1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
24 static struct kmem_cache
*extent_state_cache
;
25 static struct kmem_cache
*extent_buffer_cache
;
26 static struct bio_set
*btrfs_bioset
;
28 #ifdef CONFIG_BTRFS_DEBUG
29 static LIST_HEAD(buffers
);
30 static LIST_HEAD(states
);
32 static DEFINE_SPINLOCK(leak_lock
);
35 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
39 spin_lock_irqsave(&leak_lock
, flags
);
41 spin_unlock_irqrestore(&leak_lock
, flags
);
45 void btrfs_leak_debug_del(struct list_head
*entry
)
49 spin_lock_irqsave(&leak_lock
, flags
);
51 spin_unlock_irqrestore(&leak_lock
, flags
);
55 void btrfs_leak_debug_check(void)
57 struct extent_state
*state
;
58 struct extent_buffer
*eb
;
60 while (!list_empty(&states
)) {
61 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
62 printk(KERN_ERR
"BTRFS: state leak: start %llu end %llu "
63 "state %lu in tree %p refs %d\n",
64 state
->start
, state
->end
, state
->state
, state
->tree
,
65 atomic_read(&state
->refs
));
66 list_del(&state
->leak_list
);
67 kmem_cache_free(extent_state_cache
, state
);
70 while (!list_empty(&buffers
)) {
71 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
72 printk(KERN_ERR
"BTRFS: buffer leak start %llu len %lu "
74 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
75 list_del(&eb
->leak_list
);
76 kmem_cache_free(extent_buffer_cache
, eb
);
80 #define btrfs_debug_check_extent_io_range(tree, start, end) \
81 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
82 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
83 struct extent_io_tree
*tree
, u64 start
, u64 end
)
91 inode
= tree
->mapping
->host
;
92 isize
= i_size_read(inode
);
93 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
94 printk_ratelimited(KERN_DEBUG
95 "BTRFS: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
96 caller
, btrfs_ino(inode
), isize
, start
, end
);
100 #define btrfs_leak_debug_add(new, head) do {} while (0)
101 #define btrfs_leak_debug_del(entry) do {} while (0)
102 #define btrfs_leak_debug_check() do {} while (0)
103 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
106 #define BUFFER_LRU_MAX 64
111 struct rb_node rb_node
;
114 struct extent_page_data
{
116 struct extent_io_tree
*tree
;
117 get_extent_t
*get_extent
;
118 unsigned long bio_flags
;
120 /* tells writepage not to lock the state bits for this range
121 * it still does the unlocking
123 unsigned int extent_locked
:1;
125 /* tells the submit_bio code to use a WRITE_SYNC */
126 unsigned int sync_io
:1;
129 static noinline
void flush_write_bio(void *data
);
130 static inline struct btrfs_fs_info
*
131 tree_fs_info(struct extent_io_tree
*tree
)
135 return btrfs_sb(tree
->mapping
->host
->i_sb
);
138 int __init
extent_io_init(void)
140 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
141 sizeof(struct extent_state
), 0,
142 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
143 if (!extent_state_cache
)
146 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
147 sizeof(struct extent_buffer
), 0,
148 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
149 if (!extent_buffer_cache
)
150 goto free_state_cache
;
152 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
153 offsetof(struct btrfs_io_bio
, bio
));
155 goto free_buffer_cache
;
157 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
163 bioset_free(btrfs_bioset
);
167 kmem_cache_destroy(extent_buffer_cache
);
168 extent_buffer_cache
= NULL
;
171 kmem_cache_destroy(extent_state_cache
);
172 extent_state_cache
= NULL
;
176 void extent_io_exit(void)
178 btrfs_leak_debug_check();
181 * Make sure all delayed rcu free are flushed before we
185 if (extent_state_cache
)
186 kmem_cache_destroy(extent_state_cache
);
187 if (extent_buffer_cache
)
188 kmem_cache_destroy(extent_buffer_cache
);
190 bioset_free(btrfs_bioset
);
193 void extent_io_tree_init(struct extent_io_tree
*tree
,
194 struct address_space
*mapping
)
196 tree
->state
= RB_ROOT
;
198 tree
->dirty_bytes
= 0;
199 spin_lock_init(&tree
->lock
);
200 tree
->mapping
= mapping
;
203 static struct extent_state
*alloc_extent_state(gfp_t mask
)
205 struct extent_state
*state
;
207 state
= kmem_cache_alloc(extent_state_cache
, mask
);
213 btrfs_leak_debug_add(&state
->leak_list
, &states
);
214 atomic_set(&state
->refs
, 1);
215 init_waitqueue_head(&state
->wq
);
216 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
220 void free_extent_state(struct extent_state
*state
)
224 if (atomic_dec_and_test(&state
->refs
)) {
225 WARN_ON(state
->tree
);
226 btrfs_leak_debug_del(&state
->leak_list
);
227 trace_free_extent_state(state
, _RET_IP_
);
228 kmem_cache_free(extent_state_cache
, state
);
232 static struct rb_node
*tree_insert(struct rb_root
*root
,
233 struct rb_node
*search_start
,
235 struct rb_node
*node
,
236 struct rb_node
***p_in
,
237 struct rb_node
**parent_in
)
240 struct rb_node
*parent
= NULL
;
241 struct tree_entry
*entry
;
243 if (p_in
&& parent_in
) {
249 p
= search_start
? &search_start
: &root
->rb_node
;
252 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
254 if (offset
< entry
->start
)
256 else if (offset
> entry
->end
)
263 rb_link_node(node
, parent
, p
);
264 rb_insert_color(node
, root
);
268 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
269 struct rb_node
**prev_ret
,
270 struct rb_node
**next_ret
,
271 struct rb_node
***p_ret
,
272 struct rb_node
**parent_ret
)
274 struct rb_root
*root
= &tree
->state
;
275 struct rb_node
**n
= &root
->rb_node
;
276 struct rb_node
*prev
= NULL
;
277 struct rb_node
*orig_prev
= NULL
;
278 struct tree_entry
*entry
;
279 struct tree_entry
*prev_entry
= NULL
;
283 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
286 if (offset
< entry
->start
)
288 else if (offset
> entry
->end
)
301 while (prev
&& offset
> prev_entry
->end
) {
302 prev
= rb_next(prev
);
303 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
310 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
311 while (prev
&& offset
< prev_entry
->start
) {
312 prev
= rb_prev(prev
);
313 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
320 static inline struct rb_node
*
321 tree_search_for_insert(struct extent_io_tree
*tree
,
323 struct rb_node
***p_ret
,
324 struct rb_node
**parent_ret
)
326 struct rb_node
*prev
= NULL
;
329 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
335 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
338 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
341 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
342 struct extent_state
*other
)
344 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
345 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
350 * utility function to look for merge candidates inside a given range.
351 * Any extents with matching state are merged together into a single
352 * extent in the tree. Extents with EXTENT_IO in their state field
353 * are not merged because the end_io handlers need to be able to do
354 * operations on them without sleeping (or doing allocations/splits).
356 * This should be called with the tree lock held.
358 static void merge_state(struct extent_io_tree
*tree
,
359 struct extent_state
*state
)
361 struct extent_state
*other
;
362 struct rb_node
*other_node
;
364 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
367 other_node
= rb_prev(&state
->rb_node
);
369 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
370 if (other
->end
== state
->start
- 1 &&
371 other
->state
== state
->state
) {
372 merge_cb(tree
, state
, other
);
373 state
->start
= other
->start
;
375 rb_erase(&other
->rb_node
, &tree
->state
);
376 free_extent_state(other
);
379 other_node
= rb_next(&state
->rb_node
);
381 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
382 if (other
->start
== state
->end
+ 1 &&
383 other
->state
== state
->state
) {
384 merge_cb(tree
, state
, other
);
385 state
->end
= other
->end
;
387 rb_erase(&other
->rb_node
, &tree
->state
);
388 free_extent_state(other
);
393 static void set_state_cb(struct extent_io_tree
*tree
,
394 struct extent_state
*state
, unsigned long *bits
)
396 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
397 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
400 static void clear_state_cb(struct extent_io_tree
*tree
,
401 struct extent_state
*state
, unsigned long *bits
)
403 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
404 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
407 static void set_state_bits(struct extent_io_tree
*tree
,
408 struct extent_state
*state
, unsigned long *bits
);
411 * insert an extent_state struct into the tree. 'bits' are set on the
412 * struct before it is inserted.
414 * This may return -EEXIST if the extent is already there, in which case the
415 * state struct is freed.
417 * The tree lock is not taken internally. This is a utility function and
418 * probably isn't what you want to call (see set/clear_extent_bit).
420 static int insert_state(struct extent_io_tree
*tree
,
421 struct extent_state
*state
, u64 start
, u64 end
,
423 struct rb_node
**parent
,
426 struct rb_node
*node
;
429 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
431 state
->start
= start
;
434 set_state_bits(tree
, state
, bits
);
436 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
438 struct extent_state
*found
;
439 found
= rb_entry(node
, struct extent_state
, rb_node
);
440 printk(KERN_ERR
"BTRFS: found node %llu %llu on insert of "
442 found
->start
, found
->end
, start
, end
);
446 merge_state(tree
, state
);
450 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
453 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
454 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
458 * split a given extent state struct in two, inserting the preallocated
459 * struct 'prealloc' as the newly created second half. 'split' indicates an
460 * offset inside 'orig' where it should be split.
463 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
464 * are two extent state structs in the tree:
465 * prealloc: [orig->start, split - 1]
466 * orig: [ split, orig->end ]
468 * The tree locks are not taken by this function. They need to be held
471 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
472 struct extent_state
*prealloc
, u64 split
)
474 struct rb_node
*node
;
476 split_cb(tree
, orig
, split
);
478 prealloc
->start
= orig
->start
;
479 prealloc
->end
= split
- 1;
480 prealloc
->state
= orig
->state
;
483 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
484 &prealloc
->rb_node
, NULL
, NULL
);
486 free_extent_state(prealloc
);
489 prealloc
->tree
= tree
;
493 static struct extent_state
*next_state(struct extent_state
*state
)
495 struct rb_node
*next
= rb_next(&state
->rb_node
);
497 return rb_entry(next
, struct extent_state
, rb_node
);
503 * utility function to clear some bits in an extent state struct.
504 * it will optionally wake up any one waiting on this state (wake == 1).
506 * If no bits are set on the state struct after clearing things, the
507 * struct is freed and removed from the tree
509 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
510 struct extent_state
*state
,
511 unsigned long *bits
, int wake
)
513 struct extent_state
*next
;
514 unsigned long bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
516 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
517 u64 range
= state
->end
- state
->start
+ 1;
518 WARN_ON(range
> tree
->dirty_bytes
);
519 tree
->dirty_bytes
-= range
;
521 clear_state_cb(tree
, state
, bits
);
522 state
->state
&= ~bits_to_clear
;
525 if (state
->state
== 0) {
526 next
= next_state(state
);
528 rb_erase(&state
->rb_node
, &tree
->state
);
530 free_extent_state(state
);
535 merge_state(tree
, state
);
536 next
= next_state(state
);
541 static struct extent_state
*
542 alloc_extent_state_atomic(struct extent_state
*prealloc
)
545 prealloc
= alloc_extent_state(GFP_ATOMIC
);
550 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
552 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
553 "Extent tree was modified by another "
554 "thread while locked.");
558 * clear some bits on a range in the tree. This may require splitting
559 * or inserting elements in the tree, so the gfp mask is used to
560 * indicate which allocations or sleeping are allowed.
562 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
563 * the given range from the tree regardless of state (ie for truncate).
565 * the range [start, end] is inclusive.
567 * This takes the tree lock, and returns 0 on success and < 0 on error.
569 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
570 unsigned long bits
, int wake
, int delete,
571 struct extent_state
**cached_state
,
574 struct extent_state
*state
;
575 struct extent_state
*cached
;
576 struct extent_state
*prealloc
= NULL
;
577 struct rb_node
*node
;
582 btrfs_debug_check_extent_io_range(tree
, start
, end
);
584 if (bits
& EXTENT_DELALLOC
)
585 bits
|= EXTENT_NORESERVE
;
588 bits
|= ~EXTENT_CTLBITS
;
589 bits
|= EXTENT_FIRST_DELALLOC
;
591 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
594 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
595 prealloc
= alloc_extent_state(mask
);
600 spin_lock(&tree
->lock
);
602 cached
= *cached_state
;
605 *cached_state
= NULL
;
609 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
610 cached
->end
> start
) {
612 atomic_dec(&cached
->refs
);
617 free_extent_state(cached
);
620 * this search will find the extents that end after
623 node
= tree_search(tree
, start
);
626 state
= rb_entry(node
, struct extent_state
, rb_node
);
628 if (state
->start
> end
)
630 WARN_ON(state
->end
< start
);
631 last_end
= state
->end
;
633 /* the state doesn't have the wanted bits, go ahead */
634 if (!(state
->state
& bits
)) {
635 state
= next_state(state
);
640 * | ---- desired range ---- |
642 * | ------------- state -------------- |
644 * We need to split the extent we found, and may flip
645 * bits on second half.
647 * If the extent we found extends past our range, we
648 * just split and search again. It'll get split again
649 * the next time though.
651 * If the extent we found is inside our range, we clear
652 * the desired bit on it.
