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/version.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 /* temporary define until extent_map moves out of btrfs */
22 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
23 unsigned long extra_flags
,
24 void (*ctor
)(void *, struct kmem_cache
*,
27 static struct kmem_cache
*extent_state_cache
;
28 static struct kmem_cache
*extent_buffer_cache
;
30 static LIST_HEAD(buffers
);
31 static LIST_HEAD(states
);
35 static spinlock_t leak_lock
= SPIN_LOCK_UNLOCKED
;
38 #define BUFFER_LRU_MAX 64
43 struct rb_node rb_node
;
46 struct extent_page_data
{
48 struct extent_io_tree
*tree
;
49 get_extent_t
*get_extent
;
51 /* tells writepage not to lock the state bits for this range
52 * it still does the unlocking
57 int __init
extent_io_init(void)
59 extent_state_cache
= btrfs_cache_create("extent_state",
60 sizeof(struct extent_state
), 0,
62 if (!extent_state_cache
)
65 extent_buffer_cache
= btrfs_cache_create("extent_buffers",
66 sizeof(struct extent_buffer
), 0,
68 if (!extent_buffer_cache
)
69 goto free_state_cache
;
73 kmem_cache_destroy(extent_state_cache
);
77 void extent_io_exit(void)
79 struct extent_state
*state
;
80 struct extent_buffer
*eb
;
82 while (!list_empty(&states
)) {
83 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
84 printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state
->start
, state
->end
, state
->state
, state
->tree
, atomic_read(&state
->refs
));
85 list_del(&state
->leak_list
);
86 kmem_cache_free(extent_state_cache
, state
);
90 while (!list_empty(&buffers
)) {
91 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
92 printk("buffer leak start %Lu len %lu refs %d\n", eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
93 list_del(&eb
->leak_list
);
94 kmem_cache_free(extent_buffer_cache
, eb
);
96 if (extent_state_cache
)
97 kmem_cache_destroy(extent_state_cache
);
98 if (extent_buffer_cache
)
99 kmem_cache_destroy(extent_buffer_cache
);
102 void extent_io_tree_init(struct extent_io_tree
*tree
,
103 struct address_space
*mapping
, gfp_t mask
)
105 tree
->state
.rb_node
= NULL
;
106 tree
->buffer
.rb_node
= NULL
;
108 tree
->dirty_bytes
= 0;
109 spin_lock_init(&tree
->lock
);
110 spin_lock_init(&tree
->buffer_lock
);
111 tree
->mapping
= mapping
;
113 EXPORT_SYMBOL(extent_io_tree_init
);
115 struct extent_state
*alloc_extent_state(gfp_t mask
)
117 struct extent_state
*state
;
122 state
= kmem_cache_alloc(extent_state_cache
, mask
);
129 spin_lock_irqsave(&leak_lock
, flags
);
130 list_add(&state
->leak_list
, &states
);
131 spin_unlock_irqrestore(&leak_lock
, flags
);
133 atomic_set(&state
->refs
, 1);
134 init_waitqueue_head(&state
->wq
);
137 EXPORT_SYMBOL(alloc_extent_state
);
139 void free_extent_state(struct extent_state
*state
)
143 if (atomic_dec_and_test(&state
->refs
)) {
147 WARN_ON(state
->tree
);
149 spin_lock_irqsave(&leak_lock
, flags
);
150 list_del(&state
->leak_list
);
151 spin_unlock_irqrestore(&leak_lock
, flags
);
153 kmem_cache_free(extent_state_cache
, state
);
156 EXPORT_SYMBOL(free_extent_state
);
158 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
159 struct rb_node
*node
)
161 struct rb_node
** p
= &root
->rb_node
;
162 struct rb_node
* parent
= NULL
;
163 struct tree_entry
*entry
;
167 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
169 if (offset
< entry
->start
)
171 else if (offset
> entry
->end
)
177 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
178 rb_link_node(node
, parent
, p
);
179 rb_insert_color(node
, root
);
183 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
184 struct rb_node
**prev_ret
,
185 struct rb_node
**next_ret
)
187 struct rb_root
*root
= &tree
->state
;
188 struct rb_node
* n
= root
->rb_node
;
189 struct rb_node
*prev
= NULL
;
190 struct rb_node
*orig_prev
= NULL
;
191 struct tree_entry
*entry
;
192 struct tree_entry
*prev_entry
= NULL
;
195 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
199 if (offset
< entry
->start
)
201 else if (offset
> entry
->end
)
210 while(prev
&& offset
> prev_entry
->end
) {
211 prev
= rb_next(prev
);
212 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
219 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
220 while(prev
&& offset
< prev_entry
->start
) {
221 prev
= rb_prev(prev
);
222 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
229 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
232 struct rb_node
*prev
= NULL
;
235 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
->set_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("end < start %Lu %Lu\n", end
, start
);
369 if (bits
& EXTENT_DIRTY
)
370 tree
->dirty_bytes
+= end
- start
+ 1;
371 set_state_cb(tree
, state
, bits
);
372 state
->state
|= bits
;
373 state
->start
= start
;
375 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
377 struct extent_state
*found
;
378 found
= rb_entry(node
, struct extent_state
, rb_node
);
379 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, start
, end
);
380 free_extent_state(state
);
384 merge_state(tree
, state
);
389 * split a given extent state struct in two, inserting the preallocated
390 * struct 'prealloc' as the newly created second half. 'split' indicates an
391 * offset inside 'orig' where it should be split.
394 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
395 * are two extent state structs in the tree:
396 * prealloc: [orig->start, split - 1]
397 * orig: [ split, orig->end ]
399 * The tree locks are not taken by this function. They need to be held
402 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
403 struct extent_state
*prealloc
, u64 split
)
405 struct rb_node
*node
;
406 prealloc
->start
= orig
->start
;
407 prealloc
->end
= split
- 1;
408 prealloc
->state
= orig
->state
;
411 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
413 struct extent_state
*found
;
414 found
= rb_entry(node
, struct extent_state
, rb_node
);
415 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, prealloc
->start
, prealloc
->end
);
416 free_extent_state(prealloc
);
419 prealloc
->tree
= tree
;
424 * utility function to clear some bits in an extent state struct.
425 * it will optionally wake up any one waiting on this state (wake == 1), or
426 * forcibly remove the state from the tree (delete == 1).
428 * If no bits are set on the state struct after clearing things, the
429 * struct is freed and removed from the tree
431 static int clear_state_bit(struct extent_io_tree
*tree
,
432 struct extent_state
*state
, int bits
, int wake
,
435 int ret
= state
->state
& bits
;
437 if ((bits
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
438 u64 range
= state
->end
- state
->start
+ 1;
439 WARN_ON(range
> tree
->dirty_bytes
);
440 tree
->dirty_bytes
-= range
;
442 clear_state_cb(tree
, state
, bits
);
443 state
->state
&= ~bits
;
446 if (delete || state
->state
== 0) {
448 clear_state_cb(tree
, state
, state
->state
);
449 rb_erase(&state
->rb_node
, &tree
->state
);
451 free_extent_state(state
);
456 merge_state(tree
, state
);
462 * clear some bits on a range in the tree. This may require splitting
463 * or inserting elements in the tree, so the gfp mask is used to
464 * indicate which allocations or sleeping are allowed.
466 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
467 * the given range from the tree regardless of state (ie for truncate).
469 * the range [start, end] is inclusive.
471 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
472 * bits were already set, or zero if none of the bits were already set.
474 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
475 int bits
, int wake
, int delete, gfp_t mask
)
477 struct extent_state
*state
;
478 struct extent_state
*prealloc
= NULL
;
479 struct rb_node
*node
;
485 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
486 prealloc
= alloc_extent_state(mask
);
491 spin_lock_irqsave(&tree
->lock
, flags
);
493 * this search will find the extents that end after
496 node
= tree_search(tree
, start
);
499 state
= rb_entry(node
, struct extent_state
, rb_node
);
500 if (state
->start
> end
)
502 WARN_ON(state
->end
< start
);
505 * | ---- desired range ---- |
507 * | ------------- state -------------- |
509 * We need to split the extent we found, and may flip
510 * bits on second half.
512 * If the extent we found extends past our range, we
513 * just split and search again. It'll get split again
514 * the next time though.
516 * If the extent we found is inside our range, we clear
517 * the desired bit on it.
520 if (state
->start
< start
) {
522 prealloc
= alloc_extent_state(GFP_ATOMIC
);
523 err
= split_state(tree
, state
, prealloc
, start
);
524 BUG_ON(err
== -EEXIST
);
528 if (state
->end
<= end
) {
529 start
= state
->end
+ 1;
530 set
|= clear_state_bit(tree
, state
, bits
,
533 start
= state
->start
;
538 * | ---- desired range ---- |
540 * We need to split the extent, and clear the bit
543 if (state
->start
<= end
&& state
->end
> end
) {
545 prealloc
= alloc_extent_state(GFP_ATOMIC
);
546 err
= split_state(tree
, state
, prealloc
, end
+ 1);
547 BUG_ON(err
== -EEXIST
);
551 set
|= clear_state_bit(tree
, prealloc
, bits
,
557 start
= state
->end
+ 1;
558 set
|= clear_state_bit(tree
, state
, bits
, wake
, delete);
562 spin_unlock_irqrestore(&tree
->lock
, flags
);
564 free_extent_state(prealloc
);
571 spin_unlock_irqrestore(&tree
->lock
, flags
);
572 if (mask
& __GFP_WAIT
)
576 EXPORT_SYMBOL(clear_extent_bit
);
578 static int wait_on_state(struct extent_io_tree
*tree
,
579 struct extent_state
*state
)
580 __releases(tree
->lock
)
581 __acquires(tree
->lock
)
584 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
585 spin_unlock_irq(&tree
->lock
);
587 spin_lock_irq(&tree
->lock
);
588 finish_wait(&state
->wq
, &wait
);
593 * waits for one or more bits to clear on a range in the state tree.
