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
);
1648 cached_state
= NULL
;
1650 max_bytes
= PAGE_CACHE_SIZE
;
1658 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1660 /* step three, lock the state bits for the whole range */
1661 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1663 /* then test to make sure it is all still delalloc */
1664 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1665 EXTENT_DELALLOC
, 1, cached_state
);
1667 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1668 &cached_state
, GFP_NOFS
);
1669 __unlock_for_delalloc(inode
, locked_page
,
1670 delalloc_start
, delalloc_end
);
1674 free_extent_state(cached_state
);
1675 *start
= delalloc_start
;
1676 *end
= delalloc_end
;
1681 int extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1682 struct page
*locked_page
,
1683 unsigned long clear_bits
,
1684 unsigned long page_ops
)
1686 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1688 struct page
*pages
[16];
1689 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1690 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1691 unsigned long nr_pages
= end_index
- index
+ 1;
1694 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1698 while (nr_pages
> 0) {
1699 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1700 min_t(unsigned long,
1701 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1702 for (i
= 0; i
< ret
; i
++) {
1704 if (page_ops
& PAGE_SET_PRIVATE2
)
1705 SetPagePrivate2(pages
[i
]);
1707 if (pages
[i
] == locked_page
) {
1708 page_cache_release(pages
[i
]);
1711 if (page_ops
& PAGE_CLEAR_DIRTY
)
1712 clear_page_dirty_for_io(pages
[i
]);
1713 if (page_ops
& PAGE_SET_WRITEBACK
)
1714 set_page_writeback(pages
[i
]);
1715 if (page_ops
& PAGE_END_WRITEBACK
)
1716 end_page_writeback(pages
[i
]);
1717 if (page_ops
& PAGE_UNLOCK
)
1718 unlock_page(pages
[i
]);
1719 page_cache_release(pages
[i
]);
1729 * count the number of bytes in the tree that have a given bit(s)
1730 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1731 * cached. The total number found is returned.
1733 u64
count_range_bits(struct extent_io_tree
*tree
,
1734 u64
*start
, u64 search_end
, u64 max_bytes
,
1735 unsigned long bits
, int contig
)
1737 struct rb_node
*node
;
1738 struct extent_state
*state
;
1739 u64 cur_start
= *start
;
1740 u64 total_bytes
= 0;
1744 if (search_end
<= cur_start
) {
1749 spin_lock(&tree
->lock
);
1750 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1751 total_bytes
= tree
->dirty_bytes
;
1755 * this search will find all the extents that end after
1758 node
= tree_search(tree
, cur_start
);
1763 state
= rb_entry(node
, struct extent_state
, rb_node
);
1764 if (state
->start
> search_end
)
1766 if (contig
&& found
&& state
->start
> last
+ 1)
1768 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1769 total_bytes
+= min(search_end
, state
->end
) + 1 -
1770 max(cur_start
, state
->start
);
1771 if (total_bytes
>= max_bytes
)
1774 *start
= max(cur_start
, state
->start
);
1778 } else if (contig
&& found
) {
1781 node
= rb_next(node
);
1786 spin_unlock(&tree
->lock
);
1791 * set the private field for a given byte offset in the tree. If there isn't
1792 * an extent_state there already, this does nothing.
1794 static int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1796 struct rb_node
*node
;
1797 struct extent_state
*state
;
1800 spin_lock(&tree
->lock
);
1802 * this search will find all the extents that end after
1805 node
= tree_search(tree
, start
);
1810 state
= rb_entry(node
, struct extent_state
, rb_node
);
1811 if (state
->start
!= start
) {
1815 state
->private = private;
1817 spin_unlock(&tree
->lock
);
1821 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1823 struct rb_node
*node
;
1824 struct extent_state
*state
;
1827 spin_lock(&tree
->lock
);
1829 * this search will find all the extents that end after
1832 node
= tree_search(tree
, start
);
1837 state
= rb_entry(node
, struct extent_state
, rb_node
);
1838 if (state
->start
!= start
) {
1842 *private = state
->private;
1844 spin_unlock(&tree
->lock
);
1849 * searches a range in the state tree for a given mask.
1850 * If 'filled' == 1, this returns 1 only if every extent in the tree
1851 * has the bits set. Otherwise, 1 is returned if any bit in the
1852 * range is found set.
1854 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1855 unsigned long bits
, int filled
, struct extent_state
*cached
)
1857 struct extent_state
*state
= NULL
;
1858 struct rb_node
*node
;
1861 spin_lock(&tree
->lock
);
1862 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1863 cached
->end
> start
)
1864 node
= &cached
->rb_node
;
1866 node
= tree_search(tree
, start
);
1867 while (node
&& start
<= end
) {
1868 state
= rb_entry(node
, struct extent_state
, rb_node
);
1870 if (filled
&& state
->start
> start
) {
1875 if (state
->start
> end
)
1878 if (state
->state
& bits
) {
1882 } else if (filled
) {
1887 if (state
->end
== (u64
)-1)
1890 start
= state
->end
+ 1;
1893 node
= rb_next(node
);
1900 spin_unlock(&tree
->lock
);
1905 * helper function to set a given page up to date if all the
1906 * extents in the tree for that page are up to date
1908 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1910 u64 start
= page_offset(page
);
1911 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1912 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1913 SetPageUptodate(page
);
1917 * When IO fails, either with EIO or csum verification fails, we
1918 * try other mirrors that might have a good copy of the data. This
1919 * io_failure_record is used to record state as we go through all the
1920 * mirrors. If another mirror has good data, the page is set up to date
1921 * and things continue. If a good mirror can't be found, the original
1922 * bio end_io callback is called to indicate things have failed.
1924 struct io_failure_record
{
1929 unsigned long bio_flags
;
1935 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
1940 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1942 set_state_private(failure_tree
, rec
->start
, 0);
1943 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1944 rec
->start
+ rec
->len
- 1,
1945 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1949 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1950 rec
->start
+ rec
->len
- 1,
1951 EXTENT_DAMAGED
, GFP_NOFS
);
1959 static void repair_io_failure_callback(struct bio
*bio
, int err
)
1961 complete(bio
->bi_private
);
1965 * this bypasses the standard btrfs submit functions deliberately, as
1966 * the standard behavior is to write all copies in a raid setup. here we only
1967 * want to write the one bad copy. so we do the mapping for ourselves and issue
1968 * submit_bio directly.
1969 * to avoid any synchronization issues, wait for the data after writing, which
1970 * actually prevents the read that triggered the error from finishing.
1971 * currently, there can be no more than two copies of every data bit. thus,
1972 * exactly one rewrite is required.
1974 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 start
,
1975 u64 length
, u64 logical
, struct page
*page
,
1979 struct btrfs_device
*dev
;
1980 DECLARE_COMPLETION_ONSTACK(compl);
1983 struct btrfs_bio
*bbio
= NULL
;
1984 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
1987 BUG_ON(!mirror_num
);
1989 /* we can't repair anything in raid56 yet */
1990 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
1993 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
1996 bio
->bi_private
= &compl;
1997 bio
->bi_end_io
= repair_io_failure_callback
;
1999 map_length
= length
;
2001 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2002 &map_length
, &bbio
, mirror_num
);
2007 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2008 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2009 bio
->bi_sector
= sector
;
2010 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2012 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2016 bio
->bi_bdev
= dev
->bdev
;
2017 bio_add_page(bio
, page
, length
, start
- page_offset(page
));
2018 btrfsic_submit_bio(WRITE_SYNC
, bio
);
2019 wait_for_completion(&compl);
2021 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
2022 /* try to remap that extent elsewhere? */
2024 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2028 printk_ratelimited_in_rcu(KERN_INFO
"btrfs read error corrected: ino %lu off %llu "
2029 "(dev %s sector %llu)\n", page
->mapping
->host
->i_ino
,
2030 start
, rcu_str_deref(dev
->name
), sector
);
2036 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2039 u64 start
= eb
->start
;
2040 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2043 for (i
= 0; i
< num_pages
; i
++) {
2044 struct page
*p
= extent_buffer_page(eb
, i
);
2045 ret
= repair_io_failure(root
->fs_info
, start
, PAGE_CACHE_SIZE
,
2046 start
, p
, mirror_num
);
2049 start
+= PAGE_CACHE_SIZE
;
2056 * each time an IO finishes, we do a fast check in the IO failure tree
2057 * to see if we need to process or clean up an io_failure_record
2059 static int clean_io_failure(u64 start
, struct page
*page
)
2062 u64 private_failure
;
2063 struct io_failure_record
*failrec
;
2064 struct btrfs_fs_info
*fs_info
;
2065 struct extent_state
*state
;
2069 struct inode
*inode
= page
->mapping
->host
;
2072 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2073 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2077 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2082 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2083 BUG_ON(!failrec
->this_mirror
);
2085 if (failrec
->in_validation
) {
2086 /* there was no real error, just free the record */
2087 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2093 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2094 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2097 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2099 if (state
&& state
->start
<= failrec
->start
&&
2100 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2101 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2102 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2104 if (num_copies
> 1) {
2105 ret
= repair_io_failure(fs_info
, start
, failrec
->len
,
2106 failrec
->logical
, page
,
2107 failrec
->failed_mirror
);
2115 ret
= free_io_failure(inode
, failrec
, did_repair
);
2121 * this is a generic handler for readpage errors (default
2122 * readpage_io_failed_hook). if other copies exist, read those and write back
2123 * good data to the failed position. does not investigate in remapping the
2124 * failed extent elsewhere, hoping the device will be smart enough to do this as
2128 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2129 struct page
*page
, u64 start
, u64 end
,
2132 struct io_failure_record
*failrec
= NULL
;
2134 struct extent_map
*em
;
2135 struct inode
*inode
= page
->mapping
->host
;
2136 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2137 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2138 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2140 struct btrfs_io_bio
*btrfs_failed_bio
;
2141 struct btrfs_io_bio
*btrfs_bio
;
2147 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2149 ret
= get_state_private(failure_tree
, start
, &private);
2151 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2154 failrec
->start
= start
;
2155 failrec
->len
= end
- start
+ 1;
2156 failrec
->this_mirror
= 0;
2157 failrec
->bio_flags
= 0;
2158 failrec
->in_validation
= 0;
2160 read_lock(&em_tree
->lock
);
2161 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2163 read_unlock(&em_tree
->lock
);
2168 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
2169 free_extent_map(em
);
2172 read_unlock(&em_tree
->lock
);
2178 logical
= start
- em
->start
;
2179 logical
= em
->block_start
+ logical
;
2180 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2181 logical
= em
->block_start
;
2182 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2183 extent_set_compress_type(&failrec
->bio_flags
,
2186 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2187 "len=%llu\n", logical
, start
, failrec
->len
);
2188 failrec
->logical
= logical
;
2189 free_extent_map(em
);
2191 /* set the bits in the private failure tree */
2192 ret
= set_extent_bits(failure_tree
, start
, end
,
2193 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2195 ret
= set_state_private(failure_tree
, start
,
2196 (u64
)(unsigned long)failrec
);
2197 /* set the bits in the inode's tree */
2199 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2206 failrec
= (struct io_failure_record
*)(unsigned long)private;
2207 pr_debug("bio_readpage_error: (found) logical=%llu, "
2208 "start=%llu, len=%llu, validation=%d\n",
2209 failrec
->logical
, failrec
->start
, failrec
->len
,
2210 failrec
->in_validation
);
2212 * when data can be on disk more than twice, add to failrec here
2213 * (e.g. with a list for failed_mirror) to make
2214 * clean_io_failure() clean all those errors at once.
