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"
18 /* temporary define until extent_map moves out of btrfs */
19 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
20 unsigned long extra_flags
,
21 void (*ctor
)(void *, struct kmem_cache
*,
24 static struct kmem_cache
*extent_state_cache
;
25 static struct kmem_cache
*extent_buffer_cache
;
27 static LIST_HEAD(buffers
);
28 static LIST_HEAD(states
);
29 static spinlock_t leak_lock
= SPIN_LOCK_UNLOCKED
;
31 #define BUFFER_LRU_MAX 64
36 struct rb_node rb_node
;
39 struct extent_page_data
{
41 struct extent_io_tree
*tree
;
42 get_extent_t
*get_extent
;
45 int __init
extent_io_init(void)
47 extent_state_cache
= btrfs_cache_create("extent_state",
48 sizeof(struct extent_state
), 0,
50 if (!extent_state_cache
)
53 extent_buffer_cache
= btrfs_cache_create("extent_buffers",
54 sizeof(struct extent_buffer
), 0,
56 if (!extent_buffer_cache
)
57 goto free_state_cache
;
61 kmem_cache_destroy(extent_state_cache
);
65 void extent_io_exit(void)
67 struct extent_state
*state
;
68 struct extent_buffer
*eb
;
70 while (!list_empty(&states
)) {
71 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
72 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
));
73 list_del(&state
->leak_list
);
74 kmem_cache_free(extent_state_cache
, state
);
78 while (!list_empty(&buffers
)) {
79 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
80 printk("buffer leak start %Lu len %lu refs %d\n", eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
81 list_del(&eb
->leak_list
);
82 kmem_cache_free(extent_buffer_cache
, eb
);
84 if (extent_state_cache
)
85 kmem_cache_destroy(extent_state_cache
);
86 if (extent_buffer_cache
)
87 kmem_cache_destroy(extent_buffer_cache
);
90 void extent_io_tree_init(struct extent_io_tree
*tree
,
91 struct address_space
*mapping
, gfp_t mask
)
93 tree
->state
.rb_node
= NULL
;
95 tree
->dirty_bytes
= 0;
96 spin_lock_init(&tree
->lock
);
97 spin_lock_init(&tree
->lru_lock
);
98 tree
->mapping
= mapping
;
99 INIT_LIST_HEAD(&tree
->buffer_lru
);
103 EXPORT_SYMBOL(extent_io_tree_init
);
105 void extent_io_tree_empty_lru(struct extent_io_tree
*tree
)
107 struct extent_buffer
*eb
;
108 while(!list_empty(&tree
->buffer_lru
)) {
109 eb
= list_entry(tree
->buffer_lru
.next
, struct extent_buffer
,
111 list_del_init(&eb
->lru
);
112 free_extent_buffer(eb
);
115 EXPORT_SYMBOL(extent_io_tree_empty_lru
);
117 struct extent_state
*alloc_extent_state(gfp_t mask
)
119 struct extent_state
*state
;
122 state
= kmem_cache_alloc(extent_state_cache
, mask
);
128 spin_lock_irqsave(&leak_lock
, flags
);
129 list_add(&state
->leak_list
, &states
);
130 spin_unlock_irqrestore(&leak_lock
, flags
);
132 atomic_set(&state
->refs
, 1);
133 init_waitqueue_head(&state
->wq
);
136 EXPORT_SYMBOL(alloc_extent_state
);
138 void free_extent_state(struct extent_state
*state
)
142 if (atomic_dec_and_test(&state
->refs
)) {
144 WARN_ON(state
->tree
);
145 spin_lock_irqsave(&leak_lock
, flags
);
146 list_del(&state
->leak_list
);
147 spin_unlock_irqrestore(&leak_lock
, flags
);
148 kmem_cache_free(extent_state_cache
, state
);
151 EXPORT_SYMBOL(free_extent_state
);
153 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
154 struct rb_node
*node
)
156 struct rb_node
** p
= &root
->rb_node
;
157 struct rb_node
* parent
= NULL
;
158 struct tree_entry
*entry
;
162 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
164 if (offset
< entry
->start
)
166 else if (offset
> entry
->end
)
172 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
173 rb_link_node(node
, parent
, p
);
174 rb_insert_color(node
, root
);
178 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
179 struct rb_node
**prev_ret
,
180 struct rb_node
**next_ret
)
182 struct rb_root
*root
= &tree
->state
;
183 struct rb_node
* n
= root
->rb_node
;
184 struct rb_node
*prev
= NULL
;
185 struct rb_node
*orig_prev
= NULL
;
186 struct tree_entry
*entry
;
187 struct tree_entry
*prev_entry
= NULL
;
190 struct extent_state
*state
;
192 if (state
->start
<= offset
&& offset
<= state
->end
)
193 return &tree
->last
->rb_node
;
196 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
200 if (offset
< entry
->start
)
202 else if (offset
> entry
->end
)
205 tree
->last
= rb_entry(n
, struct extent_state
, rb_node
);
212 while(prev
&& offset
> prev_entry
->end
) {
213 prev
= rb_next(prev
);
214 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
221 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
222 while(prev
&& offset
< prev_entry
->start
) {
223 prev
= rb_prev(prev
);
224 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
231 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
234 struct rb_node
*prev
= NULL
;
237 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
240 tree
->last
= rb_entry(prev
, struct extent_state
,
249 * utility function to look for merge candidates inside a given range.
250 * Any extents with matching state are merged together into a single
251 * extent in the tree. Extents with EXTENT_IO in their state field
252 * are not merged because the end_io handlers need to be able to do
253 * operations on them without sleeping (or doing allocations/splits).
255 * This should be called with the tree lock held.
257 static int merge_state(struct extent_io_tree
*tree
,
258 struct extent_state
*state
)
260 struct extent_state
*other
;
261 struct rb_node
*other_node
;
263 if (state
->state
& EXTENT_IOBITS
)
266 other_node
= rb_prev(&state
->rb_node
);
268 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
269 if (other
->end
== state
->start
- 1 &&
270 other
->state
== state
->state
) {
271 state
->start
= other
->start
;
273 if (tree
->last
== other
)
275 rb_erase(&other
->rb_node
, &tree
->state
);
276 free_extent_state(other
);
279 other_node
= rb_next(&state
->rb_node
);
281 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
282 if (other
->start
== state
->end
+ 1 &&
283 other
->state
== state
->state
) {
284 other
->start
= state
->start
;
286 if (tree
->last
== state
)
288 rb_erase(&state
->rb_node
, &tree
->state
);
289 free_extent_state(state
);
295 static void set_state_cb(struct extent_io_tree
*tree
,
296 struct extent_state
*state
,
299 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
300 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
->start
,
301 state
->end
, state
->state
, bits
);
305 static void clear_state_cb(struct extent_io_tree
*tree
,
306 struct extent_state
*state
,
309 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
310 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
->start
,
311 state
->end
, state
->state
, bits
);
316 * insert an extent_state struct into the tree. 'bits' are set on the
317 * struct before it is inserted.
319 * This may return -EEXIST if the extent is already there, in which case the
320 * state struct is freed.
322 * The tree lock is not taken internally. This is a utility function and
323 * probably isn't what you want to call (see set/clear_extent_bit).
325 static int insert_state(struct extent_io_tree
*tree
,
326 struct extent_state
*state
, u64 start
, u64 end
,
329 struct rb_node
*node
;
332 printk("end < start %Lu %Lu\n", end
, start
);
335 if (bits
& EXTENT_DIRTY
)
336 tree
->dirty_bytes
+= end
- start
+ 1;
337 set_state_cb(tree
, state
, bits
);
338 state
->state
|= bits
;
339 state
->start
= start
;
341 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
343 struct extent_state
*found
;
344 found
= rb_entry(node
, struct extent_state
, rb_node
);
345 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, start
, end
);
346 free_extent_state(state
);
351 merge_state(tree
, state
);
356 * split a given extent state struct in two, inserting the preallocated
357 * struct 'prealloc' as the newly created second half. 'split' indicates an
358 * offset inside 'orig' where it should be split.
361 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
362 * are two extent state structs in the tree:
363 * prealloc: [orig->start, split - 1]
364 * orig: [ split, orig->end ]
366 * The tree locks are not taken by this function. They need to be held
369 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
370 struct extent_state
*prealloc
, u64 split
)
372 struct rb_node
*node
;
373 prealloc
->start
= orig
->start
;
374 prealloc
->end
= split
- 1;
375 prealloc
->state
= orig
->state
;
378 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
380 struct extent_state
*found
;
381 found
= rb_entry(node
, struct extent_state
, rb_node
);
382 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, prealloc
->start
, prealloc
->end
);
383 free_extent_state(prealloc
);
386 prealloc
->tree
= tree
;
391 * utility function to clear some bits in an extent state struct.
392 * it will optionally wake up any one waiting on this state (wake == 1), or
393 * forcibly remove the state from the tree (delete == 1).
