1 #include <linux/bitops.h>
2 #include <linux/slab.h>
6 #include <linux/pagemap.h>
7 #include <linux/page-flags.h>
8 #include <linux/module.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 /* temporary define until extent_map moves out of btrfs */
21 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
22 unsigned long extra_flags
,
23 void (*ctor
)(void *, struct kmem_cache
*,
26 static struct kmem_cache
*extent_state_cache
;
27 static struct kmem_cache
*extent_buffer_cache
;
29 static LIST_HEAD(buffers
);
30 static LIST_HEAD(states
);
34 static DEFINE_SPINLOCK(leak_lock
);
37 #define BUFFER_LRU_MAX 64
42 struct rb_node rb_node
;
45 struct extent_page_data
{
47 struct extent_io_tree
*tree
;
48 get_extent_t
*get_extent
;
50 /* tells writepage not to lock the state bits for this range
51 * it still does the unlocking
56 int __init
extent_io_init(void)
58 extent_state_cache
= btrfs_cache_create("extent_state",
59 sizeof(struct extent_state
), 0,
61 if (!extent_state_cache
)
64 extent_buffer_cache
= btrfs_cache_create("extent_buffers",
65 sizeof(struct extent_buffer
), 0,
67 if (!extent_buffer_cache
)
68 goto free_state_cache
;
72 kmem_cache_destroy(extent_state_cache
);
76 void extent_io_exit(void)
78 struct extent_state
*state
;
79 struct extent_buffer
*eb
;
81 while (!list_empty(&states
)) {
82 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
83 printk(KERN_ERR
"btrfs state leak: start %llu end %llu "
84 "state %lu in tree %p refs %d\n",
85 (unsigned long long)state
->start
,
86 (unsigned long long)state
->end
,
87 state
->state
, state
->tree
, atomic_read(&state
->refs
));
88 list_del(&state
->leak_list
);
89 kmem_cache_free(extent_state_cache
, state
);
93 while (!list_empty(&buffers
)) {
94 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
95 printk(KERN_ERR
"btrfs buffer leak start %llu len %lu "
96 "refs %d\n", (unsigned long long)eb
->start
,
97 eb
->len
, atomic_read(&eb
->refs
));
98 list_del(&eb
->leak_list
);
99 kmem_cache_free(extent_buffer_cache
, eb
);
101 if (extent_state_cache
)
102 kmem_cache_destroy(extent_state_cache
);
103 if (extent_buffer_cache
)
104 kmem_cache_destroy(extent_buffer_cache
);
107 void extent_io_tree_init(struct extent_io_tree
*tree
,
108 struct address_space
*mapping
, gfp_t mask
)
110 tree
->state
.rb_node
= NULL
;
111 tree
->buffer
.rb_node
= NULL
;
113 tree
->dirty_bytes
= 0;
114 spin_lock_init(&tree
->lock
);
115 spin_lock_init(&tree
->buffer_lock
);
116 tree
->mapping
= mapping
;
119 static struct extent_state
*alloc_extent_state(gfp_t mask
)
121 struct extent_state
*state
;
126 state
= kmem_cache_alloc(extent_state_cache
, mask
);
133 spin_lock_irqsave(&leak_lock
, flags
);
134 list_add(&state
->leak_list
, &states
);
135 spin_unlock_irqrestore(&leak_lock
, flags
);
137 atomic_set(&state
->refs
, 1);
138 init_waitqueue_head(&state
->wq
);
142 static void free_extent_state(struct extent_state
*state
)
146 if (atomic_dec_and_test(&state
->refs
)) {
150 WARN_ON(state
->tree
);
152 spin_lock_irqsave(&leak_lock
, flags
);
153 list_del(&state
->leak_list
);
154 spin_unlock_irqrestore(&leak_lock
, flags
);
156 kmem_cache_free(extent_state_cache
, state
);
160 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
161 struct rb_node
*node
)
163 struct rb_node
**p
= &root
->rb_node
;
164 struct rb_node
*parent
= NULL
;
165 struct tree_entry
*entry
;
169 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
171 if (offset
< entry
->start
)
173 else if (offset
> entry
->end
)
179 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
180 rb_link_node(node
, parent
, p
);
181 rb_insert_color(node
, root
);
185 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
186 struct rb_node
**prev_ret
,
187 struct rb_node
**next_ret
)
189 struct rb_root
*root
= &tree
->state
;
190 struct rb_node
*n
= root
->rb_node
;
191 struct rb_node
*prev
= NULL
;
192 struct rb_node
*orig_prev
= NULL
;
193 struct tree_entry
*entry
;
194 struct tree_entry
*prev_entry
= NULL
;
197 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
201 if (offset
< entry
->start
)
203 else if (offset
> entry
->end
)
211 while (prev
&& offset
> prev_entry
->end
) {
212 prev
= rb_next(prev
);
213 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
220 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
221 while (prev
&& offset
< prev_entry
->start
) {
222 prev
= rb_prev(prev
);
223 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
230 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
233 struct rb_node
*prev
= NULL
;
236 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
242 static struct extent_buffer
*buffer_tree_insert(struct extent_io_tree
*tree
,
243 u64 offset
, struct rb_node
*node
)
245 struct rb_root
*root
= &tree
->buffer
;
246 struct rb_node
**p
= &root
->rb_node
;
247 struct rb_node
*parent
= NULL
;
248 struct extent_buffer
*eb
;
252 eb
= rb_entry(parent
, struct extent_buffer
, rb_node
);
254 if (offset
< eb
->start
)
256 else if (offset
> eb
->start
)
262 rb_link_node(node
, parent
, p
);
263 rb_insert_color(node
, root
);
267 static struct extent_buffer
*buffer_search(struct extent_io_tree
*tree
,
270 struct rb_root
*root
= &tree
->buffer
;
271 struct rb_node
*n
= root
->rb_node
;
272 struct extent_buffer
*eb
;
275 eb
= rb_entry(n
, struct extent_buffer
, rb_node
);
276 if (offset
< eb
->start
)
278 else if (offset
> eb
->start
)
287 * utility function to look for merge candidates inside a given range.
288 * Any extents with matching state are merged together into a single
289 * extent in the tree. Extents with EXTENT_IO in their state field
290 * are not merged because the end_io handlers need to be able to do
291 * operations on them without sleeping (or doing allocations/splits).
293 * This should be called with the tree lock held.
295 static int merge_state(struct extent_io_tree
*tree
,
296 struct extent_state
*state
)
298 struct extent_state
*other
;
299 struct rb_node
*other_node
;
301 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
304 other_node
= rb_prev(&state
->rb_node
);
306 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
307 if (other
->end
== state
->start
- 1 &&
308 other
->state
== state
->state
) {
309 state
->start
= other
->start
;
311 rb_erase(&other
->rb_node
, &tree
->state
);
312 free_extent_state(other
);
315 other_node
= rb_next(&state
->rb_node
);
317 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
318 if (other
->start
== state
->end
+ 1 &&
319 other
->state
== state
->state
) {
320 other
->start
= state
->start
;
322 rb_erase(&state
->rb_node
, &tree
->state
);
323 free_extent_state(state
);
329 static void set_state_cb(struct extent_io_tree
*tree
,
330 struct extent_state
*state
,
333 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
334 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
->start
,
335 state
->end
, state
->state
, bits
);
339 static void clear_state_cb(struct extent_io_tree
*tree
,
340 struct extent_state
*state
,
343 if (tree
->ops
&& tree
->ops
->clear_bit_hook
) {
344 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
->start
,
345 state
->end
, state
->state
, bits
);
350 * insert an extent_state struct into the tree. 'bits' are set on the
351 * struct before it is inserted.
353 * This may return -EEXIST if the extent is already there, in which case the
354 * state struct is freed.
356 * The tree lock is not taken internally. This is a utility function and
357 * probably isn't what you want to call (see set/clear_extent_bit).
359 static int insert_state(struct extent_io_tree
*tree
,
360 struct extent_state
*state
, u64 start
, u64 end
,
363 struct rb_node
*node
;
366 printk(KERN_ERR
"btrfs end < start %llu %llu\n",
367 (unsigned long long)end
,
368 (unsigned long long)start
);
371 if (bits
& EXTENT_DIRTY
)
372 tree
->dirty_bytes
+= end
- start
+ 1;
373 set_state_cb(tree
, state
, bits
);
374 state
->state
|= bits
;
375 state
->start
= start
;
377 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
379 struct extent_state
*found
;
380 found
= rb_entry(node
, struct extent_state
, rb_node
);
381 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
382 "%llu %llu\n", (unsigned long long)found
->start
,
383 (unsigned long long)found
->end
,
384 (unsigned long long)start
, (unsigned long long)end
);
385 free_extent_state(state
);
389 merge_state(tree
, state
);
394 * split a given extent state struct in two, inserting the preallocated
395 * struct 'prealloc' as the newly created second half. 'split' indicates an
396 * offset inside 'orig' where it should be split.
399 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
400 * are two extent state structs in the tree:
401 * prealloc: [orig->start, split - 1]
402 * orig: [ split, orig->end ]
404 * The tree locks are not taken by this function. They need to be held
407 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
408 struct extent_state
*prealloc
, u64 split
)
410 struct rb_node
*node
;
411 prealloc
->start
= orig
->start
;
412 prealloc
->end
= split
- 1;
413 prealloc
->state
= orig
->state
;
416 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
418 struct extent_state
*found
;
419 found
= rb_entry(node
, struct extent_state
, rb_node
);
420 free_extent_state(prealloc
);
423 prealloc
->tree
= tree
;
428 * utility function to clear some bits in an extent state struct.
