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
);
95 EXPORT_SYMBOL(extent_io_tree_init
);
97 void extent_io_tree_empty_lru(struct extent_io_tree
*tree
)
99 struct extent_buffer
*eb
;
100 while(!list_empty(&tree
->buffer_lru
)) {
101 eb
= list_entry(tree
->buffer_lru
.next
, struct extent_buffer
,
103 list_del_init(&eb
->lru
);
104 free_extent_buffer(eb
);
107 EXPORT_SYMBOL(extent_io_tree_empty_lru
);
109 struct extent_state
*alloc_extent_state(gfp_t mask
)
111 struct extent_state
*state
;
113 state
= kmem_cache_alloc(extent_state_cache
, mask
);
114 if (!state
|| IS_ERR(state
))
120 atomic_set(&state
->refs
, 1);
121 init_waitqueue_head(&state
->wq
);
124 EXPORT_SYMBOL(alloc_extent_state
);
126 void free_extent_state(struct extent_state
*state
)
130 if (atomic_dec_and_test(&state
->refs
)) {
131 WARN_ON(state
->tree
);
132 kmem_cache_free(extent_state_cache
, state
);
135 EXPORT_SYMBOL(free_extent_state
);
137 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
138 struct rb_node
*node
)
140 struct rb_node
** p
= &root
->rb_node
;
141 struct rb_node
* parent
= NULL
;
142 struct tree_entry
*entry
;
146 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
148 if (offset
< entry
->start
)
150 else if (offset
> entry
->end
)
156 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
157 rb_link_node(node
, parent
, p
);
158 rb_insert_color(node
, root
);
162 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
163 struct rb_node
**prev_ret
,
164 struct rb_node
**next_ret
)
166 struct rb_root
*root
= &tree
->state
;
167 struct rb_node
* n
= root
->rb_node
;
168 struct rb_node
*prev
= NULL
;
169 struct rb_node
*orig_prev
= NULL
;
170 struct tree_entry
*entry
;
171 struct tree_entry
*prev_entry
= NULL
;
174 struct extent_state
*state
;
176 if (state
->start
<= offset
&& offset
<= state
->end
)
177 return &tree
->last
->rb_node
;
180 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
184 if (offset
< entry
->start
)
186 else if (offset
> entry
->end
)
189 tree
->last
= rb_entry(n
, struct extent_state
, rb_node
);
196 while(prev
&& offset
> prev_entry
->end
) {
197 prev
= rb_next(prev
);
198 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
205 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
206 while(prev
&& offset
< prev_entry
->start
) {
207 prev
= rb_prev(prev
);
208 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
215 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
218 struct rb_node
*prev
= NULL
;
221 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
224 tree
->last
= rb_entry(prev
, struct extent_state
,
233 * utility function to look for merge candidates inside a given range.
234 * Any extents with matching state are merged together into a single
235 * extent in the tree. Extents with EXTENT_IO in their state field
236 * are not merged because the end_io handlers need to be able to do
237 * operations on them without sleeping (or doing allocations/splits).
239 * This should be called with the tree lock held.
241 static int merge_state(struct extent_io_tree
*tree
,
242 struct extent_state
*state
)
244 struct extent_state
*other
;
245 struct rb_node
*other_node
;
247 if (state
->state
& EXTENT_IOBITS
)
250 other_node
= rb_prev(&state
->rb_node
);
252 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
253 if (other
->end
== state
->start
- 1 &&
254 other
->state
== state
->state
) {
255 state
->start
= other
->start
;
257 if (tree
->last
== other
)
259 rb_erase(&other
->rb_node
, &tree
->state
);
260 free_extent_state(other
);
263 other_node
= rb_next(&state
->rb_node
);
265 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
266 if (other
->start
== state
->end
+ 1 &&
267 other
->state
== state
->state
) {
268 other
->start
= state
->start
;
270 if (tree
->last
== state
)
272 rb_erase(&state
->rb_node
, &tree
->state
);
273 free_extent_state(state
);
279 static void set_state_cb(struct extent_io_tree
*tree
,
280 struct extent_state
*state
,
283 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
284 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
->start
,
285 state
->end
, state
->state
, bits
);
289 static void clear_state_cb(struct extent_io_tree
*tree
,
290 struct extent_state
*state
,
293 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
294 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
->start
,
295 state
->end
, state
->state
, bits
);
300 * insert an extent_state struct into the tree. 'bits' are set on the
301 * struct before it is inserted.
303 * This may return -EEXIST if the extent is already there, in which case the
304 * state struct is freed.
306 * The tree lock is not taken internally. This is a utility function and
307 * probably isn't what you want to call (see set/clear_extent_bit).
309 static int insert_state(struct extent_io_tree
*tree
,
310 struct extent_state
*state
, u64 start
, u64 end
,
313 struct rb_node
*node
;
316 printk("end < start %Lu %Lu\n", end
, start
);
319 if (bits
& EXTENT_DIRTY
)
320 tree
->dirty_bytes
+= end
- start
+ 1;
321 set_state_cb(tree
, state
, bits
);
322 state
->state
|= bits
;
323 state
->start
= start
;
325 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
327 struct extent_state
*found
;
328 found
= rb_entry(node
, struct extent_state
, rb_node
);
329 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, start
, end
);
330 free_extent_state(state
);
335 merge_state(tree
, state
);
340 * split a given extent state struct in two, inserting the preallocated
341 * struct 'prealloc' as the newly created second half. 'split' indicates an
342 * offset inside 'orig' where it should be split.
345 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
346 * are two extent state structs in the tree:
347 * prealloc: [orig->start, split - 1]
348 * orig: [ split, orig->end ]
350 * The tree locks are not taken by this function. They need to be held
353 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
354 struct extent_state
*prealloc
, u64 split
)
356 struct rb_node
*node
;
357 prealloc
->start
= orig
->start
;
358 prealloc
->end
= split
- 1;
359 prealloc
->state
= orig
->state
;
362 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
364 struct extent_state
*found
;
365 found
= rb_entry(node
, struct extent_state
, rb_node
);
366 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, prealloc
->start
, prealloc
->end
);
367 free_extent_state(prealloc
);
370 prealloc
->tree
= tree
;
375 * utility function to clear some bits in an extent state struct.
376 * it will optionally wake up any one waiting on this state (wake == 1), or
377 * forcibly remove the state from the tree (delete == 1).
379 * If no bits are set on the state struct after clearing things, the
380 * struct is freed and removed from the tree
382 static int clear_state_bit(struct extent_io_tree
*tree
,
383 struct extent_state
*state
, int bits
, int wake
,
386 int ret
= state
->state
& bits
;
388 if ((bits
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
389 u64 range
= state
->end
- state
->start
+ 1;
390 WARN_ON(range
> tree
->dirty_bytes
);
391 tree
->dirty_bytes
-= range
;
393 clear_state_cb(tree
, state
, bits
);
394 state
->state
&= ~bits
;
397 if (delete || state
->state
== 0) {
399 if (tree
->last
== state
)
401 rb_erase(&state
->rb_node
, &tree
->state
);
403 free_extent_state(state
);
408 merge_state(tree
, state
);
414 * clear some bits on a range in the tree. This may require splitting
415 * or inserting elements in the tree, so the gfp mask is used to
416 * indicate which allocations or sleeping are allowed.