395 * If no bits are set on the state struct after clearing things, the
396 * struct is freed and removed from the tree
398 static int clear_state_bit(struct extent_io_tree
*tree
,
399 struct extent_state
*state
, int bits
, int wake
,
402 int ret
= state
->state
& bits
;
404 if ((bits
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
405 u64 range
= state
->end
- state
->start
+ 1;
406 WARN_ON(range
> tree
->dirty_bytes
);
407 tree
->dirty_bytes
-= range
;
409 clear_state_cb(tree
, state
, bits
);
410 state
->state
&= ~bits
;
413 if (delete || state
->state
== 0) {
415 clear_state_cb(tree
, state
, state
->state
);
416 if (tree
->last
== state
) {
417 tree
->last
= extent_state_next(state
);
419 rb_erase(&state
->rb_node
, &tree
->state
);
421 free_extent_state(state
);
426 merge_state(tree
, state
);
432 * clear some bits on a range in the tree. This may require splitting
433 * or inserting elements in the tree, so the gfp mask is used to
434 * indicate which allocations or sleeping are allowed.
436 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
437 * the given range from the tree regardless of state (ie for truncate).
439 * the range [start, end] is inclusive.
441 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
442 * bits were already set, or zero if none of the bits were already set.
444 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
445 int bits
, int wake
, int delete, gfp_t mask
)
447 struct extent_state
*state
;
448 struct extent_state
*prealloc
= NULL
;
449 struct rb_node
*node
;
455 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
456 prealloc
= alloc_extent_state(mask
);
461 spin_lock_irqsave(&tree
->lock
, flags
);
463 * this search will find the extents that end after
466 node
= tree_search(tree
, start
);
469 state
= rb_entry(node
, struct extent_state
, rb_node
);
470 if (state
->start
> end
)
472 WARN_ON(state
->end
< start
);
475 * | ---- desired range ---- |
477 * | ------------- state -------------- |
479 * We need to split the extent we found, and may flip
480 * bits on second half.
482 * If the extent we found extends past our range, we
483 * just split and search again. It'll get split again
484 * the next time though.
486 * If the extent we found is inside our range, we clear
487 * the desired bit on it.
490 if (state
->start
< start
) {
492 prealloc
= alloc_extent_state(GFP_ATOMIC
);
493 err
= split_state(tree
, state
, prealloc
, start
);
494 BUG_ON(err
== -EEXIST
);
498 if (state
->end
<= end
) {
499 start
= state
->end
+ 1;
500 set
|= clear_state_bit(tree
, state
, bits
,
503 start
= state
->start
;
508 * | ---- desired range ---- |
510 * We need to split the extent, and clear the bit
513 if (state
->start
<= end
&& state
->end
> end
) {
515 prealloc
= alloc_extent_state(GFP_ATOMIC
);
516 err
= split_state(tree
, state
, prealloc
, end
+ 1);
517 BUG_ON(err
== -EEXIST
);
521 set
|= clear_state_bit(tree
, prealloc
, bits
,
527 start
= state
->end
+ 1;
528 set
|= clear_state_bit(tree
, state
, bits
, wake
, delete);
532 spin_unlock_irqrestore(&tree
->lock
, flags
);
534 free_extent_state(prealloc
);
541 spin_unlock_irqrestore(&tree
->lock
, flags
);
542 if (mask
& __GFP_WAIT
)
546 EXPORT_SYMBOL(clear_extent_bit
);
548 static int wait_on_state(struct extent_io_tree
*tree
,
549 struct extent_state
*state
)
552 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
553 spin_unlock_irq(&tree
->lock
);
555 spin_lock_irq(&tree
->lock
);
556 finish_wait(&state
->wq
, &wait
);
561 * waits for one or more bits to clear on a range in the state tree.
562 * The range [start, end] is inclusive.
563 * The tree lock is taken by this function
565 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
567 struct extent_state
*state
;
568 struct rb_node
*node
;
570 spin_lock_irq(&tree
->lock
);
574 * this search will find all the extents that end after
577 node
= tree_search(tree
, start
);
581 state
= rb_entry(node
, struct extent_state
, rb_node
);
583 if (state
->start
> end
)
586 if (state
->state
& bits
) {
587 start
= state
->start
;
588 atomic_inc(&state
->refs
);
589 wait_on_state(tree
, state
);
590 free_extent_state(state
);
593 start
= state
->end
+ 1;
598 if (need_resched()) {
599 spin_unlock_irq(&tree
->lock
);
601 spin_lock_irq(&tree
->lock
);
605 spin_unlock_irq(&tree
->lock
);
608 EXPORT_SYMBOL(wait_extent_bit
);
610 static void set_state_bits(struct extent_io_tree
*tree
,
611 struct extent_state
*state
,
614 if ((bits
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
615 u64 range
= state
->end
- state
->start
+ 1;
616 tree
->dirty_bytes
+= range
;
618 set_state_cb(tree
, state
, bits
);
619 state
->state
|= bits
;
623 * set some bits on a range in the tree. This may require allocations
624 * or sleeping, so the gfp mask is used to indicate what is allowed.
626 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
627 * range already has the desired bits set. The start of the existing
628 * range is returned in failed_start in this case.
630 * [start, end] is inclusive
631 * This takes the tree lock.
633 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
,
634 int exclusive
, u64
*failed_start
, gfp_t mask
)
636 struct extent_state
*state
;
637 struct extent_state
*prealloc
= NULL
;
638 struct rb_node
*node
;
645 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
646 prealloc
= alloc_extent_state(mask
);
651 spin_lock_irqsave(&tree
->lock
, flags
);
653 * this search will find all the extents that end after
656 node
= tree_search(tree
, start
);
658 err
= insert_state(tree
, prealloc
, start
, end
, bits
);
660 BUG_ON(err
== -EEXIST
);
664 state
= rb_entry(node
, struct extent_state
, rb_node
);
665 last_start
= state
->start
;
666 last_end
= state
->end
;
669 * | ---- desired range ---- |
672 * Just lock what we found and keep going
674 if (state
->start
== start
&& state
->end
<= end
) {
675 set
= state
->state
& bits
;
676 if (set
&& exclusive
) {
677 *failed_start
= state
->start
;
681 set_state_bits(tree
, state
, bits
);
682 start
= state
->end
+ 1;
683 merge_state(tree
, state
);
688 * | ---- desired range ---- |
691 * | ------------- state -------------- |
693 * We need to split the extent we found, and may flip bits on
696 * If the extent we found extends past our
697 * range, we just split and search again. It'll get split
698 * again the next time though.
700 * If the extent we found is inside our range, we set the
703 if (state
->start
< start
) {
704 set
= state
->state
& bits
;
705 if (exclusive
&& set
) {
706 *failed_start
= start
;
710 err
= split_state(tree
, state
, prealloc
, start
);
711 BUG_ON(err
== -EEXIST
);
715 if (state
->end
<= end
) {
716 set_state_bits(tree
, state
, bits
);
717 start
= state
->end
+ 1;
718 merge_state(tree
, state
);
720 start
= state
->start
;
725 * | ---- desired range ---- |
726 * | state | or | state |
728 * There's a hole, we need to insert something in it and
729 * ignore the extent we found.
