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
);
30 #define BUFFER_LRU_MAX 64
35 struct rb_node rb_node
;
38 struct extent_page_data
{
40 struct extent_io_tree
*tree
;
41 get_extent_t
*get_extent
;
44 int __init
extent_io_init(void)
46 extent_state_cache
= btrfs_cache_create("extent_state",
47 sizeof(struct extent_state
), 0,
49 if (!extent_state_cache
)
52 extent_buffer_cache
= btrfs_cache_create("extent_buffers",
53 sizeof(struct extent_buffer
), 0,
55 if (!extent_buffer_cache
)
56 goto free_state_cache
;
60 kmem_cache_destroy(extent_state_cache
);
64 void extent_io_exit(void)
66 struct extent_state
*state
;
68 while (!list_empty(&states
)) {
69 state
= list_entry(states
.next
, struct extent_state
, list
);
70 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
));
71 list_del(&state
->list
);
72 kmem_cache_free(extent_state_cache
, state
);
76 if (extent_state_cache
)
77 kmem_cache_destroy(extent_state_cache
);
78 if (extent_buffer_cache
)
79 kmem_cache_destroy(extent_buffer_cache
);
82 void extent_io_tree_init(struct extent_io_tree
*tree
,
83 struct address_space
*mapping
, gfp_t mask
)
85 tree
->state
.rb_node
= NULL
;
87 tree
->dirty_bytes
= 0;
88 spin_lock_init(&tree
->lock
);
89 spin_lock_init(&tree
->lru_lock
);
90 tree
->mapping
= mapping
;
91 INIT_LIST_HEAD(&tree
->buffer_lru
);
94 EXPORT_SYMBOL(extent_io_tree_init
);
96 void extent_io_tree_empty_lru(struct extent_io_tree
*tree
)
98 struct extent_buffer
*eb
;
99 while(!list_empty(&tree
->buffer_lru
)) {
100 eb
= list_entry(tree
->buffer_lru
.next
, struct extent_buffer
,
102 list_del_init(&eb
->lru
);
103 free_extent_buffer(eb
);
106 EXPORT_SYMBOL(extent_io_tree_empty_lru
);
108 struct extent_state
*alloc_extent_state(gfp_t mask
)
110 struct extent_state
*state
;
112 state
= kmem_cache_alloc(extent_state_cache
, mask
);
113 if (!state
|| IS_ERR(state
))
119 atomic_set(&state
->refs
, 1);
120 init_waitqueue_head(&state
->wq
);
123 EXPORT_SYMBOL(alloc_extent_state
);
125 void free_extent_state(struct extent_state
*state
)
129 if (atomic_dec_and_test(&state
->refs
)) {
130 WARN_ON(state
->tree
);
131 kmem_cache_free(extent_state_cache
, state
);
134 EXPORT_SYMBOL(free_extent_state
);
136 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
137 struct rb_node
*node
)
139 struct rb_node
** p
= &root
->rb_node
;
140 struct rb_node
* parent
= NULL
;
141 struct tree_entry
*entry
;
145 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
147 if (offset
< entry
->start
)
149 else if (offset
> entry
->end
)
155 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
156 rb_link_node(node
, parent
, p
);
157 rb_insert_color(node
, root
);
161 static struct rb_node
*__tree_search(struct rb_root
*root
, u64 offset
,
162 struct rb_node
**prev_ret
,
163 struct rb_node
**next_ret
)
165 struct rb_node
* n
= root
->rb_node
;
166 struct rb_node
*prev
= NULL
;
167 struct rb_node
*orig_prev
= NULL
;
168 struct tree_entry
*entry
;
169 struct tree_entry
*prev_entry
= NULL
;
172 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
176 if (offset
< entry
->start
)
178 else if (offset
> entry
->end
)
186 while(prev
&& offset
> prev_entry
->end
) {
187 prev
= rb_next(prev
);
188 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
195 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
196 while(prev
&& offset
< prev_entry
->start
) {
197 prev
= rb_prev(prev
);
198 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
205 static inline struct rb_node
*tree_search(struct rb_root
*root
, u64 offset
)
207 struct rb_node
*prev
= NULL
;
210 ret
= __tree_search(root
, offset
, &prev
, NULL
);
217 * utility function to look for merge candidates inside a given range.
218 * Any extents with matching state are merged together into a single
219 * extent in the tree. Extents with EXTENT_IO in their state field
220 * are not merged because the end_io handlers need to be able to do
221 * operations on them without sleeping (or doing allocations/splits).
223 * This should be called with the tree lock held.
225 static int merge_state(struct extent_io_tree
*tree
,
226 struct extent_state
*state
)
228 struct extent_state
*other
;
229 struct rb_node
*other_node
;
231 if (state
->state
& EXTENT_IOBITS
)
234 other_node
= rb_prev(&state
->rb_node
);
236 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
237 if (other
->end
== state
->start
- 1 &&
238 other
->state
== state
->state
) {
239 state
->start
= other
->start
;
241 rb_erase(&other
->rb_node
, &tree
->state
);
242 free_extent_state(other
);
245 other_node
= rb_next(&state
->rb_node
);
247 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
248 if (other
->start
== state
->end
+ 1 &&
249 other
->state
== state
->state
) {
250 other
->start
= state
->start
;
252 rb_erase(&state
->rb_node
, &tree
->state
);
253 free_extent_state(state
);
260 * insert an extent_state struct into the tree. 'bits' are set on the
261 * struct before it is inserted.
263 * This may return -EEXIST if the extent is already there, in which case the
264 * state struct is freed.
266 * The tree lock is not taken internally. This is a utility function and
267 * probably isn't what you want to call (see set/clear_extent_bit).
269 static int insert_state(struct extent_io_tree
*tree
,
270 struct extent_state
*state
, u64 start
, u64 end
,
273 struct rb_node
*node
;
276 printk("end < start %Lu %Lu\n", end
, start
);
279 if (bits
& EXTENT_DIRTY
)
280 tree
->dirty_bytes
+= end
- start
+ 1;
281 state
->state
|= bits
;
282 state
->start
= start
;
284 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
286 struct extent_state
*found
;
287 found
= rb_entry(node
, struct extent_state
, rb_node
);
288 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, start
, end
);
289 free_extent_state(state
);
293 merge_state(tree
, state
);
298 * split a given extent state struct in two, inserting the preallocated
299 * struct 'prealloc' as the newly created second half. 'split' indicates an
300 * offset inside 'orig' where it should be split.
303 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
304 * are two extent state structs in the tree:
305 * prealloc: [orig->start, split - 1]
306 * orig: [ split, orig->end ]
308 * The tree locks are not taken by this function. They need to be held
311 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
312 struct extent_state
*prealloc
, u64 split
)
314 struct rb_node
*node
;
315 prealloc
->start
= orig
->start
;
316 prealloc
->end
= split
- 1;
317 prealloc
->state
= orig
->state
;
320 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
322 struct extent_state
*found
;
323 found
= rb_entry(node
, struct extent_state
, rb_node
);
324 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, prealloc
->start
, prealloc
->end
);
325 free_extent_state(prealloc
);
328 prealloc
->tree
= tree
;
333 * utility function to clear some bits in an extent state struct.
334 * it will optionally wake up any one waiting on this state (wake == 1), or
335 * forcibly remove the state from the tree (delete == 1).
337 * If no bits are set on the state struct after clearing things, the
338 * struct is freed and removed from the tree
340 static int clear_state_bit(struct extent_io_tree
*tree
,
341 struct extent_state
*state
, int bits
, int wake
,
344 int ret
= state
->state
& bits
;
346 if ((bits
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
347 u64 range
= state
->end
- state
->start
+ 1;
348 WARN_ON(range
> tree
->dirty_bytes
);
349 tree
->dirty_bytes
-= range
;
351 state
->state
&= ~bits
;
354 if (delete || state
->state
== 0) {
356 rb_erase(&state
->rb_node
, &tree
->state
);
358 free_extent_state(state
);
363 merge_state(tree
, state
);
369 * clear some bits on a range in the tree. This may require splitting
370 * or inserting elements in the tree, so the gfp mask is used to
371 * indicate which allocations or sleeping are allowed.
373 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
374 * the given range from the tree regardless of state (ie for truncate).
376 * the range [start, end] is inclusive.
378 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
379 * bits were already set, or zero if none of the bits were already set.
381 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
382 int bits
, int wake
, int delete, gfp_t mask
)
384 struct extent_state
*state
;
385 struct extent_state
*prealloc
= NULL
;
386 struct rb_node
*node
;
392 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
393 prealloc
= alloc_extent_state(mask
);
398 spin_lock_irqsave(&tree
->lock
, flags
);
400 * this search will find the extents that end after
403 node
= tree_search(&tree
->state
, start
);
406 state
= rb_entry(node
, struct extent_state
, rb_node
);
407 if (state
->start
> end
)
409 WARN_ON(state
->end
< start
);
412 * | ---- desired range ---- |
414 * | ------------- state -------------- |
416 * We need to split the extent we found, and may flip
417 * bits on second half.