429 * it will optionally wake up any one waiting on this state (wake == 1), or
430 * forcibly remove the state from the tree (delete == 1).
432 * If no bits are set on the state struct after clearing things, the
433 * struct is freed and removed from the tree
435 static int clear_state_bit(struct extent_io_tree
*tree
,
436 struct extent_state
*state
, int bits
, int wake
,
439 int ret
= state
->state
& bits
;
441 if ((bits
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
442 u64 range
= state
->end
- state
->start
+ 1;
443 WARN_ON(range
> tree
->dirty_bytes
);
444 tree
->dirty_bytes
-= range
;
446 clear_state_cb(tree
, state
, bits
);
447 state
->state
&= ~bits
;
450 if (delete || state
->state
== 0) {
452 clear_state_cb(tree
, state
, state
->state
);
453 rb_erase(&state
->rb_node
, &tree
->state
);
455 free_extent_state(state
);
460 merge_state(tree
, state
);
466 * clear some bits on a range in the tree. This may require splitting
467 * or inserting elements in the tree, so the gfp mask is used to
468 * indicate which allocations or sleeping are allowed.
470 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
471 * the given range from the tree regardless of state (ie for truncate).
473 * the range [start, end] is inclusive.
475 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
476 * bits were already set, or zero if none of the bits were already set.
478 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
479 int bits
, int wake
, int delete, gfp_t mask
)
481 struct extent_state
*state
;
482 struct extent_state
*prealloc
= NULL
;
483 struct rb_node
*node
;
488 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
489 prealloc
= alloc_extent_state(mask
);
494 spin_lock(&tree
->lock
);
496 * this search will find the extents that end after
499 node
= tree_search(tree
, start
);
502 state
= rb_entry(node
, struct extent_state
, rb_node
);
503 if (state
->start
> end
)
505 WARN_ON(state
->end
< start
);
508 * | ---- desired range ---- |
510 * | ------------- state -------------- |
512 * We need to split the extent we found, and may flip
513 * bits on second half.
515 * If the extent we found extends past our range, we
516 * just split and search again. It'll get split again
517 * the next time though.
519 * If the extent we found is inside our range, we clear
520 * the desired bit on it.
523 if (state
->start
< start
) {
525 prealloc
= alloc_extent_state(GFP_ATOMIC
);
526 err
= split_state(tree
, state
, prealloc
, start
);
527 BUG_ON(err
== -EEXIST
);
531 if (state
->end
<= end
) {
532 start
= state
->end
+ 1;
533 set
|= clear_state_bit(tree
, state
, bits
,
536 start
= state
->start
;
541 * | ---- desired range ---- |
543 * We need to split the extent, and clear the bit
546 if (state
->start
<= end
&& state
->end
> end
) {
548 prealloc
= alloc_extent_state(GFP_ATOMIC
);
549 err
= split_state(tree
, state
, prealloc
, end
+ 1);
550 BUG_ON(err
== -EEXIST
);
554 set
|= clear_state_bit(tree
, prealloc
, bits
,
560 start
= state
->end
+ 1;
561 set
|= clear_state_bit(tree
, state
, bits
, wake
, delete);
565 spin_unlock(&tree
->lock
);
567 free_extent_state(prealloc
);
574 spin_unlock(&tree
->lock
);
575 if (mask
& __GFP_WAIT
)
580 static int wait_on_state(struct extent_io_tree
*tree
,
581 struct extent_state
*state
)
582 __releases(tree
->lock
)
583 __acquires(tree
->lock
)
586 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
587 spin_unlock(&tree
->lock
);
589 spin_lock(&tree
->lock
);
590 finish_wait(&state
->wq
, &wait
);
595 * waits for one or more bits to clear on a range in the state tree.
596 * The range [start, end] is inclusive.
597 * The tree lock is taken by this function
599 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
601 struct extent_state
*state
;
602 struct rb_node
*node
;
604 spin_lock(&tree
->lock
);
608 * this search will find all the extents that end after
611 node
= tree_search(tree
, start
);
615 state
= rb_entry(node
, struct extent_state
, rb_node
);
617 if (state
->start
> end
)
620 if (state
->state
& bits
) {
621 start
= state
->start
;
622 atomic_inc(&state
->refs
);
623 wait_on_state(tree
, state
);
624 free_extent_state(state
);
627 start
= state
->end
+ 1;
632 if (need_resched()) {
633 spin_unlock(&tree
->lock
);
635 spin_lock(&tree
->lock
);
639 spin_unlock(&tree
->lock
);
643 static void set_state_bits(struct extent_io_tree
*tree
,
644 struct extent_state
*state
,
647 if ((bits
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
648 u64 range
= state
->end
- state
->start
+ 1;
649 tree
->dirty_bytes
+= range
;
651 set_state_cb(tree
, state
, bits
);
652 state
->state
|= bits
;
656 * set some bits on a range in the tree. This may require allocations
657 * or sleeping, so the gfp mask is used to indicate what is allowed.
659 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
660 * range already has the desired bits set. The start of the existing
661 * range is returned in failed_start in this case.
663 * [start, end] is inclusive
664 * This takes the tree lock.
666 static int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
667 int bits
, int exclusive
, u64
*failed_start
,
670 struct extent_state
*state
;
671 struct extent_state
*prealloc
= NULL
;
672 struct rb_node
*node
;
678 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
679 prealloc
= alloc_extent_state(mask
);
684 spin_lock(&tree
->lock
);
686 * this search will find all the extents that end after
689 node
= tree_search(tree
, start
);
691 err
= insert_state(tree
, prealloc
, start
, end
, bits
);
693 BUG_ON(err
== -EEXIST
);
697 state
= rb_entry(node
, struct extent_state
, rb_node
);
698 last_start
= state
->start
;
699 last_end
= state
->end
;
702 * | ---- desired range ---- |
705 * Just lock what we found and keep going
707 if (state
->start
== start
&& state
->end
<= end
) {
708 set
= state
->state
& bits
;
709 if (set
&& exclusive
) {
710 *failed_start
= state
->start
;
714 set_state_bits(tree
, state
, bits
);
715 start
= state
->end
+ 1;
716 merge_state(tree
, state
);
721 * | ---- desired range ---- |
724 * | ------------- state -------------- |
726 * We need to split the extent we found, and may flip bits on
729 * If the extent we found extends past our
730 * range, we just split and search again. It'll get split
731 * again the next time though.
733 * If the extent we found is inside our range, we set the
736 if (state
->start
< start
) {
737 set
= state
->state
& bits
;
738 if (exclusive
&& set
) {
739 *failed_start
= start
;
743 err
= split_state(tree
, state
, prealloc
, start
);
744 BUG_ON(err
== -EEXIST
);
748 if (state
->end
<= end
) {
749 set_state_bits(tree
, state
, bits
);
750 start
= state
->end
+ 1;
751 merge_state(tree
, state
);
753 start
= state
->start
;
758 * | ---- desired range ---- |
759 * | state | or | state |
761 * There's a hole, we need to insert something in it and
762 * ignore the extent we found.
764 if (state
->start
> start
) {
766 if (end
< last_start
)
769 this_end
= last_start
- 1;
770 err
= insert_state(tree
, prealloc
, start
, this_end
,
773 BUG_ON(err
== -EEXIST
);
776 start
= this_end
+ 1;
780 * | ---- desired range ---- |
782 * We need to split the extent, and set the bit
785 if (state
->start
<= end
&& state
->end
> end
) {
786 set
= state
->state
& bits
;
787 if (exclusive
&& set
) {
788 *failed_start
= start
;
792 err
= split_state(tree
, state
, prealloc
, end
+ 1);
793 BUG_ON(err
== -EEXIST
);
795 set_state_bits(tree
, prealloc
, bits
);
796 merge_state(tree
, prealloc
);
804 spin_unlock(&tree
->lock
);
806 free_extent_state(prealloc
);
813 spin_unlock(&tree
->lock
);
814 if (mask
& __GFP_WAIT
)
819 /* wrappers around set/clear extent bit */
820 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
823 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
827 int set_extent_ordered(struct extent_io_tree
*tree
, u64 start
, u64 end
,
830 return set_extent_bit(tree
, start
, end
, EXTENT_ORDERED
, 0, NULL
, mask
);
833 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
834 int bits
, gfp_t mask
)
836 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
840 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
841 int bits
, gfp_t mask
)
843 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, mask
);
846 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
849 return set_extent_bit(tree
, start
, end
,
850 EXTENT_DELALLOC
| EXTENT_DIRTY
,
854 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
857 return clear_extent_bit(tree
, start
, end
,
858 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, mask
);
861 int clear_extent_ordered(struct extent_io_tree
*tree
, u64 start
, u64 end
,
864 return clear_extent_bit(tree
, start
, end
, EXTENT_ORDERED
, 1, 0, mask
);
867 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
870 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
874 static int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
877 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0, mask
);
880 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
883 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, NULL
,
887 static int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
,
890 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0, mask
);
893 static int set_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
896 return set_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
,
900 static int clear_extent_writeback(struct extent_io_tree
*tree
, u64 start
,
903 return clear_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 1, 0, mask
);
906 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
908 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
912 * either insert or lock state struct between start and end use mask to tell
913 * us if waiting is desired.
