fiemap always move forward and don't
4166 * overlap, so adjust the offsets here
4168 em_start
= max(em
->start
, off
);
4171 * record the offset from the start of the extent
4172 * for adjusting the disk offset below. Only do this if the
4173 * extent isn't compressed since our in ram offset may be past
4174 * what we have actually allocated on disk.
4176 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4177 offset_in_extent
= em_start
- em
->start
;
4178 em_end
= extent_map_end(em
);
4179 em_len
= em_end
- em_start
;
4180 emflags
= em
->flags
;
4185 * bump off for our next call to get_extent
4187 off
= extent_map_end(em
);
4191 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4193 flags
|= FIEMAP_EXTENT_LAST
;
4194 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4195 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4196 FIEMAP_EXTENT_NOT_ALIGNED
);
4197 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4198 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4199 FIEMAP_EXTENT_UNKNOWN
);
4201 disko
= em
->block_start
+ offset_in_extent
;
4203 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4204 flags
|= FIEMAP_EXTENT_ENCODED
;
4206 free_extent_map(em
);
4208 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4209 (last
== (u64
)-1 && isize
<= em_end
)) {
4210 flags
|= FIEMAP_EXTENT_LAST
;
4214 /* now scan forward to see if this is really the last extent. */
4215 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4222 flags
|= FIEMAP_EXTENT_LAST
;
4225 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4231 free_extent_map(em
);
4233 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4234 &cached_state
, GFP_NOFS
);
4238 static void __free_extent_buffer(struct extent_buffer
*eb
)
4240 btrfs_leak_debug_del(&eb
->leak_list
);
4241 kmem_cache_free(extent_buffer_cache
, eb
);
4244 static int extent_buffer_under_io(struct extent_buffer
*eb
)
4246 return (atomic_read(&eb
->io_pages
) ||
4247 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4248 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4252 * Helper for releasing extent buffer page.
4254 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4255 unsigned long start_idx
)
4257 unsigned long index
;
4258 unsigned long num_pages
;
4260 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4262 BUG_ON(extent_buffer_under_io(eb
));
4264 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4265 index
= start_idx
+ num_pages
;
4266 if (start_idx
>= index
)
4271 page
= extent_buffer_page(eb
, index
);
4272 if (page
&& mapped
) {
4273 spin_lock(&page
->mapping
->private_lock
);
4275 * We do this since we'll remove the pages after we've
4276 * removed the eb from the radix tree, so we could race
4277 * and have this page now attached to the new eb. So
4278 * only clear page_private if it's still connected to
4281 if (PagePrivate(page
) &&
4282 page
->private == (unsigned long)eb
) {
4283 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4284 BUG_ON(PageDirty(page
));
4285 BUG_ON(PageWriteback(page
));
4287 * We need to make sure we haven't be attached
4290 ClearPagePrivate(page
);
4291 set_page_private(page
, 0);
4292 /* One for the page private */
4293 page_cache_release(page
);
4295 spin_unlock(&page
->mapping
->private_lock
);
4299 /* One for when we alloced the page */
4300 page_cache_release(page
);
4302 } while (index
!= start_idx
);
4306 * Helper for releasing the extent buffer.