2217 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2218 failrec
->logical
, failrec
->len
);
2219 if (num_copies
== 1) {
2221 * we only have a single copy of the data, so don't bother with
2222 * all the retry and error correction code that follows. no
2223 * matter what the error is, it is very likely to persist.
2225 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2226 num_copies
, failrec
->this_mirror
, failed_mirror
);
2227 free_io_failure(inode
, failrec
, 0);
2232 * there are two premises:
2233 * a) deliver good data to the caller
2234 * b) correct the bad sectors on disk
2236 if (failed_bio
->bi_vcnt
> 1) {
2238 * to fulfill b), we need to know the exact failing sectors, as
2239 * we don't want to rewrite any more than the failed ones. thus,
2240 * we need separate read requests for the failed bio
2242 * if the following BUG_ON triggers, our validation request got
2243 * merged. we need separate requests for our algorithm to work.
2245 BUG_ON(failrec
->in_validation
);
2246 failrec
->in_validation
= 1;
2247 failrec
->this_mirror
= failed_mirror
;
2248 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2251 * we're ready to fulfill a) and b) alongside. get a good copy
2252 * of the failed sector and if we succeed, we have setup
2253 * everything for repair_io_failure to do the rest for us.
2255 if (failrec
->in_validation
) {
2256 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2257 failrec
->in_validation
= 0;
2258 failrec
->this_mirror
= 0;
2260 failrec
->failed_mirror
= failed_mirror
;
2261 failrec
->this_mirror
++;
2262 if (failrec
->this_mirror
== failed_mirror
)
2263 failrec
->this_mirror
++;
2264 read_mode
= READ_SYNC
;
2267 if (failrec
->this_mirror
> num_copies
) {
2268 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2269 num_copies
, failrec
->this_mirror
, failed_mirror
);
2270 free_io_failure(inode
, failrec
, 0);
2274 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2276 free_io_failure(inode
, failrec
, 0);
2279 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2280 bio
->bi_sector
= failrec
->logical
>> 9;
2281 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2284 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2285 if (btrfs_failed_bio
->csum
) {
2286 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2287 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2289 btrfs_bio
= btrfs_io_bio(bio
);
2290 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2291 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2292 phy_offset
*= csum_size
;
2293 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ phy_offset
,
2297 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2299 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2300 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2301 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2303 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2304 failrec
->this_mirror
,
2305 failrec
->bio_flags
, 0);
2309 /* lots and lots of room for performance fixes in the end_bio funcs */
2311 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2313 int uptodate
= (err
== 0);
2314 struct extent_io_tree
*tree
;
2317 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2319 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2320 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2321 end
, NULL
, uptodate
);
2327 ClearPageUptodate(page
);
2329 ret
= ret
< 0 ? ret
: -EIO
;
2330 mapping_set_error(page
->mapping
, ret
);
2336 * after a writepage IO is done, we need to:
2337 * clear the uptodate bits on error
2338 * clear the writeback bits in the extent tree for this IO
2339 * end_page_writeback if the page has no more pending IO
2341 * Scheduling is not allowed, so the extent state tree is expected
2342 * to have one and only one object corresponding to this IO.
2344 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2346 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2347 struct extent_io_tree
*tree
;
2352 struct page
*page
= bvec
->bv_page
;
2353 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2355 /* We always issue full-page reads, but if some block
2356 * in a page fails to read, blk_update_request() will
2357 * advance bv_offset and adjust bv_len to compensate.
2358 * Print a warning for nonzero offsets, and an error
2359 * if they don't add up to a full page. */
2360 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2361 printk("%s page write in btrfs with offset %u and length %u\n",
2362 bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
2363 ? KERN_ERR
"partial" : KERN_INFO
"incomplete",
2364 bvec
->bv_offset
, bvec
->bv_len
);
2366 start
= page_offset(page
);
2367 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2369 if (--bvec
>= bio
->bi_io_vec
)
2370 prefetchw(&bvec
->bv_page
->flags
);
2372 if (end_extent_writepage(page
, err
, start
, end
))
2375 end_page_writeback(page
);
2376 } while (bvec
>= bio
->bi_io_vec
);
2382 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2385 struct extent_state
*cached
= NULL
;
2386 u64 end
= start
+ len
- 1;
2388 if (uptodate
&& tree
->track_uptodate
)
2389 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2390 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2394 * after a readpage IO is done, we need to:
2395 * clear the uptodate bits on error
2396 * set the uptodate bits if things worked
2397 * set the page up to date if all extents in the tree are uptodate
2398 * clear the lock bit in the extent tree
2399 * unlock the page if there are no other extents locked for it
2401 * Scheduling is not allowed, so the extent state tree is expected
2402 * to have one and only one object corresponding to this IO.
2404 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2406 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2407 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2408 struct bio_vec
*bvec
= bio
->bi_io_vec
;
2409 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2410 struct extent_io_tree
*tree
;
2415 u64 extent_start
= 0;
2424 struct page
*page
= bvec
->bv_page
;
2425 struct inode
*inode
= page
->mapping
->host
;
2427 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2428 "mirror=%lu\n", (u64
)bio
->bi_sector
, err
,
2429 io_bio
->mirror_num
);
2430 tree
= &BTRFS_I(inode
)->io_tree
;
2432 /* We always issue full-page reads, but if some block
2433 * in a page fails to read, blk_update_request() will
2434 * advance bv_offset and adjust bv_len to compensate.
2435 * Print a warning for nonzero offsets, and an error
2436 * if they don't add up to a full page. */
2437 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2438 printk("%s page read in btrfs with offset %u and length %u\n",
2439 bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
2440 ? KERN_ERR
"partial" : KERN_INFO
"incomplete",
2441 bvec
->bv_offset
, bvec
->bv_len
);
2443 start
= page_offset(page
);
2444 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2447 if (++bvec
<= bvec_end
)
2448 prefetchw(&bvec
->bv_page
->flags
);
2450 mirror
= io_bio
->mirror_num
;
2451 if (likely(uptodate
&& tree
->ops
&&
2452 tree
->ops
->readpage_end_io_hook
)) {
2453 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2459 clean_io_failure(start
, page
);
2462 if (likely(uptodate
))
2465 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2466 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2468 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2472 * The generic bio_readpage_error handles errors the
2473 * following way: If possible, new read requests are
2474 * created and submitted and will end up in
2475 * end_bio_extent_readpage as well (if we're lucky, not
2476 * in the !uptodate case). In that case it returns 0 and
2477 * we just go on with the next page in our bio. If it
2478 * can't handle the error it will return -EIO and we
2479 * remain responsible for that page.