915 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
920 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
921 &failed_start
, mask
);
922 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
923 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
924 start
= failed_start
;
928 WARN_ON(start
> end
);
933 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
939 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
940 &failed_start
, mask
);
941 if (err
== -EEXIST
) {
942 if (failed_start
> start
)
943 clear_extent_bit(tree
, start
, failed_start
- 1,
944 EXTENT_LOCKED
, 1, 0, mask
);
950 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
953 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, mask
);
957 * helper function to set pages and extents in the tree dirty
959 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
961 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
962 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
965 while (index
<= end_index
) {
966 page
= find_get_page(tree
->mapping
, index
);
968 __set_page_dirty_nobuffers(page
);
969 page_cache_release(page
);
972 set_extent_dirty(tree
, start
, end
, GFP_NOFS
);
977 * helper function to set both pages and extents in the tree writeback
979 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
981 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
982 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
985 while (index
<= end_index
) {
986 page
= find_get_page(tree
->mapping
, index
);
988 set_page_writeback(page
);
989 page_cache_release(page
);
992 set_extent_writeback(tree
, start
, end
, GFP_NOFS
);
997 * find the first offset in the io tree with 'bits' set. zero is
998 * returned if we find something, and *start_ret and *end_ret are
999 * set to reflect the state struct that was found.
1001 * If nothing was found, 1 is returned, < 0 on error
1003 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1004 u64
*start_ret
, u64
*end_ret
, int bits
)
1006 struct rb_node
*node
;
1007 struct extent_state
*state
;
1010 spin_lock(&tree
->lock
);
1012 * this search will find all the extents that end after
1015 node
= tree_search(tree
, start
);
1020 state
= rb_entry(node
, struct extent_state
, rb_node
);
1021 if (state
->end
>= start
&& (state
->state
& bits
)) {
1022 *start_ret
= state
->start
;
1023 *end_ret
= state
->end
;
1027 node
= rb_next(node
);
1032 spin_unlock(&tree
->lock
);
1036 /* find the first state struct with 'bits' set after 'start', and
1037 * return it. tree->lock must be held. NULL will returned if
1038 * nothing was found after 'start'
1040 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
1041 u64 start
, int bits
)
1043 struct rb_node
*node
;
1044 struct extent_state
*state
;
1047 * this search will find all the extents that end after
1050 node
= tree_search(tree
, start
);
1055 state
= rb_entry(node
, struct extent_state
, rb_node
);
1056 if (state
->end
>= start
&& (state
->state
& bits
))
1059 node
= rb_next(node
);
1068 * find a contiguous range of bytes in the file marked as delalloc, not
1069 * more than 'max_bytes'. start and end are used to return the range,
1071 * 1 is returned if we find something, 0 if nothing was in the tree
1073 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1074 u64
*start
, u64
*end
, u64 max_bytes
)
1076 struct rb_node
*node
;
1077 struct extent_state
*state
;
1078 u64 cur_start
= *start
;
1080 u64 total_bytes
= 0;
1082 spin_lock(&tree
->lock
);
1085 * this search will find all the extents that end after
1088 node
= tree_search(tree
, cur_start
);
1096 state
= rb_entry(node
, struct extent_state
, rb_node
);
1097 if (found
&& (state
->start
!= cur_start
||
1098 (state
->state
& EXTENT_BOUNDARY
))) {
1101 if (!(state
->state
& EXTENT_DELALLOC
)) {
1107 *start
= state
->start
;
1110 cur_start
= state
->end
+ 1;
1111 node
= rb_next(node
);
1114 total_bytes
+= state
->end
- state
->start
+ 1;
1115 if (total_bytes
>= max_bytes
)
1119 spin_unlock(&tree
->lock
);
1123 static noinline
int __unlock_for_delalloc(struct inode
*inode
,
1124 struct page
*locked_page
,
1128 struct page
*pages
[16];
1129 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1130 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1131 unsigned long nr_pages
= end_index
- index
+ 1;
1134 if (index
== locked_page
->index
&& end_index
== index
)
1137 while (nr_pages
> 0) {
1138 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1139 min_t(unsigned long, nr_pages
,
1140 ARRAY_SIZE(pages
)), pages
);
1141 for (i
= 0; i
< ret
; i
++) {
1142 if (pages
[i
] != locked_page
)
1143 unlock_page(pages
[i
]);
1144 page_cache_release(pages
[i
]);
1153 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1154 struct page
*locked_page
,
1158 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1159 unsigned long start_index
= index
;
1160 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1161 unsigned long pages_locked
= 0;
1162 struct page
*pages
[16];
1163 unsigned long nrpages
;
1167 /* the caller is responsible for locking the start index */
1168 if (index
== locked_page
->index
&& index
== end_index
)
1171 /* skip the page at the start index */
1172 nrpages
= end_index
- index
+ 1;
1173 while (nrpages
> 0) {
1174 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1175 min_t(unsigned long,
1176 nrpages
, ARRAY_SIZE(pages
)), pages
);
1181 /* now we have an array of pages, lock them all */
1182 for (i
= 0; i
< ret
; i
++) {
1184 * the caller is taking responsibility for
1187 if (pages
[i
] != locked_page
) {
1188 lock_page(pages
[i
]);
1189 if (!PageDirty(pages
[i
]) ||
1190 pages
[i
]->mapping
!= inode
->i_mapping
) {
1192 unlock_page(pages
[i
]);
1193 page_cache_release(pages
[i
]);
1197 page_cache_release(pages
[i
]);
1206 if (ret
&& pages_locked
) {
1207 __unlock_for_delalloc(inode
, locked_page
,
1209 ((u64
)(start_index
+ pages_locked
- 1)) <<
1216 * find a contiguous range of bytes in the file marked as delalloc, not
1217 * more than 'max_bytes'. start and end are used to return the range,
1219 * 1 is returned if we find something, 0 if nothing was in the tree
1221 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1222 struct extent_io_tree
*tree
,
1223 struct page
*locked_page
,
1224 u64
*start
, u64
*end
,
1234 /* step one, find a bunch of delalloc bytes starting at start */
1235 delalloc_start
= *start
;
1237 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1239 if (!found
|| delalloc_end
<= *start
) {
1240 *start
= delalloc_start
;
1241 *end
= delalloc_end
;
1246 * start comes from the offset of locked_page. We have to lock
1247 * pages in order, so we can't process delalloc bytes before
1250 if (delalloc_start
< *start
)
1251 delalloc_start
= *start
;
1254 * make sure to limit the number of pages we try to lock down
1257 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
)
1258 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1260 /* step two, lock all the pages after the page that has start */
1261 ret
= lock_delalloc_pages(inode
, locked_page
,
1262 delalloc_start
, delalloc_end
);
1263 if (ret
== -EAGAIN
) {
1264 /* some of the pages are gone, lets avoid looping by
1265 * shortening the size of the delalloc range we're searching
1268 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1269 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1279 /* step three, lock the state bits for the whole range */
1280 lock_extent(tree
, delalloc_start
, delalloc_end
, GFP_NOFS
);
1282 /* then test to make sure it is all still delalloc */
1283 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1284 EXTENT_DELALLOC
, 1);
1286 unlock_extent(tree
, delalloc_start
, delalloc_end
, GFP_NOFS
);
1287 __unlock_for_delalloc(inode
, locked_page
,
1288 delalloc_start
, delalloc_end
);
1292 *start
= delalloc_start
;
1293 *end
= delalloc_end
;
1298 int extent_clear_unlock_delalloc(struct inode
*inode
,
1299 struct extent_io_tree
*tree
,
1300 u64 start
, u64 end
, struct page
*locked_page
,
1303 int clear_delalloc
, int clear_dirty
,
1308 struct page
*pages
[16];
1309 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1310 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1311 unsigned long nr_pages
= end_index
- index
+ 1;
1316 clear_bits
|= EXTENT_LOCKED
;
1318 clear_bits
|= EXTENT_DIRTY
;
1321 clear_bits
|= EXTENT_DELALLOC
;
1323 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, GFP_NOFS
);
1324 if (!(unlock_pages
|| clear_dirty
|| set_writeback
|| end_writeback
))
1327 while (nr_pages
> 0) {
1328 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1329 min_t(unsigned long,
1330 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1331 for (i
= 0; i
< ret
; i
++) {
1332 if (pages
[i
] == locked_page
) {
1333 page_cache_release(pages
[i
]);
1337 clear_page_dirty_for_io(pages
[i
]);
1339 set_page_writeback(pages
[i
]);
1341 end_page_writeback(pages
[i
]);
1343 unlock_page(pages
[i
]);
1344 page_cache_release(pages
[i
]);
1354 * count the number of bytes in the tree that have a given bit(s)
1355 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1356 * cached. The total number found is returned.
1358 u64
count_range_bits(struct extent_io_tree
*tree
,
1359 u64
*start
, u64 search_end
, u64 max_bytes
,
1362 struct rb_node
*node
;
1363 struct extent_state
*state
;
1364 u64 cur_start
= *start
;
1365 u64 total_bytes
= 0;
1368 if (search_end
<= cur_start
) {
1373 spin_lock(&tree
->lock
);
1374 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1375 total_bytes
= tree
->dirty_bytes
;
1379 * this search will find all the extents that end after
1382 node
= tree_search(tree
, cur_start
);
1387 state
= rb_entry(node
, struct extent_state
, rb_node
);
1388 if (state
->start
> search_end
)
1390 if (state
->end
>= cur_start
&& (state
->state
& bits
)) {
1391 total_bytes
+= min(search_end
, state
->end
) + 1 -
1392 max(cur_start
, state
->start
);
1393 if (total_bytes
>= max_bytes
)
1396 *start
= state
->start
;
1400 node
= rb_next(node
);
1405 spin_unlock(&tree
->lock
);
1411 * helper function to lock both pages and extents in the tree.
1412 * pages must be locked first.