4308 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4310 btrfs_release_extent_buffer_page(eb
, 0);
4311 __free_extent_buffer(eb
);
4314 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
4319 struct extent_buffer
*eb
= NULL
;
4321 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
4328 rwlock_init(&eb
->lock
);
4329 atomic_set(&eb
->write_locks
, 0);
4330 atomic_set(&eb
->read_locks
, 0);
4331 atomic_set(&eb
->blocking_readers
, 0);
4332 atomic_set(&eb
->blocking_writers
, 0);
4333 atomic_set(&eb
->spinning_readers
, 0);
4334 atomic_set(&eb
->spinning_writers
, 0);
4335 eb
->lock_nested
= 0;
4336 init_waitqueue_head(&eb
->write_lock_wq
);
4337 init_waitqueue_head(&eb
->read_lock_wq
);
4339 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4341 spin_lock_init(&eb
->refs_lock
);
4342 atomic_set(&eb
->refs
, 1);
4343 atomic_set(&eb
->io_pages
, 0);
4346 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4348 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4349 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4350 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4355 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4359 struct extent_buffer
*new;
4360 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4362 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_NOFS
);
4366 for (i
= 0; i
< num_pages
; i
++) {
4367 p
= alloc_page(GFP_NOFS
);
4369 btrfs_release_extent_buffer(new);
4372 attach_extent_buffer_page(new, p
);
4373 WARN_ON(PageDirty(p
));
4378 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4379 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4380 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4385 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
4387 struct extent_buffer
*eb
;
4388 unsigned long num_pages
= num_extent_pages(0, len
);
4391 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_NOFS
);
4395 for (i
= 0; i
< num_pages
; i
++) {
4396 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4400 set_extent_buffer_uptodate(eb
);
4401 btrfs_set_header_nritems(eb
, 0);
4402 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4407 __free_page(eb
->pages
[i
- 1]);
4408 __free_extent_buffer(eb
);
4412 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4415 /* the ref bit is tricky. We have to make sure it is set
4416 * if we have the buffer dirty. Otherwise the
4417 * code to free a buffer can end up dropping a dirty
4420 * Once the ref bit is set, it won't go away while the
4421 * buffer is dirty or in writeback, and it also won't
4422 * go away while we have the reference count on the
4425 * We can't just set the ref bit without bumping the
4426 * ref on the eb because free_extent_buffer might
4427 * see the ref bit and try to clear it. If this happens
4428 * free_extent_buffer might end up dropping our original
4429 * ref by mistake and freeing the page before we are able
4430 * to add one more ref.
4432 * So bump the ref count first, then set the bit. If someone
4433 * beat us to it, drop the ref we added.
4435 refs
= atomic_read(&eb
->refs
);
4436 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4439 spin_lock(&eb
->refs_lock
);
4440 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4441 atomic_inc(&eb
->refs
);
4442 spin_unlock(&eb
->refs_lock
);
4445 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
)
4447 unsigned long num_pages
, i
;
4449 check_buffer_tree_ref(eb
);
4451 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4452 for (i
= 0; i
< num_pages
; i
++) {
4453 struct page
*p
= extent_buffer_page(eb
, i
);
4454 mark_page_accessed(p
);
4458 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
4459 u64 start
, unsigned long len
)
4461 unsigned long num_pages
= num_extent_pages(start
, len
);
4463 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4464 struct extent_buffer
*eb
;
4465 struct extent_buffer
*exists
= NULL
;
4467 struct address_space
*mapping
= tree
->mapping
;
4472 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4473 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4475 mark_extent_buffer_accessed(eb
);
4480 eb
= __alloc_extent_buffer(tree
, start
, len
, GFP_NOFS
);
4484 for (i
= 0; i
< num_pages
; i
++, index
++) {
4485 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4489 spin_lock(&mapping
->private_lock
);
4490 if (PagePrivate(p
)) {
4492 * We could have already allocated an eb for this page
4493 * and attached one so lets see if we can get a ref on
4494 * the existing eb, and if we can we know it's good and
4495 * we can just return that one, else we know we can just
4496 * overwrite page->private.
4498 exists
= (struct extent_buffer
*)p
->private;
4499 if (atomic_inc_not_zero(&exists
->refs
)) {
4500 spin_unlock(&mapping
->private_lock
);
4502 page_cache_release(p
);
4503 mark_extent_buffer_accessed(exists
);
4508 * Do this so attach doesn't complain and we need to
4509 * drop the ref the old guy had.
4511 ClearPagePrivate(p
);
4512 WARN_ON(PageDirty(p
));
4513 page_cache_release(p
);
4515 attach_extent_buffer_page(eb
, p
);
4516 spin_unlock(&mapping
->private_lock
);
4517 WARN_ON(PageDirty(p
));
4518 mark_page_accessed(p
);
4520 if (!PageUptodate(p
))
4524 * see below about how we avoid a nasty race with release page
4525 * and why we unlock later
4529 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4531 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4535 spin_lock(&tree
->buffer_lock
);
4536 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
4537 if (ret
== -EEXIST
) {
4538 exists
= radix_tree_lookup(&tree
->buffer
,
4539 start
>> PAGE_CACHE_SHIFT
);
4540 if (!atomic_inc_not_zero(&exists
->refs
)) {
4541 spin_unlock(&tree
->buffer_lock
);
4542 radix_tree_preload_end();
4546 spin_unlock(&tree
->buffer_lock
);
4547 radix_tree_preload_end();
4548 mark_extent_buffer_accessed(exists
);
4551 /* add one reference for the tree */
4552 check_buffer_tree_ref(eb
);
4553 spin_unlock(&tree
->buffer_lock
);
4554 radix_tree_preload_end();
4557 * there is a race where release page may have
4558 * tried to find this extent buffer in the radix
4559 * but failed. It will tell the VM it is safe to
4560 * reclaim the, and it will clear the page private bit.