418 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
419 * the given range from the tree regardless of state (ie for truncate).
421 * the range [start, end] is inclusive.
423 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
424 * bits were already set, or zero if none of the bits were already set.
426 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
427 int bits
, int wake
, int delete, gfp_t mask
)
429 struct extent_state
*state
;
430 struct extent_state
*prealloc
= NULL
;
431 struct rb_node
*node
;
437 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
438 prealloc
= alloc_extent_state(mask
);
443 spin_lock_irqsave(&tree
->lock
, flags
);
445 * this search will find the extents that end after
448 node
= tree_search(tree
, start
);
451 state
= rb_entry(node
, struct extent_state
, rb_node
);
452 if (state
->start
> end
)
454 WARN_ON(state
->end
< start
);
457 * | ---- desired range ---- |
459 * | ------------- state -------------- |
461 * We need to split the extent we found, and may flip
462 * bits on second half.
464 * If the extent we found extends past our range, we
465 * just split and search again. It'll get split again
466 * the next time though.
468 * If the extent we found is inside our range, we clear
469 * the desired bit on it.
472 if (state
->start
< start
) {
474 prealloc
= alloc_extent_state(GFP_ATOMIC
);
475 err
= split_state(tree
, state
, prealloc
, start
);
476 BUG_ON(err
== -EEXIST
);
480 if (state
->end
<= end
) {
481 start
= state
->end
+ 1;
482 set
|= clear_state_bit(tree
, state
, bits
,
485 start
= state
->start
;
490 * | ---- desired range ---- |
492 * We need to split the extent, and clear the bit
495 if (state
->start
<= end
&& state
->end
> end
) {
497 prealloc
= alloc_extent_state(GFP_ATOMIC
);
498 err
= split_state(tree
, state
, prealloc
, end
+ 1);
499 BUG_ON(err
== -EEXIST
);
503 set
|= clear_state_bit(tree
, prealloc
, bits
,
509 start
= state
->end
+ 1;
510 set
|= clear_state_bit(tree
, state
, bits
, wake
, delete);
514 spin_unlock_irqrestore(&tree
->lock
, flags
);
516 free_extent_state(prealloc
);
523 spin_unlock_irqrestore(&tree
->lock
, flags
);
524 if (mask
& __GFP_WAIT
)
528 EXPORT_SYMBOL(clear_extent_bit
);
530 static int wait_on_state(struct extent_io_tree
*tree
,
531 struct extent_state
*state
)
534 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
535 spin_unlock_irq(&tree
->lock
);
537 spin_lock_irq(&tree
->lock
);
538 finish_wait(&state
->wq
, &wait
);
543 * waits for one or more bits to clear on a range in the state tree.
544 * The range [start, end] is inclusive.
545 * The tree lock is taken by this function
547 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
549 struct extent_state
*state
;
550 struct rb_node
*node
;
552 spin_lock_irq(&tree
->lock
);
556 * this search will find all the extents that end after
559 node
= tree_search(tree
, start
);
563 state
= rb_entry(node
, struct extent_state
, rb_node
);
565 if (state
->start
> end
)
568 if (state
->state
& bits
) {
569 start
= state
->start
;
570 atomic_inc(&state
->refs
);
571 wait_on_state(tree
, state
);
572 free_extent_state(state
);
575 start
= state
->end
+ 1;
580 if (need_resched()) {
581 spin_unlock_irq(&tree
->lock
);
583 spin_lock_irq(&tree
->lock
);
587 spin_unlock_irq(&tree
->lock
);
590 EXPORT_SYMBOL(wait_extent_bit
);
592 static void set_state_bits(struct extent_io_tree
*tree
,
593 struct extent_state
*state
,
596 if ((bits
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
597 u64 range
= state
->end
- state
->start
+ 1;
598 tree
->dirty_bytes
+= range
;
600 set_state_cb(tree
, state
, bits
);
601 state
->state
|= bits
;
605 * set some bits on a range in the tree. This may require allocations
606 * or sleeping, so the gfp mask is used to indicate what is allowed.
608 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
609 * range already has the desired bits set. The start of the existing
610 * range is returned in failed_start in this case.
612 * [start, end] is inclusive
613 * This takes the tree lock.
615 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
,
616 int exclusive
, u64
*failed_start
, gfp_t mask
)
618 struct extent_state
*state
;
619 struct extent_state
*prealloc
= NULL
;
620 struct rb_node
*node
;
627 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
628 prealloc
= alloc_extent_state(mask
);
633 spin_lock_irqsave(&tree
->lock
, flags
);
635 * this search will find all the extents that end after
638 node
= tree_search(tree
, start
);
640 err
= insert_state(tree
, prealloc
, start
, end
, bits
);
642 BUG_ON(err
== -EEXIST
);
646 state
= rb_entry(node
, struct extent_state
, rb_node
);
647 last_start
= state
->start
;
648 last_end
= state
->end
;
651 * | ---- desired range ---- |
654 * Just lock what we found and keep going
656 if (state
->start
== start
&& state
->end
<= end
) {
657 set
= state
->state
& bits
;
658 if (set
&& exclusive
) {
659 *failed_start
= state
->start
;
663 set_state_bits(tree
, state
, bits
);
664 start
= state
->end
+ 1;
665 merge_state(tree
, state
);
670 * | ---- desired range ---- |
673 * | ------------- state -------------- |
675 * We need to split the extent we found, and may flip bits on
678 * If the extent we found extends past our
679 * range, we just split and search again. It'll get split
680 * again the next time though.
682 * If the extent we found is inside our range, we set the
685 if (state
->start
< start
) {
686 set
= state
->state
& bits
;
687 if (exclusive
&& set
) {
688 *failed_start
= start
;
692 err
= split_state(tree
, state
, prealloc
, start
);
693 BUG_ON(err
== -EEXIST
);
697 if (state
->end
<= end
) {
698 set_state_bits(tree
, state
, bits
);
699 start
= state
->end
+ 1;
700 merge_state(tree
, state
);
702 start
= state
->start
;
707 * | ---- desired range ---- |
708 * | state | or | state |
710 * There's a hole, we need to insert something in it and
711 * ignore the extent we found.
