1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 #include "check-integrity.h"
22 #include "rcu-string.h"
24 static struct kmem_cache
*extent_state_cache
;
25 static struct kmem_cache
*extent_buffer_cache
;
26 static struct bio_set
*btrfs_bioset
;
28 #ifdef CONFIG_BTRFS_DEBUG
29 static LIST_HEAD(buffers
);
30 static LIST_HEAD(states
);
32 static DEFINE_SPINLOCK(leak_lock
);
35 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
39 spin_lock_irqsave(&leak_lock
, flags
);
41 spin_unlock_irqrestore(&leak_lock
, flags
);
45 void btrfs_leak_debug_del(struct list_head
*entry
)
49 spin_lock_irqsave(&leak_lock
, flags
);
51 spin_unlock_irqrestore(&leak_lock
, flags
);
55 void btrfs_leak_debug_check(void)
57 struct extent_state
*state
;
58 struct extent_buffer
*eb
;
60 while (!list_empty(&states
)) {
61 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
62 printk(KERN_ERR
"btrfs state leak: start %llu end %llu "
63 "state %lu in tree %p refs %d\n",
64 (unsigned long long)state
->start
,
65 (unsigned long long)state
->end
,
66 state
->state
, state
->tree
, atomic_read(&state
->refs
));
67 list_del(&state
->leak_list
);
68 kmem_cache_free(extent_state_cache
, state
);
71 while (!list_empty(&buffers
)) {
72 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
73 printk(KERN_ERR
"btrfs buffer leak start %llu len %lu "
74 "refs %d\n", (unsigned long long)eb
->start
,
75 eb
->len
, atomic_read(&eb
->refs
));
76 list_del(&eb
->leak_list
);
77 kmem_cache_free(extent_buffer_cache
, eb
);
81 #define btrfs_debug_check_extent_io_range(inode, start, end) \
82 __btrfs_debug_check_extent_io_range(__func__, (inode), (start), (end))
83 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
84 struct inode
*inode
, u64 start
, u64 end
)
86 u64 isize
= i_size_read(inode
);
88 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
89 printk_ratelimited(KERN_DEBUG
90 "btrfs: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
92 (unsigned long long)btrfs_ino(inode
),
93 (unsigned long long)isize
,
94 (unsigned long long)start
,
95 (unsigned long long)end
);
99 #define btrfs_leak_debug_add(new, head) do {} while (0)
100 #define btrfs_leak_debug_del(entry) do {} while (0)
101 #define btrfs_leak_debug_check() do {} while (0)
102 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
105 #define BUFFER_LRU_MAX 64
110 struct rb_node rb_node
;
113 struct extent_page_data
{
115 struct extent_io_tree
*tree
;
116 get_extent_t
*get_extent
;
117 unsigned long bio_flags
;
119 /* tells writepage not to lock the state bits for this range
120 * it still does the unlocking
122 unsigned int extent_locked
:1;
124 /* tells the submit_bio code to use a WRITE_SYNC */
125 unsigned int sync_io
:1;
128 static noinline
void flush_write_bio(void *data
);
129 static inline struct btrfs_fs_info
*
130 tree_fs_info(struct extent_io_tree
*tree
)
132 return btrfs_sb(tree
->mapping
->host
->i_sb
);
135 int __init
extent_io_init(void)
137 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
138 sizeof(struct extent_state
), 0,
139 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
140 if (!extent_state_cache
)
143 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
144 sizeof(struct extent_buffer
), 0,
145 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
146 if (!extent_buffer_cache
)
147 goto free_state_cache
;
149 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
150 offsetof(struct btrfs_io_bio
, bio
));
152 goto free_buffer_cache
;
156 kmem_cache_destroy(extent_buffer_cache
);
157 extent_buffer_cache
= NULL
;
160 kmem_cache_destroy(extent_state_cache
);
161 extent_state_cache
= NULL
;
165 void extent_io_exit(void)
167 btrfs_leak_debug_check();
170 * Make sure all delayed rcu free are flushed before we
174 if (extent_state_cache
)
175 kmem_cache_destroy(extent_state_cache
);
176 if (extent_buffer_cache
)
177 kmem_cache_destroy(extent_buffer_cache
);
179 bioset_free(btrfs_bioset
);
182 void extent_io_tree_init(struct extent_io_tree
*tree
,
183 struct address_space
*mapping
)
185 tree
->state
= RB_ROOT
;
186 INIT_RADIX_TREE(&tree
->buffer
, GFP_ATOMIC
);
188 tree
->dirty_bytes
= 0;
189 spin_lock_init(&tree
->lock
);
190 spin_lock_init(&tree
->buffer_lock
);
191 tree
->mapping
= mapping
;
194 static struct extent_state
*alloc_extent_state(gfp_t mask
)
196 struct extent_state
*state
;
198 state
= kmem_cache_alloc(extent_state_cache
, mask
);
204 btrfs_leak_debug_add(&state
->leak_list
, &states
);
205 atomic_set(&state
->refs
, 1);
206 init_waitqueue_head(&state
->wq
);
207 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
211 void free_extent_state(struct extent_state
*state
)
215 if (atomic_dec_and_test(&state
->refs
)) {
216 WARN_ON(state
->tree
);
217 btrfs_leak_debug_del(&state
->leak_list
);
218 trace_free_extent_state(state
, _RET_IP_
);
219 kmem_cache_free(extent_state_cache
, state
);
223 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
224 struct rb_node
*node
)
226 struct rb_node
**p
= &root
->rb_node
;
227 struct rb_node
*parent
= NULL
;
228 struct tree_entry
*entry
;
232 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
234 if (offset
< entry
->start
)
236 else if (offset
> entry
->end
)
242 rb_link_node(node
, parent
, p
);
243 rb_insert_color(node
, root
);
247 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
248 struct rb_node
**prev_ret
,
249 struct rb_node
**next_ret
)
251 struct rb_root
*root
= &tree
->state
;
252 struct rb_node
*n
= root
->rb_node
;
253 struct rb_node
*prev
= NULL
;
254 struct rb_node
*orig_prev
= NULL
;
255 struct tree_entry
*entry
;
256 struct tree_entry
*prev_entry
= NULL
;
259 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
263 if (offset
< entry
->start
)
265 else if (offset
> entry
->end
)
273 while (prev
&& offset
> prev_entry
->end
) {
274 prev
= rb_next(prev
);
275 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
282 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
283 while (prev
&& offset
< prev_entry
->start
) {
284 prev
= rb_prev(prev
);
285 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
292 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
295 struct rb_node
*prev
= NULL
;
298 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
304 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
305 struct extent_state
*other
)
307 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
308 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
313 * utility function to look for merge candidates inside a given range.
314 * Any extents with matching state are merged together into a single
315 * extent in the tree. Extents with EXTENT_IO in their state field
316 * are not merged because the end_io handlers need to be able to do
317 * operations on them without sleeping (or doing allocations/splits).
319 * This should be called with the tree lock held.
321 static void merge_state(struct extent_io_tree
*tree
,
322 struct extent_state
*state
)
324 struct extent_state
*other
;
325 struct rb_node
*other_node
;
327 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
330 other_node
= rb_prev(&state
->rb_node
);
332 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
333 if (other
->end
== state
->start
- 1 &&
334 other
->state
== state
->state
) {
335 merge_cb(tree
, state
, other
);
336 state
->start
= other
->start
;
338 rb_erase(&other
->rb_node
, &tree
->state
);
339 free_extent_state(other
);
342 other_node
= rb_next(&state
->rb_node
);
344 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
345 if (other
->start
== state
->end
+ 1 &&
346 other
->state
== state
->state
) {
347 merge_cb(tree
, state
, other
);
348 state
->end
= other
->end
;
350 rb_erase(&other
->rb_node
, &tree
->state
);
351 free_extent_state(other
);
356 static void set_state_cb(struct extent_io_tree
*tree
,
357 struct extent_state
*state
, unsigned long *bits
)
359 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
360 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
363 static void clear_state_cb(struct extent_io_tree
*tree
,
364 struct extent_state
*state
, unsigned long *bits
)
366 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
367 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
370 static void set_state_bits(struct extent_io_tree
*tree
,
371 struct extent_state
*state
, unsigned long *bits
);
374 * insert an extent_state struct into the tree. 'bits' are set on the
375 * struct before it is inserted.
377 * This may return -EEXIST if the extent is already there, in which case the
378 * state struct is freed.
380 * The tree lock is not taken internally. This is a utility function and
381 * probably isn't what you want to call (see set/clear_extent_bit).
383 static int insert_state(struct extent_io_tree
*tree
,
384 struct extent_state
*state
, u64 start
, u64 end
,
387 struct rb_node
*node
;
390 WARN(1, KERN_ERR
"btrfs end < start %llu %llu\n",
391 (unsigned long long)end
,
392 (unsigned long long)start
);
393 state
->start
= start
;
396 set_state_bits(tree
, state
, bits
);
398 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
400 struct extent_state
*found
;
401 found
= rb_entry(node
, struct extent_state
, rb_node
);
402 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
403 "%llu %llu\n", (unsigned long long)found
->start
,
404 (unsigned long long)found
->end
,
405 (unsigned long long)start
, (unsigned long long)end
);
409 merge_state(tree
, state
);
413 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
416 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
417 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
421 * split a given extent state struct in two, inserting the preallocated
422 * struct 'prealloc' as the newly created second half. 'split' indicates an
423 * offset inside 'orig' where it should be split.
426 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
427 * are two extent state structs in the tree:
428 * prealloc: [orig->start, split - 1]
429 * orig: [ split, orig->end ]
431 * The tree locks are not taken by this function. They need to be held
434 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
435 struct extent_state
*prealloc
, u64 split
)
437 struct rb_node
*node
;
439 split_cb(tree
, orig
, split
);
441 prealloc
->start
= orig
->start
;
442 prealloc
->end
= split
- 1;
443 prealloc
->state
= orig
->state
;
446 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
448 free_extent_state(prealloc
);
451 prealloc
->tree
= tree
;
455 static struct extent_state
*next_state(struct extent_state
*state
)
457 struct rb_node
*next
= rb_next(&state
->rb_node
);
459 return rb_entry(next
, struct extent_state
, rb_node
);
465 * utility function to clear some bits in an extent state struct.
466 * it will optionally wake up any one waiting on this state (wake == 1).
468 * If no bits are set on the state struct after clearing things, the
469 * struct is freed and removed from the tree
471 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
472 struct extent_state
*state
,
473 unsigned long *bits
, int wake
)
475 struct extent_state
*next
;
476 unsigned long bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
478 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
479 u64 range
= state
->end
- state
->start
+ 1;
480 WARN_ON(range
> tree
->dirty_bytes
);
481 tree
->dirty_bytes
-= range
;
483 clear_state_cb(tree
, state
, bits
);
484 state
->state
&= ~bits_to_clear
;
487 if (state
->state
== 0) {
488 next
= next_state(state
);
490 rb_erase(&state
->rb_node
, &tree
->state
);
492 free_extent_state(state
);
497 merge_state(tree
, state
);
498 next
= next_state(state
);
503 static struct extent_state
*
504 alloc_extent_state_atomic(struct extent_state
*prealloc
)
507 prealloc
= alloc_extent_state(GFP_ATOMIC
);
512 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
514 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
515 "Extent tree was modified by another "
516 "thread while locked.");
520 * clear some bits on a range in the tree. This may require splitting
521 * or inserting elements in the tree, so the gfp mask is used to
522 * indicate which allocations or sleeping are allowed.
524 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
525 * the given range from the tree regardless of state (ie for truncate).
527 * the range [start, end] is inclusive.
529 * This takes the tree lock, and returns 0 on success and < 0 on error.
531 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
532 unsigned long bits
, int wake
, int delete,
533 struct extent_state
**cached_state
,
536 struct extent_state
*state
;
537 struct extent_state
*cached
;
538 struct extent_state
*prealloc
= NULL
;
539 struct rb_node
*node
;
544 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
547 bits
|= ~EXTENT_CTLBITS
;
548 bits
|= EXTENT_FIRST_DELALLOC
;
550 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
553 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
554 prealloc
= alloc_extent_state(mask
);
559 spin_lock(&tree
->lock
);
561 cached
= *cached_state
;
564 *cached_state
= NULL
;
568 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
569 cached
->end
> start
) {
571 atomic_dec(&cached
->refs
);
576 free_extent_state(cached
);
579 * this search will find the extents that end after
582 node
= tree_search(tree
, start
);
585 state
= rb_entry(node
, struct extent_state
, rb_node
);
587 if (state
->start
> end
)
589 WARN_ON(state
->end
< start
);
590 last_end
= state
->end
;
592 /* the state doesn't have the wanted bits, go ahead */
593 if (!(state
->state
& bits
)) {
594 state
= next_state(state
);
599 * | ---- desired range ---- |
601 * | ------------- state -------------- |
603 * We need to split the extent we found, and may flip
604 * bits on second half.
606 * If the extent we found extends past our range, we
607 * just split and search again. It'll get split again
608 * the next time though.
610 * If the extent we found is inside our range, we clear
611 * the desired bit on it.