1414 static int lock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1416 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1417 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1421 while (index
<= end_index
) {
1422 page
= grab_cache_page(tree
->mapping
, index
);
1428 err
= PTR_ERR(page
);
1433 lock_extent(tree
, start
, end
, GFP_NOFS
);
1438 * we failed above in getting the page at 'index', so we undo here
1439 * up to but not including the page at 'index'
1442 index
= start
>> PAGE_CACHE_SHIFT
;
1443 while (index
< end_index
) {
1444 page
= find_get_page(tree
->mapping
, index
);
1446 page_cache_release(page
);
1453 * helper function to unlock both pages and extents in the tree.
1455 static int unlock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1457 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1458 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1461 while (index
<= end_index
) {
1462 page
= find_get_page(tree
->mapping
, index
);
1464 page_cache_release(page
);
1467 unlock_extent(tree
, start
, end
, GFP_NOFS
);
1473 * set the private field for a given byte offset in the tree. If there isn't
1474 * an extent_state there already, this does nothing.
1476 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1478 struct rb_node
*node
;
1479 struct extent_state
*state
;
1482 spin_lock(&tree
->lock
);
1484 * this search will find all the extents that end after
1487 node
= tree_search(tree
, start
);
1492 state
= rb_entry(node
, struct extent_state
, rb_node
);
1493 if (state
->start
!= start
) {
1497 state
->private = private;
1499 spin_unlock(&tree
->lock
);
1503 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1505 struct rb_node
*node
;
1506 struct extent_state
*state
;
1509 spin_lock(&tree
->lock
);
1511 * this search will find all the extents that end after
1514 node
= tree_search(tree
, start
);
1519 state
= rb_entry(node
, struct extent_state
, rb_node
);
1520 if (state
->start
!= start
) {
1524 *private = state
->private;
1526 spin_unlock(&tree
->lock
);
1531 * searches a range in the state tree for a given mask.
1532 * If 'filled' == 1, this returns 1 only if every extent in the tree
1533 * has the bits set. Otherwise, 1 is returned if any bit in the
1534 * range is found set.
1536 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1537 int bits
, int filled
)
1539 struct extent_state
*state
= NULL
;
1540 struct rb_node
*node
;
1543 spin_lock(&tree
->lock
);
1544 node
= tree_search(tree
, start
);
1545 while (node
&& start
<= end
) {
1546 state
= rb_entry(node
, struct extent_state
, rb_node
);
1548 if (filled
&& state
->start
> start
) {
1553 if (state
->start
> end
)
1556 if (state
->state
& bits
) {
1560 } else if (filled
) {
1564 start
= state
->end
+ 1;
1567 node
= rb_next(node
);
1574 spin_unlock(&tree
->lock
);
1579 * helper function to set a given page up to date if all the
1580 * extents in the tree for that page are up to date
1582 static int check_page_uptodate(struct extent_io_tree
*tree
,
1585 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1586 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1587 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1))
1588 SetPageUptodate(page
);
1593 * helper function to unlock a page if all the extents in the tree
1594 * for that page are unlocked
1596 static int check_page_locked(struct extent_io_tree
*tree
,
1599 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1600 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1601 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0))
1607 * helper function to end page writeback if all the extents
1608 * in the tree for that page are done with writeback
1610 static int check_page_writeback(struct extent_io_tree
*tree
,
1613 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1614 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1615 if (!test_range_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 0))
1616 end_page_writeback(page
);
1620 /* lots and lots of room for performance fixes in the end_bio funcs */
1623 * after a writepage IO is done, we need to:
1624 * clear the uptodate bits on error
1625 * clear the writeback bits in the extent tree for this IO
1626 * end_page_writeback if the page has no more pending IO
1628 * Scheduling is not allowed, so the extent state tree is expected
1629 * to have one and only one object corresponding to this IO.
1631 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1633 int uptodate
= err
== 0;
1634 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1635 struct extent_io_tree
*tree
;
1642 struct page
*page
= bvec
->bv_page
;
1643 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1645 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1647 end
= start
+ bvec
->bv_len
- 1;
1649 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1654 if (--bvec
>= bio
->bi_io_vec
)
1655 prefetchw(&bvec
->bv_page
->flags
);
1656 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1657 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
1658 end
, NULL
, uptodate
);
1663 if (!uptodate
&& tree
->ops
&&
1664 tree
->ops
->writepage_io_failed_hook
) {
1665 ret
= tree
->ops
->writepage_io_failed_hook(bio
, page
,
1668 uptodate
= (err
== 0);
1674 clear_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1675 ClearPageUptodate(page
);
1679 clear_extent_writeback(tree
, start
, end
, GFP_ATOMIC
);
1682 end_page_writeback(page
);
1684 check_page_writeback(tree
, page
);
1685 } while (bvec
>= bio
->bi_io_vec
);
1691 * after a readpage IO is done, we need to:
1692 * clear the uptodate bits on error
1693 * set the uptodate bits if things worked
1694 * set the page up to date if all extents in the tree are uptodate
1695 * clear the lock bit in the extent tree
1696 * unlock the page if there are no other extents locked for it
1698 * Scheduling is not allowed, so the extent state tree is expected
1699 * to have one and only one object corresponding to this IO.
1701 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1703 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1704 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1705 struct extent_io_tree
*tree
;
1715 struct page
*page
= bvec
->bv_page
;
1716 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1718 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1720 end
= start
+ bvec
->bv_len
- 1;
1722 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1727 if (--bvec
>= bio
->bi_io_vec
)
1728 prefetchw(&bvec
->bv_page
->flags
);
1730 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1731 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1736 if (!uptodate
&& tree
->ops
&&
1737 tree
->ops
->readpage_io_failed_hook
) {
1738 ret
= tree
->ops
->readpage_io_failed_hook(bio
, page
,
1742 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1750 set_extent_uptodate(tree
, start
, end
,
1753 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1757 SetPageUptodate(page
);
1759 ClearPageUptodate(page
);
1765 check_page_uptodate(tree
, page
);
1767 ClearPageUptodate(page
);
1770 check_page_locked(tree
, page
);
1772 } while (bvec
>= bio
->bi_io_vec
);
1778 * IO done from prepare_write is pretty simple, we just unlock
1779 * the structs in the extent tree when done, and set the uptodate bits
1782 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1784 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1785 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1786 struct extent_io_tree
*tree
;
1791 struct page
*page
= bvec
->bv_page
;
1792 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1794 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1796 end
= start
+ bvec
->bv_len
- 1;
1798 if (--bvec
>= bio
->bi_io_vec
)
1799 prefetchw(&bvec
->bv_page
->flags
);
1802 set_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1804 ClearPageUptodate(page
);
1808 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1810 } while (bvec
>= bio
->bi_io_vec
);
1816 extent_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1821 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1823 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1824 while (!bio
&& (nr_vecs
/= 2))
1825 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1830 bio
->bi_bdev
= bdev
;
1831 bio
->bi_sector
= first_sector
;
1836 static int submit_one_bio(int rw
, struct bio
*bio
, int mirror_num
,
1837 unsigned long bio_flags
)
1840 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1841 struct page
*page
= bvec
->bv_page
;
1842 struct extent_io_tree
*tree
= bio
->bi_private
;
1846 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1847 end
= start
+ bvec
->bv_len
- 1;
1849 bio
->bi_private
= NULL
;
1853 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
1854 tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
1855 mirror_num
, bio_flags
);
1857 submit_bio(rw
, bio
);
1858 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1864 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1865 struct page
*page
, sector_t sector
,
1866 size_t size
, unsigned long offset
,
1867 struct block_device
*bdev
,
1868 struct bio
**bio_ret
,
1869 unsigned long max_pages
,
1870 bio_end_io_t end_io_func
,
1872 unsigned long prev_bio_flags
,
1873 unsigned long bio_flags
)
1879 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
1880 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
1881 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
1883 if (bio_ret
&& *bio_ret
) {
1886 contig
= bio
->bi_sector
== sector
;
1888 contig
= bio
->bi_sector
+ (bio
->bi_size
>> 9) ==
1891 if (prev_bio_flags
!= bio_flags
|| !contig
||
1892 (tree
->ops
&& tree
->ops
->merge_bio_hook
&&
1893 tree
->ops
->merge_bio_hook(page
, offset
, page_size
, bio
,
1895 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
1896 ret
= submit_one_bio(rw
, bio
, mirror_num
,
1903 if (this_compressed
)
1906 nr
= bio_get_nr_vecs(bdev
);
1908 bio
= extent_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1910 bio_add_page(bio
, page
, page_size
, offset
);
1911 bio
->bi_end_io
= end_io_func
;
1912 bio
->bi_private
= tree
;
1917 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
1922 void set_page_extent_mapped(struct page
*page
)
1924 if (!PagePrivate(page
)) {
1925 SetPagePrivate(page
);
1926 page_cache_get(page
);
