1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
25 static struct kmem_cache
*extent_state_cache
;
26 static struct kmem_cache
*extent_buffer_cache
;
28 static LIST_HEAD(buffers
);
29 static LIST_HEAD(states
);
33 static DEFINE_SPINLOCK(leak_lock
);
36 #define BUFFER_LRU_MAX 64
41 struct rb_node rb_node
;
44 struct extent_page_data
{
46 struct extent_io_tree
*tree
;
47 get_extent_t
*get_extent
;
48 unsigned long bio_flags
;
50 /* tells writepage not to lock the state bits for this range
51 * it still does the unlocking
53 unsigned int extent_locked
:1;
55 /* tells the submit_bio code to use a WRITE_SYNC */
56 unsigned int sync_io
:1;
59 static noinline
void flush_write_bio(void *data
);
60 static inline struct btrfs_fs_info
*
61 tree_fs_info(struct extent_io_tree
*tree
)
63 return btrfs_sb(tree
->mapping
->host
->i_sb
);
66 int __init
extent_io_init(void)
68 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
69 sizeof(struct extent_state
), 0,
70 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
71 if (!extent_state_cache
)
74 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
75 sizeof(struct extent_buffer
), 0,
76 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
77 if (!extent_buffer_cache
)
78 goto free_state_cache
;
82 kmem_cache_destroy(extent_state_cache
);
86 void extent_io_exit(void)
88 struct extent_state
*state
;
89 struct extent_buffer
*eb
;
91 while (!list_empty(&states
)) {
92 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
93 printk(KERN_ERR
"btrfs state leak: start %llu end %llu "
94 "state %lu in tree %p refs %d\n",
95 (unsigned long long)state
->start
,
96 (unsigned long long)state
->end
,
97 state
->state
, state
->tree
, atomic_read(&state
->refs
));
98 list_del(&state
->leak_list
);
99 kmem_cache_free(extent_state_cache
, state
);
103 while (!list_empty(&buffers
)) {
104 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
105 printk(KERN_ERR
"btrfs buffer leak start %llu len %lu "
106 "refs %d\n", (unsigned long long)eb
->start
,
107 eb
->len
, atomic_read(&eb
->refs
));
108 list_del(&eb
->leak_list
);
109 kmem_cache_free(extent_buffer_cache
, eb
);
113 * Make sure all delayed rcu free are flushed before we
117 if (extent_state_cache
)
118 kmem_cache_destroy(extent_state_cache
);
119 if (extent_buffer_cache
)
120 kmem_cache_destroy(extent_buffer_cache
);
123 void extent_io_tree_init(struct extent_io_tree
*tree
,
124 struct address_space
*mapping
)
126 tree
->state
= RB_ROOT
;
127 INIT_RADIX_TREE(&tree
->buffer
, GFP_ATOMIC
);
129 tree
->dirty_bytes
= 0;
130 spin_lock_init(&tree
->lock
);
131 spin_lock_init(&tree
->buffer_lock
);
132 tree
->mapping
= mapping
;
135 static struct extent_state
*alloc_extent_state(gfp_t mask
)
137 struct extent_state
*state
;
142 state
= kmem_cache_alloc(extent_state_cache
, mask
);
149 spin_lock_irqsave(&leak_lock
, flags
);
150 list_add(&state
->leak_list
, &states
);
151 spin_unlock_irqrestore(&leak_lock
, flags
);
153 atomic_set(&state
->refs
, 1);
154 init_waitqueue_head(&state
->wq
);
155 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
159 void free_extent_state(struct extent_state
*state
)
163 if (atomic_dec_and_test(&state
->refs
)) {
167 WARN_ON(state
->tree
);
169 spin_lock_irqsave(&leak_lock
, flags
);
170 list_del(&state
->leak_list
);
171 spin_unlock_irqrestore(&leak_lock
, flags
);
173 trace_free_extent_state(state
, _RET_IP_
);
174 kmem_cache_free(extent_state_cache
, state
);
178 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
179 struct rb_node
*node
)
181 struct rb_node
**p
= &root
->rb_node
;
182 struct rb_node
*parent
= NULL
;
183 struct tree_entry
*entry
;
187 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
189 if (offset
< entry
->start
)
191 else if (offset
> entry
->end
)
197 rb_link_node(node
, parent
, p
);
198 rb_insert_color(node
, root
);
202 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
203 struct rb_node
**prev_ret
,
204 struct rb_node
**next_ret
)
206 struct rb_root
*root
= &tree
->state
;
207 struct rb_node
*n
= root
->rb_node
;
208 struct rb_node
*prev
= NULL
;
209 struct rb_node
*orig_prev
= NULL
;
210 struct tree_entry
*entry
;
211 struct tree_entry
*prev_entry
= NULL
;
214 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
218 if (offset
< entry
->start
)
220 else if (offset
> entry
->end
)
228 while (prev
&& offset
> prev_entry
->end
) {
229 prev
= rb_next(prev
);
230 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
237 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
238 while (prev
&& offset
< prev_entry
->start
) {
239 prev
= rb_prev(prev
);
240 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
247 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
250 struct rb_node
*prev
= NULL
;
253 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
259 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
260 struct extent_state
*other
)
262 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
263 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
268 * utility function to look for merge candidates inside a given range.
269 * Any extents with matching state are merged together into a single
270 * extent in the tree. Extents with EXTENT_IO in their state field
271 * are not merged because the end_io handlers need to be able to do
272 * operations on them without sleeping (or doing allocations/splits).
274 * This should be called with the tree lock held.
276 static void merge_state(struct extent_io_tree
*tree
,
277 struct extent_state
*state
)
279 struct extent_state
*other
;
280 struct rb_node
*other_node
;
282 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
285 other_node
= rb_prev(&state
->rb_node
);
287 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
288 if (other
->end
== state
->start
- 1 &&
289 other
->state
== state
->state
) {
290 merge_cb(tree
, state
, other
);
291 state
->start
= other
->start
;
293 rb_erase(&other
->rb_node
, &tree
->state
);
294 free_extent_state(other
);
297 other_node
= rb_next(&state
->rb_node
);
299 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
300 if (other
->start
== state
->end
+ 1 &&
301 other
->state
== state
->state
) {
302 merge_cb(tree
, state
, other
);
303 state
->end
= other
->end
;
305 rb_erase(&other
->rb_node
, &tree
->state
);
306 free_extent_state(other
);
311 static void set_state_cb(struct extent_io_tree
*tree
,
312 struct extent_state
*state
, int *bits
)
314 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
315 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
318 static void clear_state_cb(struct extent_io_tree
*tree
,
319 struct extent_state
*state
, int *bits
)
321 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
322 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
325 static void set_state_bits(struct extent_io_tree
*tree
,
326 struct extent_state
*state
, int *bits
);
329 * insert an extent_state struct into the tree. 'bits' are set on the
330 * struct before it is inserted.
332 * This may return -EEXIST if the extent is already there, in which case the
333 * state struct is freed.
335 * The tree lock is not taken internally. This is a utility function and
336 * probably isn't what you want to call (see set/clear_extent_bit).
338 static int insert_state(struct extent_io_tree
*tree
,
339 struct extent_state
*state
, u64 start
, u64 end
,
342 struct rb_node
*node
;
345 WARN(1, KERN_ERR
"btrfs end < start %llu %llu\n",
346 (unsigned long long)end
,
347 (unsigned long long)start
);
348 state
->start
= start
;
351 set_state_bits(tree
, state
, bits
);
353 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
355 struct extent_state
*found
;
356 found
= rb_entry(node
, struct extent_state
, rb_node
);
357 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
358 "%llu %llu\n", (unsigned long long)found
->start
,
359 (unsigned long long)found
->end
,
360 (unsigned long long)start
, (unsigned long long)end
);
364 merge_state(tree
, state
);
368 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
371 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
372 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
376 * split a given extent state struct in two, inserting the preallocated
377 * struct 'prealloc' as the newly created second half. 'split' indicates an
378 * offset inside 'orig' where it should be split.
381 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
382 * are two extent state structs in the tree:
383 * prealloc: [orig->start, split - 1]
384 * orig: [ split, orig->end ]
386 * The tree locks are not taken by this function. They need to be held
389 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
390 struct extent_state
*prealloc
, u64 split
)
392 struct rb_node
*node
;
394 split_cb(tree
, orig
, split
);
396 prealloc
->start
= orig
->start
;
397 prealloc
->end
= split
- 1;
398 prealloc
->state
= orig
->state
;
401 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
403 free_extent_state(prealloc
);
406 prealloc
->tree
= tree
;
410 static struct extent_state
*next_state(struct extent_state
*state
)
412 struct rb_node
*next
= rb_next(&state
->rb_node
);
414 return rb_entry(next
, struct extent_state
, rb_node
);
420 * utility function to clear some bits in an extent state struct.
421 * it will optionally wake up any one waiting on this state (wake == 1).
423 * If no bits are set on the state struct after clearing things, the
424 * struct is freed and removed from the tree
426 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
427 struct extent_state
*state
,
430 struct extent_state
*next
;
431 int bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
433 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
434 u64 range
= state
->end
- state
->start
+ 1;
435 WARN_ON(range
> tree
->dirty_bytes
);
436 tree
->dirty_bytes
-= range
;
438 clear_state_cb(tree
, state
, bits
);
439 state
->state
&= ~bits_to_clear
;
442 if (state
->state
== 0) {
443 next
= next_state(state
);
445 rb_erase(&state
->rb_node
, &tree
->state
);
447 free_extent_state(state
);
452 merge_state(tree
, state
);
453 next
= next_state(state
);
458 static struct extent_state
*
459 alloc_extent_state_atomic(struct extent_state
*prealloc
)
462 prealloc
= alloc_extent_state(GFP_ATOMIC
);
467 void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
469 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
470 "Extent tree was modified by another "
471 "thread while locked.");
475 * clear some bits on a range in the tree. This may require splitting
476 * or inserting elements in the tree, so the gfp mask is used to
477 * indicate which allocations or sleeping are allowed.
479 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
480 * the given range from the tree regardless of state (ie for truncate).
482 * the range [start, end] is inclusive.
484 * This takes the tree lock, and returns 0 on success and < 0 on error.
486 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
487 int bits
, int wake
, int delete,
488 struct extent_state
**cached_state
,
491 struct extent_state
*state
;
492 struct extent_state
*cached
;
493 struct extent_state
*prealloc
= NULL
;
494 struct rb_node
*node
;
500 bits
|= ~EXTENT_CTLBITS
;
501 bits
|= EXTENT_FIRST_DELALLOC
;
503 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
506 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
507 prealloc
= alloc_extent_state(mask
);
512 spin_lock(&tree
->lock
);
514 cached
= *cached_state
;
517 *cached_state
= NULL
;
521 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
522 cached
->end
> start
) {
524 atomic_dec(&cached
->refs
);
529 free_extent_state(cached
);
532 * this search will find the extents that end after
535 node
= tree_search(tree
, start
);
538 state
= rb_entry(node
, struct extent_state
, rb_node
);
540 if (state
->start
> end
)
542 WARN_ON(state
->end
< start
);
543 last_end
= state
->end
;
545 /* the state doesn't have the wanted bits, go ahead */
546 if (!(state
->state
& bits
)) {
547 state
= next_state(state
);
552 * | ---- desired range ---- |
554 * | ------------- state -------------- |
556 * We need to split the extent we found, and may flip
557 * bits on second half.
559 * If the extent we found extends past our range, we
560 * just split and search again. It'll get split again
561 * the next time though.
563 * If the extent we found is inside our range, we clear
564 * the desired bit on it.
567 if (state
->start
< start
) {
568 prealloc
= alloc_extent_state_atomic(prealloc
);
570 err
= split_state(tree
, state
, prealloc
, start
);
572 extent_io_tree_panic(tree
, err
);
577 if (state
->end
<= end
) {
578 state
= clear_state_bit(tree
, state
, &bits
, wake
);
584 * | ---- desired range ---- |
586 * We need to split the extent, and clear the bit
589 if (state
->start
<= end
&& state
->end
> end
) {
590 prealloc
= alloc_extent_state_atomic(prealloc
);
592 err
= split_state(tree
, state
, prealloc
, end
+ 1);
594 extent_io_tree_panic(tree
, err
);
599 clear_state_bit(tree
, prealloc
, &bits
, wake
);
605 state
= clear_state_bit(tree
, state
, &bits
, wake
);
607 if (last_end
== (u64
)-1)
609 start
= last_end
+ 1;
610 if (start
<= end
&& state
&& !need_resched())
615 spin_unlock(&tree
->lock
);
617 free_extent_state(prealloc
);
624 spin_unlock(&tree
->lock
);
625 if (mask
& __GFP_WAIT
)
630 static void wait_on_state(struct extent_io_tree
*tree
,
631 struct extent_state
*state
)
632 __releases(tree
->lock
)
633 __acquires(tree
->lock
)
636 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
637 spin_unlock(&tree
->lock
);
639 spin_lock(&tree
->lock
);
640 finish_wait(&state
->wq
, &wait
);
644 * waits for one or more bits to clear on a range in the state tree.