731 if (state
->start
> start
) {
733 if (end
< last_start
)
736 this_end
= last_start
-1;
737 err
= insert_state(tree
, prealloc
, start
, this_end
,
740 BUG_ON(err
== -EEXIST
);
743 start
= this_end
+ 1;
747 * | ---- desired range ---- |
749 * We need to split the extent, and set the bit
752 if (state
->start
<= end
&& state
->end
> end
) {
753 set
= state
->state
& bits
;
754 if (exclusive
&& set
) {
755 *failed_start
= start
;
759 err
= split_state(tree
, state
, prealloc
, end
+ 1);
760 BUG_ON(err
== -EEXIST
);
762 set_state_bits(tree
, prealloc
, bits
);
763 merge_state(tree
, prealloc
);
771 spin_unlock_irqrestore(&tree
->lock
, flags
);
773 free_extent_state(prealloc
);
780 spin_unlock_irqrestore(&tree
->lock
, flags
);
781 if (mask
& __GFP_WAIT
)
785 EXPORT_SYMBOL(set_extent_bit
);
787 /* wrappers around set/clear extent bit */
788 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
791 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
794 EXPORT_SYMBOL(set_extent_dirty
);
796 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
797 int bits
, gfp_t mask
)
799 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
802 EXPORT_SYMBOL(set_extent_bits
);
804 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
805 int bits
, gfp_t mask
)
807 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, mask
);
809 EXPORT_SYMBOL(clear_extent_bits
);
811 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
814 return set_extent_bit(tree
, start
, end
,
815 EXTENT_DELALLOC
| EXTENT_DIRTY
, 0, NULL
,
818 EXPORT_SYMBOL(set_extent_delalloc
);
820 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
823 return clear_extent_bit(tree
, start
, end
,
824 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, mask
);
826 EXPORT_SYMBOL(clear_extent_dirty
);
828 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
831 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
834 EXPORT_SYMBOL(set_extent_new
);
836 int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
839 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0, mask
);
841 EXPORT_SYMBOL(clear_extent_new
);
843 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
846 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, NULL
,
849 EXPORT_SYMBOL(set_extent_uptodate
);
851 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
854 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0, mask
);
856 EXPORT_SYMBOL(clear_extent_uptodate
);
858 int set_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
861 return set_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
,
864 EXPORT_SYMBOL(set_extent_writeback
);
866 int clear_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
869 return clear_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 1, 0, mask
);
871 EXPORT_SYMBOL(clear_extent_writeback
);
873 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
875 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
877 EXPORT_SYMBOL(wait_on_extent_writeback
);
879 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
884 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
885 &failed_start
, mask
);
886 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
887 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
888 start
= failed_start
;
892 WARN_ON(start
> end
);
896 EXPORT_SYMBOL(lock_extent
);
898 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
901 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, mask
);
903 EXPORT_SYMBOL(unlock_extent
);
906 * helper function to set pages and extents in the tree dirty
908 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
910 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
911 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
914 while (index
<= end_index
) {
915 page
= find_get_page(tree
->mapping
, index
);
917 __set_page_dirty_nobuffers(page
);
918 page_cache_release(page
);
921 set_extent_dirty(tree
, start
, end
, GFP_NOFS
);
924 EXPORT_SYMBOL(set_range_dirty
);
927 * helper function to set both pages and extents in the tree writeback
929 int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
931 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
932 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
935 while (index
<= end_index
) {
936 page
= find_get_page(tree
->mapping
, index
);
938 set_page_writeback(page
);
939 page_cache_release(page
);
942 set_extent_writeback(tree
, start
, end
, GFP_NOFS
);
945 EXPORT_SYMBOL(set_range_writeback
);
947 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
948 u64
*start_ret
, u64
*end_ret
, int bits
)
950 struct rb_node
*node
;
951 struct extent_state
*state
;
954 spin_lock_irq(&tree
->lock
);
956 * this search will find all the extents that end after
959 node
= tree_search(tree
, start
);
965 state
= rb_entry(node
, struct extent_state
, rb_node
);
966 if (state
->end
>= start
&& (state
->state
& bits
)) {
967 *start_ret
= state
->start
;
968 *end_ret
= state
->end
;
972 node
= rb_next(node
);
977 spin_unlock_irq(&tree
->lock
);
980 EXPORT_SYMBOL(find_first_extent_bit
);
982 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
985 struct rb_node
*node
;
986 struct extent_state
*state
;
989 * this search will find all the extents that end after
992 node
= tree_search(tree
, start
);
998 state
= rb_entry(node
, struct extent_state
, rb_node
);
999 if (state
->end
>= start
&& (state
->state
& bits
)) {
1002 node
= rb_next(node
);
1009 EXPORT_SYMBOL(find_first_extent_bit_state
);
1011 u64
find_lock_delalloc_range(struct extent_io_tree
*tree
,
1012 u64
*start
, u64
*end
, u64 max_bytes
)
1014 struct rb_node
*node
;
1015 struct extent_state
*state
;
1016 u64 cur_start
= *start
;
1018 u64 total_bytes
= 0;
1020 spin_lock_irq(&tree
->lock
);
1022 * this search will find all the extents that end after
1026 node
= tree_search(tree
, cur_start
);
1034 state
= rb_entry(node
, struct extent_state
, rb_node
);
1035 if (found
&& state
->start
!= cur_start
) {
1038 if (!(state
->state
& EXTENT_DELALLOC
)) {
1044 struct extent_state
*prev_state
;
1045 struct rb_node
*prev_node
= node
;
1047 prev_node
= rb_prev(prev_node
);
1050 prev_state
= rb_entry(prev_node
,
1051 struct extent_state
,
1053 if (!(prev_state
->state
& EXTENT_DELALLOC
))
1059 if (state
->state
& EXTENT_LOCKED
) {
1061 atomic_inc(&state
->refs
);
1062 prepare_to_wait(&state
->wq
, &wait
,
1063 TASK_UNINTERRUPTIBLE
);
1064 spin_unlock_irq(&tree
->lock
);
1066 spin_lock_irq(&tree
->lock
);
1067 finish_wait(&state
->wq
, &wait
);
1068 free_extent_state(state
);
1071 set_state_cb(tree
, state
, EXTENT_LOCKED
);
1072 state
->state
|= EXTENT_LOCKED
;
1074 *start
= state
->start
;
1077 cur_start
= state
->end
+ 1;
1078 node
= rb_next(node
);
1081 total_bytes
+= state
->end
- state
->start
+ 1;
1082 if (total_bytes
>= max_bytes
)
1086 spin_unlock_irq(&tree
->lock
);
1090 u64
count_range_bits(struct extent_io_tree
*tree
,
1091 u64
*start
, u64 search_end
, u64 max_bytes
,
1094 struct rb_node
*node
;
1095 struct extent_state
*state
;
1096 u64 cur_start
= *start
;
1097 u64 total_bytes
= 0;
1100 if (search_end
<= cur_start
) {
1101 printk("search_end %Lu start %Lu\n", search_end
, cur_start
);
1106 spin_lock_irq(&tree
->lock
);
1107 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1108 total_bytes
= tree
->dirty_bytes
;
1112 * this search will find all the extents that end after
1115 node
= tree_search(tree
, cur_start
);
1121 state
= rb_entry(node
, struct extent_state
, rb_node
);
1122 if (state
->start
> search_end
)
1124 if (state
->end
>= cur_start
&& (state
->state
& bits
)) {
1125 total_bytes
+= min(search_end
, state
->end
) + 1 -
1126 max(cur_start
, state
->start
);
1127 if (total_bytes
>= max_bytes
)
1130 *start
= state
->start
;
1134 node
= rb_next(node
);
1139 spin_unlock_irq(&tree
->lock
);
1143 * helper function to lock both pages and extents in the tree.
1144 * pages must be locked first.
1146 int lock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1148 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1149 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1153 while (index
<= end_index
) {
1154 page
= grab_cache_page(tree
->mapping
, index
);
1160 err
= PTR_ERR(page
);
1165 lock_extent(tree
, start
, end
, GFP_NOFS
);
1170 * we failed above in getting the page at 'index', so we undo here
1171 * up to but not including the page at 'index'
1174 index
= start
>> PAGE_CACHE_SHIFT
;
1175 while (index
< end_index
) {
1176 page
= find_get_page(tree
->mapping
, index
);
1178 page_cache_release(page
);
1183 EXPORT_SYMBOL(lock_range
);
1186 * helper function to unlock both pages and extents in the tree.
1188 int unlock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1190 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1191 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1194 while (index
<= end_index
) {
1195 page
= find_get_page(tree
->mapping
, index
);
1197 page_cache_release(page
);
1200 unlock_extent(tree
, start
, end
, GFP_NOFS
);
1203 EXPORT_SYMBOL(unlock_range
);
1205 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1207 struct rb_node
*node
;
1208 struct extent_state
*state
;
1211 spin_lock_irq(&tree
->lock
);
1213 * this search will find all the extents that end after
1216 node
= tree_search(tree
, start
);
1221 state
= rb_entry(node
, struct extent_state
, rb_node
);
1222 if (state
->start
!= start
) {
1226 state
->private = private;
1228 spin_unlock_irq(&tree
->lock
);
1232 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1234 struct rb_node
*node
;
1235 struct extent_state
*state
;
1238 spin_lock_irq(&tree
->lock
);
1240 * this search will find all the extents that end after
1243 node
= tree_search(tree
, start
);
1248 state
= rb_entry(node
, struct extent_state
, rb_node
);
1249 if (state
->start
!= start
) {
1253 *private = state
->private;
1255 spin_unlock_irq(&tree
->lock
);
1260 * searches a range in the state tree for a given mask.
1261 * If 'filled' == 1, this returns 1 only if every extent in the tree
1262 * has the bits set. Otherwise, 1 is returned if any bit in the
1263 * range is found set.
1265 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1266 int bits
, int filled
)
1268 struct extent_state
*state
= NULL
;
1269 struct rb_node
*node
;
1271 unsigned long flags
;
1273 spin_lock_irqsave(&tree
->lock
, flags
);
1274 node
= tree_search(tree
, start
);
1275 while (node
&& start
<= end
) {
1276 state
= rb_entry(node
, struct extent_state
, rb_node
);
1278 if (filled
&& state
->start
> start
) {
1283 if (state
->start
> end
)
1286 if (state
->state
& bits
) {
1290 } else if (filled
) {
1294 start
= state
->end
+ 1;
1297 node
= rb_next(node
);
1304 spin_unlock_irqrestore(&tree
->lock
, flags
);
1307 EXPORT_SYMBOL(test_range_bit
);
1310 * helper function to set a given page up to date if all the
1311 * extents in the tree for that page are up to date
1313 static int check_page_uptodate(struct extent_io_tree
*tree
,
1316 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1317 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1318 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1))
1319 SetPageUptodate(page
);
1324 * helper function to unlock a page if all the extents in the tree
1325 * for that page are unlocked
1327 static int check_page_locked(struct extent_io_tree
*tree
,
1330 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1331 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1332 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0))
1338 * helper function to end page writeback if all the extents
1339 * in the tree for that page are done with writeback
1341 static int check_page_writeback(struct extent_io_tree
*tree
,
1344 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1345 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1346 if (!test_range_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 0))
1347 end_page_writeback(page
);
1351 /* lots and lots of room for performance fixes in the end_bio funcs */
1354 * after a writepage IO is done, we need to:
1355 * clear the uptodate bits on error
1356 * clear the writeback bits in the extent tree for this IO
1357 * end_page_writeback if the page has no more pending IO
1359 * Scheduling is not allowed, so the extent state tree is expected
1360 * to have one and only one object corresponding to this IO.