2481 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2485 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2492 if (likely(uptodate
)) {
2493 loff_t i_size
= i_size_read(inode
);
2494 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2497 /* Zero out the end if this page straddles i_size */
2498 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2499 if (page
->index
== end_index
&& offset
)
2500 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2501 SetPageUptodate(page
);
2503 ClearPageUptodate(page
);
2509 if (unlikely(!uptodate
)) {
2511 endio_readpage_release_extent(tree
,
2517 endio_readpage_release_extent(tree
, start
,
2518 end
- start
+ 1, 0);
2519 } else if (!extent_len
) {
2520 extent_start
= start
;
2521 extent_len
= end
+ 1 - start
;
2522 } else if (extent_start
+ extent_len
== start
) {
2523 extent_len
+= end
+ 1 - start
;
2525 endio_readpage_release_extent(tree
, extent_start
,
2526 extent_len
, uptodate
);
2527 extent_start
= start
;
2528 extent_len
= end
+ 1 - start
;
2530 } while (bvec
<= bvec_end
);
2533 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2536 io_bio
->end_io(io_bio
, err
);
2541 * this allocates from the btrfs_bioset. We're returning a bio right now
2542 * but you can call btrfs_io_bio for the appropriate container_of magic
2545 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2548 struct btrfs_io_bio
*btrfs_bio
;
2551 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2553 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2554 while (!bio
&& (nr_vecs
/= 2)) {
2555 bio
= bio_alloc_bioset(gfp_flags
,
2556 nr_vecs
, btrfs_bioset
);
2562 bio
->bi_bdev
= bdev
;
2563 bio
->bi_sector
= first_sector
;
2564 btrfs_bio
= btrfs_io_bio(bio
);
2565 btrfs_bio
->csum
= NULL
;
2566 btrfs_bio
->csum_allocated
= NULL
;
2567 btrfs_bio
->end_io
= NULL
;
2572 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2574 return bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2578 /* this also allocates from the btrfs_bioset */
2579 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2581 struct btrfs_io_bio
*btrfs_bio
;
2584 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2586 btrfs_bio
= btrfs_io_bio(bio
);
2587 btrfs_bio
->csum
= NULL
;
2588 btrfs_bio
->csum_allocated
= NULL
;
2589 btrfs_bio
->end_io
= NULL
;
2595 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2596 int mirror_num
, unsigned long bio_flags
)
2599 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2600 struct page
*page
= bvec
->bv_page
;
2601 struct extent_io_tree
*tree
= bio
->bi_private
;
2604 start
= page_offset(page
) + bvec
->bv_offset
;
2606 bio
->bi_private
= NULL
;
2610 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2611 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2612 mirror_num
, bio_flags
, start
);
2614 btrfsic_submit_bio(rw
, bio
);
2616 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2622 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2623 unsigned long offset
, size_t size
, struct bio
*bio
,
2624 unsigned long bio_flags
)
2627 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2628 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2635 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2636 struct page
*page
, sector_t sector
,
2637 size_t size
, unsigned long offset
,
2638 struct block_device
*bdev
,
2639 struct bio
**bio_ret
,
2640 unsigned long max_pages
,
2641 bio_end_io_t end_io_func
,
2643 unsigned long prev_bio_flags
,
2644 unsigned long bio_flags
)
2650 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2651 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2652 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2654 if (bio_ret
&& *bio_ret
) {
2657 contig
= bio
->bi_sector
== sector
;
2659 contig
= bio_end_sector(bio
) == sector
;
2661 if (prev_bio_flags
!= bio_flags
|| !contig
||
2662 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2663 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2664 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2673 if (this_compressed
)
2676 nr
= bio_get_nr_vecs(bdev
);
2678 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2682 bio_add_page(bio
, page
, page_size
, offset
);
2683 bio
->bi_end_io
= end_io_func
;
2684 bio
->bi_private
= tree
;
2689 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2694 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2697 if (!PagePrivate(page
)) {
2698 SetPagePrivate(page
);
2699 page_cache_get(page
);
2700 set_page_private(page
, (unsigned long)eb
);
2702 WARN_ON(page
->private != (unsigned long)eb
);
2706 void set_page_extent_mapped(struct page
*page
)
2708 if (!PagePrivate(page
)) {
2709 SetPagePrivate(page
);
2710 page_cache_get(page
);
2711 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2715 static struct extent_map
*
2716 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2717 u64 start
, u64 len
, get_extent_t
*get_extent
,
2718 struct extent_map
**em_cached
)
2720 struct extent_map
*em
;
2722 if (em_cached
&& *em_cached
) {
2724 if (em
->in_tree
&& start
>= em
->start
&&
2725 start
< extent_map_end(em
)) {
2726 atomic_inc(&em
->refs
);
2730 free_extent_map(em
);
2734 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2735 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2737 atomic_inc(&em
->refs
);
2743 * basic readpage implementation. Locked extent state structs are inserted
2744 * into the tree that are removed when the IO is done (by the end_io
2746 * XXX JDM: This needs looking at to ensure proper page locking
2748 static int __do_readpage(struct extent_io_tree
*tree
,
2750 get_extent_t
*get_extent
,
2751 struct extent_map
**em_cached
,
2752 struct bio
**bio
, int mirror_num
,
2753 unsigned long *bio_flags
, int rw
)
2755 struct inode
*inode
= page
->mapping
->host
;
2756 u64 start
= page_offset(page
);
2757 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2761 u64 last_byte
= i_size_read(inode
);
2765 struct extent_map
*em
;
2766 struct block_device
*bdev
;
2769 int parent_locked
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2770 size_t pg_offset
= 0;
2772 size_t disk_io_size
;
2773 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2774 unsigned long this_bio_flag
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2776 set_page_extent_mapped(page
);
2779 if (!PageUptodate(page
)) {
2780 if (cleancache_get_page(page
) == 0) {
2781 BUG_ON(blocksize
!= PAGE_SIZE
);
2782 unlock_extent(tree
, start
, end
);
2787 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2789 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2792 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2793 userpage
= kmap_atomic(page
);
2794 memset(userpage
+ zero_offset
, 0, iosize
);
2795 flush_dcache_page(page
);
2796 kunmap_atomic(userpage
);
2799 while (cur
<= end
) {
2800 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2802 if (cur
>= last_byte
) {
2804 struct extent_state
*cached
= NULL
;
2806 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2807 userpage
= kmap_atomic(page
);
2808 memset(userpage
+ pg_offset
, 0, iosize
);
2809 flush_dcache_page(page
);
2810 kunmap_atomic(userpage
);
2811 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2814 unlock_extent_cached(tree
, cur
,
2819 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2820 end
- cur
+ 1, get_extent
, em_cached
);
2821 if (IS_ERR_OR_NULL(em
)) {
2824 unlock_extent(tree
, cur
, end
);
2827 extent_offset
= cur
- em
->start
;
2828 BUG_ON(extent_map_end(em
) <= cur
);
2831 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2832 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2833 extent_set_compress_type(&this_bio_flag
,
2837 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2838 cur_end
= min(extent_map_end(em
) - 1, end
);
2839 iosize
= ALIGN(iosize
, blocksize
);
2840 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2841 disk_io_size
= em
->block_len
;
2842 sector
= em
->block_start
>> 9;
2844 sector
= (em
->block_start
+ extent_offset
) >> 9;
2845 disk_io_size
= iosize
;
2848 block_start
= em
->block_start
;
2849 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2850 block_start
= EXTENT_MAP_HOLE
;
2851 free_extent_map(em
);
2854 /* we've found a hole, just zero and go on */
2855 if (block_start
== EXTENT_MAP_HOLE
) {
2857 struct extent_state
*cached
= NULL
;
2859 userpage
= kmap_atomic(page
);
2860 memset(userpage
+ pg_offset
, 0, iosize
);
2861 flush_dcache_page(page
);
2862 kunmap_atomic(userpage
);
2864 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2866 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2869 pg_offset
+= iosize
;
2872 /* the get_extent function already copied into the page */
2873 if (test_range_bit(tree
, cur
, cur_end
,
2874 EXTENT_UPTODATE
, 1, NULL
)) {
2875 check_page_uptodate(tree
, page
);
2877 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2879 pg_offset
+= iosize
;
2882 /* we have an inline extent but it didn't get marked up
2883 * to date. Error out
2885 if (block_start
== EXTENT_MAP_INLINE
) {
2888 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2890 pg_offset
+= iosize
;
2895 ret
= submit_extent_page(rw