655 if (state
->start
< start
) {
656 prealloc
= alloc_extent_state_atomic(prealloc
);
658 err
= split_state(tree
, state
, prealloc
, start
);
660 extent_io_tree_panic(tree
, err
);
665 if (state
->end
<= end
) {
666 state
= clear_state_bit(tree
, state
, &bits
, wake
);
672 * | ---- desired range ---- |
674 * We need to split the extent, and clear the bit
677 if (state
->start
<= end
&& state
->end
> end
) {
678 prealloc
= alloc_extent_state_atomic(prealloc
);
680 err
= split_state(tree
, state
, prealloc
, end
+ 1);
682 extent_io_tree_panic(tree
, err
);
687 clear_state_bit(tree
, prealloc
, &bits
, wake
);
693 state
= clear_state_bit(tree
, state
, &bits
, wake
);
695 if (last_end
== (u64
)-1)
697 start
= last_end
+ 1;
698 if (start
<= end
&& state
&& !need_resched())
703 spin_unlock(&tree
->lock
);
705 free_extent_state(prealloc
);
712 spin_unlock(&tree
->lock
);
713 if (mask
& __GFP_WAIT
)
718 static void wait_on_state(struct extent_io_tree
*tree
,
719 struct extent_state
*state
)
720 __releases(tree
->lock
)
721 __acquires(tree
->lock
)
724 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
725 spin_unlock(&tree
->lock
);
727 spin_lock(&tree
->lock
);
728 finish_wait(&state
->wq
, &wait
);
732 * waits for one or more bits to clear on a range in the state tree.
733 * The range [start, end] is inclusive.
734 * The tree lock is taken by this function
736 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
739 struct extent_state
*state
;
740 struct rb_node
*node
;
742 btrfs_debug_check_extent_io_range(tree
, start
, end
);
744 spin_lock(&tree
->lock
);
748 * this search will find all the extents that end after
751 node
= tree_search(tree
, start
);
755 state
= rb_entry(node
, struct extent_state
, rb_node
);
757 if (state
->start
> end
)
760 if (state
->state
& bits
) {
761 start
= state
->start
;
762 atomic_inc(&state
->refs
);
763 wait_on_state(tree
, state
);
764 free_extent_state(state
);
767 start
= state
->end
+ 1;
772 cond_resched_lock(&tree
->lock
);
775 spin_unlock(&tree
->lock
);
778 static void set_state_bits(struct extent_io_tree
*tree
,
779 struct extent_state
*state
,
782 unsigned long bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
784 set_state_cb(tree
, state
, bits
);
785 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
786 u64 range
= state
->end
- state
->start
+ 1;
787 tree
->dirty_bytes
+= range
;
789 state
->state
|= bits_to_set
;
792 static void cache_state(struct extent_state
*state
,
793 struct extent_state
**cached_ptr
)
795 if (cached_ptr
&& !(*cached_ptr
)) {
796 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
798 atomic_inc(&state
->refs
);
804 * set some bits on a range in the tree. This may require allocations or
805 * sleeping, so the gfp mask is used to indicate what is allowed.
807 * If any of the exclusive bits are set, this will fail with -EEXIST if some
808 * part of the range already has the desired bits set. The start of the
809 * existing range is returned in failed_start in this case.
811 * [start, end] is inclusive This takes the tree lock.
814 static int __must_check
815 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
816 unsigned long bits
, unsigned long exclusive_bits
,
817 u64
*failed_start
, struct extent_state
**cached_state
,
820 struct extent_state
*state
;
821 struct extent_state
*prealloc
= NULL
;
822 struct rb_node
*node
;
824 struct rb_node
*parent
;
829 btrfs_debug_check_extent_io_range(tree
, start
, end
);
831 bits
|= EXTENT_FIRST_DELALLOC
;
833 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
834 prealloc
= alloc_extent_state(mask
);
838 spin_lock(&tree
->lock
);
839 if (cached_state
&& *cached_state
) {
840 state
= *cached_state
;
841 if (state
->start
<= start
&& state
->end
> start
&&
843 node
= &state
->rb_node
;
848 * this search will find all the extents that end after
851 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
853 prealloc
= alloc_extent_state_atomic(prealloc
);
855 err
= insert_state(tree
, prealloc
, start
, end
,
858 extent_io_tree_panic(tree
, err
);
860 cache_state(prealloc
, cached_state
);
864 state
= rb_entry(node
, struct extent_state
, rb_node
);
866 last_start
= state
->start
;
867 last_end
= state
->end
;
870 * | ---- desired range ---- |
873 * Just lock what we found and keep going
875 if (state
->start
== start
&& state
->end
<= end
) {
876 if (state
->state
& exclusive_bits
) {
877 *failed_start
= state
->start
;
882 set_state_bits(tree
, state
, &bits
);
883 cache_state(state
, cached_state
);
884 merge_state(tree
, state
);
885 if (last_end
== (u64
)-1)
887 start
= last_end
+ 1;
888 state
= next_state(state
);
889 if (start
< end
&& state
&& state
->start
== start
&&
896 * | ---- desired range ---- |
899 * | ------------- state -------------- |
901 * We need to split the extent we found, and may flip bits on
904 * If the extent we found extends past our
905 * range, we just split and search again. It'll get split
906 * again the next time though.
908 * If the extent we found is inside our range, we set the
911 if (state
->start
< start
) {
912 if (state
->state
& exclusive_bits
) {
913 *failed_start
= start
;
918 prealloc
= alloc_extent_state_atomic(prealloc
);
920 err
= split_state(tree
, state
, prealloc
, start
);
922 extent_io_tree_panic(tree
, err
);
927 if (state
->end
<= end
) {
928 set_state_bits(tree
, state
, &bits
);
929 cache_state(state
, cached_state
);
930 merge_state(tree
, state
);
931 if (last_end
== (u64
)-1)
933 start
= last_end
+ 1;
934 state
= next_state(state
);
935 if (start
< end
&& state
&& state
->start
== start
&&
942 * | ---- desired range ---- |
943 * | state | or | state |
945 * There's a hole, we need to insert something in it and
946 * ignore the extent we found.
948 if (state
->start
> start
) {
950 if (end
< last_start
)
953 this_end
= last_start
- 1;
955 prealloc
= alloc_extent_state_atomic(prealloc
);
959 * Avoid to free 'prealloc' if it can be merged with
962 err
= insert_state(tree
, prealloc
, start
, this_end
,
965 extent_io_tree_panic(tree
, err
);
967 cache_state(prealloc
, cached_state
);
969 start
= this_end
+ 1;
973 * | ---- desired range ---- |
975 * We need to split the extent, and set the bit
978 if (state
->start
<= end
&& state
->end
> end
) {
979 if (state
->state
& exclusive_bits
) {
980 *failed_start
= start
;
985 prealloc
= alloc_extent_state_atomic(prealloc
);
987 err
= split_state(tree
, state
, prealloc
, end
+ 1);
989 extent_io_tree_panic(tree
, err
);
991 set_state_bits(tree
, prealloc
, &bits
);
992 cache_state(prealloc
, cached_state
);
993 merge_state(tree
, prealloc
);
1001 spin_unlock(&tree
->lock
);
1003 free_extent_state(prealloc
);
1010 spin_unlock(&tree
->lock
);
1011 if (mask
& __GFP_WAIT
)
1016 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1017 unsigned long bits
, u64
* failed_start
,
1018 struct extent_state
**cached_state
, gfp_t mask
)
1020 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1021 cached_state
, mask
);
1026 * convert_extent_bit - convert all bits in a given range from one bit to
1028 * @tree: the io tree to search
1029 * @start: the start offset in bytes
1030 * @end: the end offset in bytes (inclusive)
1031 * @bits: the bits to set in this range
1032 * @clear_bits: the bits to clear in this range
1033 * @cached_state: state that we're going to cache
1034 * @mask: the allocation mask
1036 * This will go through and set bits for the given range. If any states exist
1037 * already in this range they are set with the given bit and cleared of the
1038 * clear_bits. This is only meant to be used by things that are mergeable, ie
1039 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1040 * boundary bits like LOCK.
1042 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1043 unsigned long bits
, unsigned long clear_bits
,
1044 struct extent_state
**cached_state
, gfp_t mask
)
1046 struct extent_state
*state
;
1047 struct extent_state
*prealloc
= NULL
;
1048 struct rb_node
*node
;
1050 struct rb_node
*parent
;
1055 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1058 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
1059 prealloc
= alloc_extent_state(mask
);
1064 spin_lock(&tree
->lock
);
1065 if (cached_state
&& *cached_state
) {
1066 state
= *cached_state
;
1067 if (state
->start
<= start
&& state
->end
> start
&&
1069 node
= &state
->rb_node
;
1075 * this search will find all the extents that end after
1078 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1080 prealloc
= alloc_extent_state_atomic(prealloc
);
1085 err
= insert_state(tree
, prealloc
, start
, end
,
1086 &p
, &parent
, &bits
);
1088 extent_io_tree_panic(tree
, err
);
1089 cache_state(prealloc
, cached_state
);
1093 state
= rb_entry(node
, struct extent_state
, rb_node
);
1095 last_start
= state
->start
;
1096 last_end
= state
->end
;
1099 * | ---- desired range ---- |
1102 * Just lock what we found and keep going
1104 if (state
->start
== start
&& state
->end
<= end
) {
1105 set_state_bits(tree
, state
, &bits
);
1106 cache_state(state
, cached_state
);
1107 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1108 if (last_end
== (u64
)-1)
1110 start
= last_end
+ 1;
1111 if (start
< end
&& state
&& state
->start
== start
&&
1118 * | ---- desired range ---- |
1121 * | ------------- state -------------- |
1123 * We need to split the extent we found, and may flip bits on
1126 * If the extent we found extends past our
1127 * range, we just split and search again. It'll get split
1128 * again the next time though.
1130 * If the extent we found is inside our range, we set the
1131 * desired bit on it.
1133 if (state
->start
< start
) {
1134 prealloc
= alloc_extent_state_atomic(prealloc
);
1139 err
= split_state(tree
, state
, prealloc
, start
);
1141 extent_io_tree_panic(tree
, err
);
1145 if (state
->end
<= end
) {
1146 set_state_bits(tree
, state
, &bits
);
1147 cache_state(state
, cached_state
);
1148 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1149 if (last_end
== (u64
)-1)
1151 start
= last_end
+ 1;
1152 if (start
< end
&& state
&& state
->start
== start
&&
1159 * | ---- desired range ---- |
1160 * | state | or | state |
1162 * There's a hole, we need to insert something in it and
1163 * ignore the extent we found.
1165 if (state
->start
> start
) {
1167 if (end
< last_start
)
1170 this_end
= last_start
- 1;
1172 prealloc
= alloc_extent_state_atomic(prealloc
);
1179 * Avoid to free 'prealloc' if it can be merged with
1182 err
= insert_state(tree
, prealloc
, start
, this_end
,
1185 extent_io_tree_panic(tree
, err
);
1186 cache_state(prealloc
, cached_state
);
1188 start
= this_end
+ 1;
1192 * | ---- desired range ---- |
1194 * We need to split the extent, and set the bit
1197 if (state
->start
<= end
&& state
->end
> end
) {
1198 prealloc
= alloc_extent_state_atomic(prealloc
);
1204 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1206 extent_io_tree_panic(tree
, err
);
1208 set_state_bits(tree
, prealloc
, &bits
);
1209 cache_state(prealloc
, cached_state
);
1210 clear_state_bit(tree
, prealloc
, &clear_bits
, 0);
1218 spin_unlock(&tree
->lock
);
1220 free_extent_state(prealloc
);
1227 spin_unlock(&tree
->lock
);
1228 if (mask
& __GFP_WAIT
)
1233 /* wrappers around set/clear extent bit */
1234 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1237 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, NULL
,
1241 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1242 unsigned long bits
, gfp_t mask
)
1244 return set_extent_bit(tree
, start
, end
, bits
, NULL
,
1248 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1249 unsigned long bits
, gfp_t mask
)
1251 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
1254 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1255 struct extent_state
**cached_state
, gfp_t mask
)
1257 return set_extent_bit(tree
, start
, end
,
1258 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1259 NULL
, cached_state
, mask
);
1262 int set_extent_defrag(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1263 struct extent_state
**cached_state
, gfp_t mask
)
1265 return set_extent_bit(tree
, start
, end
,
1266 EXTENT_DELALLOC
| EXTENT_UPTODATE
| EXTENT_DEFRAG
,
1267 NULL
, cached_state
, mask
);
1270 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1273 return clear_extent_bit(tree
, start
, end
,
1274 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1275 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1278 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1281 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, NULL
,
1285 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1286 struct extent_state
**cached_state
, gfp_t mask
)
1288 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, NULL
,
1289 cached_state
, mask
);
1292 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1293 struct extent_state
**cached_state
, gfp_t mask
)
1295 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1296 cached_state
, mask
);
1300 * either insert or lock state struct between start and end use mask to tell
1301 * us if waiting is desired.