713 if (state
->start
> start
) {
715 if (end
< last_start
)
718 this_end
= last_start
-1;
719 err
= insert_state(tree
, prealloc
, start
, this_end
,
722 BUG_ON(err
== -EEXIST
);
725 start
= this_end
+ 1;
729 * | ---- desired range ---- |
731 * We need to split the extent, and set the bit
734 if (state
->start
<= end
&& state
->end
> end
) {
735 set
= state
->state
& bits
;
736 if (exclusive
&& set
) {
737 *failed_start
= start
;
741 err
= split_state(tree
, state
, prealloc
, end
+ 1);
742 BUG_ON(err
== -EEXIST
);
744 set_state_bits(tree
, prealloc
, bits
);
745 merge_state(tree
, prealloc
);
753 spin_unlock_irqrestore(&tree
->lock
, flags
);
755 free_extent_state(prealloc
);
762 spin_unlock_irqrestore(&tree
->lock
, flags
);
763 if (mask
& __GFP_WAIT
)
767 EXPORT_SYMBOL(set_extent_bit
);
769 /* wrappers around set/clear extent bit */
770 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
773 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
776 EXPORT_SYMBOL(set_extent_dirty
);
778 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
779 int bits
, gfp_t mask
)
781 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
784 EXPORT_SYMBOL(set_extent_bits
);
786 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
787 int bits
, gfp_t mask
)
789 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, mask
);
791 EXPORT_SYMBOL(clear_extent_bits
);
793 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
796 return set_extent_bit(tree
, start
, end
,
797 EXTENT_DELALLOC
| EXTENT_DIRTY
, 0, NULL
,
800 EXPORT_SYMBOL(set_extent_delalloc
);
802 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
805 return clear_extent_bit(tree
, start
, end
,
806 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, mask
);
808 EXPORT_SYMBOL(clear_extent_dirty
);
810 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
813 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
816 EXPORT_SYMBOL(set_extent_new
);
818 int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
821 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0, mask
);
823 EXPORT_SYMBOL(clear_extent_new
);
825 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
828 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, NULL
,
831 EXPORT_SYMBOL(set_extent_uptodate
);
833 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
836 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0, mask
);
838 EXPORT_SYMBOL(clear_extent_uptodate
);
840 int set_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
843 return set_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
,
846 EXPORT_SYMBOL(set_extent_writeback
);
848 int clear_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
851 return clear_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 1, 0, mask
);
853 EXPORT_SYMBOL(clear_extent_writeback
);
855 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
857 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
859 EXPORT_SYMBOL(wait_on_extent_writeback
);
861 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
866 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
867 &failed_start
, mask
);
868 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
869 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
870 start
= failed_start
;
874 WARN_ON(start
> end
);
878 EXPORT_SYMBOL(lock_extent
);
880 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
883 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, mask
);
885 EXPORT_SYMBOL(unlock_extent
);
888 * helper function to set pages and extents in the tree dirty
890 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
892 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
893 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
896 while (index
<= end_index
) {
897 page
= find_get_page(tree
->mapping
, index
);
899 __set_page_dirty_nobuffers(page
);
900 page_cache_release(page
);
903 set_extent_dirty(tree
, start
, end
, GFP_NOFS
);
906 EXPORT_SYMBOL(set_range_dirty
);
909 * helper function to set both pages and extents in the tree writeback
911 int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
913 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
914 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
917 while (index
<= end_index
) {
918 page
= find_get_page(tree
->mapping
, index
);
920 set_page_writeback(page
);
921 page_cache_release(page
);
924 set_extent_writeback(tree
, start
, end
, GFP_NOFS
);
927 EXPORT_SYMBOL(set_range_writeback
);
929 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
930 u64
*start_ret
, u64
*end_ret
, int bits
)
932 struct rb_node
*node
;
933 struct extent_state
*state
;
936 spin_lock_irq(&tree
->lock
);
938 * this search will find all the extents that end after
941 node
= tree_search(tree
, start
);
942 if (!node
|| IS_ERR(node
)) {
947 state
= rb_entry(node
, struct extent_state
, rb_node
);
948 if (state
->end
>= start
&& (state
->state
& bits
)) {
949 *start_ret
= state
->start
;
950 *end_ret
= state
->end
;
954 node
= rb_next(node
);
959 spin_unlock_irq(&tree
->lock
);
962 EXPORT_SYMBOL(find_first_extent_bit
);
964 u64
find_lock_delalloc_range(struct extent_io_tree
*tree
,
965 u64
*start
, u64
*end
, u64 max_bytes
)
967 struct rb_node
*node
;
968 struct extent_state
*state
;
969 u64 cur_start
= *start
;
973 spin_lock_irq(&tree
->lock
);
975 * this search will find all the extents that end after
979 node
= tree_search(tree
, cur_start
);
980 if (!node
|| IS_ERR(node
)) {
986 state
= rb_entry(node
, struct extent_state
, rb_node
);
987 if (found
&& state
->start
!= cur_start
) {
990 if (!(state
->state
& EXTENT_DELALLOC
)) {
996 struct extent_state
*prev_state
;
997 struct rb_node
*prev_node
= node
;
999 prev_node
= rb_prev(prev_node
);
1002 prev_state
= rb_entry(prev_node
,
1003 struct extent_state
,
1005 if (!(prev_state
->state
& EXTENT_DELALLOC
))
1011 if (state
->state
& EXTENT_LOCKED
) {
1013 atomic_inc(&state
->refs
);
1014 prepare_to_wait(&state
->wq
, &wait
,
1015 TASK_UNINTERRUPTIBLE
);
1016 spin_unlock_irq(&tree
->lock
);
1018 spin_lock_irq(&tree
->lock
);
1019 finish_wait(&state
->wq
, &wait
);
1020 free_extent_state(state
);
1023 set_state_cb(tree
, state
, EXTENT_LOCKED
);
1024 state
->state
|= EXTENT_LOCKED
;
1026 *start
= state
->start
;
1029 cur_start
= state
->end
+ 1;
1030 node
= rb_next(node
);
1033 total_bytes
+= state
->end
- state
->start
+ 1;
1034 if (total_bytes
>= max_bytes
)
1038 spin_unlock_irq(&tree
->lock
);
1042 u64
count_range_bits(struct extent_io_tree
*tree
,
1043 u64
*start
, u64 search_end
, u64 max_bytes
,
1046 struct rb_node
*node
;
1047 struct extent_state
*state
;
1048 u64 cur_start
= *start
;
1049 u64 total_bytes
= 0;
1052 if (search_end
<= cur_start
) {
1053 printk("search_end %Lu start %Lu\n", search_end
, cur_start
);
1058 spin_lock_irq(&tree
->lock
);
1059 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1060 total_bytes
= tree
->dirty_bytes
;
1064 * this search will find all the extents that end after
1067 node
= tree_search(tree
, cur_start
);
1068 if (!node
|| IS_ERR(node
)) {
1073 state
= rb_entry(node
, struct extent_state
, rb_node
);
1074 if (state
->start
> search_end
)
1076 if (state
->end
>= cur_start
&& (state
->state
& bits
)) {
1077 total_bytes
+= min(search_end
, state
->end
) + 1 -
1078 max(cur_start
, state
->start
);
1079 if (total_bytes
>= max_bytes
)
1082 *start
= state
->start
;
1086 node
= rb_next(node
);
1091 spin_unlock_irq(&tree
->lock
);
1095 * helper function to lock both pages and extents in the tree.
1096 * pages must be locked first.
1098 int lock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1100 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1101 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1105 while (index
<= end_index
) {
1106 page
= grab_cache_page(tree
->mapping
, index
);
1112 err
= PTR_ERR(page
);
1117 lock_extent(tree
, start
, end
, GFP_NOFS
);
1122 * we failed above in getting the page at 'index', so we undo here
1123 * up to but not including the page at 'index'
1126 index
= start
>> PAGE_CACHE_SHIFT
;
1127 while (index
< end_index
) {
1128 page
= find_get_page(tree
->mapping
, index
);
1130 page_cache_release(page
);
1135 EXPORT_SYMBOL(lock_range
);
1138 * helper function to unlock both pages and extents in the tree.