, tree
, page
,
2896 sector
, disk_io_size
, pg_offset
,
2898 end_bio_extent_readpage
, mirror_num
,
2903 *bio_flags
= this_bio_flag
;
2907 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2910 pg_offset
+= iosize
;
2914 if (!PageError(page
))
2915 SetPageUptodate(page
);
2921 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
2922 struct page
*pages
[], int nr_pages
,
2924 get_extent_t
*get_extent
,
2925 struct extent_map
**em_cached
,
2926 struct bio
**bio
, int mirror_num
,
2927 unsigned long *bio_flags
, int rw
)
2929 struct inode
*inode
;
2930 struct btrfs_ordered_extent
*ordered
;
2933 inode
= pages
[0]->mapping
->host
;
2935 lock_extent(tree
, start
, end
);
2936 ordered
= btrfs_lookup_ordered_range(inode
, start
,
2940 unlock_extent(tree
, start
, end
);
2941 btrfs_start_ordered_extent(inode
, ordered
, 1);
2942 btrfs_put_ordered_extent(ordered
);
2945 for (index
= 0; index
< nr_pages
; index
++) {
2946 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
2947 mirror_num
, bio_flags
, rw
);
2948 page_cache_release(pages
[index
]);
2952 static void __extent_readpages(struct extent_io_tree
*tree
,
2953 struct page
*pages
[],
2954 int nr_pages
, get_extent_t
*get_extent
,
2955 struct extent_map
**em_cached
,
2956 struct bio
**bio
, int mirror_num
,
2957 unsigned long *bio_flags
, int rw
)
2963 int first_index
= 0;
2965 for (index
= 0; index
< nr_pages
; index
++) {
2966 page_start
= page_offset(pages
[index
]);
2969 end
= start
+ PAGE_CACHE_SIZE
- 1;
2970 first_index
= index
;
2971 } else if (end
+ 1 == page_start
) {
2972 end
+= PAGE_CACHE_SIZE
;
2974 __do_contiguous_readpages(tree
, &pages
[first_index
],
2975 index
- first_index
, start
,
2976 end
, get_extent
, em_cached
,
2977 bio
, mirror_num
, bio_flags
,
2980 end
= start
+ PAGE_CACHE_SIZE
- 1;
2981 first_index
= index
;
2986 __do_contiguous_readpages(tree
, &pages
[first_index
],
2987 index
- first_index
, start
,
2988 end
, get_extent
, em_cached
, bio
,
2989 mirror_num
, bio_flags
, rw
);
2992 static int __extent_read_full_page(struct extent_io_tree
*tree
,
2994 get_extent_t
*get_extent
,
2995 struct bio
**bio
, int mirror_num
,
2996 unsigned long *bio_flags
, int rw
)
2998 struct inode
*inode
= page
->mapping
->host
;
2999 struct btrfs_ordered_extent
*ordered
;
3000 u64 start
= page_offset(page
);
3001 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3005 lock_extent(tree
, start
, end
);
3006 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
3009 unlock_extent(tree
, start
, end
);
3010 btrfs_start_ordered_extent(inode
, ordered
, 1);
3011 btrfs_put_ordered_extent(ordered
);
3014 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3019 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3020 get_extent_t
*get_extent
, int mirror_num
)
3022 struct bio
*bio
= NULL
;
3023 unsigned long bio_flags
= 0;
3026 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3029 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3033 int extent_read_full_page_nolock(struct extent_io_tree
*tree
, struct page
*page
,
3034 get_extent_t
*get_extent
, int mirror_num
)
3036 struct bio
*bio
= NULL
;
3037 unsigned long bio_flags
= EXTENT_BIO_PARENT_LOCKED
;
3040 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, &bio
, mirror_num
,
3043 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3047 static noinline
void update_nr_written(struct page
*page
,
3048 struct writeback_control
*wbc
,
3049 unsigned long nr_written
)
3051 wbc
->nr_to_write
-= nr_written
;
3052 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
3053 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
3054 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
3058 * the writepage semantics are similar to regular writepage. extent
3059 * records are inserted to lock ranges in the tree, and as dirty areas
3060 * are found, they are marked writeback. Then the lock bits are removed
3061 * and the end_io handler clears the writeback ranges
3063 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3066 struct inode
*inode
= page
->mapping
->host
;
3067 struct extent_page_data
*epd
= data
;
3068 struct extent_io_tree
*tree
= epd
->tree
;
3069 u64 start
= page_offset(page
);
3071 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3075 u64 last_byte
= i_size_read(inode
);
3079 struct extent_state
*cached_state
= NULL
;
3080 struct extent_map
*em
;
3081 struct block_device
*bdev
;
3084 size_t pg_offset
= 0;
3086 loff_t i_size
= i_size_read(inode
);
3087 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
3093 unsigned long nr_written
= 0;
3094 bool fill_delalloc
= true;
3096 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3097 write_flags
= WRITE_SYNC
;
3099 write_flags
= WRITE
;
3101 trace___extent_writepage(page
, inode
, wbc
);
3103 WARN_ON(!PageLocked(page
));
3105 ClearPageError(page
);
3107 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
3108 if (page
->index
> end_index
||
3109 (page
->index
== end_index
&& !pg_offset
)) {
3110 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
3115 if (page
->index
== end_index
) {
3118 userpage
= kmap_atomic(page
);
3119 memset(userpage
+ pg_offset
, 0,
3120 PAGE_CACHE_SIZE
- pg_offset
);
3121 kunmap_atomic(userpage
);
3122 flush_dcache_page(page
);
3126 set_page_extent_mapped(page
);
3128 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
3129 fill_delalloc
= false;
3131 delalloc_start
= start
;
3134 if (!epd
->extent_locked
&& fill_delalloc
) {
3135 u64 delalloc_to_write
= 0;
3137 * make sure the wbc mapping index is at least updated
3140 update_nr_written(page
, wbc
, 0);
3142 while (delalloc_end
< page_end
) {
3143 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3148 if (nr_delalloc
== 0) {
3149 delalloc_start
= delalloc_end
+ 1;
3152 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3157 /* File system has been set read-only */
3163 * delalloc_end is already one less than the total
3164 * length, so we don't subtract one from
3167 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3170 delalloc_start
= delalloc_end
+ 1;
3172 if (wbc
->nr_to_write
< delalloc_to_write
) {
3175 if (delalloc_to_write
< thresh
* 2)
3176 thresh
= delalloc_to_write
;
3177 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3181 /* did the fill delalloc function already unlock and start
3187 * we've unlocked the page, so we can't update
3188 * the mapping's writeback index, just update
3191 wbc
->nr_to_write
-= nr_written
;
3195 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3196 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3199 /* Fixup worker will requeue */
3201 wbc
->pages_skipped
++;
3203 redirty_page_for_writepage(wbc
, page
);
3204 update_nr_written(page
, wbc
, nr_written
);
3212 * we don't want to touch the inode after unlocking the page,
3213 * so we update the mapping writeback index now
3215 update_nr_written(page
, wbc
, nr_written
+ 1);
3218 if (last_byte
<= start
) {
3219 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3220 tree
->ops
->writepage_end_io_hook(page
, start
,
3225 blocksize
= inode
->i_sb
->s_blocksize
;
3227 while (cur
<= end
) {
3228 if (cur
>= last_byte
) {
3229 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3230 tree
->ops
->writepage_end_io_hook(page
, cur
,
3234 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3236 if (IS_ERR_OR_NULL(em
)) {
3241 extent_offset
= cur
- em
->start
;
3242 BUG_ON(extent_map_end(em
) <= cur
);
3244 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
3245 iosize
= ALIGN(iosize
, blocksize
);
3246 sector
= (em
->block_start
+ extent_offset
) >> 9;
3248 block_start
= em
->block_start
;
3249 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3250 free_extent_map(em
);
3254 * compressed and inline extents are written through other
3257 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3258 block_start
== EXTENT_MAP_INLINE
) {
3260 * end_io notification does not happen here for
3261 * compressed extents
3263 if (!compressed
&& tree
->ops
&&
3264 tree
->ops
->writepage_end_io_hook
)
3265 tree
->ops
->writepage_end_io_hook(page
, cur
,
3268 else if (compressed
) {
3269 /* we don't want to end_page_writeback on
3270 * a compressed extent. this happens
3277 pg_offset
+= iosize
;
3280 /* leave this out until we have a page_mkwrite call */
3281 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
3282 EXTENT_DIRTY
, 0, NULL
)) {
3284 pg_offset
+= iosize
;
3288 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3289 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3297 unsigned long max_nr
= end_index
+ 1;
3299 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3300 if (!PageWriteback(page
)) {
3301 printk(KERN_ERR
"btrfs warning page %lu not "
3302 "writeback, cur %llu end %llu\n",
3303 page
->index
, cur
, end
);
3306 ret
= submit_extent_page(write_flags
, tree
, page
,
3307 sector
, iosize
, pg_offset
,
3308 bdev
, &epd
->bio
, max_nr
,
3309 end_bio_extent_writepage
,
3315 pg_offset
+= iosize
;
3320 /* make sure the mapping tag for page dirty gets cleared */
3321 set_page_writeback(page
);
3322 end_page_writeback(page
);
3328 /* drop our reference on any cached states */