594 * The range [start, end] is inclusive.
595 * The tree lock is taken by this function
597 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
599 struct extent_state
*state
;
600 struct rb_node
*node
;
602 spin_lock_irq(&tree
->lock
);
606 * this search will find all the extents that end after
609 node
= tree_search(tree
, start
);
613 state
= rb_entry(node
, struct extent_state
, rb_node
);
615 if (state
->start
> end
)
618 if (state
->state
& bits
) {
619 start
= state
->start
;
620 atomic_inc(&state
->refs
);
621 wait_on_state(tree
, state
);
622 free_extent_state(state
);
625 start
= state
->end
+ 1;
630 if (need_resched()) {
631 spin_unlock_irq(&tree
->lock
);
633 spin_lock_irq(&tree
->lock
);
637 spin_unlock_irq(&tree
->lock
);
640 EXPORT_SYMBOL(wait_extent_bit
);
642 static void set_state_bits(struct extent_io_tree
*tree
,
643 struct extent_state
*state
,
646 if ((bits
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
647 u64 range
= state
->end
- state
->start
+ 1;
648 tree
->dirty_bytes
+= range
;
650 set_state_cb(tree
, state
, bits
);
651 state
->state
|= bits
;
655 * set some bits on a range in the tree. This may require allocations
656 * or sleeping, so the gfp mask is used to indicate what is allowed.
658 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
659 * range already has the desired bits set. The start of the existing
660 * range is returned in failed_start in this case.
662 * [start, end] is inclusive
663 * This takes the tree lock.
665 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
,
666 int exclusive
, u64
*failed_start
, gfp_t mask
)
668 struct extent_state
*state
;
669 struct extent_state
*prealloc
= NULL
;
670 struct rb_node
*node
;
677 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
678 prealloc
= alloc_extent_state(mask
);
683 spin_lock_irqsave(&tree
->lock
, flags
);
685 * this search will find all the extents that end after
688 node
= tree_search(tree
, start
);
690 err
= insert_state(tree
, prealloc
, start
, end
, bits
);
692 BUG_ON(err
== -EEXIST
);
696 state
= rb_entry(node
, struct extent_state
, rb_node
);
697 last_start
= state
->start
;
698 last_end
= state
->end
;
701 * | ---- desired range ---- |
704 * Just lock what we found and keep going
706 if (state
->start
== start
&& state
->end
<= end
) {
707 set
= state
->state
& bits
;
708 if (set
&& exclusive
) {
709 *failed_start
= state
->start
;
713 set_state_bits(tree
, state
, bits
);
714 start
= state
->end
+ 1;
715 merge_state(tree
, state
);
720 * | ---- desired range ---- |
723 * | ------------- state -------------- |
725 * We need to split the extent we found, and may flip bits on
728 * If the extent we found extends past our
729 * range, we just split and search again. It'll get split
730 * again the next time though.
732 * If the extent we found is inside our range, we set the
735 if (state
->start
< start
) {
736 set
= state
->state
& bits
;
737 if (exclusive
&& set
) {
738 *failed_start
= start
;
742 err
= split_state(tree
, state
, prealloc
, start
);
743 BUG_ON(err
== -EEXIST
);
747 if (state
->end
<= end
) {
748 set_state_bits(tree
, state
, bits
);
749 start
= state
->end
+ 1;
750 merge_state(tree
, state
);
752 start
= state
->start
;
757 * | ---- desired range ---- |
758 * | state | or | state |
760 * There's a hole, we need to insert something in it and
761 * ignore the extent we found.
763 if (state
->start
> start
) {
765 if (end
< last_start
)
768 this_end
= last_start
-1;
769 err
= insert_state(tree
, prealloc
, start
, this_end
,
772 BUG_ON(err
== -EEXIST
);
775 start
= this_end
+ 1;
779 * | ---- desired range ---- |
781 * We need to split the extent, and set the bit
784 if (state
->start
<= end
&& state
->end
> end
) {
785 set
= state
->state
& bits
;
786 if (exclusive
&& set
) {
787 *failed_start
= start
;
791 err
= split_state(tree
, state
, prealloc
, end
+ 1);
792 BUG_ON(err
== -EEXIST
);
794 set_state_bits(tree
, prealloc
, bits
);
795 merge_state(tree
, prealloc
);
803 spin_unlock_irqrestore(&tree
->lock
, flags
);
805 free_extent_state(prealloc
);
812 spin_unlock_irqrestore(&tree
->lock
, flags
);
813 if (mask
& __GFP_WAIT
)
817 EXPORT_SYMBOL(set_extent_bit
);
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
,
826 EXPORT_SYMBOL(set_extent_dirty
);
828 int set_extent_ordered(struct extent_io_tree
*tree
, u64 start
, u64 end
,
831 return set_extent_bit(tree
, start
, end
, EXTENT_ORDERED
, 0, NULL
, mask
);
833 EXPORT_SYMBOL(set_extent_ordered
);
835 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
836 int bits
, gfp_t mask
)
838 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
841 EXPORT_SYMBOL(set_extent_bits
);
843 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
844 int bits
, gfp_t mask
)
846 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, mask
);
848 EXPORT_SYMBOL(clear_extent_bits
);
850 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
853 return set_extent_bit(tree
, start
, end
,
854 EXTENT_DELALLOC
| EXTENT_DIRTY
,
857 EXPORT_SYMBOL(set_extent_delalloc
);
859 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
862 return clear_extent_bit(tree
, start
, end
,
863 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, mask
);
865 EXPORT_SYMBOL(clear_extent_dirty
);
867 int clear_extent_ordered(struct extent_io_tree
*tree
, u64 start
, u64 end
,
870 return clear_extent_bit(tree
, start
, end
, EXTENT_ORDERED
, 1, 0, mask
);
872 EXPORT_SYMBOL(clear_extent_ordered
);
874 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
877 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
880 EXPORT_SYMBOL(set_extent_new
);
882 int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
885 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0, mask
);
887 EXPORT_SYMBOL(clear_extent_new
);
889 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
892 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, NULL
,
895 EXPORT_SYMBOL(set_extent_uptodate
);
897 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
900 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0, mask
);
902 EXPORT_SYMBOL(clear_extent_uptodate
);
904 int set_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
907 return set_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
,
910 EXPORT_SYMBOL(set_extent_writeback
);
912 int clear_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
915 return clear_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 1, 0, mask
);
917 EXPORT_SYMBOL(clear_extent_writeback
);
919 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
921 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
923 EXPORT_SYMBOL(wait_on_extent_writeback
);
926 * either insert or lock state struct between start and end use mask to tell
927 * us if waiting is desired.
929 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
934 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
935 &failed_start
, mask
);
936 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
937 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
938 start
= failed_start
;
942 WARN_ON(start
> end
);
946 EXPORT_SYMBOL(lock_extent
);
948 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
954 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
955 &failed_start
, mask
);
956 if (err
== -EEXIST
) {
957 if (failed_start
> start
)
958 clear_extent_bit(tree
, start
, failed_start
- 1,
959 EXTENT_LOCKED
, 1, 0, mask
);
964 EXPORT_SYMBOL(try_lock_extent
);
966 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
969 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, mask
);
971 EXPORT_SYMBOL(unlock_extent
);
974 * helper function to set pages and extents in the tree dirty
976 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
978 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
979 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
982 while (index
<= end_index
) {
983 page
= find_get_page(tree
->mapping
, index
);
985 __set_page_dirty_nobuffers(page
);
986 page_cache_release(page
);
989 set_extent_dirty(tree
, start
, end
, GFP_NOFS
);
992 EXPORT_SYMBOL(set_range_dirty
);
995 * helper function to set both pages and extents in the tree writeback
997 int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
999 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1000 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1003 while (index
<= end_index
) {
1004 page
= find_get_page(tree
->mapping
, index
);
1006 set_page_writeback(page
);
1007 page_cache_release(page
);
1010 set_extent_writeback(tree
, start
, end
, GFP_NOFS
);
1013 EXPORT_SYMBOL(set_range_writeback
);
1016 * find the first offset in the io tree with 'bits' set. zero is
1017 * returned if we find something, and *start_ret and *end_ret are
1018 * set to reflect the state struct that was found.