1140 int unlock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1142 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1143 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1146 while (index
<= end_index
) {
1147 page
= find_get_page(tree
->mapping
, index
);
1149 page_cache_release(page
);
1152 unlock_extent(tree
, start
, end
, GFP_NOFS
);
1155 EXPORT_SYMBOL(unlock_range
);
1157 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1159 struct rb_node
*node
;
1160 struct extent_state
*state
;
1163 spin_lock_irq(&tree
->lock
);
1165 * this search will find all the extents that end after
1168 node
= tree_search(tree
, start
);
1169 if (!node
|| IS_ERR(node
)) {
1173 state
= rb_entry(node
, struct extent_state
, rb_node
);
1174 if (state
->start
!= start
) {
1178 state
->private = private;
1180 spin_unlock_irq(&tree
->lock
);
1184 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1186 struct rb_node
*node
;
1187 struct extent_state
*state
;
1190 spin_lock_irq(&tree
->lock
);
1192 * this search will find all the extents that end after
1195 node
= tree_search(tree
, start
);
1196 if (!node
|| IS_ERR(node
)) {
1200 state
= rb_entry(node
, struct extent_state
, rb_node
);
1201 if (state
->start
!= start
) {
1205 *private = state
->private;
1207 spin_unlock_irq(&tree
->lock
);
1212 * searches a range in the state tree for a given mask.
1213 * If 'filled' == 1, this returns 1 only if every extent in the tree
1214 * has the bits set. Otherwise, 1 is returned if any bit in the
1215 * range is found set.
1217 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1218 int bits
, int filled
)
1220 struct extent_state
*state
= NULL
;
1221 struct rb_node
*node
;
1223 unsigned long flags
;
1225 spin_lock_irqsave(&tree
->lock
, flags
);
1226 node
= tree_search(tree
, start
);
1227 while (node
&& start
<= end
) {
1228 state
= rb_entry(node
, struct extent_state
, rb_node
);
1230 if (filled
&& state
->start
> start
) {
1235 if (state
->start
> end
)
1238 if (state
->state
& bits
) {
1242 } else if (filled
) {
1246 start
= state
->end
+ 1;
1249 node
= rb_next(node
);
1256 spin_unlock_irqrestore(&tree
->lock
, flags
);
1259 EXPORT_SYMBOL(test_range_bit
);
1262 * helper function to set a given page up to date if all the
1263 * extents in the tree for that page are up to date
1265 static int check_page_uptodate(struct extent_io_tree
*tree
,
1268 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1269 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1270 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1))
1271 SetPageUptodate(page
);
1276 * helper function to unlock a page if all the extents in the tree
1277 * for that page are unlocked
1279 static int check_page_locked(struct extent_io_tree
*tree
,
1282 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1283 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1284 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0))
1290 * helper function to end page writeback if all the extents
1291 * in the tree for that page are done with writeback
1293 static int check_page_writeback(struct extent_io_tree
*tree
,
1296 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1297 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1298 if (!test_range_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 0))
1299 end_page_writeback(page
);
1303 /* lots and lots of room for performance fixes in the end_bio funcs */
1306 * after a writepage IO is done, we need to:
1307 * clear the uptodate bits on error
1308 * clear the writeback bits in the extent tree for this IO
1309 * end_page_writeback if the page has no more pending IO
1311 * Scheduling is not allowed, so the extent state tree is expected
1312 * to have one and only one object corresponding to this IO.
1314 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1315 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1317 static int end_bio_extent_writepage(struct bio
*bio
,
1318 unsigned int bytes_done
, int err
)
1321 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1322 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1323 struct extent_state
*state
= bio
->bi_private
;
1324 struct extent_io_tree
*tree
= state
->tree
;
1325 struct rb_node
*node
;
1330 unsigned long flags
;
1332 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1337 struct page
*page
= bvec
->bv_page
;
1338 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1340 end
= start
+ bvec
->bv_len
- 1;
1342 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1347 if (--bvec
>= bio
->bi_io_vec
)
1348 prefetchw(&bvec
->bv_page
->flags
);
1351 clear_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1352 ClearPageUptodate(page
);
1356 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1357 tree
->ops
->writepage_end_io_hook(page
, start
, end
,
1362 * bios can get merged in funny ways, and so we need to
1363 * be careful with the state variable. We know the
1364 * state won't be merged with others because it has
1365 * WRITEBACK set, but we can't be sure each biovec is
1366 * sequential in the file. So, if our cached state
1367 * doesn't match the expected end, search the tree
1368 * for the correct one.
1371 spin_lock_irqsave(&tree
->lock
, flags
);
1372 if (!state
|| state
->end
!= end
) {
1374 node
= __etree_search(tree
, start
, NULL
, NULL
);
1376 state
= rb_entry(node
, struct extent_state
,
1378 if (state
->end
!= end
||
1379 !(state
->state
& EXTENT_WRITEBACK
))
1383 spin_unlock_irqrestore(&tree
->lock
, flags
);
1384 clear_extent_writeback(tree
, start
,
1391 struct extent_state
*clear
= state
;
1393 node
= rb_prev(&state
->rb_node
);
1395 state
= rb_entry(node
,
1396 struct extent_state
,
1402 clear_state_bit(tree
, clear
, EXTENT_WRITEBACK
,
1413 /* before releasing the lock, make sure the next state
1414 * variable has the expected bits set and corresponds
1415 * to the correct offsets in the file
1417 if (state
&& (state
->end
+ 1 != start
||
1418 !state
->state
& EXTENT_WRITEBACK
)) {
1421 spin_unlock_irqrestore(&tree
->lock
, flags
);
1425 end_page_writeback(page
);
1427 check_page_writeback(tree
, page
);
1428 } while (bvec
>= bio
->bi_io_vec
);
1430 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1436 * after a readpage IO is done, we need to:
1437 * clear the uptodate bits on error
1438 * set the uptodate bits if things worked
1439 * set the page up to date if all extents in the tree are uptodate
1440 * clear the lock bit in the extent tree
1441 * unlock the page if there are no other extents locked for it
1443 * Scheduling is not allowed, so the extent state tree is expected
1444 * to have one and only one object corresponding to this IO.