645 * The range [start, end] is inclusive.
646 * The tree lock is taken by this function
648 void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
650 struct extent_state
*state
;
651 struct rb_node
*node
;
653 spin_lock(&tree
->lock
);
657 * this search will find all the extents that end after
660 node
= tree_search(tree
, start
);
664 state
= rb_entry(node
, struct extent_state
, rb_node
);
666 if (state
->start
> end
)
669 if (state
->state
& bits
) {
670 start
= state
->start
;
671 atomic_inc(&state
->refs
);
672 wait_on_state(tree
, state
);
673 free_extent_state(state
);
676 start
= state
->end
+ 1;
681 cond_resched_lock(&tree
->lock
);
684 spin_unlock(&tree
->lock
);
687 static void set_state_bits(struct extent_io_tree
*tree
,
688 struct extent_state
*state
,
691 int bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
693 set_state_cb(tree
, state
, bits
);
694 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
695 u64 range
= state
->end
- state
->start
+ 1;
696 tree
->dirty_bytes
+= range
;
698 state
->state
|= bits_to_set
;
701 static void cache_state(struct extent_state
*state
,
702 struct extent_state
**cached_ptr
)
704 if (cached_ptr
&& !(*cached_ptr
)) {
705 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
707 atomic_inc(&state
->refs
);
712 static void uncache_state(struct extent_state
**cached_ptr
)
714 if (cached_ptr
&& (*cached_ptr
)) {
715 struct extent_state
*state
= *cached_ptr
;
717 free_extent_state(state
);
722 * set some bits on a range in the tree. This may require allocations or
723 * sleeping, so the gfp mask is used to indicate what is allowed.
725 * If any of the exclusive bits are set, this will fail with -EEXIST if some
726 * part of the range already has the desired bits set. The start of the
727 * existing range is returned in failed_start in this case.
729 * [start, end] is inclusive This takes the tree lock.
732 static int __must_check
733 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
734 int bits
, int exclusive_bits
, u64
*failed_start
,
735 struct extent_state
**cached_state
, gfp_t mask
)
737 struct extent_state
*state
;
738 struct extent_state
*prealloc
= NULL
;
739 struct rb_node
*node
;
744 bits
|= EXTENT_FIRST_DELALLOC
;
746 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
747 prealloc
= alloc_extent_state(mask
);
751 spin_lock(&tree
->lock
);
752 if (cached_state
&& *cached_state
) {
753 state
= *cached_state
;
754 if (state
->start
<= start
&& state
->end
> start
&&
756 node
= &state
->rb_node
;
761 * this search will find all the extents that end after
764 node
= tree_search(tree
, start
);
766 prealloc
= alloc_extent_state_atomic(prealloc
);
768 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
770 extent_io_tree_panic(tree
, err
);
775 state
= rb_entry(node
, struct extent_state
, rb_node
);
777 last_start
= state
->start
;
778 last_end
= state
->end
;
781 * | ---- desired range ---- |
784 * Just lock what we found and keep going
786 if (state
->start
== start
&& state
->end
<= end
) {
787 if (state
->state
& exclusive_bits
) {
788 *failed_start
= state
->start
;
793 set_state_bits(tree
, state
, &bits
);
794 cache_state(state
, cached_state
);
795 merge_state(tree
, state
);
796 if (last_end
== (u64
)-1)
798 start
= last_end
+ 1;
799 state
= next_state(state
);
800 if (start
< end
&& state
&& state
->start
== start
&&
807 * | ---- desired range ---- |
810 * | ------------- state -------------- |
812 * We need to split the extent we found, and may flip bits on
815 * If the extent we found extends past our
816 * range, we just split and search again. It'll get split
817 * again the next time though.
819 * If the extent we found is inside our range, we set the
822 if (state
->start
< start
) {
823 if (state
->state
& exclusive_bits
) {
824 *failed_start
= start
;
829 prealloc
= alloc_extent_state_atomic(prealloc
);
831 err
= split_state(tree
, state
, prealloc
, start
);
833 extent_io_tree_panic(tree
, err
);
838 if (state
->end
<= end
) {
839 set_state_bits(tree
, state
, &bits
);
840 cache_state(state
, cached_state
);
841 merge_state(tree
, state
);
842 if (last_end
== (u64
)-1)
844 start
= last_end
+ 1;
845 state
= next_state(state
);
846 if (start
< end
&& state
&& state
->start
== start
&&
853 * | ---- desired range ---- |
854 * | state | or | state |
856 * There's a hole, we need to insert something in it and
857 * ignore the extent we found.
859 if (state
->start
> start
) {
861 if (end
< last_start
)
864 this_end
= last_start
- 1;
866 prealloc
= alloc_extent_state_atomic(prealloc
);
870 * Avoid to free 'prealloc' if it can be merged with
873 err
= insert_state(tree
, prealloc
, start
, this_end
,
876 extent_io_tree_panic(tree
, err
);
878 cache_state(prealloc
, cached_state
);
880 start
= this_end
+ 1;
884 * | ---- desired range ---- |
886 * We need to split the extent, and set the bit
889 if (state
->start
<= end
&& state
->end
> end
) {
890 if (state
->state
& exclusive_bits
) {
891 *failed_start
= start
;
896 prealloc
= alloc_extent_state_atomic(prealloc
);
898 err
= split_state(tree
, state
, prealloc
, end
+ 1);
900 extent_io_tree_panic(tree
, err
);
902 set_state_bits(tree
, prealloc
, &bits
);
903 cache_state(prealloc
, cached_state
);
904 merge_state(tree
, prealloc
);
912 spin_unlock(&tree
->lock
);
914 free_extent_state(prealloc
);
921 spin_unlock(&tree
->lock
);
922 if (mask
& __GFP_WAIT
)
927 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
,
928 u64
*failed_start
, struct extent_state
**cached_state
,
931 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
937 * convert_extent_bit - convert all bits in a given range from one bit to
939 * @tree: the io tree to search
940 * @start: the start offset in bytes
941 * @end: the end offset in bytes (inclusive)
942 * @bits: the bits to set in this range
943 * @clear_bits: the bits to clear in this range
944 * @cached_state: state that we're going to cache
945 * @mask: the allocation mask
947 * This will go through and set bits for the given range. If any states exist
948 * already in this range they are set with the given bit and cleared of the
949 * clear_bits. This is only meant to be used by things that are mergeable, ie
950 * converting from say DELALLOC to DIRTY. This is not meant to be used with
951 * boundary bits like LOCK.
953 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
954 int bits
, int clear_bits
,
955 struct extent_state
**cached_state
, gfp_t mask
)
957 struct extent_state
*state
;
958 struct extent_state
*prealloc
= NULL
;
959 struct rb_node
*node
;
965 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
966 prealloc
= alloc_extent_state(mask
);
971 spin_lock(&tree
->lock
);
972 if (cached_state
&& *cached_state
) {
973 state
= *cached_state
;
974 if (state
->start
<= start
&& state
->end
> start
&&
976 node
= &state
->rb_node
;
982 * this search will find all the extents that end after
985 node
= tree_search(tree
, start
);
987 prealloc
= alloc_extent_state_atomic(prealloc
);
992 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
995 extent_io_tree_panic(tree
, err
);
998 state
= rb_entry(node
, struct extent_state
, rb_node
);
1000 last_start
= state
->start
;
1001 last_end
= state
->end
;
1004 * | ---- desired range ---- |
1007 * Just lock what we found and keep going
1009 if (state
->start
== start
&& state
->end
<= end
) {
1010 set_state_bits(tree
, state
, &bits
);
1011 cache_state(state
, cached_state
);
1012 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1013 if (last_end
== (u64
)-1)
1015 start
= last_end
+ 1;
1016 if (start
< end
&& state
&& state
->start
== start
&&
1023 * | ---- desired range ---- |
1026 * | ------------- state -------------- |
1028 * We need to split the extent we found, and may flip bits on
1031 * If the extent we found extends past our
1032 * range, we just split and search again. It'll get split
1033 * again the next time though.
1035 * If the extent we found is inside our range, we set the
1036 * desired bit on it.
1038 if (state
->start
< start
) {
1039 prealloc
= alloc_extent_state_atomic(prealloc
);
1044 err
= split_state(tree
, state
, prealloc
, start
);
1046 extent_io_tree_panic(tree
, err
);
1050 if (state
->end
<= end
) {
1051 set_state_bits(tree
, state
, &bits
);
1052 cache_state(state
, cached_state
);
1053 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1054 if (last_end
== (u64
)-1)
1056 start
= last_end
+ 1;
1057 if (start
< end
&& state
&& state
->start
== start
&&
1064 * | ---- desired range ---- |
1065 * | state | or | state |
1067 * There's a hole, we need to insert something in it and
1068 * ignore the extent we found.
1070 if (state
->start
> start
) {
1072 if (end
< last_start
)
1075 this_end
= last_start
- 1;
1077 prealloc
= alloc_extent_state_atomic(prealloc
);
1084 * Avoid to free 'prealloc' if it can be merged with
1087 err
= insert_state(tree
, prealloc
, start
, this_end
,
1090 extent_io_tree_panic(tree
, err
);
1091 cache_state(prealloc
, cached_state
);
1093 start
= this_end
+ 1;
1097 * | ---- desired range ---- |
1099 * We need to split the extent, and set the bit
1102 if (state
->start
<= end
&& state
->end
> end
) {
1103 prealloc
= alloc_extent_state_atomic(prealloc
);
1109 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1111 extent_io_tree_panic(tree
, err
);
1113 set_state_bits(tree
, prealloc
, &bits
);
1114 cache_state(prealloc
, cached_state
);
1115 clear_state_bit(tree
, prealloc
, &clear_bits
, 0);
1123 spin_unlock(&tree
->lock
);
1125 free_extent_state(prealloc
);
1132 spin_unlock(&tree
->lock
);
1133 if (mask
& __GFP_WAIT
)
1138 /* wrappers around set/clear extent bit */
1139 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1142 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, NULL
,
1146 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1147 int bits
, gfp_t mask
)
1149 return set_extent_bit(tree
, start
, end
, bits
, NULL
,
1153 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1154 int bits
, gfp_t mask
)
1156 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
1159 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1160 struct extent_state
**cached_state
, gfp_t mask
)
1162 return set_extent_bit(tree
, start
, end
,
1163 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1164 NULL
, cached_state
, mask
);
1167 int set_extent_defrag(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1168 struct extent_state
**cached_state
, gfp_t mask
)
1170 return set_extent_bit(tree
, start
, end
,
1171 EXTENT_DELALLOC
| EXTENT_UPTODATE
| EXTENT_DEFRAG
,
1172 NULL
, cached_state
, mask
);
1175 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1178 return clear_extent_bit(tree
, start
, end
,
1179 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1180 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1183 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1186 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, NULL
,
1190 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1191 struct extent_state
**cached_state
, gfp_t mask
)
1193 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0,
1194 cached_state
, mask
);
1197 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1198 struct extent_state
**cached_state
, gfp_t mask
)
1200 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1201 cached_state
, mask
);
1205 * either insert or lock state struct between start and end use mask to tell
1206 * us if waiting is desired.
1208 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1209 int bits
, struct extent_state
**cached_state
)
1214 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1215 EXTENT_LOCKED
, &failed_start
,
1216 cached_state
, GFP_NOFS
);
1217 if (err
== -EEXIST
) {
1218 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1219 start
= failed_start
;
1222 WARN_ON(start
> end
);
1227 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1229 return lock_extent_bits(tree
, start
, end
, 0, NULL
);
1232 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1237 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1238 &failed_start
, NULL
, GFP_NOFS
);
1239 if (err
== -EEXIST
) {
1240 if (failed_start
> start
)
1241 clear_extent_bit(tree
, start
, failed_start
- 1,
1242 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1248 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1249 struct extent_state
**cached
, gfp_t mask
)
1251 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1255 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1257 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1262 * helper function to set both pages and extents in the tree writeback
1264 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1266 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1267 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1270 while (index
<= end_index
) {
1271 page
= find_get_page(tree
->mapping
, index
);
1272 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1273 set_page_writeback(page
);
1274 page_cache_release(page
);
1280 /* find the first state struct with 'bits' set after 'start', and
1281 * return it. tree->lock must be held. NULL will returned if
1282 * nothing was found after 'start'
1284 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
1285 u64 start
, int bits
)
1287 struct rb_node
*node
;
1288 struct extent_state
*state
;
1291 * this search will find all the extents that end after
1294 node
= tree_search(tree
, start
);
1299 state
= rb_entry(node
, struct extent_state
, rb_node
);
1300 if (state
->end
>= start
&& (state
->state
& bits
))
1303 node
= rb_next(node
);
1312 * find the first offset in the io tree with 'bits' set. zero is
1313 * returned if we find something, and *start_ret and *end_ret are
1314 * set to reflect the state struct that was found.