1020 * If nothing was found, 1 is returned, < 0 on error
1022 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1023 u64
*start_ret
, u64
*end_ret
, int bits
)
1025 struct rb_node
*node
;
1026 struct extent_state
*state
;
1029 spin_lock_irq(&tree
->lock
);
1031 * this search will find all the extents that end after
1034 node
= tree_search(tree
, start
);
1040 state
= rb_entry(node
, struct extent_state
, rb_node
);
1041 if (state
->end
>= start
&& (state
->state
& bits
)) {
1042 *start_ret
= state
->start
;
1043 *end_ret
= state
->end
;
1047 node
= rb_next(node
);
1052 spin_unlock_irq(&tree
->lock
);
1055 EXPORT_SYMBOL(find_first_extent_bit
);
1057 /* find the first state struct with 'bits' set after 'start', and
1058 * return it. tree->lock must be held. NULL will returned if
1059 * nothing was found after 'start'
1061 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
1062 u64 start
, int bits
)
1064 struct rb_node
*node
;
1065 struct extent_state
*state
;
1068 * this search will find all the extents that end after
1071 node
= tree_search(tree
, start
);
1077 state
= rb_entry(node
, struct extent_state
, rb_node
);
1078 if (state
->end
>= start
&& (state
->state
& bits
)) {
1081 node
= rb_next(node
);
1088 EXPORT_SYMBOL(find_first_extent_bit_state
);
1091 * find a contiguous range of bytes in the file marked as delalloc, not
1092 * more than 'max_bytes'. start and end are used to return the range,
1094 * 1 is returned if we find something, 0 if nothing was in the tree
1096 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1097 u64
*start
, u64
*end
, u64 max_bytes
)
1099 struct rb_node
*node
;
1100 struct extent_state
*state
;
1101 u64 cur_start
= *start
;
1103 u64 total_bytes
= 0;
1105 spin_lock_irq(&tree
->lock
);
1108 * this search will find all the extents that end after
1111 node
= tree_search(tree
, cur_start
);
1119 state
= rb_entry(node
, struct extent_state
, rb_node
);
1120 if (found
&& (state
->start
!= cur_start
||
1121 (state
->state
& EXTENT_BOUNDARY
))) {
1124 if (!(state
->state
& EXTENT_DELALLOC
)) {
1130 *start
= state
->start
;
1133 cur_start
= state
->end
+ 1;
1134 node
= rb_next(node
);
1137 total_bytes
+= state
->end
- state
->start
+ 1;
1138 if (total_bytes
>= max_bytes
)
1142 spin_unlock_irq(&tree
->lock
);
1146 static noinline
int __unlock_for_delalloc(struct inode
*inode
,
1147 struct page
*locked_page
,
1151 struct page
*pages
[16];
1152 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1153 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1154 unsigned long nr_pages
= end_index
- index
+ 1;
1157 if (index
== locked_page
->index
&& end_index
== index
)
1160 while(nr_pages
> 0) {
1161 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1162 min_t(unsigned long, nr_pages
,
1163 ARRAY_SIZE(pages
)), pages
);
1164 for (i
= 0; i
< ret
; i
++) {
1165 if (pages
[i
] != locked_page
)
1166 unlock_page(pages
[i
]);
1167 page_cache_release(pages
[i
]);
1176 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1177 struct page
*locked_page
,
1181 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1182 unsigned long start_index
= index
;
1183 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1184 unsigned long pages_locked
= 0;
1185 struct page
*pages
[16];
1186 unsigned long nrpages
;
1190 /* the caller is responsible for locking the start index */
1191 if (index
== locked_page
->index
&& index
== end_index
)
1194 /* skip the page at the start index */
1195 nrpages
= end_index
- index
+ 1;
1196 while(nrpages
> 0) {
1197 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1198 min_t(unsigned long,
1199 nrpages
, ARRAY_SIZE(pages
)), pages
);
1204 /* now we have an array of pages, lock them all */
1205 for (i
= 0; i
< ret
; i
++) {
1207 * the caller is taking responsibility for
1210 if (pages
[i
] != locked_page
) {
1211 lock_page(pages
[i
]);
1212 if (!PageDirty(pages
[i
]) ||
1213 pages
[i
]->mapping
!= inode
->i_mapping
) {
1215 unlock_page(pages
[i
]);
1216 page_cache_release(pages
[i
]);
1220 page_cache_release(pages
[i
]);
1229 if (ret
&& pages_locked
) {
1230 __unlock_for_delalloc(inode
, locked_page
,
1232 ((u64
)(start_index
+ pages_locked
- 1)) <<
1239 * find a contiguous range of bytes in the file marked as delalloc, not
1240 * more than 'max_bytes'. start and end are used to return the range,
1242 * 1 is returned if we find something, 0 if nothing was in the tree
1244 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1245 struct extent_io_tree
*tree
,
1246 struct page
*locked_page
,
1247 u64
*start
, u64
*end
,
1257 /* step one, find a bunch of delalloc bytes starting at start */
1258 delalloc_start
= *start
;
1260 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1262 if (!found
|| delalloc_end
<= *start
) {
1263 *start
= delalloc_start
;
1264 *end
= delalloc_end
;
1269 * start comes from the offset of locked_page. We have to lock
1270 * pages in order, so we can't process delalloc bytes before
1273 if (delalloc_start
< *start
) {
1274 delalloc_start
= *start
;
1278 * make sure to limit the number of pages we try to lock down
1281 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
) {
1282 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1284 /* step two, lock all the pages after the page that has start */
1285 ret
= lock_delalloc_pages(inode
, locked_page
,
1286 delalloc_start
, delalloc_end
);
1287 if (ret
== -EAGAIN
) {
1288 /* some of the pages are gone, lets avoid looping by
1289 * shortening the size of the delalloc range we're searching
1292 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1293 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1303 /* step three, lock the state bits for the whole range */
1304 lock_extent(tree
, delalloc_start
, delalloc_end
, GFP_NOFS
);
1306 /* then test to make sure it is all still delalloc */
1307 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1308 EXTENT_DELALLOC
, 1);
1310 unlock_extent(tree
, delalloc_start
, delalloc_end
, GFP_NOFS
);
1311 __unlock_for_delalloc(inode
, locked_page
,
1312 delalloc_start
, delalloc_end
);
1316 *start
= delalloc_start
;
1317 *end
= delalloc_end
;
1322 int extent_clear_unlock_delalloc(struct inode
*inode
,
1323 struct extent_io_tree
*tree
,
1324 u64 start
, u64 end
, struct page
*locked_page
,
1327 int clear_delalloc
, int clear_dirty
,
1332 struct page
*pages
[16];
1333 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1334 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1335 unsigned long nr_pages
= end_index
- index
+ 1;
1340 clear_bits
|= EXTENT_LOCKED
;
1342 clear_bits
|= EXTENT_DIRTY
;
1345 clear_bits
|= EXTENT_DELALLOC
;
1347 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, GFP_NOFS
);
1348 if (!(unlock_pages
|| clear_dirty
|| set_writeback
|| end_writeback
))
1351 while(nr_pages
> 0) {
1352 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1353 min_t(unsigned long,
1354 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1355 for (i
= 0; i
< ret
; i
++) {
1356 if (pages
[i
] == locked_page
) {
1357 page_cache_release(pages
[i
]);
1361 clear_page_dirty_for_io(pages
[i
]);
1363 set_page_writeback(pages
[i
]);
1365 end_page_writeback(pages
[i
]);
1367 unlock_page(pages
[i
]);
1368 page_cache_release(pages
[i
]);
1376 EXPORT_SYMBOL(extent_clear_unlock_delalloc
);
1379 * count the number of bytes in the tree that have a given bit(s)
1380 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1381 * cached. The total number found is returned.
1383 u64
count_range_bits(struct extent_io_tree
*tree
,
1384 u64
*start
, u64 search_end
, u64 max_bytes
,
1387 struct rb_node
*node
;
1388 struct extent_state
*state
;
1389 u64 cur_start
= *start
;
1390 u64 total_bytes
= 0;
1393 if (search_end
<= cur_start
) {
1394 printk("search_end %Lu start %Lu\n", search_end
, cur_start
);
1399 spin_lock_irq(&tree
->lock
);
1400 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1401 total_bytes
= tree
->dirty_bytes
;
1405 * this search will find all the extents that end after
1408 node
= tree_search(tree
, cur_start
);
1414 state
= rb_entry(node
, struct extent_state
, rb_node
);
1415 if (state
->start
> search_end
)
1417 if (state
->end
>= cur_start
&& (state
->state
& bits
)) {
1418 total_bytes
+= min(search_end
, state
->end
) + 1 -
1419 max(cur_start
, state
->start
);
1420 if (total_bytes
>= max_bytes
)
1423 *start
= state
->start
;
1427 node
= rb_next(node
);
1432 spin_unlock_irq(&tree
->lock
);
1436 * helper function to lock both pages and extents in the tree.
1437 * pages must be locked first.
1439 int lock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1441 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1442 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1446 while (index
<= end_index
) {
1447 page
= grab_cache_page(tree
->mapping
, index
);
1453 err
= PTR_ERR(page
);
1458 lock_extent(tree
, start
, end
, GFP_NOFS
);
1463 * we failed above in getting the page at 'index', so we undo here
1464 * up to but not including the page at 'index'
1467 index
= start
>> PAGE_CACHE_SHIFT
;
1468 while (index
< end_index
) {
1469 page
= find_get_page(tree
->mapping
, index
);
1471 page_cache_release(page
);
1476 EXPORT_SYMBOL(lock_range
);
1479 * helper function to unlock both pages and extents in the tree.
1481 int unlock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1483 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1484 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1487 while (index
<= end_index
) {
1488 page
= find_get_page(tree
->mapping
, index
);
1490 page_cache_release(page
);
1493 unlock_extent(tree
, start
, end
, GFP_NOFS
);
1496 EXPORT_SYMBOL(unlock_range
);
1499 * set the private field for a given byte offset in the tree. If there isn't
1500 * an extent_state there already, this does nothing.
1502 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1504 struct rb_node
*node
;
1505 struct extent_state
*state
;
1508 spin_lock_irq(&tree
->lock
);
1510 * this search will find all the extents that end after
1513 node
= tree_search(tree
, start
);
1518 state
= rb_entry(node
, struct extent_state
, rb_node
);
1519 if (state
->start
!= start
) {
1523 state
->private = private;
1525 spin_unlock_irq(&tree
->lock
);
1529 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1531 struct rb_node
*node
;
1532 struct extent_state
*state
;
1535 spin_lock_irq(&tree
->lock
);
1537 * this search will find all the extents that end after
1540 node
= tree_search(tree
, start
);
1545 state
= rb_entry(node
, struct extent_state
, rb_node
);
1546 if (state
->start
!= start
) {
1550 *private = state
->private;
1552 spin_unlock_irq(&tree
->lock
);
1557 * searches a range in the state tree for a given mask.