419 * If the extent we found extends past our range, we
420 * just split and search again. It'll get split again
421 * the next time though.
423 * If the extent we found is inside our range, we clear
424 * the desired bit on it.
427 if (state
->start
< start
) {
429 prealloc
= alloc_extent_state(GFP_ATOMIC
);
430 err
= split_state(tree
, state
, prealloc
, start
);
431 BUG_ON(err
== -EEXIST
);
435 if (state
->end
<= end
) {
436 start
= state
->end
+ 1;
437 set
|= clear_state_bit(tree
, state
, bits
,
440 start
= state
->start
;
445 * | ---- desired range ---- |
447 * We need to split the extent, and clear the bit
450 if (state
->start
<= end
&& state
->end
> end
) {
452 prealloc
= alloc_extent_state(GFP_ATOMIC
);
453 err
= split_state(tree
, state
, prealloc
, end
+ 1);
454 BUG_ON(err
== -EEXIST
);
458 set
|= clear_state_bit(tree
, prealloc
, bits
,
464 start
= state
->end
+ 1;
465 set
|= clear_state_bit(tree
, state
, bits
, wake
, delete);
469 spin_unlock_irqrestore(&tree
->lock
, flags
);
471 free_extent_state(prealloc
);
478 spin_unlock_irqrestore(&tree
->lock
, flags
);
479 if (mask
& __GFP_WAIT
)
483 EXPORT_SYMBOL(clear_extent_bit
);
485 static int wait_on_state(struct extent_io_tree
*tree
,
486 struct extent_state
*state
)
489 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
490 spin_unlock_irq(&tree
->lock
);
492 spin_lock_irq(&tree
->lock
);
493 finish_wait(&state
->wq
, &wait
);
498 * waits for one or more bits to clear on a range in the state tree.
499 * The range [start, end] is inclusive.
500 * The tree lock is taken by this function
502 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
504 struct extent_state
*state
;
505 struct rb_node
*node
;
507 spin_lock_irq(&tree
->lock
);
511 * this search will find all the extents that end after
514 node
= tree_search(&tree
->state
, start
);
518 state
= rb_entry(node
, struct extent_state
, rb_node
);
520 if (state
->start
> end
)
523 if (state
->state
& bits
) {
524 start
= state
->start
;
525 atomic_inc(&state
->refs
);
526 wait_on_state(tree
, state
);
527 free_extent_state(state
);
530 start
= state
->end
+ 1;
535 if (need_resched()) {
536 spin_unlock_irq(&tree
->lock
);
538 spin_lock_irq(&tree
->lock
);
542 spin_unlock_irq(&tree
->lock
);
545 EXPORT_SYMBOL(wait_extent_bit
);
547 static void set_state_bits(struct extent_io_tree
*tree
,
548 struct extent_state
*state
,
551 if ((bits
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
552 u64 range
= state
->end
- state
->start
+ 1;
553 tree
->dirty_bytes
+= range
;
555 state
->state
|= bits
;
559 * set some bits on a range in the tree. This may require allocations
560 * or sleeping, so the gfp mask is used to indicate what is allowed.
562 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
563 * range already has the desired bits set. The start of the existing
564 * range is returned in failed_start in this case.
566 * [start, end] is inclusive
567 * This takes the tree lock.
569 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
,
570 int exclusive
, u64
*failed_start
, gfp_t mask
)
572 struct extent_state
*state
;
573 struct extent_state
*prealloc
= NULL
;
574 struct rb_node
*node
;
581 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
582 prealloc
= alloc_extent_state(mask
);
587 spin_lock_irqsave(&tree
->lock
, flags
);
589 * this search will find all the extents that end after
592 node
= tree_search(&tree
->state
, start
);
594 err
= insert_state(tree
, prealloc
, start
, end
, bits
);
596 BUG_ON(err
== -EEXIST
);
600 state
= rb_entry(node
, struct extent_state
, rb_node
);
601 last_start
= state
->start
;
602 last_end
= state
->end
;
605 * | ---- desired range ---- |
608 * Just lock what we found and keep going
610 if (state
->start
== start
&& state
->end
<= end
) {
611 set
= state
->state
& bits
;
612 if (set
&& exclusive
) {
613 *failed_start
= state
->start
;
617 set_state_bits(tree
, state
, bits
);
618 start
= state
->end
+ 1;
619 merge_state(tree
, state
);
624 * | ---- desired range ---- |
627 * | ------------- state -------------- |
629 * We need to split the extent we found, and may flip bits on
632 * If the extent we found extends past our
633 * range, we just split and search again. It'll get split
634 * again the next time though.
636 * If the extent we found is inside our range, we set the
639 if (state
->start
< start
) {
640 set
= state
->state
& bits
;
641 if (exclusive
&& set
) {
642 *failed_start
= start
;
646 err
= split_state(tree
, state
, prealloc
, start
);
647 BUG_ON(err
== -EEXIST
);
651 if (state
->end
<= end
) {
652 set_state_bits(tree
, state
, bits
);
653 start
= state
->end
+ 1;
654 merge_state(tree
, state
);
656 start
= state
->start
;
661 * | ---- desired range ---- |
662 * | state | or | state |
664 * There's a hole, we need to insert something in it and
665 * ignore the extent we found.
667 if (state
->start
> start
) {
669 if (end
< last_start
)
672 this_end
= last_start
-1;
673 err
= insert_state(tree
, prealloc
, start
, this_end
,
676 BUG_ON(err
== -EEXIST
);
679 start
= this_end
+ 1;
683 * | ---- desired range ---- |
685 * We need to split the extent, and set the bit
688 if (state
->start
<= end
&& state
->end
> end
) {
689 set
= state
->state
& bits
;
690 if (exclusive
&& set
) {
691 *failed_start
= start
;
695 err
= split_state(tree
, state
, prealloc
, end
+ 1);
696 BUG_ON(err
== -EEXIST
);
698 set_state_bits(tree
, prealloc
, bits
);
699 merge_state(tree
, prealloc
);
707 spin_unlock_irqrestore(&tree
->lock
, flags
);
709 free_extent_state(prealloc
);
716 spin_unlock_irqrestore(&tree
->lock
, flags
);
717 if (mask
& __GFP_WAIT
)
721 EXPORT_SYMBOL(set_extent_bit
);
723 /* wrappers around set/clear extent bit */
724 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
727 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
730 EXPORT_SYMBOL(set_extent_dirty
);
732 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
733 int bits
, gfp_t mask
)
735 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
738 EXPORT_SYMBOL(set_extent_bits
);
740 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
741 int bits
, gfp_t mask
)
743 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, mask
);
745 EXPORT_SYMBOL(clear_extent_bits
);
747 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
750 return set_extent_bit(tree
, start
, end
,
751 EXTENT_DELALLOC
| EXTENT_DIRTY
, 0, NULL
,
754 EXPORT_SYMBOL(set_extent_delalloc
);
756 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
759 return clear_extent_bit(tree
, start
, end
,
760 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, mask
);
762 EXPORT_SYMBOL(clear_extent_dirty
);
764 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
767 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
770 EXPORT_SYMBOL(set_extent_new
);
772 int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
775 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0, mask
);
777 EXPORT_SYMBOL(clear_extent_new
);
779 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
782 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, NULL
,
785 EXPORT_SYMBOL(set_extent_uptodate
);
787 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
790 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0, mask
);
792 EXPORT_SYMBOL(clear_extent_uptodate
);
794 int set_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
797 return set_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
,
800 EXPORT_SYMBOL(set_extent_writeback
);
802 int clear_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
805 return clear_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 1, 0, mask
);
807 EXPORT_SYMBOL(clear_extent_writeback
);
809 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
811 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
813 EXPORT_SYMBOL(wait_on_extent_writeback
);
815 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
820 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
821 &failed_start
, mask
);
822 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
823 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
824 start
= failed_start
;
828 WARN_ON(start
> end
);
832 EXPORT_SYMBOL(lock_extent
);
834 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
837 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, mask
);
839 EXPORT_SYMBOL(unlock_extent
);
842 * helper function to set pages and extents in the tree dirty