1316 * If nothing was found, 1 is returned. If found something, return 0.
1318 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1319 u64
*start_ret
, u64
*end_ret
, int bits
,
1320 struct extent_state
**cached_state
)
1322 struct extent_state
*state
;
1326 spin_lock(&tree
->lock
);
1327 if (cached_state
&& *cached_state
) {
1328 state
= *cached_state
;
1329 if (state
->end
== start
- 1 && state
->tree
) {
1330 n
= rb_next(&state
->rb_node
);
1332 state
= rb_entry(n
, struct extent_state
,
1334 if (state
->state
& bits
)
1338 free_extent_state(*cached_state
);
1339 *cached_state
= NULL
;
1342 free_extent_state(*cached_state
);
1343 *cached_state
= NULL
;
1346 state
= find_first_extent_bit_state(tree
, start
, bits
);
1349 cache_state(state
, cached_state
);
1350 *start_ret
= state
->start
;
1351 *end_ret
= state
->end
;
1355 spin_unlock(&tree
->lock
);
1360 * find a contiguous range of bytes in the file marked as delalloc, not
1361 * more than 'max_bytes'. start and end are used to return the range,
1363 * 1 is returned if we find something, 0 if nothing was in the tree
1365 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1366 u64
*start
, u64
*end
, u64 max_bytes
,
1367 struct extent_state
**cached_state
)
1369 struct rb_node
*node
;
1370 struct extent_state
*state
;
1371 u64 cur_start
= *start
;
1373 u64 total_bytes
= 0;
1375 spin_lock(&tree
->lock
);
1378 * this search will find all the extents that end after
1381 node
= tree_search(tree
, cur_start
);
1389 state
= rb_entry(node
, struct extent_state
, rb_node
);
1390 if (found
&& (state
->start
!= cur_start
||
1391 (state
->state
& EXTENT_BOUNDARY
))) {
1394 if (!(state
->state
& EXTENT_DELALLOC
)) {
1400 *start
= state
->start
;
1401 *cached_state
= state
;
1402 atomic_inc(&state
->refs
);
1406 cur_start
= state
->end
+ 1;
1407 node
= rb_next(node
);
1410 total_bytes
+= state
->end
- state
->start
+ 1;
1411 if (total_bytes
>= max_bytes
)
1415 spin_unlock(&tree
->lock
);
1419 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1420 struct page
*locked_page
,
1424 struct page
*pages
[16];
1425 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1426 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1427 unsigned long nr_pages
= end_index
- index
+ 1;
1430 if (index
== locked_page
->index
&& end_index
== index
)
1433 while (nr_pages
> 0) {
1434 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1435 min_t(unsigned long, nr_pages
,
1436 ARRAY_SIZE(pages
)), pages
);
1437 for (i
= 0; i
< ret
; i
++) {
1438 if (pages
[i
] != locked_page
)
1439 unlock_page(pages
[i
]);
1440 page_cache_release(pages
[i
]);
1448 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1449 struct page
*locked_page
,
1453 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1454 unsigned long start_index
= index
;
1455 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1456 unsigned long pages_locked
= 0;
1457 struct page
*pages
[16];
1458 unsigned long nrpages
;
1462 /* the caller is responsible for locking the start index */
1463 if (index
== locked_page
->index
&& index
== end_index
)
1466 /* skip the page at the start index */
1467 nrpages
= end_index
- index
+ 1;
1468 while (nrpages
> 0) {
1469 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1470 min_t(unsigned long,
1471 nrpages
, ARRAY_SIZE(pages
)), pages
);
1476 /* now we have an array of pages, lock them all */
1477 for (i
= 0; i
< ret
; i
++) {
1479 * the caller is taking responsibility for
1482 if (pages
[i
] != locked_page
) {
1483 lock_page(pages
[i
]);
1484 if (!PageDirty(pages
[i
]) ||
1485 pages
[i
]->mapping
!= inode
->i_mapping
) {
1487 unlock_page(pages
[i
]);
1488 page_cache_release(pages
[i
]);
1492 page_cache_release(pages
[i
]);
1501 if (ret
&& pages_locked
) {
1502 __unlock_for_delalloc(inode
, locked_page
,
1504 ((u64
)(start_index
+ pages_locked
- 1)) <<
1511 * find a contiguous range of bytes in the file marked as delalloc, not
1512 * more than 'max_bytes'. start and end are used to return the range,
1514 * 1 is returned if we find something, 0 if nothing was in the tree
1516 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1517 struct extent_io_tree
*tree
,
1518 struct page
*locked_page
,
1519 u64
*start
, u64
*end
,
1525 struct extent_state
*cached_state
= NULL
;
1530 /* step one, find a bunch of delalloc bytes starting at start */
1531 delalloc_start
= *start
;
1533 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1534 max_bytes
, &cached_state
);
1535 if (!found
|| delalloc_end
<= *start
) {
1536 *start
= delalloc_start
;
1537 *end
= delalloc_end
;
1538 free_extent_state(cached_state
);
1543 * start comes from the offset of locked_page. We have to lock
1544 * pages in order, so we can't process delalloc bytes before
1547 if (delalloc_start
< *start
)
1548 delalloc_start
= *start
;
1551 * make sure to limit the number of pages we try to lock down
1554 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
)
1555 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1557 /* step two, lock all the pages after the page that has start */
1558 ret
= lock_delalloc_pages(inode
, locked_page
,
1559 delalloc_start
, delalloc_end
);
1560 if (ret
== -EAGAIN
) {
1561 /* some of the pages are gone, lets avoid looping by
1562 * shortening the size of the delalloc range we're searching
1564 free_extent_state(cached_state
);
1566 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1567 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1575 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1577 /* step three, lock the state bits for the whole range */
1578 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1580 /* then test to make sure it is all still delalloc */
1581 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1582 EXTENT_DELALLOC
, 1, cached_state
);
1584 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1585 &cached_state
, GFP_NOFS
);
1586 __unlock_for_delalloc(inode
, locked_page
,
1587 delalloc_start
, delalloc_end
);
1591 free_extent_state(cached_state
);
1592 *start
= delalloc_start
;
1593 *end
= delalloc_end
;
1598 int extent_clear_unlock_delalloc(struct inode
*inode
,
1599 struct extent_io_tree
*tree
,
1600 u64 start
, u64 end
, struct page
*locked_page
,
1604 struct page
*pages
[16];
1605 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1606 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1607 unsigned long nr_pages
= end_index
- index
+ 1;
1611 if (op
& EXTENT_CLEAR_UNLOCK
)
1612 clear_bits
|= EXTENT_LOCKED
;
1613 if (op
& EXTENT_CLEAR_DIRTY
)
1614 clear_bits
|= EXTENT_DIRTY
;
1616 if (op
& EXTENT_CLEAR_DELALLOC
)
1617 clear_bits
|= EXTENT_DELALLOC
;
1619 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1620 if (!(op
& (EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
1621 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
|
1622 EXTENT_SET_PRIVATE2
)))
1625 while (nr_pages
> 0) {
1626 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1627 min_t(unsigned long,
1628 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1629 for (i
= 0; i
< ret
; i
++) {
1631 if (op
& EXTENT_SET_PRIVATE2
)
1632 SetPagePrivate2(pages
[i
]);
1634 if (pages
[i
] == locked_page
) {
1635 page_cache_release(pages
[i
]);
1638 if (op
& EXTENT_CLEAR_DIRTY
)
1639 clear_page_dirty_for_io(pages
[i
]);
1640 if (op
& EXTENT_SET_WRITEBACK
)
1641 set_page_writeback(pages
[i
]);
1642 if (op
& EXTENT_END_WRITEBACK
)
1643 end_page_writeback(pages
[i
]);
1644 if (op
& EXTENT_CLEAR_UNLOCK_PAGE
)
1645 unlock_page(pages
[i
]);
1646 page_cache_release(pages
[i
]);
1656 * count the number of bytes in the tree that have a given bit(s)
1657 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1658 * cached. The total number found is returned.
1660 u64
count_range_bits(struct extent_io_tree
*tree
,
1661 u64
*start
, u64 search_end
, u64 max_bytes
,
1662 unsigned long bits
, int contig
)
1664 struct rb_node
*node
;
1665 struct extent_state
*state
;
1666 u64 cur_start
= *start
;
1667 u64 total_bytes
= 0;
1671 if (search_end
<= cur_start
) {
1676 spin_lock(&tree
->lock
);
1677 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1678 total_bytes
= tree
->dirty_bytes
;
1682 * this search will find all the extents that end after
1685 node
= tree_search(tree
, cur_start
);
1690 state
= rb_entry(node
, struct extent_state
, rb_node
);
1691 if (state
->start
> search_end
)
1693 if (contig
&& found
&& state
->start
> last
+ 1)
1695 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1696 total_bytes
+= min(search_end
, state
->end
) + 1 -
1697 max(cur_start
, state
->start
);
1698 if (total_bytes
>= max_bytes
)
1701 *start
= max(cur_start
, state
->start
);
1705 } else if (contig
&& found
) {
1708 node
= rb_next(node
);
1713 spin_unlock(&tree
->lock
);
1718 * set the private field for a given byte offset in the tree. If there isn't
1719 * an extent_state there already, this does nothing.
1721 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1723 struct rb_node
*node
;
1724 struct extent_state
*state
;
1727 spin_lock(&tree
->lock
);
1729 * this search will find all the extents that end after
1732 node
= tree_search(tree
, start
);
1737 state
= rb_entry(node
, struct extent_state
, rb_node
);
1738 if (state
->start
!= start
) {
1742 state
->private = private;
1744 spin_unlock(&tree
->lock
);
1748 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1750 struct rb_node
*node
;
1751 struct extent_state
*state
;
1754 spin_lock(&tree
->lock
);
1756 * this search will find all the extents that end after
1759 node
= tree_search(tree
, start
);
1764 state
= rb_entry(node
, struct extent_state
, rb_node
);
1765 if (state
->start
!= start
) {
1769 *private = state
->private;
1771 spin_unlock(&tree
->lock
);
1776 * searches a range in the state tree for a given mask.
1777 * If 'filled' == 1, this returns 1 only if every extent in the tree
1778 * has the bits set. Otherwise, 1 is returned if any bit in the
1779 * range is found set.
1781 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1782 int bits
, int filled
, struct extent_state
*cached
)
1784 struct extent_state
*state
= NULL
;
1785 struct rb_node
*node
;
1788 spin_lock(&tree
->lock
);
1789 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1790 cached
->end
> start
)
1791 node
= &cached
->rb_node
;
1793 node
= tree_search(tree
, start
);
1794 while (node
&& start
<= end
) {
1795 state
= rb_entry(node
, struct extent_state
, rb_node
);
1797 if (filled
&& state
->start
> start
) {
1802 if (state
->start
> end
)
1805 if (state
->state
& bits
) {
1809 } else if (filled
) {
1814 if (state
->end
== (u64
)-1)
1817 start
= state
->end
+ 1;
1820 node
= rb_next(node
);
1827 spin_unlock(&tree
->lock
);
1832 * helper function to set a given page up to date if all the
1833 * extents in the tree for that page are up to date
1835 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1837 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1838 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1839 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1840 SetPageUptodate(page
);
1844 * helper function to unlock a page if all the extents in the tree
1845 * for that page are unlocked
1847 static void check_page_locked(struct extent_io_tree
*tree
, struct page
*page
)
1849 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1850 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1851 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0, NULL
))
1856 * helper function to end page writeback if all the extents
1857 * in the tree for that page are done with writeback
1859 static void check_page_writeback(struct extent_io_tree
*tree
,
1862 end_page_writeback(page
);
1866 * When IO fails, either with EIO or csum verification fails, we
1867 * try other mirrors that might have a good copy of the data. This
1868 * io_failure_record is used to record state as we go through all the
1869 * mirrors. If another mirror has good data, the page is set up to date
1870 * and things continue. If a good mirror can't be found, the original
1871 * bio end_io callback is called to indicate things have failed.
1873 struct io_failure_record
{
1878 unsigned long bio_flags
;
1884 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
1889 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1891 set_state_private(failure_tree
, rec
->start
, 0);
1892 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1893 rec
->start
+ rec
->len
- 1,
1894 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1899 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1900 rec
->start
+ rec
->len
- 1,
1901 EXTENT_DAMAGED
, GFP_NOFS
);
1910 static void repair_io_failure_callback(struct bio
*bio
, int err
)
1912 complete(bio
->bi_private
);
1916 * this bypasses the standard btrfs submit functions deliberately, as
1917 * the standard behavior is to write all copies in a raid setup. here we only
1918 * want to write the one bad copy. so we do the mapping for ourselves and issue
1919 * submit_bio directly.