1303 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1304 unsigned long bits
, struct extent_state
**cached_state
)
1309 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1310 EXTENT_LOCKED
, &failed_start
,
1311 cached_state
, GFP_NOFS
);
1312 if (err
== -EEXIST
) {
1313 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1314 start
= failed_start
;
1317 WARN_ON(start
> end
);
1322 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1324 return lock_extent_bits(tree
, start
, end
, 0, NULL
);
1327 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1332 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1333 &failed_start
, NULL
, GFP_NOFS
);
1334 if (err
== -EEXIST
) {
1335 if (failed_start
> start
)
1336 clear_extent_bit(tree
, start
, failed_start
- 1,
1337 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1343 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1344 struct extent_state
**cached
, gfp_t mask
)
1346 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1350 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1352 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1356 int extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1358 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1359 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1362 while (index
<= end_index
) {
1363 page
= find_get_page(inode
->i_mapping
, index
);
1364 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1365 clear_page_dirty_for_io(page
);
1366 page_cache_release(page
);
1372 int extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1374 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1375 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1378 while (index
<= end_index
) {
1379 page
= find_get_page(inode
->i_mapping
, index
);
1380 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1381 account_page_redirty(page
);
1382 __set_page_dirty_nobuffers(page
);
1383 page_cache_release(page
);
1390 * helper function to set both pages and extents in the tree writeback
1392 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1394 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1395 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1398 while (index
<= end_index
) {
1399 page
= find_get_page(tree
->mapping
, index
);
1400 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1401 set_page_writeback(page
);
1402 page_cache_release(page
);
1408 /* find the first state struct with 'bits' set after 'start', and
1409 * return it. tree->lock must be held. NULL will returned if
1410 * nothing was found after 'start'
1412 static struct extent_state
*
1413 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1414 u64 start
, unsigned long bits
)
1416 struct rb_node
*node
;
1417 struct extent_state
*state
;
1420 * this search will find all the extents that end after
1423 node
= tree_search(tree
, start
);
1428 state
= rb_entry(node
, struct extent_state
, rb_node
);
1429 if (state
->end
>= start
&& (state
->state
& bits
))
1432 node
= rb_next(node
);
1441 * find the first offset in the io tree with 'bits' set. zero is
1442 * returned if we find something, and *start_ret and *end_ret are
1443 * set to reflect the state struct that was found.
1445 * If nothing was found, 1 is returned. If found something, return 0.
1447 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1448 u64
*start_ret
, u64
*end_ret
, unsigned long bits
,
1449 struct extent_state
**cached_state
)
1451 struct extent_state
*state
;
1455 spin_lock(&tree
->lock
);
1456 if (cached_state
&& *cached_state
) {
1457 state
= *cached_state
;
1458 if (state
->end
== start
- 1 && state
->tree
) {
1459 n
= rb_next(&state
->rb_node
);
1461 state
= rb_entry(n
, struct extent_state
,
1463 if (state
->state
& bits
)
1467 free_extent_state(*cached_state
);
1468 *cached_state
= NULL
;
1471 free_extent_state(*cached_state
);
1472 *cached_state
= NULL
;
1475 state
= find_first_extent_bit_state(tree
, start
, bits
);
1478 cache_state(state
, cached_state
);
1479 *start_ret
= state
->start
;
1480 *end_ret
= state
->end
;
1484 spin_unlock(&tree
->lock
);
1489 * find a contiguous range of bytes in the file marked as delalloc, not
1490 * more than 'max_bytes'. start and end are used to return the range,
1492 * 1 is returned if we find something, 0 if nothing was in the tree
1494 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1495 u64
*start
, u64
*end
, u64 max_bytes
,
1496 struct extent_state
**cached_state
)
1498 struct rb_node
*node
;
1499 struct extent_state
*state
;
1500 u64 cur_start
= *start
;
1502 u64 total_bytes
= 0;
1504 spin_lock(&tree
->lock
);
1507 * this search will find all the extents that end after
1510 node
= tree_search(tree
, cur_start
);
1518 state
= rb_entry(node
, struct extent_state
, rb_node
);
1519 if (found
&& (state
->start
!= cur_start
||
1520 (state
->state
& EXTENT_BOUNDARY
))) {
1523 if (!(state
->state
& EXTENT_DELALLOC
)) {
1529 *start
= state
->start
;
1530 *cached_state
= state
;
1531 atomic_inc(&state
->refs
);
1535 cur_start
= state
->end
+ 1;
1536 node
= rb_next(node
);
1537 total_bytes
+= state
->end
- state
->start
+ 1;
1538 if (total_bytes
>= max_bytes
)
1544 spin_unlock(&tree
->lock
);
1548 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1549 struct page
*locked_page
,
1553 struct page
*pages
[16];
1554 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1555 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1556 unsigned long nr_pages
= end_index
- index
+ 1;
1559 if (index
== locked_page
->index
&& end_index
== index
)
1562 while (nr_pages
> 0) {
1563 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1564 min_t(unsigned long, nr_pages
,
1565 ARRAY_SIZE(pages
)), pages
);
1566 for (i
= 0; i
< ret
; i
++) {
1567 if (pages
[i
] != locked_page
)
1568 unlock_page(pages
[i
]);
1569 page_cache_release(pages
[i
]);
1577 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1578 struct page
*locked_page
,
1582 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1583 unsigned long start_index
= index
;
1584 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1585 unsigned long pages_locked
= 0;
1586 struct page
*pages
[16];
1587 unsigned long nrpages
;
1591 /* the caller is responsible for locking the start index */
1592 if (index
== locked_page
->index
&& index
== end_index
)
1595 /* skip the page at the start index */
1596 nrpages
= end_index
- index
+ 1;
1597 while (nrpages
> 0) {
1598 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1599 min_t(unsigned long,
1600 nrpages
, ARRAY_SIZE(pages
)), pages
);
1605 /* now we have an array of pages, lock them all */
1606 for (i
= 0; i
< ret
; i
++) {
1608 * the caller is taking responsibility for
1611 if (pages
[i
] != locked_page
) {
1612 lock_page(pages
[i
]);
1613 if (!PageDirty(pages
[i
]) ||
1614 pages
[i
]->mapping
!= inode
->i_mapping
) {
1616 unlock_page(pages
[i
]);
1617 page_cache_release(pages
[i
]);
1621 page_cache_release(pages
[i
]);
1630 if (ret
&& pages_locked
) {
1631 __unlock_for_delalloc(inode
, locked_page
,
1633 ((u64
)(start_index
+ pages_locked
- 1)) <<
1640 * find a contiguous range of bytes in the file marked as delalloc, not
1641 * more than 'max_bytes'. start and end are used to return the range,
1643 * 1 is returned if we find something, 0 if nothing was in the tree
1645 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1646 struct extent_io_tree
*tree
,
1647 struct page
*locked_page
, u64
*start
,
1648 u64
*end
, u64 max_bytes
)
1653 struct extent_state
*cached_state
= NULL
;
1658 /* step one, find a bunch of delalloc bytes starting at start */
1659 delalloc_start
= *start
;
1661 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1662 max_bytes
, &cached_state
);
1663 if (!found
|| delalloc_end
<= *start
) {
1664 *start
= delalloc_start
;
1665 *end
= delalloc_end
;
1666 free_extent_state(cached_state
);
1671 * start comes from the offset of locked_page. We have to lock
1672 * pages in order, so we can't process delalloc bytes before
1675 if (delalloc_start
< *start
)
1676 delalloc_start
= *start
;
1679 * make sure to limit the number of pages we try to lock down
1681 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1682 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1684 /* step two, lock all the pages after the page that has start */
1685 ret
= lock_delalloc_pages(inode
, locked_page
,
1686 delalloc_start
, delalloc_end
);
1687 if (ret
== -EAGAIN
) {
1688 /* some of the pages are gone, lets avoid looping by
1689 * shortening the size of the delalloc range we're searching
1691 free_extent_state(cached_state
);
1693 max_bytes
= PAGE_CACHE_SIZE
;
1701 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1703 /* step three, lock the state bits for the whole range */
1704 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1706 /* then test to make sure it is all still delalloc */
1707 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1708 EXTENT_DELALLOC
, 1, cached_state
);
1710 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1711 &cached_state
, GFP_NOFS
);
1712 __unlock_for_delalloc(inode
, locked_page
,
1713 delalloc_start
, delalloc_end
);
1717 free_extent_state(cached_state
);
1718 *start
= delalloc_start
;
1719 *end
= delalloc_end
;
1724 int extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1725 struct page
*locked_page
,
1726 unsigned long clear_bits
,
1727 unsigned long page_ops
)
1729 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1731 struct page
*pages
[16];
1732 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1733 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1734 unsigned long nr_pages
= end_index
- index
+ 1;
1737 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1741 while (nr_pages
> 0) {
1742 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1743 min_t(unsigned long,
1744 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1745 for (i
= 0; i
< ret
; i
++) {
1747 if (page_ops
& PAGE_SET_PRIVATE2
)
1748 SetPagePrivate2(pages
[i
]);
1750 if (pages
[i
] == locked_page
) {
1751 page_cache_release(pages
[i
]);
1754 if (page_ops
& PAGE_CLEAR_DIRTY
)
1755 clear_page_dirty_for_io(pages
[i
]);
1756 if (page_ops
& PAGE_SET_WRITEBACK
)
1757 set_page_writeback(pages
[i
]);
1758 if (page_ops
& PAGE_END_WRITEBACK
)
1759 end_page_writeback(pages
[i
]);
1760 if (page_ops
& PAGE_UNLOCK
)
1761 unlock_page(pages
[i
]);
1762 page_cache_release(pages
[i
]);
1772 * count the number of bytes in the tree that have a given bit(s)
1773 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1774 * cached. The total number found is returned.
1776 u64
count_range_bits(struct extent_io_tree
*tree
,
1777 u64
*start
, u64 search_end
, u64 max_bytes
,
1778 unsigned long bits
, int contig
)
1780 struct rb_node
*node
;
1781 struct extent_state
*state
;
1782 u64 cur_start
= *start
;
1783 u64 total_bytes
= 0;
1787 if (WARN_ON(search_end
<= cur_start
))
1790 spin_lock(&tree
->lock
);
1791 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1792 total_bytes
= tree
->dirty_bytes
;
1796 * this search will find all the extents that end after
1799 node
= tree_search(tree
, cur_start
);
1804 state
= rb_entry(node
, struct extent_state
, rb_node
);
1805 if (state
->start
> search_end
)
1807 if (contig
&& found
&& state
->start
> last
+ 1)
1809 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1810 total_bytes
+= min(search_end
, state
->end
) + 1 -
1811 max(cur_start
, state
->start
);
1812 if (total_bytes
>= max_bytes
)
1815 *start
= max(cur_start
, state
->start
);
1819 } else if (contig
&& found
) {
1822 node
= rb_next(node
);
1827 spin_unlock(&tree
->lock
);
1832 * set the private field for a given byte offset in the tree. If there isn't
1833 * an extent_state there already, this does nothing.
1835 static int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1837 struct rb_node
*node
;
1838 struct extent_state
*state
;
1841 spin_lock(&tree
->lock
);
1843 * this search will find all the extents that end after
1846 node
= tree_search(tree
, start
);
1851 state
= rb_entry(node
, struct extent_state
, rb_node
);
1852 if (state
->start
!= start
) {
1856 state
->private = private;
1858 spin_unlock(&tree
->lock
);
1862 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1864 struct rb_node
*node
;
1865 struct extent_state
*state
;
1868 spin_lock(&tree
->lock
);
1870 * this search will find all the extents that end after
1873 node
= tree_search(tree
, start
);
1878 state
= rb_entry(node
, struct extent_state
, rb_node
);
1879 if (state
->start
!= start
) {
1883 *private = state
->private;
1885 spin_unlock(&tree
->lock
);
1890 * searches a range in the state tree for a given mask.
1891 * If 'filled' == 1, this returns 1 only if every extent in the tree
1892 * has the bits set. Otherwise, 1 is returned if any bit in the
1893 * range is found set.
1895 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1896 unsigned long bits
, int filled
, struct extent_state
*cached
)
1898 struct extent_state
*state
= NULL
;
1899 struct rb_node
*node
;
1902 spin_lock(&tree
->lock
);
1903 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1904 cached
->end
> start
)
1905 node
= &cached
->rb_node
;
1907 node
= tree_search(tree
, start
);
1908 while (node
&& start
<= end
) {
1909 state
= rb_entry(node
, struct extent_state
, rb_node
);
1911 if (filled
&& state
->start
> start
) {
1916 if (state
->start
> end
)
1919 if (state
->state
& bits
) {
1923 } else if (filled
) {
1928 if (state
->end
== (u64
)-1)
1931 start
= state
->end
+ 1;
1934 node
= rb_next(node
);
1941 spin_unlock(&tree
->lock
);
1946 * helper function to set a given page up to date if all the
1947 * extents in the tree for that page are up to date
1949 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1951 u64 start
= page_offset(page
);
1952 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1953 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1954 SetPageUptodate(page
);
1958 * When IO fails, either with EIO or csum verification fails, we
1959 * try other mirrors that might have a good copy of the data. This
1960 * io_failure_record is used to record state as we go through all the
1961 * mirrors. If another mirror has good data, the page is set up to date
1962 * and things continue. If a good mirror can't be found, the original
1963 * bio end_io callback is called to indicate things have failed.
1965 struct io_failure_record
{
1970 unsigned long bio_flags
;
1976 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
1981 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1983 set_state_private(failure_tree
, rec
->start
, 0);
1984 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1985 rec
->start
+ rec
->len
- 1,
1986 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1990 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1991 rec
->start
+ rec
->len
- 1,
1992 EXTENT_DAMAGED
, GFP_NOFS
);
2001 * this bypasses the standard btrfs submit functions deliberately, as
2002 * the standard behavior is to write all copies in a raid setup. here we only
2003 * want to write the one bad copy. so we do the mapping for ourselves and issue
2004 * submit_bio directly.