1362 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1363 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1365 static int end_bio_extent_writepage(struct bio
*bio
,
1366 unsigned int bytes_done
, int err
)
1369 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1370 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1371 struct extent_state
*state
= bio
->bi_private
;
1372 struct extent_io_tree
*tree
= state
->tree
;
1373 struct rb_node
*node
;
1378 unsigned long flags
;
1380 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1385 struct page
*page
= bvec
->bv_page
;
1386 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1388 end
= start
+ bvec
->bv_len
- 1;
1390 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1395 if (--bvec
>= bio
->bi_io_vec
)
1396 prefetchw(&bvec
->bv_page
->flags
);
1399 clear_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1400 ClearPageUptodate(page
);
1404 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1405 tree
->ops
->writepage_end_io_hook(page
, start
, end
,
1410 * bios can get merged in funny ways, and so we need to
1411 * be careful with the state variable. We know the
1412 * state won't be merged with others because it has
1413 * WRITEBACK set, but we can't be sure each biovec is
1414 * sequential in the file. So, if our cached state
1415 * doesn't match the expected end, search the tree
1416 * for the correct one.
1419 spin_lock_irqsave(&tree
->lock
, flags
);
1420 if (!state
|| state
->end
!= end
) {
1422 node
= __etree_search(tree
, start
, NULL
, NULL
);
1424 state
= rb_entry(node
, struct extent_state
,
1426 if (state
->end
!= end
||
1427 !(state
->state
& EXTENT_WRITEBACK
))
1431 spin_unlock_irqrestore(&tree
->lock
, flags
);
1432 clear_extent_writeback(tree
, start
,
1439 struct extent_state
*clear
= state
;
1441 node
= rb_prev(&state
->rb_node
);
1443 state
= rb_entry(node
,
1444 struct extent_state
,
1450 clear_state_bit(tree
, clear
, EXTENT_WRITEBACK
,
1461 /* before releasing the lock, make sure the next state
1462 * variable has the expected bits set and corresponds
1463 * to the correct offsets in the file
1465 if (state
&& (state
->end
+ 1 != start
||
1466 !(state
->state
& EXTENT_WRITEBACK
))) {
1469 spin_unlock_irqrestore(&tree
->lock
, flags
);
1473 end_page_writeback(page
);
1475 check_page_writeback(tree
, page
);
1476 } while (bvec
>= bio
->bi_io_vec
);
1478 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1484 * after a readpage IO is done, we need to:
1485 * clear the uptodate bits on error
1486 * set the uptodate bits if things worked
1487 * set the page up to date if all extents in the tree are uptodate
1488 * clear the lock bit in the extent tree
1489 * unlock the page if there are no other extents locked for it
1491 * Scheduling is not allowed, so the extent state tree is expected
1492 * to have one and only one object corresponding to this IO.
1494 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1495 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1497 static int end_bio_extent_readpage(struct bio
*bio
,
1498 unsigned int bytes_done
, int err
)
1501 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1502 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1503 struct extent_state
*state
= bio
->bi_private
;
1504 struct extent_io_tree
*tree
= state
->tree
;
1505 struct rb_node
*node
;
1509 unsigned long flags
;
1513 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1519 struct page
*page
= bvec
->bv_page
;
1520 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1522 end
= start
+ bvec
->bv_len
- 1;
1524 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1529 if (--bvec
>= bio
->bi_io_vec
)
1530 prefetchw(&bvec
->bv_page
->flags
);
1532 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1533 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1538 if (!uptodate
&& tree
->ops
&&
1539 tree
->ops
->readpage_io_failed_hook
) {
1540 ret
= tree
->ops
->readpage_io_failed_hook(bio
, page
,
1545 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1550 spin_lock_irqsave(&tree
->lock
, flags
);
1551 if (!state
|| state
->end
!= end
) {
1553 node
= __etree_search(tree
, start
, NULL
, NULL
);
1555 state
= rb_entry(node
, struct extent_state
,
1557 if (state
->end
!= end
||
1558 !(state
->state
& EXTENT_LOCKED
))
1562 spin_unlock_irqrestore(&tree
->lock
, flags
);
1564 set_extent_uptodate(tree
, start
, end
,
1566 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1573 struct extent_state
*clear
= state
;
1575 node
= rb_prev(&state
->rb_node
);
1577 state
= rb_entry(node
,
1578 struct extent_state
,
1584 set_state_cb(tree
, clear
, EXTENT_UPTODATE
);
1585 clear
->state
|= EXTENT_UPTODATE
;
1587 clear_state_bit(tree
, clear
, EXTENT_LOCKED
,
1598 /* before releasing the lock, make sure the next state
1599 * variable has the expected bits set and corresponds
1600 * to the correct offsets in the file
1602 if (state
&& (state
->end
+ 1 != start
||
1603 !(state
->state
& EXTENT_LOCKED
))) {
1606 spin_unlock_irqrestore(&tree
->lock
, flags
);
1610 SetPageUptodate(page
);
1612 ClearPageUptodate(page
);
1618 check_page_uptodate(tree
, page
);
1620 ClearPageUptodate(page
);
1623 check_page_locked(tree
, page
);
1625 } while (bvec
>= bio
->bi_io_vec
);
1628 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1634 * IO done from prepare_write is pretty simple, we just unlock
1635 * the structs in the extent tree when done, and set the uptodate bits
1638 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1639 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1641 static int end_bio_extent_preparewrite(struct bio
*bio
,
1642 unsigned int bytes_done
, int err
)
1645 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1646 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1647 struct extent_state
*state
= bio
->bi_private
;
1648 struct extent_io_tree
*tree
= state
->tree
;
1652 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1658 struct page
*page
= bvec
->bv_page
;
1659 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1661 end
= start
+ bvec
->bv_len
- 1;
1663 if (--bvec
>= bio
->bi_io_vec
)
1664 prefetchw(&bvec
->bv_page
->flags
);
1667 set_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1669 ClearPageUptodate(page
);
1673 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1675 } while (bvec
>= bio
->bi_io_vec
);
1678 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1684 extent_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1689 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1691 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1692 while (!bio
&& (nr_vecs
/= 2))
1693 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1697 bio
->bi_bdev
= bdev
;
1698 bio
->bi_sector
= first_sector
;
1703 static int submit_one_bio(int rw
, struct bio
*bio
, int mirror_num
)
1707 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1708 struct page
*page
= bvec
->bv_page
;
1709 struct extent_io_tree
*tree
= bio
->bi_private
;
1710 struct rb_node
*node
;
1711 struct extent_state
*state
;
1715 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1716 end
= start
+ bvec
->bv_len
- 1;
1718 spin_lock_irq(&tree
->lock
);
1719 node
= __etree_search(tree
, start
, NULL
, NULL
);
1721 state
= rb_entry(node
, struct extent_state
, rb_node
);
1722 while(state
->end
< end
) {
1723 node
= rb_next(node
);
1724 state
= rb_entry(node
, struct extent_state
, rb_node
);
1726 BUG_ON(state
->end
!= end
);
1727 spin_unlock_irq(&tree
->lock
);
1729 bio
->bi_private
= state
;
1733 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1734 if (maxsector
< bio
->bi_sector
) {
1735 printk("sector too large max %Lu got %llu\n", maxsector
,
1736 (unsigned long long)bio
->bi_sector
);
1739 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
1740 tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
1743 submit_bio(rw
, bio
);
1744 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1750 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1751 struct page
*page
, sector_t sector
,
1752 size_t size
, unsigned long offset
,
1753 struct block_device
*bdev
,
1754 struct bio
**bio_ret
,
1755 unsigned long max_pages
,
1756 bio_end_io_t end_io_func
,
1763 if (bio_ret
&& *bio_ret
) {
1765 if (bio
->bi_sector
+ (bio
->bi_size
>> 9) != sector
||
1766 (tree
->ops
&& tree
->ops
->merge_bio_hook
&&
1767 tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
)) ||
1768 bio_add_page(bio
, page
, size
, offset
) < size
) {
1769 ret
= submit_one_bio(rw
, bio
, mirror_num
);
1775 nr
= bio_get_nr_vecs(bdev
);
1776 bio