1920 * to avoid any synchonization issues, wait for the data after writing, which
1921 * actually prevents the read that triggered the error from finishing.
1922 * currently, there can be no more than two copies of every data bit. thus,
1923 * exactly one rewrite is required.
1925 int repair_io_failure(struct btrfs_mapping_tree
*map_tree
, u64 start
,
1926 u64 length
, u64 logical
, struct page
*page
,
1930 struct btrfs_device
*dev
;
1931 DECLARE_COMPLETION_ONSTACK(compl);
1934 struct btrfs_bio
*bbio
= NULL
;
1937 BUG_ON(!mirror_num
);
1939 bio
= bio_alloc(GFP_NOFS
, 1);
1942 bio
->bi_private
= &compl;
1943 bio
->bi_end_io
= repair_io_failure_callback
;
1945 map_length
= length
;
1947 ret
= btrfs_map_block(map_tree
, WRITE
, logical
,
1948 &map_length
, &bbio
, mirror_num
);
1953 BUG_ON(mirror_num
!= bbio
->mirror_num
);
1954 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
1955 bio
->bi_sector
= sector
;
1956 dev
= bbio
->stripes
[mirror_num
-1].dev
;
1958 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
1962 bio
->bi_bdev
= dev
->bdev
;
1963 bio_add_page(bio
, page
, length
, start
-page_offset(page
));
1964 btrfsic_submit_bio(WRITE_SYNC
, bio
);
1965 wait_for_completion(&compl);
1967 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
1968 /* try to remap that extent elsewhere? */
1970 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
1974 printk_ratelimited_in_rcu(KERN_INFO
"btrfs read error corrected: ino %lu off %llu "
1975 "(dev %s sector %llu)\n", page
->mapping
->host
->i_ino
,
1976 start
, rcu_str_deref(dev
->name
), sector
);
1982 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
1985 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
1986 u64 start
= eb
->start
;
1987 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
1990 for (i
= 0; i
< num_pages
; i
++) {
1991 struct page
*p
= extent_buffer_page(eb
, i
);
1992 ret
= repair_io_failure(map_tree
, start
, PAGE_CACHE_SIZE
,
1993 start
, p
, mirror_num
);
1996 start
+= PAGE_CACHE_SIZE
;
2003 * each time an IO finishes, we do a fast check in the IO failure tree
2004 * to see if we need to process or clean up an io_failure_record
2006 static int clean_io_failure(u64 start
, struct page
*page
)
2009 u64 private_failure
;
2010 struct io_failure_record
*failrec
;
2011 struct btrfs_mapping_tree
*map_tree
;
2012 struct extent_state
*state
;
2016 struct inode
*inode
= page
->mapping
->host
;
2019 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2020 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2024 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2029 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2030 BUG_ON(!failrec
->this_mirror
);
2032 if (failrec
->in_validation
) {
2033 /* there was no real error, just free the record */
2034 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2040 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2041 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2044 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2046 if (state
&& state
->start
== failrec
->start
) {
2047 map_tree
= &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
;
2048 num_copies
= btrfs_num_copies(map_tree
, failrec
->logical
,
2050 if (num_copies
> 1) {
2051 ret
= repair_io_failure(map_tree
, start
, failrec
->len
,
2052 failrec
->logical
, page
,
2053 failrec
->failed_mirror
);
2060 ret
= free_io_failure(inode
, failrec
, did_repair
);
2066 * this is a generic handler for readpage errors (default
2067 * readpage_io_failed_hook). if other copies exist, read those and write back
2068 * good data to the failed position. does not investigate in remapping the
2069 * failed extent elsewhere, hoping the device will be smart enough to do this as
2073 static int bio_readpage_error(struct bio
*failed_bio
, struct page
*page
,
2074 u64 start
, u64 end
, int failed_mirror
,
2075 struct extent_state
*state
)
2077 struct io_failure_record
*failrec
= NULL
;
2079 struct extent_map
*em
;
2080 struct inode
*inode
= page
->mapping
->host
;
2081 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2082 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2083 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2090 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2092 ret
= get_state_private(failure_tree
, start
, &private);
2094 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2097 failrec
->start
= start
;
2098 failrec
->len
= end
- start
+ 1;
2099 failrec
->this_mirror
= 0;
2100 failrec
->bio_flags
= 0;
2101 failrec
->in_validation
= 0;
2103 read_lock(&em_tree
->lock
);
2104 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2106 read_unlock(&em_tree
->lock
);
2111 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
2112 free_extent_map(em
);
2115 read_unlock(&em_tree
->lock
);
2121 logical
= start
- em
->start
;
2122 logical
= em
->block_start
+ logical
;
2123 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2124 logical
= em
->block_start
;
2125 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2126 extent_set_compress_type(&failrec
->bio_flags
,
2129 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2130 "len=%llu\n", logical
, start
, failrec
->len
);
2131 failrec
->logical
= logical
;
2132 free_extent_map(em
);
2134 /* set the bits in the private failure tree */
2135 ret
= set_extent_bits(failure_tree
, start
, end
,
2136 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2138 ret
= set_state_private(failure_tree
, start
,
2139 (u64
)(unsigned long)failrec
);
2140 /* set the bits in the inode's tree */
2142 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2149 failrec
= (struct io_failure_record
*)(unsigned long)private;
2150 pr_debug("bio_readpage_error: (found) logical=%llu, "
2151 "start=%llu, len=%llu, validation=%d\n",
2152 failrec
->logical
, failrec
->start
, failrec
->len
,
2153 failrec
->in_validation
);
2155 * when data can be on disk more than twice, add to failrec here
2156 * (e.g. with a list for failed_mirror) to make
2157 * clean_io_failure() clean all those errors at once.
2160 num_copies
= btrfs_num_copies(
2161 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
2162 failrec
->logical
, failrec
->len
);
2163 if (num_copies
== 1) {
2165 * we only have a single copy of the data, so don't bother with
2166 * all the retry and error correction code that follows. no
2167 * matter what the error is, it is very likely to persist.
2169 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2170 "state=%p, num_copies=%d, next_mirror %d, "
2171 "failed_mirror %d\n", state
, num_copies
,
2172 failrec
->this_mirror
, failed_mirror
);
2173 free_io_failure(inode
, failrec
, 0);
2178 spin_lock(&tree
->lock
);
2179 state
= find_first_extent_bit_state(tree
, failrec
->start
,
2181 if (state
&& state
->start
!= failrec
->start
)
2183 spin_unlock(&tree
->lock
);
2187 * there are two premises:
2188 * a) deliver good data to the caller
2189 * b) correct the bad sectors on disk
2191 if (failed_bio
->bi_vcnt
> 1) {
2193 * to fulfill b), we need to know the exact failing sectors, as
2194 * we don't want to rewrite any more than the failed ones. thus,
2195 * we need separate read requests for the failed bio
2197 * if the following BUG_ON triggers, our validation request got
2198 * merged. we need separate requests for our algorithm to work.
2200 BUG_ON(failrec
->in_validation
);
2201 failrec
->in_validation
= 1;
2202 failrec
->this_mirror
= failed_mirror
;
2203 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2206 * we're ready to fulfill a) and b) alongside. get a good copy
2207 * of the failed sector and if we succeed, we have setup
2208 * everything for repair_io_failure to do the rest for us.
2210 if (failrec
->in_validation
) {
2211 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2212 failrec
->in_validation
= 0;
2213 failrec
->this_mirror
= 0;
2215 failrec
->failed_mirror
= failed_mirror
;
2216 failrec
->this_mirror
++;
2217 if (failrec
->this_mirror
== failed_mirror
)
2218 failrec
->this_mirror
++;
2219 read_mode
= READ_SYNC
;
2222 if (!state
|| failrec
->this_mirror
> num_copies
) {
2223 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2224 "next_mirror %d, failed_mirror %d\n", state
,
2225 num_copies
, failrec
->this_mirror
, failed_mirror
);
2226 free_io_failure(inode
, failrec
, 0);
2230 bio
= bio_alloc(GFP_NOFS
, 1);
2232 free_io_failure(inode
, failrec
, 0);
2235 bio
->bi_private
= state
;
2236 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2237 bio
->bi_sector
= failrec
->logical
>> 9;
2238 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2241 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2243 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2244 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2245 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2247 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2248 failrec
->this_mirror
,
2249 failrec
->bio_flags
, 0);
2253 /* lots and lots of room for performance fixes in the end_bio funcs */
2255 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2257 int uptodate
= (err
== 0);
2258 struct extent_io_tree
*tree
;
2261 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2263 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2264 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2265 end
, NULL
, uptodate
);
2271 ClearPageUptodate(page
);
2278 * after a writepage IO is done, we need to:
2279 * clear the uptodate bits on error
2280 * clear the writeback bits in the extent tree for this IO
2281 * end_page_writeback if the page has no more pending IO
2283 * Scheduling is not allowed, so the extent state tree is expected
2284 * to have one and only one object corresponding to this IO.
2286 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2288 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2289 struct extent_io_tree
*tree
;
2295 struct page
*page
= bvec
->bv_page
;
2296 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2298 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
2300 end
= start
+ bvec
->bv_len
- 1;
2302 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
2307 if (--bvec
>= bio
->bi_io_vec
)
2308 prefetchw(&bvec
->bv_page
->flags
);
2310 if (end_extent_writepage(page
, err
, start
, end
))
2314 end_page_writeback(page
);
2316 check_page_writeback(tree
, page
);
2317 } while (bvec
>= bio
->bi_io_vec
);
2323 * after a readpage IO is done, we need to:
2324 * clear the uptodate bits on error
2325 * set the uptodate bits if things worked
2326 * set the page up to date if all extents in the tree are uptodate
2327 * clear the lock bit in the extent tree
2328 * unlock the page if there are no other extents locked for it
2330 * Scheduling is not allowed, so the extent state tree is expected
2331 * to have one and only one object corresponding to this IO.
2333 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2335 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2336 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2337 struct bio_vec
*bvec
= bio
->bi_io_vec
;
2338 struct extent_io_tree
*tree
;
2349 struct page
*page
= bvec
->bv_page
;
2350 struct extent_state
*cached
= NULL
;
2351 struct extent_state
*state
;
2353 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2354 "mirror=%ld\n", (u64
)bio
->bi_sector
, err
,
2355 (long int)bio
->bi_bdev
);
2356 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2358 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
2360 end
= start
+ bvec
->bv_len
- 1;
2362 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
2367 if (++bvec
<= bvec_end
)
2368 prefetchw(&bvec
->bv_page
->flags
);
2370 spin_lock(&tree
->lock
);
2371 state
= find_first_extent_bit_state(tree
, start
, EXTENT_LOCKED
);
2372 if (state
&& state
->start
== start
) {
2374 * take a reference on the state, unlock will drop
2377 cache_state(state
, &cached
);
2379 spin_unlock(&tree
->lock
);
2381 mirror
= (int)(unsigned long)bio
->bi_bdev
;
2382 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
2383 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
2388 clean_io_failure(start
, page
);
2391 if (!uptodate
&& tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2392 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2394 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2396 } else if (!uptodate
) {
2398 * The generic bio_readpage_error handles errors the
2399 * following way: If possible, new read requests are
2400 * created and submitted and will end up in
2401 * end_bio_extent_readpage as well (if we're lucky, not
2402 * in the !uptodate case). In that case it returns 0 and
2403 * we just go on with the next page in our bio. If it
2404 * can't handle the error it will return -EIO and we
2405 * remain responsible for that page.
2407 ret
= bio_readpage_error(bio
, page
, start
, end
, mirror
, NULL
);
2410 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2413 uncache_state(&cached
);
2418 if (uptodate
&& tree
->track_uptodate
) {
2419 set_extent_uptodate(tree
, start
, end
, &cached
,
2422 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2426 SetPageUptodate(page
);
2428 ClearPageUptodate(page
);
2434 check_page_uptodate(tree
, page
);
2436 ClearPageUptodate(page
);
2439 check_page_locked(tree
, page
);
2441 } while (bvec
<= bvec_end
);
2447 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2452 bio
= bio_alloc(gfp_flags
, nr_vecs
);
2454 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2455 while (!bio
&& (nr_vecs
/= 2))
2456 bio
= bio_alloc(gfp_flags
, nr_vecs
);
2461 bio
->bi_bdev
= bdev
;
2462 bio
->bi_sector
= first_sector
;
2468 * Since writes are async, they will only return -ENOMEM.
2469 * Reads can return the full range of I/O error conditions.