1558 * If 'filled' == 1, this returns 1 only if every extent in the tree
1559 * has the bits set. Otherwise, 1 is returned if any bit in the
1560 * range is found set.
1562 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1563 int bits
, int filled
)
1565 struct extent_state
*state
= NULL
;
1566 struct rb_node
*node
;
1568 unsigned long flags
;
1570 spin_lock_irqsave(&tree
->lock
, flags
);
1571 node
= tree_search(tree
, start
);
1572 while (node
&& start
<= end
) {
1573 state
= rb_entry(node
, struct extent_state
, rb_node
);
1575 if (filled
&& state
->start
> start
) {
1580 if (state
->start
> end
)
1583 if (state
->state
& bits
) {
1587 } else if (filled
) {
1591 start
= state
->end
+ 1;
1594 node
= rb_next(node
);
1601 spin_unlock_irqrestore(&tree
->lock
, flags
);
1604 EXPORT_SYMBOL(test_range_bit
);
1607 * helper function to set a given page up to date if all the
1608 * extents in the tree for that page are up to date
1610 static int check_page_uptodate(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_UPTODATE
, 1))
1616 SetPageUptodate(page
);
1621 * helper function to unlock a page if all the extents in the tree
1622 * for that page are unlocked
1624 static int check_page_locked(struct extent_io_tree
*tree
,
1627 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1628 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1629 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0))
1635 * helper function to end page writeback if all the extents
1636 * in the tree for that page are done with writeback
1638 static int check_page_writeback(struct extent_io_tree
*tree
,
1641 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1642 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1643 if (!test_range_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 0))
1644 end_page_writeback(page
);
1648 /* lots and lots of room for performance fixes in the end_bio funcs */
1651 * after a writepage IO is done, we need to:
1652 * clear the uptodate bits on error
1653 * clear the writeback bits in the extent tree for this IO
1654 * end_page_writeback if the page has no more pending IO
1656 * Scheduling is not allowed, so the extent state tree is expected
1657 * to have one and only one object corresponding to this IO.
1659 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1661 int uptodate
= err
== 0;
1662 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1663 struct extent_io_tree
*tree
;
1670 struct page
*page
= bvec
->bv_page
;
1671 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1673 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1675 end
= start
+ bvec
->bv_len
- 1;
1677 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1682 if (--bvec
>= bio
->bi_io_vec
)
1683 prefetchw(&bvec
->bv_page
->flags
);
1684 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1685 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
1686 end
, NULL
, uptodate
);
1691 if (!uptodate
&& tree
->ops
&&
1692 tree
->ops
->writepage_io_failed_hook
) {
1693 ret
= tree
->ops
->writepage_io_failed_hook(bio
, page
,
1696 uptodate
= (err
== 0);
1702 clear_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1703 ClearPageUptodate(page
);
1707 clear_extent_writeback(tree
, start
, end
, GFP_ATOMIC
);
1710 end_page_writeback(page
);
1712 check_page_writeback(tree
, page
);
1713 } while (bvec
>= bio
->bi_io_vec
);
1719 * after a readpage IO is done, we need to:
1720 * clear the uptodate bits on error
1721 * set the uptodate bits if things worked
1722 * set the page up to date if all extents in the tree are uptodate
1723 * clear the lock bit in the extent tree
1724 * unlock the page if there are no other extents locked for it
1726 * Scheduling is not allowed, so the extent state tree is expected
1727 * to have one and only one object corresponding to this IO.
1729 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1731 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1732 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1733 struct extent_io_tree
*tree
;
1740 struct page
*page
= bvec
->bv_page
;
1741 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1743 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1745 end
= start
+ bvec
->bv_len
- 1;
1747 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1752 if (--bvec
>= bio
->bi_io_vec
)
1753 prefetchw(&bvec
->bv_page
->flags
);
1755 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1756 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1761 if (!uptodate
&& tree
->ops
&&
1762 tree
->ops
->readpage_io_failed_hook
) {
1763 ret
= tree
->ops
->readpage_io_failed_hook(bio
, page
,
1767 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1773 set_extent_uptodate(tree
, start
, end
,
1776 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1780 SetPageUptodate(page
);
1782 ClearPageUptodate(page
);
1788 check_page_uptodate(tree
, page
);
1790 ClearPageUptodate(page
);
1793 check_page_locked(tree
, page
);
1795 } while (bvec
>= bio
->bi_io_vec
);
1801 * IO done from prepare_write is pretty simple, we just unlock
1802 * the structs in the extent tree when done, and set the uptodate bits
1805 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1807 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1808 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1809 struct extent_io_tree
*tree
;
1814 struct page
*page
= bvec
->bv_page
;
1815 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1817 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1819 end
= start
+ bvec
->bv_len
- 1;
1821 if (--bvec
>= bio
->bi_io_vec
)
1822 prefetchw(&bvec
->bv_page
->flags
);
1825 set_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1827 ClearPageUptodate(page
);
1831 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1833 } while (bvec
>= bio
->bi_io_vec
);
1839 extent_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1844 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1846 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1847 while (!bio
&& (nr_vecs
/= 2))
1848 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1853 bio
->bi_bdev
= bdev
;
1854 bio
->bi_sector
= first_sector
;
1859 static int submit_one_bio(int rw
, struct bio
*bio
, int mirror_num
,
1860 unsigned long bio_flags
)
1863 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1864 struct page
*page
= bvec
->bv_page
;
1865 struct extent_io_tree
*tree
= bio
->bi_private
;
1869 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1870 end
= start
+ bvec
->bv_len
- 1;
1872 bio
->bi_private
= NULL
;
1876 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
1877 tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
1878 mirror_num
, bio_flags
);
1880 submit_bio(rw
, bio
);
1881 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1887 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1888 struct page
*page
, sector_t sector
,
1889 size_t size
, unsigned long offset
,
1890 struct block_device
*bdev
,
1891 struct bio
**bio_ret
,
1892 unsigned long max_pages
,
1893 bio_end_io_t end_io_func
,
1895 unsigned long prev_bio_flags
,
1896 unsigned long bio_flags
)
1902 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
1903 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
1904 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
1906 if (bio_ret
&& *bio_ret
) {
1909 contig
= bio
->bi_sector
== sector
;
1911 contig
= bio
->bi_sector
+ (bio
->bi_size
>> 9) ==
1914 if (prev_bio_flags
!= bio_flags
|| !contig
||
1915 (tree
->ops
&& tree
->ops
->merge_bio_hook
&&
1916 tree
->ops
->merge_bio_hook(page
, offset
, page_size
, bio
,
1918 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
1919 ret
= submit_one_bio(rw
, bio
, mirror_num
,
1926 if (this_compressed
)
1929 nr
= bio_get_nr_vecs(bdev
);
1931 bio
= extent_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1933 printk("failed to allocate bio nr %d\n", nr
);
1936 bio_add_page(bio
, page
, page_size
, offset
);
1937 bio
->bi_end_io
= end_io_func
;
1938 bio
->bi_private
= tree
;
1943 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
1949 void set_page_extent_mapped(struct page
*page
)
1951 if (!PagePrivate(page
)) {
1952 SetPagePrivate(page
);
1953 page_cache_get(page
);
1954 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1957 EXPORT_SYMBOL(set_page_extent_mapped
);
1959 void set_page_extent_head(struct page
*page
, unsigned long len
)
1961 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1965 * basic readpage implementation. Locked extent state structs are inserted
1966 * into the tree that are removed when the IO is done (by the end_io
1969 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1971 get_extent_t
*get_extent
,
1972 struct bio
**bio
, int mirror_num
,
1973 unsigned long *bio_flags
)
1975 struct inode
*inode
= page
->mapping
->host
;
1976 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1977 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1981 u64 last_byte
= i_size_read(inode
);
1985 struct extent_map
*em
;
1986 struct block_device
*bdev
;
1989 size_t page_offset
= 0;
1991 size_t disk_io_size
;
1992 size_t blocksize
= inode
->i_sb
->s_blocksize
;
1993 unsigned long this_bio_flag
= 0;
1995 set_page_extent_mapped(page
);
1998 lock_extent(tree
, start
, end
, GFP_NOFS
);
2000 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2002 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2005 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2006 userpage