2005 * to avoid any synchronization issues, wait for the data after writing, which
2006 * actually prevents the read that triggered the error from finishing.
2007 * currently, there can be no more than two copies of every data bit. thus,
2008 * exactly one rewrite is required.
2010 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 start
,
2011 u64 length
, u64 logical
, struct page
*page
,
2015 struct btrfs_device
*dev
;
2018 struct btrfs_bio
*bbio
= NULL
;
2019 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2022 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2023 BUG_ON(!mirror_num
);
2025 /* we can't repair anything in raid56 yet */
2026 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2029 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2033 map_length
= length
;
2035 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2036 &map_length
, &bbio
, mirror_num
);
2041 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2042 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2043 bio
->bi_sector
= sector
;
2044 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2046 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2050 bio
->bi_bdev
= dev
->bdev
;
2051 bio_add_page(bio
, page
, length
, start
- page_offset(page
));
2053 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
)) {
2054 /* try to remap that extent elsewhere? */
2056 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2060 printk_ratelimited_in_rcu(KERN_INFO
2061 "BTRFS: read error corrected: ino %lu off %llu "
2062 "(dev %s sector %llu)\n", page
->mapping
->host
->i_ino
,
2063 start
, rcu_str_deref(dev
->name
), sector
);
2069 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2072 u64 start
= eb
->start
;
2073 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2076 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2079 for (i
= 0; i
< num_pages
; i
++) {
2080 struct page
*p
= extent_buffer_page(eb
, i
);
2081 ret
= repair_io_failure(root
->fs_info
, start
, PAGE_CACHE_SIZE
,
2082 start
, p
, mirror_num
);
2085 start
+= PAGE_CACHE_SIZE
;
2092 * each time an IO finishes, we do a fast check in the IO failure tree
2093 * to see if we need to process or clean up an io_failure_record
2095 static int clean_io_failure(u64 start
, struct page
*page
)
2098 u64 private_failure
;
2099 struct io_failure_record
*failrec
;
2100 struct inode
*inode
= page
->mapping
->host
;
2101 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2102 struct extent_state
*state
;
2108 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2109 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2113 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2118 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2119 BUG_ON(!failrec
->this_mirror
);
2121 if (failrec
->in_validation
) {
2122 /* there was no real error, just free the record */
2123 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2128 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2131 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2132 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2135 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2137 if (state
&& state
->start
<= failrec
->start
&&
2138 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2139 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2141 if (num_copies
> 1) {
2142 ret
= repair_io_failure(fs_info
, start
, failrec
->len
,
2143 failrec
->logical
, page
,
2144 failrec
->failed_mirror
);
2152 ret
= free_io_failure(inode
, failrec
, did_repair
);
2158 * this is a generic handler for readpage errors (default
2159 * readpage_io_failed_hook). if other copies exist, read those and write back
2160 * good data to the failed position. does not investigate in remapping the
2161 * failed extent elsewhere, hoping the device will be smart enough to do this as
2165 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2166 struct page
*page
, u64 start
, u64 end
,
2169 struct io_failure_record
*failrec
= NULL
;
2171 struct extent_map
*em
;
2172 struct inode
*inode
= page
->mapping
->host
;
2173 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2174 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2175 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2177 struct btrfs_io_bio
*btrfs_failed_bio
;
2178 struct btrfs_io_bio
*btrfs_bio
;
2184 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2186 ret
= get_state_private(failure_tree
, start
, &private);
2188 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2191 failrec
->start
= start
;
2192 failrec
->len
= end
- start
+ 1;
2193 failrec
->this_mirror
= 0;
2194 failrec
->bio_flags
= 0;
2195 failrec
->in_validation
= 0;
2197 read_lock(&em_tree
->lock
);
2198 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2200 read_unlock(&em_tree
->lock
);
2205 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2206 free_extent_map(em
);
2209 read_unlock(&em_tree
->lock
);
2215 logical
= start
- em
->start
;
2216 logical
= em
->block_start
+ logical
;
2217 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2218 logical
= em
->block_start
;
2219 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2220 extent_set_compress_type(&failrec
->bio_flags
,
2223 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2224 "len=%llu\n", logical
, start
, failrec
->len
);
2225 failrec
->logical
= logical
;
2226 free_extent_map(em
);
2228 /* set the bits in the private failure tree */
2229 ret
= set_extent_bits(failure_tree
, start
, end
,
2230 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2232 ret
= set_state_private(failure_tree
, start
,
2233 (u64
)(unsigned long)failrec
);
2234 /* set the bits in the inode's tree */
2236 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2243 failrec
= (struct io_failure_record
*)(unsigned long)private;
2244 pr_debug("bio_readpage_error: (found) logical=%llu, "
2245 "start=%llu, len=%llu, validation=%d\n",
2246 failrec
->logical
, failrec
->start
, failrec
->len
,
2247 failrec
->in_validation
);
2249 * when data can be on disk more than twice, add to failrec here
2250 * (e.g. with a list for failed_mirror) to make
2251 * clean_io_failure() clean all those errors at once.
2254 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2255 failrec
->logical
, failrec
->len
);
2256 if (num_copies
== 1) {
2258 * we only have a single copy of the data, so don't bother with
2259 * all the retry and error correction code that follows. no
2260 * matter what the error is, it is very likely to persist.
2262 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2263 num_copies
, failrec
->this_mirror
, failed_mirror
);
2264 free_io_failure(inode
, failrec
, 0);
2269 * there are two premises:
2270 * a) deliver good data to the caller
2271 * b) correct the bad sectors on disk
2273 if (failed_bio
->bi_vcnt
> 1) {
2275 * to fulfill b), we need to know the exact failing sectors, as
2276 * we don't want to rewrite any more than the failed ones. thus,
2277 * we need separate read requests for the failed bio
2279 * if the following BUG_ON triggers, our validation request got
2280 * merged. we need separate requests for our algorithm to work.
2282 BUG_ON(failrec
->in_validation
);
2283 failrec
->in_validation
= 1;
2284 failrec
->this_mirror
= failed_mirror
;
2285 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2288 * we're ready to fulfill a) and b) alongside. get a good copy
2289 * of the failed sector and if we succeed, we have setup
2290 * everything for repair_io_failure to do the rest for us.
2292 if (failrec
->in_validation
) {
2293 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2294 failrec
->in_validation
= 0;
2295 failrec
->this_mirror
= 0;
2297 failrec
->failed_mirror
= failed_mirror
;
2298 failrec
->this_mirror
++;
2299 if (failrec
->this_mirror
== failed_mirror
)
2300 failrec
->this_mirror
++;
2301 read_mode
= READ_SYNC
;
2304 if (failrec
->this_mirror
> num_copies
) {
2305 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2306 num_copies
, failrec
->this_mirror
, failed_mirror
);
2307 free_io_failure(inode
, failrec
, 0);
2311 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2313 free_io_failure(inode
, failrec
, 0);
2316 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2317 bio
->bi_sector
= failrec
->logical
>> 9;
2318 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2321 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2322 if (btrfs_failed_bio
->csum
) {
2323 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2324 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2326 btrfs_bio
= btrfs_io_bio(bio
);
2327 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2328 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2329 phy_offset
*= csum_size
;
2330 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ phy_offset
,
2334 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2336 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2337 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2338 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2340 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2341 failrec
->this_mirror
,
2342 failrec
->bio_flags
, 0);
2346 /* lots and lots of room for performance fixes in the end_bio funcs */
2348 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2350 int uptodate
= (err
== 0);
2351 struct extent_io_tree
*tree
;
2354 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2356 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2357 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2358 end
, NULL
, uptodate
);
2364 ClearPageUptodate(page
);
2371 * after a writepage IO is done, we need to:
2372 * clear the uptodate bits on error
2373 * clear the writeback bits in the extent tree for this IO
2374 * end_page_writeback if the page has no more pending IO
2376 * Scheduling is not allowed, so the extent state tree is expected
2377 * to have one and only one object corresponding to this IO.
2379 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2381 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2386 struct page
*page
= bvec
->bv_page
;
2388 /* We always issue full-page reads, but if some block
2389 * in a page fails to read, blk_update_request() will
2390 * advance bv_offset and adjust bv_len to compensate.
2391 * Print a warning for nonzero offsets, and an error
2392 * if they don't add up to a full page. */
2393 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2394 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2395 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2396 "partial page write in btrfs with offset %u and length %u",
2397 bvec
->bv_offset
, bvec
->bv_len
);
2399 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2400 "incomplete page write in btrfs with offset %u and "
2402 bvec
->bv_offset
, bvec
->bv_len
);
2405 start
= page_offset(page
);
2406 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2408 if (--bvec
>= bio
->bi_io_vec
)
2409 prefetchw(&bvec
->bv_page
->flags
);
2411 if (end_extent_writepage(page
, err
, start
, end
))
2414 end_page_writeback(page
);
2415 } while (bvec
>= bio
->bi_io_vec
);
2421 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2424 struct extent_state
*cached
= NULL
;
2425 u64 end
= start
+ len
- 1;
2427 if (uptodate
&& tree
->track_uptodate
)
2428 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2429 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2433 * after a readpage IO is done, we need to:
2434 * clear the uptodate bits on error
2435 * set the uptodate bits if things worked
2436 * set the page up to date if all extents in the tree are uptodate
2437 * clear the lock bit in the extent tree
2438 * unlock the page if there are no other extents locked for it
2440 * Scheduling is not allowed, so the extent state tree is expected
2441 * to have one and only one object corresponding to this IO.
2443 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2445 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2446 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2447 struct bio_vec
*bvec
= bio
->bi_io_vec
;
2448 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2449 struct extent_io_tree
*tree
;
2454 u64 extent_start
= 0;
2463 struct page
*page
= bvec
->bv_page
;
2464 struct inode
*inode
= page
->mapping
->host
;
2466 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2467 "mirror=%lu\n", (u64
)bio
->bi_sector
, err
,
2468 io_bio
->mirror_num
);
2469 tree
= &BTRFS_I(inode
)->io_tree
;
2471 /* We always issue full-page reads, but if some block
2472 * in a page fails to read, blk_update_request() will
2473 * advance bv_offset and adjust bv_len to compensate.
2474 * Print a warning for nonzero offsets, and an error
2475 * if they don't add up to a full page. */
2476 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2477 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2478 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2479 "partial page read in btrfs with offset %u and length %u",
2480 bvec
->bv_offset
, bvec
->bv_len
);
2482 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2483 "incomplete page read in btrfs with offset %u and "
2485 bvec
->bv_offset
, bvec
->bv_len
);
2488 start
= page_offset(page
);
2489 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2492 if (++bvec
<= bvec_end
)
2493 prefetchw(&bvec
->bv_page
->flags
);
2495 mirror
= io_bio
->mirror_num
;
2496 if (likely(uptodate
&& tree
->ops
&&
2497 tree
->ops
->readpage_end_io_hook
)) {
2498 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2504 clean_io_failure(start
, page
);
2507 if (likely(uptodate
))
2510 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2511 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2513 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2517 * The generic bio_readpage_error handles errors the
2518 * following way: If possible, new read requests are
2519 * created and submitted and will end up in
2520 * end_bio_extent_readpage as well (if we're lucky, not
2521 * in the !uptodate case). In that case it returns 0 and
2522 * we just go on with the next page in our bio. If it
2523 * can't handle the error it will return -EIO and we
2524 * remain responsible for that page.