1446 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1447 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1449 static int end_bio_extent_readpage(struct bio
*bio
,
1450 unsigned int bytes_done
, int err
)
1453 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1454 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1455 struct extent_state
*state
= bio
->bi_private
;
1456 struct extent_io_tree
*tree
= state
->tree
;
1457 struct rb_node
*node
;
1461 unsigned long flags
;
1465 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1471 struct page
*page
= bvec
->bv_page
;
1472 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1474 end
= start
+ bvec
->bv_len
- 1;
1476 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1481 if (--bvec
>= bio
->bi_io_vec
)
1482 prefetchw(&bvec
->bv_page
->flags
);
1484 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1485 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1491 spin_lock_irqsave(&tree
->lock
, flags
);
1492 if (!state
|| state
->end
!= end
) {
1494 node
= __etree_search(tree
, start
, NULL
, NULL
);
1496 state
= rb_entry(node
, struct extent_state
,
1498 if (state
->end
!= end
||
1499 !(state
->state
& EXTENT_LOCKED
))
1503 spin_unlock_irqrestore(&tree
->lock
, flags
);
1504 set_extent_uptodate(tree
, start
, end
,
1506 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1513 struct extent_state
*clear
= state
;
1515 node
= rb_prev(&state
->rb_node
);
1517 state
= rb_entry(node
,
1518 struct extent_state
,
1523 set_state_cb(tree
, clear
, EXTENT_UPTODATE
);
1524 clear
->state
|= EXTENT_UPTODATE
;
1525 clear_state_bit(tree
, clear
, EXTENT_LOCKED
,
1536 /* before releasing the lock, make sure the next state
1537 * variable has the expected bits set and corresponds
1538 * to the correct offsets in the file
1540 if (state
&& (state
->end
+ 1 != start
||
1541 !state
->state
& EXTENT_WRITEBACK
)) {
1544 spin_unlock_irqrestore(&tree
->lock
, flags
);
1548 SetPageUptodate(page
);
1550 ClearPageUptodate(page
);
1556 check_page_uptodate(tree
, page
);
1558 ClearPageUptodate(page
);
1561 check_page_locked(tree
, page
);
1563 } while (bvec
>= bio
->bi_io_vec
);
1566 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1572 * IO done from prepare_write is pretty simple, we just unlock
1573 * the structs in the extent tree when done, and set the uptodate bits
1576 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1577 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1579 static int end_bio_extent_preparewrite(struct bio
*bio
,
1580 unsigned int bytes_done
, int err
)
1583 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1584 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1585 struct extent_state
*state
= bio
->bi_private
;
1586 struct extent_io_tree
*tree
= state
->tree
;
1590 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1596 struct page
*page
= bvec
->bv_page
;
1597 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1599 end
= start
+ bvec
->bv_len
- 1;
1601 if (--bvec
>= bio
->bi_io_vec
)
1602 prefetchw(&bvec
->bv_page
->flags
);
1605 set_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1607 ClearPageUptodate(page
);
1611 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1613 } while (bvec
>= bio
->bi_io_vec
);
1616 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1622 extent_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1627 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1629 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1630 while (!bio
&& (nr_vecs
/= 2))
1631 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1635 bio
->bi_bdev
= bdev
;
1636 bio
->bi_sector
= first_sector
;
1641 static int submit_one_bio(int rw
, struct bio
*bio
)
1645 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1646 struct page
*page
= bvec
->bv_page
;
1647 struct extent_io_tree
*tree
= bio
->bi_private
;
1648 struct rb_node
*node
;
1649 struct extent_state
*state
;
1653 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1654 end
= start
+ bvec
->bv_len
- 1;
1656 spin_lock_irq(&tree
->lock
);
1657 node
= __etree_search(tree
, start
, NULL
, NULL
);
1659 state
= rb_entry(node
, struct extent_state
, rb_node
);
1660 while(state
->end
< end
) {
1661 node
= rb_next(node
);
1662 state
= rb_entry(node
, struct extent_state
, rb_node
);
1664 BUG_ON(state
->end
!= end
);
1665 spin_unlock_irq(&tree
->lock
);
1667 bio
->bi_private
= state
;
1671 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1672 if (maxsector
< bio
->bi_sector
) {
1673 printk("sector too large max %Lu got %llu\n", maxsector
,
1674 (unsigned long long)bio
->bi_sector
);
1678 submit_bio(rw
, bio
);
1679 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1685 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1686 struct page
*page
, sector_t sector
,
1687 size_t size
, unsigned long offset
,
1688 struct block_device
*bdev
,
1689 struct bio
**bio_ret
,
1690 unsigned long max_pages
,
1691 bio_end_io_t end_io_func
)
1697 if (bio_ret
&& *bio_ret
) {
1699 if (bio
->bi_sector
+ (bio
->bi_size
>> 9) != sector
||
1700 bio_add_page(bio
, page
, size
, offset
) < size
) {
1701 ret
= submit_one_bio(rw
, bio
);
1707 nr
= min_t(int, max_pages
, bio_get_nr_vecs(bdev
));
1708 bio
= extent_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1710 printk("failed to allocate bio nr %d\n", nr
);
1714 bio_add_page(bio
, page
, size
, offset
);
1715 bio
->bi_end_io
= end_io_func
;
1716 bio
->bi_private
= tree
;
1721 ret
= submit_one_bio(rw
, bio
);
1727 void set_page_extent_mapped(struct page
*page
)
1729 if (!PagePrivate(page
)) {
1730 SetPagePrivate(page
);
1731 WARN_ON(!page
->mapping
->a_ops
->invalidatepage
);
1732 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1733 page_cache_get(page
);
1737 void set_page_extent_head(struct page
*page
, unsigned long len
)
1739 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1743 * basic readpage implementation. Locked extent state structs are inserted
1744 * into the tree that are removed when the IO is done (by the end_io
1747 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1749 get_extent_t
*get_extent
,
1752 struct inode
*inode
= page
->mapping
->host
;
1753 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1754 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1758 u64 last_byte
= i_size_read(inode
);
1762 struct extent_map
*em
;
1763 struct block_device
*bdev
;
1766 size_t page_offset
= 0;
1768 size_t blocksize
= inode
->i_sb
->s_blocksize
;
1770 set_page_extent_mapped(page
);
1773 lock_extent(tree
, start
, end
, GFP_NOFS
);
1775 while (cur
<= end
) {
1776 if (cur
>= last_byte
) {
1778 iosize
= PAGE_CACHE_SIZE
- page_offset
;
1779 userpage
= kmap_atomic(page
, KM_USER0
);
1780 memset(userpage
+ page_offset
, 0, iosize
);
1781 flush_dcache_page(page
);
1782 kunmap_atomic(userpage
, KM_USER0
);