= extent_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1778 printk("failed to allocate bio nr %d\n", nr
);
1782 bio_add_page(bio
, page
, size
, offset
);
1783 bio
->bi_end_io
= end_io_func
;
1784 bio
->bi_private
= tree
;
1789 ret
= submit_one_bio(rw
, bio
, mirror_num
);
1795 void set_page_extent_mapped(struct page
*page
)
1797 if (!PagePrivate(page
)) {
1798 SetPagePrivate(page
);
1799 WARN_ON(!page
->mapping
->a_ops
->invalidatepage
);
1800 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1801 page_cache_get(page
);
1805 void set_page_extent_head(struct page
*page
, unsigned long len
)
1807 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1811 * basic readpage implementation. Locked extent state structs are inserted
1812 * into the tree that are removed when the IO is done (by the end_io
1815 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1817 get_extent_t
*get_extent
,
1818 struct bio
**bio
, int mirror_num
)
1820 struct inode
*inode
= page
->mapping
->host
;
1821 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1822 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1826 u64 last_byte
= i_size_read(inode
);
1830 struct extent_map
*em
;
1831 struct block_device
*bdev
;
1834 size_t page_offset
= 0;
1836 size_t blocksize
= inode
->i_sb
->s_blocksize
;
1838 set_page_extent_mapped(page
);
1841 lock_extent(tree
, start
, end
, GFP_NOFS
);
1843 while (cur
<= end
) {
1844 if (cur
>= last_byte
) {
1846 iosize
= PAGE_CACHE_SIZE
- page_offset
;
1847 userpage
= kmap_atomic(page
, KM_USER0
);
1848 memset(userpage
+ page_offset
, 0, iosize
);
1849 flush_dcache_page(page
);
1850 kunmap_atomic(userpage
, KM_USER0
);
1851 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1853 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1856 em
= get_extent(inode
, page
, page_offset
, cur
,
1858 if (IS_ERR(em
) || !em
) {
1860 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
1864 extent_offset
= cur
- em
->start
;
1865 BUG_ON(extent_map_end(em
) <= cur
);
1868 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
1869 cur_end
= min(extent_map_end(em
) - 1, end
);
1870 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
1871 sector
= (em
->block_start
+ extent_offset
) >> 9;
1873 block_start
= em
->block_start
;
1874 free_extent_map(em
);
1877 /* we've found a hole, just zero and go on */
1878 if (block_start
== EXTENT_MAP_HOLE
) {
1880 userpage
= kmap_atomic(page
, KM_USER0
);
1881 memset(userpage
+ page_offset
, 0, iosize
);
1882 flush_dcache_page(page
);
1883 kunmap_atomic(userpage
, KM_USER0
);
1885 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1887 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1889 page_offset
+= iosize
;
1892 /* the get_extent function already copied into the page */
1893 if (test_range_bit(tree
, cur
, cur_end
, EXTENT_UPTODATE
, 1)) {
1894 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1896 page_offset
+= iosize
;
1899 /* we have an inline extent but it didn't get marked up
1900 * to date. Error out
1902 if (block_start
== EXTENT_MAP_INLINE
) {
1904 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1906 page_offset
+= iosize
;
1911 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
1912 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
1916 unsigned long nr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
1918 ret
= submit_extent_page(READ
, tree
, page
,
1919 sector
, iosize
, page_offset
,
1921 end_bio_extent_readpage
, mirror_num
);
1926 page_offset
+= iosize
;
1930 if (!PageError(page
))
1931 SetPageUptodate(page
);
1937 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
1938 get_extent_t
*get_extent
)
1940 struct bio
*bio
= NULL
;
1943 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, 0);
1945 submit_one_bio(READ
, bio
, 0);
1948 EXPORT_SYMBOL(extent_read_full_page
);
1951 * the writepage semantics are similar to regular writepage. extent
1952 * records are inserted to lock ranges in the tree, and as dirty areas
1953 * are found, they are marked writeback. Then the lock bits are removed
1954 * and the end_io handler clears the writeback ranges
1956 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
1959 struct inode
*inode
= page
->mapping
->host
;
1960 struct extent_page_data
*epd
= data
;
1961 struct extent_io_tree
*tree
= epd
->tree
;
1962 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1964 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1968 u64 last_byte
= i_size_read(inode
);
1972 struct extent_map
*em
;
1973 struct block_device
*bdev
;
1976 size_t page_offset
= 0;
1978 loff_t i_size
= i_size_read(inode
);
1979 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1983 WARN_ON(!PageLocked(page
));
1984 if (page
->index
> end_index
) {
1985 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
1990 if (page
->index
== end_index
) {
1993 size_t offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
1995 userpage
= kmap_atomic(page
, KM_USER0
);
1996 memset(userpage
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
1997 flush_dcache_page(page
);
1998 kunmap_atomic(userpage
, KM_USER0
);
2001 set_page_extent_mapped(page
);
2003 delalloc_start
= start
;
2005 while(delalloc_end
< page_end
) {
2006 nr_delalloc
= find_lock_delalloc_range(tree
, &delalloc_start
,
2009 if (nr_delalloc
== 0) {
2010 delalloc_start
= delalloc_end
+ 1;
2013 tree
->ops
->fill_delalloc(inode
, delalloc_start
,
2015 clear_extent_bit(tree
, delalloc_start
,
2017 EXTENT_LOCKED
| EXTENT_DELALLOC
,
2019 delalloc_start
= delalloc_end
+ 1;
2021 lock_extent(tree
, start
, page_end
, GFP_NOFS
);
2024 if (test_range_bit(tree
, start
, page_end
, EXTENT_DELALLOC
, 0)) {
2025 printk("found delalloc bits after lock_extent\n");
2028 if (last_byte
<= start
) {
2029 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
2033 set_extent_uptodate(tree
, start
, page_end
, GFP_NOFS
);
2034 blocksize
= inode
->i_sb
->s_blocksize
;
2036 while (cur
<= end
) {
2037 if (cur
>= last_byte
) {
2038 clear_extent_dirty(tree
, cur
, page_end
, GFP_NOFS
);
2041 em
= epd
->get_extent(inode
, page
, page_offset
, cur
,
2043 if (IS_ERR(em
) || !em
) {
2048 extent_offset
= cur
- em
->start
;
2049 BUG_ON(extent_map_end(em
) <= cur
);
2051 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2052 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2053 sector
= (em
->block_start
+ extent_offset
) >> 9;
2055 block_start
= em
->block_start
;
2056 free_extent_map(em
);
2059 if (block_start
== EXTENT_MAP_HOLE
||
2060 block_start
== EXTENT_MAP_INLINE
) {
2061 clear_extent_dirty(tree
, cur
,
2062 cur
+ iosize
- 1, GFP_NOFS
);
2064 page_offset
+= iosize
;
2068 /* leave this out until we have a page_mkwrite call */
2069 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2072 page_offset
+= iosize
;
2075 clear_extent_dirty(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2076 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2077 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2085 unsigned long max_nr
= end_index
+ 1;
2086 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2087 if (!PageWriteback(page
)) {
2088 printk("warning page %lu not writeback, "
2089 "cur %llu end %llu\n", page
->index
,
2090 (unsigned long long)cur
,
2091 (unsigned long long)end
);
2094 ret
= submit_extent_page(WRITE
, tree
, page
, sector
,
2095 iosize
, page_offset
, bdev
,
2097 end_bio_extent_writepage
, 0);
2102 page_offset
+= iosize
;
2107 /* make sure the mapping tag for page dirty gets cleared */
2108 set_page_writeback(page
);
2109 end_page_writeback(page
);
2111 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
2116 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,20)
2117 /* Taken directly from 2.6.23 for 2.6.18 back port */
2118 typedef int (*writepage_t
)(struct page
*page
, struct writeback_control
*wbc
,
2122 * write_cache_pages - walk the list of dirty pages of the given address space
2123 * and write all of them.
2124 * @mapping: address space structure to write
2125 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2126 * @writepage: function called for each page
2127 * @data: data passed to writepage function
2129 * If a page is already under I/O, write_cache_pages() skips it, even
2130 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2131 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2132 * and msync() need to guarantee that all the data which was dirty at the time
2133 * the call was made get new I/O started against them. If wbc->sync_mode is
2134 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2135 * existing IO to complete.