844 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
846 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
847 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
850 while (index
<= end_index
) {
851 page
= find_get_page(tree
->mapping
, index
);
853 __set_page_dirty_nobuffers(page
);
854 page_cache_release(page
);
857 set_extent_dirty(tree
, start
, end
, GFP_NOFS
);
860 EXPORT_SYMBOL(set_range_dirty
);
863 * helper function to set both pages and extents in the tree writeback
865 int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
867 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
868 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
871 while (index
<= end_index
) {
872 page
= find_get_page(tree
->mapping
, index
);
874 set_page_writeback(page
);
875 page_cache_release(page
);
878 set_extent_writeback(tree
, start
, end
, GFP_NOFS
);
881 EXPORT_SYMBOL(set_range_writeback
);
883 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
884 u64
*start_ret
, u64
*end_ret
, int bits
)
886 struct rb_node
*node
;
887 struct extent_state
*state
;
890 spin_lock_irq(&tree
->lock
);
892 * this search will find all the extents that end after
895 node
= tree_search(&tree
->state
, start
);
896 if (!node
|| IS_ERR(node
)) {
901 state
= rb_entry(node
, struct extent_state
, rb_node
);
902 if (state
->end
>= start
&& (state
->state
& bits
)) {
903 *start_ret
= state
->start
;
904 *end_ret
= state
->end
;
908 node
= rb_next(node
);
913 spin_unlock_irq(&tree
->lock
);
916 EXPORT_SYMBOL(find_first_extent_bit
);
918 u64
find_lock_delalloc_range(struct extent_io_tree
*tree
,
919 u64
*start
, u64
*end
, u64 max_bytes
)
921 struct rb_node
*node
;
922 struct extent_state
*state
;
923 u64 cur_start
= *start
;
927 spin_lock_irq(&tree
->lock
);
929 * this search will find all the extents that end after
933 node
= tree_search(&tree
->state
, cur_start
);
934 if (!node
|| IS_ERR(node
)) {
940 state
= rb_entry(node
, struct extent_state
, rb_node
);
941 if (found
&& state
->start
!= cur_start
) {
944 if (!(state
->state
& EXTENT_DELALLOC
)) {
950 struct extent_state
*prev_state
;
951 struct rb_node
*prev_node
= node
;
953 prev_node
= rb_prev(prev_node
);
956 prev_state
= rb_entry(prev_node
,
959 if (!(prev_state
->state
& EXTENT_DELALLOC
))
965 if (state
->state
& EXTENT_LOCKED
) {
967 atomic_inc(&state
->refs
);
968 prepare_to_wait(&state
->wq
, &wait
,
969 TASK_UNINTERRUPTIBLE
);
970 spin_unlock_irq(&tree
->lock
);
972 spin_lock_irq(&tree
->lock
);
973 finish_wait(&state
->wq
, &wait
);
974 free_extent_state(state
);
977 state
->state
|= EXTENT_LOCKED
;
979 *start
= state
->start
;
982 cur_start
= state
->end
+ 1;
983 node
= rb_next(node
);
986 total_bytes
+= state
->end
- state
->start
+ 1;
987 if (total_bytes
>= max_bytes
)
991 spin_unlock_irq(&tree
->lock
);
995 u64
count_range_bits(struct extent_io_tree
*tree
,
996 u64
*start
, u64 search_end
, u64 max_bytes
,
999 struct rb_node
*node
;
1000 struct extent_state
*state
;
1001 u64 cur_start
= *start
;
1002 u64 total_bytes
= 0;
1005 if (search_end
<= cur_start
) {
1006 printk("search_end %Lu start %Lu\n", search_end
, cur_start
);
1011 spin_lock_irq(&tree
->lock
);
1012 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1013 total_bytes
= tree
->dirty_bytes
;
1017 * this search will find all the extents that end after
1020 node
= tree_search(&tree
->state
, cur_start
);
1021 if (!node
|| IS_ERR(node
)) {
1026 state
= rb_entry(node
, struct extent_state
, rb_node
);
1027 if (state
->start
> search_end
)
1029 if (state
->end
>= cur_start
&& (state
->state
& bits
)) {
1030 total_bytes
+= min(search_end
, state
->end
) + 1 -
1031 max(cur_start
, state
->start
);
1032 if (total_bytes
>= max_bytes
)
1035 *start
= state
->start
;
1039 node
= rb_next(node
);
1044 spin_unlock_irq(&tree
->lock
);
1048 * helper function to lock both pages and extents in the tree.
1049 * pages must be locked first.
1051 int lock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1053 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1054 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1058 while (index
<= end_index
) {
1059 page
= grab_cache_page(tree
->mapping
, index
);
1065 err
= PTR_ERR(page
);
1070 lock_extent(tree
, start
, end
, GFP_NOFS
);
1075 * we failed above in getting the page at 'index', so we undo here
1076 * up to but not including the page at 'index'
1079 index
= start
>> PAGE_CACHE_SHIFT
;
1080 while (index
< end_index
) {
1081 page
= find_get_page(tree
->mapping
, index
);
1083 page_cache_release(page
);
1088 EXPORT_SYMBOL(lock_range
);
1091 * helper function to unlock both pages and extents in the tree.
1093 int unlock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1095 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1096 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1099 while (index
<= end_index
) {
1100 page
= find_get_page(tree
->mapping
, index
);
1102 page_cache_release(page
);
1105 unlock_extent(tree
, start
, end
, GFP_NOFS
);
1108 EXPORT_SYMBOL(unlock_range
);
1110 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1112 struct rb_node
*node
;
1113 struct extent_state
*state
;
1116 spin_lock_irq(&tree
->lock
);
1118 * this search will find all the extents that end after
1121 node
= tree_search(&tree
->state
, start
);
1122 if (!node
|| IS_ERR(node
)) {
1126 state
= rb_entry(node
, struct extent_state
, rb_node
);
1127 if (state
->start
!= start
) {
1131 state
->private = private;
1133 spin_unlock_irq(&tree
->lock
);
1137 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1139 struct rb_node
*node
;
1140 struct extent_state
*state
;
1143 spin_lock_irq(&tree
->lock
);
1145 * this search will find all the extents that end after
1148 node
= tree_search(&tree
->state
, start
);
1149 if (!node
|| IS_ERR(node
)) {
1153 state
= rb_entry(node
, struct extent_state
, rb_node
);
1154 if (state
->start
!= start
) {
1158 *private = state
->private;
1160 spin_unlock_irq(&tree
->lock
);
1165 * searches a range in the state tree for a given mask.
1166 * If 'filled' == 1, this returns 1 only if every extent in the tree
1167 * has the bits set. Otherwise, 1 is returned if any bit in the
1168 * range is found set.
1170 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1171 int bits
, int filled
)
1173 struct extent_state
*state
= NULL
;
1174 struct rb_node
*node
;
1176 unsigned long flags
;
1178 spin_lock_irqsave(&tree
->lock
, flags
);
1179 node
= tree_search(&tree
->state
, start
);
1180 while (node
&& start
<= end
) {
1181 state
= rb_entry(node
, struct extent_state
, rb_node
);
1183 if (filled
&& state
->start
> start
) {
1188 if (state
->start
> end
)
1191 if (state
->state
& bits
) {
1195 } else if (filled
) {
1199 start
= state
->end
+ 1;
1202 node
= rb_next(node
);
1209 spin_unlock_irqrestore(&tree
->lock
, flags
);
1212 EXPORT_SYMBOL(test_range_bit
);
1215 * helper function to set a given page up to date if all the
1216 * extents in the tree for that page are up to date
1218 static int check_page_uptodate(struct extent_io_tree
*tree
,
1221 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1222 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1223 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1))
1224 SetPageUptodate(page
);
1229 * helper function to unlock a page if all the extents in the tree
1230 * for that page are unlocked
1232 static int check_page_locked(struct extent_io_tree
*tree
,
1235 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1236 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1237 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0))
1243 * helper function to end page writeback if all the extents
1244 * in the tree for that page are done with writeback
1246 static int check_page_writeback(struct extent_io_tree
*tree
,
1249 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1250 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1251 if (!test_range_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 0))
1252 end_page_writeback(page
);
1256 /* lots and lots of room for performance fixes in the end_bio funcs */
1259 * after a writepage IO is done, we need to:
1260 * clear the uptodate bits on error
1261 * clear the writeback bits in the extent tree for this IO
1262 * end_page_writeback if the page has no more pending IO
1264 * Scheduling is not allowed, so the extent state tree is expected
1265 * to have one and only one object corresponding to this IO.