1927 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1931 static void set_page_extent_head(struct page
*page
, unsigned long len
)
1933 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1937 * basic readpage implementation. Locked extent state structs are inserted
1938 * into the tree that are removed when the IO is done (by the end_io
1941 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1943 get_extent_t
*get_extent
,
1944 struct bio
**bio
, int mirror_num
,
1945 unsigned long *bio_flags
)
1947 struct inode
*inode
= page
->mapping
->host
;
1948 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1949 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1953 u64 last_byte
= i_size_read(inode
);
1957 struct extent_map
*em
;
1958 struct block_device
*bdev
;
1961 size_t page_offset
= 0;
1963 size_t disk_io_size
;
1964 size_t blocksize
= inode
->i_sb
->s_blocksize
;
1965 unsigned long this_bio_flag
= 0;
1967 set_page_extent_mapped(page
);
1970 lock_extent(tree
, start
, end
, GFP_NOFS
);
1972 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
1974 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
1977 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
1978 userpage
= kmap_atomic(page
, KM_USER0
);
1979 memset(userpage
+ zero_offset
, 0, iosize
);
1980 flush_dcache_page(page
);
1981 kunmap_atomic(userpage
, KM_USER0
);
1984 while (cur
<= end
) {
1985 if (cur
>= last_byte
) {
1987 iosize
= PAGE_CACHE_SIZE
- page_offset
;
1988 userpage
= kmap_atomic(page
, KM_USER0
);
1989 memset(userpage
+ page_offset
, 0, iosize
);
1990 flush_dcache_page(page
);
1991 kunmap_atomic(userpage
, KM_USER0
);
1992 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1994 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1997 em
= get_extent(inode
, page
, page_offset
, cur
,
1999 if (IS_ERR(em
) || !em
) {
2001 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
2004 extent_offset
= cur
- em
->start
;
2005 BUG_ON(extent_map_end(em
) <= cur
);
2008 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
2009 this_bio_flag
= EXTENT_BIO_COMPRESSED
;
2011 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2012 cur_end
= min(extent_map_end(em
) - 1, end
);
2013 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2014 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2015 disk_io_size
= em
->block_len
;
2016 sector
= em
->block_start
>> 9;
2018 sector
= (em
->block_start
+ extent_offset
) >> 9;
2019 disk_io_size
= iosize
;
2022 block_start
= em
->block_start
;
2023 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2024 block_start
= EXTENT_MAP_HOLE
;
2025 free_extent_map(em
);
2028 /* we've found a hole, just zero and go on */
2029 if (block_start
== EXTENT_MAP_HOLE
) {
2031 userpage
= kmap_atomic(page
, KM_USER0
);
2032 memset(userpage
+ page_offset
, 0, iosize
);
2033 flush_dcache_page(page
);
2034 kunmap_atomic(userpage
, KM_USER0
);
2036 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2038 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2040 page_offset
+= iosize
;
2043 /* the get_extent function already copied into the page */
2044 if (test_range_bit(tree
, cur
, cur_end
, EXTENT_UPTODATE
, 1)) {
2045 check_page_uptodate(tree
, page
);
2046 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2048 page_offset
+= iosize
;
2051 /* we have an inline extent but it didn't get marked up
2052 * to date. Error out
2054 if (block_start
== EXTENT_MAP_INLINE
) {
2056 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2058 page_offset
+= iosize
;
2063 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
2064 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
2068 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2070 ret
= submit_extent_page(READ
, tree
, page
,
2071 sector
, disk_io_size
, page_offset
,
2073 end_bio_extent_readpage
, mirror_num
,
2077 *bio_flags
= this_bio_flag
;
2082 page_offset
+= iosize
;
2085 if (!PageError(page
))
2086 SetPageUptodate(page
);
2092 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2093 get_extent_t
*get_extent
)
2095 struct bio
*bio
= NULL
;
2096 unsigned long bio_flags
= 0;
2099 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, 0,
2102 submit_one_bio(READ
, bio
, 0, bio_flags
);
2107 * the writepage semantics are similar to regular writepage. extent
2108 * records are inserted to lock ranges in the tree, and as dirty areas
2109 * are found, they are marked writeback. Then the lock bits are removed
2110 * and the end_io handler clears the writeback ranges
2112 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
2115 struct inode
*inode
= page
->mapping
->host
;
2116 struct extent_page_data
*epd
= data
;
2117 struct extent_io_tree
*tree
= epd
->tree
;
2118 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2120 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2124 u64 last_byte
= i_size_read(inode
);
2129 struct extent_map
*em
;
2130 struct block_device
*bdev
;
2133 size_t pg_offset
= 0;
2135 loff_t i_size
= i_size_read(inode
);
2136 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2141 unsigned long nr_written
= 0;
2143 WARN_ON(!PageLocked(page
));
2144 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
2145 if (page
->index
> end_index
||
2146 (page
->index
== end_index
&& !pg_offset
)) {
2147 page
->mapping
->a_ops
->invalidatepage(page
, 0);
2152 if (page
->index
== end_index
) {
2155 userpage
= kmap_atomic(page
, KM_USER0
);
2156 memset(userpage
+ pg_offset
, 0,
2157 PAGE_CACHE_SIZE
- pg_offset
);
2158 kunmap_atomic(userpage
, KM_USER0
);
2159 flush_dcache_page(page
);
2163 set_page_extent_mapped(page
);
2165 delalloc_start
= start
;
2168 if (!epd
->extent_locked
) {
2169 while (delalloc_end
< page_end
) {
2170 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
2175 if (nr_delalloc
== 0) {
2176 delalloc_start
= delalloc_end
+ 1;
2179 tree
->ops
->fill_delalloc(inode
, page
, delalloc_start
,
2180 delalloc_end
, &page_started
,
2182 delalloc_start
= delalloc_end
+ 1;
2185 /* did the fill delalloc function already unlock and start
2190 goto update_nr_written
;
2193 lock_extent(tree
, start
, page_end
, GFP_NOFS
);
2195 unlock_start
= start
;
2197 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
2198 ret
= tree
->ops
->writepage_start_hook(page
, start
,
2200 if (ret
== -EAGAIN
) {
2201 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
2202 redirty_page_for_writepage(wbc
, page
);
2205 goto update_nr_written
;
2212 if (test_range_bit(tree
, start
, page_end
, EXTENT_DELALLOC
, 0))
2213 printk(KERN_ERR
"btrfs delalloc bits after lock_extent\n");
2215 if (last_byte
<= start
) {
2216 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
2217 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
2218 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2219 tree
->ops
->writepage_end_io_hook(page
, start
,
2221 unlock_start
= page_end
+ 1;
2225 set_extent_uptodate(tree
, start
, page_end
, GFP_NOFS
);
2226 blocksize
= inode
->i_sb
->s_blocksize
;
2228 while (cur
<= end
) {
2229 if (cur
>= last_byte
) {
2230 clear_extent_dirty(tree
, cur
, page_end
, GFP_NOFS
);
2231 unlock_extent(tree
, unlock_start
, page_end
, GFP_NOFS
);
2232 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2233 tree
->ops
->writepage_end_io_hook(page
, cur
,
2235 unlock_start
= page_end
+ 1;
2238 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
2240 if (IS_ERR(em
) || !em
) {
2245 extent_offset
= cur
- em
->start
;
2246 BUG_ON(extent_map_end(em
) <= cur
);
2248 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2249 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2250 sector
= (em
->block_start
+ extent_offset
) >> 9;
2252 block_start
= em
->block_start
;
2253 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
2254 free_extent_map(em
);
2258 * compressed and inline extents are written through other
2261 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
2262 block_start
== EXTENT_MAP_INLINE
) {
2263 clear_extent_dirty(tree
, cur
,
2264 cur
+ iosize
- 1, GFP_NOFS
);
2266 unlock_extent(tree
, unlock_start
, cur
+ iosize
- 1,
2270 * end_io notification does not happen here for
2271 * compressed extents
2273 if (!compressed
&& tree
->ops
&&
2274 tree
->ops
->writepage_end_io_hook
)
2275 tree
->ops
->writepage_end_io_hook(page
, cur
,
2278 else if (compressed
) {
2279 /* we don't want to end_page_writeback on
2280 * a compressed extent. this happens
2287 pg_offset
+= iosize
;
2291 /* leave this out until we have a page_mkwrite call */
2292 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2295 pg_offset
+= iosize
;
2299 clear_extent_dirty(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2300 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2301 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2309 unsigned long max_nr
= end_index
+ 1;
2311 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2312 if (!PageWriteback(page
)) {
2313 printk(KERN_ERR
"btrfs warning page %lu not "
2314 "writeback, cur %llu end %llu\n",
2315 page
->index
, (unsigned long long)cur
,
2316 (unsigned long long)end
);
2319 ret
= submit_extent_page(WRITE
, tree
, page
, sector
,
2320 iosize
, pg_offset
, bdev
,
2322 end_bio_extent_writepage
,
2328 pg_offset
+= iosize
;
2333 /* make sure the mapping tag for page dirty gets cleared */
2334 set_page_writeback(page
);
2335 end_page_writeback(page
);
2337 if (unlock_start
<= page_end
)
2338 unlock_extent(tree
, unlock_start
, page_end
, GFP_NOFS
);
2342 wbc
->nr_to_write
-= nr_written
;
2343 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
2344 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
2345 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
2350 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2351 * @mapping: address space structure to write
2352 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2353 * @writepage: function called for each page
2354 * @data: data passed to writepage function
2356 * If a page is already under I/O, write_cache_pages() skips it, even
2357 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2358 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2359 * and msync() need to guarantee that all the data which was dirty at the time
2360 * the call was made get new I/O started against them. If wbc->sync_mode is
2361 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2362 * existing IO to complete.