2137 static int write_cache_pages(struct address_space
*mapping
,
2138 struct writeback_control
*wbc
, writepage_t writepage
,
2141 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
2144 struct pagevec pvec
;
2147 pgoff_t end
; /* Inclusive */
2149 int range_whole
= 0;
2151 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2152 wbc
->encountered_congestion
= 1;
2156 pagevec_init(&pvec
, 0);
2157 if (wbc
->range_cyclic
) {
2158 index
= mapping
->writeback_index
; /* Start from prev offset */
2161 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2162 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2163 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2168 while (!done
&& (index
<= end
) &&
2169 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2170 PAGECACHE_TAG_DIRTY
,
2171 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2175 for (i
= 0; i
< nr_pages
; i
++) {
2176 struct page
*page
= pvec
.pages
[i
];
2179 * At this point we hold neither mapping->tree_lock nor
2180 * lock on the page itself: the page may be truncated or
2181 * invalidated (changing page->mapping to NULL), or even
2182 * swizzled back from swapper_space to tmpfs file
2187 if (unlikely(page
->mapping
!= mapping
)) {
2192 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2198 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2199 wait_on_page_writeback(page
);
2201 if (PageWriteback(page
) ||
2202 !clear_page_dirty_for_io(page
)) {
2207 ret
= (*writepage
)(page
, wbc
, data
);
2209 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2213 if (ret
|| (--(wbc
->nr_to_write
) <= 0))
2215 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2216 wbc
->encountered_congestion
= 1;
2220 pagevec_release(&pvec
);
2223 if (!scanned
&& !done
) {
2225 * We hit the last page and there is more work to be done: wrap
2226 * back to the start of the file
2232 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2233 mapping
->writeback_index
= index
;
2238 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2239 get_extent_t
*get_extent
,
2240 struct writeback_control
*wbc
)
2243 struct address_space
*mapping
= page
->mapping
;
2244 struct extent_page_data epd
= {
2247 .get_extent
= get_extent
,
2249 struct writeback_control wbc_writepages
= {
2251 .sync_mode
= WB_SYNC_NONE
,
2252 .older_than_this
= NULL
,
2254 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2255 .range_end
= (loff_t
)-1,
2259 ret
= __extent_writepage(page
, wbc
, &epd
);
2261 write_cache_pages(mapping
, &wbc_writepages
, __extent_writepage
, &epd
);
2263 submit_one_bio(WRITE
, epd
.bio
, 0);
2267 EXPORT_SYMBOL(extent_write_full_page
);
2270 int extent_writepages(struct extent_io_tree
*tree
,
2271 struct address_space
*mapping
,
2272 get_extent_t
*get_extent
,
2273 struct writeback_control
*wbc
)
2276 struct extent_page_data epd
= {
2279 .get_extent
= get_extent
,
2282 ret
= write_cache_pages(mapping
, wbc
, __extent_writepage
, &epd
);
2284 submit_one_bio(WRITE
, epd
.bio
, 0);
2288 EXPORT_SYMBOL(extent_writepages
);
2290 int extent_readpages(struct extent_io_tree
*tree
,
2291 struct address_space
*mapping
,
2292 struct list_head
*pages
, unsigned nr_pages
,
2293 get_extent_t get_extent
)
2295 struct bio
*bio
= NULL
;
2297 struct pagevec pvec
;
2299 pagevec_init(&pvec
, 0);
2300 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2301 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2303 prefetchw(&page
->flags
);
2304 list_del(&page
->lru
);
2306 * what we want to do here is call add_to_page_cache_lru,
2307 * but that isn't exported, so we reproduce it here
2309 if (!add_to_page_cache(page
, mapping
,
2310 page
->index
, GFP_KERNEL
)) {
2312 /* open coding of lru_cache_add, also not exported */
2313 page_cache_get(page
);
2314 if (!pagevec_add(&pvec
, page
))
2315 __pagevec_lru_add(&pvec
);
2316 __extent_read_full_page(tree
, page
, get_extent
,
2319 page_cache_release(page
);
2321 if (pagevec_count(&pvec
))
2322 __pagevec_lru_add(&pvec
);
2323 BUG_ON(!list_empty(pages
));
2325 submit_one_bio(READ
, bio
, 0);
2328 EXPORT_SYMBOL(extent_readpages
);
2331 * basic invalidatepage code, this waits on any locked or writeback
2332 * ranges corresponding to the page, and then deletes any extent state
2333 * records from the tree
2335 int extent_invalidatepage(struct extent_io_tree
*tree
,
2336 struct page
*page
, unsigned long offset
)
2338 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2339 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2340 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2342 start
+= (offset
+ blocksize
-1) & ~(blocksize
- 1);
2346 lock_extent(tree
, start
, end
, GFP_NOFS
);
2347 wait_on_extent_writeback(tree
, start
, end
);
2348 clear_extent_bit(tree
, start
, end
,
2349 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
,
2353 EXPORT_SYMBOL(extent_invalidatepage
);
2356 * simple commit_write call, set_range_dirty is used to mark both
2357 * the pages and the extent records as dirty
2359 int extent_commit_write(struct extent_io_tree
*tree
,
2360 struct inode
*inode
, struct page
*page
,
2361 unsigned from
, unsigned to
)
2363 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2365 set_page_extent_mapped(page
);
2366 set_page_dirty(page
);
2368 if (pos
> inode
->i_size
) {
2369 i_size_write(inode
, pos
);
2370 mark_inode_dirty(inode
);
2374 EXPORT_SYMBOL(extent_commit_write
);
2376 int extent_prepare_write(struct extent_io_tree
*tree
,
2377 struct inode
*inode
, struct page
*page
,
2378 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2380 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2381 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2383 u64 orig_block_start
;
2386 struct extent_map
*em
;
2387 unsigned blocksize
= 1 << inode
->i_blkbits
;
2388 size_t page_offset
= 0;
2389 size_t block_off_start
;
2390 size_t block_off_end
;
2396 set_page_extent_mapped(page
);
2398 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2399 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2400 orig_block_start
= block_start
;
2402 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2403 while(block_start
<= block_end
) {
2404 em
= get_extent(inode
, page
, page_offset
, block_start
,
2405 block_end
- block_start
+ 1, 1);
2406 if (IS_ERR(em
) || !em
) {
2409 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2410 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2411 block_off_end
= block_off_start
+ blocksize
;
2412 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2414 if (!PageUptodate(page
) && isnew
&&
2415 (block_off_end
> to
|| block_off_start
< from
)) {
2418 kaddr
= kmap_atomic(page
, KM_USER0
);
2419 if (block_off_end
> to
)
2420 memset(kaddr
+ to
, 0, block_off_end
- to
);
2421 if (block_off_start
< from
)
2422 memset(kaddr
+ block_off_start
, 0,
2423 from
- block_off_start
);
2424 flush_dcache_page(page
);
2425 kunmap_atomic(kaddr
, KM_USER0
);
2427 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2428 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2429 !isnew
&& !PageUptodate(page
) &&
2430 (block_off_end
> to
|| block_off_start
< from
) &&
2431 !test_range_bit(tree
, block_start
, cur_end
,
2432 EXTENT_UPTODATE
, 1)) {
2434 u64 extent_offset
= block_start
- em
->start
;
2436 sector
= (em
->block_start
+ extent_offset
) >> 9;
2437 iosize
= (cur_end
- block_start
+ blocksize
) &
2438 ~((u64
)blocksize
- 1);
2440 * we've already got the extent locked, but we
2441 * need to split the state such that our end_bio
2442 * handler can clear the lock.
2444 set_extent_bit(tree
, block_start
,
2445 block_start
+ iosize
- 1,
2446 EXTENT_LOCKED
, 0, NULL
, GFP_NOFS
);
2447 ret
= submit_extent_page(READ
, tree
, page
,
2448 sector
, iosize
, page_offset
, em
->bdev
,
2450 end_bio_extent_preparewrite
, 0);
2452 block_start
= block_start
+ iosize
;
2454 set_extent_uptodate(tree
, block_start
, cur_end
,
2456 unlock_extent(tree
, block_start
, cur_end
, GFP_NOFS
);
2457 block_start
= cur_end
+ 1;
2459 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2460 free_extent_map(em
);
2463 wait_extent_bit(tree
, orig_block_start
,
2464 block_end
, EXTENT_LOCKED
);
2466 check_page_uptodate(tree
, page
);
2468 /* FIXME, zero out newly allocated blocks on error */
2471 EXPORT_SYMBOL(extent_prepare_write
);
2474 * a helper for releasepage. As long as there are no locked extents
2475 * in the range corresponding to the page, both state records and extent
2476 * map records are removed
2478 int try_release_extent_mapping(struct extent_map_tree
*map
,
2479 struct extent_io_tree
*tree
, struct page
*page
,
2482 struct extent_map
*em
;
2483 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2484 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2485 u64 orig_start
= start
;
2487 if ((mask
& __GFP_WAIT
) &&
2488 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2490 while (start
<= end
) {
2491 len
= end
- start
+ 1;
2492 spin_lock(&map
->lock
);
2493 em
= lookup_extent_mapping(map
, start
, len
);
2494 if (!em
|| IS_ERR(em
)) {
2495 spin_unlock(&map
->lock
);
2498 if (em
->start
!= start
) {
2499 spin_unlock(&map
->lock
);
2500 free_extent_map(em