1267 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1268 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1270 static int end_bio_extent_writepage(struct bio
*bio
,
1271 unsigned int bytes_done
, int err
)
1274 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1275 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1276 struct extent_state
*state
= bio
->bi_private
;
1277 struct extent_io_tree
*tree
= state
->tree
;
1278 struct rb_node
*node
;
1283 unsigned long flags
;
1285 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1290 struct page
*page
= bvec
->bv_page
;
1291 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1293 end
= start
+ bvec
->bv_len
- 1;
1295 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1300 if (--bvec
>= bio
->bi_io_vec
)
1301 prefetchw(&bvec
->bv_page
->flags
);
1304 clear_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1305 ClearPageUptodate(page
);
1309 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1310 tree
->ops
->writepage_end_io_hook(page
, start
, end
,
1315 * bios can get merged in funny ways, and so we need to
1316 * be careful with the state variable. We know the
1317 * state won't be merged with others because it has
1318 * WRITEBACK set, but we can't be sure each biovec is
1319 * sequential in the file. So, if our cached state
1320 * doesn't match the expected end, search the tree
1321 * for the correct one.
1324 spin_lock_irqsave(&tree
->lock
, flags
);
1325 if (!state
|| state
->end
!= end
) {
1327 node
= __tree_search(&tree
->state
, start
, NULL
, NULL
);
1329 state
= rb_entry(node
, struct extent_state
,
1331 if (state
->end
!= end
||
1332 !(state
->state
& EXTENT_WRITEBACK
))
1336 spin_unlock_irqrestore(&tree
->lock
, flags
);
1337 clear_extent_writeback(tree
, start
,
1344 struct extent_state
*clear
= state
;
1346 node
= rb_prev(&state
->rb_node
);
1348 state
= rb_entry(node
,
1349 struct extent_state
,
1355 clear_state_bit(tree
, clear
, EXTENT_WRITEBACK
,
1366 /* before releasing the lock, make sure the next state
1367 * variable has the expected bits set and corresponds
1368 * to the correct offsets in the file
1370 if (state
&& (state
->end
+ 1 != start
||
1371 !state
->state
& EXTENT_WRITEBACK
)) {
1374 spin_unlock_irqrestore(&tree
->lock
, flags
);
1378 end_page_writeback(page
);
1380 check_page_writeback(tree
, page
);
1381 } while (bvec
>= bio
->bi_io_vec
);
1383 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1389 * after a readpage IO is done, we need to:
1390 * clear the uptodate bits on error
1391 * set the uptodate bits if things worked
1392 * set the page up to date if all extents in the tree are uptodate
1393 * clear the lock bit in the extent tree
1394 * unlock the page if there are no other extents locked for it
1396 * Scheduling is not allowed, so the extent state tree is expected
1397 * to have one and only one object corresponding to this IO.
1399 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1400 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1402 static int end_bio_extent_readpage(struct bio
*bio
,
1403 unsigned int bytes_done
, int err
)
1406 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1407 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1408 struct extent_state
*state
= bio
->bi_private
;
1409 struct extent_io_tree
*tree
= state
->tree
;
1410 struct rb_node
*node
;
1414 unsigned long flags
;
1418 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1424 struct page
*page
= bvec
->bv_page
;
1425 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1427 end
= start
+ bvec
->bv_len
- 1;
1429 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1434 if (--bvec
>= bio
->bi_io_vec
)
1435 prefetchw(&bvec
->bv_page
->flags
);
1437 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1438 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1444 spin_lock_irqsave(&tree
->lock
, flags
);
1445 if (!state
|| state
->end
!= end
) {
1447 node
= __tree_search(&tree
->state
, start
, NULL
, NULL
);
1449 state
= rb_entry(node
, struct extent_state
,
1451 if (state
->end
!= end
||
1452 !(state
->state
& EXTENT_LOCKED
))
1456 spin_unlock_irqrestore(&tree
->lock
, flags
);
1457 set_extent_uptodate(tree
, start
, end
,
1459 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1466 struct extent_state
*clear
= state
;
1468 node
= rb_prev(&state
->rb_node
);
1470 state
= rb_entry(node
,
1471 struct extent_state
,
1476 clear
->state
|= EXTENT_UPTODATE
;
1477 clear_state_bit(tree
, clear
, EXTENT_LOCKED
,
1488 /* before releasing the lock, make sure the next state
1489 * variable has the expected bits set and corresponds
1490 * to the correct offsets in the file
1492 if (state
&& (state
->end
+ 1 != start
||
1493 !state
->state
& EXTENT_WRITEBACK
)) {
1496 spin_unlock_irqrestore(&tree
->lock
, flags
);
1500 SetPageUptodate(page
);
1502 ClearPageUptodate(page
);
1508 check_page_uptodate(tree
, page
);
1510 ClearPageUptodate(page
);
1513 check_page_locked(tree
, page
);
1515 } while (bvec
>= bio
->bi_io_vec
);
1518 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1524 * IO done from prepare_write is pretty simple, we just unlock
1525 * the structs in the extent tree when done, and set the uptodate bits
1528 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1529 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1531 static int end_bio_extent_preparewrite(struct bio
*bio
,
1532 unsigned int bytes_done
, int err
)
1535 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1536 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1537 struct extent_state
*state
= bio
->bi_private
;
1538 struct extent_io_tree
*tree
= state
->tree
;
1542 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1548 struct page
*page
= bvec
->bv_page
;
1549 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1551 end
= start
+ bvec
->bv_len
- 1;
1553 if (--bvec
>= bio
->bi_io_vec
)
1554 prefetchw(&bvec
->bv_page
->flags
);
1557 set_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1559 ClearPageUptodate(page
);
1563 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1565 } while (bvec
>= bio
->bi_io_vec
);
1568 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1574 extent_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1579 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1581 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1582 while (!bio
&& (nr_vecs
/= 2))
1583 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1587 bio
->bi_bdev
= bdev
;
1588 bio
->bi_sector
= first_sector
;
1593 static int submit_one_bio(int rw
, struct bio
*bio
)
1597 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1598 struct page
*page
= bvec
->bv_page
;
1599 struct extent_io_tree
*tree
= bio
->bi_private
;
1600 struct rb_node
*node
;
1601 struct extent_state
*state
;
1605 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1606 end
= start
+ bvec
->bv_len
- 1;
1608 spin_lock_irq(&tree
->lock
);
1609 node
= __tree_search(&tree
->state
, start
, NULL
, NULL
);
1611 state
= rb_entry(node
, struct extent_state
, rb_node
);
1612 while(state
->end
< end
) {
1613 node
= rb_next(node
);
1614 state
= rb_entry(node
, struct extent_state
, rb_node
);
1616 BUG_ON(state
->end
!= end
);
1617 spin_unlock_irq(&tree
->lock
);
1619 bio
->bi_private
= state
;
1623 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1624 if (maxsector
< bio
->bi_sector
) {
1625 printk("sector too large max %Lu got %llu\n", maxsector
,
1626 (unsigned long long)bio
->bi_sector
);
1630 submit_bio(rw
, bio
);
1631 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1637 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1638 struct page
*page
, sector_t sector
,
1639 size_t size
, unsigned long offset
,
1640 struct block_device
*bdev
,
1641 struct bio
**bio_ret
,
1642 unsigned long max_pages
,
1643 bio_end_io_t end_io_func
)
1649 if (bio_ret
&& *bio_ret
) {
1651 if (bio
->bi_sector
+ (bio
->bi_size
>> 9) != sector
||
1652 bio_add_page(bio
, page
, size
, offset
) < size
) {
1653 ret
= submit_one_bio(rw
, bio
);
1659 nr
= min_t(int, max_pages
, bio_get_nr_vecs(bdev
));
1660 bio
= extent_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1662 printk("failed to allocate bio nr %d\n", nr
);
1666 bio_add_page(bio
, page
, size
, offset
);
1667 bio
->bi_end_io
= end_io_func
;
1668 bio