2364 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
2365 struct address_space
*mapping
,
2366 struct writeback_control
*wbc
,
2367 writepage_t writepage
, void *data
,
2368 void (*flush_fn
)(void *))
2370 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
2373 struct pagevec pvec
;
2376 pgoff_t end
; /* Inclusive */
2378 int range_whole
= 0;
2380 pagevec_init(&pvec
, 0);
2381 if (wbc
->range_cyclic
) {
2382 index
= mapping
->writeback_index
; /* Start from prev offset */
2385 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2386 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2387 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2392 while (!done
&& (index
<= end
) &&
2393 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2394 PAGECACHE_TAG_DIRTY
, min(end
- index
,
2395 (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2399 for (i
= 0; i
< nr_pages
; i
++) {
2400 struct page
*page
= pvec
.pages
[i
];
2403 * At this point we hold neither mapping->tree_lock nor
2404 * lock on the page itself: the page may be truncated or
2405 * invalidated (changing page->mapping to NULL), or even
2406 * swizzled back from swapper_space to tmpfs file
2409 if (tree
->ops
&& tree
->ops
->write_cache_pages_lock_hook
)
2410 tree
->ops
->write_cache_pages_lock_hook(page
);
2414 if (unlikely(page
->mapping
!= mapping
)) {
2419 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2425 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
2426 if (PageWriteback(page
))
2428 wait_on_page_writeback(page
);
2431 if (PageWriteback(page
) ||
2432 !clear_page_dirty_for_io(page
)) {
2437 ret
= (*writepage
)(page
, wbc
, data
);
2439 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2443 if (ret
|| wbc
->nr_to_write
<= 0)
2445 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2446 wbc
->encountered_congestion
= 1;
2450 pagevec_release(&pvec
);
2453 if (!scanned
&& !done
) {
2455 * We hit the last page and there is more work to be done: wrap
2456 * back to the start of the file
2465 static noinline
void flush_write_bio(void *data
)
2467 struct extent_page_data
*epd
= data
;
2469 submit_one_bio(WRITE
, epd
->bio
, 0, 0);
2474 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2475 get_extent_t
*get_extent
,
2476 struct writeback_control
*wbc
)
2479 struct address_space
*mapping
= page
->mapping
;
2480 struct extent_page_data epd
= {
2483 .get_extent
= get_extent
,
2486 struct writeback_control wbc_writepages
= {
2488 .sync_mode
= WB_SYNC_NONE
,
2489 .older_than_this
= NULL
,
2491 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2492 .range_end
= (loff_t
)-1,
2496 ret
= __extent_writepage(page
, wbc
, &epd
);
2498 extent_write_cache_pages(tree
, mapping
, &wbc_writepages
,
2499 __extent_writepage
, &epd
, flush_write_bio
);
2501 submit_one_bio(WRITE
, epd
.bio
, 0, 0);
2505 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
2506 u64 start
, u64 end
, get_extent_t
*get_extent
,
2510 struct address_space
*mapping
= inode
->i_mapping
;
2512 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
2515 struct extent_page_data epd
= {
2518 .get_extent
= get_extent
,
2521 struct writeback_control wbc_writepages
= {
2522 .bdi
= inode
->i_mapping
->backing_dev_info
,
2524 .older_than_this
= NULL
,
2525 .nr_to_write
= nr_pages
* 2,
2526 .range_start
= start
,
2527 .range_end
= end
+ 1,
2530 while (start
<= end
) {
2531 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
2532 if (clear_page_dirty_for_io(page
))
2533 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
2535 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2536 tree
->ops
->writepage_end_io_hook(page
, start
,
2537 start
+ PAGE_CACHE_SIZE
- 1,
2541 page_cache_release(page
);
2542 start
+= PAGE_CACHE_SIZE
;
2546 submit_one_bio(WRITE
, epd
.bio
, 0, 0);
2550 int extent_writepages(struct extent_io_tree
*tree
,
2551 struct address_space
*mapping
,
2552 get_extent_t
*get_extent
,
2553 struct writeback_control
*wbc
)
2556 struct extent_page_data epd
= {
2559 .get_extent
= get_extent
,
2563 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
2564 __extent_writepage
, &epd
,
2567 submit_one_bio(WRITE
, epd
.bio
, 0, 0);
2571 int extent_readpages(struct extent_io_tree
*tree
,
2572 struct address_space
*mapping
,
2573 struct list_head
*pages
, unsigned nr_pages
,
2574 get_extent_t get_extent
)
2576 struct bio
*bio
= NULL
;
2578 struct pagevec pvec
;
2579 unsigned long bio_flags
= 0;
2581 pagevec_init(&pvec
, 0);
2582 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2583 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2585 prefetchw(&page
->flags
);
2586 list_del(&page
->lru
);
2588 * what we want to do here is call add_to_page_cache_lru,
2589 * but that isn't exported, so we reproduce it here
2591 if (!add_to_page_cache(page
, mapping
,
2592 page
->index
, GFP_KERNEL
)) {
2594 /* open coding of lru_cache_add, also not exported */
2595 page_cache_get(page
);
2596 if (!pagevec_add(&pvec
, page
))
2597 __pagevec_lru_add_file(&pvec
);
2598 __extent_read_full_page(tree
, page
, get_extent
,
2599 &bio
, 0, &bio_flags
);
2601 page_cache_release(page
);
2603 if (pagevec_count(&pvec
))
2604 __pagevec_lru_add_file(&pvec
);
2605 BUG_ON(!list_empty(pages
));
2607 submit_one_bio(READ
, bio
, 0, bio_flags
);
2612 * basic invalidatepage code, this waits on any locked or writeback
2613 * ranges corresponding to the page, and then deletes any extent state
2614 * records from the tree
2616 int extent_invalidatepage(struct extent_io_tree
*tree
,
2617 struct page
*page
, unsigned long offset
)
2619 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2620 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2621 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2623 start
+= (offset
+ blocksize
- 1) & ~(blocksize
- 1);
2627 lock_extent(tree
, start
, end
, GFP_NOFS
);
2628 wait_on_extent_writeback(tree
, start
, end
);
2629 clear_extent_bit(tree
, start
, end
,
2630 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
,
2636 * simple commit_write call, set_range_dirty is used to mark both
2637 * the pages and the extent records as dirty
2639 int extent_commit_write(struct extent_io_tree
*tree
,
2640 struct inode
*inode
, struct page
*page
,
2641 unsigned from
, unsigned to
)
2643 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2645 set_page_extent_mapped(page
);
2646 set_page_dirty(page
);
2648 if (pos
> inode
->i_size
) {
2649 i_size_write(inode
, pos
);
2650 mark_inode_dirty(inode
);
2655 int extent_prepare_write(struct extent_io_tree
*tree
,
2656 struct inode
*inode
, struct page
*page
,
2657 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2659 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2660 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2662 u64 orig_block_start
;
2665 struct extent_map
*em
;
2666 unsigned blocksize
= 1 << inode
->i_blkbits
;
2667 size_t page_offset
= 0;
2668 size_t block_off_start
;
2669 size_t block_off_end
;
2675 set_page_extent_mapped(page
);
2677 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2678 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2679 orig_block_start
= block_start
;
2681 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2682 while (block_start
<= block_end
) {
2683 em
= get_extent(inode
, page
, page_offset
, block_start
,
2684 block_end
- block_start
+ 1, 1);
2685 if (IS_ERR(em
) || !em
)
2688 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2689 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2690 block_off_end
= block_off_start
+ blocksize
;
2691 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2693 if (!PageUptodate(page
) && isnew
&&
2694 (block_off_end
> to
|| block_off_start
< from
)) {
2697 kaddr
= kmap_atomic(page
, KM_USER0
);
2698 if (block_off_end
> to
)
2699 memset(kaddr
+ to
, 0, block_off_end
- to
);
2700 if (block_off_start
< from
)
2701 memset(kaddr
+ block_off_start
, 0,
2702 from
- block_off_start
);
2703 flush_dcache_page(page
);
2704 kunmap_atomic(kaddr
, KM_USER0
);
2706 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2707 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2708 !isnew
&& !PageUptodate(page
) &&
2709 (block_off_end
> to
|| block_off_start
< from
) &&
2710 !test_range_bit(tree
, block_start
, cur_end
,
2711 EXTENT_UPTODATE
, 1)) {
2713 u64 extent_offset
= block_start
- em
->start
;
2715 sector
= (em
->block_start
+ extent_offset
) >> 9;
2716 iosize
= (cur_end
- block_start
+ blocksize
) &
2717 ~((u64
)blocksize
- 1);
2719 * we've already got the extent locked, but we
2720 * need to split the state such that our end_bio
2721 * handler can clear the lock.