614 if (state
->start
< start
) {
615 prealloc
= alloc_extent_state_atomic(prealloc
);
617 err
= split_state(tree
, state
, prealloc
, start
);
619 extent_io_tree_panic(tree
, err
);
624 if (state
->end
<= end
) {
625 state
= clear_state_bit(tree
, state
, &bits
, wake
);
631 * | ---- desired range ---- |
633 * We need to split the extent, and clear the bit
636 if (state
->start
<= end
&& state
->end
> end
) {
637 prealloc
= alloc_extent_state_atomic(prealloc
);
639 err
= split_state(tree
, state
, prealloc
, end
+ 1);
641 extent_io_tree_panic(tree
, err
);
646 clear_state_bit(tree
, prealloc
, &bits
, wake
);
652 state
= clear_state_bit(tree
, state
, &bits
, wake
);
654 if (last_end
== (u64
)-1)
656 start
= last_end
+ 1;
657 if (start
<= end
&& state
&& !need_resched())
662 spin_unlock(&tree
->lock
);
664 free_extent_state(prealloc
);
671 spin_unlock(&tree
->lock
);
672 if (mask
& __GFP_WAIT
)
677 static void wait_on_state(struct extent_io_tree
*tree
,
678 struct extent_state
*state
)
679 __releases(tree
->lock
)
680 __acquires(tree
->lock
)
683 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
684 spin_unlock(&tree
->lock
);
686 spin_lock(&tree
->lock
);
687 finish_wait(&state
->wq
, &wait
);
691 * waits for one or more bits to clear on a range in the state tree.
692 * The range [start, end] is inclusive.
693 * The tree lock is taken by this function
695 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
698 struct extent_state
*state
;
699 struct rb_node
*node
;
701 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
703 spin_lock(&tree
->lock
);
707 * this search will find all the extents that end after
710 node
= tree_search(tree
, start
);
714 state
= rb_entry(node
, struct extent_state
, rb_node
);
716 if (state
->start
> end
)
719 if (state
->state
& bits
) {
720 start
= state
->start
;
721 atomic_inc(&state
->refs
);
722 wait_on_state(tree
, state
);
723 free_extent_state(state
);
726 start
= state
->end
+ 1;
731 cond_resched_lock(&tree
->lock
);
734 spin_unlock(&tree
->lock
);
737 static void set_state_bits(struct extent_io_tree
*tree
,
738 struct extent_state
*state
,
741 unsigned long bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
743 set_state_cb(tree
, state
, bits
);
744 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
745 u64 range
= state
->end
- state
->start
+ 1;
746 tree
->dirty_bytes
+= range
;
748 state
->state
|= bits_to_set
;
751 static void cache_state(struct extent_state
*state
,
752 struct extent_state
**cached_ptr
)
754 if (cached_ptr
&& !(*cached_ptr
)) {
755 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
757 atomic_inc(&state
->refs
);
762 static void uncache_state(struct extent_state
**cached_ptr
)
764 if (cached_ptr
&& (*cached_ptr
)) {
765 struct extent_state
*state
= *cached_ptr
;
767 free_extent_state(state
);
772 * set some bits on a range in the tree. This may require allocations or
773 * sleeping, so the gfp mask is used to indicate what is allowed.
775 * If any of the exclusive bits are set, this will fail with -EEXIST if some
776 * part of the range already has the desired bits set. The start of the
777 * existing range is returned in failed_start in this case.
779 * [start, end] is inclusive This takes the tree lock.
782 static int __must_check
783 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
784 unsigned long bits
, unsigned long exclusive_bits
,
785 u64
*failed_start
, struct extent_state
**cached_state
,
788 struct extent_state
*state
;
789 struct extent_state
*prealloc
= NULL
;
790 struct rb_node
*node
;
795 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
797 bits
|= EXTENT_FIRST_DELALLOC
;
799 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
800 prealloc
= alloc_extent_state(mask
);
804 spin_lock(&tree
->lock
);
805 if (cached_state
&& *cached_state
) {
806 state
= *cached_state
;
807 if (state
->start
<= start
&& state
->end
> start
&&
809 node
= &state
->rb_node
;
814 * this search will find all the extents that end after
817 node
= tree_search(tree
, start
);
819 prealloc
= alloc_extent_state_atomic(prealloc
);
821 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
823 extent_io_tree_panic(tree
, err
);
828 state
= rb_entry(node
, struct extent_state
, rb_node
);
830 last_start
= state
->start
;
831 last_end
= state
->end
;
834 * | ---- desired range ---- |
837 * Just lock what we found and keep going
839 if (state
->start
== start
&& state
->end
<= end
) {
840 if (state
->state
& exclusive_bits
) {
841 *failed_start
= state
->start
;
846 set_state_bits(tree
, state
, &bits
);
847 cache_state(state
, cached_state
);
848 merge_state(tree
, state
);
849 if (last_end
== (u64
)-1)
851 start
= last_end
+ 1;
852 state
= next_state(state
);
853 if (start
< end
&& state
&& state
->start
== start
&&
860 * | ---- desired range ---- |
863 * | ------------- state -------------- |
865 * We need to split the extent we found, and may flip bits on
868 * If the extent we found extends past our
869 * range, we just split and search again. It'll get split
870 * again the next time though.
872 * If the extent we found is inside our range, we set the
875 if (state
->start
< start
) {
876 if (state
->state
& exclusive_bits
) {
877 *failed_start
= start
;
882 prealloc
= alloc_extent_state_atomic(prealloc
);
884 err
= split_state(tree
, state
, prealloc
, start
);
886 extent_io_tree_panic(tree
, err
);
891 if (state
->end
<= end
) {
892 set_state_bits(tree
, state
, &bits
);
893 cache_state(state
, cached_state
);
894 merge_state(tree
, state
);
895 if (last_end
== (u64
)-1)
897 start
= last_end
+ 1;
898 state
= next_state(state
);
899 if (start
< end
&& state
&& state
->start
== start
&&
906 * | ---- desired range ---- |
907 * | state | or | state |
909 * There's a hole, we need to insert something in it and
910 * ignore the extent we found.
912 if (state
->start
> start
) {
914 if (end
< last_start
)
917 this_end
= last_start
- 1;
919 prealloc
= alloc_extent_state_atomic(prealloc
);
923 * Avoid to free 'prealloc' if it can be merged with
926 err
= insert_state(tree
, prealloc
, start
, this_end
,
929 extent_io_tree_panic(tree
, err
);
931 cache_state(prealloc
, cached_state
);
933 start
= this_end
+ 1;
937 * | ---- desired range ---- |
939 * We need to split the extent, and set the bit
942 if (state
->start
<= end
&& state
->end
> end
) {
943 if (state
->state
& exclusive_bits
) {
944 *failed_start
= start
;
949 prealloc
= alloc_extent_state_atomic(prealloc
);
951 err
= split_state(tree
, state
, prealloc
, end
+ 1);
953 extent_io_tree_panic(tree
, err
);
955 set_state_bits(tree
, prealloc
, &bits
);
956 cache_state(prealloc
, cached_state
);
957 merge_state(tree
, prealloc
);
965 spin_unlock(&tree
->lock
);
967 free_extent_state(prealloc
);
974 spin_unlock(&tree
->lock
);
975 if (mask
& __GFP_WAIT
)
980 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
981 unsigned long bits
, u64
* failed_start
,
982 struct extent_state
**cached_state
, gfp_t mask
)
984 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
990 * convert_extent_bit - convert all bits in a given range from one bit to
992 * @tree: the io tree to search
993 * @start: the start offset in bytes
994 * @end: the end offset in bytes (inclusive)
995 * @bits: the bits to set in this range
996 * @clear_bits: the bits to clear in this range
997 * @cached_state: state that we're going to cache
998 * @mask: the allocation mask
1000 * This will go through and set bits for the given range. If any states exist
1001 * already in this range they are set with the given bit and cleared of the
1002 * clear_bits. This is only meant to be used by things that are mergeable, ie
1003 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1004 * boundary bits like LOCK.
1006 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1007 unsigned long bits
, unsigned long clear_bits
,
1008 struct extent_state
**cached_state
, gfp_t mask
)
1010 struct extent_state
*state
;
1011 struct extent_state
*prealloc
= NULL
;
1012 struct rb_node
*node
;
1017 btrfs_debug_check_extent_io_range(tree
->mapping
->host
, start
, end
);
1020 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
1021 prealloc
= alloc_extent_state(mask
);
1026 spin_lock(&tree
->lock
);
1027 if (cached_state
&& *cached_state
) {
1028 state
= *cached_state
;
1029 if (state
->start
<= start
&& state
->end
> start
&&
1031 node
= &state
->rb_node
;
1037 * this search will find all the extents that end after
1040 node
= tree_search(tree
, start
);
1042 prealloc
= alloc_extent_state_atomic(prealloc
);
1047 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
1050 extent_io_tree_panic(tree
, err
);
1053 state
= rb_entry(node
, struct extent_state
, rb_node
);
1055 last_start
= state
->start
;
1056 last_end
= state
->end
;
1059 * | ---- desired range ---- |
1062 * Just lock what we found and keep going
1064 if (state
->start
== start
&& state
->end
<= end
) {
1065 set_state_bits(tree
, state
, &bits
);
1066 cache_state(state
, cached_state
);
1067 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1068 if (last_end
== (u64
)-1)
1070 start
= last_end
+ 1;
1071 if (start
< end
&& state
&& state
->start
== start
&&
1078 * | ---- desired range ---- |
1081 * | ------------- state -------------- |
1083 * We need to split the extent we found, and may flip bits on
1086 * If the extent we found extends past our
1087 * range, we just split and search again. It'll get split
1088 * again the next time though.
1090 * If the extent we found is inside our range, we set the
1091 * desired bit on it.
1093 if (state
->start
< start
) {
1094 prealloc
= alloc_extent_state_atomic(prealloc
);
1099 err
= split_state(tree
, state
, prealloc
, start
);
1101 extent_io_tree_panic(tree
, err
);
1105 if (state
->end
<= end
) {
1106 set_state_bits(tree
, state
, &bits
);
1107 cache_state(state
, cached_state
);
1108 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1109 if (last_end
== (u64
)-1)
1111 start
= last_end
+ 1;
1112 if (start
< end
&& state
&& state
->start
== start
&&
1119 * | ---- desired range ---- |
1120 * | state | or | state |
1122 * There's a hole, we need to insert something in it and
1123 * ignore the extent we found.
1125 if (state
->start
> start
) {
1127 if (end
< last_start
)
1130 this_end
= last_start
- 1;
1132 prealloc
= alloc_extent_state_atomic(prealloc
);
1139 * Avoid to free 'prealloc' if it can be merged with
1142 err
= insert_state(tree
, prealloc
, start
, this_end
,
1145 extent_io_tree_panic(tree
, err
);
1146 cache_state(prealloc
, cached_state
);
1148 start
= this_end
+ 1;
1152 * | ---- desired range ---- |
1154 * We need to split the extent, and set the bit
1157 if (state
->start
<= end
&& state
->end
> end
) {
1158 prealloc
= alloc_extent_state_atomic(prealloc
);
1164 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1166 extent_io_tree_panic(tree
, err
);
1168 set_state_bits(tree
, prealloc
, &bits
);
1169 cache_state(prealloc
, cached_state
);
1170 clear_state_bit(tree
, prealloc
, &clear_bits
, 0);
1178 spin_unlock(&tree
->lock
);
1180 free_extent_state(prealloc
);
1187 spin_unlock(&tree
->lock
);
1188 if (mask
& __GFP_WAIT
)
1193 /* wrappers around set/clear extent bit */
1194 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1197 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, NULL
,
1201 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1202 unsigned long bits
, gfp_t mask
)
1204 return set_extent_bit(tree
, start
, end
, bits
, NULL
,
1208 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1209 unsigned long bits
, gfp_t mask
)
1211 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
1214 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1215 struct extent_state
**cached_state
, gfp_t mask
)
1217 return set_extent_bit(tree
, start
, end
,
1218 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1219 NULL
, cached_state
, mask
);
1222 int set_extent_defrag(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1223 struct extent_state
**cached_state
, gfp_t mask
)
1225 return set_extent_bit(tree
, start
, end
,
1226 EXTENT_DELALLOC
| EXTENT_UPTODATE
| EXTENT_DEFRAG
,
1227 NULL
, cached_state
, mask
);
1230 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1233 return clear_extent_bit(tree
, start
, end
,
1234 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1235 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1238 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1241 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, NULL
,
1245 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1246 struct extent_state
**cached_state
, gfp_t mask
)
1248 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, NULL
,
1249 cached_state
, mask
);
1252 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1253 struct extent_state
**cached_state
, gfp_t mask
)
1255 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1256 cached_state
, mask
);
1260 * either insert or lock state struct between start and end use mask to tell
1261 * us if waiting is desired.