);
2503 if (!test_range_bit(tree
, em
->start
,
2504 extent_map_end(em
) - 1,
2505 EXTENT_LOCKED
, 0)) {
2506 remove_extent_mapping(map
, em
);
2507 /* once for the rb tree */
2508 free_extent_map(em
);
2510 start
= extent_map_end(em
);
2511 spin_unlock(&map
->lock
);
2514 free_extent_map(em
);
2517 if (test_range_bit(tree
, orig_start
, end
, EXTENT_IOBITS
, 0))
2520 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2522 clear_extent_bit(tree
, orig_start
, end
, EXTENT_UPTODATE
,
2527 EXPORT_SYMBOL(try_release_extent_mapping
);
2529 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2530 get_extent_t
*get_extent
)
2532 struct inode
*inode
= mapping
->host
;
2533 u64 start
= iblock
<< inode
->i_blkbits
;
2534 sector_t sector
= 0;
2535 struct extent_map
*em
;
2537 em
= get_extent(inode
, NULL
, 0, start
, (1 << inode
->i_blkbits
), 0);
2538 if (!em
|| IS_ERR(em
))
2541 if (em
->block_start
== EXTENT_MAP_INLINE
||
2542 em
->block_start
== EXTENT_MAP_HOLE
)
2545 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2547 free_extent_map(em
);
2551 static int add_lru(struct extent_io_tree
*tree
, struct extent_buffer
*eb
)
2553 if (list_empty(&eb
->lru
)) {
2554 extent_buffer_get(eb
);
2555 list_add(&eb
->lru
, &tree
->buffer_lru
);
2557 if (tree
->lru_size
>= BUFFER_LRU_MAX
) {
2558 struct extent_buffer
*rm
;
2559 rm
= list_entry(tree
->buffer_lru
.prev
,
2560 struct extent_buffer
, lru
);
2562 list_del_init(&rm
->lru
);
2563 free_extent_buffer(rm
);
2566 list_move(&eb
->lru
, &tree
->buffer_lru
);
2569 static struct extent_buffer
*find_lru(struct extent_io_tree
*tree
,
2570 u64 start
, unsigned long len
)
2572 struct list_head
*lru
= &tree
->buffer_lru
;
2573 struct list_head
*cur
= lru
->next
;
2574 struct extent_buffer
*eb
;
2576 if (list_empty(lru
))
2580 eb
= list_entry(cur
, struct extent_buffer
, lru
);
2581 if (eb
->start
== start
&& eb
->len
== len
) {
2582 extent_buffer_get(eb
);
2586 } while (cur
!= lru
);
2590 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
2592 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
2593 (start
>> PAGE_CACHE_SHIFT
);
2596 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
2600 struct address_space
*mapping
;
2603 return eb
->first_page
;
2604 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
2605 mapping
= eb
->first_page
->mapping
;
2606 read_lock_irq(&mapping
->tree_lock
);
2607 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
2608 read_unlock_irq(&mapping
->tree_lock
);
2612 int release_extent_buffer_tail_pages(struct extent_buffer
*eb
)
2614 unsigned long num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2620 for (i
= 1; i
< num_pages
; i
++) {
2621 page
= extent_buffer_page(eb
, i
);
2622 page_cache_release(page
);
2628 int invalidate_extent_lru(struct extent_io_tree
*tree
, u64 start
,
2631 struct list_head
*lru
= &tree
->buffer_lru
;
2632 struct list_head
*cur
= lru
->next
;
2633 struct extent_buffer
*eb
;
2636 spin_lock(&tree
->lru_lock
);
2637 if (list_empty(lru
))
2641 eb
= list_entry(cur
, struct extent_buffer
, lru
);
2642 if (eb
->start
<= start
&& eb
->start
+ eb
->len
> start
) {
2643 eb
->flags
&= ~EXTENT_UPTODATE
;
2646 } while (cur
!= lru
);
2648 spin_unlock(&tree
->lru_lock
);
2652 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
2657 struct extent_buffer
*eb
= NULL
;
2658 unsigned long flags
;
2660 spin_lock(&tree
->lru_lock
);
2661 eb
= find_lru(tree
, start
, len
);
2662 spin_unlock(&tree
->lru_lock
);
2667 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
2668 INIT_LIST_HEAD(&eb
->lru
);
2671 spin_lock_irqsave(&leak_lock
, flags
);
2672 list_add(&eb
->leak_list
, &buffers
);
2673 spin_unlock_irqrestore(&leak_lock
, flags
);
2674 atomic_set(&eb
->refs
, 1);
2679 static void __free_extent_buffer(struct extent_buffer
*eb
)
2681 unsigned long flags
;
2682 spin_lock_irqsave(&leak_lock
, flags
);
2683 list_del(&eb
->leak_list
);
2684 spin_unlock_irqrestore(&leak_lock
, flags
);
2685 kmem_cache_free(extent_buffer_cache
, eb
);
2688 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
2689 u64 start
, unsigned long len
,
2693 unsigned long num_pages
= num_extent_pages(start
, len
);
2695 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2696 struct extent_buffer
*eb
;
2698 struct address_space
*mapping
= tree
->mapping
;
2701 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2705 if (eb
->flags
& EXTENT_BUFFER_FILLED
)
2709 eb
->first_page
= page0
;
2712 page_cache_get(page0
);
2713 mark_page_accessed(page0
);
2714 set_page_extent_mapped(page0
);
2715 set_page_extent_head(page0
, len
);
2716 uptodate
= PageUptodate(page0
);
2720 for (; i
< num_pages
; i
++, index
++) {
2721 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
2726 set_page_extent_mapped(p
);
2727 mark_page_accessed(p
);
2730 set_page_extent_head(p
, len
);
2732 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2734 if (!PageUptodate(p
))
2739 eb
->flags
|= EXTENT_UPTODATE
;
2740 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2743 spin_lock(&tree
->lru_lock
);
2745 spin_unlock(&tree
->lru_lock
);
2749 spin_lock(&tree
->lru_lock
);
2750 list_del_init(&eb
->lru
);
2751 spin_unlock(&tree
->lru_lock
);
2752 if (!atomic_dec_and_test(&eb
->refs
))
2754 for (index
= 1; index
< i
; index
++) {
2755 page_cache_release(extent_buffer_page(eb
, index
));
2758 page_cache_release(extent_buffer_page(eb
, 0));
2759 __free_extent_buffer(eb
);
2762 EXPORT_SYMBOL(alloc_extent_buffer
);
2764 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
2765 u64 start
, unsigned long len
,
2768 unsigned long num_pages
= num_extent_pages(start
, len
);
2770 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2771 struct extent_buffer
*eb
;
2773 struct address_space
*mapping
= tree
->mapping
;
2776 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2780 if (eb
->flags
& EXTENT_BUFFER_FILLED
)
2783 for (i
= 0; i
< num_pages
; i
++, index
++) {
2784 p
= find_lock_page(mapping
, index
);
2788 set_page_extent_mapped(p
);
2789 mark_page_accessed(p
);
2793 set_page_extent_head(p
, len
);
2795 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2798 if (!PageUptodate(p
))
2803 eb
->flags
|= EXTENT_UPTODATE
;
2804 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2807 spin_lock(&tree
->lru_lock
);
2809 spin_unlock(&tree
->lru_lock
);
2812 spin_lock(&tree
->lru_lock
);
2813 list_del_init(&eb
->lru
);
2814 spin_unlock(&tree
->lru_lock
);
2815 if (!atomic_dec_and_test(&eb
->refs
))
2817 for (index
= 1; index
< i
; index
++) {
2818 page_cache_release(extent_buffer_page(eb
, index
));
2821 page_cache_release(extent_buffer_page(eb
, 0));
2822 __free_extent_buffer(eb
);
2825 EXPORT_SYMBOL(find_extent_buffer
);
2827 void free_extent_buffer(struct extent_buffer
*eb
)
2830 unsigned long num_pages
;
2835 if (!atomic_dec_and_test(&eb
->refs
))
2838 WARN_ON(!list_empty(&eb
->lru
));
2839 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2841 for (i
= 1; i
< num_pages
; i
++) {
2842 page_cache_release(extent_buffer_page(eb
, i
));
2844 page_cache_release(extent_buffer_page(eb
, 0));
2845 __free_extent_buffer(eb
);
2847 EXPORT_SYMBOL(free_extent_buffer
);
2849 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
2850 struct extent_buffer
*eb
)
2854 unsigned long num_pages
;
2857 u64 start
= eb
->start
;
2858 u64 end
= start
+ eb
->len
- 1;
2860 set
= clear_extent_dirty(tree
, start
, end
, GFP_NOFS
);
2861 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2863 for (i
= 0; i
< num_pages
; i
++) {
2864 page
= extent_buffer_page(eb
, i
);
2867 set_page_extent_head(page
, eb
->len
);
2869 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2872 * if we're on the last page or the first page and the
2873 * block isn't aligned on a page boundary, do extra checks
2874 * to make sure we don't clean page that is partially dirty
2876 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2877 ((i
== num_pages
- 1) &&
2878 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2879 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2880 end
= start
+ PAGE_CACHE_SIZE
- 1;
2881 if (test_range_bit(tree
, start
, end
,
2887 clear_page_dirty_for_io(page
);
2888 read_lock_irq(&page
->mapping
->tree_lock
);
2889 if (!PageDirty(page
)) {
2890 radix_tree_tag_clear(&page
->mapping
->page_tree
,
2892 PAGECACHE_TAG_DIRTY
);
2894 read_unlock_irq(&page
->mapping
->tree_lock
);
2899 EXPORT_SYMBOL(clear_extent_buffer_dirty
);
2901 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
2902 struct extent_buffer
*eb
)
2904 return wait_on_extent_writeback(tree
, eb
->start
,
2905 eb
->start
+ eb
->len
- 1);
2907 EXPORT_SYMBOL(wait_on_extent_buffer_writeback
);
2909 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
2910 struct extent_buffer
*eb
)
2913 unsigned long num_pages
;
2915 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2916 for (i
= 0; i
< num_pages
; i
++) {
2917 struct page
*page
= extent_buffer_page(eb
, i
);
2918 /* writepage may need to do something special for the
2919 * first page, we have to make sure page->private is
2920 * properly set. releasepage may drop page->private
2921 * on us if the page isn't already dirty.