4561 * We must make sure to set the page private bit properly
4562 * after the extent buffer is in the radix tree so
4563 * it doesn't get lost
4565 SetPageChecked(eb
->pages
[0]);
4566 for (i
= 1; i
< num_pages
; i
++) {
4567 p
= extent_buffer_page(eb
, i
);
4568 ClearPageChecked(p
);
4571 unlock_page(eb
->pages
[0]);
4575 for (i
= 0; i
< num_pages
; i
++) {
4577 unlock_page(eb
->pages
[i
]);
4580 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4581 btrfs_release_extent_buffer(eb
);
4585 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
4586 u64 start
, unsigned long len
)
4588 struct extent_buffer
*eb
;
4591 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4592 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4594 mark_extent_buffer_accessed(eb
);
4602 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4604 struct extent_buffer
*eb
=
4605 container_of(head
, struct extent_buffer
, rcu_head
);
4607 __free_extent_buffer(eb
);
4610 /* Expects to have eb->eb_lock already held */
4611 static int release_extent_buffer(struct extent_buffer
*eb
)
4613 WARN_ON(atomic_read(&eb
->refs
) == 0);
4614 if (atomic_dec_and_test(&eb
->refs
)) {
4615 if (test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
)) {
4616 spin_unlock(&eb
->refs_lock
);
4618 struct extent_io_tree
*tree
= eb
->tree
;
4620 spin_unlock(&eb
->refs_lock
);
4622 spin_lock(&tree
->buffer_lock
);
4623 radix_tree_delete(&tree
->buffer
,
4624 eb
->start
>> PAGE_CACHE_SHIFT
);
4625 spin_unlock(&tree
->buffer_lock
);
4628 /* Should be safe to release our pages at this point */
4629 btrfs_release_extent_buffer_page(eb
, 0);
4630 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4633 spin_unlock(&eb
->refs_lock
);
4638 void free_extent_buffer(struct extent_buffer
*eb
)
4646 refs
= atomic_read(&eb
->refs
);
4649 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
4654 spin_lock(&eb
->refs_lock
);
4655 if (atomic_read(&eb
->refs
) == 2 &&
4656 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4657 atomic_dec(&eb
->refs
);
4659 if (atomic_read(&eb
->refs
) == 2 &&
4660 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4661 !extent_buffer_under_io(eb
) &&
4662 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4663 atomic_dec(&eb
->refs
);
4666 * I know this is terrible, but it's temporary until we stop tracking
4667 * the uptodate bits and such for the extent buffers.
4669 release_extent_buffer(eb
);
4672 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4677 spin_lock(&eb
->refs_lock
);
4678 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4680 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4681 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4682 atomic_dec(&eb
->refs
);
4683 release_extent_buffer(eb
);
4686 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4689 unsigned long num_pages
;
4692 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4694 for (i
= 0; i
< num_pages
; i
++) {
4695 page
= extent_buffer_page(eb
, i
);
4696 if (!PageDirty(page
))
4700 WARN_ON(!PagePrivate(page
));
4702 clear_page_dirty_for_io(page
);
4703 spin_lock_irq(&page
->mapping
->tree_lock
);
4704 if (!PageDirty(page
)) {
4705 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4707 PAGECACHE_TAG_DIRTY
);
4709 spin_unlock_irq(&page
->mapping
->tree_lock
);
4710 ClearPageError(page
);
4713 WARN_ON(atomic_read(&eb
->refs
) == 0);
4716 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4719 unsigned long num_pages
;
4722 check_buffer_tree_ref(eb
);
4724 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4726 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4727 WARN_ON(atomic_read(&eb
->refs
) == 0);
4728 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4730 for (i
= 0; i
< num_pages
; i
++)
4731 set_page_dirty(extent_buffer_page(eb
, i
));
4735 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4739 unsigned long num_pages
;
4741 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4742 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4743 for (i
= 0; i
< num_pages
; i
++) {
4744 page
= extent_buffer_page(eb
, i
);
4746 ClearPageUptodate(page
);
4751 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4755 unsigned long num_pages
;
4757 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4758 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4759 for (i
= 0; i
< num_pages
; i
++) {
4760 page
= extent_buffer_page(eb
, i
);
4761 SetPageUptodate(page
);
4766 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4768 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4771 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4772 struct extent_buffer
*eb
, u64 start
, int wait
,
4773 get_extent_t
*get_extent
, int mirror_num
)
4776 unsigned long start_i
;
4780 int locked_pages