3329 free_extent_state(cached_state
);
3333 static int eb_wait(void *word
)
3339 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3341 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3342 TASK_UNINTERRUPTIBLE
);
3345 static int lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3346 struct btrfs_fs_info
*fs_info
,
3347 struct extent_page_data
*epd
)
3349 unsigned long i
, num_pages
;
3353 if (!btrfs_try_tree_write_lock(eb
)) {
3355 flush_write_bio(epd
);
3356 btrfs_tree_lock(eb
);
3359 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3360 btrfs_tree_unlock(eb
);
3364 flush_write_bio(epd
);
3368 wait_on_extent_buffer_writeback(eb
);
3369 btrfs_tree_lock(eb
);
3370 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3372 btrfs_tree_unlock(eb
);
3377 * We need to do this to prevent races in people who check if the eb is
3378 * under IO since we can end up having no IO bits set for a short period
3381 spin_lock(&eb
->refs_lock
);
3382 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3383 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3384 spin_unlock(&eb
->refs_lock
);
3385 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3386 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3388 fs_info
->dirty_metadata_batch
);
3391 spin_unlock(&eb
->refs_lock
);
3394 btrfs_tree_unlock(eb
);
3399 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3400 for (i
= 0; i
< num_pages
; i
++) {
3401 struct page
*p
= extent_buffer_page(eb
, i
);
3403 if (!trylock_page(p
)) {
3405 flush_write_bio(epd
);
3415 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3417 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3418 smp_mb__after_clear_bit();
3419 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3422 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3424 int uptodate
= err
== 0;
3425 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
3426 struct extent_buffer
*eb
;
3430 struct page
*page
= bvec
->bv_page
;
3433 eb
= (struct extent_buffer
*)page
->private;
3435 done
= atomic_dec_and_test(&eb
->io_pages
);
3437 if (!uptodate
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3438 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3439 ClearPageUptodate(page
);
3443 end_page_writeback(page
);
3448 end_extent_buffer_writeback(eb
);
3449 } while (bvec
>= bio
->bi_io_vec
);
3455 static int write_one_eb(struct extent_buffer
*eb
,
3456 struct btrfs_fs_info
*fs_info
,
3457 struct writeback_control
*wbc
,
3458 struct extent_page_data
*epd
)
3460 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3461 u64 offset
= eb
->start
;
3462 unsigned long i
, num_pages
;
3463 unsigned long bio_flags
= 0;
3464 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3467 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3468 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3469 atomic_set(&eb
->io_pages
, num_pages
);
3470 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3471 bio_flags
= EXTENT_BIO_TREE_LOG
;
3473 for (i
= 0; i
< num_pages
; i
++) {
3474 struct page
*p
= extent_buffer_page(eb
, i
);
3476 clear_page_dirty_for_io(p
);
3477 set_page_writeback(p
);
3478 ret
= submit_extent_page(rw
, eb
->tree
, p
, offset
>> 9,
3479 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3480 -1, end_bio_extent_buffer_writepage
,
3481 0, epd
->bio_flags
, bio_flags
);
3482 epd
->bio_flags
= bio_flags
;
3484 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3486 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3487 end_extent_buffer_writeback(eb
);
3491 offset
+= PAGE_CACHE_SIZE
;
3492 update_nr_written(p
, wbc
, 1);
3496 if (unlikely(ret
)) {
3497 for (; i
< num_pages
; i
++) {
3498 struct page
*p
= extent_buffer_page(eb
, i
);
3506 int btree_write_cache_pages(struct address_space
*mapping
,
3507 struct writeback_control
*wbc
)
3509 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3510 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3511 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3512 struct extent_page_data epd
= {
3516 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3521 int nr_to_write_done
= 0;
3522 struct pagevec pvec
;
3525 pgoff_t end
; /* Inclusive */
3529 pagevec_init(&pvec
, 0);
3530 if (wbc
->range_cyclic
) {
3531 index
= mapping
->writeback_index
; /* Start from prev offset */
3534 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3535 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3538 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3539 tag
= PAGECACHE_TAG_TOWRITE
;
3541 tag
= PAGECACHE_TAG_DIRTY
;
3543 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3544 tag_pages_for_writeback(mapping
, index
, end
);
3545 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3546 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3547 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3551 for (i
= 0; i
< nr_pages
; i
++) {
3552 struct page
*page
= pvec
.pages
[i
];
3554 if (!PagePrivate(page
))
3557 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3562 spin_lock(&mapping
->private_lock
);
3563 if (!PagePrivate(page
)) {
3564 spin_unlock(&mapping
->private_lock
);
3568 eb
= (struct extent_buffer
*)page
->private;
3571 * Shouldn't happen and normally this would be a BUG_ON
3572 * but no sense in crashing the users box for something
3573 * we can survive anyway.
3576 spin_unlock(&mapping
->private_lock
);
3581 if (eb
== prev_eb
) {
3582 spin_unlock(&mapping
->private_lock
);
3586 ret
= atomic_inc_not_zero(&eb
->refs
);
3587 spin_unlock(&mapping
->private_lock
);
3592 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3594 free_extent_buffer(eb
);
3598 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3601 free_extent_buffer(eb
);
3604 free_extent_buffer(eb
);
3607 * the filesystem may choose to bump up nr_to_write.
3608 * We have to make sure to honor the new nr_to_write
3611 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3613 pagevec_release(&pvec
);
3616 if (!scanned
&& !done
) {
3618 * We hit the last page and there is more work to be done: wrap
3619 * back to the start of the file
3625 flush_write_bio(&epd
);
3630 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3631 * @mapping: address space structure to write
3632 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3633 * @writepage: function called for each page
3634 * @data: data passed to writepage function
3636 * If a page is already under I/O, write_cache_pages() skips it, even
3637 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3638 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3639 * and msync() need to guarantee that all the data which was dirty at the time
3640 * the call was made get new I/O started against them. If wbc->sync_mode is
3641 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3642 * existing IO to complete.
3644 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3645 struct address_space
*mapping
,
3646 struct writeback_control
*wbc
,
3647 writepage_t writepage
, void *data
,
3648 void (*flush_fn
)(void *))
3650 struct inode
*inode
= mapping
->host
;
3653 int nr_to_write_done
= 0;
3654 struct pagevec pvec
;
3657 pgoff_t end
; /* Inclusive */
3662 * We have to hold onto the inode so that ordered extents can do their
3663 * work when the IO finishes. The alternative to this is failing to add
3664 * an ordered extent if the igrab() fails there and that is a huge pain
3665 * to deal with, so instead just hold onto the inode throughout the
3666 * writepages operation. If it fails here we are freeing up the inode
3667 * anyway and we'd rather not waste our time writing out stuff that is
3668 * going to be truncated anyway.
3673 pagevec_init(&pvec
, 0);
3674 if (wbc
->range_cyclic
) {
3675 index
= mapping
->writeback_index
; /* Start from prev offset */
3678 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3679 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3682 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3683 tag
= PAGECACHE_TAG_TOWRITE
;
3685 tag
= PAGECACHE_TAG_DIRTY
;
3687 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3688 tag_pages_for_writeback(mapping
, index
, end
);
3689 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3690 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3691 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3695 for (i
= 0; i
< nr_pages
; i
++) {
3696 struct page
*page
= pvec
.pages
[i
];
3699 * At this point we hold neither mapping->tree_lock nor
3700 * lock on the page itself: the page may be truncated or
3701 * invalidated (changing page->mapping to NULL), or even
3702 * swizzled back from swapper_space to tmpfs file
3705 if (!trylock_page(page
)) {
3710 if (unlikely(page
->mapping
!= mapping
)) {
3715 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3721 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3722 if (PageWriteback(page
))
3724 wait_on_page_writeback(page
);
3727 if (PageWriteback(page
) ||
3728 !clear_page_dirty_for_io(page
)) {
3733 ret
= (*writepage
)(page
, wbc
, data
);
3735 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3743 * the filesystem may choose to bump up nr_to_write.