2526 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2530 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2537 if (likely(uptodate
)) {
2538 loff_t i_size
= i_size_read(inode
);
2539 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2542 /* Zero out the end if this page straddles i_size */
2543 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2544 if (page
->index
== end_index
&& offset
)
2545 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2546 SetPageUptodate(page
);
2548 ClearPageUptodate(page
);
2554 if (unlikely(!uptodate
)) {
2556 endio_readpage_release_extent(tree
,
2562 endio_readpage_release_extent(tree
, start
,
2563 end
- start
+ 1, 0);
2564 } else if (!extent_len
) {
2565 extent_start
= start
;
2566 extent_len
= end
+ 1 - start
;
2567 } else if (extent_start
+ extent_len
== start
) {
2568 extent_len
+= end
+ 1 - start
;
2570 endio_readpage_release_extent(tree
, extent_start
,
2571 extent_len
, uptodate
);
2572 extent_start
= start
;
2573 extent_len
= end
+ 1 - start
;
2575 } while (bvec
<= bvec_end
);
2578 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2581 io_bio
->end_io(io_bio
, err
);
2586 * this allocates from the btrfs_bioset. We're returning a bio right now
2587 * but you can call btrfs_io_bio for the appropriate container_of magic
2590 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2593 struct btrfs_io_bio
*btrfs_bio
;
2596 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2598 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2599 while (!bio
&& (nr_vecs
/= 2)) {
2600 bio
= bio_alloc_bioset(gfp_flags
,
2601 nr_vecs
, btrfs_bioset
);
2607 bio
->bi_bdev
= bdev
;
2608 bio
->bi_sector
= first_sector
;
2609 btrfs_bio
= btrfs_io_bio(bio
);
2610 btrfs_bio
->csum
= NULL
;
2611 btrfs_bio
->csum_allocated
= NULL
;
2612 btrfs_bio
->end_io
= NULL
;
2617 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2619 return bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2623 /* this also allocates from the btrfs_bioset */
2624 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2626 struct btrfs_io_bio
*btrfs_bio
;
2629 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2631 btrfs_bio
= btrfs_io_bio(bio
);
2632 btrfs_bio
->csum
= NULL
;
2633 btrfs_bio
->csum_allocated
= NULL
;
2634 btrfs_bio
->end_io
= NULL
;
2640 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2641 int mirror_num
, unsigned long bio_flags
)
2644 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2645 struct page
*page
= bvec
->bv_page
;
2646 struct extent_io_tree
*tree
= bio
->bi_private
;
2649 start
= page_offset(page
) + bvec
->bv_offset
;
2651 bio
->bi_private
= NULL
;
2655 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2656 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2657 mirror_num
, bio_flags
, start
);
2659 btrfsic_submit_bio(rw
, bio
);
2661 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2667 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2668 unsigned long offset
, size_t size
, struct bio
*bio
,
2669 unsigned long bio_flags
)
2672 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2673 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2680 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2681 struct page
*page
, sector_t sector
,
2682 size_t size
, unsigned long offset
,
2683 struct block_device
*bdev
,
2684 struct bio
**bio_ret
,
2685 unsigned long max_pages
,
2686 bio_end_io_t end_io_func
,
2688 unsigned long prev_bio_flags
,
2689 unsigned long bio_flags
)
2695 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2696 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2697 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2699 if (bio_ret
&& *bio_ret
) {
2702 contig
= bio
->bi_sector
== sector
;
2704 contig
= bio_end_sector(bio
) == sector
;
2706 if (prev_bio_flags
!= bio_flags
|| !contig
||
2707 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2708 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2709 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2718 if (this_compressed
)
2721 nr
= bio_get_nr_vecs(bdev
);
2723 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2727 bio_add_page(bio
, page
, page_size
, offset
);
2728 bio
->bi_end_io
= end_io_func
;
2729 bio
->bi_private
= tree
;
2734 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2739 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2742 if (!PagePrivate(page
)) {
2743 SetPagePrivate(page
);
2744 page_cache_get(page
);
2745 set_page_private(page
, (unsigned long)eb
);
2747 WARN_ON(page
->private != (unsigned long)eb
);
2751 void set_page_extent_mapped(struct page
*page
)
2753 if (!PagePrivate(page
)) {
2754 SetPagePrivate(page
);
2755 page_cache_get(page
);
2756 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2760 static struct extent_map
*
2761 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2762 u64 start
, u64 len
, get_extent_t
*get_extent
,
2763 struct extent_map
**em_cached
)
2765 struct extent_map
*em
;
2767 if (em_cached
&& *em_cached
) {
2769 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2770 start
< extent_map_end(em
)) {
2771 atomic_inc(&em
->refs
);
2775 free_extent_map(em
);
2779 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2780 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2782 atomic_inc(&em
->refs
);
2788 * basic readpage implementation. Locked extent state structs are inserted
2789 * into the tree that are removed when the IO is done (by the end_io
2791 * XXX JDM: This needs looking at to ensure proper page locking
2793 static int __do_readpage(struct extent_io_tree
*tree
,
2795 get_extent_t
*get_extent
,
2796 struct extent_map
**em_cached
,
2797 struct bio
**bio
, int mirror_num
,
2798 unsigned long *bio_flags
, int rw
)
2800 struct inode
*inode
= page
->mapping
->host
;
2801 u64 start
= page_offset(page
);
2802 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2806 u64 last_byte
= i_size_read(inode
);
2810 struct extent_map
*em
;
2811 struct block_device
*bdev
;
2814 int parent_locked
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2815 size_t pg_offset
= 0;
2817 size_t disk_io_size
;
2818 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2819 unsigned long this_bio_flag
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2821 set_page_extent_mapped(page
);
2824 if (!PageUptodate(page
)) {
2825 if (cleancache_get_page(page
) == 0) {
2826 BUG_ON(blocksize
!= PAGE_SIZE
);
2827 unlock_extent(tree
, start
, end
);
2832 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2834 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2837 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2838 userpage
= kmap_atomic(page
);
2839 memset(userpage
+ zero_offset
, 0, iosize
);
2840 flush_dcache_page(page
);
2841 kunmap_atomic(userpage
);
2844 while (cur
<= end
) {
2845 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2847 if (cur
>= last_byte
) {
2849 struct extent_state
*cached
= NULL
;
2851 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2852 userpage
= kmap_atomic(page
);
2853 memset(userpage
+ pg_offset
, 0, iosize
);
2854 flush_dcache_page(page
);
2855 kunmap_atomic(userpage
);
2856 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2859 unlock_extent_cached(tree
, cur
,
2864 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2865 end
- cur
+ 1, get_extent
, em_cached
);
2866 if (IS_ERR_OR_NULL(em
)) {
2869 unlock_extent(tree
, cur
, end
);
2872 extent_offset
= cur
- em
->start
;
2873 BUG_ON(extent_map_end(em
) <= cur
);
2876 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2877 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2878 extent_set_compress_type(&this_bio_flag
,
2882 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2883 cur_end
= min(extent_map_end(em
) - 1, end
);
2884 iosize
= ALIGN(iosize
, blocksize
);
2885 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2886 disk_io_size
= em
->block_len
;
2887 sector
= em
->block_start
>> 9;
2889 sector
= (em
->block_start
+ extent_offset
) >> 9;
2890 disk_io_size
= iosize
;
2893 block_start
= em
->block_start
;
2894 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2895 block_start
= EXTENT_MAP_HOLE
;
2896 free_extent_map(em
);
2899 /* we've found a hole, just zero and go on */
2900 if (block_start
== EXTENT_MAP_HOLE
) {
2902 struct extent_state
*cached
= NULL
;
2904 userpage
= kmap_atomic(page
);
2905 memset(userpage
+ pg_offset
, 0, iosize
);
2906 flush_dcache_page(page
);
2907 kunmap_atomic(userpage
);
2909 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2911 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2914 pg_offset
+= iosize
;
2917 /* the get_extent function already copied into the page */
2918 if (test_range_bit(tree
, cur
, cur_end
,
2919 EXTENT_UPTODATE
, 1, NULL
)) {
2920 check_page_uptodate(tree
, page
);
2922 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2924 pg_offset
+= iosize
;
2927 /* we have an inline extent but it didn't get marked up
2928 * to date. Error out
2930 if (block_start
== EXTENT_MAP_INLINE
) {
2933 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2935 pg_offset
+= iosize
;
2940 ret
= submit_extent_page(rw
, tree
, page
,
2941 sector
, disk_io_size
, pg_offset
,
2943 end_bio_extent_readpage
, mirror_num
,
2948 *bio_flags
= this_bio_flag
;
2952 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2955 pg_offset
+= iosize
;
2959 if (!PageError(page
))
2960 SetPageUptodate(page
);
2966 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
2967 struct page
*pages
[], int nr_pages
,
2969 get_extent_t
*get_extent
,
2970 struct extent_map
**em_cached
,
2971 struct bio
**bio
, int mirror_num
,
2972 unsigned long *bio_flags
, int rw
)
2974 struct inode
*inode
;
2975 struct btrfs_ordered_extent
*ordered
;
2978 inode
= pages
[0]->mapping
->host
;
2980 lock_extent(tree
, start
, end
);
2981 ordered
= btrfs_lookup_ordered_range(inode
, start
,
2985 unlock_extent(tree
, start
, end
);
2986 btrfs_start_ordered_extent(inode
, ordered
, 1);
2987 btrfs_put_ordered_extent(ordered
);
2990 for (index
= 0; index
< nr_pages
; index
++) {
2991 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
2992 mirror_num
, bio_flags
, rw
);
2993 page_cache_release(pages
[index
]);
2997 static void __extent_readpages(struct extent_io_tree
*tree
,
2998 struct page
*pages
[],
2999 int nr_pages
, get_extent_t
*get_extent
,
3000 struct extent_map
**em_cached
,
3001 struct bio
**bio
, int mirror_num
,
3002 unsigned long *bio_flags
, int rw
)
3008 int first_index
= 0;
3010 for (index
= 0; index
< nr_pages
; index
++) {
3011 page_start
= page_offset(pages
[index
]);
3014 end
= start
+ PAGE_CACHE_SIZE
- 1;
3015 first_index
= index
;
3016 } else if (end
+ 1 == page_start
) {
3017 end
+= PAGE_CACHE_SIZE
;
3019 __do_contiguous_readpages(tree
, &pages
[first_index
],
3020 index
- first_index
, start
,
3021 end
, get_extent
, em_cached
,
3022 bio
, mirror_num
, bio_flags
,
3025 end
= start
+ PAGE_CACHE_SIZE
- 1;
3026 first_index
= index
;
3031 __do_contiguous_readpages(tree
, &pages
[first_index
],
3032 index
- first_index
, start
,
3033 end
, get_extent
, em_cached
, bio
,
3034 mirror_num
, bio_flags
, rw
);
3037 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3039 get_extent_t
*get_extent
,
3040 struct bio
**bio
, int mirror_num
,
3041 unsigned long *bio_flags
, int rw
)
3043 struct inode
*inode
= page
->mapping
->host
;
3044 struct btrfs_ordered_extent
*ordered
;
3045 u64 start
= page_offset(page
);
3046 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3050 lock_extent(tree
, start
, end
);
3051 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
3054 unlock_extent(tree
, start
, end
);
3055 btrfs_start_ordered_extent(inode
, ordered
, 1);
3056 btrfs_put_ordered_extent(ordered
);
3059 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3064 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3065 get_extent_t
*get_extent
, int mirror_num
)
3067 struct bio
*bio
= NULL
;
3068 unsigned long bio_flags
= 0;
3071 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3074 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3078 int extent_read_full_page_nolock(struct extent_io_tree
*tree
, struct page
*page
,
3079 get_extent_t
*get_extent
, int mirror_num
)
3081 struct bio
*bio
= NULL
;
3082 unsigned long bio_flags
= EXTENT_BIO_PARENT_LOCKED
;
3085 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, &bio
, mirror_num
,
3088 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3092 static noinline
void update_nr_written(struct page
*page
,
3093 struct writeback_control
*wbc
,
3094 unsigned long nr_written
)
3096 wbc
->nr_to_write
-= nr_written
;
3097 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
3098 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
3099 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
3103 * the writepage semantics are similar to regular writepage. extent
3104 * records are inserted to lock ranges in the tree, and as dirty areas
3105 * are found, they are marked writeback. Then the lock bits are removed
3106 * and the end_io handler clears the writeback ranges
3108 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3111 struct inode
*inode
= page
->mapping
->host
;
3112 struct extent_page_data
*epd
= data
;
3113 struct extent_io_tree
*tree
= epd
->tree
;
3114 u64 start
= page_offset(page
);
3116 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3120 u64 last_byte
= i_size_read(inode
);
3124 struct extent_state
*cached_state
= NULL
;
3125 struct extent_map
*em
;
3126 struct block_device
*bdev
;
3129 size_t pg_offset
= 0;
3131 loff_t i_size
= i_size_read(inode
);
3132 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
3138 unsigned long nr_written
= 0;
3139 bool fill_delalloc
= true;
3141 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3142 write_flags
= WRITE_SYNC
;
3144 write_flags
= WRITE
;
3146 trace___extent_writepage(page
, inode
, wbc
);
3148 WARN_ON(!PageLocked(page
));
3150 ClearPageError(page
);
3152 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
3153 if (page
->index
> end_index
||
3154 (page
->index
== end_index
&& !pg_offset
)) {
3155 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
3160 if (page
->index
== end_index
) {
3163 userpage
= kmap_atomic(page
);
3164 memset(userpage
+ pg_offset
, 0,
3165 PAGE_CACHE_SIZE
- pg_offset
);
3166 kunmap_atomic(userpage
);
3167 flush_dcache_page(page
);
3171 set_page_extent_mapped(page
);
3173 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
3174 fill_delalloc
= false;
3176 delalloc_start
= start
;
3179 if (!epd
->extent_locked
&& fill_delalloc
) {
3180 u64 delalloc_to_write
= 0;
3182 * make sure the wbc mapping index is at least updated
3185 update_nr_written(page
, wbc
, 0);
3187 while (delalloc_end
< page_end
) {
3188 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3193 if (nr_delalloc
== 0) {
3194 delalloc_start
= delalloc_end
+ 1;
3197 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3202 /* File system has been set read-only */
3208 * delalloc_end is already one less than the total
3209 * length, so we don't subtract one from
3212 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3215 delalloc_start
= delalloc_end
+ 1;
3217 if (wbc
->nr_to_write
< delalloc_to_write
) {
3220 if (delalloc_to_write
< thresh
* 2)
3221 thresh
= delalloc_to_write
;
3222 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3226 /* did the fill delalloc function already unlock and start
3232 * we've unlocked the page, so we can't update
3233 * the mapping's writeback index, just update
3236 wbc
->nr_to_write
-= nr_written
;
3240 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3241 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3244 /* Fixup worker will requeue */
3246 wbc
->pages_skipped
++;
3248 redirty_page_for_writepage(wbc
, page
);
3249 update_nr_written(page
, wbc
, nr_written
);
3257 * we don't want to touch the inode after unlocking the page,
3258 * so we update the mapping writeback index now
3260 update_nr_written(page
, wbc
, nr_written
+ 1);
3263 if (last_byte
<= start
) {
3264 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3265 tree
->ops
->writepage_end_io_hook(page
, start
,
3270 blocksize