= kmap_atomic(page
, KM_USER0
);
2007 memset(userpage
+ zero_offset
, 0, iosize
);
2008 flush_dcache_page(page
);
2009 kunmap_atomic(userpage
, KM_USER0
);
2012 while (cur
<= end
) {
2013 if (cur
>= last_byte
) {
2015 iosize
= PAGE_CACHE_SIZE
- page_offset
;
2016 userpage
= kmap_atomic(page
, KM_USER0
);
2017 memset(userpage
+ page_offset
, 0, iosize
);
2018 flush_dcache_page(page
);
2019 kunmap_atomic(userpage
, KM_USER0
);
2020 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2022 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2025 em
= get_extent(inode
, page
, page_offset
, cur
,
2027 if (IS_ERR(em
) || !em
) {
2029 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
2032 extent_offset
= cur
- em
->start
;
2033 if (extent_map_end(em
) <= cur
) {
2034 printk("bad mapping em [%Lu %Lu] cur %Lu\n", em
->start
, extent_map_end(em
), cur
);
2036 BUG_ON(extent_map_end(em
) <= cur
);
2038 printk("2bad mapping end %Lu cur %Lu\n", end
, cur
);
2042 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
2043 this_bio_flag
= EXTENT_BIO_COMPRESSED
;
2045 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2046 cur_end
= min(extent_map_end(em
) - 1, end
);
2047 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2048 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2049 disk_io_size
= em
->block_len
;
2050 sector
= em
->block_start
>> 9;
2052 sector
= (em
->block_start
+ extent_offset
) >> 9;
2053 disk_io_size
= iosize
;
2056 block_start
= em
->block_start
;
2057 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2058 block_start
= EXTENT_MAP_HOLE
;
2059 free_extent_map(em
);
2062 /* we've found a hole, just zero and go on */
2063 if (block_start
== EXTENT_MAP_HOLE
) {
2065 userpage
= kmap_atomic(page
, KM_USER0
);
2066 memset(userpage
+ page_offset
, 0, iosize
);
2067 flush_dcache_page(page
);
2068 kunmap_atomic(userpage
, KM_USER0
);
2070 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2072 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2074 page_offset
+= iosize
;
2077 /* the get_extent function already copied into the page */
2078 if (test_range_bit(tree
, cur
, cur_end
, EXTENT_UPTODATE
, 1)) {
2079 check_page_uptodate(tree
, page
);
2080 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2082 page_offset
+= iosize
;
2085 /* we have an inline extent but it didn't get marked up
2086 * to date. Error out
2088 if (block_start
== EXTENT_MAP_INLINE
) {
2090 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2092 page_offset
+= iosize
;
2097 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
2098 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
2102 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2104 ret
= submit_extent_page(READ
, tree
, page
,
2105 sector
, disk_io_size
, page_offset
,
2107 end_bio_extent_readpage
, mirror_num
,
2111 *bio_flags
= this_bio_flag
;
2116 page_offset
+= iosize
;
2119 if (!PageError(page
))
2120 SetPageUptodate(page
);
2126 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2127 get_extent_t
*get_extent
)
2129 struct bio
*bio
= NULL
;
2130 unsigned long bio_flags
= 0;
2133 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, 0,
2136 submit_one_bio(READ
, bio
, 0, bio_flags
);
2139 EXPORT_SYMBOL(extent_read_full_page
);
2142 * the writepage semantics are similar to regular writepage. extent
2143 * records are inserted to lock ranges in the tree, and as dirty areas
2144 * are found, they are marked writeback. Then the lock bits are removed
2145 * and the end_io handler clears the writeback ranges
2147 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
2150 struct inode
*inode
= page
->mapping
->host
;
2151 struct extent_page_data
*epd
= data
;
2152 struct extent_io_tree
*tree
= epd
->tree
;
2153 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2155 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2159 u64 last_byte
= i_size_read(inode
);
2164 struct extent_map
*em
;
2165 struct block_device
*bdev
;
2168 size_t pg_offset
= 0;
2170 loff_t i_size
= i_size_read(inode
);
2171 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2176 unsigned long nr_written
= 0;
2178 WARN_ON(!PageLocked(page
));
2179 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
2180 if (page
->index
> end_index
||
2181 (page
->index
== end_index
&& !pg_offset
)) {
2182 page
->mapping
->a_ops
->invalidatepage(page
, 0);
2187 if (page
->index
== end_index
) {
2190 userpage
= kmap_atomic(page
, KM_USER0
);
2191 memset(userpage
+ pg_offset
, 0,
2192 PAGE_CACHE_SIZE
- pg_offset
);
2193 kunmap_atomic(userpage
, KM_USER0
);
2194 flush_dcache_page(page
);
2198 set_page_extent_mapped(page
);
2200 delalloc_start
= start
;
2203 if (!epd
->extent_locked
) {
2204 while(delalloc_end
< page_end
) {
2205 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
2210 if (nr_delalloc
== 0) {
2211 delalloc_start
= delalloc_end
+ 1;
2214 tree
->ops
->fill_delalloc(inode
, page
, delalloc_start
,
2215 delalloc_end
, &page_started
,
2217 delalloc_start
= delalloc_end
+ 1;
2220 /* did the fill delalloc function already unlock and start
2225 goto update_nr_written
;
2228 lock_extent(tree
, start
, page_end
, GFP_NOFS
);
2230 unlock_start
= start
;
2232 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
2233 ret
= tree
->ops
->writepage_start_hook(page
, start
,
2235 if (ret
== -EAGAIN
) {
2236 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
2237 redirty_page_for_writepage(wbc
, page
);
2240 goto update_nr_written
;
2247 if (test_range_bit(tree
, start
, page_end
, EXTENT_DELALLOC
, 0)) {
2248 printk("found delalloc bits after lock_extent\n");
2251 if (last_byte
<= start
) {
2252 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
2253 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
2254 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2255 tree
->ops
->writepage_end_io_hook(page
, start
,
2257 unlock_start
= page_end
+ 1;
2261 set_extent_uptodate(tree
, start
, page_end
, GFP_NOFS
);
2262 blocksize
= inode
->i_sb
->s_blocksize
;
2264 while (cur
<= end
) {
2265 if (cur
>= last_byte
) {
2266 clear_extent_dirty(tree
, cur
, page_end
, GFP_NOFS
);
2267 unlock_extent(tree
, unlock_start
, page_end
, GFP_NOFS
);
2268 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2269 tree
->ops
->writepage_end_io_hook(page
, cur
,
2271 unlock_start
= page_end
+ 1;
2274 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
2276 if (IS_ERR(em
) || !em
) {
2281 extent_offset
= cur
- em
->start
;
2282 BUG_ON(extent_map_end(em
) <= cur
);
2284 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2285 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2286 sector
= (em
->block_start
+ extent_offset
) >> 9;
2288 block_start
= em
->block_start
;
2289 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
2290 free_extent_map(em
);
2294 * compressed and inline extents are written through other
2297 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
2298 block_start
== EXTENT_MAP_INLINE
) {
2299 clear_extent_dirty(tree
, cur
,
2300 cur
+ iosize
- 1, GFP_NOFS
);
2302 unlock_extent(tree
, unlock_start
, cur
+ iosize
-1,
2306 * end_io notification does not happen here for
2307 * compressed extents
2309 if (!compressed
&& tree
->ops
&&
2310 tree
->ops
->writepage_end_io_hook
)
2311 tree
->ops
->writepage_end_io_hook(page
, cur
,
2314 else if (compressed
) {
2315 /* we don't want to end_page_writeback on
2316 * a compressed extent. this happens
2323 pg_offset
+= iosize
;
2327 /* leave this out until we have a page_mkwrite call */
2328 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2331 pg_offset
+= iosize
;
2335 clear_extent_dirty(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2336 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2337 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2345 unsigned long max_nr
= end_index
+ 1;
2347 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2348 if (!PageWriteback(page
)) {
2349 printk("warning page %lu not writeback, "
2350 "cur %llu end %llu\n", page
->index
,
2351 (unsigned long long)cur
,
2352 (unsigned long long)end
);
2355 ret
= submit_extent_page(WRITE
, tree
, page
, sector
,
2356 iosize
, pg_offset
, bdev
,
2358 end_bio_extent_writepage
,
2364 pg_offset
+= iosize
;
2369 /* make sure the mapping tag for page dirty gets cleared */
2370 set_page_writeback(page
);
2371 end_page_writeback(page
);
2373 if (unlock_start
<= page_end
)
2374 unlock_extent(tree
, unlock_start
, page_end
, GFP_NOFS
);
2378 wbc
->nr_to_write
-= nr_written
;
2379 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
2380 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
2381 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
2386 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2387 * @mapping: address space structure to write
2388 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2389 * @writepage: function called for each page
2390 * @data: data passed to writepage function
2392 * If a page is already under I/O, write_cache_pages() skips it, even
2393 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2394 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2395 * and msync() need to guarantee that all the data which was dirty at the time
2396 * the call was made get new I/O started against them. If wbc->sync_mode is
2397 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2398 * existing IO to complete.