3744 * We have to make sure to honor the new nr_to_write
3747 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3749 pagevec_release(&pvec
);
3752 if (!scanned
&& !done
) {
3754 * We hit the last page and there is more work to be done: wrap
3755 * back to the start of the file
3761 btrfs_add_delayed_iput(inode
);
3765 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3774 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
3775 BUG_ON(ret
< 0); /* -ENOMEM */
3780 static noinline
void flush_write_bio(void *data
)
3782 struct extent_page_data
*epd
= data
;
3783 flush_epd_write_bio(epd
);
3786 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3787 get_extent_t
*get_extent
,
3788 struct writeback_control
*wbc
)
3791 struct extent_page_data epd
= {
3794 .get_extent
= get_extent
,
3796 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3800 ret
= __extent_writepage(page
, wbc
, &epd
);
3802 flush_epd_write_bio(&epd
);
3806 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3807 u64 start
, u64 end
, get_extent_t
*get_extent
,
3811 struct address_space
*mapping
= inode
->i_mapping
;
3813 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3816 struct extent_page_data epd
= {
3819 .get_extent
= get_extent
,
3821 .sync_io
= mode
== WB_SYNC_ALL
,
3824 struct writeback_control wbc_writepages
= {
3826 .nr_to_write
= nr_pages
* 2,
3827 .range_start
= start
,
3828 .range_end
= end
+ 1,
3831 while (start
<= end
) {
3832 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3833 if (clear_page_dirty_for_io(page
))
3834 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3836 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3837 tree
->ops
->writepage_end_io_hook(page
, start
,
3838 start
+ PAGE_CACHE_SIZE
- 1,
3842 page_cache_release(page
);
3843 start
+= PAGE_CACHE_SIZE
;
3846 flush_epd_write_bio(&epd
);
3850 int extent_writepages(struct extent_io_tree
*tree
,
3851 struct address_space
*mapping
,
3852 get_extent_t
*get_extent
,
3853 struct writeback_control
*wbc
)
3856 struct extent_page_data epd
= {
3859 .get_extent
= get_extent
,
3861 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3865 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3866 __extent_writepage
, &epd
,
3868 flush_epd_write_bio(&epd
);
3872 int extent_readpages(struct extent_io_tree
*tree
,
3873 struct address_space
*mapping
,
3874 struct list_head
*pages
, unsigned nr_pages
,
3875 get_extent_t get_extent
)
3877 struct bio
*bio
= NULL
;
3879 unsigned long bio_flags
= 0;
3880 struct page
*pagepool
[16];
3882 struct extent_map
*em_cached
= NULL
;
3885 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3886 page
= list_entry(pages
->prev
, struct page
, lru
);
3888 prefetchw(&page
->flags
);
3889 list_del(&page
->lru
);
3890 if (add_to_page_cache_lru(page
, mapping
,
3891 page
->index
, GFP_NOFS
)) {
3892 page_cache_release(page
);
3896 pagepool
[nr
++] = page
;
3897 if (nr
< ARRAY_SIZE(pagepool
))
3899 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3900 &bio
, 0, &bio_flags
, READ
);
3904 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3905 &bio
, 0, &bio_flags
, READ
);
3908 free_extent_map(em_cached
);
3910 BUG_ON(!list_empty(pages
));
3912 return submit_one_bio(READ
, bio
, 0, bio_flags
);
3917 * basic invalidatepage code, this waits on any locked or writeback
3918 * ranges corresponding to the page, and then deletes any extent state
3919 * records from the tree
3921 int extent_invalidatepage(struct extent_io_tree
*tree
,
3922 struct page
*page
, unsigned long offset
)
3924 struct extent_state
*cached_state
= NULL
;
3925 u64 start
= page_offset(page
);
3926 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3927 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
3929 start
+= ALIGN(offset
, blocksize
);
3933 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
3934 wait_on_page_writeback(page
);
3935 clear_extent_bit(tree
, start
, end
,
3936 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3937 EXTENT_DO_ACCOUNTING
,
3938 1, 1, &cached_state
, GFP_NOFS
);
3943 * a helper for releasepage, this tests for areas of the page that
3944 * are locked or under IO and drops the related state bits if it is safe
3947 static int try_release_extent_state(struct extent_map_tree
*map
,
3948 struct extent_io_tree
*tree
,
3949 struct page
*page
, gfp_t mask
)
3951 u64 start
= page_offset(page
);
3952 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3955 if (test_range_bit(tree
, start
, end
,
3956 EXTENT_IOBITS
, 0, NULL
))
3959 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
3962 * at this point we can safely clear everything except the
3963 * locked bit and the nodatasum bit
3965 ret
= clear_extent_bit(tree
, start
, end
,
3966 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
3969 /* if clear_extent_bit failed for enomem reasons,
3970 * we can't allow the release to continue.
3981 * a helper for releasepage. As long as there are no locked extents
3982 * in the range corresponding to the page, both state records and extent
3983 * map records are removed
3985 int try_release_extent_mapping(struct extent_map_tree
*map
,
3986 struct extent_io_tree
*tree
, struct page
*page
,
3989 struct extent_map
*em
;
3990 u64 start
= page_offset(page
);
3991 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3993 if ((mask
& __GFP_WAIT
) &&
3994 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
3996 while (start
<= end
) {
3997 len
= end
- start
+ 1;
3998 write_lock(&map
->lock
);
3999 em
= lookup_extent_mapping(map
, start
, len
);
4001 write_unlock(&map
->lock
);
4004 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4005 em
->start
!= start
) {
4006 write_unlock(&map
->lock
);
4007 free_extent_map(em
);
4010 if (!test_range_bit(tree
, em
->start
,
4011 extent_map_end(em
) - 1,
4012 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4014 remove_extent_mapping(map
, em
);
4015 /* once for the rb tree */
4016 free_extent_map(em
);
4018 start
= extent_map_end(em
);
4019 write_unlock(&map
->lock
);
4022 free_extent_map(em
);
4025 return try_release_extent_state(map
, tree
, page
, mask
);
4029 * helper function for fiemap, which doesn't want to see any holes.
4030 * This maps until we find something past 'last'
4032 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4035 get_extent_t
*get_extent
)
4037 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4038 struct extent_map
*em
;
4045 len
= last
- offset
;
4048 len
= ALIGN(len
, sectorsize
);
4049 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4050 if (IS_ERR_OR_NULL(em
))
4053 /* if this isn't a hole return it */
4054 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4055 em
->block_start
!= EXTENT_MAP_HOLE
) {
4059 /* this is a hole, advance to the next extent */
4060 offset
= extent_map_end(em
);
4061 free_extent_map(em
);
4068 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4069 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4073 u64 max
= start
+ len
;
4077 u64 last_for_get_extent
= 0;
4079 u64 isize
= i_size_read(inode
);
4080 struct btrfs_key found_key
;
4081 struct extent_map
*em
= NULL
;
4082 struct extent_state
*cached_state
= NULL
;
4083 struct btrfs_path
*path
;
4084 struct btrfs_file_extent_item
*item
;
4089 unsigned long emflags
;
4094 path
= btrfs_alloc_path();
4097 path
->leave_spinning
= 1;
4099 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
4100 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
4103 * lookup the last file extent. We're not using i_size here
4104 * because there might be preallocation past i_size
4106 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
4107 path
, btrfs_ino(inode
), -1, 0);
4109 btrfs_free_path(path
);
4114 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4115 struct btrfs_file_extent_item
);
4116 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4117 found_type
= btrfs_key_type(&found_key
);
4119 /* No extents, but there might be delalloc bits */
4120 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4121 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4122 /* have to trust i_size as the end */
4124 last_for_get_extent
= isize
;
4127 * remember the start of the last extent. There are a
4128 * bunch of different factors that go into the length of the
4129 * extent, so its much less complex to remember where it started
4131 last
= found_key
.offset
;
4132 last_for_get_extent
= last
+ 1;
4134 btrfs_free_path(path
);
4137 * we might have some extents allocated but more delalloc past those
4138 * extents. so, we trust isize unless the start of the last extent is
4143 last_for_get_extent
= isize
;
4146 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1, 0,
4149 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4159 u64 offset_in_extent
= 0;
4161 /* break if the extent we found is outside the range */
4162 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4166 * get_extent may return an extent that starts before our
4167 * requested range. We have to make sure the ranges
4168 * we return to fiemap always move forward and don't
4169 * overlap, so adjust the offsets here
4171 em_start
= max(em
->start
, off
);
4174 * record the offset from the start of the extent
4175 * for adjusting the disk offset below. Only do this if the
4176 * extent isn't compressed since our in ram offset may be past
4177 * what we have actually allocated on disk.
4179 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4180 offset_in_extent
= em_start
- em
->start
;
4181 em_end
= extent_map_end(em
);
4182 em_len
= em_end
- em_start
;
4183 emflags
= em
->flags
;
4188 * bump off for our next call to get_extent
4190 off
= extent_map_end(em
);
4194 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4196 flags
|= FIEMAP_EXTENT_LAST
;
4197 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4198 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4199 FIEMAP_EXTENT_NOT_ALIGNED
);
4200 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4201 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4202 FIEMAP_EXTENT_UNKNOWN
);
4204 disko
= em
->block_start
+ offset_in_extent
;
4206 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4207 flags
|= FIEMAP_EXTENT_ENCODED
;
4209 free_extent_map(em
);
4211 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4212 (last
== (u64
)-1 && isize
<= em_end
)) {
4213 flags
|= FIEMAP_EXTENT_LAST
;
4217 /* now scan forward to see if this is really the last extent. */
4218 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4225 flags
|= FIEMAP_EXTENT_LAST
;
4228 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4234 free_extent_map(em
);
4236 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4237 &cached_state
, GFP_NOFS
);
4241 static void __free_extent_buffer(struct extent_buffer
*eb
)
4243 btrfs_leak_debug_del(&eb
->leak_list
);
4244 kmem_cache_free(extent_buffer_cache
, eb
);
4247 static int extent_buffer_under_io(struct extent_buffer
*eb
)
4249 return (atomic_read(&eb
->io_pages
) ||
4250 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4251 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4255 * Helper for releasing extent buffer page.
4257 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4258 unsigned long start_idx
)
4260 unsigned long index
;
4261 unsigned long num_pages
;
4263 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4265 BUG_ON(extent_buffer_under_io(eb
));
4267 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4268 index
= start_idx
+ num_pages
;
4269 if (start_idx
>= index
)
4274 page
= extent_buffer_page(eb
, index
);
4275 if (page
&& mapped
) {
4276 spin_lock(&page
->mapping
->private_lock
);
4278 * We do this since we'll remove the pages after we've
4279 * removed the eb from the radix tree, so we could race
4280 * and have this page now attached to the new eb. So
4281 * only clear page_private if it's still connected to
4284 if (PagePrivate(page
) &&
4285 page
->private == (unsigned long)eb
) {
4286 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4287 BUG_ON(PageDirty(page
));
4288 BUG_ON(PageWriteback(page
));
4290 * We need to make sure we haven't be attached
4293 ClearPagePrivate(page
);
4294 set_page_private(page
, 0);
4295 /* One for the page private */
4296 page_cache_release(page
);
4298 spin_unlock(&page
->mapping
->private_lock
);
4302 /* One for when we alloced the page */
4303 page_cache_release(page
);
4305 } while (index
!= start_idx
);
4309 * Helper for releasing the extent buffer.