= inode
->i_sb
->s_blocksize
;
3272 while (cur
<= end
) {
3273 if (cur
>= last_byte
) {
3274 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3275 tree
->ops
->writepage_end_io_hook(page
, cur
,
3279 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3281 if (IS_ERR_OR_NULL(em
)) {
3286 extent_offset
= cur
- em
->start
;
3287 BUG_ON(extent_map_end(em
) <= cur
);
3289 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
3290 iosize
= ALIGN(iosize
, blocksize
);
3291 sector
= (em
->block_start
+ extent_offset
) >> 9;
3293 block_start
= em
->block_start
;
3294 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3295 free_extent_map(em
);
3299 * compressed and inline extents are written through other
3302 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3303 block_start
== EXTENT_MAP_INLINE
) {
3305 * end_io notification does not happen here for
3306 * compressed extents
3308 if (!compressed
&& tree
->ops
&&
3309 tree
->ops
->writepage_end_io_hook
)
3310 tree
->ops
->writepage_end_io_hook(page
, cur
,
3313 else if (compressed
) {
3314 /* we don't want to end_page_writeback on
3315 * a compressed extent. this happens
3322 pg_offset
+= iosize
;
3325 /* leave this out until we have a page_mkwrite call */
3326 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
3327 EXTENT_DIRTY
, 0, NULL
)) {
3329 pg_offset
+= iosize
;
3333 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3334 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3342 unsigned long max_nr
= end_index
+ 1;
3344 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3345 if (!PageWriteback(page
)) {
3346 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3347 "page %lu not writeback, cur %llu end %llu",
3348 page
->index
, cur
, end
);
3351 ret
= submit_extent_page(write_flags
, tree
, page
,
3352 sector
, iosize
, pg_offset
,
3353 bdev
, &epd
->bio
, max_nr
,
3354 end_bio_extent_writepage
,
3360 pg_offset
+= iosize
;
3365 /* make sure the mapping tag for page dirty gets cleared */
3366 set_page_writeback(page
);
3367 end_page_writeback(page
);
3373 /* drop our reference on any cached states */
3374 free_extent_state(cached_state
);
3378 static int eb_wait(void *word
)
3384 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3386 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3387 TASK_UNINTERRUPTIBLE
);
3390 static int lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3391 struct btrfs_fs_info
*fs_info
,
3392 struct extent_page_data
*epd
)
3394 unsigned long i
, num_pages
;
3398 if (!btrfs_try_tree_write_lock(eb
)) {
3400 flush_write_bio(epd
);
3401 btrfs_tree_lock(eb
);
3404 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3405 btrfs_tree_unlock(eb
);
3409 flush_write_bio(epd
);
3413 wait_on_extent_buffer_writeback(eb
);
3414 btrfs_tree_lock(eb
);
3415 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3417 btrfs_tree_unlock(eb
);
3422 * We need to do this to prevent races in people who check if the eb is
3423 * under IO since we can end up having no IO bits set for a short period
3426 spin_lock(&eb
->refs_lock
);
3427 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3428 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3429 spin_unlock(&eb
->refs_lock
);
3430 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3431 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3433 fs_info
->dirty_metadata_batch
);
3436 spin_unlock(&eb
->refs_lock
);
3439 btrfs_tree_unlock(eb
);
3444 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3445 for (i
= 0; i
< num_pages
; i
++) {
3446 struct page
*p
= extent_buffer_page(eb
, i
);
3448 if (!trylock_page(p
)) {
3450 flush_write_bio(epd
);
3460 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3462 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3463 smp_mb__after_clear_bit();
3464 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3467 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3469 int uptodate
= err
== 0;
3470 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
3471 struct extent_buffer
*eb
;
3475 struct page
*page
= bvec
->bv_page
;
3478 eb
= (struct extent_buffer
*)page
->private;
3480 done
= atomic_dec_and_test(&eb
->io_pages
);
3482 if (!uptodate
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3483 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3484 ClearPageUptodate(page
);
3488 end_page_writeback(page
);
3493 end_extent_buffer_writeback(eb
);
3494 } while (bvec
>= bio
->bi_io_vec
);
3500 static int write_one_eb(struct extent_buffer
*eb
,
3501 struct btrfs_fs_info
*fs_info
,
3502 struct writeback_control
*wbc
,
3503 struct extent_page_data
*epd
)
3505 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3506 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3507 u64 offset
= eb
->start
;
3508 unsigned long i
, num_pages
;
3509 unsigned long bio_flags
= 0;
3510 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3513 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3514 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3515 atomic_set(&eb
->io_pages
, num_pages
);
3516 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3517 bio_flags
= EXTENT_BIO_TREE_LOG
;
3519 for (i
= 0; i
< num_pages
; i
++) {
3520 struct page
*p
= extent_buffer_page(eb
, i
);
3522 clear_page_dirty_for_io(p
);
3523 set_page_writeback(p
);
3524 ret
= submit_extent_page(rw
, tree
, p
, offset
>> 9,
3525 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3526 -1, end_bio_extent_buffer_writepage
,
3527 0, epd
->bio_flags
, bio_flags
);
3528 epd
->bio_flags
= bio_flags
;
3530 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3532 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3533 end_extent_buffer_writeback(eb
);
3537 offset
+= PAGE_CACHE_SIZE
;
3538 update_nr_written(p
, wbc
, 1);
3542 if (unlikely(ret
)) {
3543 for (; i
< num_pages
; i
++) {
3544 struct page
*p
= extent_buffer_page(eb
, i
);
3552 int btree_write_cache_pages(struct address_space
*mapping
,
3553 struct writeback_control
*wbc
)
3555 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3556 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3557 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3558 struct extent_page_data epd
= {
3562 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3567 int nr_to_write_done
= 0;
3568 struct pagevec pvec
;
3571 pgoff_t end
; /* Inclusive */
3575 pagevec_init(&pvec
, 0);
3576 if (wbc
->range_cyclic
) {
3577 index
= mapping
->writeback_index
; /* Start from prev offset */
3580 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3581 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3584 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3585 tag
= PAGECACHE_TAG_TOWRITE
;
3587 tag
= PAGECACHE_TAG_DIRTY
;
3589 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3590 tag_pages_for_writeback(mapping
, index
, end
);
3591 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3592 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3593 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3597 for (i
= 0; i
< nr_pages
; i
++) {
3598 struct page
*page
= pvec
.pages
[i
];
3600 if (!PagePrivate(page
))
3603 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3608 spin_lock(&mapping
->private_lock
);
3609 if (!PagePrivate(page
)) {
3610 spin_unlock(&mapping
->private_lock
);
3614 eb
= (struct extent_buffer
*)page
->private;
3617 * Shouldn't happen and normally this would be a BUG_ON
3618 * but no sense in crashing the users box for something
3619 * we can survive anyway.
3622 spin_unlock(&mapping
->private_lock
);
3626 if (eb
== prev_eb
) {
3627 spin_unlock(&mapping
->private_lock
);
3631 ret
= atomic_inc_not_zero(&eb
->refs
);
3632 spin_unlock(&mapping
->private_lock
);
3637 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3639 free_extent_buffer(eb
);
3643 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3646 free_extent_buffer(eb
);
3649 free_extent_buffer(eb
);
3652 * the filesystem may choose to bump up nr_to_write.
3653 * We have to make sure to honor the new nr_to_write
3656 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3658 pagevec_release(&pvec
);
3661 if (!scanned
&& !done
) {
3663 * We hit the last page and there is more work to be done: wrap
3664 * back to the start of the file
3670 flush_write_bio(&epd
);
3675 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3676 * @mapping: address space structure to write
3677 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3678 * @writepage: function called for each page
3679 * @data: data passed to writepage function
3681 * If a page is already under I/O, write_cache_pages() skips it, even
3682 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3683 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3684 * and msync() need to guarantee that all the data which was dirty at the time
3685 * the call was made get new I/O started against them. If wbc->sync_mode is
3686 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3687 * existing IO to complete.
3689 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3690 struct address_space
*mapping
,
3691 struct writeback_control
*wbc
,
3692 writepage_t writepage
, void *data
,
3693 void (*flush_fn
)(void *))
3695 struct inode
*inode
= mapping
->host
;
3698 int nr_to_write_done
= 0;
3699 struct pagevec pvec
;
3702 pgoff_t end
; /* Inclusive */
3707 * We have to hold onto the inode so that ordered extents can do their
3708 * work when the IO finishes. The alternative to this is failing to add
3709 * an ordered extent if the igrab() fails there and that is a huge pain
3710 * to deal with, so instead just hold onto the inode throughout the
3711 * writepages operation. If it fails here we are freeing up the inode
3712 * anyway and we'd rather not waste our time writing out stuff that is
3713 * going to be truncated anyway.
3718 pagevec_init(&pvec
, 0);
3719 if (wbc
->range_cyclic
) {
3720 index
= mapping
->writeback_index
; /* Start from prev offset */
3723 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3724 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3727 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3728 tag
= PAGECACHE_TAG_TOWRITE
;
3730 tag
= PAGECACHE_TAG_DIRTY
;
3732 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3733 tag_pages_for_writeback(mapping
, index
, end
);
3734 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3735 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3736 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3740 for (i
= 0; i
< nr_pages
; i
++) {
3741 struct page
*page
= pvec
.pages
[i
];
3744 * At this point we hold neither mapping->tree_lock nor
3745 * lock on the page itself: the page may be truncated or
3746 * invalidated (changing page->mapping to NULL), or even
3747 * swizzled back from swapper_space to tmpfs file
3750 if (!trylock_page(page
)) {
3755 if (unlikely(page
->mapping
!= mapping
)) {
3760 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3766 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3767 if (PageWriteback(page
))
3769 wait_on_page_writeback(page
);
3772 if (PageWriteback(page
) ||
3773 !clear_page_dirty_for_io(page
)) {
3778 ret
= (*writepage
)(page
, wbc
, data
);
3780 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3788 * the filesystem may choose to bump up nr_to_write.
3789 * We have to make sure to honor the new nr_to_write
3792 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3794 pagevec_release(&pvec
);
3797 if (!scanned
&& !done
) {
3799 * We hit the last page and there is more work to be done: wrap
3800 * back to the start of the file
3806 btrfs_add_delayed_iput(inode
);
3810 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3819 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
3820 BUG_ON(ret
< 0); /* -ENOMEM */
3825 static noinline
void flush_write_bio(void *data
)
3827 struct extent_page_data
*epd
= data
;
3828 flush_epd_write_bio(epd
);
3831 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3832 get_extent_t
*get_extent
,
3833 struct writeback_control
*wbc
)
3836 struct extent_page_data epd
= {
3839 .get_extent
= get_extent
,
3841 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3845 ret
= __extent_writepage(page
, wbc
, &epd
);
3847 flush_epd_write_bio(&epd
);
3851 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3852 u64 start
, u64 end
, get_extent_t
*get_extent
,
3856 struct address_space
*mapping
= inode
->i_mapping
;
3858 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3861 struct extent_page_data epd
= {
3864 .get_extent
= get_extent
,
3866 .sync_io
= mode
== WB_SYNC_ALL
,
3869 struct writeback_control wbc_writepages
= {
3871 .nr_to_write
= nr_pages
* 2,
3872 .range_start
= start
,
3873 .range_end
= end
+ 1,
3876 while (start
<= end
) {
3877 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3878 if (clear_page_dirty_for_io(page
))
3879 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3881 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3882 tree
->ops
->writepage_end_io_hook(page
, start
,
3883 start
+ PAGE_CACHE_SIZE
- 1,
3887 page_cache_release(page
);
3888 start
+= PAGE_CACHE_SIZE
;
3891 flush_epd_write_bio(&epd
);
3895 int extent_writepages(struct extent_io_tree
*tree
,
3896 struct address_space
*mapping
,
3897 get_extent_t
*get_extent
,
3898 struct writeback_control
*wbc
)
3901 struct extent_page_data epd
= {
3904 .get_extent
= get_extent
,
3906 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3910 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3911 __extent_writepage
, &epd
,
3913 flush_epd_write_bio(&epd
);
3917 int extent_readpages(struct extent_io_tree
*tree
,
3918 struct address_space
*mapping
,
3919 struct list_head
*pages
, unsigned nr_pages
,
3920 get_extent_t get_extent
)
3922 struct bio
*bio
= NULL
;
3924 unsigned long bio_flags
= 0;
3925 struct page
*pagepool
[16];
3927 struct extent_map
*em_cached
= NULL
;
3930 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3931 page
= list_entry(pages
->prev
, struct page
, lru
);
3933 prefetchw(&page
->flags
);
3934 list_del(&page
->lru
);
3935 if (add_to_page_cache_lru(page
, mapping
,
3936 page
->index
, GFP_NOFS
)) {
3937 page_cache_release(page
);
3941 pagepool
[nr
++] = page
;
3942 if (nr
< ARRAY_SIZE(pagepool
))
3944 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3945 &bio
, 0, &bio_flags
, READ
);
3949 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
3950 &bio
, 0, &bio_flags
, READ
);
3953 free_extent_map(em_cached
);
3955 BUG_ON(!list_empty(pages
));
3957 return submit_one_bio(READ
, bio
, 0, bio_flags
);
3962 * basic invalidatepage code, this waits on any locked or writeback
3963 * ranges corresponding to the page, and then deletes any extent state
3964 * records from the tree
3966 int extent_invalidatepage(struct extent_io_tree
*tree
,
3967 struct page
*page
, unsigned long offset
)
3969 struct extent_state
*cached_state
= NULL
;
3970 u64 start
= page_offset(page
);
3971 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3972 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
3974 start
+= ALIGN(offset
, blocksize
);
3978 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
3979 wait_on_page_writeback(page
);
3980 clear_extent_bit(tree
, start
, end
,
3981 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3982 EXTENT_DO_ACCOUNTING
,
3983 1, 1, &cached_state
, GFP_NOFS
);
3988 * a helper for releasepage, this tests for areas of the page that
3989 * are locked or under IO and drops the related state bits if it is safe
3992 static int try_release_extent_state(struct extent_map_tree
*map
,
3993 struct extent_io_tree
*tree
,
3994 struct page
*page
, gfp_t mask
)
3996 u64 start
= page_offset(page
);
3997 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4000 if (test_range_bit(tree
, start
, end
,
4001 EXTENT_IOBITS
, 0, NULL
))
4004 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4007 * at this point we can safely clear everything except the
4008 * locked bit and the nodatasum bit
4010 ret
= clear_extent_bit(tree
, start
, end
,
4011 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4014 /* if clear_extent_bit failed for enomem reasons,
4015 * we can't allow the release to continue.