2400 int extent_write_cache_pages(struct extent_io_tree
*tree
,
2401 struct address_space
*mapping
,
2402 struct writeback_control
*wbc
,
2403 writepage_t writepage
, void *data
,
2404 void (*flush_fn
)(void *))
2406 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
2409 struct pagevec pvec
;
2412 pgoff_t end
; /* Inclusive */
2414 int range_whole
= 0;
2416 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2417 wbc
->encountered_congestion
= 1;
2421 pagevec_init(&pvec
, 0);
2422 if (wbc
->range_cyclic
) {
2423 index
= mapping
->writeback_index
; /* Start from prev offset */
2426 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2427 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2428 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2433 while (!done
&& (index
<= end
) &&
2434 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2435 PAGECACHE_TAG_DIRTY
,
2436 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2440 for (i
= 0; i
< nr_pages
; i
++) {
2441 struct page
*page
= pvec
.pages
[i
];
2444 * At this point we hold neither mapping->tree_lock nor
2445 * lock on the page itself: the page may be truncated or
2446 * invalidated (changing page->mapping to NULL), or even
2447 * swizzled back from swapper_space to tmpfs file
2450 if (tree
->ops
&& tree
->ops
->write_cache_pages_lock_hook
)
2451 tree
->ops
->write_cache_pages_lock_hook(page
);
2455 if (unlikely(page
->mapping
!= mapping
)) {
2460 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2466 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
2467 if (PageWriteback(page
))
2469 wait_on_page_writeback(page
);
2472 if (PageWriteback(page
) ||
2473 !clear_page_dirty_for_io(page
)) {
2478 ret
= (*writepage
)(page
, wbc
, data
);
2480 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2484 if (ret
|| wbc
->nr_to_write
<= 0)
2486 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2487 wbc
->encountered_congestion
= 1;
2491 pagevec_release(&pvec
);
2494 if (!scanned
&& !done
) {
2496 * We hit the last page and there is more work to be done: wrap
2497 * back to the start of the file
2505 EXPORT_SYMBOL(extent_write_cache_pages
);
2507 static noinline
void flush_write_bio(void *data
)
2509 struct extent_page_data
*epd
= data
;
2511 submit_one_bio(WRITE
, epd
->bio
, 0, 0);
2516 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2517 get_extent_t
*get_extent
,
2518 struct writeback_control
*wbc
)
2521 struct address_space
*mapping
= page
->mapping
;
2522 struct extent_page_data epd
= {
2525 .get_extent
= get_extent
,
2528 struct writeback_control wbc_writepages
= {
2530 .sync_mode
= WB_SYNC_NONE
,
2531 .older_than_this
= NULL
,
2533 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2534 .range_end
= (loff_t
)-1,
2538 ret
= __extent_writepage(page
, wbc
, &epd
);
2540 extent_write_cache_pages(tree
, mapping
, &wbc_writepages
,
2541 __extent_writepage
, &epd
, flush_write_bio
);
2543 submit_one_bio(WRITE
, epd
.bio
, 0, 0);
2547 EXPORT_SYMBOL(extent_write_full_page
);
2549 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
2550 u64 start
, u64 end
, get_extent_t
*get_extent
,
2554 struct address_space
*mapping
= inode
->i_mapping
;
2556 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
2559 struct extent_page_data epd
= {
2562 .get_extent
= get_extent
,
2565 struct writeback_control wbc_writepages
= {
2566 .bdi
= inode
->i_mapping
->backing_dev_info
,
2568 .older_than_this
= NULL
,
2569 .nr_to_write
= nr_pages
* 2,
2570 .range_start
= start
,
2571 .range_end
= end
+ 1,
2574 while(start
<= end
) {
2575 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
2576 if (clear_page_dirty_for_io(page
))
2577 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
2579 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2580 tree
->ops
->writepage_end_io_hook(page
, start
,
2581 start
+ PAGE_CACHE_SIZE
- 1,
2585 page_cache_release(page
);
2586 start
+= PAGE_CACHE_SIZE
;
2590 submit_one_bio(WRITE
, epd
.bio
, 0, 0);
2593 EXPORT_SYMBOL(extent_write_locked_range
);
2596 int extent_writepages(struct extent_io_tree
*tree
,
2597 struct address_space
*mapping
,
2598 get_extent_t
*get_extent
,
2599 struct writeback_control
*wbc
)
2602 struct extent_page_data epd
= {
2605 .get_extent
= get_extent
,
2609 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
2610 __extent_writepage
, &epd
,
2613 submit_one_bio(WRITE
, epd
.bio
, 0, 0);
2617 EXPORT_SYMBOL(extent_writepages
);
2619 int extent_readpages(struct extent_io_tree
*tree
,
2620 struct address_space
*mapping
,
2621 struct list_head
*pages
, unsigned nr_pages
,
2622 get_extent_t get_extent
)
2624 struct bio
*bio
= NULL
;
2626 struct pagevec pvec
;
2627 unsigned long bio_flags
= 0;
2629 pagevec_init(&pvec
, 0);
2630 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2631 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2633 prefetchw(&page
->flags
);
2634 list_del(&page
->lru
);
2636 * what we want to do here is call add_to_page_cache_lru,
2637 * but that isn't exported, so we reproduce it here
2639 if (!add_to_page_cache(page
, mapping
,
2640 page
->index
, GFP_KERNEL
)) {
2642 /* open coding of lru_cache_add, also not exported */
2643 page_cache_get(page
);
2644 if (!pagevec_add(&pvec
, page
))
2645 __pagevec_lru_add_file(&pvec
);
2646 __extent_read_full_page(tree
, page
, get_extent
,
2647 &bio
, 0, &bio_flags
);
2649 page_cache_release(page
);
2651 if (pagevec_count(&pvec
))
2652 __pagevec_lru_add_file(&pvec
);
2653 BUG_ON(!list_empty(pages
));
2655 submit_one_bio(READ
, bio
, 0, bio_flags
);
2658 EXPORT_SYMBOL(extent_readpages
);
2661 * basic invalidatepage code, this waits on any locked or writeback
2662 * ranges corresponding to the page, and then deletes any extent state
2663 * records from the tree
2665 int extent_invalidatepage(struct extent_io_tree
*tree
,
2666 struct page
*page
, unsigned long offset
)
2668 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2669 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2670 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2672 start
+= (offset
+ blocksize
-1) & ~(blocksize
- 1);
2676 lock_extent(tree
, start
, end
, GFP_NOFS
);
2677 wait_on_extent_writeback(tree
, start
, end
);
2678 clear_extent_bit(tree
, start
, end
,
2679 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
,
2683 EXPORT_SYMBOL(extent_invalidatepage
);
2686 * simple commit_write call, set_range_dirty is used to mark both
2687 * the pages and the extent records as dirty
2689 int extent_commit_write(struct extent_io_tree
*tree
,
2690 struct inode
*inode
, struct page
*page
,
2691 unsigned from
, unsigned to
)
2693 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2695 set_page_extent_mapped(page
);
2696 set_page_dirty(page
);
2698 if (pos
> inode
->i_size
) {
2699 i_size_write(inode
, pos
);
2700 mark_inode_dirty(inode
);
2704 EXPORT_SYMBOL(extent_commit_write
);
2706 int extent_prepare_write(struct extent_io_tree
*tree
,
2707 struct inode
*inode
, struct page
*page
,
2708 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2710 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2711 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2713 u64 orig_block_start
;
2716 struct extent_map
*em
;
2717 unsigned blocksize
= 1 << inode
->i_blkbits
;
2718 size_t page_offset
= 0;
2719 size_t block_off_start
;
2720 size_t block_off_end
;
2726 set_page_extent_mapped(page
);
2728 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2729 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2730 orig_block_start
= block_start
;
2732 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2733 while(block_start
<= block_end
) {
2734 em
= get_extent(inode
, page
, page_offset
, block_start
,
2735 block_end
- block_start
+ 1, 1);
2736 if (IS_ERR(em
) || !em
) {
2739 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2740 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2741 block_off_end
= block_off_start
+ blocksize
;
2742 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2744 if (!PageUptodate(page
) && isnew
&&
2745 (block_off_end
> to
|| block_off_start
< from
)) {
2748 kaddr
= kmap_atomic(page
, KM_USER0
);
2749 if (block_off_end
> to
)
2750 memset(kaddr
+ to
, 0, block_off_end
- to
);
2751 if (block_off_start
< from
)
2752 memset(kaddr
+ block_off_start
, 0,
2753 from
- block_off_start
);
2754 flush_dcache_page(page
);
2755 kunmap_atomic(kaddr
, KM_USER0
);
2757 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2758 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2759 !isnew
&& !PageUptodate(page
) &&
2760 (block_off_end
> to
|| block_off_start
< from
) &&
2761 !test_range_bit(tree
, block_start
, cur_end
,
2762 EXTENT_UPTODATE
, 1)) {
2764 u64 extent_offset
= block_start
- em
->start
;
2766 sector
= (em
->block_start
+ extent_offset
) >> 9;
2767 iosize
= (cur_end
- block_start
+ blocksize
) &
2768 ~((u64
)blocksize
- 1);
2770 * we've already got the extent locked, but we
2771 * need to split the state such that our end_bio
2772 * handler can clear the lock.
2774 set_extent_bit(tree
, block_start
,
2775 block_start
+ iosize
- 1,
2776 EXTENT_LOCKED
, 0, NULL
, GFP_NOFS
);
2777 ret
= submit_extent_page(READ
, tree
, page
,
2778 sector
, iosize
, page_offset
, em
->bdev
,
2780 end_bio_extent_preparewrite
, 0,
2783 block_start
= block_start
+ iosize
;
2785 set_extent_uptodate(tree
, block_start
, cur_end
,
2787 unlock_extent(tree
, block_start
, cur_end
, GFP_NOFS
);
2788 block_start
= cur_end
+ 1;
2790 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2791 free_extent_map(em
);
2794 wait_extent_bit(tree
, orig_block_start
,
2795 block_end
, EXTENT_LOCKED
);
2797 check_page_uptodate(tree
, page
);
2799 /* FIXME, zero out newly allocated blocks on error */
2802 EXPORT_SYMBOL(extent_prepare_write
);
2805 * a helper for releasepage, this tests for areas of the page that
2806 * are locked or under IO and drops the related state bits if it is safe
2809 int try_release_extent_state(struct extent_map_tree
*map
,
2810 struct extent_io_tree
*tree
, struct page
*page
,
2813 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2814 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2817 if (test_range_bit(tree
, start
, end
,
2818 EXTENT_IOBITS
| EXTENT_ORDERED
, 0))
2821 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2823 clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
,
2828 EXPORT_SYMBOL(try_release_extent_state
);
2831 * a helper for releasepage. As long as there are no locked extents
2832 * in the range corresponding to the page, both state records and extent
2833 * map records are removed
2835 int try_release_extent_mapping(struct extent_map_tree
*map
,
2836 struct extent_io_tree
*tree
, struct page
*page
,
2839 struct extent_map
*em
;
2840 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2841 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2843 if ((mask
& __GFP_WAIT
) &&
2844 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2846 while (start
<= end
) {
2847 len
= end
- start
+ 1;
2848 spin_lock(&map
->lock
);
2849 em
= lookup_extent_mapping(map
, start
, len
);
2850 if (!em
|| IS_ERR(em
)) {
2851 spin_unlock(&map
->lock
);
2854 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
2855 em
->start
!= start
) {
2856 spin_unlock(&map
->lock
);
2857 free_extent_map(em
);
2860 if (!test_range_bit(tree
, em
->start
,
2861 extent_map_end(em
) - 1,
2862 EXTENT_LOCKED
| EXTENT_WRITEBACK
|
2865 remove_extent_mapping(map
, em
);
2866 /* once for the rb tree */
2867 free_extent_map(em
);
2869 start
= extent_map_end(em
);
2870 spin_unlock(&map
->lock
);
2873 free_extent_map(em
);
2876 return try_release_extent_state(map
, tree
, page
, mask
);
2878 EXPORT_SYMBOL(try_release_extent_mapping
);
2880 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2881 get_extent_t
*get_extent
)
2883 struct inode
*inode
= mapping
->host
;
2884 u64 start
= iblock
<< inode
->i_blkbits
;
2885 sector_t sector
= 0;
2886 size_t blksize
= (1 << inode
->i_blkbits
);
2887 struct extent_map
*em
;
2889 lock_extent(&BTRFS_I(inode
)->io_tree
, start
, start
+ blksize
- 1,
2891 em
= get_extent(inode
, NULL
, 0, start
, blksize
, 0);
2892 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
, start
+ blksize
- 1,
2894 if (!em
|| IS_ERR(em
))
2897 if (em
->block_start
> EXTENT_MAP_LAST_BYTE
)
2900 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2902 free_extent_map(em
);
2906 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
2910 struct address_space
*mapping
;
2913 return eb
->first_page
;
2914 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
2915 mapping
= eb
->first_page
->mapping
;
2920 * extent_buffer_page is only called after pinning the page
2921 * by increasing the reference count. So we know the page must
2922 * be in the radix tree.