1263 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1264 unsigned long bits
, struct extent_state
**cached_state
)
1269 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1270 EXTENT_LOCKED
, &failed_start
,
1271 cached_state
, GFP_NOFS
);
1272 if (err
== -EEXIST
) {
1273 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1274 start
= failed_start
;
1277 WARN_ON(start
> end
);
1282 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1284 return lock_extent_bits(tree
, start
, end
, 0, NULL
);
1287 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1292 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1293 &failed_start
, NULL
, GFP_NOFS
);
1294 if (err
== -EEXIST
) {
1295 if (failed_start
> start
)
1296 clear_extent_bit(tree
, start
, failed_start
- 1,
1297 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1303 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1304 struct extent_state
**cached
, gfp_t mask
)
1306 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1310 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1312 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1316 int extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1318 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1319 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1322 while (index
<= end_index
) {
1323 page
= find_get_page(inode
->i_mapping
, index
);
1324 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1325 clear_page_dirty_for_io(page
);
1326 page_cache_release(page
);
1332 int extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1334 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1335 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1338 while (index
<= end_index
) {
1339 page
= find_get_page(inode
->i_mapping
, index
);
1340 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1341 account_page_redirty(page
);
1342 __set_page_dirty_nobuffers(page
);
1343 page_cache_release(page
);
1350 * helper function to set both pages and extents in the tree writeback
1352 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1354 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1355 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1358 while (index
<= end_index
) {
1359 page
= find_get_page(tree
->mapping
, index
);
1360 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1361 set_page_writeback(page
);
1362 page_cache_release(page
);
1368 /* find the first state struct with 'bits' set after 'start', and
1369 * return it. tree->lock must be held. NULL will returned if
1370 * nothing was found after 'start'
1372 static struct extent_state
*
1373 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1374 u64 start
, unsigned long bits
)
1376 struct rb_node
*node
;
1377 struct extent_state
*state
;
1380 * this search will find all the extents that end after
1383 node
= tree_search(tree
, start
);
1388 state
= rb_entry(node
, struct extent_state
, rb_node
);
1389 if (state
->end
>= start
&& (state
->state
& bits
))
1392 node
= rb_next(node
);
1401 * find the first offset in the io tree with 'bits' set. zero is
1402 * returned if we find something, and *start_ret and *end_ret are
1403 * set to reflect the state struct that was found.
1405 * If nothing was found, 1 is returned. If found something, return 0.
1407 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1408 u64
*start_ret
, u64
*end_ret
, unsigned long bits
,
1409 struct extent_state
**cached_state
)
1411 struct extent_state
*state
;
1415 spin_lock(&tree
->lock
);
1416 if (cached_state
&& *cached_state
) {
1417 state
= *cached_state
;
1418 if (state
->end
== start
- 1 && state
->tree
) {
1419 n
= rb_next(&state
->rb_node
);
1421 state
= rb_entry(n
, struct extent_state
,
1423 if (state
->state
& bits
)
1427 free_extent_state(*cached_state
);
1428 *cached_state
= NULL
;
1431 free_extent_state(*cached_state
);
1432 *cached_state
= NULL
;
1435 state
= find_first_extent_bit_state(tree
, start
, bits
);
1438 cache_state(state
, cached_state
);
1439 *start_ret
= state
->start
;
1440 *end_ret
= state
->end
;
1444 spin_unlock(&tree
->lock
);
1449 * find a contiguous range of bytes in the file marked as delalloc, not
1450 * more than 'max_bytes'. start and end are used to return the range,
1452 * 1 is returned if we find something, 0 if nothing was in the tree
1454 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1455 u64
*start
, u64
*end
, u64 max_bytes
,
1456 struct extent_state
**cached_state
)
1458 struct rb_node
*node
;
1459 struct extent_state
*state
;
1460 u64 cur_start
= *start
;
1462 u64 total_bytes
= 0;
1464 spin_lock(&tree
->lock
);
1467 * this search will find all the extents that end after
1470 node
= tree_search(tree
, cur_start
);
1478 state
= rb_entry(node
, struct extent_state
, rb_node
);
1479 if (found
&& (state
->start
!= cur_start
||
1480 (state
->state
& EXTENT_BOUNDARY
))) {
1483 if (!(state
->state
& EXTENT_DELALLOC
)) {
1489 *start
= state
->start
;
1490 *cached_state
= state
;
1491 atomic_inc(&state
->refs
);
1495 cur_start
= state
->end
+ 1;
1496 node
= rb_next(node
);
1499 total_bytes
+= state
->end
- state
->start
+ 1;
1500 if (total_bytes
>= max_bytes
)
1504 spin_unlock(&tree
->lock
);
1508 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1509 struct page
*locked_page
,
1513 struct page
*pages
[16];
1514 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1515 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1516 unsigned long nr_pages
= end_index
- index
+ 1;
1519 if (index
== locked_page
->index
&& end_index
== index
)
1522 while (nr_pages
> 0) {
1523 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1524 min_t(unsigned long, nr_pages
,
1525 ARRAY_SIZE(pages
)), pages
);
1526 for (i
= 0; i
< ret
; i
++) {
1527 if (pages
[i
] != locked_page
)
1528 unlock_page(pages
[i
]);
1529 page_cache_release(pages
[i
]);
1537 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1538 struct page
*locked_page
,
1542 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1543 unsigned long start_index
= index
;
1544 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1545 unsigned long pages_locked
= 0;
1546 struct page
*pages
[16];
1547 unsigned long nrpages
;
1551 /* the caller is responsible for locking the start index */
1552 if (index
== locked_page
->index
&& index
== end_index
)
1555 /* skip the page at the start index */
1556 nrpages
= end_index
- index
+ 1;
1557 while (nrpages
> 0) {
1558 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1559 min_t(unsigned long,
1560 nrpages
, ARRAY_SIZE(pages
)), pages
);
1565 /* now we have an array of pages, lock them all */
1566 for (i
= 0; i
< ret
; i
++) {
1568 * the caller is taking responsibility for
1571 if (pages
[i
] != locked_page
) {
1572 lock_page(pages
[i
]);
1573 if (!PageDirty(pages
[i
]) ||
1574 pages
[i
]->mapping
!= inode
->i_mapping
) {
1576 unlock_page(pages
[i
]);
1577 page_cache_release(pages
[i
]);
1581 page_cache_release(pages
[i
]);
1590 if (ret
&& pages_locked
) {
1591 __unlock_for_delalloc(inode
, locked_page
,
1593 ((u64
)(start_index
+ pages_locked
- 1)) <<
1600 * find a contiguous range of bytes in the file marked as delalloc, not
1601 * more than 'max_bytes'. start and end are used to return the range,
1603 * 1 is returned if we find something, 0 if nothing was in the tree
1605 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1606 struct extent_io_tree
*tree
,
1607 struct page
*locked_page
,
1608 u64
*start
, u64
*end
,
1614 struct extent_state
*cached_state
= NULL
;
1619 /* step one, find a bunch of delalloc bytes starting at start */
1620 delalloc_start
= *start
;
1622 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1623 max_bytes
, &cached_state
);
1624 if (!found
|| delalloc_end
<= *start
) {
1625 *start
= delalloc_start
;
1626 *end
= delalloc_end
;
1627 free_extent_state(cached_state
);
1632 * start comes from the offset of locked_page. We have to lock
1633 * pages in order, so we can't process delalloc bytes before
1636 if (delalloc_start
< *start
)
1637 delalloc_start
= *start
;
1640 * make sure to limit the number of pages we try to lock down
1643 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
)
1644 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1646 /* step two, lock all the pages after the page that has start */
1647 ret
= lock_delalloc_pages(inode
, locked_page
,
1648 delalloc_start
, delalloc_end
);
1649 if (ret
== -EAGAIN
) {
1650 /* some of the pages are gone, lets avoid looping by
1651 * shortening the size of the delalloc range we're searching
1653 free_extent_state(cached_state
);
1655 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1656 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1664 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1666 /* step three, lock the state bits for the whole range */
1667 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1669 /* then test to make sure it is all still delalloc */
1670 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1671 EXTENT_DELALLOC
, 1, cached_state
);
1673 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1674 &cached_state
, GFP_NOFS
);
1675 __unlock_for_delalloc(inode
, locked_page
,
1676 delalloc_start
, delalloc_end
);
1680 free_extent_state(cached_state
);
1681 *start
= delalloc_start
;
1682 *end
= delalloc_end
;
1687 int extent_clear_unlock_delalloc(struct inode
*inode
,
1688 struct extent_io_tree
*tree
,
1689 u64 start
, u64 end
, struct page
*locked_page
,
1693 struct page
*pages
[16];
1694 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1695 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1696 unsigned long nr_pages
= end_index
- index
+ 1;
1698 unsigned long clear_bits
= 0;
1700 if (op
& EXTENT_CLEAR_UNLOCK
)
1701 clear_bits
|= EXTENT_LOCKED
;
1702 if (op
& EXTENT_CLEAR_DIRTY
)
1703 clear_bits
|= EXTENT_DIRTY
;
1705 if (op
& EXTENT_CLEAR_DELALLOC
)
1706 clear_bits
|= EXTENT_DELALLOC
;
1708 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1709 if (!(op
& (EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
1710 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
|
1711 EXTENT_SET_PRIVATE2
)))
1714 while (nr_pages
> 0) {
1715 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1716 min_t(unsigned long,
1717 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1718 for (i
= 0; i
< ret
; i
++) {
1720 if (op
& EXTENT_SET_PRIVATE2
)
1721 SetPagePrivate2(pages
[i
]);
1723 if (pages
[i
] == locked_page
) {
1724 page_cache_release(pages
[i
]);
1727 if (op
& EXTENT_CLEAR_DIRTY
)
1728 clear_page_dirty_for_io(pages
[i
]);
1729 if (op
& EXTENT_SET_WRITEBACK
)
1730 set_page_writeback(pages
[i
]);
1731 if (op
& EXTENT_END_WRITEBACK
)
1732 end_page_writeback(pages
[i
]);
1733 if (op
& EXTENT_CLEAR_UNLOCK_PAGE
)
1734 unlock_page(pages
[i
]);
1735 page_cache_release(pages
[i
]);
1745 * count the number of bytes in the tree that have a given bit(s)
1746 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1747 * cached. The total number found is returned.
1749 u64
count_range_bits(struct extent_io_tree
*tree
,
1750 u64
*start
, u64 search_end
, u64 max_bytes
,
1751 unsigned long bits
, int contig
)
1753 struct rb_node
*node
;
1754 struct extent_state
*state
;
1755 u64 cur_start
= *start
;
1756 u64 total_bytes
= 0;
1760 if (search_end
<= cur_start
) {
1765 spin_lock(&tree
->lock
);
1766 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1767 total_bytes
= tree
->dirty_bytes
;
1771 * this search will find all the extents that end after
1774 node
= tree_search(tree
, cur_start
);
1779 state
= rb_entry(node
, struct extent_state
, rb_node
);
1780 if (state
->start
> search_end
)
1782 if (contig
&& found
&& state
->start
> last
+ 1)
1784 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1785 total_bytes
+= min(search_end
, state
->end
) + 1 -
1786 max(cur_start
, state
->start
);
1787 if (total_bytes
>= max_bytes
)
1790 *start
= max(cur_start
, state
->start
);
1794 } else if (contig
&& found
) {
1797 node
= rb_next(node
);
1802 spin_unlock(&tree
->lock
);
1807 * set the private field for a given byte offset in the tree. If there isn't
1808 * an extent_state there already, this does nothing.
1810 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1812 struct rb_node
*node
;
1813 struct extent_state
*state
;
1816 spin_lock(&tree
->lock
);
1818 * this search will find all the extents that end after
1821 node
= tree_search(tree
, start
);
1826 state
= rb_entry(node
, struct extent_state
, rb_node
);
1827 if (state
->start
!= start
) {
1831 state
->private = private;
1833 spin_unlock(&tree
->lock
);
1837 void extent_cache_csums_dio(struct extent_io_tree
*tree
, u64 start
, u32 csums
[],
1840 struct rb_node
*node
;
1841 struct extent_state
*state
;
1843 spin_lock(&tree
->lock
);
1845 * this search will find all the extents that end after
1848 node
= tree_search(tree
, start
);
1851 state
= rb_entry(node
, struct extent_state
, rb_node
);
1852 BUG_ON(state
->start
!= start
);
1855 state
->private = *csums
++;
1857 state
= next_state(state
);
1859 spin_unlock(&tree
->lock
);
1862 static inline u64
__btrfs_get_bio_offset(struct bio
*bio
, int bio_index
)
1864 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio_index
;
1866 return page_offset(bvec
->bv_page
) + bvec
->bv_offset
;
1869 void extent_cache_csums(struct extent_io_tree
*tree
, struct bio
*bio
, int bio_index
,
1870 u32 csums
[], int count
)
1872 struct rb_node
*node
;
1873 struct extent_state
*state
= NULL
;
1876 spin_lock(&tree
->lock
);
1878 start
= __btrfs_get_bio_offset(bio
, bio_index
);
1879 if (state
== NULL
|| state
->start
!= start
) {
1880 node
= tree_search(tree
, start
);
1883 state
= rb_entry(node
, struct extent_state
, rb_node
);
1884 BUG_ON(state
->start
!= start
);
1886 state
->private = *csums
++;
1890 state
= next_state(state
);
1892 spin_unlock(&tree
->lock
);
1895 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1897 struct rb_node
*node
;
1898 struct extent_state
*state
;
1901 spin_lock(&tree
->lock
);
1903 * this search will find all the extents that end after
1906 node
= tree_search(tree
, start
);
1911 state
= rb_entry(node
, struct extent_state
, rb_node
);
1912 if (state
->start
!= start
) {
1916 *private = state
->private;
1918 spin_unlock(&tree
->lock
);
1923 * searches a range in the state tree for a given mask.