4026 * a helper for releasepage. As long as there are no locked extents
4027 * in the range corresponding to the page, both state records and extent
4028 * map records are removed
4030 int try_release_extent_mapping(struct extent_map_tree
*map
,
4031 struct extent_io_tree
*tree
, struct page
*page
,
4034 struct extent_map
*em
;
4035 u64 start
= page_offset(page
);
4036 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4038 if ((mask
& __GFP_WAIT
) &&
4039 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
4041 while (start
<= end
) {
4042 len
= end
- start
+ 1;
4043 write_lock(&map
->lock
);
4044 em
= lookup_extent_mapping(map
, start
, len
);
4046 write_unlock(&map
->lock
);
4049 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4050 em
->start
!= start
) {
4051 write_unlock(&map
->lock
);
4052 free_extent_map(em
);
4055 if (!test_range_bit(tree
, em
->start
,
4056 extent_map_end(em
) - 1,
4057 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4059 remove_extent_mapping(map
, em
);
4060 /* once for the rb tree */
4061 free_extent_map(em
);
4063 start
= extent_map_end(em
);
4064 write_unlock(&map
->lock
);
4067 free_extent_map(em
);
4070 return try_release_extent_state(map
, tree
, page
, mask
);
4074 * helper function for fiemap, which doesn't want to see any holes.
4075 * This maps until we find something past 'last'
4077 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4080 get_extent_t
*get_extent
)
4082 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4083 struct extent_map
*em
;
4090 len
= last
- offset
;
4093 len
= ALIGN(len
, sectorsize
);
4094 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4095 if (IS_ERR_OR_NULL(em
))
4098 /* if this isn't a hole return it */
4099 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4100 em
->block_start
!= EXTENT_MAP_HOLE
) {
4104 /* this is a hole, advance to the next extent */
4105 offset
= extent_map_end(em
);
4106 free_extent_map(em
);
4113 static noinline
int count_ext_ref(u64 inum
, u64 offset
, u64 root_id
, void *ctx
)
4115 unsigned long cnt
= *((unsigned long *)ctx
);
4118 *((unsigned long *)ctx
) = cnt
;
4120 /* Now we're sure that the extent is shared. */
4126 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4127 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4131 u64 max
= start
+ len
;
4135 u64 last_for_get_extent
= 0;
4137 u64 isize
= i_size_read(inode
);
4138 struct btrfs_key found_key
;
4139 struct extent_map
*em
= NULL
;
4140 struct extent_state
*cached_state
= NULL
;
4141 struct btrfs_path
*path
;
4150 path
= btrfs_alloc_path();
4153 path
->leave_spinning
= 1;
4155 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
4156 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
4159 * lookup the last file extent. We're not using i_size here
4160 * because there might be preallocation past i_size
4162 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
4163 path
, btrfs_ino(inode
), -1, 0);
4165 btrfs_free_path(path
);
4170 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4171 found_type
= btrfs_key_type(&found_key
);
4173 /* No extents, but there might be delalloc bits */
4174 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4175 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4176 /* have to trust i_size as the end */
4178 last_for_get_extent
= isize
;
4181 * remember the start of the last extent. There are a
4182 * bunch of different factors that go into the length of the
4183 * extent, so its much less complex to remember where it started
4185 last
= found_key
.offset
;
4186 last_for_get_extent
= last
+ 1;
4188 btrfs_release_path(path
);
4191 * we might have some extents allocated but more delalloc past those
4192 * extents. so, we trust isize unless the start of the last extent is
4197 last_for_get_extent
= isize
;
4200 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1, 0,
4203 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4213 u64 offset_in_extent
= 0;
4215 /* break if the extent we found is outside the range */
4216 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4220 * get_extent may return an extent that starts before our
4221 * requested range. We have to make sure the ranges
4222 * we return to fiemap always move forward and don't
4223 * overlap, so adjust the offsets here
4225 em_start
= max(em
->start
, off
);
4228 * record the offset from the start of the extent
4229 * for adjusting the disk offset below. Only do this if the
4230 * extent isn't compressed since our in ram offset may be past
4231 * what we have actually allocated on disk.
4233 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4234 offset_in_extent
= em_start
- em
->start
;
4235 em_end
= extent_map_end(em
);
4236 em_len
= em_end
- em_start
;
4241 * bump off for our next call to get_extent
4243 off
= extent_map_end(em
);
4247 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4249 flags
|= FIEMAP_EXTENT_LAST
;
4250 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4251 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4252 FIEMAP_EXTENT_NOT_ALIGNED
);
4253 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4254 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4255 FIEMAP_EXTENT_UNKNOWN
);
4257 unsigned long ref_cnt
= 0;
4259 disko
= em
->block_start
+ offset_in_extent
;
4262 * As btrfs supports shared space, this information
4263 * can be exported to userspace tools via
4264 * flag FIEMAP_EXTENT_SHARED.
4266 ret
= iterate_inodes_from_logical(
4268 BTRFS_I(inode
)->root
->fs_info
,
4269 path
, count_ext_ref
, &ref_cnt
);
4270 if (ret
< 0 && ret
!= -ENOENT
)
4274 flags
|= FIEMAP_EXTENT_SHARED
;
4276 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4277 flags
|= FIEMAP_EXTENT_ENCODED
;
4279 free_extent_map(em
);
4281 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4282 (last
== (u64
)-1 && isize
<= em_end
)) {
4283 flags
|= FIEMAP_EXTENT_LAST
;
4287 /* now scan forward to see if this is really the last extent. */
4288 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4295 flags
|= FIEMAP_EXTENT_LAST
;
4298 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4304 free_extent_map(em
);
4306 btrfs_free_path(path
);
4307 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4308 &cached_state
, GFP_NOFS
);
4312 static void __free_extent_buffer(struct extent_buffer
*eb
)
4314 btrfs_leak_debug_del(&eb
->leak_list
);
4315 kmem_cache_free(extent_buffer_cache
, eb
);
4318 int extent_buffer_under_io(struct extent_buffer
*eb
)
4320 return (atomic_read(&eb
->io_pages
) ||
4321 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4322 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4326 * Helper for releasing extent buffer page.
4328 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4329 unsigned long start_idx
)
4331 unsigned long index
;
4332 unsigned long num_pages
;
4334 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4336 BUG_ON(extent_buffer_under_io(eb
));
4338 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4339 index
= start_idx
+ num_pages
;
4340 if (start_idx
>= index
)
4345 page
= extent_buffer_page(eb
, index
);
4346 if (page
&& mapped
) {
4347 spin_lock(&page
->mapping
->private_lock
);
4349 * We do this since we'll remove the pages after we've
4350 * removed the eb from the radix tree, so we could race
4351 * and have this page now attached to the new eb. So
4352 * only clear page_private if it's still connected to
4355 if (PagePrivate(page
) &&
4356 page
->private == (unsigned long)eb
) {
4357 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4358 BUG_ON(PageDirty(page
));
4359 BUG_ON(PageWriteback(page
));
4361 * We need to make sure we haven't be attached
4364 ClearPagePrivate(page
);
4365 set_page_private(page
, 0);
4366 /* One for the page private */
4367 page_cache_release(page
);
4369 spin_unlock(&page
->mapping
->private_lock
);
4373 /* One for when we alloced the page */
4374 page_cache_release(page
);
4376 } while (index
!= start_idx
);
4380 * Helper for releasing the extent buffer.
4382 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4384 btrfs_release_extent_buffer_page(eb
, 0);
4385 __free_extent_buffer(eb
);
4388 static struct extent_buffer
*
4389 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4390 unsigned long len
, gfp_t mask
)
4392 struct extent_buffer
*eb
= NULL
;
4394 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
4399 eb
->fs_info
= fs_info
;
4401 rwlock_init(&eb
->lock
);
4402 atomic_set(&eb
->write_locks
, 0);
4403 atomic_set(&eb
->read_locks
, 0);
4404 atomic_set(&eb
->blocking_readers
, 0);
4405 atomic_set(&eb
->blocking_writers
, 0);
4406 atomic_set(&eb
->spinning_readers
, 0);
4407 atomic_set(&eb
->spinning_writers
, 0);
4408 eb
->lock_nested
= 0;
4409 init_waitqueue_head(&eb
->write_lock_wq
);
4410 init_waitqueue_head(&eb
->read_lock_wq
);
4412 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4414 spin_lock_init(&eb
->refs_lock
);
4415 atomic_set(&eb
->refs
, 1);
4416 atomic_set(&eb
->io_pages
, 0);
4419 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4421 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4422 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4423 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4428 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4432 struct extent_buffer
*new;
4433 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4435 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_NOFS
);
4439 for (i
= 0; i
< num_pages
; i
++) {
4440 p
= alloc_page(GFP_NOFS
);
4442 btrfs_release_extent_buffer(new);
4445 attach_extent_buffer_page(new, p
);
4446 WARN_ON(PageDirty(p
));
4451 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4452 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4453 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4458 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
4460 struct extent_buffer
*eb
;
4461 unsigned long num_pages
= num_extent_pages(0, len
);
4464 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_NOFS
);
4468 for (i
= 0; i
< num_pages
; i
++) {
4469 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4473 set_extent_buffer_uptodate(eb
);
4474 btrfs_set_header_nritems(eb
, 0);
4475 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4480 __free_page(eb
->pages
[i
- 1]);
4481 __free_extent_buffer(eb
);
4485 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4488 /* the ref bit is tricky. We have to make sure it is set
4489 * if we have the buffer dirty. Otherwise the
4490 * code to free a buffer can end up dropping a dirty
4493 * Once the ref bit is set, it won't go away while the
4494 * buffer is dirty or in writeback, and it also won't
4495 * go away while we have the reference count on the
4498 * We can't just set the ref bit without bumping the
4499 * ref on the eb because free_extent_buffer might
4500 * see the ref bit and try to clear it. If this happens
4501 * free_extent_buffer might end up dropping our original
4502 * ref by mistake and freeing the page before we are able
4503 * to add one more ref.
4505 * So bump the ref count first, then set the bit. If someone
4506 * beat us to it, drop the ref we added.
4508 refs
= atomic_read(&eb
->refs
);
4509 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4512 spin_lock(&eb
->refs_lock
);
4513 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4514 atomic_inc(&eb
->refs
);
4515 spin_unlock(&eb
->refs_lock
);
4518 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
)
4520 unsigned long num_pages
, i
;
4522 check_buffer_tree_ref(eb
);
4524 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4525 for (i
= 0; i
< num_pages
; i
++) {
4526 struct page
*p
= extent_buffer_page(eb
, i
);
4527 mark_page_accessed(p
);
4531 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4534 struct extent_buffer
*eb
;
4537 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4538 start
>> PAGE_CACHE_SHIFT
);
4539 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4541 mark_extent_buffer_accessed(eb
);
4549 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4550 u64 start
, unsigned long len
)
4552 unsigned long num_pages
= num_extent_pages(start
, len
);
4554 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4555 struct extent_buffer
*eb
;
4556 struct extent_buffer
*exists
= NULL
;
4558 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4562 eb
= find_extent_buffer(fs_info
, start
);
4566 eb
= __alloc_extent_buffer(fs_info
, start
, len
, GFP_NOFS
);
4570 for (i
= 0; i
< num_pages
; i
++, index
++) {
4571 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4575 spin_lock(&mapping
->private_lock
);
4576 if (PagePrivate(p
)) {
4578 * We could have already allocated an eb for this page
4579 * and attached one so lets see if we can get a ref on
4580 * the existing eb, and if we can we know it's good and
4581 * we can just return that one, else we know we can just
4582 * overwrite page->private.
4584 exists
= (struct extent_buffer
*)p
->private;
4585 if (atomic_inc_not_zero(&exists
->refs
)) {
4586 spin_unlock(&mapping
->private_lock
);
4588 page_cache_release(p
);
4589 mark_extent_buffer_accessed(exists
);
4594 * Do this so attach doesn't complain and we need to
4595 * drop the ref the old guy had.