2925 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
2931 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
2933 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
2934 (start
>> PAGE_CACHE_SHIFT
);
2937 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
2942 struct extent_buffer
*eb
= NULL
;
2944 unsigned long flags
;
2947 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
2950 mutex_init(&eb
->mutex
);
2952 spin_lock_irqsave(&leak_lock
, flags
);
2953 list_add(&eb
->leak_list
, &buffers
);
2954 spin_unlock_irqrestore(&leak_lock
, flags
);
2956 atomic_set(&eb
->refs
, 1);
2961 static void __free_extent_buffer(struct extent_buffer
*eb
)
2964 unsigned long flags
;
2965 spin_lock_irqsave(&leak_lock
, flags
);
2966 list_del(&eb
->leak_list
);
2967 spin_unlock_irqrestore(&leak_lock
, flags
);
2969 kmem_cache_free(extent_buffer_cache
, eb
);
2972 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
2973 u64 start
, unsigned long len
,
2977 unsigned long num_pages
= num_extent_pages(start
, len
);
2979 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2980 struct extent_buffer
*eb
;
2981 struct extent_buffer
*exists
= NULL
;
2983 struct address_space
*mapping
= tree
->mapping
;
2986 spin_lock(&tree
->buffer_lock
);
2987 eb
= buffer_search(tree
, start
);
2989 atomic_inc(&eb
->refs
);
2990 spin_unlock(&tree
->buffer_lock
);
2991 mark_page_accessed(eb
->first_page
);
2994 spin_unlock(&tree
->buffer_lock
);
2996 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
3001 eb
->first_page
= page0
;
3004 page_cache_get(page0
);
3005 mark_page_accessed(page0
);
3006 set_page_extent_mapped(page0
);
3007 set_page_extent_head(page0
, len
);
3008 uptodate
= PageUptodate(page0
);
3012 for (; i
< num_pages
; i
++, index
++) {
3013 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
3018 set_page_extent_mapped(p
);
3019 mark_page_accessed(p
);
3022 set_page_extent_head(p
, len
);
3024 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
3026 if (!PageUptodate(p
))
3031 eb
->flags
|= EXTENT_UPTODATE
;
3032 eb
->flags
|= EXTENT_BUFFER_FILLED
;
3034 spin_lock(&tree
->buffer_lock
);
3035 exists
= buffer_tree_insert(tree
, start
, &eb
->rb_node
);
3037 /* add one reference for the caller */
3038 atomic_inc(&exists
->refs
);
3039 spin_unlock(&tree
->buffer_lock
);
3042 spin_unlock(&tree
->buffer_lock
);
3044 /* add one reference for the tree */
3045 atomic_inc(&eb
->refs
);
3049 if (!atomic_dec_and_test(&eb
->refs
))
3051 for (index
= 1; index
< i
; index
++)
3052 page_cache_release(extent_buffer_page(eb
, index
));
3053 page_cache_release(extent_buffer_page(eb
, 0));
3054 __free_extent_buffer(eb
);
3057 EXPORT_SYMBOL(alloc_extent_buffer
);
3059 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
3060 u64 start
, unsigned long len
,
3063 struct extent_buffer
*eb
;
3065 spin_lock(&tree
->buffer_lock
);
3066 eb
= buffer_search(tree
, start
);
3068 atomic_inc(&eb
->refs
);
3069 spin_unlock(&tree
->buffer_lock
);
3072 mark_page_accessed(eb
->first_page
);
3076 EXPORT_SYMBOL(find_extent_buffer
);
3078 void free_extent_buffer(struct extent_buffer
*eb
)
3083 if (!atomic_dec_and_test(&eb
->refs
))
3088 EXPORT_SYMBOL(free_extent_buffer
);
3090 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
3091 struct extent_buffer
*eb
)
3095 unsigned long num_pages
;
3098 u64 start
= eb
->start
;
3099 u64 end
= start
+ eb
->len
- 1;
3101 set
= clear_extent_dirty(tree
, start
, end
, GFP_NOFS
);
3102 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3104 for (i
= 0; i
< num_pages
; i
++) {
3105 page
= extent_buffer_page(eb
, i
);
3106 if (!set
&& !PageDirty(page
))
3111 set_page_extent_head(page
, eb
->len
);
3113 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
3116 * if we're on the last page or the first page and the
3117 * block isn't aligned on a page boundary, do extra checks
3118 * to make sure we don't clean page that is partially dirty
3120 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
3121 ((i
== num_pages
- 1) &&
3122 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
3123 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3124 end
= start
+ PAGE_CACHE_SIZE
- 1;
3125 if (test_range_bit(tree
, start
, end
,
3131 clear_page_dirty_for_io(page
);
3132 spin_lock_irq(&page
->mapping
->tree_lock
);
3133 if (!PageDirty(page
)) {
3134 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3136 PAGECACHE_TAG_DIRTY
);
3138 spin_unlock_irq(&page
->mapping
->tree_lock
);
3143 EXPORT_SYMBOL(clear_extent_buffer_dirty
);
3145 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
3146 struct extent_buffer
*eb
)
3148 return wait_on_extent_writeback(tree
, eb
->start
,
3149 eb
->start
+ eb
->len
- 1);
3151 EXPORT_SYMBOL(wait_on_extent_buffer_writeback
);
3153 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
3154 struct extent_buffer
*eb
)
3157 unsigned long num_pages
;
3159 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3160 for (i
= 0; i
< num_pages
; i
++) {
3161 struct page
*page
= extent_buffer_page(eb
, i
);
3162 /* writepage may need to do something special for the
3163 * first page, we have to make sure page->private is
3164 * properly set. releasepage may drop page->private
3165 * on us if the page isn't already dirty.