= 0;
4781 int all_uptodate
= 1;
4782 unsigned long num_pages
;
4783 unsigned long num_reads
= 0;
4784 struct bio
*bio
= NULL
;
4785 unsigned long bio_flags
= 0;
4787 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4791 WARN_ON(start
< eb
->start
);
4792 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4793 (eb
->start
>> PAGE_CACHE_SHIFT
);
4798 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4799 for (i
= start_i
; i
< num_pages
; i
++) {
4800 page
= extent_buffer_page(eb
, i
);
4801 if (wait
== WAIT_NONE
) {
4802 if (!trylock_page(page
))
4808 if (!PageUptodate(page
)) {
4815 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4819 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4820 eb
->read_mirror
= 0;
4821 atomic_set(&eb
->io_pages
, num_reads
);
4822 for (i
= start_i
; i
< num_pages
; i
++) {
4823 page
= extent_buffer_page(eb
, i
);
4824 if (!PageUptodate(page
)) {
4825 ClearPageError(page
);
4826 err
= __extent_read_full_page(tree
, page
,
4828 mirror_num
, &bio_flags
,
4838 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
4844 if (ret
|| wait
!= WAIT_COMPLETE
)
4847 for (i
= start_i
; i
< num_pages
; i
++) {
4848 page
= extent_buffer_page(eb
, i
);
4849 wait_on_page_locked(page
);
4850 if (!PageUptodate(page
))
4858 while (locked_pages
> 0) {
4859 page
= extent_buffer_page(eb
, i
);
4867 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
4868 unsigned long start
,
4875 char *dst
= (char *)dstv
;
4876 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4877 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4879 WARN_ON(start
> eb
->len
);
4880 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4882 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
4885 page
= extent_buffer_page(eb
, i
);
4887 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4888 kaddr
= page_address(page
);
4889 memcpy(dst
, kaddr
+ offset
, cur
);
4898 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
4899 unsigned long min_len
, char **map
,
4900 unsigned long *map_start
,
4901 unsigned long *map_len
)
4903 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
4906 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4907 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4908 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
4915 offset
= start_offset
;
4919 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
4922 if (start
+ min_len
> eb
->len
) {
4923 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
4925 eb
->start
, eb
->len
, start
, min_len
);
4929 p
= extent_buffer_page(eb
, i
);
4930 kaddr
= page_address(p
);
4931 *map
= kaddr
+ offset
;
4932 *map_len
= PAGE_CACHE_SIZE
- offset
;
4936 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
4937 unsigned long start
,
4944 char *ptr
= (char *)ptrv
;
4945 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4946 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4949 WARN_ON(start
> eb
->len
);
4950 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4952 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
4955 page
= extent_buffer_page(eb
, i
);
4957 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4959 kaddr
= page_address(page
);
4960 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
4972 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
4973 unsigned long start
, unsigned long len
)
4979 char *src
= (char *)srcv
;
4980 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4981 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4983 WARN_ON(start
> eb
->len
);
4984 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4986 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
4989 page
= extent_buffer_page(eb
, i
);
4990 WARN_ON(!PageUptodate(page
));
4992 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4993 kaddr
= page_address(page
);
4994 memcpy(kaddr
+ offset
, src
, cur
);
5003 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5004 unsigned long start
, unsigned long len
)
5010 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5011 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5013 WARN_ON(start
> eb
->len
);
5014 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5016 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5019 page
= extent_buffer_page(eb
, i
);
5020 WARN_ON(!PageUptodate(page
));
5022 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5023 kaddr
= page_address(page
);
5024 memset(kaddr
+ offset
, c
, cur
);
5032 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5033 unsigned long dst_offset
, unsigned long src_offset
,
5036 u64 dst_len
= dst
->len
;
5041 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5042 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5044 WARN_ON(src
->len
!= dst_len
);
5046 offset
= (start_offset