2471 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2472 int mirror_num
, unsigned long bio_flags
)
2475 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2476 struct page
*page
= bvec
->bv_page
;
2477 struct extent_io_tree
*tree
= bio
->bi_private
;
2480 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
2482 bio
->bi_private
= NULL
;
2486 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2487 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2488 mirror_num
, bio_flags
, start
);
2490 btrfsic_submit_bio(rw
, bio
);
2492 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2498 static int merge_bio(struct extent_io_tree
*tree
, struct page
*page
,
2499 unsigned long offset
, size_t size
, struct bio
*bio
,
2500 unsigned long bio_flags
)
2503 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2504 ret
= tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
,
2511 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2512 struct page
*page
, sector_t sector
,
2513 size_t size
, unsigned long offset
,
2514 struct block_device
*bdev
,
2515 struct bio
**bio_ret
,
2516 unsigned long max_pages
,
2517 bio_end_io_t end_io_func
,
2519 unsigned long prev_bio_flags
,
2520 unsigned long bio_flags
)
2526 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2527 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2528 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2530 if (bio_ret
&& *bio_ret
) {
2533 contig
= bio
->bi_sector
== sector
;
2535 contig
= bio
->bi_sector
+ (bio
->bi_size
>> 9) ==
2538 if (prev_bio_flags
!= bio_flags
|| !contig
||
2539 merge_bio(tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2540 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2541 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2550 if (this_compressed
)
2553 nr
= bio_get_nr_vecs(bdev
);
2555 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2559 bio_add_page(bio
, page
, page_size
, offset
);
2560 bio
->bi_end_io
= end_io_func
;
2561 bio
->bi_private
= tree
;
2566 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2571 void attach_extent_buffer_page(struct extent_buffer
*eb
, struct page
*page
)
2573 if (!PagePrivate(page
)) {
2574 SetPagePrivate(page
);
2575 page_cache_get(page
);
2576 set_page_private(page
, (unsigned long)eb
);
2578 WARN_ON(page
->private != (unsigned long)eb
);
2582 void set_page_extent_mapped(struct page
*page
)
2584 if (!PagePrivate(page
)) {
2585 SetPagePrivate(page
);
2586 page_cache_get(page
);
2587 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2592 * basic readpage implementation. Locked extent state structs are inserted
2593 * into the tree that are removed when the IO is done (by the end_io
2595 * XXX JDM: This needs looking at to ensure proper page locking
2597 static int __extent_read_full_page(struct extent_io_tree
*tree
,
2599 get_extent_t
*get_extent
,
2600 struct bio
**bio
, int mirror_num
,
2601 unsigned long *bio_flags
)
2603 struct inode
*inode
= page
->mapping
->host
;
2604 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2605 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2609 u64 last_byte
= i_size_read(inode
);
2613 struct extent_map
*em
;
2614 struct block_device
*bdev
;
2615 struct btrfs_ordered_extent
*ordered
;
2618 size_t pg_offset
= 0;
2620 size_t disk_io_size
;
2621 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2622 unsigned long this_bio_flag
= 0;
2624 set_page_extent_mapped(page
);
2626 if (!PageUptodate(page
)) {
2627 if (cleancache_get_page(page
) == 0) {
2628 BUG_ON(blocksize
!= PAGE_SIZE
);
2635 lock_extent(tree
, start
, end
);
2636 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
2639 unlock_extent(tree
, start
, end
);
2640 btrfs_start_ordered_extent(inode
, ordered
, 1);
2641 btrfs_put_ordered_extent(ordered
);
2644 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2646 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2649 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2650 userpage
= kmap_atomic(page
);
2651 memset(userpage
+ zero_offset
, 0, iosize
);
2652 flush_dcache_page(page
);
2653 kunmap_atomic(userpage
);
2656 while (cur
<= end
) {
2657 if (cur
>= last_byte
) {
2659 struct extent_state
*cached
= NULL
;
2661 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2662 userpage
= kmap_atomic(page
);
2663 memset(userpage
+ pg_offset
, 0, iosize
);
2664 flush_dcache_page(page
);
2665 kunmap_atomic(userpage
);
2666 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2668 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2672 em
= get_extent(inode
, page
, pg_offset
, cur
,
2674 if (IS_ERR_OR_NULL(em
)) {
2676 unlock_extent(tree
, cur
, end
);
2679 extent_offset
= cur
- em
->start
;
2680 BUG_ON(extent_map_end(em
) <= cur
);
2683 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2684 this_bio_flag
= EXTENT_BIO_COMPRESSED
;
2685 extent_set_compress_type(&this_bio_flag
,
2689 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2690 cur_end
= min(extent_map_end(em
) - 1, end
);
2691 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2692 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2693 disk_io_size
= em
->block_len
;
2694 sector
= em
->block_start
>> 9;
2696 sector
= (em
->block_start
+ extent_offset
) >> 9;
2697 disk_io_size
= iosize
;
2700 block_start
= em
->block_start
;
2701 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2702 block_start
= EXTENT_MAP_HOLE
;
2703 free_extent_map(em
);
2706 /* we've found a hole, just zero and go on */
2707 if (block_start
== EXTENT_MAP_HOLE
) {
2709 struct extent_state
*cached
= NULL
;
2711 userpage
= kmap_atomic(page
);
2712 memset(userpage
+ pg_offset
, 0, iosize
);
2713 flush_dcache_page(page
);
2714 kunmap_atomic(userpage
);
2716 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2718 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2721 pg_offset
+= iosize
;
2724 /* the get_extent function already copied into the page */
2725 if (test_range_bit(tree
, cur
, cur_end
,
2726 EXTENT_UPTODATE
, 1, NULL
)) {
2727 check_page_uptodate(tree
, page
);
2728 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2730 pg_offset
+= iosize
;
2733 /* we have an inline extent but it didn't get marked up
2734 * to date. Error out
2736 if (block_start
== EXTENT_MAP_INLINE
) {
2738 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2740 pg_offset
+= iosize
;
2745 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
2746 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
2750 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2752 ret
= submit_extent_page(READ
, tree
, page
,
2753 sector
, disk_io_size
, pg_offset
,
2755 end_bio_extent_readpage
, mirror_num
,
2760 *bio_flags
= this_bio_flag
;
2765 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2768 pg_offset
+= iosize
;
2772 if (!PageError(page
))
2773 SetPageUptodate(page
);
2779 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2780 get_extent_t
*get_extent
, int mirror_num
)
2782 struct bio
*bio
= NULL
;
2783 unsigned long bio_flags
= 0;
2786 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
2789 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
2793 static noinline
void update_nr_written(struct page
*page
,
2794 struct writeback_control
*wbc
,
2795 unsigned long nr_written
)
2797 wbc
->nr_to_write
-= nr_written
;
2798 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
2799 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
2800 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
2804 * the writepage semantics are similar to regular writepage. extent
2805 * records are inserted to lock ranges in the tree, and as dirty areas
2806 * are found, they are marked writeback. Then the lock bits are removed
2807 * and the end_io handler clears the writeback ranges
2809 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
2812 struct inode
*inode
= page
->mapping
->host
;
2813 struct extent_page_data
*epd
= data
;
2814 struct extent_io_tree
*tree
= epd
->tree
;
2815 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2817 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2821 u64 last_byte
= i_size_read(inode
);
2825 struct extent_state
*cached_state
= NULL
;
2826 struct extent_map
*em
;
2827 struct block_device
*bdev
;
2830 size_t pg_offset
= 0;
2832 loff_t i_size
= i_size_read(inode
);
2833 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2839 unsigned long nr_written
= 0;
2840 bool fill_delalloc
= true;
2842 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2843 write_flags
= WRITE_SYNC
;
2845 write_flags
= WRITE
;
2847 trace___extent_writepage(page
, inode
, wbc
);
2849 WARN_ON(!PageLocked(page
));
2851 ClearPageError(page
);
2853 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
2854 if (page
->index
> end_index
||
2855 (page
->index
== end_index
&& !pg_offset
)) {
2856 page
->mapping
->a_ops
->invalidatepage(page
, 0);
2861 if (page
->index
== end_index
) {
2864 userpage
= kmap_atomic(page
);
2865 memset(userpage
+ pg_offset
, 0,
2866 PAGE_CACHE_SIZE
- pg_offset
);
2867 kunmap_atomic(userpage
);
2868 flush_dcache_page(page
);
2872 set_page_extent_mapped(page
);
2874 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
2875 fill_delalloc
= false;
2877 delalloc_start
= start
;
2880 if (!epd
->extent_locked
&& fill_delalloc
) {
2881 u64 delalloc_to_write
= 0;
2883 * make sure the wbc mapping index is at least updated
2886 update_nr_written(page
, wbc
, 0);
2888 while (delalloc_end
< page_end
) {
2889 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
2894 if (nr_delalloc
== 0) {
2895 delalloc_start
= delalloc_end
+ 1;
2898 ret
= tree
->ops
->fill_delalloc(inode
, page
,
2903 /* File system has been set read-only */
2909 * delalloc_end is already one less than the total
2910 * length, so we don't subtract one from
2913 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
2916 delalloc_start
= delalloc_end
+ 1;
2918 if (wbc
->nr_to_write
< delalloc_to_write
) {
2921 if (delalloc_to_write
< thresh
* 2)
2922 thresh
= delalloc_to_write
;
2923 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
2927 /* did the fill delalloc function already unlock and start
2933 * we've unlocked the page, so we can't update
2934 * the mapping's writeback index, just update
2937 wbc
->nr_to_write
-= nr_written
;
2941 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
2942 ret
= tree
->ops
->writepage_start_hook(page
, start
,
2945 /* Fixup worker will requeue */
2947 wbc
->pages_skipped
++;
2949 redirty_page_for_writepage(wbc
, page
);
2950 update_nr_written(page
, wbc
, nr_written
);
2958 * we don't want to touch the inode after unlocking the page,
2959 * so we update the mapping writeback index now
2961 update_nr_written(page
, wbc
, nr_written
+ 1);
2964 if (last_byte
<= start
) {
2965 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2966 tree
->ops
->writepage_end_io_hook(page
, start
,
2971 blocksize
= inode
->i_sb
->s_blocksize
;
2973 while (cur
<= end
) {
2974 if (cur
>= last_byte
) {
2975 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2976 tree
->ops
->writepage_end_io_hook(page
, cur
,
2980 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
2982 if (IS_ERR_OR_NULL(em
)) {
2987 extent_offset
= cur
- em
->start
;
2988 BUG_ON(extent_map_end(em
) <= cur
);
2990 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2991 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2992 sector
= (em
->block_start
+ extent_offset
) >> 9;
2994 block_start
= em
->block_start
;
2995 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
2996 free_extent_map(em
);
3000 * compressed and inline extents are written through other
3003 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3004 block_start
== EXTENT_MAP_INLINE
) {
3006 * end_io notification does not happen here for
3007 * compressed extents
3009 if (!compressed
&& tree
->ops
&&
3010 tree
->ops
->writepage_end_io_hook
)
3011 tree
->ops
->writepage_end_io_hook(page
, cur
,
3014 else if (compressed
) {
3015 /* we don't want to end_page_writeback on
3016 * a compressed extent. this happens
3023 pg_offset
+= iosize
;
3026 /* leave this out until we have a page_mkwrite call */
3027 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
3028 EXTENT_DIRTY
, 0, NULL
)) {
3030 pg_offset
+= iosize
;
3034 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3035 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3043 unsigned long max_nr
= end_index
+ 1;
3045 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3046 if (!PageWriteback(page
)) {
3047 printk(KERN_ERR
"btrfs warning page %lu not "
3048 "writeback, cur %llu end %llu\n",
3049 page
->index
, (unsigned long long)cur
,
3050 (unsigned long long)end
);
3053 ret
= submit_extent_page(write_flags
, tree
, page
,
3054 sector
, iosize
, pg_offset
,
3055 bdev
, &epd
->bio
, max_nr
,
3056 end_bio_extent_writepage
,
3062 pg_offset
+= iosize
;
3067 /* make sure the mapping tag for page dirty gets cleared */
3068 set_page_writeback(page
);
3069 end_page_writeback(page
);
3075 /* drop our reference on any cached states */
3076 free_extent_state(cached_state
);
3080 static int eb_wait(void *word
)
3086 static void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3088 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3089 TASK_UNINTERRUPTIBLE
);
3092 static int lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3093 struct btrfs_fs_info
*fs_info
,
3094 struct extent_page_data
*epd
)
3096 unsigned long i
, num_pages
;
3100 if (!btrfs_try_tree_write_lock(eb
)) {
3102 flush_write_bio(epd
);
3103 btrfs_tree_lock(eb
);
3106 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3107 btrfs_tree_unlock(eb
);
3111 flush_write_bio(epd
);
3115 wait_on_extent_buffer_writeback(eb
);
3116 btrfs_tree_lock(eb
);
3117 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3119 btrfs_tree_unlock(eb
);
3124 * We need to do this to prevent races in people who check if the eb is
3125 * under IO since we can end up having no IO bits set for a short period
3128 spin_lock(&eb
->refs_lock
);
3129 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3130 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3131 spin_unlock(&eb
->refs_lock
);
3132 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3133 spin_lock(&fs_info
->delalloc_lock
);
3134 if (fs_info
->dirty_metadata_bytes
>= eb
->len
)
3135 fs_info
->dirty_metadata_bytes
-= eb
->len
;
3138 spin_unlock(&fs_info
->delalloc_lock
);
3141 spin_unlock(&eb
->refs_lock
);
3144 btrfs_tree_unlock(eb
);
3149 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3150 for (i
= 0; i
< num_pages
; i
++) {
3151 struct page
*p
= extent_buffer_page(eb
, i
);
3153 if (!trylock_page(p
)) {
3155 flush_write_bio(epd
);
3165 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3167 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3168 smp_mb__after_clear_bit();
3169 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3172 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3174 int uptodate
= err
== 0;
3175 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
3176 struct extent_buffer
*eb
;
3180 struct page
*page
= bvec
->bv_page
;
3183 eb
= (struct extent_buffer
*)page
->private;
3185 done
= atomic_dec_and_test(&eb
->io_pages
);
3187 if (!uptodate
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3188 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3189 ClearPageUptodate(page
);
3193 end_page_writeback(page
);
3198 end_extent_buffer_writeback(eb
);
3199 } while (bvec
>= bio
->bi_io_vec
);
3205 static int write_one_eb(struct extent_buffer
*eb
,
3206 struct btrfs_fs_info
*fs_info
,
3207 struct writeback_control
*wbc
,
3208 struct extent_page_data
*epd
)
3210 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3211 u64 offset
= eb
->start
;
3212 unsigned long i
, num_pages
;
3213 unsigned long bio_flags
= 0;
3214 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
);
3217 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3218 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3219 atomic_set(&eb
->io_pages
, num_pages
);
3220 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3221 bio_flags
= EXTENT_BIO_TREE_LOG
;
3223 for (i
= 0; i
< num_pages
; i
++) {
3224 struct page
*p
= extent_buffer_page(eb
, i
);
3226 clear_page_dirty_for_io(p
);
3227 set_page_writeback(p
);
3228 ret
= submit_extent_page(rw
, eb
->tree
, p
, offset
>> 9,
3229 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3230 -1, end_bio_extent_buffer_writepage
,
3231 0, epd
->bio_flags
, bio_flags
);
3232 epd
->bio_flags
= bio_flags
;
3234 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3236 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3237 end_extent_buffer_writeback(eb
);
3241 offset
+= PAGE_CACHE_SIZE
;
3242 update_nr_written(p
, wbc
, 1);
3246 if (unlikely(ret
)) {
3247 for (; i
< num_pages
; i
++) {
3248 struct page
*p
= extent_buffer_page(eb
, i
);
3256 int btree_write_cache_pages(struct address_space
*mapping
,
3257 struct writeback_control
*wbc
)
3259 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3260 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3261 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3262 struct extent_page_data epd
= {
3266 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3271 int nr_to_write_done
= 0;
3272 struct pagevec pvec
;
3275 pgoff_t end
; /* Inclusive */
3279 pagevec_init(&pvec
, 0);
3280 if (wbc
->range_cyclic
) {
3281 index
= mapping
->writeback_index
; /* Start from prev offset */
3284 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3285 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3288 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3289 tag
= PAGECACHE_TAG_TOWRITE
;
3291 tag
= PAGECACHE_TAG_DIRTY
;
3293 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3294 tag_pages_for_writeback(mapping
, index
, end
);
3295 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3296 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3297 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3301 for (i
= 0; i
< nr_pages
; i
++) {
3302 struct page
*page
= pvec
.pages
[i
];
3304 if (!PagePrivate(page
))
3307 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3312 spin_lock(&mapping
->private_lock
);
3313 if (!PagePrivate(page
)) {
3314 spin_unlock(&mapping
->private_lock
);
3318 eb
= (struct extent_buffer
*)page
->private;
3321 * Shouldn't happen and normally this would be a BUG_ON
3322 * but no sense in crashing the users box for something
3323 * we can survive anyway.