->bi_private
= tree
;
1673 ret
= submit_one_bio(rw
, bio
);
1679 void set_page_extent_mapped(struct page
*page
)
1681 if (!PagePrivate(page
)) {
1682 SetPagePrivate(page
);
1683 WARN_ON(!page
->mapping
->a_ops
->invalidatepage
);
1684 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1685 page_cache_get(page
);
1689 void set_page_extent_head(struct page
*page
, unsigned long len
)
1691 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1695 * basic readpage implementation. Locked extent state structs are inserted
1696 * into the tree that are removed when the IO is done (by the end_io
1699 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1701 get_extent_t
*get_extent
,
1704 struct inode
*inode
= page
->mapping
->host
;
1705 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1706 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1710 u64 last_byte
= i_size_read(inode
);
1714 struct extent_map
*em
;
1715 struct block_device
*bdev
;
1718 size_t page_offset
= 0;
1720 size_t blocksize
= inode
->i_sb
->s_blocksize
;
1722 set_page_extent_mapped(page
);
1725 lock_extent(tree
, start
, end
, GFP_NOFS
);
1727 while (cur
<= end
) {
1728 if (cur
>= last_byte
) {
1730 iosize
= PAGE_CACHE_SIZE
- page_offset
;
1731 userpage
= kmap_atomic(page
, KM_USER0
);
1732 memset(userpage
+ page_offset
, 0, iosize
);
1733 flush_dcache_page(page
);
1734 kunmap_atomic(userpage
, KM_USER0
);
1735 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1737 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1740 em
= get_extent(inode
, page
, page_offset
, cur
,
1742 if (IS_ERR(em
) || !em
) {
1744 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
1748 extent_offset
= cur
- em
->start
;
1749 BUG_ON(extent_map_end(em
) <= cur
);
1752 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
1753 cur_end
= min(extent_map_end(em
) - 1, end
);
1754 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
1755 sector
= (em
->block_start
+ extent_offset
) >> 9;
1757 block_start
= em
->block_start
;
1758 free_extent_map(em
);
1761 /* we've found a hole, just zero and go on */
1762 if (block_start
== EXTENT_MAP_HOLE
) {
1764 userpage
= kmap_atomic(page
, KM_USER0
);
1765 memset(userpage
+ page_offset
, 0, iosize
);
1766 flush_dcache_page(page
);
1767 kunmap_atomic(userpage
, KM_USER0
);
1769 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1771 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1773 page_offset
+= iosize
;
1776 /* the get_extent function already copied into the page */
1777 if (test_range_bit(tree
, cur
, cur_end
, EXTENT_UPTODATE
, 1)) {
1778 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1780 page_offset
+= iosize
;
1783 /* we have an inline extent but it didn't get marked up
1784 * to date. Error out
1786 if (block_start
== EXTENT_MAP_INLINE
) {
1788 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1790 page_offset
+= iosize
;
1795 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
1796 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
1800 unsigned long nr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
1802 ret
= submit_extent_page(READ
, tree
, page
,
1803 sector
, iosize
, page_offset
,
1805 end_bio_extent_readpage
);
1810 page_offset
+= iosize
;
1814 if (!PageError(page
))
1815 SetPageUptodate(page
);
1821 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
1822 get_extent_t
*get_extent
)
1824 struct bio
*bio
= NULL
;
1827 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
);
1829 submit_one_bio(READ
, bio
);
1832 EXPORT_SYMBOL(extent_read_full_page
);
1835 * the writepage semantics are similar to regular writepage. extent
1836 * records are inserted to lock ranges in the tree, and as dirty areas
1837 * are found, they are marked writeback. Then the lock bits are removed
1838 * and the end_io handler clears the writeback ranges
1840 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
1843 struct inode
*inode
= page
->mapping
->host
;
1844 struct extent_page_data
*epd
= data
;
1845 struct extent_io_tree
*tree
= epd
->tree
;
1846 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1848 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1852 u64 last_byte
= i_size_read(inode
);
1856 struct extent_map
*em
;
1857 struct block_device
*bdev
;
1860 size_t page_offset
= 0;
1862 loff_t i_size
= i_size_read(inode
);
1863 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1867 WARN_ON(!PageLocked(page
));
1868 if (page
->index
> end_index
) {
1869 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
1874 if (page
->index
== end_index
) {
1877 size_t offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
1879 userpage
= kmap_atomic(page
, KM_USER0
);
1880 memset(userpage
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
1881 flush_dcache_page(page
);
1882 kunmap_atomic(userpage
, KM_USER0
);
1885 set_page_extent_mapped(page
);
1887 delalloc_start
= start
;
1889 while(delalloc_end
< page_end
) {
1890 nr_delalloc
= find_lock_delalloc_range(tree
, &delalloc_start
,
1893 if (nr_delalloc
== 0) {
1894 delalloc_start
= delalloc_end
+ 1;
1897 tree
->ops
->fill_delalloc(inode
, delalloc_start
,
1899 clear_extent_bit(tree
, delalloc_start
,
1901 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1903 delalloc_start
= delalloc_end
+ 1;
1905 lock_extent(tree
, start
, page_end
, GFP_NOFS
);
1908 if (test_range_bit(tree
, start
, page_end
, EXTENT_DELALLOC
, 0)) {
1909 printk("found delalloc bits after lock_extent\n");
1912 if (last_byte
<= start
) {
1913 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
1917 set_extent_uptodate(tree
, start
, page_end
, GFP_NOFS
);
1918 blocksize
= inode
->i_sb
->s_blocksize
;
1920 while (cur
<= end
) {
1921 if (cur
>= last_byte
) {
1922 clear_extent_dirty(tree
, cur
, page_end
, GFP_NOFS
);
1925 em
= epd
->get_extent(inode
, page
, page_offset
, cur
,
1927 if (IS_ERR(em
) || !em
) {
1932 extent_offset
= cur
- em
->start
;
1933 BUG_ON(extent_map_end(em
) <= cur
);
1935 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
1936 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
1937 sector
= (em
->block_start
+ extent_offset
) >> 9;
1939 block_start
= em
->block_start
;
1940 free_extent_map(em
);
1943 if (block_start
== EXTENT_MAP_HOLE
||
1944 block_start
== EXTENT_MAP_INLINE
) {
1945 clear_extent_dirty(tree
, cur
,
1946 cur
+ iosize
- 1, GFP_NOFS
);
1948 page_offset
+= iosize
;
1952 /* leave this out until we have a page_mkwrite call */
1953 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
1956 page_offset
+= iosize
;
1959 clear_extent_dirty(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1960 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
1961 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
1969 unsigned long max_nr
= end_index
+ 1;
1970 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
1971 if (!PageWriteback(page
)) {
1972 printk("warning page %lu not writeback, "
1973 "cur %llu end %llu\n", page
->index
,
1974 (unsigned long long)cur
,
1975 (unsigned long long)end
);
1978 ret
= submit_extent_page(WRITE
, tree
, page
, sector
,
1979 iosize
, page_offset
, bdev
,
1981 end_bio_extent_writepage
);
1986 page_offset
+= iosize
;
1991 /* make sure the mapping tag for page dirty gets cleared */
1992 set_page_writeback(page
);
1993 end_page_writeback(page
);
1995 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
2000 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2002 /* Taken directly from 2.6.23 for 2.6.18 back port */
2003 typedef int (*writepage_t
)(struct page
*page
, struct writeback_control
*wbc
,
2007 * write_cache_pages - walk the list of dirty pages of the given address space
2008 * and write all of them.
2009 * @mapping: address space structure to write
2010 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2011 * @writepage: function called for each page
2012 * @data: data passed to writepage function
2014 * If a page is already under I/O, write_cache_pages() skips it, even
2015 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2016 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2017 * and msync() need to guarantee that all the data which was dirty at the time
2018 * the call was made get new I/O started against them. If wbc->sync_mode is
2019 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2020 * existing IO to complete.