1924 * If 'filled' == 1, this returns 1 only if every extent in the tree
1925 * has the bits set. Otherwise, 1 is returned if any bit in the
1926 * range is found set.
1928 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1929 unsigned long bits
, int filled
, struct extent_state
*cached
)
1931 struct extent_state
*state
= NULL
;
1932 struct rb_node
*node
;
1935 spin_lock(&tree
->lock
);
1936 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1937 cached
->end
> start
)
1938 node
= &cached
->rb_node
;
1940 node
= tree_search(tree
, start
);
1941 while (node
&& start
<= end
) {
1942 state
= rb_entry(node
, struct extent_state
, rb_node
);
1944 if (filled
&& state
->start
> start
) {
1949 if (state
->start
> end
)
1952 if (state
->state
& bits
) {
1956 } else if (filled
) {
1961 if (state
->end
== (u64
)-1)
1964 start
= state
->end
+ 1;
1967 node
= rb_next(node
);
1974 spin_unlock(&tree
->lock
);
1979 * helper function to set a given page up to date if all the
1980 * extents in the tree for that page are up to date
1982 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1984 u64 start
= page_offset(page
);
1985 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1986 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1987 SetPageUptodate(page
);
1991 * When IO fails, either with EIO or csum verification fails, we
1992 * try other mirrors that might have a good copy of the data. This
1993 * io_failure_record is used to record state as we go through all the
1994 * mirrors. If another mirror has good data, the page is set up to date
1995 * and things continue. If a good mirror can't be found, the original
1996 * bio end_io callback is called to indicate things have failed.
1998 struct io_failure_record
{
2003 unsigned long bio_flags
;
2009 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
2014 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2016 set_state_private(failure_tree
, rec
->start
, 0);
2017 ret
= clear_extent_bits(failure_tree
, rec
->start
,
2018 rec
->start
+ rec
->len
- 1,
2019 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2023 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
2024 rec
->start
+ rec
->len
- 1,
2025 EXTENT_DAMAGED
, GFP_NOFS
);
2033 static void repair_io_failure_callback(struct bio
*bio
, int err
)
2035 complete(bio
->bi_private
);
2039 * this bypasses the standard btrfs submit functions deliberately, as
2040 * the standard behavior is to write all copies in a raid setup. here we only
2041 * want to write the one bad copy. so we do the mapping for ourselves and issue
2042 * submit_bio directly.
2043 * to avoid any synchronization issues, wait for the data after writing, which
2044 * actually prevents the read that triggered the error from finishing.
2045 * currently, there can be no more than two copies of every data bit. thus,
2046 * exactly one rewrite is required.
2048 int repair_io_failure(struct btrfs_fs_info
*fs_info
, u64 start
,
2049 u64 length
, u64 logical
, struct page
*page
,
2053 struct btrfs_device
*dev
;
2054 DECLARE_COMPLETION_ONSTACK(compl);
2057 struct btrfs_bio
*bbio
= NULL
;
2058 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2061 BUG_ON(!mirror_num
);
2063 /* we can't repair anything in raid56 yet */
2064 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2067 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2070 bio
->bi_private
= &compl;
2071 bio
->bi_end_io
= repair_io_failure_callback
;
2073 map_length
= length
;
2075 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2076 &map_length
, &bbio
, mirror_num
);
2081 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2082 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2083 bio
->bi_sector
= sector
;
2084 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2086 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2090 bio
->bi_bdev
= dev
->bdev
;
2091 bio_add_page(bio
, page
, length
, start
- page_offset(page
));
2092 btrfsic_submit_bio(WRITE_SYNC
, bio
);
2093 wait_for_completion(&compl);
2095 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
2096 /* try to remap that extent elsewhere? */
2098 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2102 printk_ratelimited_in_rcu(KERN_INFO
"btrfs read error corrected: ino %lu off %llu "
2103 "(dev %s sector %llu)\n", page
->mapping
->host
->i_ino
,
2104 start
, rcu_str_deref(dev
->name
), sector
);
2110 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2113 u64 start
= eb
->start
;
2114 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2117 for (i
= 0; i
< num_pages
; i
++) {
2118 struct page
*p
= extent_buffer_page(eb
, i
);
2119 ret
= repair_io_failure(root
->fs_info
, start
, PAGE_CACHE_SIZE
,
2120 start
, p
, mirror_num
);
2123 start
+= PAGE_CACHE_SIZE
;
2130 * each time an IO finishes, we do a fast check in the IO failure tree
2131 * to see if we need to process or clean up an io_failure_record
2133 static int clean_io_failure(u64 start
, struct page
*page
)
2136 u64 private_failure
;
2137 struct io_failure_record
*failrec
;
2138 struct btrfs_fs_info
*fs_info
;
2139 struct extent_state
*state
;
2143 struct inode
*inode
= page
->mapping
->host
;
2146 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2147 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2151 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2156 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2157 BUG_ON(!failrec
->this_mirror
);
2159 if (failrec
->in_validation
) {
2160 /* there was no real error, just free the record */
2161 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2167 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2168 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2171 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2173 if (state
&& state
->start
== failrec
->start
) {
2174 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2175 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2177 if (num_copies
> 1) {
2178 ret
= repair_io_failure(fs_info
, start
, failrec
->len
,
2179 failrec
->logical
, page
,
2180 failrec
->failed_mirror
);
2188 ret
= free_io_failure(inode
, failrec
, did_repair
);
2194 * this is a generic handler for readpage errors (default
2195 * readpage_io_failed_hook). if other copies exist, read those and write back
2196 * good data to the failed position. does not investigate in remapping the
2197 * failed extent elsewhere, hoping the device will be smart enough to do this as
2201 static int bio_readpage_error(struct bio
*failed_bio
, struct page
*page
,
2202 u64 start
, u64 end
, int failed_mirror
,
2203 struct extent_state
*state
)
2205 struct io_failure_record
*failrec
= NULL
;
2207 struct extent_map
*em
;
2208 struct inode
*inode
= page
->mapping
->host
;
2209 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2210 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2211 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2218 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2220 ret
= get_state_private(failure_tree
, start
, &private);
2222 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2225 failrec
->start
= start
;
2226 failrec
->len
= end
- start
+ 1;
2227 failrec
->this_mirror
= 0;
2228 failrec
->bio_flags
= 0;
2229 failrec
->in_validation
= 0;
2231 read_lock(&em_tree
->lock
);
2232 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2234 read_unlock(&em_tree
->lock
);
2239 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
2240 free_extent_map(em
);
2243 read_unlock(&em_tree
->lock
);
2249 logical
= start
- em
->start
;
2250 logical
= em
->block_start
+ logical
;
2251 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2252 logical
= em
->block_start
;
2253 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2254 extent_set_compress_type(&failrec
->bio_flags
,
2257 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2258 "len=%llu\n", logical
, start
, failrec
->len
);
2259 failrec
->logical
= logical
;
2260 free_extent_map(em
);
2262 /* set the bits in the private failure tree */
2263 ret
= set_extent_bits(failure_tree
, start
, end
,
2264 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2266 ret
= set_state_private(failure_tree
, start
,
2267 (u64
)(unsigned long)failrec
);
2268 /* set the bits in the inode's tree */
2270 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2277 failrec
= (struct io_failure_record
*)(unsigned long)private;
2278 pr_debug("bio_readpage_error: (found) logical=%llu, "
2279 "start=%llu, len=%llu, validation=%d\n",
2280 failrec
->logical
, failrec
->start
, failrec
->len
,
2281 failrec
->in_validation
);
2283 * when data can be on disk more than twice, add to failrec here
2284 * (e.g. with a list for failed_mirror) to make
2285 * clean_io_failure() clean all those errors at once.
2288 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2289 failrec
->logical
, failrec
->len
);
2290 if (num_copies
== 1) {
2292 * we only have a single copy of the data, so don't bother with
2293 * all the retry and error correction code that follows. no
2294 * matter what the error is, it is very likely to persist.
2296 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2297 "state=%p, num_copies=%d, next_mirror %d, "
2298 "failed_mirror %d\n", state
, num_copies
,
2299 failrec
->this_mirror
, failed_mirror
);
2300 free_io_failure(inode
, failrec
, 0);
2305 spin_lock(&tree
->lock
);
2306 state
= find_first_extent_bit_state(tree
, failrec
->start
,
2308 if (state
&& state
->start
!= failrec
->start
)
2310 spin_unlock(&tree
->lock
);
2314 * there are two premises:
2315 * a) deliver good data to the caller
2316 * b) correct the bad sectors on disk
2318 if (failed_bio
->bi_vcnt
> 1) {
2320 * to fulfill b), we need to know the exact failing sectors, as
2321 * we don't want to rewrite any more than the failed ones. thus,
2322 * we need separate read requests for the failed bio
2324 * if the following BUG_ON triggers, our validation request got
2325 * merged. we need separate requests for our algorithm to work.
2327 BUG_ON(failrec
->in_validation
);
2328 failrec
->in_validation
= 1;
2329 failrec
->this_mirror
= failed_mirror
;
2330 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2333 * we're ready to fulfill a) and b) alongside. get a good copy
2334 * of the failed sector and if we succeed, we have setup
2335 * everything for repair_io_failure to do the rest for us.
2337 if (failrec
->in_validation
) {
2338 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2339 failrec
->in_validation
= 0;
2340 failrec
->this_mirror
= 0;
2342 failrec
->failed_mirror
= failed_mirror
;
2343 failrec
->this_mirror
++;
2344 if (failrec
->this_mirror
== failed_mirror
)
2345 failrec
->this_mirror
++;
2346 read_mode
= READ_SYNC
;
2349 if (!state
|| failrec
->this_mirror
> num_copies
) {
2350 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2351 "next_mirror %d, failed_mirror %d\n", state
,
2352 num_copies
, failrec
->this_mirror
, failed_mirror
);
2353 free_io_failure(inode
, failrec
, 0);
2357 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2359 free_io_failure(inode
, failrec
, 0);
2362 bio
->bi_private
= state
;
2363 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2364 bio
->bi_sector
= failrec
->logical
>> 9;
2365 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2368 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2370 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2371 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2372 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2374 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2375 failrec
->this_mirror
,
2376 failrec
->bio_flags
, 0);
2380 /* lots and lots of room for performance fixes in the end_bio funcs */
2382 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2384 int uptodate
= (err
== 0);
2385 struct extent_io_tree
*tree
;
2388 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2390 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2391 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2392 end
, NULL
, uptodate
);
2398 ClearPageUptodate(page
);
2405 * after a writepage IO is done, we need to:
2406 * clear the uptodate bits on error
2407 * clear the writeback bits in the extent tree for this IO
2408 * end_page_writeback if the page has no more pending IO
2410 * Scheduling is not allowed, so the extent state tree is expected
2411 * to have one and only one object corresponding to this IO.
2413 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2415 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2416 struct extent_io_tree
*tree
;
2421 struct page
*page
= bvec
->bv_page
;
2422 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2424 /* We always issue full-page reads, but if some block
2425 * in a page fails to read, blk_update_request() will
2426 * advance bv_offset and adjust bv_len to compensate.
2427 * Print a warning for nonzero offsets, and an error
2428 * if they don't add up to a full page. */
2429 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2430 printk("%s page write in btrfs with offset %u and length %u\n",
2431 bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
2432 ? KERN_ERR
"partial" : KERN_INFO
"incomplete",
2433 bvec
->bv_offset
, bvec
->bv_len
);
2435 start
= page_offset(page
);
2436 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2438 if (--bvec
>= bio
->bi_io_vec
)
2439 prefetchw(&bvec
->bv_page
->flags
);
2441 if (end_extent_writepage(page
, err
, start
, end
))
2444 end_page_writeback(page
);
2445 } while (bvec
>= bio
->bi_io_vec
);
2451 * after a readpage IO is done, we need to:
2452 * clear the uptodate bits on error
2453 * set the uptodate bits if things worked
2454 * set the page up to date if all extents in the tree are uptodate
2455 * clear the lock bit in the extent tree
2456 * unlock the page if there are no other extents locked for it
2458 * Scheduling is not allowed, so the extent state tree is expected
2459 * to have one and only one object corresponding to this IO.
2461 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2463 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2464 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2465 struct bio_vec
*bvec
= bio
->bi_io_vec
;
2466 struct extent_io_tree
*tree
;
2476 struct page
*page
= bvec
->bv_page
;
2477 struct extent_state
*cached
= NULL
;
2478 struct extent_state
*state
;
2479 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2480 struct inode
*inode
= page
->mapping
->host
;
2482 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2483 "mirror=%lu\n", (u64
)bio
->bi_sector
, err
,
2484 io_bio
->mirror_num
);
2485 tree
= &BTRFS_I(inode
)->io_tree
;
2487 /* We always issue full-page reads, but if some block
2488 * in a page fails to read, blk_update_request() will
2489 * advance bv_offset and adjust bv_len to compensate.