2723 set_extent_bit(tree
, block_start
,
2724 block_start
+ iosize
- 1,
2725 EXTENT_LOCKED
, 0, NULL
, GFP_NOFS
);
2726 ret
= submit_extent_page(READ
, tree
, page
,
2727 sector
, iosize
, page_offset
, em
->bdev
,
2729 end_bio_extent_preparewrite
, 0,
2732 block_start
= block_start
+ iosize
;
2734 set_extent_uptodate(tree
, block_start
, cur_end
,
2736 unlock_extent(tree
, block_start
, cur_end
, GFP_NOFS
);
2737 block_start
= cur_end
+ 1;
2739 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2740 free_extent_map(em
);
2743 wait_extent_bit(tree
, orig_block_start
,
2744 block_end
, EXTENT_LOCKED
);
2746 check_page_uptodate(tree
, page
);
2748 /* FIXME, zero out newly allocated blocks on error */
2753 * a helper for releasepage, this tests for areas of the page that
2754 * are locked or under IO and drops the related state bits if it is safe
2757 int try_release_extent_state(struct extent_map_tree
*map
,
2758 struct extent_io_tree
*tree
, struct page
*page
,
2761 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2762 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2765 if (test_range_bit(tree
, start
, end
,
2766 EXTENT_IOBITS
| EXTENT_ORDERED
, 0))
2769 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2771 clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
,
2778 * a helper for releasepage. As long as there are no locked extents
2779 * in the range corresponding to the page, both state records and extent
2780 * map records are removed
2782 int try_release_extent_mapping(struct extent_map_tree
*map
,
2783 struct extent_io_tree
*tree
, struct page
*page
,
2786 struct extent_map
*em
;
2787 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2788 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2790 if ((mask
& __GFP_WAIT
) &&
2791 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2793 while (start
<= end
) {
2794 len
= end
- start
+ 1;
2795 spin_lock(&map
->lock
);
2796 em
= lookup_extent_mapping(map
, start
, len
);
2797 if (!em
|| IS_ERR(em
)) {
2798 spin_unlock(&map
->lock
);
2801 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
2802 em
->start
!= start
) {
2803 spin_unlock(&map
->lock
);
2804 free_extent_map(em
);
2807 if (!test_range_bit(tree
, em
->start
,
2808 extent_map_end(em
) - 1,
2809 EXTENT_LOCKED
| EXTENT_WRITEBACK
|
2812 remove_extent_mapping(map
, em
);
2813 /* once for the rb tree */
2814 free_extent_map(em
);
2816 start
= extent_map_end(em
);
2817 spin_unlock(&map
->lock
);
2820 free_extent_map(em
);
2823 return try_release_extent_state(map
, tree
, page
, mask
);
2826 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2827 get_extent_t
*get_extent
)
2829 struct inode
*inode
= mapping
->host
;
2830 u64 start
= iblock
<< inode
->i_blkbits
;
2831 sector_t sector
= 0;
2832 size_t blksize
= (1 << inode
->i_blkbits
);
2833 struct extent_map
*em
;
2835 lock_extent(&BTRFS_I(inode
)->io_tree
, start
, start
+ blksize
- 1,
2837 em
= get_extent(inode
, NULL
, 0, start
, blksize
, 0);
2838 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
, start
+ blksize
- 1,
2840 if (!em
|| IS_ERR(em
))
2843 if (em
->block_start
> EXTENT_MAP_LAST_BYTE
)
2846 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2848 free_extent_map(em
);
2852 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
2853 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
2857 u64 max
= start
+ len
;
2860 struct extent_map
*em
= NULL
;
2862 u64 em_start
= 0, em_len
= 0;
2863 unsigned long emflags
;
2869 lock_extent(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
,
2871 em
= get_extent(inode
, NULL
, 0, off
, max
- off
, 0);
2879 off
= em
->start
+ em
->len
;
2883 em_start
= em
->start
;
2889 switch (em
->block_start
) {
2890 case EXTENT_MAP_LAST_BYTE
:
2892 flags
|= FIEMAP_EXTENT_LAST
;
2894 case EXTENT_MAP_HOLE
:
2895 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2897 case EXTENT_MAP_INLINE
:
2898 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
2899 FIEMAP_EXTENT_NOT_ALIGNED
);
2901 case EXTENT_MAP_DELALLOC
:
2902 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2903 FIEMAP_EXTENT_UNKNOWN
);
2906 disko
= em
->block_start
;
2909 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
2910 flags
|= FIEMAP_EXTENT_ENCODED
;
2912 emflags
= em
->flags
;
2913 free_extent_map(em
);
2917 em
= get_extent(inode
, NULL
, 0, off
, max
- off
, 0);
2924 emflags
= em
->flags
;
2926 if (test_bit(EXTENT_FLAG_VACANCY
, &emflags
)) {
2927 flags
|= FIEMAP_EXTENT_LAST
;
2931 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
2937 free_extent_map(em
);
2939 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
,
2944 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
2948 struct address_space
*mapping
;
2951 return eb
->first_page
;
2952 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
2953 mapping
= eb
->first_page
->mapping
;
2958 * extent_buffer_page is only called after pinning the page
2959 * by increasing the reference count. So we know the page must
2960 * be in the radix tree.
2963 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
2969 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
2971 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
2972 (start
>> PAGE_CACHE_SHIFT
);
2975 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
2980 struct extent_buffer
*eb
= NULL
;
2982 unsigned long flags
;
2985 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
2988 spin_lock_init(&eb
->lock
);
2989 init_waitqueue_head(&eb
->lock_wq
);
2992 spin_lock_irqsave(&leak_lock
, flags
);
2993 list_add(&eb
->leak_list
, &buffers
);
2994 spin_unlock_irqrestore(&leak_lock
, flags
);
2996 atomic_set(&eb
->refs
, 1);
3001 static void __free_extent_buffer(struct extent_buffer
*eb
)
3004 unsigned long flags
;
3005 spin_lock_irqsave(&leak_lock
, flags
);
3006 list_del(&eb
->leak_list
);
3007 spin_unlock_irqrestore(&leak_lock
, flags
);
3009 kmem_cache_free(extent_buffer_cache
, eb
);
3012 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
3013 u64 start
, unsigned long len
,
3017 unsigned long num_pages
= num_extent_pages(start
, len
);
3019 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
3020 struct extent_buffer
*eb
;
3021 struct extent_buffer
*exists
= NULL
;
3023 struct address_space
*mapping
= tree
->mapping
;
3026 spin_lock(&tree
->buffer_lock
);
3027 eb
= buffer_search(tree
, start
);
3029 atomic_inc(&eb
->refs
);
3030 spin_unlock(&tree
->buffer_lock
);
3031 mark_page_accessed(eb
->first_page
);
3034 spin_unlock(&tree
->buffer_lock
);
3036 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
3041 eb
->first_page
= page0
;
3044 page_cache_get(page0
);
3045 mark_page_accessed(page0
);
3046 set_page_extent_mapped(page0
);
3047 set_page_extent_head(page0
, len
);
3048 uptodate
= PageUptodate(page0
);
3052 for (; i
< num_pages
; i
++, index
++) {
3053 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
3058 set_page_extent_mapped(p
);
3059 mark_page_accessed(p
);
3062 set_page_extent_head(p
, len
);
3064 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
3066 if (!PageUptodate(p
))
3071 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3073 spin_lock(&tree
->buffer_lock
);
3074 exists
= buffer_tree_insert(tree
, start
, &eb
->rb_node
);
3076 /* add one reference for the caller */
3077 atomic_inc(&exists
->refs
);
3078 spin_unlock(&tree
->buffer_lock
);
3081 spin_unlock(&tree
->buffer_lock
);
3083 /* add one reference for the tree */
3084 atomic_inc(&eb
->refs
);
3088 if (!atomic_dec_and_test(&eb
->refs
))
3090 for (index
= 1; index
< i
; index
++)
3091 page_cache_release(extent_buffer_page(eb
, index
));
3092 page_cache_release(extent_buffer_page(eb
, 0));
3093 __free_extent_buffer(eb
);
3097 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
3098 u64 start
, unsigned long len
,
3101 struct extent_buffer
*eb
;
3103 spin_lock(&tree
->buffer_lock
);
3104 eb
= buffer_search(tree
, start
);
3106 atomic_inc(&eb
->refs
);
3107 spin_unlock(&tree
->buffer_lock
);
3110 mark_page_accessed(eb
->first_page
);
3115 void free_extent_buffer(struct extent_buffer
*eb
)
3120 if (!atomic_dec_and_test(&eb
->refs
))
3126 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
3127 struct extent_buffer
*eb
)
3131 unsigned long num_pages
;
3134 u64 start
= eb
->start
;
3135 u64 end
= start
+ eb
->len
- 1;
3137 set
= clear_extent_dirty(tree
, start
, end
, GFP_NOFS
);
3138 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3140 for (i
= 0; i
< num_pages
; i
++) {
3141 page
= extent_buffer_page(eb
, i
);
3142 if (!set
&& !PageDirty(page
))
3147 set_page_extent_head(page
, eb
->len
);
3149 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
3152 * if we're on the last page or the first page and the
3153 * block isn't aligned on a page boundary, do extra checks
3154 * to make sure we don't clean page that is partially dirty
3156 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
3157 ((i
== num_pages
- 1) &&
3158 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
3159 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3160 end
= start
+ PAGE_CACHE_SIZE
- 1;
3161 if (test_range_bit(tree
, start
, end
,
3167 clear_page_dirty_for_io(page
);
3168 spin_lock_irq(&page
->mapping
->tree_lock
);
3169 if (!PageDirty(page
)) {
3170 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3172 PAGECACHE_TAG_DIRTY
);
3174 spin_unlock_irq(&page
->mapping
->tree_lock
);
3180 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
3181 struct extent_buffer
*eb
)
3183 return wait_on_extent_writeback(tree
, eb
->start
,
3184 eb
->start
+ eb
->len
- 1);
3187 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
3188 struct extent_buffer
*eb
)
3191 unsigned long num_pages
;
3193 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3194 for (i
= 0; i
< num_pages
; i
++) {
3195 struct page
*page
= extent_buffer_page(eb
, i
);
3196 /* writepage may need to do something special for the
3197 * first page, we have to make sure page->private is
3198 * properly set. releasepage may drop page->private
3199 * on us if the page isn't already dirty.