2022 static int write_cache_pages(struct address_space
*mapping
,
2023 struct writeback_control
*wbc
, writepage_t writepage
,
2026 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
2029 struct pagevec pvec
;
2032 pgoff_t end
; /* Inclusive */
2034 int range_whole
= 0;
2036 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2037 wbc
->encountered_congestion
= 1;
2041 pagevec_init(&pvec
, 0);
2042 if (wbc
->range_cyclic
) {
2043 index
= mapping
->writeback_index
; /* Start from prev offset */
2046 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2047 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2048 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2053 while (!done
&& (index
<= end
) &&
2054 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2055 PAGECACHE_TAG_DIRTY
,
2056 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2060 for (i
= 0; i
< nr_pages
; i
++) {
2061 struct page
*page
= pvec
.pages
[i
];
2064 * At this point we hold neither mapping->tree_lock nor
2065 * lock on the page itself: the page may be truncated or
2066 * invalidated (changing page->mapping to NULL), or even
2067 * swizzled back from swapper_space to tmpfs file
2072 if (unlikely(page
->mapping
!= mapping
)) {
2077 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2083 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2084 wait_on_page_writeback(page
);
2086 if (PageWriteback(page
) ||
2087 !clear_page_dirty_for_io(page
)) {
2092 ret
= (*writepage
)(page
, wbc
, data
);
2094 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2098 if (ret
|| (--(wbc
->nr_to_write
) <= 0))
2100 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2101 wbc
->encountered_congestion
= 1;
2105 pagevec_release(&pvec
);
2108 if (!scanned
&& !done
) {
2110 * We hit the last page and there is more work to be done: wrap
2111 * back to the start of the file
2117 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2118 mapping
->writeback_index
= index
;
2123 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2124 get_extent_t
*get_extent
,
2125 struct writeback_control
*wbc
)
2128 struct address_space
*mapping
= page
->mapping
;
2129 struct extent_page_data epd
= {
2132 .get_extent
= get_extent
,
2134 struct writeback_control wbc_writepages
= {
2136 .sync_mode
= WB_SYNC_NONE
,
2137 .older_than_this
= NULL
,
2139 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2140 .range_end
= (loff_t
)-1,
2144 ret
= __extent_writepage(page
, wbc
, &epd
);
2146 write_cache_pages(mapping
, &wbc_writepages
, __extent_writepage
, &epd
);
2148 submit_one_bio(WRITE
, epd
.bio
);
2152 EXPORT_SYMBOL(extent_write_full_page
);
2155 int extent_writepages(struct extent_io_tree
*tree
,
2156 struct address_space
*mapping
,
2157 get_extent_t
*get_extent
,
2158 struct writeback_control
*wbc
)
2161 struct extent_page_data epd
= {
2164 .get_extent
= get_extent
,
2167 ret
= write_cache_pages(mapping
, wbc
, __extent_writepage
, &epd
);
2169 submit_one_bio(WRITE
, epd
.bio
);
2173 EXPORT_SYMBOL(extent_writepages
);
2175 int extent_readpages(struct extent_io_tree
*tree
,
2176 struct address_space
*mapping
,
2177 struct list_head
*pages
, unsigned nr_pages
,
2178 get_extent_t get_extent
)
2180 struct bio
*bio
= NULL
;
2182 struct pagevec pvec
;
2184 pagevec_init(&pvec
, 0);
2185 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2186 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2188 prefetchw(&page
->flags
);
2189 list_del(&page
->lru
);
2191 * what we want to do here is call add_to_page_cache_lru,
2192 * but that isn't exported, so we reproduce it here
2194 if (!add_to_page_cache(page
, mapping
,
2195 page
->index
, GFP_KERNEL
)) {
2197 /* open coding of lru_cache_add, also not exported */
2198 page_cache_get(page
);
2199 if (!pagevec_add(&pvec
, page
))
2200 __pagevec_lru_add(&pvec
);
2201 __extent_read_full_page(tree
, page
, get_extent
, &bio
);
2203 page_cache_release(page
);
2205 if (pagevec_count(&pvec
))
2206 __pagevec_lru_add(&pvec
);
2207 BUG_ON(!list_empty(pages
));
2209 submit_one_bio(READ
, bio
);
2212 EXPORT_SYMBOL(extent_readpages
);
2215 * basic invalidatepage code, this waits on any locked or writeback
2216 * ranges corresponding to the page, and then deletes any extent state
2217 * records from the tree
2219 int extent_invalidatepage(struct extent_io_tree
*tree
,
2220 struct page
*page
, unsigned long offset
)
2222 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2223 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2224 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2226 start
+= (offset
+ blocksize
-1) & ~(blocksize
- 1);
2230 lock_extent(tree
, start
, end
, GFP_NOFS
);
2231 wait_on_extent_writeback(tree
, start
, end
);
2232 clear_extent_bit(tree
, start
, end
,
2233 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
,
2237 EXPORT_SYMBOL(extent_invalidatepage
);
2240 * simple commit_write call, set_range_dirty is used to mark both
2241 * the pages and the extent records as dirty
2243 int extent_commit_write(struct extent_io_tree
*tree
,
2244 struct inode
*inode
, struct page
*page
,
2245 unsigned from
, unsigned to
)
2247 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2249 set_page_extent_mapped(page
);
2250 set_page_dirty(page
);
2252 if (pos
> inode
->i_size
) {
2253 i_size_write(inode
, pos
);
2254 mark_inode_dirty(inode
);
2258 EXPORT_SYMBOL(extent_commit_write
);
2260 int extent_prepare_write(struct extent_io_tree
*tree
,
2261 struct inode
*inode
, struct page
*page
,
2262 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2264 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2265 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2267 u64 orig_block_start
;
2270 struct extent_map
*em
;
2271 unsigned blocksize
= 1 << inode
->i_blkbits
;
2272 size_t page_offset
= 0;
2273 size_t block_off_start
;
2274 size_t block_off_end
;
2280 set_page_extent_mapped(page
);
2282 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2283 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2284 orig_block_start
= block_start
;
2286 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2287 while(block_start
<= block_end
) {
2288 em
= get_extent(inode
, page
, page_offset
, block_start
,
2289 block_end
- block_start
+ 1, 1);
2290 if (IS_ERR(em
) || !em
) {
2293 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2294 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2295 block_off_end
= block_off_start
+ blocksize
;
2296 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2298 if (!PageUptodate(page
) && isnew
&&
2299 (block_off_end
> to
|| block_off_start
< from
)) {
2302 kaddr
= kmap_atomic(page
, KM_USER0
);
2303 if (block_off_end
> to
)
2304 memset(kaddr
+ to
, 0, block_off_end
- to
);
2305 if (block_off_start
< from
)
2306 memset(kaddr
+ block_off_start
, 0,
2307 from
- block_off_start
);
2308 flush_dcache_page(page
);
2309 kunmap_atomic(kaddr
, KM_USER0
);
2311 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2312 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2313 !isnew
&& !PageUptodate(page
) &&
2314 (block_off_end
> to
|| block_off_start
< from
) &&
2315 !test_range_bit(tree
, block_start
, cur_end
,
2316 EXTENT_UPTODATE
, 1)) {
2318 u64 extent_offset
= block_start
- em
->start
;
2320 sector
= (em
->block_start
+ extent_offset
) >> 9;
2321 iosize
= (cur_end
- block_start
+ blocksize
) &
2322 ~((u64
)blocksize
- 1);
2324 * we've already got the extent locked, but we
2325 * need to split the state such that our end_bio
2326 * handler can clear the lock.
2328 set_extent_bit(tree
, block_start
,
2329 block_start
+ iosize
- 1,
2330 EXTENT_LOCKED
, 0, NULL
, GFP_NOFS
);
2331 ret
= submit_extent_page(READ
, tree
, page
,
2332 sector
, iosize
, page_offset
, em
->bdev
,
2334 end_bio_extent_preparewrite
);
2336 block_start
= block_start
+ iosize
;
2338 set_extent_uptodate(tree
, block_start
, cur_end
,
2340 unlock_extent(tree
, block_start
, cur_end
, GFP_NOFS
);
2341 block_start
= cur_end
+ 1;
2343 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2344 free_extent_map(em
);
2347 wait_extent_bit(tree
, orig_block_start
,
2348 block_end
, EXTENT_LOCKED
);
2350 check_page_uptodate(tree
, page
);
2352 /* FIXME, zero out newly allocated blocks on error */
2355 EXPORT_SYMBOL(extent_prepare_write
);
2358 * a helper for releasepage. As long as there are no locked extents
2359 * in the range corresponding to the page, both state records and extent
2360 * map records are removed
2362 int try_release_extent_mapping(struct extent_map_tree
*map
,
2363 struct extent_io_tree
*tree
, struct page
*page
,
2366 struct extent_map
*em
;