2490 * Print a warning for nonzero offsets, and an error
2491 * if they don't add up to a full page. */
2492 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2493 printk("%s page read in btrfs with offset %u and length %u\n",
2494 bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
2495 ? KERN_ERR
"partial" : KERN_INFO
"incomplete",
2496 bvec
->bv_offset
, bvec
->bv_len
);
2498 start
= page_offset(page
);
2499 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2501 if (++bvec
<= bvec_end
)
2502 prefetchw(&bvec
->bv_page
->flags
);
2504 spin_lock(&tree
->lock
);
2505 state
= find_first_extent_bit_state(tree
, start
, EXTENT_LOCKED
);
2506 if (state
&& state
->start
== start
) {
2508 * take a reference on the state, unlock will drop
2511 cache_state(state
, &cached
);
2513 spin_unlock(&tree
->lock
);
2515 mirror
= io_bio
->mirror_num
;
2516 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
2517 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
2522 clean_io_failure(start
, page
);
2525 if (!uptodate
&& tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2526 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2528 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2530 } else if (!uptodate
) {
2532 * The generic bio_readpage_error handles errors the
2533 * following way: If possible, new read requests are
2534 * created and submitted and will end up in
2535 * end_bio_extent_readpage as well (if we're lucky, not
2536 * in the !uptodate case). In that case it returns 0 and
2537 * we just go on with the next page in our bio. If it
2538 * can't handle the error it will return -EIO and we
2539 * remain responsible for that page.
2541 ret
= bio_readpage_error(bio
, page
, start
, end
, mirror
, NULL
);
2544 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2547 uncache_state(&cached
);
2552 if (uptodate
&& tree
->track_uptodate
) {
2553 set_extent_uptodate(tree
, start
, end
, &cached
,
2556 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2559 loff_t i_size
= i_size_read(inode
);
2560 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2563 /* Zero out the end if this page straddles i_size */
2564 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2565 if (page
->index
== end_index
&& offset
)
2566 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2567 SetPageUptodate(page
);
2569 ClearPageUptodate(page
);
2573 } while (bvec
<= bvec_end
);
2579 * this allocates from the btrfs_bioset. We're returning a bio right now
2580 * but you can call btrfs_io_bio for the appropriate container_of magic
2583 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2588 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2590 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2591 while (!bio
&& (nr_vecs
/= 2)) {
2592 bio
= bio_alloc_bioset(gfp_flags
,
2593 nr_vecs
, btrfs_bioset
);
2599 bio
->bi_bdev
= bdev
;
2600 bio
->bi_sector
= first_sector
;
2605 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2607 return bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2611 /* this also allocates from the btrfs_bioset */
2612 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2614 return bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2618 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2619 int mirror_num
, unsigned long bio_flags
)
2622 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2623 struct page
*page
= bvec
->bv_page
;
2624 struct extent_io_tree
*tree
= bio
->bi_private
;
2627 start
= page_offset(page
) + bvec
->bv_offset
;
2629 bio
->bi_private
= NULL
;
2633 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2634 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2635 mirror_num
, bio_flags
, start
);
2637 btrfsic_submit_bio(rw
, bio
);
2639 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2645 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2646 unsigned long offset
, size_t size
, struct bio
*bio
,
2647 unsigned long bio_flags
)
2650 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2651 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2658 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2659 struct page
*page
, sector_t sector
,
2660 size_t size
, unsigned long offset
,
2661 struct block_device
*bdev
,
2662 struct bio
**bio_ret
,
2663 unsigned long max_pages
,
2664 bio_end_io_t end_io_func
,
2666 unsigned long prev_bio_flags
,
2667 unsigned long bio_flags
)
2673 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2674 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2675 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2677 if (bio_ret
&& *bio_ret
) {
2680 contig
= bio
->bi_sector
== sector
;
2682 contig
= bio
->bi_sector
+ (bio
->bi_size
>> 9) ==
2685 if (prev_bio_flags
!= bio_flags
|| !contig
||
2686 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2687 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2688 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2697 if (this_compressed
)
2700 nr
= bio_get_nr_vecs(bdev
);
2702 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2706 bio_add_page(bio
, page
, page_size
, offset
);
2707 bio
->bi_end_io
= end_io_func
;
2708 bio
->bi_private
= tree
;
2713 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2718 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2721 if (!PagePrivate(page
)) {
2722 SetPagePrivate(page
);
2723 page_cache_get(page
);
2724 set_page_private(page
, (unsigned long)eb
);
2726 WARN_ON(page
->private != (unsigned long)eb
);
2730 void set_page_extent_mapped(struct page
*page
)
2732 if (!PagePrivate(page
)) {
2733 SetPagePrivate(page
);
2734 page_cache_get(page
);
2735 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2740 * basic readpage implementation. Locked extent state structs are inserted
2741 * into the tree that are removed when the IO is done (by the end_io
2743 * XXX JDM: This needs looking at to ensure proper page locking
2745 static int __extent_read_full_page(struct extent_io_tree
*tree
,
2747 get_extent_t
*get_extent
,
2748 struct bio
**bio
, int mirror_num
,
2749 unsigned long *bio_flags
, int rw
)
2751 struct inode
*inode
= page
->mapping
->host
;
2752 u64 start
= page_offset(page
);
2753 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2757 u64 last_byte
= i_size_read(inode
);
2761 struct extent_map
*em
;
2762 struct block_device
*bdev
;
2763 struct btrfs_ordered_extent
*ordered
;
2766 size_t pg_offset
= 0;
2768 size_t disk_io_size
;
2769 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2770 unsigned long this_bio_flag
= 0;
2772 set_page_extent_mapped(page
);
2774 if (!PageUptodate(page
)) {
2775 if (cleancache_get_page(page
) == 0) {
2776 BUG_ON(blocksize
!= PAGE_SIZE
);
2783 lock_extent(tree
, start
, end
);
2784 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
2787 unlock_extent(tree
, start
, end
);
2788 btrfs_start_ordered_extent(inode
, ordered
, 1);
2789 btrfs_put_ordered_extent(ordered
);
2792 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2794 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2797 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2798 userpage
= kmap_atomic(page
);
2799 memset(userpage
+ zero_offset
, 0, iosize
);
2800 flush_dcache_page(page
);
2801 kunmap_atomic(userpage
);
2804 while (cur
<= end
) {
2805 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2807 if (cur
>= last_byte
) {
2809 struct extent_state
*cached
= NULL
;
2811 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2812 userpage
= kmap_atomic(page
);
2813 memset(userpage
+ pg_offset
, 0, iosize
);
2814 flush_dcache_page(page
);
2815 kunmap_atomic(userpage
);
2816 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2818 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2822 em
= get_extent(inode
, page
, pg_offset
, cur
,
2824 if (IS_ERR_OR_NULL(em
)) {
2826 unlock_extent(tree
, cur
, end
);
2829 extent_offset
= cur
- em
->start
;
2830 BUG_ON(extent_map_end(em
) <= cur
);
2833 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2834 this_bio_flag
= EXTENT_BIO_COMPRESSED
;
2835 extent_set_compress_type(&this_bio_flag
,
2839 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2840 cur_end
= min(extent_map_end(em
) - 1, end
);
2841 iosize
= ALIGN(iosize
, blocksize
);
2842 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2843 disk_io_size
= em
->block_len
;
2844 sector
= em
->block_start
>> 9;
2846 sector
= (em
->block_start
+ extent_offset
) >> 9;
2847 disk_io_size
= iosize
;
2850 block_start
= em
->block_start
;
2851 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2852 block_start
= EXTENT_MAP_HOLE
;
2853 free_extent_map(em
);
2856 /* we've found a hole, just zero and go on */
2857 if (block_start
== EXTENT_MAP_HOLE
) {
2859 struct extent_state
*cached
= NULL
;
2861 userpage
= kmap_atomic(page
);
2862 memset(userpage
+ pg_offset
, 0, iosize
);
2863 flush_dcache_page(page
);
2864 kunmap_atomic(userpage
);
2866 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2868 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2871 pg_offset
+= iosize
;
2874 /* the get_extent function already copied into the page */
2875 if (test_range_bit(tree
, cur
, cur_end
,
2876 EXTENT_UPTODATE
, 1, NULL
)) {
2877 check_page_uptodate(tree
, page
);
2878 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2880 pg_offset
+= iosize
;
2883 /* we have an inline extent but it didn't get marked up
2884 * to date. Error out
2886 if (block_start
== EXTENT_MAP_INLINE
) {
2888 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2890 pg_offset
+= iosize
;
2895 ret
= submit_extent_page(rw
, tree
, page
,
2896 sector
, disk_io_size
, pg_offset
,
2898 end_bio_extent_readpage
, mirror_num
,
2903 *bio_flags
= this_bio_flag
;
2906 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2909 pg_offset
+= iosize
;
2913 if (!PageError(page
))
2914 SetPageUptodate(page
);
2920 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2921 get_extent_t
*get_extent
, int mirror_num
)
2923 struct bio
*bio
= NULL
;
2924 unsigned long bio_flags
= 0;
2927 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
2930 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
2934 static noinline
void update_nr_written(struct page
*page
,
2935 struct writeback_control
*wbc
,
2936 unsigned long nr_written
)
2938 wbc
->nr_to_write
-= nr_written
;
2939 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
2940 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
2941 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
2945 * the writepage semantics are similar to regular writepage. extent
2946 * records are inserted to lock ranges in the tree, and as dirty areas
2947 * are found, they are marked writeback. Then the lock bits are removed
2948 * and the end_io handler clears the writeback ranges
2950 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
2953 struct inode
*inode
= page
->mapping
->host
;
2954 struct extent_page_data
*epd
= data
;
2955 struct extent_io_tree
*tree
= epd
->tree
;
2956 u64 start
= page_offset(page
);
2958 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2962 u64 last_byte
= i_size_read(inode
);
2966 struct extent_state
*cached_state
= NULL
;
2967 struct extent_map
*em
;
2968 struct block_device
*bdev
;
2971 size_t pg_offset
= 0;
2973 loff_t i_size
= i_size_read(inode
);
2974 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2980 unsigned long nr_written
= 0;
2981 bool fill_delalloc
= true;
2983 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2984 write_flags
= WRITE_SYNC
;
2986 write_flags
= WRITE
;
2988 trace___extent_writepage(page
, inode
, wbc
);
2990 WARN_ON(!PageLocked(page
));
2992 ClearPageError(page
);
2994 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
2995 if (page
->index
> end_index
||
2996 (page
->index
== end_index
&& !pg_offset
)) {
2997 page
->mapping
->a_ops
->invalidatepage(page
, 0);
3002 if (page
->index
== end_index
) {
3005 userpage
= kmap_atomic(page
);
3006 memset(userpage
+ pg_offset
, 0,
3007 PAGE_CACHE_SIZE
- pg_offset
);
3008 kunmap_atomic(userpage
);
3009 flush_dcache_page(page
);
3013 set_page_extent_mapped(page
);
3015 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
3016 fill_delalloc
= false;
3018 delalloc_start
= start
;
3021 if (!epd
->extent_locked
&& fill_delalloc
) {
3022 u64 delalloc_to_write
= 0;
3024 * make sure the wbc mapping index is at least updated
3027 update_nr_written(page
, wbc
, 0);
3029 while (delalloc_end
< page_end
) {
3030 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3035 if (nr_delalloc
== 0) {
3036 delalloc_start
= delalloc_end
+ 1;
3039 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3044 /* File system has been set read-only */
3050 * delalloc_end is already one less than the total
3051 * length, so we don't subtract one from
3054 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3057 delalloc_start
= delalloc_end
+ 1;
3059 if (wbc
->nr_to_write
< delalloc_to_write
) {
3062 if (delalloc_to_write
< thresh
* 2)
3063 thresh
= delalloc_to_write
;
3064 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3068 /* did the fill delalloc function already unlock and start
3074 * we've unlocked the page, so we can't update
3075 * the mapping's writeback index, just update
3078 wbc
->nr_to_write
-= nr_written
;
3082 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3083 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3086 /* Fixup worker will requeue */
3088 wbc
->pages_skipped
++;
3090 redirty_page_for_writepage(wbc
, page
);
3091 update_nr_written(page
, wbc
, nr_written
);
3099 * we don't want to touch the inode after unlocking the page,
3100 * so we update the mapping writeback index now
3102 update_nr_written(page
, wbc
, nr_written
+ 1);
3105 if (last_byte
<= start
) {
3106 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3107 tree
->ops
->writepage_end_io_hook(page
, start
,
3112 blocksize
= inode
->i_sb
->s_blocksize
;
3114 while (cur
<= end
) {
3115 if (cur
>= last_byte
) {
3116 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3117 tree
->ops
->writepage_end_io_hook(page
, cur
,
3121 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3123 if (IS_ERR_OR_NULL(em
)) {
3128 extent_offset
= cur
- em
->start
;
3129 BUG_ON(extent_map_end(em
) <= cur
);
3131 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
3132 iosize
= ALIGN(iosize
, blocksize
);
3133 sector
= (em
->block_start
+ extent_offset
) >> 9;
3135 block_start
= em
->block_start
;
3136 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3137 free_extent_map(em
);
3141 * compressed and inline extents are written through other
3144 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3145 block_start
== EXTENT_MAP_INLINE
) {
3147 * end_io notification does not happen here for
3148 * compressed extents
3150 if (!compressed
&& tree
->ops
&&
3151 tree
->ops
->writepage_end_io_hook
)
3152 tree
->ops
->writepage_end_io_hook(page
, cur
,
3155 else if (compressed
) {
3156 /* we don't want to end_page_writeback on
3157 * a compressed extent. this happens
3164 pg_offset
+= iosize
;