1783 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1785 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1788 em
= get_extent(inode
, page
, page_offset
, cur
,
1790 if (IS_ERR(em
) || !em
) {
1792 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
1796 extent_offset
= cur
- em
->start
;
1797 BUG_ON(extent_map_end(em
) <= cur
);
1800 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
1801 cur_end
= min(extent_map_end(em
) - 1, end
);
1802 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
1803 sector
= (em
->block_start
+ extent_offset
) >> 9;
1805 block_start
= em
->block_start
;
1806 free_extent_map(em
);
1809 /* we've found a hole, just zero and go on */
1810 if (block_start
== EXTENT_MAP_HOLE
) {
1812 userpage
= kmap_atomic(page
, KM_USER0
);
1813 memset(userpage
+ page_offset
, 0, iosize
);
1814 flush_dcache_page(page
);
1815 kunmap_atomic(userpage
, KM_USER0
);
1817 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1819 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1821 page_offset
+= iosize
;
1824 /* the get_extent function already copied into the page */
1825 if (test_range_bit(tree
, cur
, cur_end
, EXTENT_UPTODATE
, 1)) {
1826 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1828 page_offset
+= iosize
;
1831 /* we have an inline extent but it didn't get marked up
1832 * to date. Error out
1834 if (block_start
== EXTENT_MAP_INLINE
) {
1836 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1838 page_offset
+= iosize
;
1843 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
1844 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
1848 unsigned long nr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
1850 ret
= submit_extent_page(READ
, tree
, page
,
1851 sector
, iosize
, page_offset
,
1853 end_bio_extent_readpage
);
1858 page_offset
+= iosize
;
1862 if (!PageError(page
))
1863 SetPageUptodate(page
);
1869 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
1870 get_extent_t
*get_extent
)
1872 struct bio
*bio
= NULL
;
1875 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
);
1877 submit_one_bio(READ
, bio
);
1880 EXPORT_SYMBOL(extent_read_full_page
);
1883 * the writepage semantics are similar to regular writepage. extent
1884 * records are inserted to lock ranges in the tree, and as dirty areas
1885 * are found, they are marked writeback. Then the lock bits are removed
1886 * and the end_io handler clears the writeback ranges
1888 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
1891 struct inode
*inode
= page
->mapping
->host
;
1892 struct extent_page_data
*epd
= data
;
1893 struct extent_io_tree
*tree
= epd
->tree
;
1894 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1896 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1900 u64 last_byte
= i_size_read(inode
);
1904 struct extent_map
*em
;
1905 struct block_device
*bdev
;
1908 size_t page_offset
= 0;
1910 loff_t i_size
= i_size_read(inode
);
1911 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1915 WARN_ON(!PageLocked(page
));
1916 if (page
->index
> end_index
) {
1917 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
1922 if (page
->index
== end_index
) {
1925 size_t offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
1927 userpage
= kmap_atomic(page
, KM_USER0
);
1928 memset(userpage
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
1929 flush_dcache_page(page
);
1930 kunmap_atomic(userpage
, KM_USER0
);
1933 set_page_extent_mapped(page
);
1935 delalloc_start
= start
;
1937 while(delalloc_end
< page_end
) {
1938 nr_delalloc
= find_lock_delalloc_range(tree
, &delalloc_start
,
1941 if (nr_delalloc
== 0) {
1942 delalloc_start
= delalloc_end
+ 1;
1945 tree
->ops
->fill_delalloc(inode
, delalloc_start
,
1947 clear_extent_bit(tree
, delalloc_start
,
1949 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1951 delalloc_start
= delalloc_end
+ 1;
1953 lock_extent(tree
, start
, page_end
, GFP_NOFS
);
1956 if (test_range_bit(tree
, start
, page_end
, EXTENT_DELALLOC
, 0)) {
1957 printk("found delalloc bits after lock_extent\n");
1960 if (last_byte
<= start
) {
1961 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
1965 set_extent_uptodate(tree
, start
, page_end
, GFP_NOFS
);
1966 blocksize
= inode
->i_sb
->s_blocksize
;
1968 while (cur
<= end
) {
1969 if (cur
>= last_byte
) {
1970 clear_extent_dirty(tree
, cur
, page_end
, GFP_NOFS
);
1973 em
= epd
->get_extent(inode
, page
, page_offset
, cur
,
1975 if (IS_ERR(em
) || !em
) {
1980 extent_offset
= cur
- em
->start
;
1981 BUG_ON(extent_map_end(em
) <= cur
);
1983 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
1984 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
1985 sector
= (em
->block_start
+ extent_offset
) >> 9;
1987 block_start
= em
->block_start
;
1988 free_extent_map(em
);
1991 if (block_start
== EXTENT_MAP_HOLE
||
1992 block_start
== EXTENT_MAP_INLINE
) {
1993 clear_extent_dirty(tree
, cur
,
1994 cur
+ iosize
- 1, GFP_NOFS
);
1996 page_offset
+= iosize
;
2000 /* leave this out until we have a page_mkwrite call */
2001 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2004 page_offset
+= iosize
;
2007 clear_extent_dirty(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2008 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2009 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2017 unsigned long max_nr
= end_index
+ 1;
2018 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2019 if (!PageWriteback(page
)) {
2020 printk("warning page %lu not writeback, "
2021 "cur %llu end %llu\n", page
->index
,
2022 (unsigned long long)cur
,
2023 (unsigned long long)end
);
2026 ret
= submit_extent_page(WRITE
, tree
, page
, sector
,
2027 iosize
, page_offset
, bdev
,
2029 end_bio_extent_writepage
);
2034 page_offset
+= iosize
;
2039 /* make sure the mapping tag for page dirty gets cleared */
2040 set_page_writeback(page
);
2041 end_page_writeback(page
);
2043 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
2048 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2050 /* Taken directly from 2.6.23 for 2.6.18 back port */
2051 typedef int (*writepage_t
)(struct page
*page
, struct writeback_control
*wbc
,
2055 * write_cache_pages - walk the list of dirty pages of the given address space
2056 * and write all of them.
2057 * @mapping: address space structure to write
2058 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2059 * @writepage: function called for each page
2060 * @data: data passed to writepage function
2062 * If a page is already under I/O, write_cache_pages() skips it, even
2063 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2064 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2065 * and msync() need to guarantee that all the data which was dirty at the time
2066 * the call was made get new I/O started against them. If wbc->sync_mode is
2067 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2068 * existing IO to complete.