2367 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2368 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2369 u64 orig_start
= start
;
2372 if ((mask
& __GFP_WAIT
) &&
2373 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2374 while (start
<= end
) {
2375 spin_lock(&map
->lock
);
2376 em
= lookup_extent_mapping(map
, start
, end
);
2377 if (!em
|| IS_ERR(em
)) {
2378 spin_unlock(&map
->lock
);
2381 if (em
->start
!= start
) {
2382 spin_unlock(&map
->lock
);
2383 free_extent_map(em
);
2386 if (!test_range_bit(tree
, em
->start
,
2387 extent_map_end(em
) - 1,
2388 EXTENT_LOCKED
, 0)) {
2389 remove_extent_mapping(map
, em
);
2390 /* once for the rb tree */
2391 free_extent_map(em
);
2393 start
= extent_map_end(em
);
2394 spin_unlock(&map
->lock
);
2397 free_extent_map(em
);
2400 if (test_range_bit(tree
, orig_start
, end
, EXTENT_IOBITS
, 0))
2403 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2405 clear_extent_bit(tree
, orig_start
, end
, EXTENT_UPTODATE
,
2410 EXPORT_SYMBOL(try_release_extent_mapping
);
2412 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2413 get_extent_t
*get_extent
)
2415 struct inode
*inode
= mapping
->host
;
2416 u64 start
= iblock
<< inode
->i_blkbits
;
2417 sector_t sector
= 0;
2418 struct extent_map
*em
;
2420 em
= get_extent(inode
, NULL
, 0, start
, (1 << inode
->i_blkbits
), 0);
2421 if (!em
|| IS_ERR(em
))
2424 if (em
->block_start
== EXTENT_MAP_INLINE
||
2425 em
->block_start
== EXTENT_MAP_HOLE
)
2428 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2429 printk("bmap finds %Lu %Lu block %Lu\n", em
->start
, em
->len
, em
->block_start
);
2431 free_extent_map(em
);
2435 static int add_lru(struct extent_io_tree
*tree
, struct extent_buffer
*eb
)
2437 if (list_empty(&eb
->lru
)) {
2438 extent_buffer_get(eb
);
2439 list_add(&eb
->lru
, &tree
->buffer_lru
);
2441 if (tree
->lru_size
>= BUFFER_LRU_MAX
) {
2442 struct extent_buffer
*rm
;
2443 rm
= list_entry(tree
->buffer_lru
.prev
,
2444 struct extent_buffer
, lru
);
2446 list_del_init(&rm
->lru
);
2447 free_extent_buffer(rm
);
2450 list_move(&eb
->lru
, &tree
->buffer_lru
);
2453 static struct extent_buffer
*find_lru(struct extent_io_tree
*tree
,
2454 u64 start
, unsigned long len
)
2456 struct list_head
*lru
= &tree
->buffer_lru
;
2457 struct list_head
*cur
= lru
->next
;
2458 struct extent_buffer
*eb
;
2460 if (list_empty(lru
))
2464 eb
= list_entry(cur
, struct extent_buffer
, lru
);
2465 if (eb
->start
== start
&& eb
->len
== len
) {
2466 extent_buffer_get(eb
);
2470 } while (cur
!= lru
);
2474 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
2476 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
2477 (start
>> PAGE_CACHE_SHIFT
);
2480 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
2484 struct address_space
*mapping
;
2487 return eb
->first_page
;
2488 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
2489 mapping
= eb
->first_page
->mapping
;
2490 read_lock_irq(&mapping
->tree_lock
);
2491 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
2492 read_unlock_irq(&mapping
->tree_lock
);
2496 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
2501 struct extent_buffer
*eb
= NULL
;
2503 spin_lock(&tree
->lru_lock
);
2504 eb
= find_lru(tree
, start
, len
);
2505 spin_unlock(&tree
->lru_lock
);
2510 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
2511 INIT_LIST_HEAD(&eb
->lru
);
2514 atomic_set(&eb
->refs
, 1);
2519 static void __free_extent_buffer(struct extent_buffer
*eb
)
2521 kmem_cache_free(extent_buffer_cache
, eb
);
2524 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
2525 u64 start
, unsigned long len
,
2529 unsigned long num_pages
= num_extent_pages(start
, len
);
2531 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2532 struct extent_buffer
*eb
;
2534 struct address_space
*mapping
= tree
->mapping
;
2537 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2538 if (!eb
|| IS_ERR(eb
))
2541 if (eb
->flags
& EXTENT_BUFFER_FILLED
)
2545 eb
->first_page
= page0
;
2548 page_cache_get(page0
);
2549 mark_page_accessed(page0
);
2550 set_page_extent_mapped(page0
);
2551 WARN_ON(!PageUptodate(page0
));
2552 set_page_extent_head(page0
, len
);
2556 for (; i
< num_pages
; i
++, index
++) {
2557 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
2562 set_page_extent_mapped(p
);
2563 mark_page_accessed(p
);
2566 set_page_extent_head(p
, len
);
2568 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2570 if (!PageUptodate(p
))
2575 eb
->flags
|= EXTENT_UPTODATE
;
2576 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2579 spin_lock(&tree
->lru_lock
);
2581 spin_unlock(&tree
->lru_lock
);
2585 spin_lock(&tree
->lru_lock
);
2586 list_del_init(&eb
->lru
);
2587 spin_unlock(&tree
->lru_lock
);
2588 if (!atomic_dec_and_test(&eb
->refs
))
2590 for (index
= 1; index
< i
; index
++) {
2591 page_cache_release(extent_buffer_page(eb
, index
));
2594 page_cache_release(extent_buffer_page(eb
, 0));
2595 __free_extent_buffer(eb
);
2598 EXPORT_SYMBOL(alloc_extent_buffer
);
2600 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
2601 u64 start
, unsigned long len
,
2604 unsigned long num_pages
= num_extent_pages(start
, len
);
2606 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2607 struct extent_buffer
*eb
;
2609 struct address_space
*mapping
= tree
->mapping
;
2612 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2613 if (!eb
|| IS_ERR(eb
))
2616 if (eb
->flags
& EXTENT_BUFFER_FILLED
)
2619 for (i
= 0; i
< num_pages
; i
++, index
++) {
2620 p
= find_lock_page(mapping
, index
);
2624 set_page_extent_mapped(p
);
2625 mark_page_accessed(p
);
2629 set_page_extent_head(p
, len
);
2631 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2634 if (!PageUptodate(p
))
2639 eb
->flags
|= EXTENT_UPTODATE
;
2640 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2643 spin_lock(&tree
->lru_lock
);
2645 spin_unlock(&tree
->lru_lock
);
2648 spin_lock(&tree
->lru_lock
);
2649 list_del_init(&eb
->lru
);
2650 spin_unlock(&tree
->lru_lock
);
2651 if (!atomic_dec_and_test(&eb
->refs
))
2653 for (index
= 1; index
< i
; index
++) {
2654 page_cache_release(extent_buffer_page(eb
, index
));
2657 page_cache_release(extent_buffer_page(eb
, 0));
2658 __free_extent_buffer(eb
);
2661 EXPORT_SYMBOL(find_extent_buffer
);
2663 void free_extent_buffer(struct extent_buffer
*eb
)
2666 unsigned long num_pages
;
2671 if (!atomic_dec_and_test(&eb
->refs
))
2674 WARN_ON(!list_empty(&eb
->lru
));
2675 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2677 for (i
= 1; i
< num_pages
; i
++) {
2678 page_cache_release(extent_buffer_page(eb
, i
));
2680 page_cache_release(extent_buffer_page(eb
, 0));
2681 __free_extent_buffer(eb
);
2683 EXPORT_SYMBOL(free_extent_buffer
);
2685 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
2686 struct extent_buffer
*eb
)
2690 unsigned long num_pages
;
2693 u64 start
= eb
->start
;
2694 u64 end
= start
+ eb
->len
- 1;
2696 set
= clear_extent_dirty(tree
, start
, end
, GFP_NOFS
);
2697 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2699 for (i
= 0; i
< num_pages
; i
++) {
2700 page
= extent_buffer_page(eb
, i
);
2703 set_page_extent_head(page
, eb
->len
);
2705 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2708 * if we're on the last page or the first page and the
2709 * block isn't aligned on a page boundary, do extra checks
2710 * to make sure we don't clean page that is partially dirty
2712 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2713 ((i
== num_pages
- 1) &&
2714 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2715 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2716 end
= start
+ PAGE_CACHE_SIZE
- 1;
2717 if (test_range_bit(tree
, start
, end
,
2723 clear_page_dirty_for_io(page
);
2724 read_lock_irq(&page
->mapping
->tree_lock
);
2725 if (!PageDirty(page
)) {
2726 radix_tree_tag_clear(&page
->mapping
->page_tree
,
2728 PAGECACHE_TAG_DIRTY
);
2730 read_unlock_irq(&page
->mapping
->tree_lock
);
2735 EXPORT_SYMBOL(clear_extent_buffer_dirty
);
2737 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
2738 struct extent_buffer
*eb
)
2740 return wait_on_extent_writeback(tree
, eb
->start
,
2741 eb
->start
+ eb
->len
- 1);
2743 EXPORT_SYMBOL(wait_on_extent_buffer_writeback
);
2745 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
2746 struct extent_buffer
*eb
)
2749 unsigned long num_pages
;
2751 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2752 for (i
= 0; i
< num_pages
; i
++) {
2753 struct page
*page
= extent_buffer_page(eb
, i
);
2754 /* writepage may need to do something special for the
2755 * first page, we have to make sure page->private is
2756 * properly set. releasepage may drop page->private
2757 * on us if the page isn't already dirty.