4597 ClearPagePrivate(p
);
4598 WARN_ON(PageDirty(p
));
4599 page_cache_release(p
);
4601 attach_extent_buffer_page(eb
, p
);
4602 spin_unlock(&mapping
->private_lock
);
4603 WARN_ON(PageDirty(p
));
4604 mark_page_accessed(p
);
4606 if (!PageUptodate(p
))
4610 * see below about how we avoid a nasty race with release page
4611 * and why we unlock later
4615 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4617 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4621 spin_lock(&fs_info
->buffer_lock
);
4622 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4623 start
>> PAGE_CACHE_SHIFT
, eb
);
4624 spin_unlock(&fs_info
->buffer_lock
);
4625 radix_tree_preload_end();
4626 if (ret
== -EEXIST
) {
4627 exists
= find_extent_buffer(fs_info
, start
);
4633 /* add one reference for the tree */
4634 check_buffer_tree_ref(eb
);
4635 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4638 * there is a race where release page may have
4639 * tried to find this extent buffer in the radix
4640 * but failed. It will tell the VM it is safe to
4641 * reclaim the, and it will clear the page private bit.
4642 * We must make sure to set the page private bit properly
4643 * after the extent buffer is in the radix tree so
4644 * it doesn't get lost
4646 SetPageChecked(eb
->pages
[0]);
4647 for (i
= 1; i
< num_pages
; i
++) {
4648 p
= extent_buffer_page(eb
, i
);
4649 ClearPageChecked(p
);
4652 unlock_page(eb
->pages
[0]);
4656 for (i
= 0; i
< num_pages
; i
++) {
4658 unlock_page(eb
->pages
[i
]);
4661 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4662 btrfs_release_extent_buffer(eb
);
4666 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4668 struct extent_buffer
*eb
=
4669 container_of(head
, struct extent_buffer
, rcu_head
);
4671 __free_extent_buffer(eb
);
4674 /* Expects to have eb->eb_lock already held */
4675 static int release_extent_buffer(struct extent_buffer
*eb
)
4677 WARN_ON(atomic_read(&eb
->refs
) == 0);
4678 if (atomic_dec_and_test(&eb
->refs
)) {
4679 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
4680 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
4682 spin_unlock(&eb
->refs_lock
);
4684 spin_lock(&fs_info
->buffer_lock
);
4685 radix_tree_delete(&fs_info
->buffer_radix
,
4686 eb
->start
>> PAGE_CACHE_SHIFT
);
4687 spin_unlock(&fs_info
->buffer_lock
);
4689 spin_unlock(&eb
->refs_lock
);
4692 /* Should be safe to release our pages at this point */
4693 btrfs_release_extent_buffer_page(eb
, 0);
4694 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4697 spin_unlock(&eb
->refs_lock
);
4702 void free_extent_buffer(struct extent_buffer
*eb
)
4710 refs
= atomic_read(&eb
->refs
);
4713 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
4718 spin_lock(&eb
->refs_lock
);
4719 if (atomic_read(&eb
->refs
) == 2 &&
4720 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4721 atomic_dec(&eb
->refs
);
4723 if (atomic_read(&eb
->refs
) == 2 &&
4724 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4725 !extent_buffer_under_io(eb
) &&
4726 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4727 atomic_dec(&eb
->refs
);
4730 * I know this is terrible, but it's temporary until we stop tracking
4731 * the uptodate bits and such for the extent buffers.
4733 release_extent_buffer(eb
);
4736 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4741 spin_lock(&eb
->refs_lock
);
4742 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4744 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4745 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4746 atomic_dec(&eb
->refs
);
4747 release_extent_buffer(eb
);
4750 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4753 unsigned long num_pages
;
4756 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4758 for (i
= 0; i
< num_pages
; i
++) {
4759 page
= extent_buffer_page(eb
, i
);
4760 if (!PageDirty(page
))
4764 WARN_ON(!PagePrivate(page
));
4766 clear_page_dirty_for_io(page
);
4767 spin_lock_irq(&page
->mapping
->tree_lock
);
4768 if (!PageDirty(page
)) {
4769 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4771 PAGECACHE_TAG_DIRTY
);
4773 spin_unlock_irq(&page
->mapping
->tree_lock
);
4774 ClearPageError(page
);
4777 WARN_ON(atomic_read(&eb
->refs
) == 0);
4780 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4783 unsigned long num_pages
;
4786 check_buffer_tree_ref(eb
);
4788 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4790 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4791 WARN_ON(atomic_read(&eb
->refs
) == 0);
4792 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4794 for (i
= 0; i
< num_pages
; i
++)
4795 set_page_dirty(extent_buffer_page(eb
, i
));
4799 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4803 unsigned long num_pages
;
4805 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4806 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4807 for (i
= 0; i
< num_pages
; i
++) {
4808 page
= extent_buffer_page(eb
, i
);
4810 ClearPageUptodate(page
);
4815 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4819 unsigned long num_pages
;
4821 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4822 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4823 for (i
= 0; i
< num_pages
; i
++) {
4824 page
= extent_buffer_page(eb
, i
);
4825 SetPageUptodate(page
);
4830 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4832 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4835 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4836 struct extent_buffer
*eb
, u64 start
, int wait
,
4837 get_extent_t
*get_extent
, int mirror_num
)
4840 unsigned long start_i
;
4844 int locked_pages
= 0;
4845 int all_uptodate
= 1;
4846 unsigned long num_pages
;
4847 unsigned long num_reads
= 0;
4848 struct bio
*bio
= NULL
;
4849 unsigned long bio_flags
= 0;
4851 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4855 WARN_ON(start
< eb
->start
);
4856 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4857 (eb
->start
>> PAGE_CACHE_SHIFT
);
4862 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4863 for (i
= start_i
; i
< num_pages
; i
++) {
4864 page
= extent_buffer_page(eb
, i
);
4865 if (wait
== WAIT_NONE
) {
4866 if (!trylock_page(page
))
4872 if (!PageUptodate(page
)) {
4879 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4883 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4884 eb
->read_mirror
= 0;
4885 atomic_set(&eb
->io_pages
, num_reads
);
4886 for (i
= start_i
; i
< num_pages
; i
++) {
4887 page
= extent_buffer_page(eb
, i
);
4888 if (!PageUptodate(page
)) {
4889 ClearPageError(page
);
4890 err
= __extent_read_full_page(tree
, page
,
4892 mirror_num
, &bio_flags
,
4902 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
4908 if (ret
|| wait
!= WAIT_COMPLETE
)
4911 for (i
= start_i
; i
< num_pages
; i
++) {
4912 page
= extent_buffer_page(eb
, i
);
4913 wait_on_page_locked(page
);
4914 if (!PageUptodate(page
))
4922 while (locked_pages
> 0) {
4923 page
= extent_buffer_page(eb
, i
);
4931 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
4932 unsigned long start
,
4939 char *dst
= (char *)dstv
;
4940 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4941 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4943 WARN_ON(start
> eb
->len
);
4944 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4946 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
4949 page
= extent_buffer_page(eb
, i
);
4951 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4952 kaddr
= page_address(page
);
4953 memcpy(dst
, kaddr
+ offset
, cur
);
4962 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
4963 unsigned long min_len
, char **map
,
4964 unsigned long *map_start
,
4965 unsigned long *map_len
)
4967 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
4970 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4971 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4972 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
4979 offset
= start_offset
;
4983 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
4986 if (start
+ min_len
> eb
->len
) {
4987 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
4989 eb
->start
, eb
->len
, start
, min_len
);
4993 p
= extent_buffer_page(eb
, i
);
4994 kaddr
= page_address(p
);
4995 *map
= kaddr
+ offset
;
4996 *map_len
= PAGE_CACHE_SIZE
- offset
;
5000 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5001 unsigned long start
,
5008 char *ptr
= (char *)ptrv
;
5009 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5010 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5013 WARN_ON(start
> eb
->len
);
5014 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5016 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5019 page
= extent_buffer_page(eb
, i
);
5021 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5023 kaddr
= page_address(page
);
5024 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5036 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5037 unsigned long start
, unsigned long len
)
5043 char *src
= (char *)srcv
;
5044 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5045 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5047 WARN_ON(start
> eb
->len
);
5048 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5050 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5053 page
= extent_buffer_page(eb
, i
);
5054 WARN_ON(!PageUptodate(page
));
5056 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5057 kaddr
= page_address(page
);
5058 memcpy(kaddr
+ offset
, src
, cur
);
5067 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5068 unsigned long start
, unsigned long len
)
5074 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5075 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5077 WARN_ON(start
> eb
->len
);
5078 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5080 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5083 page
= extent_buffer_page(eb
, i
);
5084 WARN_ON(!PageUptodate(page
));
5086 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5087 kaddr
= page_address(page
);
5088 memset(kaddr
+ offset
, c
, cur
);
5096 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5097 unsigned long dst_offset
, unsigned long src_offset
,
5100 u64 dst_len
= dst
->len
;
5105 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5106 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5108 WARN_ON(src
->len
!= dst_len
);
5110 offset
= (start_offset
+ dst_offset
) &
5111 (PAGE_CACHE_SIZE
- 1);
5114 page
= extent_buffer_page(dst
, i
);
5115 WARN_ON(!PageUptodate(page
));
5117 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
5119 kaddr
= page_address(page
);
5120 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5129 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5131 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5132 return distance
< len
;
5135 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5136 unsigned long dst_off
, unsigned long src_off
,
5139 char *dst_kaddr
= page_address(dst_page
);
5141 int must_memmove
= 0;
5143 if (dst_page
!= src_page
) {
5144 src_kaddr
= page_address(src_page
);
5146 src_kaddr
= dst_kaddr
;
5147 if (areas_overlap(src_off
, dst_off
, len
))
5152 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5154 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5157 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5158 unsigned long src_offset
, unsigned long len
)
5161 size_t dst_off_in_page
;
5162 size_t src_off_in_page
;
5163 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5164 unsigned long dst_i
;
5165 unsigned long src_i
;
5167 if (src_offset
+ len
> dst
->len
) {
5168 printk(KERN_ERR
"BTRFS: memmove bogus src_offset %lu move "
5169 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
5172 if (dst_offset
+ len
> dst
->len
) {
5173 printk(KERN_ERR
"BTRFS: memmove bogus dst_offset %lu move "
5174 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
5179 dst_off_in_page
= (start_offset
+ dst_offset
) &
5180 (PAGE_CACHE_SIZE
- 1);
5181 src_off_in_page
= (start_offset
+ src_offset
) &
5182 (PAGE_CACHE_SIZE
- 1);
5184 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5185 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5187 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5189 cur
= min_t(unsigned long, cur
,
5190 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5192 copy_pages(extent_buffer_page(dst
, dst_i
),
5193 extent_buffer_page(dst
, src_i
),
5194 dst_off_in_page
, src_off_in_page
, cur
);
5202 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5203 unsigned long src_offset
, unsigned long len
)
5206 size_t dst_off_in_page
;
5207 size_t src_off_in_page
;
5208 unsigned long dst_end
= dst_offset
+ len
- 1;
5209 unsigned long src_end
= src_offset
+ len
- 1;
5210 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5211 unsigned long dst_i
;
5212 unsigned long src_i
;
5214 if (src_offset
+ len
> dst
->len
) {
5215 printk(KERN_ERR
"BTRFS: memmove bogus src_offset %lu move "
5216 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
5219 if (dst_offset
+ len
> dst
->len
) {
5220 printk(KERN_ERR
"BTRFS: memmove bogus dst_offset %lu move "
5221 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
5224 if (dst_offset
< src_offset
) {
5225 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5229 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5230 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5232 dst_off_in_page
= (start_offset
+ dst_end
) &
5233 (PAGE_CACHE_SIZE
- 1);
5234 src_off_in_page
= (start_offset
+ src_end
) &
5235 (PAGE_CACHE_SIZE
- 1);
5237 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5238 cur
= min(cur
, dst_off_in_page
+ 1);
5239 copy_pages(extent_buffer_page(dst
, dst_i
),
5240 extent_buffer_page(dst
, src_i
),
5241 dst_off_in_page
- cur
+ 1,
5242 src_off_in_page
- cur
+ 1, cur
);
5250 int try_release_extent_buffer(struct page
*page
)
5252 struct extent_buffer
*eb
;
5255 * We need to make sure noboody is attaching this page to an eb right
5258 spin_lock(&page
->mapping
->private_lock
);
5259 if (!PagePrivate(page
)) {
5260 spin_unlock(&page
->mapping
->private_lock
);
5264 eb
= (struct extent_buffer
*)page
->private;
5268 * This is a little awful but should be ok, we need to make sure that
5269 * the eb doesn't disappear out from under us while we're looking at
5272 spin_lock(&eb
->refs_lock
);
5273 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5274 spin_unlock(&eb
->refs_lock
);
5275 spin_unlock(&page
->mapping
->private_lock
);
5278 spin_unlock(&page
->mapping
->private_lock
);
5281 * If tree ref isn't set then we know the ref on this eb is a real ref,
5282 * so just return, this page will likely be freed soon anyway.
5284 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5285 spin_unlock(&eb
->refs_lock
);
5289 return release_extent_buffer(eb
);