2070 static int write_cache_pages(struct address_space
*mapping
,
2071 struct writeback_control
*wbc
, writepage_t writepage
,
2074 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
2077 struct pagevec pvec
;
2080 pgoff_t end
; /* Inclusive */
2082 int range_whole
= 0;
2084 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2085 wbc
->encountered_congestion
= 1;
2089 pagevec_init(&pvec
, 0);
2090 if (wbc
->range_cyclic
) {
2091 index
= mapping
->writeback_index
; /* Start from prev offset */
2094 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2095 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2096 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2101 while (!done
&& (index
<= end
) &&
2102 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2103 PAGECACHE_TAG_DIRTY
,
2104 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2108 for (i
= 0; i
< nr_pages
; i
++) {
2109 struct page
*page
= pvec
.pages
[i
];
2112 * At this point we hold neither mapping->tree_lock nor
2113 * lock on the page itself: the page may be truncated or
2114 * invalidated (changing page->mapping to NULL), or even
2115 * swizzled back from swapper_space to tmpfs file
2120 if (unlikely(page
->mapping
!= mapping
)) {
2125 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2131 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2132 wait_on_page_writeback(page
);
2134 if (PageWriteback(page
) ||
2135 !clear_page_dirty_for_io(page
)) {
2140 ret
= (*writepage
)(page
, wbc
, data
);
2142 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2146 if (ret
|| (--(wbc
->nr_to_write
) <= 0))
2148 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2149 wbc
->encountered_congestion
= 1;
2153 pagevec_release(&pvec
);
2156 if (!scanned
&& !done
) {
2158 * We hit the last page and there is more work to be done: wrap
2159 * back to the start of the file
2165 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2166 mapping
->writeback_index
= index
;
2171 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2172 get_extent_t
*get_extent
,
2173 struct writeback_control
*wbc
)
2176 struct address_space
*mapping
= page
->mapping
;
2177 struct extent_page_data epd
= {
2180 .get_extent
= get_extent
,
2182 struct writeback_control wbc_writepages
= {
2184 .sync_mode
= WB_SYNC_NONE
,
2185 .older_than_this
= NULL
,
2187 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2188 .range_end
= (loff_t
)-1,
2192 ret
= __extent_writepage(page
, wbc
, &epd
);
2194 write_cache_pages(mapping
, &wbc_writepages
, __extent_writepage
, &epd
);
2196 submit_one_bio(WRITE
, epd
.bio
);
2200 EXPORT_SYMBOL(extent_write_full_page
);
2203 int extent_writepages(struct extent_io_tree
*tree
,
2204 struct address_space
*mapping
,
2205 get_extent_t
*get_extent
,
2206 struct writeback_control
*wbc
)
2209 struct extent_page_data epd
= {
2212 .get_extent
= get_extent
,
2215 ret
= write_cache_pages(mapping
, wbc
, __extent_writepage
, &epd
);
2217 submit_one_bio(WRITE
, epd
.bio
);
2221 EXPORT_SYMBOL(extent_writepages
);
2223 int extent_readpages(struct extent_io_tree
*tree
,
2224 struct address_space
*mapping
,
2225 struct list_head
*pages
, unsigned nr_pages
,
2226 get_extent_t get_extent
)
2228 struct bio
*bio
= NULL
;
2230 struct pagevec pvec
;
2232 pagevec_init(&pvec
, 0);
2233 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2234 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2236 prefetchw(&page
->flags
);
2237 list_del(&page
->lru
);
2239 * what we want to do here is call add_to_page_cache_lru,
2240 * but that isn't exported, so we reproduce it here
2242 if (!add_to_page_cache(page
, mapping
,
2243 page
->index
, GFP_KERNEL
)) {
2245 /* open coding of lru_cache_add, also not exported */
2246 page_cache_get(page
);
2247 if (!pagevec_add(&pvec
, page
))
2248 __pagevec_lru_add(&pvec
);
2249 __extent_read_full_page(tree
, page
, get_extent
, &bio
);
2251 page_cache_release(page
);
2253 if (pagevec_count(&pvec
))
2254 __pagevec_lru_add(&pvec
);
2255 BUG_ON(!list_empty(pages
));
2257 submit_one_bio(READ
, bio
);
2260 EXPORT_SYMBOL(extent_readpages
);
2263 * basic invalidatepage code, this waits on any locked or writeback
2264 * ranges corresponding to the page, and then deletes any extent state
2265 * records from the tree
2267 int extent_invalidatepage(struct extent_io_tree
*tree
,
2268 struct page
*page
, unsigned long offset
)
2270 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2271 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2272 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2274 start
+= (offset
+ blocksize
-1) & ~(blocksize
- 1);
2278 lock_extent(tree
, start
, end
, GFP_NOFS
);
2279 wait_on_extent_writeback(tree
, start
, end
);
2280 clear_extent_bit(tree
, start
, end
,
2281 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
,
2285 EXPORT_SYMBOL(extent_invalidatepage
);
2288 * simple commit_write call, set_range_dirty is used to mark both
2289 * the pages and the extent records as dirty
2291 int extent_commit_write(struct extent_io_tree
*tree
,
2292 struct inode
*inode
, struct page
*page
,
2293 unsigned from
, unsigned to
)
2295 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2297 set_page_extent_mapped(page
);
2298 set_page_dirty(page
);
2300 if (pos
> inode
->i_size
) {
2301 i_size_write(inode
, pos
);
2302 mark_inode_dirty(inode
);
2306 EXPORT_SYMBOL(extent_commit_write
);
2308 int extent_prepare_write(struct extent_io_tree
*tree
,
2309 struct inode
*inode
, struct page
*page
,
2310 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2312 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2313 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2315 u64 orig_block_start
;
2318 struct extent_map
*em
;
2319 unsigned blocksize
= 1 << inode
->i_blkbits
;
2320 size_t page_offset
= 0;
2321 size_t block_off_start
;
2322 size_t block_off_end
;
2328 set_page_extent_mapped(page
);
2330 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2331 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2332 orig_block_start
= block_start
;
2334 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2335 while(block_start
<= block_end
) {
2336 em
= get_extent(inode
, page
, page_offset
, block_start
,
2337 block_end
- block_start
+ 1, 1);
2338 if (IS_ERR(em
) || !em
) {
2341 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2342 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2343 block_off_end
= block_off_start
+ blocksize
;
2344 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2346 if (!PageUptodate(page
) && isnew
&&
2347 (block_off_end
> to
|| block_off_start
< from
)) {
2350 kaddr
= kmap_atomic(page
, KM_USER0
);
2351 if (block_off_end
> to
)
2352 memset(kaddr
+ to
, 0, block_off_end
- to
);
2353 if (block_off_start
< from
)
2354 memset(kaddr
+ block_off_start
, 0,
2355 from
- block_off_start
);
2356 flush_dcache_page(page
);
2357 kunmap_atomic(kaddr
, KM_USER0
);
2359 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2360 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2361 !isnew
&& !PageUptodate(page
) &&
2362 (block_off_end
> to
|| block_off_start
< from
) &&
2363 !test_range_bit(tree
, block_start
, cur_end
,
2364 EXTENT_UPTODATE
, 1)) {
2366 u64 extent_offset
= block_start
- em
->start
;
2368 sector
= (em
->block_start
+ extent_offset
) >> 9;
2369 iosize
= (cur_end
- block_start
+ blocksize
) &
2370 ~((u64
)blocksize
- 1);
2372 * we've already got the extent locked, but we
2373 * need to split the state such that our end_bio
2374 * handler can clear the lock.