3167 /* leave this out until we have a page_mkwrite call */
3168 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
3169 EXTENT_DIRTY
, 0, NULL
)) {
3171 pg_offset
+= iosize
;
3175 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3176 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3184 unsigned long max_nr
= end_index
+ 1;
3186 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3187 if (!PageWriteback(page
)) {
3188 printk(KERN_ERR
"btrfs warning page %lu not "
3189 "writeback, cur %llu end %llu\n",
3190 page
->index
, (unsigned long long)cur
,
3191 (unsigned long long)end
);
3194 ret
= submit_extent_page(write_flags
, tree
, page
,
3195 sector
, iosize
, pg_offset
,
3196 bdev
, &epd
->bio
, max_nr
,
3197 end_bio_extent_writepage
,
3203 pg_offset
+= iosize
;
3208 /* make sure the mapping tag for page dirty gets cleared */
3209 set_page_writeback(page
);
3210 end_page_writeback(page
);
3216 /* drop our reference on any cached states */
3217 free_extent_state(cached_state
);
3221 static int eb_wait(void *word
)
3227 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3229 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3230 TASK_UNINTERRUPTIBLE
);
3233 static int lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3234 struct btrfs_fs_info
*fs_info
,
3235 struct extent_page_data
*epd
)
3237 unsigned long i
, num_pages
;
3241 if (!btrfs_try_tree_write_lock(eb
)) {
3243 flush_write_bio(epd
);
3244 btrfs_tree_lock(eb
);
3247 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3248 btrfs_tree_unlock(eb
);
3252 flush_write_bio(epd
);
3256 wait_on_extent_buffer_writeback(eb
);
3257 btrfs_tree_lock(eb
);
3258 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3260 btrfs_tree_unlock(eb
);
3265 * We need to do this to prevent races in people who check if the eb is
3266 * under IO since we can end up having no IO bits set for a short period
3269 spin_lock(&eb
->refs_lock
);
3270 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3271 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3272 spin_unlock(&eb
->refs_lock
);
3273 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3274 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3276 fs_info
->dirty_metadata_batch
);
3279 spin_unlock(&eb
->refs_lock
);
3282 btrfs_tree_unlock(eb
);
3287 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3288 for (i
= 0; i
< num_pages
; i
++) {
3289 struct page
*p
= extent_buffer_page(eb
, i
);
3291 if (!trylock_page(p
)) {
3293 flush_write_bio(epd
);
3303 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3305 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3306 smp_mb__after_clear_bit();
3307 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3310 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3312 int uptodate
= err
== 0;
3313 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
3314 struct extent_buffer
*eb
;
3318 struct page
*page
= bvec
->bv_page
;
3321 eb
= (struct extent_buffer
*)page
->private;
3323 done
= atomic_dec_and_test(&eb
->io_pages
);
3325 if (!uptodate
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3326 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3327 ClearPageUptodate(page
);
3331 end_page_writeback(page
);
3336 end_extent_buffer_writeback(eb
);
3337 } while (bvec
>= bio
->bi_io_vec
);
3343 static int write_one_eb(struct extent_buffer
*eb
,
3344 struct btrfs_fs_info
*fs_info
,
3345 struct writeback_control
*wbc
,
3346 struct extent_page_data
*epd
)
3348 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3349 u64 offset
= eb
->start
;
3350 unsigned long i
, num_pages
;
3351 unsigned long bio_flags
= 0;
3352 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3355 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3356 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3357 atomic_set(&eb
->io_pages
, num_pages
);
3358 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3359 bio_flags
= EXTENT_BIO_TREE_LOG
;
3361 for (i
= 0; i
< num_pages
; i
++) {
3362 struct page
*p
= extent_buffer_page(eb
, i
);
3364 clear_page_dirty_for_io(p
);
3365 set_page_writeback(p
);
3366 ret
= submit_extent_page(rw
, eb
->tree
, p
, offset
>> 9,
3367 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3368 -1, end_bio_extent_buffer_writepage
,
3369 0, epd
->bio_flags
, bio_flags
);
3370 epd
->bio_flags
= bio_flags
;
3372 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3374 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3375 end_extent_buffer_writeback(eb
);
3379 offset
+= PAGE_CACHE_SIZE
;
3380 update_nr_written(p
, wbc
, 1);
3384 if (unlikely(ret
)) {
3385 for (; i
< num_pages
; i
++) {
3386 struct page
*p
= extent_buffer_page(eb
, i
);
3394 int btree_write_cache_pages(struct address_space
*mapping
,
3395 struct writeback_control
*wbc
)
3397 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3398 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3399 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3400 struct extent_page_data epd
= {
3404 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3409 int nr_to_write_done
= 0;
3410 struct pagevec pvec
;
3413 pgoff_t end
; /* Inclusive */
3417 pagevec_init(&pvec
, 0);
3418 if (wbc
->range_cyclic
) {
3419 index
= mapping
->writeback_index
; /* Start from prev offset */
3422 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3423 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3426 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3427 tag
= PAGECACHE_TAG_TOWRITE
;
3429 tag
= PAGECACHE_TAG_DIRTY
;
3431 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3432 tag_pages_for_writeback(mapping
, index
, end
);
3433 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3434 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3435 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3439 for (i
= 0; i
< nr_pages
; i
++) {
3440 struct page
*page
= pvec
.pages
[i
];
3442 if (!PagePrivate(page
))
3445 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3450 spin_lock(&mapping
->private_lock
);
3451 if (!PagePrivate(page
)) {
3452 spin_unlock(&mapping
->private_lock
);
3456 eb
= (struct extent_buffer
*)page
->private;
3459 * Shouldn't happen and normally this would be a BUG_ON
3460 * but no sense in crashing the users box for something
3461 * we can survive anyway.
3464 spin_unlock(&mapping
->private_lock
);
3469 if (eb
== prev_eb
) {
3470 spin_unlock(&mapping
->private_lock
);
3474 ret
= atomic_inc_not_zero(&eb
->refs
);
3475 spin_unlock(&mapping
->private_lock
);
3480 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3482 free_extent_buffer(eb
);
3486 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3489 free_extent_buffer(eb
);
3492 free_extent_buffer(eb
);
3495 * the filesystem may choose to bump up nr_to_write.
3496 * We have to make sure to honor the new nr_to_write
3499 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3501 pagevec_release(&pvec
);
3504 if (!scanned
&& !done
) {
3506 * We hit the last page and there is more work to be done: wrap
3507 * back to the start of the file
3513 flush_write_bio(&epd
);
3518 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3519 * @mapping: address space structure to write
3520 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3521 * @writepage: function called for each page
3522 * @data: data passed to writepage function
3524 * If a page is already under I/O, write_cache_pages() skips it, even
3525 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3526 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3527 * and msync() need to guarantee that all the data which was dirty at the time
3528 * the call was made get new I/O started against them. If wbc->sync_mode is
3529 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3530 * existing IO to complete.
3532 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3533 struct address_space
*mapping
,
3534 struct writeback_control
*wbc
,
3535 writepage_t writepage
, void *data
,
3536 void (*flush_fn
)(void *))
3538 struct inode
*inode
= mapping
->host
;
3541 int nr_to_write_done
= 0;
3542 struct pagevec pvec
;
3545 pgoff_t end
; /* Inclusive */
3550 * We have to hold onto the inode so that ordered extents can do their
3551 * work when the IO finishes. The alternative to this is failing to add
3552 * an ordered extent if the igrab() fails there and that is a huge pain
3553 * to deal with, so instead just hold onto the inode throughout the
3554 * writepages operation. If it fails here we are freeing up the inode
3555 * anyway and we'd rather not waste our time writing out stuff that is
3556 * going to be truncated anyway.
3561 pagevec_init(&pvec
, 0);
3562 if (wbc
->range_cyclic
) {
3563 index
= mapping
->writeback_index
; /* Start from prev offset */
3566 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3567 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3570 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3571 tag
= PAGECACHE_TAG_TOWRITE
;
3573 tag
= PAGECACHE_TAG_DIRTY
;
3575 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3576 tag_pages_for_writeback(mapping
, index
, end
);
3577 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3578 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3579 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3583 for (i
= 0; i
< nr_pages
; i
++) {
3584 struct page
*page
= pvec
.pages
[i
];
3587 * At this point we hold neither mapping->tree_lock nor
3588 * lock on the page itself: the page may be truncated or
3589 * invalidated (changing page->mapping to NULL), or even
3590 * swizzled back from swapper_space to tmpfs file
3593 if (!trylock_page(page
)) {
3598 if (unlikely(page
->mapping
!= mapping
)) {
3603 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3609 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3610 if (PageWriteback(page
))
3612 wait_on_page_writeback(page
);
3615 if (PageWriteback(page
) ||
3616 !clear_page_dirty_for_io(page
)) {
3621 ret
= (*writepage
)(page
, wbc
, data
);
3623 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3631 * the filesystem may choose to bump up nr_to_write.
3632 * We have to make sure to honor the new nr_to_write
3635 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3637 pagevec_release(&pvec
);
3640 if (!scanned
&& !done
) {
3642 * We hit the last page and there is more work to be done: wrap
3643 * back to the start of the file
3649 btrfs_add_delayed_iput(inode
);
3653 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3662 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
3663 BUG_ON(ret
< 0); /* -ENOMEM */
3668 static noinline
void flush_write_bio(void *data
)
3670 struct extent_page_data
*epd
= data
;
3671 flush_epd_write_bio(epd
);
3674 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3675 get_extent_t
*get_extent
,
3676 struct writeback_control
*wbc
)
3679 struct extent_page_data epd
= {
3682 .get_extent
= get_extent
,
3684 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3688 ret
= __extent_writepage(page
, wbc
, &epd
);
3690 flush_epd_write_bio(&epd
);
3694 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3695 u64 start
, u64 end
, get_extent_t
*get_extent
,
3699 struct address_space
*mapping
= inode
->i_mapping
;
3701 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3704 struct extent_page_data epd
= {
3707 .get_extent
= get_extent
,
3709 .sync_io
= mode
== WB_SYNC_ALL
,
3712 struct writeback_control wbc_writepages
= {
3714 .nr_to_write
= nr_pages
* 2,
3715 .range_start
= start
,
3716 .range_end
= end
+ 1,
3719 while (start
<= end
) {
3720 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3721 if (clear_page_dirty_for_io(page
))
3722 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3724 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3725 tree
->ops
->writepage_end_io_hook(page
, start
,
3726 start
+ PAGE_CACHE_SIZE
- 1,
3730 page_cache_release(page
);
3731 start
+= PAGE_CACHE_SIZE
;
3734 flush_epd_write_bio(&epd
);
3738 int extent_writepages(struct extent_io_tree
*tree
,
3739 struct address_space
*mapping
,
3740 get_extent_t
*get_extent
,
3741 struct writeback_control
*wbc
)
3744 struct extent_page_data epd
= {
3747 .get_extent
= get_extent
,
3749 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3753 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3754 __extent_writepage
, &epd
,
3756 flush_epd_write_bio(&epd
);
3760 int extent_readpages(struct extent_io_tree
*tree
,
3761 struct address_space
*mapping
,
3762 struct list_head
*pages
, unsigned nr_pages
,
3763 get_extent_t get_extent
)
3765 struct bio
*bio
= NULL
;
3767 unsigned long bio_flags
= 0;
3768 struct page
*pagepool
[16];
3773 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3774 page
= list_entry(pages
->prev
, struct page
, lru
);
3776 prefetchw(&page
->flags
);
3777 list_del(&page
->lru
);
3778 if (add_to_page_cache_lru(page
, mapping
,
3779 page
->index
, GFP_NOFS
)) {
3780 page_cache_release(page
);
3784 pagepool
[nr
++] = page
;
3785 if (nr
< ARRAY_SIZE(pagepool
))
3787 for (i
= 0; i
< nr
; i
++) {
3788 __extent_read_full_page(tree
, pagepool
[i
], get_extent
,
3789 &bio
, 0, &bio_flags
, READ
);
3790 page_cache_release(pagepool
[i
]);
3794 for (i
= 0; i
< nr
; i
++) {
3795 __extent_read_full_page(tree
, pagepool
[i
], get_extent
,
3796 &bio
, 0, &bio_flags
, READ
);
3797 page_cache_release(pagepool
[i
]);
3800 BUG_ON(!list_empty(pages
));
3802 return submit_one_bio(READ
, bio
, 0, bio_flags
);
3807 * basic invalidatepage code, this waits on any locked or writeback
3808 * ranges corresponding to the page, and then deletes any extent state
3809 * records from the tree
3811 int extent_invalidatepage(struct extent_io_tree
*tree
,
3812 struct page
*page
, unsigned long offset
)
3814 struct extent_state
*cached_state
= NULL
;
3815 u64 start
= page_offset(page
);
3816 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3817 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
3819 start
+= ALIGN(offset
, blocksize
);
3823 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
3824 wait_on_page_writeback(page
);
3825 clear_extent_bit(tree
, start
, end
,
3826 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3827 EXTENT_DO_ACCOUNTING
,
3828 1, 1, &cached_state
, GFP_NOFS
);
3833 * a helper for releasepage, this tests for areas of the page that
3834 * are locked or under IO and drops the related state bits if it is safe
3837 static int try_release_extent_state(struct extent_map_tree
*map
,
3838 struct extent_io_tree
*tree
,
3839 struct page
*page
, gfp_t mask
)
3841 u64 start
= page_offset(page
);
3842 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3845 if (test_range_bit(tree
, start
, end
,
3846 EXTENT_IOBITS
, 0, NULL
))
3849 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
3852 * at this point we can safely clear everything except the
3853 * locked bit and the nodatasum bit
3855 ret
= clear_extent_bit(tree
, start
, end
,
3856 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
3859 /* if clear_extent_bit failed for enomem reasons,
3860 * we can't allow the release to continue.