4311 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4313 btrfs_release_extent_buffer_page(eb
, 0);
4314 __free_extent_buffer(eb
);
4317 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
4322 struct extent_buffer
*eb
= NULL
;
4324 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
4331 rwlock_init(&eb
->lock
);
4332 atomic_set(&eb
->write_locks
, 0);
4333 atomic_set(&eb
->read_locks
, 0);
4334 atomic_set(&eb
->blocking_readers
, 0);
4335 atomic_set(&eb
->blocking_writers
, 0);
4336 atomic_set(&eb
->spinning_readers
, 0);
4337 atomic_set(&eb
->spinning_writers
, 0);
4338 eb
->lock_nested
= 0;
4339 init_waitqueue_head(&eb
->write_lock_wq
);
4340 init_waitqueue_head(&eb
->read_lock_wq
);
4342 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4344 spin_lock_init(&eb
->refs_lock
);
4345 atomic_set(&eb
->refs
, 1);
4346 atomic_set(&eb
->io_pages
, 0);
4349 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4351 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4352 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4353 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4358 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4362 struct extent_buffer
*new;
4363 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4365 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_NOFS
);
4369 for (i
= 0; i
< num_pages
; i
++) {
4370 p
= alloc_page(GFP_NOFS
);
4372 btrfs_release_extent_buffer(new);
4375 attach_extent_buffer_page(new, p
);
4376 WARN_ON(PageDirty(p
));
4381 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4382 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4383 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4388 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
4390 struct extent_buffer
*eb
;
4391 unsigned long num_pages
= num_extent_pages(0, len
);
4394 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_NOFS
);
4398 for (i
= 0; i
< num_pages
; i
++) {
4399 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4403 set_extent_buffer_uptodate(eb
);
4404 btrfs_set_header_nritems(eb
, 0);
4405 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4410 __free_page(eb
->pages
[i
- 1]);
4411 __free_extent_buffer(eb
);
4415 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4418 /* the ref bit is tricky. We have to make sure it is set
4419 * if we have the buffer dirty. Otherwise the
4420 * code to free a buffer can end up dropping a dirty
4423 * Once the ref bit is set, it won't go away while the
4424 * buffer is dirty or in writeback, and it also won't
4425 * go away while we have the reference count on the
4428 * We can't just set the ref bit without bumping the
4429 * ref on the eb because free_extent_buffer might
4430 * see the ref bit and try to clear it. If this happens
4431 * free_extent_buffer might end up dropping our original
4432 * ref by mistake and freeing the page before we are able
4433 * to add one more ref.
4435 * So bump the ref count first, then set the bit. If someone
4436 * beat us to it, drop the ref we added.
4438 refs
= atomic_read(&eb
->refs
);
4439 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4442 spin_lock(&eb
->refs_lock
);
4443 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4444 atomic_inc(&eb
->refs
);
4445 spin_unlock(&eb
->refs_lock
);
4448 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
)
4450 unsigned long num_pages
, i
;
4452 check_buffer_tree_ref(eb
);
4454 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4455 for (i
= 0; i
< num_pages
; i
++) {
4456 struct page
*p
= extent_buffer_page(eb
, i
);
4457 mark_page_accessed(p
);
4461 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
4462 u64 start
, unsigned long len
)
4464 unsigned long num_pages
= num_extent_pages(start
, len
);
4466 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4467 struct extent_buffer
*eb
;
4468 struct extent_buffer
*exists
= NULL
;
4470 struct address_space
*mapping
= tree
->mapping
;
4475 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4476 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4478 mark_extent_buffer_accessed(eb
);
4483 eb
= __alloc_extent_buffer(tree
, start
, len
, GFP_NOFS
);
4487 for (i
= 0; i
< num_pages
; i
++, index
++) {
4488 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4492 spin_lock(&mapping
->private_lock
);
4493 if (PagePrivate(p
)) {
4495 * We could have already allocated an eb for this page
4496 * and attached one so lets see if we can get a ref on
4497 * the existing eb, and if we can we know it's good and
4498 * we can just return that one, else we know we can just
4499 * overwrite page->private.
4501 exists
= (struct extent_buffer
*)p
->private;
4502 if (atomic_inc_not_zero(&exists
->refs
)) {
4503 spin_unlock(&mapping
->private_lock
);
4505 page_cache_release(p
);
4506 mark_extent_buffer_accessed(exists
);
4511 * Do this so attach doesn't complain and we need to
4512 * drop the ref the old guy had.
4514 ClearPagePrivate(p
);
4515 WARN_ON(PageDirty(p
));
4516 page_cache_release(p
);
4518 attach_extent_buffer_page(eb
, p
);
4519 spin_unlock(&mapping
->private_lock
);
4520 WARN_ON(PageDirty(p
));
4521 mark_page_accessed(p
);
4523 if (!PageUptodate(p
))
4527 * see below about how we avoid a nasty race with release page
4528 * and why we unlock later
4532 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4534 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4538 spin_lock(&tree
->buffer_lock
);
4539 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
4540 if (ret
== -EEXIST
) {
4541 exists
= radix_tree_lookup(&tree
->buffer
,
4542 start
>> PAGE_CACHE_SHIFT
);
4543 if (!atomic_inc_not_zero(&exists
->refs
)) {
4544 spin_unlock(&tree
->buffer_lock
);
4545 radix_tree_preload_end();
4549 spin_unlock(&tree
->buffer_lock
);
4550 radix_tree_preload_end();
4551 mark_extent_buffer_accessed(exists
);
4554 /* add one reference for the tree */
4555 check_buffer_tree_ref(eb
);
4556 spin_unlock(&tree
->buffer_lock
);
4557 radix_tree_preload_end();
4560 * there is a race where release page may have
4561 * tried to find this extent buffer in the radix
4562 * but failed. It will tell the VM it is safe to
4563 * reclaim the, and it will clear the page private bit.
4564 * We must make sure to set the page private bit properly
4565 * after the extent buffer is in the radix tree so
4566 * it doesn't get lost
4568 SetPageChecked(eb
->pages
[0]);
4569 for (i
= 1; i
< num_pages
; i
++) {
4570 p
= extent_buffer_page(eb
, i
);
4571 ClearPageChecked(p
);
4574 unlock_page(eb
->pages
[0]);
4578 for (i
= 0; i
< num_pages
; i
++) {
4580 unlock_page(eb
->pages
[i
]);
4583 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4584 btrfs_release_extent_buffer(eb
);
4588 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
4589 u64 start
, unsigned long len
)
4591 struct extent_buffer
*eb
;
4594 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4595 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4597 mark_extent_buffer_accessed(eb
);
4605 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4607 struct extent_buffer
*eb
=
4608 container_of(head
, struct extent_buffer
, rcu_head
);
4610 __free_extent_buffer(eb
);
4613 /* Expects to have eb->eb_lock already held */
4614 static int release_extent_buffer(struct extent_buffer
*eb
)
4616 WARN_ON(atomic_read(&eb
->refs
) == 0);
4617 if (atomic_dec_and_test(&eb
->refs
)) {
4618 if (test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
)) {
4619 spin_unlock(&eb
->refs_lock
);
4621 struct extent_io_tree
*tree
= eb
->tree
;
4623 spin_unlock(&eb
->refs_lock
);
4625 spin_lock(&tree
->buffer_lock
);
4626 radix_tree_delete(&tree
->buffer
,
4627 eb
->start
>> PAGE_CACHE_SHIFT
);
4628 spin_unlock(&tree
->buffer_lock
);
4631 /* Should be safe to release our pages at this point */
4632 btrfs_release_extent_buffer_page(eb
, 0);
4633 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4636 spin_unlock(&eb
->refs_lock
);
4641 void free_extent_buffer(struct extent_buffer
*eb
)
4649 refs
= atomic_read(&eb
->refs
);
4652 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
4657 spin_lock(&eb
->refs_lock
);
4658 if (atomic_read(&eb
->refs
) == 2 &&
4659 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4660 atomic_dec(&eb
->refs
);
4662 if (atomic_read(&eb
->refs
) == 2 &&
4663 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4664 !extent_buffer_under_io(eb
) &&
4665 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4666 atomic_dec(&eb
->refs
);
4669 * I know this is terrible, but it's temporary until we stop tracking
4670 * the uptodate bits and such for the extent buffers.