+ dst_offset
) &
5047 (PAGE_CACHE_SIZE
- 1);
5050 page
= extent_buffer_page(dst
, i
);
5051 WARN_ON(!PageUptodate(page
));
5053 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
5055 kaddr
= page_address(page
);
5056 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5065 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
5066 unsigned long dst_off
, unsigned long src_off
,
5069 char *dst_kaddr
= page_address(dst_page
);
5070 if (dst_page
== src_page
) {
5071 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
5073 char *src_kaddr
= page_address(src_page
);
5074 char *p
= dst_kaddr
+ dst_off
+ len
;
5075 char *s
= src_kaddr
+ src_off
+ len
;
5082 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5084 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5085 return distance
< len
;
5088 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5089 unsigned long dst_off
, unsigned long src_off
,
5092 char *dst_kaddr
= page_address(dst_page
);
5094 int must_memmove
= 0;
5096 if (dst_page
!= src_page
) {
5097 src_kaddr
= page_address(src_page
);
5099 src_kaddr
= dst_kaddr
;
5100 if (areas_overlap(src_off
, dst_off
, len
))
5105 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5107 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5110 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5111 unsigned long src_offset
, unsigned long len
)
5114 size_t dst_off_in_page
;
5115 size_t src_off_in_page
;
5116 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5117 unsigned long dst_i
;
5118 unsigned long src_i
;
5120 if (src_offset
+ len
> dst
->len
) {
5121 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
5122 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
5125 if (dst_offset
+ len
> dst
->len
) {
5126 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
5127 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
5132 dst_off_in_page
= (start_offset
+ dst_offset
) &
5133 (PAGE_CACHE_SIZE
- 1);
5134 src_off_in_page
= (start_offset
+ src_offset
) &
5135 (PAGE_CACHE_SIZE
- 1);
5137 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5138 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5140 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5142 cur
= min_t(unsigned long, cur
,
5143 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5145 copy_pages(extent_buffer_page(dst
, dst_i
),
5146 extent_buffer_page(dst
, src_i
),
5147 dst_off_in_page
, src_off_in_page
, cur
);
5155 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5156 unsigned long src_offset
, unsigned long len
)
5159 size_t dst_off_in_page
;
5160 size_t src_off_in_page
;
5161 unsigned long dst_end
= dst_offset
+ len
- 1;
5162 unsigned long src_end
= src_offset
+ len
- 1;
5163 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5164 unsigned long dst_i
;
5165 unsigned long src_i
;
5167 if (src_offset
+ len
> dst
->len
) {
5168 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
5169 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
5172 if (dst_offset
+ len
> dst
->len
) {
5173 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
5174 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
5177 if (dst_offset
< src_offset
) {
5178 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5182 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5183 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5185 dst_off_in_page
= (start_offset
+ dst_end
) &
5186 (PAGE_CACHE_SIZE
- 1);
5187 src_off_in_page
= (start_offset
+ src_end
) &
5188 (PAGE_CACHE_SIZE
- 1);
5190 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5191 cur
= min(cur
, dst_off_in_page
+ 1);
5192 move_pages(extent_buffer_page(dst
, dst_i
),
5193 extent_buffer_page(dst
, src_i
),
5194 dst_off_in_page
- cur
+ 1,
5195 src_off_in_page
- cur
+ 1, cur
);
5203 int try_release_extent_buffer(struct page
*page
)
5205 struct extent_buffer
*eb
;
5208 * We need to make sure noboody is attaching this page to an eb right
5211 spin_lock(&page
->mapping
->private_lock
);
5212 if (!PagePrivate(page
)) {
5213 spin_unlock(&page
->mapping
->private_lock
);
5217 eb
= (struct extent_buffer
*)page
->private;
5221 * This is a little awful but should be ok, we need to make sure that
5222 * the eb doesn't disappear out from under us while we're looking at
5225 spin_lock(&eb
->refs_lock
);
5226 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5227 spin_unlock(&eb
->refs_lock
);
5228 spin_unlock(&page
->mapping
->private_lock
);
5231 spin_unlock(&page
->mapping
->private_lock
);
5234 * If tree ref isn't set then we know the ref on this eb is a real ref,
5235 * so just return, this page will likely be freed soon anyway.
5237 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5238 spin_unlock(&eb
->refs_lock
);
5242 return release_extent_buffer(eb
);