3169 set_page_extent_head(page
, eb
->len
);
3170 } else if (PagePrivate(page
) &&
3171 page
->private != EXTENT_PAGE_PRIVATE
) {
3172 set_page_extent_mapped(page
);
3174 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
3175 set_extent_dirty(tree
, page_offset(page
),
3176 page_offset(page
) + PAGE_CACHE_SIZE
-1,
3182 EXPORT_SYMBOL(set_extent_buffer_dirty
);
3184 int clear_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3185 struct extent_buffer
*eb
)
3189 unsigned long num_pages
;
3191 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3192 eb
->flags
&= ~EXTENT_UPTODATE
;
3194 clear_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3196 for (i
= 0; i
< num_pages
; i
++) {
3197 page
= extent_buffer_page(eb
, i
);
3199 ClearPageUptodate(page
);
3204 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3205 struct extent_buffer
*eb
)
3209 unsigned long num_pages
;
3211 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3213 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3215 for (i
= 0; i
< num_pages
; i
++) {
3216 page
= extent_buffer_page(eb
, i
);
3217 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
3218 ((i
== num_pages
- 1) &&
3219 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
3220 check_page_uptodate(tree
, page
);
3223 SetPageUptodate(page
);
3227 EXPORT_SYMBOL(set_extent_buffer_uptodate
);
3229 int extent_range_uptodate(struct extent_io_tree
*tree
,
3234 int pg_uptodate
= 1;
3236 unsigned long index
;
3238 ret
= test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1);
3241 while(start
<= end
) {
3242 index
= start
>> PAGE_CACHE_SHIFT
;
3243 page
= find_get_page(tree
->mapping
, index
);
3244 uptodate
= PageUptodate(page
);
3245 page_cache_release(page
);
3250 start
+= PAGE_CACHE_SIZE
;
3255 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
3256 struct extent_buffer
*eb
)
3259 unsigned long num_pages
;
3262 int pg_uptodate
= 1;
3264 if (eb
->flags
& EXTENT_UPTODATE
)
3267 ret
= test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3268 EXTENT_UPTODATE
, 1);
3272 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3273 for (i
= 0; i
< num_pages
; i
++) {
3274 page
= extent_buffer_page(eb
, i
);
3275 if (!PageUptodate(page
)) {
3282 EXPORT_SYMBOL(extent_buffer_uptodate
);
3284 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
3285 struct extent_buffer
*eb
,
3286 u64 start
, int wait
,
3287 get_extent_t
*get_extent
, int mirror_num
)
3290 unsigned long start_i
;
3294 int locked_pages
= 0;
3295 int all_uptodate
= 1;
3296 int inc_all_pages
= 0;
3297 unsigned long num_pages
;
3298 struct bio
*bio
= NULL
;
3299 unsigned long bio_flags
= 0;
3301 if (eb
->flags
& EXTENT_UPTODATE
)
3304 if (test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3305 EXTENT_UPTODATE
, 1)) {
3310 WARN_ON(start
< eb
->start
);
3311 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
3312 (eb
->start
>> PAGE_CACHE_SHIFT
);
3317 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3318 for (i
= start_i
; i
< num_pages
; i
++) {
3319 page
= extent_buffer_page(eb
, i
);
3321 if (!trylock_page(page
))
3327 if (!PageUptodate(page
)) {
3333 eb
->flags
|= EXTENT_UPTODATE
;
3335 printk("all up to date but ret is %d\n", ret
);
3340 for (i
= start_i
; i
< num_pages
; i
++) {
3341 page
= extent_buffer_page(eb
, i
);
3343 page_cache_get(page
);
3344 if (!PageUptodate(page
)) {
3347 ClearPageError(page
);
3348 err
= __extent_read_full_page(tree
, page
,
3350 mirror_num
, &bio_flags
);
3353 printk("err %d from __extent_read_full_page\n", ret
);
3361 submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3365 printk("ret %d wait %d returning\n", ret
, wait
);
3368 for (i
= start_i
; i
< num_pages
; i
++) {
3369 page
= extent_buffer_page(eb
, i
);
3370 wait_on_page_locked(page
);
3371 if (!PageUptodate(page
)) {
3372 printk("page not uptodate after wait_on_page_locked\n");
3377 eb
->flags
|= EXTENT_UPTODATE
;
3382 while(locked_pages
> 0) {
3383 page
= extent_buffer_page(eb
, i
);
3390 EXPORT_SYMBOL(read_extent_buffer_pages
);
3392 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
3393 unsigned long start
,
3400 char *dst
= (char *)dstv
;
3401 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3402 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3404 WARN_ON(start
> eb
->len
);
3405 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3407 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3410 page
= extent_buffer_page(eb
, i
);
3412 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3413 kaddr
= kmap_atomic(page
, KM_USER1
);
3414 memcpy(dst
, kaddr
+ offset
, cur
);
3415 kunmap_atomic(kaddr
, KM_USER1
);
3423 EXPORT_SYMBOL(read_extent_buffer
);
3425 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3426 unsigned long min_len
, char **token
, char **map
,
3427 unsigned long *map_start
,
3428 unsigned long *map_len
, int km
)
3430 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
3433 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3434 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3435 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
3442 offset
= start_offset
;
3446 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
3448 if (start
+ min_len
> eb
->len
) {
3449 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb
->start
, eb
->len
, start
, min_len
);
3453 p
= extent_buffer_page(eb
, i
);
3454 kaddr
= kmap_atomic(p
, km
);
3456 *map
= kaddr
+ offset
;
3457 *map_len
= PAGE_CACHE_SIZE
- offset
;
3460 EXPORT_SYMBOL(map_private_extent_buffer
);
3462 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3463 unsigned long min_len
,
3464 char **token
, char **map
,
3465 unsigned long *map_start
,
3466 unsigned long *map_len
, int km
)
3470 if (eb
->map_token
) {
3471 unmap_extent_buffer(eb
, eb
->map_token
, km
);
3472 eb
->map_token
= NULL
;
3475 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
3476 map_start
, map_len
, km
);
3478 eb
->map_token
= *token
;
3480 eb
->map_start
= *map_start
;
3481 eb
->map_len
= *map_len
;
3485 EXPORT_SYMBOL(map_extent_buffer
);
3487 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
3489 kunmap_atomic(token
, km
);
3491 EXPORT_SYMBOL(unmap_extent_buffer
);
3493 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3494 unsigned long start
,
3501 char *ptr
= (char *)ptrv
;
3502 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3503 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3506 WARN_ON(start
> eb
->len
);
3507 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3509 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3512 page
= extent_buffer_page(eb
, i
);
3514 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3516 kaddr
= kmap_atomic(page
, KM_USER0
);
3517 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3518 kunmap_atomic(kaddr
, KM_USER0
);
3529 EXPORT_SYMBOL(memcmp_extent_buffer
);
3531 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3532 unsigned long start
, unsigned long len
)
3538 char *src
= (char *)srcv
;
3539 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3540 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3542 WARN_ON(start
> eb
->len
);
3543 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3545 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3548 page
= extent_buffer_page(eb
, i
);
3549 WARN_ON(!PageUptodate(page
));
3551 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3552 kaddr
= kmap_atomic(page
, KM_USER1
);
3553 memcpy(kaddr
+ offset
, src
, cur
);
3554 kunmap_atomic(kaddr
, KM_USER1
);
3562 EXPORT_SYMBOL(write_extent_buffer
);
3564 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3565 unsigned long start
, unsigned long len
)
3571 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3572 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3574 WARN_ON(start
> eb
->len
);
3575 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3577 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3580 page
= extent_buffer_page(eb
, i
);
3581 WARN_ON(!PageUptodate(page
));
3583 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3584 kaddr
= kmap_atomic(page
, KM_USER0
);
3585 memset(kaddr
+ offset
, c
, cur
);
3586 kunmap_atomic(kaddr
, KM_USER0
);
3593 EXPORT_SYMBOL(memset_extent_buffer
);
3595 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3596 unsigned long dst_offset
, unsigned long src_offset
,
3599 u64 dst_len
= dst
->len
;
3604 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3605 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3607 WARN_ON(src
->len
!= dst_len
);
3609 offset
= (start_offset
+ dst_offset
) &
3610 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3613 page
= extent_buffer_page(dst
, i
);
3614 WARN_ON(!PageUptodate(page
));
3616 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3618 kaddr
= kmap_atomic(page
, KM_USER0
);
3619 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3620 kunmap_atomic(kaddr
, KM_USER0
);
3628 EXPORT_SYMBOL(copy_extent_buffer
);
3630 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3631 unsigned long dst_off
, unsigned long src_off
,
3634 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3635 if (dst_page
== src_page
) {
3636 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3638 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3639 char *p
= dst_kaddr
+ dst_off
+ len
;
3640 char *s
= src_kaddr
+ src_off
+ len
;
3645 kunmap_atomic(src_kaddr
, KM_USER1
);
3647 kunmap_atomic(dst_kaddr
, KM_USER0
);
3650 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3651 unsigned long dst_off
, unsigned long src_off
,
3654 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3657 if (dst_page
!= src_page
)
3658 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3660 src_kaddr
= dst_kaddr
;
3662 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3663 kunmap_atomic(dst_kaddr
, KM_USER0
);
3664 if (dst_page
!= src_page
)
3665 kunmap_atomic(src_kaddr
, KM_USER1
);
3668 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3669 unsigned long src_offset
, unsigned long len
)
3672 size_t dst_off_in_page
;
3673 size_t src_off_in_page
;
3674 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3675 unsigned long dst_i
;
3676 unsigned long src_i
;
3678 if (src_offset
+ len
> dst
->len
) {
3679 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3680 src_offset
, len
, dst
->len
);
3683 if (dst_offset
+ len
> dst
->len
) {
3684 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3685 dst_offset
, len
, dst
->len
);
3690 dst_off_in_page
= (start_offset
+ dst_offset
) &
3691 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3692 src_off_in_page
= (start_offset
+ src_offset
) &
3693 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3695 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3696 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3698 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3700 cur
= min_t(unsigned long, cur
,
3701 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3703 copy_pages(extent_buffer_page(dst
, dst_i
),
3704 extent_buffer_page(dst
, src_i
),
3705 dst_off_in_page
, src_off_in_page
, cur
);
3712 EXPORT_SYMBOL(memcpy_extent_buffer
);
3714 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3715 unsigned long src_offset
, unsigned long len
)
3718 size_t dst_off_in_page
;
3719 size_t src_off_in_page
;
3720 unsigned long dst_end
= dst_offset
+ len
- 1;
3721 unsigned long src_end
= src_offset
+ len
- 1;
3722 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3723 unsigned long dst_i
;
3724 unsigned long src_i
;
3726 if (src_offset
+ len
> dst
->len
) {
3727 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3728 src_offset
, len
, dst
->len
);
3731 if (dst_offset
+ len
> dst
->len
) {
3732 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3733 dst_offset
, len
, dst
->len
);
3736 if (dst_offset
< src_offset
) {
3737 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
3741 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
3742 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
3744 dst_off_in_page
= (start_offset
+ dst_end
) &
3745 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3746 src_off_in_page
= (start_offset
+ src_end
) &
3747 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3749 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
3750 cur
= min(cur
, dst_off_in_page
+ 1);
3751 move_pages(extent_buffer_page(dst
, dst_i
),
3752 extent_buffer_page(dst
, src_i
),
3753 dst_off_in_page
- cur
+ 1,
3754 src_off_in_page
- cur
+ 1, cur
);
3761 EXPORT_SYMBOL(memmove_extent_buffer
);
3763 int try_release_extent_buffer(struct extent_io_tree
*tree
, struct page
*page
)
3765 u64 start
= page_offset(page
);
3766 struct extent_buffer
*eb
;
3769 unsigned long num_pages
;
3771 spin_lock(&tree
->buffer_lock
);
3772 eb
= buffer_search(tree
, start
);
3776 if (atomic_read(&eb
->refs
) > 1) {
3780 /* at this point we can safely release the extent buffer */
3781 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3782 for (i
= 0; i
< num_pages
; i
++)
3783 page_cache_release(extent_buffer_page(eb
, i
));
3784 rb_erase(&eb
->rb_node
, &tree
->buffer
);
3785 __free_extent_buffer(eb
);
3787 spin_unlock(&tree
->buffer_lock
);
3790 EXPORT_SYMBOL(try_release_extent_buffer
);