2761 set_page_extent_head(page
, eb
->len
);
2762 } else if (PagePrivate(page
) &&
2763 page
->private != EXTENT_PAGE_PRIVATE
) {
2765 set_page_extent_mapped(page
);
2768 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
2772 return set_extent_dirty(tree
, eb
->start
,
2773 eb
->start
+ eb
->len
- 1, GFP_NOFS
);
2775 EXPORT_SYMBOL(set_extent_buffer_dirty
);
2777 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
2778 struct extent_buffer
*eb
)
2782 unsigned long num_pages
;
2784 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2786 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2788 for (i
= 0; i
< num_pages
; i
++) {
2789 page
= extent_buffer_page(eb
, i
);
2790 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2791 ((i
== num_pages
- 1) &&
2792 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2793 check_page_uptodate(tree
, page
);
2796 SetPageUptodate(page
);
2800 EXPORT_SYMBOL(set_extent_buffer_uptodate
);
2802 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
2803 struct extent_buffer
*eb
)
2805 if (eb
->flags
& EXTENT_UPTODATE
)
2807 return test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2808 EXTENT_UPTODATE
, 1);
2810 EXPORT_SYMBOL(extent_buffer_uptodate
);
2812 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
2813 struct extent_buffer
*eb
,
2818 unsigned long start_i
;
2822 unsigned long num_pages
;
2824 if (eb
->flags
& EXTENT_UPTODATE
)
2827 if (0 && test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2828 EXTENT_UPTODATE
, 1)) {
2833 WARN_ON(start
< eb
->start
);
2834 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
2835 (eb
->start
>> PAGE_CACHE_SHIFT
);
2840 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2841 for (i
= start_i
; i
< num_pages
; i
++) {
2842 page
= extent_buffer_page(eb
, i
);
2843 if (PageUptodate(page
)) {
2847 if (TestSetPageLocked(page
)) {
2853 if (!PageUptodate(page
)) {
2854 err
= page
->mapping
->a_ops
->readpage(NULL
, page
);
2866 for (i
= start_i
; i
< num_pages
; i
++) {
2867 page
= extent_buffer_page(eb
, i
);
2868 wait_on_page_locked(page
);
2869 if (!PageUptodate(page
)) {
2874 eb
->flags
|= EXTENT_UPTODATE
;
2877 EXPORT_SYMBOL(read_extent_buffer_pages
);
2879 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
2880 unsigned long start
,
2887 char *dst
= (char *)dstv
;
2888 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
2889 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
2890 unsigned long num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2892 WARN_ON(start
> eb
->len
);
2893 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
2895 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
2898 page
= extent_buffer_page(eb
, i
);
2899 if (!PageUptodate(page
)) {
2900 printk("page %lu not up to date i %lu, total %lu, len %lu\n", page
->index
, i
, num_pages
, eb
->len
);
2903 WARN_ON(!PageUptodate(page
));
2905 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
2906 kaddr
= kmap_atomic(page
, KM_USER1
);
2907 memcpy(dst
, kaddr
+ offset
, cur
);
2908 kunmap_atomic(kaddr
, KM_USER1
);
2916 EXPORT_SYMBOL(read_extent_buffer
);
2918 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
2919 unsigned long min_len
, char **token
, char **map
,
2920 unsigned long *map_start
,
2921 unsigned long *map_len
, int km
)
2923 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
2926 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
2927 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
2928 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
2935 offset
= start_offset
;
2939 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
2941 if (start
+ min_len
> eb
->len
) {
2942 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb
->start
, eb
->len
, start
, min_len
);
2946 p
= extent_buffer_page(eb
, i
);
2947 WARN_ON(!PageUptodate(p
));
2948 kaddr
= kmap_atomic(p
, km
);
2950 *map
= kaddr
+ offset
;
2951 *map_len
= PAGE_CACHE_SIZE
- offset
;
2954 EXPORT_SYMBOL(map_private_extent_buffer
);
2956 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
2957 unsigned long min_len
,
2958 char **token
, char **map
,
2959 unsigned long *map_start
,
2960 unsigned long *map_len
, int km
)
2964 if (eb
->map_token
) {
2965 unmap_extent_buffer(eb
, eb
->map_token
, km
);
2966 eb
->map_token
= NULL
;
2969 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
2970 map_start
, map_len
, km
);
2972 eb
->map_token
= *token
;
2974 eb
->map_start
= *map_start
;
2975 eb
->map_len
= *map_len
;
2979 EXPORT_SYMBOL(map_extent_buffer
);
2981 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
2983 kunmap_atomic(token
, km
);
2985 EXPORT_SYMBOL(unmap_extent_buffer
);
2987 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
2988 unsigned long start
,
2995 char *ptr
= (char *)ptrv
;
2996 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
2997 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3000 WARN_ON(start
> eb
->len
);
3001 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3003 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3006 page
= extent_buffer_page(eb
, i
);
3007 WARN_ON(!PageUptodate(page
));
3009 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3011 kaddr
= kmap_atomic(page
, KM_USER0
);
3012 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3013 kunmap_atomic(kaddr
, KM_USER0
);
3024 EXPORT_SYMBOL(memcmp_extent_buffer
);
3026 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3027 unsigned long start
, unsigned long len
)
3033 char *src
= (char *)srcv
;
3034 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3035 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3037 WARN_ON(start
> eb
->len
);
3038 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3040 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3043 page
= extent_buffer_page(eb
, i
);
3044 WARN_ON(!PageUptodate(page
));
3046 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3047 kaddr
= kmap_atomic(page
, KM_USER1
);
3048 memcpy(kaddr
+ offset
, src
, cur
);
3049 kunmap_atomic(kaddr
, KM_USER1
);
3057 EXPORT_SYMBOL(write_extent_buffer
);
3059 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3060 unsigned long start
, unsigned long len
)
3066 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3067 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3069 WARN_ON(start
> eb
->len
);
3070 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3072 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3075 page
= extent_buffer_page(eb
, i
);
3076 WARN_ON(!PageUptodate(page
));
3078 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3079 kaddr
= kmap_atomic(page
, KM_USER0
);
3080 memset(kaddr
+ offset
, c
, cur
);
3081 kunmap_atomic(kaddr
, KM_USER0
);
3088 EXPORT_SYMBOL(memset_extent_buffer
);
3090 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3091 unsigned long dst_offset
, unsigned long src_offset
,
3094 u64 dst_len
= dst
->len
;
3099 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3100 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3102 WARN_ON(src
->len
!= dst_len
);
3104 offset
= (start_offset
+ dst_offset
) &
3105 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3108 page
= extent_buffer_page(dst
, i
);
3109 WARN_ON(!PageUptodate(page
));
3111 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3113 kaddr
= kmap_atomic(page
, KM_USER0
);
3114 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3115 kunmap_atomic(kaddr
, KM_USER0
);
3123 EXPORT_SYMBOL(copy_extent_buffer
);
3125 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3126 unsigned long dst_off
, unsigned long src_off
,
3129 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3130 if (dst_page
== src_page
) {
3131 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3133 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3134 char *p
= dst_kaddr
+ dst_off
+ len
;
3135 char *s
= src_kaddr
+ src_off
+ len
;
3140 kunmap_atomic(src_kaddr
, KM_USER1
);
3142 kunmap_atomic(dst_kaddr
, KM_USER0
);
3145 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3146 unsigned long dst_off
, unsigned long src_off
,
3149 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3152 if (dst_page
!= src_page
)
3153 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3155 src_kaddr
= dst_kaddr
;
3157 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3158 kunmap_atomic(dst_kaddr
, KM_USER0
);
3159 if (dst_page
!= src_page
)
3160 kunmap_atomic(src_kaddr
, KM_USER1
);
3163 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3164 unsigned long src_offset
, unsigned long len
)
3167 size_t dst_off_in_page
;
3168 size_t src_off_in_page
;
3169 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3170 unsigned long dst_i
;
3171 unsigned long src_i
;
3173 if (src_offset
+ len
> dst
->len
) {
3174 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3175 src_offset
, len
, dst
->len
);
3178 if (dst_offset
+ len
> dst
->len
) {
3179 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3180 dst_offset
, len
, dst
->len
);
3185 dst_off_in_page
= (start_offset
+ dst_offset
) &
3186 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3187 src_off_in_page
= (start_offset
+ src_offset
) &
3188 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3190 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3191 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3193 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3195 cur
= min_t(unsigned long, cur
,
3196 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3198 copy_pages(extent_buffer_page(dst
, dst_i
),
3199 extent_buffer_page(dst
, src_i
),
3200 dst_off_in_page
, src_off_in_page
, cur
);
3207 EXPORT_SYMBOL(memcpy_extent_buffer
);
3209 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3210 unsigned long src_offset
, unsigned long len
)
3213 size_t dst_off_in_page
;
3214 size_t src_off_in_page
;
3215 unsigned long dst_end
= dst_offset
+ len
- 1;
3216 unsigned long src_end
= src_offset
+ len
- 1;
3217 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3218 unsigned long dst_i
;
3219 unsigned long src_i
;
3221 if (src_offset
+ len
> dst
->len
) {
3222 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3223 src_offset
, len
, dst
->len
);
3226 if (dst_offset
+ len
> dst
->len
) {
3227 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3228 dst_offset
, len
, dst
->len
);
3231 if (dst_offset
< src_offset
) {
3232 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
3236 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
3237 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
3239 dst_off_in_page
= (start_offset
+ dst_end
) &
3240 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3241 src_off_in_page
= (start_offset
+ src_end
) &
3242 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3244 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
3245 cur
= min(cur
, dst_off_in_page
+ 1);
3246 move_pages(extent_buffer_page(dst
, dst_i
),
3247 extent_buffer_page(dst
, src_i
),
3248 dst_off_in_page
- cur
+ 1,
3249 src_off_in_page
- cur
+ 1, cur
);
3256 EXPORT_SYMBOL(memmove_extent_buffer
);