3326 spin_unlock(&mapping
->private_lock
);
3331 if (eb
== prev_eb
) {
3332 spin_unlock(&mapping
->private_lock
);
3336 ret
= atomic_inc_not_zero(&eb
->refs
);
3337 spin_unlock(&mapping
->private_lock
);
3342 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3344 free_extent_buffer(eb
);
3348 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3351 free_extent_buffer(eb
);
3354 free_extent_buffer(eb
);
3357 * the filesystem may choose to bump up nr_to_write.
3358 * We have to make sure to honor the new nr_to_write
3361 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3363 pagevec_release(&pvec
);
3366 if (!scanned
&& !done
) {
3368 * We hit the last page and there is more work to be done: wrap
3369 * back to the start of the file
3375 flush_write_bio(&epd
);
3380 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3381 * @mapping: address space structure to write
3382 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3383 * @writepage: function called for each page
3384 * @data: data passed to writepage function
3386 * If a page is already under I/O, write_cache_pages() skips it, even
3387 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3388 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3389 * and msync() need to guarantee that all the data which was dirty at the time
3390 * the call was made get new I/O started against them. If wbc->sync_mode is
3391 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3392 * existing IO to complete.
3394 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3395 struct address_space
*mapping
,
3396 struct writeback_control
*wbc
,
3397 writepage_t writepage
, void *data
,
3398 void (*flush_fn
)(void *))
3400 struct inode
*inode
= mapping
->host
;
3403 int nr_to_write_done
= 0;
3404 struct pagevec pvec
;
3407 pgoff_t end
; /* Inclusive */
3412 * We have to hold onto the inode so that ordered extents can do their
3413 * work when the IO finishes. The alternative to this is failing to add
3414 * an ordered extent if the igrab() fails there and that is a huge pain
3415 * to deal with, so instead just hold onto the inode throughout the
3416 * writepages operation. If it fails here we are freeing up the inode
3417 * anyway and we'd rather not waste our time writing out stuff that is
3418 * going to be truncated anyway.
3423 pagevec_init(&pvec
, 0);
3424 if (wbc
->range_cyclic
) {
3425 index
= mapping
->writeback_index
; /* Start from prev offset */
3428 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3429 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3432 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3433 tag
= PAGECACHE_TAG_TOWRITE
;
3435 tag
= PAGECACHE_TAG_DIRTY
;
3437 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3438 tag_pages_for_writeback(mapping
, index
, end
);
3439 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3440 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3441 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3445 for (i
= 0; i
< nr_pages
; i
++) {
3446 struct page
*page
= pvec
.pages
[i
];
3449 * At this point we hold neither mapping->tree_lock nor
3450 * lock on the page itself: the page may be truncated or
3451 * invalidated (changing page->mapping to NULL), or even
3452 * swizzled back from swapper_space to tmpfs file
3456 tree
->ops
->write_cache_pages_lock_hook
) {
3457 tree
->ops
->write_cache_pages_lock_hook(page
,
3460 if (!trylock_page(page
)) {
3466 if (unlikely(page
->mapping
!= mapping
)) {
3471 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3477 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3478 if (PageWriteback(page
))
3480 wait_on_page_writeback(page
);
3483 if (PageWriteback(page
) ||
3484 !clear_page_dirty_for_io(page
)) {
3489 ret
= (*writepage
)(page
, wbc
, data
);
3491 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3499 * the filesystem may choose to bump up nr_to_write.
3500 * We have to make sure to honor the new nr_to_write
3503 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3505 pagevec_release(&pvec
);
3508 if (!scanned
&& !done
) {
3510 * We hit the last page and there is more work to be done: wrap
3511 * back to the start of the file
3517 btrfs_add_delayed_iput(inode
);
3521 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3530 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
3531 BUG_ON(ret
< 0); /* -ENOMEM */
3536 static noinline
void flush_write_bio(void *data
)
3538 struct extent_page_data
*epd
= data
;
3539 flush_epd_write_bio(epd
);
3542 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3543 get_extent_t
*get_extent
,
3544 struct writeback_control
*wbc
)
3547 struct extent_page_data epd
= {
3550 .get_extent
= get_extent
,
3552 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3556 ret
= __extent_writepage(page
, wbc
, &epd
);
3558 flush_epd_write_bio(&epd
);
3562 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3563 u64 start
, u64 end
, get_extent_t
*get_extent
,
3567 struct address_space
*mapping
= inode
->i_mapping
;
3569 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3572 struct extent_page_data epd
= {
3575 .get_extent
= get_extent
,
3577 .sync_io
= mode
== WB_SYNC_ALL
,
3580 struct writeback_control wbc_writepages
= {
3582 .nr_to_write
= nr_pages
* 2,
3583 .range_start
= start
,
3584 .range_end
= end
+ 1,
3587 while (start
<= end
) {
3588 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3589 if (clear_page_dirty_for_io(page
))
3590 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3592 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3593 tree
->ops
->writepage_end_io_hook(page
, start
,
3594 start
+ PAGE_CACHE_SIZE
- 1,
3598 page_cache_release(page
);
3599 start
+= PAGE_CACHE_SIZE
;
3602 flush_epd_write_bio(&epd
);
3606 int extent_writepages(struct extent_io_tree
*tree
,
3607 struct address_space
*mapping
,
3608 get_extent_t
*get_extent
,
3609 struct writeback_control
*wbc
)
3612 struct extent_page_data epd
= {
3615 .get_extent
= get_extent
,
3617 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3621 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3622 __extent_writepage
, &epd
,
3624 flush_epd_write_bio(&epd
);
3628 int extent_readpages(struct extent_io_tree
*tree
,
3629 struct address_space
*mapping
,
3630 struct list_head
*pages
, unsigned nr_pages
,
3631 get_extent_t get_extent
)
3633 struct bio
*bio
= NULL
;
3635 unsigned long bio_flags
= 0;
3636 struct page
*pagepool
[16];
3641 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3642 page
= list_entry(pages
->prev
, struct page
, lru
);
3644 prefetchw(&page
->flags
);
3645 list_del(&page
->lru
);
3646 if (add_to_page_cache_lru(page
, mapping
,
3647 page
->index
, GFP_NOFS
)) {
3648 page_cache_release(page
);
3652 pagepool
[nr
++] = page
;
3653 if (nr
< ARRAY_SIZE(pagepool
))
3655 for (i
= 0; i
< nr
; i
++) {
3656 __extent_read_full_page(tree
, pagepool
[i
], get_extent
,
3657 &bio
, 0, &bio_flags
);
3658 page_cache_release(pagepool
[i
]);
3662 for (i
= 0; i
< nr
; i
++) {
3663 __extent_read_full_page(tree
, pagepool
[i
], get_extent
,
3664 &bio
, 0, &bio_flags
);
3665 page_cache_release(pagepool
[i
]);
3668 BUG_ON(!list_empty(pages
));
3670 return submit_one_bio(READ
, bio
, 0, bio_flags
);
3675 * basic invalidatepage code, this waits on any locked or writeback
3676 * ranges corresponding to the page, and then deletes any extent state
3677 * records from the tree
3679 int extent_invalidatepage(struct extent_io_tree
*tree
,
3680 struct page
*page
, unsigned long offset
)
3682 struct extent_state
*cached_state
= NULL
;
3683 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
3684 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3685 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
3687 start
+= (offset
+ blocksize
- 1) & ~(blocksize
- 1);
3691 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
3692 wait_on_page_writeback(page
);
3693 clear_extent_bit(tree
, start
, end
,
3694 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3695 EXTENT_DO_ACCOUNTING
,
3696 1, 1, &cached_state
, GFP_NOFS
);
3701 * a helper for releasepage, this tests for areas of the page that
3702 * are locked or under IO and drops the related state bits if it is safe
3705 int try_release_extent_state(struct extent_map_tree
*map
,
3706 struct extent_io_tree
*tree
, struct page
*page
,
3709 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3710 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3713 if (test_range_bit(tree
, start
, end
,
3714 EXTENT_IOBITS
, 0, NULL
))
3717 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
3720 * at this point we can safely clear everything except the
3721 * locked bit and the nodatasum bit
3723 ret
= clear_extent_bit(tree
, start
, end
,
3724 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
3727 /* if clear_extent_bit failed for enomem reasons,
3728 * we can't allow the release to continue.
3739 * a helper for releasepage. As long as there are no locked extents
3740 * in the range corresponding to the page, both state records and extent
3741 * map records are removed
3743 int try_release_extent_mapping(struct extent_map_tree
*map
,
3744 struct extent_io_tree
*tree
, struct page
*page
,
3747 struct extent_map
*em
;
3748 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3749 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3751 if ((mask
& __GFP_WAIT
) &&
3752 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
3754 while (start
<= end
) {
3755 len
= end
- start
+ 1;
3756 write_lock(&map
->lock
);
3757 em
= lookup_extent_mapping(map
, start
, len
);
3759 write_unlock(&map
->lock
);
3762 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
3763 em
->start
!= start
) {
3764 write_unlock(&map
->lock
);
3765 free_extent_map(em
);
3768 if (!test_range_bit(tree
, em
->start
,
3769 extent_map_end(em
) - 1,
3770 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
3772 remove_extent_mapping(map
, em
);
3773 /* once for the rb tree */
3774 free_extent_map(em
);
3776 start
= extent_map_end(em
);
3777 write_unlock(&map
->lock
);
3780 free_extent_map(em
);
3783 return try_release_extent_state(map
, tree
, page
, mask
);
3787 * helper function for fiemap, which doesn't want to see any holes.