3871 * a helper for releasepage. As long as there are no locked extents
3872 * in the range corresponding to the page, both state records and extent
3873 * map records are removed
3875 int try_release_extent_mapping(struct extent_map_tree
*map
,
3876 struct extent_io_tree
*tree
, struct page
*page
,
3879 struct extent_map
*em
;
3880 u64 start
= page_offset(page
);
3881 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3883 if ((mask
& __GFP_WAIT
) &&
3884 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
3886 while (start
<= end
) {
3887 len
= end
- start
+ 1;
3888 write_lock(&map
->lock
);
3889 em
= lookup_extent_mapping(map
, start
, len
);
3891 write_unlock(&map
->lock
);
3894 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
3895 em
->start
!= start
) {
3896 write_unlock(&map
->lock
);
3897 free_extent_map(em
);
3900 if (!test_range_bit(tree
, em
->start
,
3901 extent_map_end(em
) - 1,
3902 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
3904 remove_extent_mapping(map
, em
);
3905 /* once for the rb tree */
3906 free_extent_map(em
);
3908 start
= extent_map_end(em
);
3909 write_unlock(&map
->lock
);
3912 free_extent_map(em
);
3915 return try_release_extent_state(map
, tree
, page
, mask
);
3919 * helper function for fiemap, which doesn't want to see any holes.
3920 * This maps until we find something past 'last'
3922 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
3925 get_extent_t
*get_extent
)
3927 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
3928 struct extent_map
*em
;
3935 len
= last
- offset
;
3938 len
= ALIGN(len
, sectorsize
);
3939 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
3940 if (IS_ERR_OR_NULL(em
))
3943 /* if this isn't a hole return it */
3944 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
3945 em
->block_start
!= EXTENT_MAP_HOLE
) {
3949 /* this is a hole, advance to the next extent */
3950 offset
= extent_map_end(em
);
3951 free_extent_map(em
);
3958 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
3959 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
3963 u64 max
= start
+ len
;
3967 u64 last_for_get_extent
= 0;
3969 u64 isize
= i_size_read(inode
);
3970 struct btrfs_key found_key
;
3971 struct extent_map
*em
= NULL
;
3972 struct extent_state
*cached_state
= NULL
;
3973 struct btrfs_path
*path
;
3974 struct btrfs_file_extent_item
*item
;
3979 unsigned long emflags
;
3984 path
= btrfs_alloc_path();
3987 path
->leave_spinning
= 1;
3989 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
3990 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
3993 * lookup the last file extent. We're not using i_size here
3994 * because there might be preallocation past i_size
3996 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
3997 path
, btrfs_ino(inode
), -1, 0);
3999 btrfs_free_path(path
);
4004 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4005 struct btrfs_file_extent_item
);
4006 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4007 found_type
= btrfs_key_type(&found_key
);
4009 /* No extents, but there might be delalloc bits */
4010 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4011 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4012 /* have to trust i_size as the end */
4014 last_for_get_extent
= isize
;
4017 * remember the start of the last extent. There are a
4018 * bunch of different factors that go into the length of the
4019 * extent, so its much less complex to remember where it started
4021 last
= found_key
.offset
;
4022 last_for_get_extent
= last
+ 1;
4024 btrfs_free_path(path
);
4027 * we might have some extents allocated but more delalloc past those
4028 * extents. so, we trust isize unless the start of the last extent is
4033 last_for_get_extent
= isize
;
4036 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1, 0,
4039 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4049 u64 offset_in_extent
;
4051 /* break if the extent we found is outside the range */
4052 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4056 * get_extent may return an extent that starts before our
4057 * requested range. We have to make sure the ranges
4058 * we return to fiemap always move forward and don't
4059 * overlap, so adjust the offsets here
4061 em_start
= max(em
->start
, off
);
4064 * record the offset from the start of the extent
4065 * for adjusting the disk offset below
4067 offset_in_extent
= em_start
- em
->start
;
4068 em_end
= extent_map_end(em
);
4069 em_len
= em_end
- em_start
;
4070 emflags
= em
->flags
;
4075 * bump off for our next call to get_extent
4077 off
= extent_map_end(em
);
4081 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4083 flags
|= FIEMAP_EXTENT_LAST
;
4084 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4085 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4086 FIEMAP_EXTENT_NOT_ALIGNED
);
4087 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4088 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4089 FIEMAP_EXTENT_UNKNOWN
);
4091 disko
= em
->block_start
+ offset_in_extent
;
4093 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4094 flags
|= FIEMAP_EXTENT_ENCODED
;
4096 free_extent_map(em
);
4098 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4099 (last
== (u64
)-1 && isize
<= em_end
)) {
4100 flags
|= FIEMAP_EXTENT_LAST
;
4104 /* now scan forward to see if this is really the last extent. */
4105 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4112 flags
|= FIEMAP_EXTENT_LAST
;
4115 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4121 free_extent_map(em
);
4123 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4124 &cached_state
, GFP_NOFS
);
4128 static void __free_extent_buffer(struct extent_buffer
*eb
)
4130 btrfs_leak_debug_del(&eb
->leak_list
);
4131 kmem_cache_free(extent_buffer_cache
, eb
);
4134 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
4139 struct extent_buffer
*eb
= NULL
;
4141 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
4148 rwlock_init(&eb
->lock
);
4149 atomic_set(&eb
->write_locks
, 0);
4150 atomic_set(&eb
->read_locks
, 0);
4151 atomic_set(&eb
->blocking_readers
, 0);
4152 atomic_set(&eb
->blocking_writers
, 0);
4153 atomic_set(&eb
->spinning_readers
, 0);
4154 atomic_set(&eb
->spinning_writers
, 0);
4155 eb
->lock_nested
= 0;
4156 init_waitqueue_head(&eb
->write_lock_wq
);
4157 init_waitqueue_head(&eb
->read_lock_wq
);
4159 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4161 spin_lock_init(&eb
->refs_lock
);
4162 atomic_set(&eb
->refs
, 1);
4163 atomic_set(&eb
->io_pages
, 0);
4166 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4168 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4169 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4170 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4175 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4179 struct extent_buffer
*new;
4180 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4182 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_ATOMIC
);
4186 for (i
= 0; i
< num_pages
; i
++) {
4187 p
= alloc_page(GFP_ATOMIC
);
4189 attach_extent_buffer_page(new, p
);
4190 WARN_ON(PageDirty(p
));
4195 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4196 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4197 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4202 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
4204 struct extent_buffer
*eb
;
4205 unsigned long num_pages
= num_extent_pages(0, len
);
4208 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_ATOMIC
);
4212 for (i
= 0; i
< num_pages
; i
++) {
4213 eb
->pages
[i
] = alloc_page(GFP_ATOMIC
);
4217 set_extent_buffer_uptodate(eb
);
4218 btrfs_set_header_nritems(eb
, 0);
4219 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4224 __free_page(eb
->pages
[i
- 1]);
4225 __free_extent_buffer(eb
);
4229 static int extent_buffer_under_io(struct extent_buffer
*eb
)
4231 return (atomic_read(&eb
->io_pages
) ||
4232 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4233 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4237 * Helper for releasing extent buffer page.
4239 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4240 unsigned long start_idx
)
4242 unsigned long index
;
4243 unsigned long num_pages
;
4245 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4247 BUG_ON(extent_buffer_under_io(eb
));
4249 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4250 index
= start_idx
+ num_pages
;
4251 if (start_idx
>= index
)
4256 page
= extent_buffer_page(eb
, index
);
4257 if (page
&& mapped
) {
4258 spin_lock(&page
->mapping
->private_lock
);
4260 * We do this since we'll remove the pages after we've
4261 * removed the eb from the radix tree, so we could race
4262 * and have this page now attached to the new eb. So
4263 * only clear page_private if it's still connected to
4266 if (PagePrivate(page
) &&
4267 page
->private == (unsigned long)eb
) {
4268 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4269 BUG_ON(PageDirty(page
));
4270 BUG_ON(PageWriteback(page
));
4272 * We need to make sure we haven't be attached
4275 ClearPagePrivate(page
);
4276 set_page_private(page
, 0);
4277 /* One for the page private */
4278 page_cache_release(page
);
4280 spin_unlock(&page
->mapping
->private_lock
);
4284 /* One for when we alloced the page */
4285 page_cache_release(page
);
4287 } while (index
!= start_idx
);
4291 * Helper for releasing the extent buffer.
4293 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4295 btrfs_release_extent_buffer_page(eb
, 0);
4296 __free_extent_buffer(eb
);
4299 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4302 /* the ref bit is tricky. We have to make sure it is set
4303 * if we have the buffer dirty. Otherwise the
4304 * code to free a buffer can end up dropping a dirty
4307 * Once the ref bit is set, it won't go away while the
4308 * buffer is dirty or in writeback, and it also won't
4309 * go away while we have the reference count on the
4312 * We can't just set the ref bit without bumping the
4313 * ref on the eb because free_extent_buffer might
4314 * see the ref bit and try to clear it. If this happens
4315 * free_extent_buffer might end up dropping our original
4316 * ref by mistake and freeing the page before we are able
4317 * to add one more ref.
4319 * So bump the ref count first, then set the bit. If someone
4320 * beat us to it, drop the ref we added.
4322 refs
= atomic_read(&eb
->refs
);
4323 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4326 spin_lock(&eb
->refs_lock
);
4327 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4328 atomic_inc(&eb
->refs
);
4329 spin_unlock(&eb
->refs_lock
);
4332 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
)
4334 unsigned long num_pages
, i
;
4336 check_buffer_tree_ref(eb
);
4338 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4339 for (i
= 0; i
< num_pages
; i
++) {
4340 struct page
*p
= extent_buffer_page(eb
, i
);
4341 mark_page_accessed(p
);
4345 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
4346 u64 start
, unsigned long len
)
4348 unsigned long num_pages
= num_extent_pages(start
, len
);
4350 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4351 struct extent_buffer
*eb
;
4352 struct extent_buffer
*exists
= NULL
;
4354 struct address_space
*mapping
= tree
->mapping
;
4359 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4360 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4362 mark_extent_buffer_accessed(eb
);
4367 eb
= __alloc_extent_buffer(tree
, start
, len
, GFP_NOFS
);
4371 for (i
= 0; i
< num_pages
; i
++, index
++) {
4372 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4376 spin_lock(&mapping
->private_lock
);
4377 if (PagePrivate(p
)) {
4379 * We could have already allocated an eb for this page
4380 * and attached one so lets see if we can get a ref on
4381 * the existing eb, and if we can we know it's good and
4382 * we can just return that one, else we know we can just
4383 * overwrite page->private.
4385 exists
= (struct extent_buffer
*)p
->private;
4386 if (atomic_inc_not_zero(&exists
->refs
)) {
4387 spin_unlock(&mapping
->private_lock
);
4389 page_cache_release(p
);
4390 mark_extent_buffer_accessed(exists
);
4395 * Do this so attach doesn't complain and we need to
4396 * drop the ref the old guy had.