2925 set_page_extent_head(page
, eb
->len
);
2926 } else if (PagePrivate(page
) &&
2927 page
->private != EXTENT_PAGE_PRIVATE
) {
2929 set_page_extent_mapped(page
);
2932 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
2936 return set_extent_dirty(tree
, eb
->start
,
2937 eb
->start
+ eb
->len
- 1, GFP_NOFS
);
2939 EXPORT_SYMBOL(set_extent_buffer_dirty
);
2941 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
2942 struct extent_buffer
*eb
)
2946 unsigned long num_pages
;
2948 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2950 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2952 for (i
= 0; i
< num_pages
; i
++) {
2953 page
= extent_buffer_page(eb
, i
);
2954 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2955 ((i
== num_pages
- 1) &&
2956 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2957 check_page_uptodate(tree
, page
);
2960 SetPageUptodate(page
);
2964 EXPORT_SYMBOL(set_extent_buffer_uptodate
);
2966 int extent_range_uptodate(struct extent_io_tree
*tree
,
2971 int pg_uptodate
= 1;
2973 unsigned long index
;
2975 ret
= test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1);
2978 while(start
<= end
) {
2979 index
= start
>> PAGE_CACHE_SHIFT
;
2980 page
= find_get_page(tree
->mapping
, index
);
2981 uptodate
= PageUptodate(page
);
2982 page_cache_release(page
);
2987 start
+= PAGE_CACHE_SIZE
;
2992 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
2993 struct extent_buffer
*eb
)
2997 unsigned long num_pages
;
3000 int pg_uptodate
= 1;
3002 if (eb
->flags
& EXTENT_UPTODATE
)
3005 ret2
= test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3006 EXTENT_UPTODATE
, 1);
3008 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3009 for (i
= 0; i
< num_pages
; i
++) {
3010 page
= extent_buffer_page(eb
, i
);
3011 if (!PageUptodate(page
)) {
3016 if ((ret
|| ret2
) && !pg_uptodate
) {
3017 printk("uptodate error2 eb %Lu ret %d ret2 %d pg_uptodate %d\n", eb
->start
, ret
, ret2
, pg_uptodate
);
3020 return (ret
|| ret2
);
3022 EXPORT_SYMBOL(extent_buffer_uptodate
);
3024 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
3025 struct extent_buffer
*eb
,
3026 u64 start
, int wait
,
3027 get_extent_t
*get_extent
, int mirror_num
)
3030 unsigned long start_i
;
3034 int locked_pages
= 0;
3035 int all_uptodate
= 1;
3036 int inc_all_pages
= 0;
3037 unsigned long num_pages
;
3038 struct bio
*bio
= NULL
;
3040 if (eb
->flags
& EXTENT_UPTODATE
)
3043 if (test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3044 EXTENT_UPTODATE
, 1)) {
3049 WARN_ON(start
< eb
->start
);
3050 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
3051 (eb
->start
>> PAGE_CACHE_SHIFT
);
3056 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3057 for (i
= start_i
; i
< num_pages
; i
++) {
3058 page
= extent_buffer_page(eb
, i
);
3060 if (TestSetPageLocked(page
))
3066 if (!PageUptodate(page
)) {
3072 eb
->flags
|= EXTENT_UPTODATE
;
3076 for (i
= start_i
; i
< num_pages
; i
++) {
3077 page
= extent_buffer_page(eb
, i
);
3079 page_cache_get(page
);
3080 if (!PageUptodate(page
)) {
3083 ClearPageError(page
);
3084 err
= __extent_read_full_page(tree
, page
,
3096 submit_one_bio(READ
, bio
, mirror_num
);
3101 for (i
= start_i
; i
< num_pages
; i
++) {
3102 page
= extent_buffer_page(eb
, i
);
3103 wait_on_page_locked(page
);
3104 if (!PageUptodate(page
)) {
3109 eb
->flags
|= EXTENT_UPTODATE
;
3114 while(locked_pages
> 0) {
3115 page
= extent_buffer_page(eb
, i
);
3122 EXPORT_SYMBOL(read_extent_buffer_pages
);
3124 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
3125 unsigned long start
,
3132 char *dst
= (char *)dstv
;
3133 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3134 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3136 WARN_ON(start
> eb
->len
);
3137 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3139 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3142 page
= extent_buffer_page(eb
, i
);
3144 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3145 kaddr
= kmap_atomic(page
, KM_USER1
);
3146 memcpy(dst
, kaddr
+ offset
, cur
);
3147 kunmap_atomic(kaddr
, KM_USER1
);
3155 EXPORT_SYMBOL(read_extent_buffer
);
3157 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3158 unsigned long min_len
, char **token
, char **map
,
3159 unsigned long *map_start
,
3160 unsigned long *map_len
, int km
)
3162 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
3165 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3166 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3167 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
3174 offset
= start_offset
;
3178 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
3180 if (start
+ min_len
> eb
->len
) {
3181 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb
->start
, eb
->len
, start
, min_len
);
3185 p
= extent_buffer_page(eb
, i
);
3186 kaddr
= kmap_atomic(p
, km
);
3188 *map
= kaddr
+ offset
;
3189 *map_len
= PAGE_CACHE_SIZE
- offset
;
3192 EXPORT_SYMBOL(map_private_extent_buffer
);
3194 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3195 unsigned long min_len
,
3196 char **token
, char **map
,
3197 unsigned long *map_start
,
3198 unsigned long *map_len
, int km
)
3202 if (eb
->map_token
) {
3203 unmap_extent_buffer(eb
, eb
->map_token
, km
);
3204 eb
->map_token
= NULL
;
3207 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
3208 map_start
, map_len
, km
);
3210 eb
->map_token
= *token
;
3212 eb
->map_start
= *map_start
;
3213 eb
->map_len
= *map_len
;
3217 EXPORT_SYMBOL(map_extent_buffer
);
3219 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
3221 kunmap_atomic(token
, km
);
3223 EXPORT_SYMBOL(unmap_extent_buffer
);
3225 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3226 unsigned long start
,
3233 char *ptr
= (char *)ptrv
;
3234 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3235 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3238 WARN_ON(start
> eb
->len
);
3239 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3241 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3244 page
= extent_buffer_page(eb
, i
);
3246 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3248 kaddr
= kmap_atomic(page
, KM_USER0
);
3249 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3250 kunmap_atomic(kaddr
, KM_USER0
);
3261 EXPORT_SYMBOL(memcmp_extent_buffer
);
3263 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3264 unsigned long start
, unsigned long len
)
3270 char *src
= (char *)srcv
;
3271 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3272 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3274 WARN_ON(start
> eb
->len
);
3275 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3277 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3280 page
= extent_buffer_page(eb
, i
);
3281 WARN_ON(!PageUptodate(page
));
3283 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3284 kaddr
= kmap_atomic(page
, KM_USER1
);
3285 memcpy(kaddr
+ offset
, src
, cur
);
3286 kunmap_atomic(kaddr
, KM_USER1
);
3294 EXPORT_SYMBOL(write_extent_buffer
);
3296 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3297 unsigned long start
, unsigned long len
)
3303 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3304 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3306 WARN_ON(start
> eb
->len
);
3307 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3309 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3312 page
= extent_buffer_page(eb
, i
);
3313 WARN_ON(!PageUptodate(page
));
3315 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3316 kaddr
= kmap_atomic(page
, KM_USER0
);
3317 memset(kaddr
+ offset
, c
, cur
);
3318 kunmap_atomic(kaddr
, KM_USER0
);
3325 EXPORT_SYMBOL(memset_extent_buffer
);
3327 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3328 unsigned long dst_offset
, unsigned long src_offset
,
3331 u64 dst_len
= dst
->len
;
3336 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3337 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3339 WARN_ON(src
->len
!= dst_len
);
3341 offset
= (start_offset
+ dst_offset
) &
3342 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3345 page
= extent_buffer_page(dst
, i
);
3346 WARN_ON(!PageUptodate(page
));
3348 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3350 kaddr
= kmap_atomic(page
, KM_USER0
);
3351 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3352 kunmap_atomic(kaddr
, KM_USER0
);
3360 EXPORT_SYMBOL(copy_extent_buffer
);
3362 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3363 unsigned long dst_off
, unsigned long src_off
,
3366 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3367 if (dst_page
== src_page
) {
3368 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3370 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3371 char *p
= dst_kaddr
+ dst_off
+ len
;
3372 char *s
= src_kaddr
+ src_off
+ len
;
3377 kunmap_atomic(src_kaddr
, KM_USER1
);
3379 kunmap_atomic(dst_kaddr
, KM_USER0
);
3382 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3383 unsigned long dst_off
, unsigned long src_off
,
3386 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3389 if (dst_page
!= src_page
)
3390 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3392 src_kaddr
= dst_kaddr
;
3394 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3395 kunmap_atomic(dst_kaddr
, KM_USER0
);
3396 if (dst_page
!= src_page
)
3397 kunmap_atomic(src_kaddr
, KM_USER1
);
3400 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3401 unsigned long src_offset
, unsigned long len
)
3404 size_t dst_off_in_page
;
3405 size_t src_off_in_page
;
3406 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3407 unsigned long dst_i
;
3408 unsigned long src_i
;
3410 if (src_offset
+ len
> dst
->len
) {
3411 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3412 src_offset
, len
, dst
->len
);
3415 if (dst_offset
+ len
> dst
->len
) {
3416 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3417 dst_offset
, len
, dst
->len
);
3422 dst_off_in_page
= (start_offset
+ dst_offset
) &
3423 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3424 src_off_in_page
= (start_offset
+ src_offset
) &
3425 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3427 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3428 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3430 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3432 cur
= min_t(unsigned long, cur
,
3433 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3435 copy_pages(extent_buffer_page(dst
, dst_i
),
3436 extent_buffer_page(dst
, src_i
),
3437 dst_off_in_page
, src_off_in_page
, cur
);
3444 EXPORT_SYMBOL(memcpy_extent_buffer
);
3446 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3447 unsigned long src_offset
, unsigned long len
)
3450 size_t dst_off_in_page
;
3451 size_t src_off_in_page
;
3452 unsigned long dst_end
= dst_offset
+ len
- 1;
3453 unsigned long src_end
= src_offset
+ len
- 1;
3454 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3455 unsigned long dst_i
;
3456 unsigned long src_i
;
3458 if (src_offset
+ len
> dst
->len
) {
3459 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3460 src_offset
, len
, dst
->len
);
3463 if (dst_offset
+ len
> dst
->len
) {
3464 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3465 dst_offset
, len
, dst
->len
);
3468 if (dst_offset
< src_offset
) {
3469 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
3473 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
3474 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
3476 dst_off_in_page
= (start_offset
+ dst_end
) &
3477 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3478 src_off_in_page
= (start_offset
+ src_end
) &
3479 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3481 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
3482 cur
= min(cur
, dst_off_in_page
+ 1);
3483 move_pages(extent_buffer_page(dst
, dst_i
),
3484 extent_buffer_page(dst
, src_i
),
3485 dst_off_in_page
- cur
+ 1,
3486 src_off_in_page
- cur
+ 1, cur
);
3493 EXPORT_SYMBOL(memmove_extent_buffer
);