3788 * This maps until we find something past 'last'
3790 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
3793 get_extent_t
*get_extent
)
3795 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
3796 struct extent_map
*em
;
3803 len
= last
- offset
;
3806 len
= (len
+ sectorsize
- 1) & ~(sectorsize
- 1);
3807 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
3808 if (IS_ERR_OR_NULL(em
))
3811 /* if this isn't a hole return it */
3812 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
3813 em
->block_start
!= EXTENT_MAP_HOLE
) {
3817 /* this is a hole, advance to the next extent */
3818 offset
= extent_map_end(em
);
3819 free_extent_map(em
);
3826 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
3827 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
3831 u64 max
= start
+ len
;
3835 u64 last_for_get_extent
= 0;
3837 u64 isize
= i_size_read(inode
);
3838 struct btrfs_key found_key
;
3839 struct extent_map
*em
= NULL
;
3840 struct extent_state
*cached_state
= NULL
;
3841 struct btrfs_path
*path
;
3842 struct btrfs_file_extent_item
*item
;
3847 unsigned long emflags
;
3852 path
= btrfs_alloc_path();
3855 path
->leave_spinning
= 1;
3857 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
3858 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
3861 * lookup the last file extent. We're not using i_size here
3862 * because there might be preallocation past i_size
3864 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
3865 path
, btrfs_ino(inode
), -1, 0);
3867 btrfs_free_path(path
);
3872 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3873 struct btrfs_file_extent_item
);
3874 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
3875 found_type
= btrfs_key_type(&found_key
);
3877 /* No extents, but there might be delalloc bits */
3878 if (found_key
.objectid
!= btrfs_ino(inode
) ||
3879 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3880 /* have to trust i_size as the end */
3882 last_for_get_extent
= isize
;
3885 * remember the start of the last extent. There are a
3886 * bunch of different factors that go into the length of the
3887 * extent, so its much less complex to remember where it started
3889 last
= found_key
.offset
;
3890 last_for_get_extent
= last
+ 1;
3892 btrfs_free_path(path
);
3895 * we might have some extents allocated but more delalloc past those
3896 * extents. so, we trust isize unless the start of the last extent is
3901 last_for_get_extent
= isize
;
3904 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
, 0,
3907 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
3917 u64 offset_in_extent
;
3919 /* break if the extent we found is outside the range */
3920 if (em
->start
>= max
|| extent_map_end(em
) < off
)
3924 * get_extent may return an extent that starts before our
3925 * requested range. We have to make sure the ranges
3926 * we return to fiemap always move forward and don't
3927 * overlap, so adjust the offsets here
3929 em_start
= max(em
->start
, off
);
3932 * record the offset from the start of the extent
3933 * for adjusting the disk offset below
3935 offset_in_extent
= em_start
- em
->start
;
3936 em_end
= extent_map_end(em
);
3937 em_len
= em_end
- em_start
;
3938 emflags
= em
->flags
;
3943 * bump off for our next call to get_extent
3945 off
= extent_map_end(em
);
3949 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
3951 flags
|= FIEMAP_EXTENT_LAST
;
3952 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
3953 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
3954 FIEMAP_EXTENT_NOT_ALIGNED
);
3955 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
3956 flags
|= (FIEMAP_EXTENT_DELALLOC
|
3957 FIEMAP_EXTENT_UNKNOWN
);
3959 disko
= em
->block_start
+ offset_in_extent
;
3961 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
3962 flags
|= FIEMAP_EXTENT_ENCODED
;
3964 free_extent_map(em
);
3966 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
3967 (last
== (u64
)-1 && isize
<= em_end
)) {
3968 flags
|= FIEMAP_EXTENT_LAST
;
3972 /* now scan forward to see if this is really the last extent. */
3973 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
3980 flags
|= FIEMAP_EXTENT_LAST
;
3983 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
3989 free_extent_map(em
);
3991 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
,
3992 &cached_state
, GFP_NOFS
);
3996 static void __free_extent_buffer(struct extent_buffer
*eb
)
3999 unsigned long flags
;
4000 spin_lock_irqsave(&leak_lock
, flags
);
4001 list_del(&eb
->leak_list
);
4002 spin_unlock_irqrestore(&leak_lock
, flags
);
4004 if (eb
->pages
&& eb
->pages
!= eb
->inline_pages
)
4006 kmem_cache_free(extent_buffer_cache
, eb
);
4009 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
4014 struct extent_buffer
*eb
= NULL
;
4016 unsigned long flags
;
4019 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
4026 rwlock_init(&eb
->lock
);
4027 atomic_set(&eb
->write_locks
, 0);
4028 atomic_set(&eb
->read_locks
, 0);
4029 atomic_set(&eb
->blocking_readers
, 0);
4030 atomic_set(&eb
->blocking_writers
, 0);
4031 atomic_set(&eb
->spinning_readers
, 0);
4032 atomic_set(&eb
->spinning_writers
, 0);
4033 eb
->lock_nested
= 0;
4034 init_waitqueue_head(&eb
->write_lock_wq
);
4035 init_waitqueue_head(&eb
->read_lock_wq
);
4038 spin_lock_irqsave(&leak_lock
, flags
);
4039 list_add(&eb
->leak_list
, &buffers
);
4040 spin_unlock_irqrestore(&leak_lock
, flags
);
4042 spin_lock_init(&eb
->refs_lock
);
4043 atomic_set(&eb
->refs
, 1);
4044 atomic_set(&eb
->io_pages
, 0);
4046 if (len
> MAX_INLINE_EXTENT_BUFFER_SIZE
) {
4047 struct page
**pages
;
4048 int num_pages
= (len
+ PAGE_CACHE_SIZE
- 1) >>
4050 pages
= kzalloc(num_pages
, mask
);
4052 __free_extent_buffer(eb
);
4057 eb
->pages
= eb
->inline_pages
;
4063 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4067 struct extent_buffer
*new;
4068 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4070 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_ATOMIC
);
4074 for (i
= 0; i
< num_pages
; i
++) {
4075 p
= alloc_page(GFP_ATOMIC
);
4077 attach_extent_buffer_page(new, p
);
4078 WARN_ON(PageDirty(p
));
4083 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4084 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4085 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4090 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
4092 struct extent_buffer
*eb
;
4093 unsigned long num_pages
= num_extent_pages(0, len
);
4096 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_ATOMIC
);
4100 for (i
= 0; i
< num_pages
; i
++) {
4101 eb
->pages
[i
] = alloc_page(GFP_ATOMIC
);
4105 set_extent_buffer_uptodate(eb
);
4106 btrfs_set_header_nritems(eb
, 0);
4107 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4112 __free_page(eb
->pages
[i
- 1]);
4113 __free_extent_buffer(eb
);
4117 static int extent_buffer_under_io(struct extent_buffer
*eb
)
4119 return (atomic_read(&eb
->io_pages
) ||
4120 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4121 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4125 * Helper for releasing extent buffer page.
4127 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4128 unsigned long start_idx
)
4130 unsigned long index
;
4131 unsigned long num_pages
;
4133 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4135 BUG_ON(extent_buffer_under_io(eb
));
4137 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4138 index
= start_idx
+ num_pages
;
4139 if (start_idx
>= index
)
4144 page
= extent_buffer_page(eb
, index
);
4145 if (page
&& mapped
) {
4146 spin_lock(&page
->mapping
->private_lock
);
4148 * We do this since we'll remove the pages after we've
4149 * removed the eb from the radix tree, so we could race
4150 * and have this page now attached to the new eb. So
4151 * only clear page_private if it's still connected to
4154 if (PagePrivate(page
) &&
4155 page
->private == (unsigned long)eb
) {
4156 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4157 BUG_ON(PageDirty(page
));
4158 BUG_ON(PageWriteback(page
));
4160 * We need to make sure we haven't be attached
4163 ClearPagePrivate(page
);
4164 set_page_private(page
, 0);
4165 /* One for the page private */
4166 page_cache_release(page
);
4168 spin_unlock(&page
->mapping
->private_lock
);
4172 /* One for when we alloced the page */
4173 page_cache_release(page
);
4175 } while (index
!= start_idx
);
4179 * Helper for releasing the extent buffer.
4181 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4183 btrfs_release_extent_buffer_page(eb
, 0);
4184 __free_extent_buffer(eb
);
4187 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4189 /* the ref bit is tricky. We have to make sure it is set
4190 * if we have the buffer dirty. Otherwise the
4191 * code to free a buffer can end up dropping a dirty
4194 * Once the ref bit is set, it won't go away while the
4195 * buffer is dirty or in writeback, and it also won't
4196 * go away while we have the reference count on the
4199 * We can't just set the ref bit without bumping the
4200 * ref on the eb because free_extent_buffer might
4201 * see the ref bit and try to clear it. If this happens
4202 * free_extent_buffer might end up dropping our original
4203 * ref by mistake and freeing the page before we are able
4204 * to add one more ref.
4206 * So bump the ref count first, then set the bit. If someone
4207 * beat us to it, drop the ref we added.
4209 spin_lock(&eb
->refs_lock
);
4210 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4211 atomic_inc(&eb
->refs
);
4212 spin_unlock(&eb
->refs_lock
);
4215 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
)
4217 unsigned long num_pages
, i
;
4219 check_buffer_tree_ref(eb
);
4221 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4222 for (i
= 0; i
< num_pages
; i
++) {
4223 struct page
*p
= extent_buffer_page(eb
, i
);
4224 mark_page_accessed(p
);
4228 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
4229 u64 start
, unsigned long len
)
4231 unsigned long num_pages
= num_extent_pages(start
, len
);
4233 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4234 struct extent_buffer
*eb
;
4235 struct extent_buffer
*exists
= NULL
;
4237 struct address_space
*mapping
= tree
->mapping
;
4242 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4243 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4245 mark_extent_buffer_accessed(eb
);
4250 eb
= __alloc_extent_buffer(tree
, start
, len
, GFP_NOFS
);
4254 for (i
= 0; i
< num_pages
; i
++, index
++) {
4255 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4259 spin_lock(&mapping
->private_lock
);
4260 if (PagePrivate(p
)) {
4262 * We could have already allocated an eb for this page
4263 * and attached one so lets see if we can get a ref on
4264 * the existing eb, and if we can we know it's good and
4265 * we can just return that one, else we know we can just
4266 * overwrite page->private.
4268 exists
= (struct extent_buffer
*)p
->private;
4269 if (atomic_inc_not_zero(&exists
->refs
)) {
4270 spin_unlock(&mapping
->private_lock
);
4272 page_cache_release(p
);
4273 mark_extent_buffer_accessed(exists
);
4278 * Do this so attach doesn't complain and we need to
4279 * drop the ref the old guy had.
4281 ClearPagePrivate(p
);
4282 WARN_ON(PageDirty(p
));
4283 page_cache_release(p
);
4285 attach_extent_buffer_page(eb
, p
);
4286 spin_unlock(&mapping
->private_lock
);
4287 WARN_ON(PageDirty(p
));
4288 mark_page_accessed(p
);
4290 if (!PageUptodate(p
))
4294 * see below about how we avoid a nasty race with release page
4295 * and why we unlock later
4299 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4301 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4305 spin_lock(&tree
->buffer_lock
);
4306 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
4307 if (ret
== -EEXIST
) {
4308 exists
= radix_tree_lookup(&tree
->buffer
,
4309 start
>> PAGE_CACHE_SHIFT
);
4310 if (!atomic_inc_not_zero(&exists
->refs
)) {
4311 spin_unlock(&tree
->buffer_lock
);
4312 radix_tree_preload_end();
4316 spin_unlock(&tree
->buffer_lock
);
4317 radix_tree_preload_end();
4318 mark_extent_buffer_accessed(exists
);
4321 /* add one reference for the tree */
4322 check_buffer_tree_ref(eb
);
4323 spin_unlock(&tree
->buffer_lock
);
4324 radix_tree_preload_end();
4327 * there is a race where release page may have
4328 * tried to find this extent buffer in the radix
4329 * but failed. It will tell the VM it is safe to
4330 * reclaim the, and it will clear the page private bit.