4672 release_extent_buffer(eb
);
4675 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4680 spin_lock(&eb
->refs_lock
);
4681 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4683 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4684 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4685 atomic_dec(&eb
->refs
);
4686 release_extent_buffer(eb
);
4689 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4692 unsigned long num_pages
;
4695 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4697 for (i
= 0; i
< num_pages
; i
++) {
4698 page
= extent_buffer_page(eb
, i
);
4699 if (!PageDirty(page
))
4703 WARN_ON(!PagePrivate(page
));
4705 clear_page_dirty_for_io(page
);
4706 spin_lock_irq(&page
->mapping
->tree_lock
);
4707 if (!PageDirty(page
)) {
4708 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4710 PAGECACHE_TAG_DIRTY
);
4712 spin_unlock_irq(&page
->mapping
->tree_lock
);
4713 ClearPageError(page
);
4716 WARN_ON(atomic_read(&eb
->refs
) == 0);
4719 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4722 unsigned long num_pages
;
4725 check_buffer_tree_ref(eb
);
4727 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4729 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4730 WARN_ON(atomic_read(&eb
->refs
) == 0);
4731 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4733 for (i
= 0; i
< num_pages
; i
++)
4734 set_page_dirty(extent_buffer_page(eb
, i
));
4738 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4742 unsigned long num_pages
;
4744 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4745 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4746 for (i
= 0; i
< num_pages
; i
++) {
4747 page
= extent_buffer_page(eb
, i
);
4749 ClearPageUptodate(page
);
4754 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4758 unsigned long num_pages
;
4760 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4761 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4762 for (i
= 0; i
< num_pages
; i
++) {
4763 page
= extent_buffer_page(eb
, i
);
4764 SetPageUptodate(page
);
4769 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4771 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4774 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4775 struct extent_buffer
*eb
, u64 start
, int wait
,
4776 get_extent_t
*get_extent
, int mirror_num
)
4779 unsigned long start_i
;
4783 int locked_pages
= 0;
4784 int all_uptodate
= 1;
4785 unsigned long num_pages
;
4786 unsigned long num_reads
= 0;
4787 struct bio
*bio
= NULL
;
4788 unsigned long bio_flags
= 0;
4790 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4794 WARN_ON(start
< eb
->start
);
4795 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4796 (eb
->start
>> PAGE_CACHE_SHIFT
);
4801 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4802 for (i
= start_i
; i
< num_pages
; i
++) {
4803 page
= extent_buffer_page(eb
, i
);
4804 if (wait
== WAIT_NONE
) {
4805 if (!trylock_page(page
))
4811 if (!PageUptodate(page
)) {
4818 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4822 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4823 eb
->read_mirror
= 0;
4824 atomic_set(&eb
->io_pages
, num_reads
);
4825 for (i
= start_i
; i
< num_pages
; i
++) {
4826 page
= extent_buffer_page(eb
, i
);
4827 if (!PageUptodate(page
)) {
4828 ClearPageError(page
);
4829 err
= __extent_read_full_page(tree
, page
,
4831 mirror_num
, &bio_flags
,
4841 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
4847 if (ret
|| wait
!= WAIT_COMPLETE
)
4850 for (i
= start_i
; i
< num_pages
; i
++) {
4851 page
= extent_buffer_page(eb
, i
);
4852 wait_on_page_locked(page
);
4853 if (!PageUptodate(page
))
4861 while (locked_pages
> 0) {
4862 page
= extent_buffer_page(eb
, i
);
4870 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
4871 unsigned long start
,
4878 char *dst
= (char *)dstv
;
4879 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4880 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4882 WARN_ON(start
> eb
->len
);
4883 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4885 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
4888 page
= extent_buffer_page(eb
, i
);
4890 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4891 kaddr
= page_address(page
);
4892 memcpy(dst
, kaddr
+ offset
, cur
);
4901 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
4902 unsigned long min_len
, char **map
,
4903 unsigned long *map_start
,
4904 unsigned long *map_len
)
4906 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
4909 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4910 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4911 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
4918 offset
= start_offset
;
4922 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
4925 if (start
+ min_len
> eb
->len
) {
4926 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
4928 eb
->start
, eb
->len
, start
, min_len
);
4932 p
= extent_buffer_page(eb
, i
);
4933 kaddr
= page_address(p
);
4934 *map
= kaddr
+ offset
;
4935 *map_len
= PAGE_CACHE_SIZE
- offset
;
4939 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
4940 unsigned long start
,
4947 char *ptr
= (char *)ptrv
;
4948 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4949 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4952 WARN_ON(start
> eb
->len
);
4953 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4955 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
4958 page
= extent_buffer_page(eb
, i
);
4960 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4962 kaddr
= page_address(page
);
4963 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
4975 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
4976 unsigned long start
, unsigned long len
)
4982 char *src
= (char *)srcv
;
4983 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4984 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4986 WARN_ON(start
> eb
->len
);
4987 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4989 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
4992 page
= extent_buffer_page(eb
, i
);
4993 WARN_ON(!PageUptodate(page
));
4995 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4996 kaddr
= page_address(page
);
4997 memcpy(kaddr
+ offset
, src
, cur
);
5006 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5007 unsigned long start
, unsigned long len
)
5013 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5014 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5016 WARN_ON(start
> eb
->len
);
5017 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5019 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5022 page
= extent_buffer_page(eb
, i
);
5023 WARN_ON(!PageUptodate(page
));
5025 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5026 kaddr
= page_address(page
);
5027 memset(kaddr
+ offset
, c
, cur
);
5035 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5036 unsigned long dst_offset
, unsigned long src_offset
,
5039 u64 dst_len
= dst
->len
;
5044 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5045 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5047 WARN_ON(src
->len
!= dst_len
);
5049 offset
= (start_offset
+ dst_offset
) &
5050 (PAGE_CACHE_SIZE
- 1);
5053 page
= extent_buffer_page(dst
, i
);
5054 WARN_ON(!PageUptodate(page
));
5056 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
5058 kaddr
= page_address(page
);
5059 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5068 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
5069 unsigned long dst_off
, unsigned long src_off
,
5072 char *dst_kaddr
= page_address(dst_page
);
5073 if (dst_page
== src_page
) {
5074 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
5076 char *src_kaddr
= page_address(src_page
);
5077 char *p
= dst_kaddr
+ dst_off
+ len
;
5078 char *s
= src_kaddr
+ src_off
+ len
;
5085 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5087 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5088 return distance
< len
;
5091 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5092 unsigned long dst_off
, unsigned long src_off
,
5095 char *dst_kaddr
= page_address(dst_page
);
5097 int must_memmove
= 0;
5099 if (dst_page
!= src_page
) {
5100 src_kaddr
= page_address(src_page
);
5102 src_kaddr
= dst_kaddr
;
5103 if (areas_overlap(src_off
, dst_off
, len
))
5108 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5110 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5113 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5114 unsigned long src_offset
, unsigned long len
)
5117 size_t dst_off_in_page
;
5118 size_t src_off_in_page
;
5119 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5120 unsigned long dst_i
;
5121 unsigned long src_i
;
5123 if (src_offset
+ len
> dst
->len
) {
5124 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
5125 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
5128 if (dst_offset
+ len
> dst
->len
) {
5129 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
5130 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
5135 dst_off_in_page
= (start_offset
+ dst_offset
) &
5136 (PAGE_CACHE_SIZE
- 1);
5137 src_off_in_page
= (start_offset
+ src_offset
) &
5138 (PAGE_CACHE_SIZE
- 1);
5140 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5141 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5143 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5145 cur
= min_t(unsigned long, cur
,
5146 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5148 copy_pages(extent_buffer_page(dst
, dst_i
),
5149 extent_buffer_page(dst
, src_i
),
5150 dst_off_in_page
, src_off_in_page
, cur
);
5158 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5159 unsigned long src_offset
, unsigned long len
)
5162 size_t dst_off_in_page
;
5163 size_t src_off_in_page
;
5164 unsigned long dst_end
= dst_offset
+ len
- 1;
5165 unsigned long src_end
= src_offset
+ len
- 1;
5166 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5167 unsigned long dst_i
;
5168 unsigned long src_i
;
5170 if (src_offset
+ len
> dst
->len
) {
5171 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
5172 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
5175 if (dst_offset
+ len
> dst
->len
) {
5176 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
5177 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
5180 if (dst_offset
< src_offset
) {
5181 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5185 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5186 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5188 dst_off_in_page
= (start_offset
+ dst_end
) &
5189 (PAGE_CACHE_SIZE
- 1);
5190 src_off_in_page
= (start_offset
+ src_end
) &
5191 (PAGE_CACHE_SIZE
- 1);
5193 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5194 cur
= min(cur
, dst_off_in_page
+ 1);
5195 move_pages(extent_buffer_page(dst
, dst_i
),
5196 extent_buffer_page(dst
, src_i
),
5197 dst_off_in_page
- cur
+ 1,
5198 src_off_in_page
- cur
+ 1, cur
);
5206 int try_release_extent_buffer(struct page
*page
)
5208 struct extent_buffer
*eb
;
5211 * We need to make sure noboody is attaching this page to an eb right
5214 spin_lock(&page
->mapping
->private_lock
);
5215 if (!PagePrivate(page
)) {
5216 spin_unlock(&page
->mapping
->private_lock
);
5220 eb
= (struct extent_buffer
*)page
->private;
5224 * This is a little awful but should be ok, we need to make sure that
5225 * the eb doesn't disappear out from under us while we're looking at
5228 spin_lock(&eb
->refs_lock
);
5229 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5230 spin_unlock(&eb
->refs_lock
);
5231 spin_unlock(&page
->mapping
->private_lock
);
5234 spin_unlock(&page
->mapping
->private_lock
);
5237 * If tree ref isn't set then we know the ref on this eb is a real ref,
5238 * so just return, this page will likely be freed soon anyway.
5240 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5241 spin_unlock(&eb
->refs_lock
);
5245 return release_extent_buffer(eb
);