3203 set_page_extent_head(page
, eb
->len
);
3204 } else if (PagePrivate(page
) &&
3205 page
->private != EXTENT_PAGE_PRIVATE
) {
3206 set_page_extent_mapped(page
);
3208 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
3209 set_extent_dirty(tree
, page_offset(page
),
3210 page_offset(page
) + PAGE_CACHE_SIZE
- 1,
3217 int clear_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3218 struct extent_buffer
*eb
)
3222 unsigned long num_pages
;
3224 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3225 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3227 clear_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3229 for (i
= 0; i
< num_pages
; i
++) {
3230 page
= extent_buffer_page(eb
, i
);
3232 ClearPageUptodate(page
);
3237 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3238 struct extent_buffer
*eb
)
3242 unsigned long num_pages
;
3244 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3246 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3248 for (i
= 0; i
< num_pages
; i
++) {
3249 page
= extent_buffer_page(eb
, i
);
3250 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
3251 ((i
== num_pages
- 1) &&
3252 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
3253 check_page_uptodate(tree
, page
);
3256 SetPageUptodate(page
);
3261 int extent_range_uptodate(struct extent_io_tree
*tree
,
3266 int pg_uptodate
= 1;
3268 unsigned long index
;
3270 ret
= test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1);
3273 while (start
<= end
) {
3274 index
= start
>> PAGE_CACHE_SHIFT
;
3275 page
= find_get_page(tree
->mapping
, index
);
3276 uptodate
= PageUptodate(page
);
3277 page_cache_release(page
);
3282 start
+= PAGE_CACHE_SIZE
;
3287 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
3288 struct extent_buffer
*eb
)
3291 unsigned long num_pages
;
3294 int pg_uptodate
= 1;
3296 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
3299 ret
= test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3300 EXTENT_UPTODATE
, 1);
3304 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3305 for (i
= 0; i
< num_pages
; i
++) {
3306 page
= extent_buffer_page(eb
, i
);
3307 if (!PageUptodate(page
)) {
3315 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
3316 struct extent_buffer
*eb
,
3317 u64 start
, int wait
,
3318 get_extent_t
*get_extent
, int mirror_num
)
3321 unsigned long start_i
;
3325 int locked_pages
= 0;
3326 int all_uptodate
= 1;
3327 int inc_all_pages
= 0;
3328 unsigned long num_pages
;
3329 struct bio
*bio
= NULL
;
3330 unsigned long bio_flags
= 0;
3332 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
3335 if (test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3336 EXTENT_UPTODATE
, 1)) {
3341 WARN_ON(start
< eb
->start
);
3342 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
3343 (eb
->start
>> PAGE_CACHE_SHIFT
);
3348 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3349 for (i
= start_i
; i
< num_pages
; i
++) {
3350 page
= extent_buffer_page(eb
, i
);
3352 if (!trylock_page(page
))
3358 if (!PageUptodate(page
))
3363 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3367 for (i
= start_i
; i
< num_pages
; i
++) {
3368 page
= extent_buffer_page(eb
, i
);
3370 page_cache_get(page
);
3371 if (!PageUptodate(page
)) {
3374 ClearPageError(page
);
3375 err
= __extent_read_full_page(tree
, page
,
3377 mirror_num
, &bio_flags
);
3386 submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3391 for (i
= start_i
; i
< num_pages
; i
++) {
3392 page
= extent_buffer_page(eb
, i
);
3393 wait_on_page_locked(page
);
3394 if (!PageUptodate(page
))
3399 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3404 while (locked_pages
> 0) {
3405 page
= extent_buffer_page(eb
, i
);
3413 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
3414 unsigned long start
,
3421 char *dst
= (char *)dstv
;
3422 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3423 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3425 WARN_ON(start
> eb
->len
);
3426 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3428 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3431 page
= extent_buffer_page(eb
, i
);
3433 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3434 kaddr
= kmap_atomic(page
, KM_USER1
);
3435 memcpy(dst
, kaddr
+ offset
, cur
);
3436 kunmap_atomic(kaddr
, KM_USER1
);
3445 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3446 unsigned long min_len
, char **token
, char **map
,
3447 unsigned long *map_start
,
3448 unsigned long *map_len
, int km
)
3450 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
3453 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3454 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3455 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
3462 offset
= start_offset
;
3466 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
3469 if (start
+ min_len
> eb
->len
) {
3470 printk(KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
3471 "wanted %lu %lu\n", (unsigned long long)eb
->start
,
3472 eb
->len
, start
, min_len
);
3476 p
= extent_buffer_page(eb
, i
);
3477 kaddr
= kmap_atomic(p
, km
);
3479 *map
= kaddr
+ offset
;
3480 *map_len
= PAGE_CACHE_SIZE
- offset
;
3484 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3485 unsigned long min_len
,
3486 char **token
, char **map
,
3487 unsigned long *map_start
,
3488 unsigned long *map_len
, int km
)
3492 if (eb
->map_token
) {
3493 unmap_extent_buffer(eb
, eb
->map_token
, km
);
3494 eb
->map_token
= NULL
;
3497 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
3498 map_start
, map_len
, km
);
3500 eb
->map_token
= *token
;
3502 eb
->map_start
= *map_start
;
3503 eb
->map_len
= *map_len
;
3508 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
3510 kunmap_atomic(token
, km
);
3513 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3514 unsigned long start
,
3521 char *ptr
= (char *)ptrv
;
3522 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3523 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3526 WARN_ON(start
> eb
->len
);
3527 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3529 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3532 page
= extent_buffer_page(eb
, i
);
3534 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3536 kaddr
= kmap_atomic(page
, KM_USER0
);
3537 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3538 kunmap_atomic(kaddr
, KM_USER0
);
3550 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3551 unsigned long start
, unsigned long len
)
3557 char *src
= (char *)srcv
;
3558 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3559 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3561 WARN_ON(start
> eb
->len
);
3562 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3564 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3567 page
= extent_buffer_page(eb
, i
);
3568 WARN_ON(!PageUptodate(page
));
3570 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3571 kaddr
= kmap_atomic(page
, KM_USER1
);
3572 memcpy(kaddr
+ offset
, src
, cur
);
3573 kunmap_atomic(kaddr
, KM_USER1
);
3582 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3583 unsigned long start
, unsigned long len
)
3589 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3590 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3592 WARN_ON(start
> eb
->len
);
3593 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3595 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3598 page
= extent_buffer_page(eb
, i
);
3599 WARN_ON(!PageUptodate(page
));
3601 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3602 kaddr
= kmap_atomic(page
, KM_USER0
);
3603 memset(kaddr
+ offset
, c
, cur
);
3604 kunmap_atomic(kaddr
, KM_USER0
);
3612 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3613 unsigned long dst_offset
, unsigned long src_offset
,
3616 u64 dst_len
= dst
->len
;
3621 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3622 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3624 WARN_ON(src
->len
!= dst_len
);
3626 offset
= (start_offset
+ dst_offset
) &
3627 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3630 page
= extent_buffer_page(dst
, i
);
3631 WARN_ON(!PageUptodate(page
));
3633 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3635 kaddr
= kmap_atomic(page
, KM_USER0
);
3636 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3637 kunmap_atomic(kaddr
, KM_USER0
);
3646 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3647 unsigned long dst_off
, unsigned long src_off
,
3650 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3651 if (dst_page
== src_page
) {
3652 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3654 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3655 char *p
= dst_kaddr
+ dst_off
+ len
;
3656 char *s
= src_kaddr
+ src_off
+ len
;
3661 kunmap_atomic(src_kaddr
, KM_USER1
);
3663 kunmap_atomic(dst_kaddr
, KM_USER0
);
3666 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3667 unsigned long dst_off
, unsigned long src_off
,
3670 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3673 if (dst_page
!= src_page
)
3674 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3676 src_kaddr
= dst_kaddr
;
3678 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3679 kunmap_atomic(dst_kaddr
, KM_USER0
);
3680 if (dst_page
!= src_page
)
3681 kunmap_atomic(src_kaddr
, KM_USER1
);
3684 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3685 unsigned long src_offset
, unsigned long len
)
3688 size_t dst_off_in_page
;
3689 size_t src_off_in_page
;
3690 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3691 unsigned long dst_i
;
3692 unsigned long src_i
;
3694 if (src_offset
+ len
> dst
->len
) {
3695 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
3696 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
3699 if (dst_offset
+ len
> dst
->len
) {
3700 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
3701 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
3706 dst_off_in_page
= (start_offset
+ dst_offset
) &
3707 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3708 src_off_in_page
= (start_offset
+ src_offset
) &
3709 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3711 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3712 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3714 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3716 cur
= min_t(unsigned long, cur
,
3717 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3719 copy_pages(extent_buffer_page(dst
, dst_i
),
3720 extent_buffer_page(dst
, src_i
),
3721 dst_off_in_page
, src_off_in_page
, cur
);
3729 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3730 unsigned long src_offset
, unsigned long len
)
3733 size_t dst_off_in_page
;
3734 size_t src_off_in_page
;
3735 unsigned long dst_end
= dst_offset
+ len
- 1;
3736 unsigned long src_end
= src_offset
+ len
- 1;
3737 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3738 unsigned long dst_i
;
3739 unsigned long src_i
;
3741 if (src_offset
+ len
> dst
->len
) {
3742 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
3743 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
3746 if (dst_offset
+ len
> dst
->len
) {
3747 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
3748 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
3751 if (dst_offset
< src_offset
) {
3752 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
3756 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
3757 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
3759 dst_off_in_page
= (start_offset
+ dst_end
) &
3760 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3761 src_off_in_page
= (start_offset
+ src_end
) &
3762 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3764 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
3765 cur
= min(cur
, dst_off_in_page
+ 1);
3766 move_pages(extent_buffer_page(dst
, dst_i
),
3767 extent_buffer_page(dst
, src_i
),
3768 dst_off_in_page
- cur
+ 1,
3769 src_off_in_page
- cur
+ 1, cur
);
3777 int try_release_extent_buffer(struct extent_io_tree
*tree
, struct page
*page
)
3779 u64 start
= page_offset(page
);
3780 struct extent_buffer
*eb
;
3783 unsigned long num_pages
;
3785 spin_lock(&tree
->buffer_lock
);
3786 eb
= buffer_search(tree
, start
);
3790 if (atomic_read(&eb
->refs
) > 1) {
3794 /* at this point we can safely release the extent buffer */
3795 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3796 for (i
= 0; i
< num_pages
; i
++)
3797 page_cache_release(extent_buffer_page(eb
, i
));
3798 rb_erase(&eb
->rb_node
, &tree
->buffer
);
3799 __free_extent_buffer(eb
);
3801 spin_unlock(&tree
->buffer_lock
);