4331 * We must make sure to set the page private bit properly
4332 * after the extent buffer is in the radix tree so
4333 * it doesn't get lost
4335 SetPageChecked(eb
->pages
[0]);
4336 for (i
= 1; i
< num_pages
; i
++) {
4337 p
= extent_buffer_page(eb
, i
);
4338 ClearPageChecked(p
);
4341 unlock_page(eb
->pages
[0]);
4345 for (i
= 0; i
< num_pages
; i
++) {
4347 unlock_page(eb
->pages
[i
]);
4350 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4351 btrfs_release_extent_buffer(eb
);
4355 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
4356 u64 start
, unsigned long len
)
4358 struct extent_buffer
*eb
;
4361 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4362 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4364 mark_extent_buffer_accessed(eb
);
4372 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4374 struct extent_buffer
*eb
=
4375 container_of(head
, struct extent_buffer
, rcu_head
);
4377 __free_extent_buffer(eb
);
4380 /* Expects to have eb->eb_lock already held */
4381 static int release_extent_buffer(struct extent_buffer
*eb
, gfp_t mask
)
4383 WARN_ON(atomic_read(&eb
->refs
) == 0);
4384 if (atomic_dec_and_test(&eb
->refs
)) {
4385 if (test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
)) {
4386 spin_unlock(&eb
->refs_lock
);
4388 struct extent_io_tree
*tree
= eb
->tree
;
4390 spin_unlock(&eb
->refs_lock
);
4392 spin_lock(&tree
->buffer_lock
);
4393 radix_tree_delete(&tree
->buffer
,
4394 eb
->start
>> PAGE_CACHE_SHIFT
);
4395 spin_unlock(&tree
->buffer_lock
);
4398 /* Should be safe to release our pages at this point */
4399 btrfs_release_extent_buffer_page(eb
, 0);
4400 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4403 spin_unlock(&eb
->refs_lock
);
4408 void free_extent_buffer(struct extent_buffer
*eb
)
4413 spin_lock(&eb
->refs_lock
);
4414 if (atomic_read(&eb
->refs
) == 2 &&
4415 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4416 atomic_dec(&eb
->refs
);
4418 if (atomic_read(&eb
->refs
) == 2 &&
4419 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4420 !extent_buffer_under_io(eb
) &&
4421 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4422 atomic_dec(&eb
->refs
);
4425 * I know this is terrible, but it's temporary until we stop tracking
4426 * the uptodate bits and such for the extent buffers.
4428 release_extent_buffer(eb
, GFP_ATOMIC
);
4431 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4436 spin_lock(&eb
->refs_lock
);
4437 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4439 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4440 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4441 atomic_dec(&eb
->refs
);
4442 release_extent_buffer(eb
, GFP_NOFS
);
4445 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4448 unsigned long num_pages
;
4451 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4453 for (i
= 0; i
< num_pages
; i
++) {
4454 page
= extent_buffer_page(eb
, i
);
4455 if (!PageDirty(page
))
4459 WARN_ON(!PagePrivate(page
));
4461 clear_page_dirty_for_io(page
);
4462 spin_lock_irq(&page
->mapping
->tree_lock
);
4463 if (!PageDirty(page
)) {
4464 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4466 PAGECACHE_TAG_DIRTY
);
4468 spin_unlock_irq(&page
->mapping
->tree_lock
);
4469 ClearPageError(page
);
4472 WARN_ON(atomic_read(&eb
->refs
) == 0);
4475 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4478 unsigned long num_pages
;
4481 check_buffer_tree_ref(eb
);
4483 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4485 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4486 WARN_ON(atomic_read(&eb
->refs
) == 0);
4487 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4489 for (i
= 0; i
< num_pages
; i
++)
4490 set_page_dirty(extent_buffer_page(eb
, i
));
4494 static int range_straddles_pages(u64 start
, u64 len
)
4496 if (len
< PAGE_CACHE_SIZE
)
4498 if (start
& (PAGE_CACHE_SIZE
- 1))
4500 if ((start
+ len
) & (PAGE_CACHE_SIZE
- 1))
4505 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4509 unsigned long num_pages
;
4511 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4512 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4513 for (i
= 0; i
< num_pages
; i
++) {
4514 page
= extent_buffer_page(eb
, i
);
4516 ClearPageUptodate(page
);
4521 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4525 unsigned long num_pages
;
4527 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4528 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4529 for (i
= 0; i
< num_pages
; i
++) {
4530 page
= extent_buffer_page(eb
, i
);
4531 SetPageUptodate(page
);
4536 int extent_range_uptodate(struct extent_io_tree
*tree
,
4541 int pg_uptodate
= 1;
4543 unsigned long index
;
4545 if (range_straddles_pages(start
, end
- start
+ 1)) {
4546 ret
= test_range_bit(tree
, start
, end
,
4547 EXTENT_UPTODATE
, 1, NULL
);
4551 while (start
<= end
) {
4552 index
= start
>> PAGE_CACHE_SHIFT
;
4553 page
= find_get_page(tree
->mapping
, index
);
4556 uptodate
= PageUptodate(page
);
4557 page_cache_release(page
);
4562 start
+= PAGE_CACHE_SIZE
;
4567 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4569 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4572 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4573 struct extent_buffer
*eb
, u64 start
, int wait
,
4574 get_extent_t
*get_extent
, int mirror_num
)
4577 unsigned long start_i
;
4581 int locked_pages
= 0;
4582 int all_uptodate
= 1;
4583 unsigned long num_pages
;
4584 unsigned long num_reads
= 0;
4585 struct bio
*bio
= NULL
;
4586 unsigned long bio_flags
= 0;
4588 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4592 WARN_ON(start
< eb
->start
);
4593 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4594 (eb
->start
>> PAGE_CACHE_SHIFT
);
4599 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4600 for (i
= start_i
; i
< num_pages
; i
++) {
4601 page
= extent_buffer_page(eb
, i
);
4602 if (wait
== WAIT_NONE
) {
4603 if (!trylock_page(page
))
4609 if (!PageUptodate(page
)) {
4616 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4620 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4621 eb
->read_mirror
= 0;
4622 atomic_set(&eb
->io_pages
, num_reads
);
4623 for (i
= start_i
; i
< num_pages
; i
++) {
4624 page
= extent_buffer_page(eb
, i
);
4625 if (!PageUptodate(page
)) {
4626 ClearPageError(page
);
4627 err
= __extent_read_full_page(tree
, page
,
4629 mirror_num
, &bio_flags
);
4638 err
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
4643 if (ret
|| wait
!= WAIT_COMPLETE
)
4646 for (i
= start_i
; i
< num_pages
; i
++) {
4647 page
= extent_buffer_page(eb
, i
);
4648 wait_on_page_locked(page
);
4649 if (!PageUptodate(page
))
4657 while (locked_pages
> 0) {
4658 page
= extent_buffer_page(eb
, i
);
4666 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
4667 unsigned long start
,
4674 char *dst
= (char *)dstv
;
4675 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4676 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4678 WARN_ON(start
> eb
->len
);
4679 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4681 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4684 page
= extent_buffer_page(eb
, i
);
4686 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4687 kaddr
= page_address(page
);
4688 memcpy(dst
, kaddr
+ offset
, cur
);
4697 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
4698 unsigned long min_len
, char **map
,
4699 unsigned long *map_start
,
4700 unsigned long *map_len
)
4702 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
4705 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4706 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4707 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
4714 offset
= start_offset
;
4718 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
4721 if (start
+ min_len
> eb
->len
) {
4722 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
4723 "wanted %lu %lu\n", (unsigned long long)eb
->start
,
4724 eb
->len
, start
, min_len
);
4728 p
= extent_buffer_page(eb
, i
);
4729 kaddr
= page_address(p
);
4730 *map
= kaddr
+ offset
;
4731 *map_len
= PAGE_CACHE_SIZE
- offset
;
4735 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
4736 unsigned long start
,
4743 char *ptr
= (char *)ptrv
;
4744 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4745 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4748 WARN_ON(start
> eb
->len
);
4749 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4751 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4754 page
= extent_buffer_page(eb
, i
);
4756 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4758 kaddr
= page_address(page
);
4759 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
4771 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
4772 unsigned long start
, unsigned long len
)
4778 char *src
= (char *)srcv
;
4779 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4780 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4782 WARN_ON(start
> eb
->len
);
4783 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4785 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4788 page
= extent_buffer_page(eb
, i
);
4789 WARN_ON(!PageUptodate(page
));
4791 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4792 kaddr
= page_address(page
);
4793 memcpy(kaddr
+ offset
, src
, cur
);
4802 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
4803 unsigned long start
, unsigned long len
)
4809 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4810 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4812 WARN_ON(start
> eb
->len
);
4813 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4815 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4818 page
= extent_buffer_page(eb
, i
);
4819 WARN_ON(!PageUptodate(page
));
4821 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4822 kaddr
= page_address(page
);
4823 memset(kaddr
+ offset
, c
, cur
);
4831 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
4832 unsigned long dst_offset
, unsigned long src_offset
,
4835 u64 dst_len
= dst
->len
;
4840 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4841 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
4843 WARN_ON(src
->len
!= dst_len
);
4845 offset
= (start_offset
+ dst_offset
) &
4846 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4849 page
= extent_buffer_page(dst
, i
);
4850 WARN_ON(!PageUptodate(page
));
4852 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
4854 kaddr
= page_address(page
);
4855 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
4864 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
4865 unsigned long dst_off
, unsigned long src_off
,
4868 char *dst_kaddr
= page_address(dst_page
);
4869 if (dst_page
== src_page
) {
4870 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
4872 char *src_kaddr
= page_address(src_page
);
4873 char *p
= dst_kaddr
+ dst_off
+ len
;
4874 char *s
= src_kaddr
+ src_off
+ len
;
4881 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
4883 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
4884 return distance
< len
;
4887 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
4888 unsigned long dst_off
, unsigned long src_off
,
4891 char *dst_kaddr
= page_address(dst_page
);
4893 int must_memmove
= 0;
4895 if (dst_page
!= src_page
) {
4896 src_kaddr
= page_address(src_page
);
4898 src_kaddr
= dst_kaddr
;
4899 if (areas_overlap(src_off
, dst_off
, len
))
4904 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
4906 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
4909 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
4910 unsigned long src_offset
, unsigned long len
)
4913 size_t dst_off_in_page
;
4914 size_t src_off_in_page
;
4915 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4916 unsigned long dst_i
;
4917 unsigned long src_i
;
4919 if (src_offset
+ len
> dst
->len
) {
4920 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
4921 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
4924 if (dst_offset
+ len
> dst
->len
) {
4925 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
4926 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
4931 dst_off_in_page
= (start_offset
+ dst_offset
) &
4932 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4933 src_off_in_page
= (start_offset
+ src_offset
) &
4934 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4936 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
4937 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
4939 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
4941 cur
= min_t(unsigned long, cur
,
4942 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
4944 copy_pages(extent_buffer_page(dst
, dst_i
),
4945 extent_buffer_page(dst
, src_i
),
4946 dst_off_in_page
, src_off_in_page
, cur
);
4954 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
4955 unsigned long src_offset
, unsigned long len
)
4958 size_t dst_off_in_page
;
4959 size_t src_off_in_page
;
4960 unsigned long dst_end
= dst_offset
+ len
- 1;
4961 unsigned long src_end
= src_offset
+ len
- 1;
4962 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4963 unsigned long dst_i
;
4964 unsigned long src_i
;
4966 if (src_offset
+ len
> dst
->len
) {
4967 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
4968 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
4971 if (dst_offset
+ len
> dst
->len
) {
4972 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
4973 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
4976 if (dst_offset
< src_offset
) {
4977 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
4981 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
4982 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
4984 dst_off_in_page
= (start_offset
+ dst_end
) &
4985 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4986 src_off_in_page
= (start_offset
+ src_end
) &
4987 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4989 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
4990 cur
= min(cur
, dst_off_in_page
+ 1);
4991 move_pages(extent_buffer_page(dst
, dst_i
),
4992 extent_buffer_page(dst
, src_i
),
4993 dst_off_in_page
- cur
+ 1,
4994 src_off_in_page
- cur
+ 1, cur
);
5002 int try_release_extent_buffer(struct page
*page
, gfp_t mask
)
5004 struct extent_buffer
*eb
;
5007 * We need to make sure noboody is attaching this page to an eb right
5010 spin_lock(&page
->mapping
->private_lock
);
5011 if (!PagePrivate(page
)) {
5012 spin_unlock(&page
->mapping
->private_lock
);
5016 eb
= (struct extent_buffer
*)page
->private;
5020 * This is a little awful but should be ok, we need to make sure that
5021 * the eb doesn't disappear out from under us while we're looking at
5024 spin_lock(&eb
->refs_lock
);
5025 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5026 spin_unlock(&eb
->refs_lock
);
5027 spin_unlock(&page
->mapping
->private_lock
);
5030 spin_unlock(&page
->mapping
->private_lock
);
5032 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
5036 * If tree ref isn't set then we know the ref on this eb is a real ref,
5037 * so just return, this page will likely be freed soon anyway.
5039 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5040 spin_unlock(&eb
->refs_lock
);
5044 return release_extent_buffer(eb
, mask
);