4398 ClearPagePrivate(p
);
4399 WARN_ON(PageDirty(p
));
4400 page_cache_release(p
);
4402 attach_extent_buffer_page(eb
, p
);
4403 spin_unlock(&mapping
->private_lock
);
4404 WARN_ON(PageDirty(p
));
4405 mark_page_accessed(p
);
4407 if (!PageUptodate(p
))
4411 * see below about how we avoid a nasty race with release page
4412 * and why we unlock later
4416 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4418 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4422 spin_lock(&tree
->buffer_lock
);
4423 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
4424 if (ret
== -EEXIST
) {
4425 exists
= radix_tree_lookup(&tree
->buffer
,
4426 start
>> PAGE_CACHE_SHIFT
);
4427 if (!atomic_inc_not_zero(&exists
->refs
)) {
4428 spin_unlock(&tree
->buffer_lock
);
4429 radix_tree_preload_end();
4433 spin_unlock(&tree
->buffer_lock
);
4434 radix_tree_preload_end();
4435 mark_extent_buffer_accessed(exists
);
4438 /* add one reference for the tree */
4439 check_buffer_tree_ref(eb
);
4440 spin_unlock(&tree
->buffer_lock
);
4441 radix_tree_preload_end();
4444 * there is a race where release page may have
4445 * tried to find this extent buffer in the radix
4446 * but failed. It will tell the VM it is safe to
4447 * reclaim the, and it will clear the page private bit.
4448 * We must make sure to set the page private bit properly
4449 * after the extent buffer is in the radix tree so
4450 * it doesn't get lost
4452 SetPageChecked(eb
->pages
[0]);
4453 for (i
= 1; i
< num_pages
; i
++) {
4454 p
= extent_buffer_page(eb
, i
);
4455 ClearPageChecked(p
);
4458 unlock_page(eb
->pages
[0]);
4462 for (i
= 0; i
< num_pages
; i
++) {
4464 unlock_page(eb
->pages
[i
]);
4467 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4468 btrfs_release_extent_buffer(eb
);
4472 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
4473 u64 start
, unsigned long len
)
4475 struct extent_buffer
*eb
;
4478 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4479 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4481 mark_extent_buffer_accessed(eb
);
4489 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4491 struct extent_buffer
*eb
=
4492 container_of(head
, struct extent_buffer
, rcu_head
);
4494 __free_extent_buffer(eb
);
4497 /* Expects to have eb->eb_lock already held */
4498 static int release_extent_buffer(struct extent_buffer
*eb
)
4500 WARN_ON(atomic_read(&eb
->refs
) == 0);
4501 if (atomic_dec_and_test(&eb
->refs
)) {
4502 if (test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
)) {
4503 spin_unlock(&eb
->refs_lock
);
4505 struct extent_io_tree
*tree
= eb
->tree
;
4507 spin_unlock(&eb
->refs_lock
);
4509 spin_lock(&tree
->buffer_lock
);
4510 radix_tree_delete(&tree
->buffer
,
4511 eb
->start
>> PAGE_CACHE_SHIFT
);
4512 spin_unlock(&tree
->buffer_lock
);
4515 /* Should be safe to release our pages at this point */
4516 btrfs_release_extent_buffer_page(eb
, 0);
4517 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4520 spin_unlock(&eb
->refs_lock
);
4525 void free_extent_buffer(struct extent_buffer
*eb
)
4533 refs
= atomic_read(&eb
->refs
);
4536 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
4541 spin_lock(&eb
->refs_lock
);
4542 if (atomic_read(&eb
->refs
) == 2 &&
4543 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4544 atomic_dec(&eb
->refs
);
4546 if (atomic_read(&eb
->refs
) == 2 &&
4547 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4548 !extent_buffer_under_io(eb
) &&
4549 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4550 atomic_dec(&eb
->refs
);
4553 * I know this is terrible, but it's temporary until we stop tracking
4554 * the uptodate bits and such for the extent buffers.
4556 release_extent_buffer(eb
);
4559 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4564 spin_lock(&eb
->refs_lock
);
4565 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4567 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4568 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4569 atomic_dec(&eb
->refs
);
4570 release_extent_buffer(eb
);
4573 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4576 unsigned long num_pages
;
4579 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4581 for (i
= 0; i
< num_pages
; i
++) {
4582 page
= extent_buffer_page(eb
, i
);
4583 if (!PageDirty(page
))
4587 WARN_ON(!PagePrivate(page
));
4589 clear_page_dirty_for_io(page
);
4590 spin_lock_irq(&page
->mapping
->tree_lock
);
4591 if (!PageDirty(page
)) {
4592 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4594 PAGECACHE_TAG_DIRTY
);
4596 spin_unlock_irq(&page
->mapping
->tree_lock
);
4597 ClearPageError(page
);
4600 WARN_ON(atomic_read(&eb
->refs
) == 0);
4603 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4606 unsigned long num_pages
;
4609 check_buffer_tree_ref(eb
);
4611 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4613 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4614 WARN_ON(atomic_read(&eb
->refs
) == 0);
4615 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4617 for (i
= 0; i
< num_pages
; i
++)
4618 set_page_dirty(extent_buffer_page(eb
, i
));
4622 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4626 unsigned long num_pages
;
4628 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4629 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4630 for (i
= 0; i
< num_pages
; i
++) {
4631 page
= extent_buffer_page(eb
, i
);
4633 ClearPageUptodate(page
);
4638 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4642 unsigned long num_pages
;
4644 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4645 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4646 for (i
= 0; i
< num_pages
; i
++) {
4647 page
= extent_buffer_page(eb
, i
);
4648 SetPageUptodate(page
);
4653 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4655 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4658 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4659 struct extent_buffer
*eb
, u64 start
, int wait
,
4660 get_extent_t
*get_extent
, int mirror_num
)
4663 unsigned long start_i
;
4667 int locked_pages
= 0;
4668 int all_uptodate
= 1;
4669 unsigned long num_pages
;
4670 unsigned long num_reads
= 0;
4671 struct bio
*bio
= NULL
;
4672 unsigned long bio_flags
= 0;
4674 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4678 WARN_ON(start
< eb
->start
);
4679 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4680 (eb
->start
>> PAGE_CACHE_SHIFT
);
4685 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4686 for (i
= start_i
; i
< num_pages
; i
++) {
4687 page
= extent_buffer_page(eb
, i
);
4688 if (wait
== WAIT_NONE
) {
4689 if (!trylock_page(page
))
4695 if (!PageUptodate(page
)) {
4702 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4706 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4707 eb
->read_mirror
= 0;
4708 atomic_set(&eb
->io_pages
, num_reads
);
4709 for (i
= start_i
; i
< num_pages
; i
++) {
4710 page
= extent_buffer_page(eb
, i
);
4711 if (!PageUptodate(page
)) {
4712 ClearPageError(page
);
4713 err
= __extent_read_full_page(tree
, page
,
4715 mirror_num
, &bio_flags
,
4725 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
4731 if (ret
|| wait
!= WAIT_COMPLETE
)
4734 for (i
= start_i
; i
< num_pages
; i
++) {
4735 page
= extent_buffer_page(eb
, i
);
4736 wait_on_page_locked(page
);
4737 if (!PageUptodate(page
))
4745 while (locked_pages
> 0) {
4746 page
= extent_buffer_page(eb
, i
);
4754 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
4755 unsigned long start
,
4762 char *dst
= (char *)dstv
;
4763 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4764 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4766 WARN_ON(start
> eb
->len
);
4767 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4769 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4772 page
= extent_buffer_page(eb
, i
);
4774 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4775 kaddr
= page_address(page
);
4776 memcpy(dst
, kaddr
+ offset
, cur
);
4785 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
4786 unsigned long min_len
, char **map
,
4787 unsigned long *map_start
,
4788 unsigned long *map_len
)
4790 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
4793 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4794 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4795 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
4802 offset
= start_offset
;
4806 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
4809 if (start
+ min_len
> eb
->len
) {
4810 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
4811 "wanted %lu %lu\n", (unsigned long long)eb
->start
,
4812 eb
->len
, start
, min_len
);
4816 p
= extent_buffer_page(eb
, i
);
4817 kaddr
= page_address(p
);
4818 *map
= kaddr
+ offset
;
4819 *map_len
= PAGE_CACHE_SIZE
- offset
;
4823 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
4824 unsigned long start
,
4831 char *ptr
= (char *)ptrv
;
4832 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4833 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4836 WARN_ON(start
> eb
->len
);
4837 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4839 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4842 page
= extent_buffer_page(eb
, i
);
4844 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4846 kaddr
= page_address(page
);
4847 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
4859 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
4860 unsigned long start
, unsigned long len
)
4866 char *src
= (char *)srcv
;
4867 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4868 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4870 WARN_ON(start
> eb
->len
);
4871 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4873 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4876 page
= extent_buffer_page(eb
, i
);
4877 WARN_ON(!PageUptodate(page
));
4879 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4880 kaddr
= page_address(page
);
4881 memcpy(kaddr
+ offset
, src
, cur
);
4890 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
4891 unsigned long start
, unsigned long len
)
4897 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4898 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4900 WARN_ON(start
> eb
->len
);
4901 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4903 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4906 page
= extent_buffer_page(eb
, i
);
4907 WARN_ON(!PageUptodate(page
));
4909 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4910 kaddr
= page_address(page
);
4911 memset(kaddr
+ offset
, c
, cur
);
4919 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
4920 unsigned long dst_offset
, unsigned long src_offset
,
4923 u64 dst_len
= dst
->len
;
4928 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4929 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
4931 WARN_ON(src
->len
!= dst_len
);
4933 offset
= (start_offset
+ dst_offset
) &
4934 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4937 page
= extent_buffer_page(dst
, i
);
4938 WARN_ON(!PageUptodate(page
));
4940 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
4942 kaddr
= page_address(page
);
4943 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
4952 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
4953 unsigned long dst_off
, unsigned long src_off
,
4956 char *dst_kaddr
= page_address(dst_page
);
4957 if (dst_page
== src_page
) {
4958 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
4960 char *src_kaddr
= page_address(src_page
);
4961 char *p
= dst_kaddr
+ dst_off
+ len
;
4962 char *s
= src_kaddr
+ src_off
+ len
;
4969 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
4971 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
4972 return distance
< len
;
4975 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
4976 unsigned long dst_off
, unsigned long src_off
,
4979 char *dst_kaddr
= page_address(dst_page
);
4981 int must_memmove
= 0;
4983 if (dst_page
!= src_page
) {
4984 src_kaddr
= page_address(src_page
);
4986 src_kaddr
= dst_kaddr
;
4987 if (areas_overlap(src_off
, dst_off
, len
))
4992 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
4994 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
4997 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
4998 unsigned long src_offset
, unsigned long len
)
5001 size_t dst_off_in_page
;
5002 size_t src_off_in_page
;
5003 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5004 unsigned long dst_i
;
5005 unsigned long src_i
;
5007 if (src_offset
+ len
> dst
->len
) {
5008 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
5009 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
5012 if (dst_offset
+ len
> dst
->len
) {
5013 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
5014 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
5019 dst_off_in_page
= (start_offset
+ dst_offset
) &
5020 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5021 src_off_in_page
= (start_offset
+ src_offset
) &
5022 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5024 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5025 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5027 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5029 cur
= min_t(unsigned long, cur
,
5030 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5032 copy_pages(extent_buffer_page(dst
, dst_i
),
5033 extent_buffer_page(dst
, src_i
),
5034 dst_off_in_page
, src_off_in_page
, cur
);
5042 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5043 unsigned long src_offset
, unsigned long len
)
5046 size_t dst_off_in_page
;
5047 size_t src_off_in_page
;
5048 unsigned long dst_end
= dst_offset
+ len
- 1;
5049 unsigned long src_end
= src_offset
+ len
- 1;
5050 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5051 unsigned long dst_i
;
5052 unsigned long src_i
;
5054 if (src_offset
+ len
> dst
->len
) {
5055 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
5056 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
5059 if (dst_offset
+ len
> dst
->len
) {
5060 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
5061 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
5064 if (dst_offset
< src_offset
) {
5065 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5069 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5070 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5072 dst_off_in_page
= (start_offset
+ dst_end
) &
5073 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5074 src_off_in_page
= (start_offset
+ src_end
) &
5075 ((unsigned long)PAGE_CACHE_SIZE
- 1);
5077 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5078 cur
= min(cur
, dst_off_in_page
+ 1);
5079 move_pages(extent_buffer_page(dst
, dst_i
),
5080 extent_buffer_page(dst
, src_i
),
5081 dst_off_in_page
- cur
+ 1,
5082 src_off_in_page
- cur
+ 1, cur
);
5090 int try_release_extent_buffer(struct page
*page
)
5092 struct extent_buffer
*eb
;
5095 * We need to make sure noboody is attaching this page to an eb right
5098 spin_lock(&page
->mapping
->private_lock
);
5099 if (!PagePrivate(page
)) {
5100 spin_unlock(&page
->mapping
->private_lock
);
5104 eb
= (struct extent_buffer
*)page
->private;
5108 * This is a little awful but should be ok, we need to make sure that
5109 * the eb doesn't disappear out from under us while we're looking at
5112 spin_lock(&eb
->refs_lock
);
5113 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5114 spin_unlock(&eb
->refs_lock
);
5115 spin_unlock(&page
->mapping
->private_lock
);
5118 spin_unlock(&page
->mapping
->private_lock
);
5121 * If tree ref isn't set then we know the ref on this eb is a real ref,
5122 * so just return, this page will likely be freed soon anyway.
5124 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5125 spin_unlock(&eb
->refs_lock
);
5129 return release_extent_buffer(eb
);