2376 set_extent_bit(tree
, block_start
,
2377 block_start
+ iosize
- 1,
2378 EXTENT_LOCKED
, 0, NULL
, GFP_NOFS
);
2379 ret
= submit_extent_page(READ
, tree
, page
,
2380 sector
, iosize
, page_offset
, em
->bdev
,
2382 end_bio_extent_preparewrite
);
2384 block_start
= block_start
+ iosize
;
2386 set_extent_uptodate(tree
, block_start
, cur_end
,
2388 unlock_extent(tree
, block_start
, cur_end
, GFP_NOFS
);
2389 block_start
= cur_end
+ 1;
2391 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2392 free_extent_map(em
);
2395 wait_extent_bit(tree
, orig_block_start
,
2396 block_end
, EXTENT_LOCKED
);
2398 check_page_uptodate(tree
, page
);
2400 /* FIXME, zero out newly allocated blocks on error */
2403 EXPORT_SYMBOL(extent_prepare_write
);
2406 * a helper for releasepage. As long as there are no locked extents
2407 * in the range corresponding to the page, both state records and extent
2408 * map records are removed
2410 int try_release_extent_mapping(struct extent_map_tree
*map
,
2411 struct extent_io_tree
*tree
, struct page
*page
,
2414 struct extent_map
*em
;
2415 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2416 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2417 u64 orig_start
= start
;
2420 if ((mask
& __GFP_WAIT
) &&
2421 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2422 while (start
<= end
) {
2423 spin_lock(&map
->lock
);
2424 em
= lookup_extent_mapping(map
, start
, end
);
2425 if (!em
|| IS_ERR(em
)) {
2426 spin_unlock(&map
->lock
);
2429 if (em
->start
!= start
) {
2430 spin_unlock(&map
->lock
);
2431 free_extent_map(em
);
2434 if (!test_range_bit(tree
, em
->start
,
2435 extent_map_end(em
) - 1,
2436 EXTENT_LOCKED
, 0)) {
2437 remove_extent_mapping(map
, em
);
2438 /* once for the rb tree */
2439 free_extent_map(em
);
2441 start
= extent_map_end(em
);
2442 spin_unlock(&map
->lock
);
2445 free_extent_map(em
);
2448 if (test_range_bit(tree
, orig_start
, end
, EXTENT_IOBITS
, 0))
2451 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2453 clear_extent_bit(tree
, orig_start
, end
, EXTENT_UPTODATE
,
2458 EXPORT_SYMBOL(try_release_extent_mapping
);
2460 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2461 get_extent_t
*get_extent
)
2463 struct inode
*inode
= mapping
->host
;
2464 u64 start
= iblock
<< inode
->i_blkbits
;
2465 sector_t sector
= 0;
2466 struct extent_map
*em
;
2468 em
= get_extent(inode
, NULL
, 0, start
, (1 << inode
->i_blkbits
), 0);
2469 if (!em
|| IS_ERR(em
))
2472 if (em
->block_start
== EXTENT_MAP_INLINE
||
2473 em
->block_start
== EXTENT_MAP_HOLE
)
2476 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2478 free_extent_map(em
);
2482 static int add_lru(struct extent_io_tree
*tree
, struct extent_buffer
*eb
)
2484 if (list_empty(&eb
->lru
)) {
2485 extent_buffer_get(eb
);
2486 list_add(&eb
->lru
, &tree
->buffer_lru
);
2488 if (tree
->lru_size
>= BUFFER_LRU_MAX
) {
2489 struct extent_buffer
*rm
;
2490 rm
= list_entry(tree
->buffer_lru
.prev
,
2491 struct extent_buffer
, lru
);
2493 list_del_init(&rm
->lru
);
2494 free_extent_buffer(rm
);
2497 list_move(&eb
->lru
, &tree
->buffer_lru
);
2500 static struct extent_buffer
*find_lru(struct extent_io_tree
*tree
,
2501 u64 start
, unsigned long len
)
2503 struct list_head
*lru
= &tree
->buffer_lru
;
2504 struct list_head
*cur
= lru
->next
;
2505 struct extent_buffer
*eb
;
2507 if (list_empty(lru
))
2511 eb
= list_entry(cur
, struct extent_buffer
, lru
);
2512 if (eb
->start
== start
&& eb
->len
== len
) {
2513 extent_buffer_get(eb
);
2517 } while (cur
!= lru
);
2521 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
2523 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
2524 (start
>> PAGE_CACHE_SHIFT
);
2527 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
2531 struct address_space
*mapping
;
2534 return eb
->first_page
;
2535 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
2536 mapping
= eb
->first_page
->mapping
;
2537 read_lock_irq(&mapping
->tree_lock
);
2538 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
2539 read_unlock_irq(&mapping
->tree_lock
);
2543 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
2548 struct extent_buffer
*eb
= NULL
;
2550 spin_lock(&tree
->lru_lock
);
2551 eb
= find_lru(tree
, start
, len
);
2552 spin_unlock(&tree
->lru_lock
);
2557 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
2558 INIT_LIST_HEAD(&eb
->lru
);
2561 atomic_set(&eb
->refs
, 1);
2566 static void __free_extent_buffer(struct extent_buffer
*eb
)
2568 kmem_cache_free(extent_buffer_cache
, eb
);
2571 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
2572 u64 start
, unsigned long len
,
2576 unsigned long num_pages
= num_extent_pages(start
, len
);
2578 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2579 struct extent_buffer
*eb
;
2581 struct address_space
*mapping
= tree
->mapping
;
2584 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2585 if (!eb
|| IS_ERR(eb
))
2588 if (eb
->flags
& EXTENT_BUFFER_FILLED
)
2592 eb
->first_page
= page0
;
2595 page_cache_get(page0
);
2596 mark_page_accessed(page0
);
2597 set_page_extent_mapped(page0
);
2598 WARN_ON(!PageUptodate(page0
));
2599 set_page_extent_head(page0
, len
);
2603 for (; i
< num_pages
; i
++, index
++) {
2604 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
2609 set_page_extent_mapped(p
);
2610 mark_page_accessed(p
);
2613 set_page_extent_head(p
, len
);
2615 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2617 if (!PageUptodate(p
))
2622 eb
->flags
|= EXTENT_UPTODATE
;
2623 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2626 spin_lock(&tree
->lru_lock
);
2628 spin_unlock(&tree
->lru_lock
);
2632 spin_lock(&tree
->lru_lock
);
2633 list_del_init(&eb
->lru
);
2634 spin_unlock(&tree
->lru_lock
);
2635 if (!atomic_dec_and_test(&eb
->refs
))
2637 for (index
= 1; index
< i
; index
++) {
2638 page_cache_release(extent_buffer_page(eb
, index
));
2641 page_cache_release(extent_buffer_page(eb
, 0));
2642 __free_extent_buffer(eb
);
2645 EXPORT_SYMBOL(alloc_extent_buffer
);
2647 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
2648 u64 start
, unsigned long len
,
2651 unsigned long num_pages
= num_extent_pages(start
, len
);
2653 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2654 struct extent_buffer
*eb
;
2656 struct address_space
*mapping
= tree
->mapping
;
2659 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2660 if (!eb
|| IS_ERR(eb
))
2663 if (eb
->flags
& EXTENT_BUFFER_FILLED
)
2666 for (i
= 0; i
< num_pages
; i
++, index
++) {
2667 p
= find_lock_page(mapping
, index
);
2671 set_page_extent_mapped(p
);
2672 mark_page_accessed(p
);
2676 set_page_extent_head(p
, len
);
2678 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2681 if (!PageUptodate(p
))
2686 eb
->flags
|= EXTENT_UPTODATE
;
2687 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2690 spin_lock(&tree
->lru_lock
);
2692 spin_unlock(&tree
->lru_lock
);
2695 spin_lock(&tree
->lru_lock
);
2696 list_del_init(&eb
->lru
);
2697 spin_unlock(&tree
->lru_lock
);
2698 if (!atomic_dec_and_test(&eb
->refs
))
2700 for (index
= 1; index
< i
; index
++) {
2701 page_cache_release(extent_buffer_page(eb
, index
));
2704 page_cache_release(extent_buffer_page(eb
, 0));
2705 __free_extent_buffer(eb
);
2708 EXPORT_SYMBOL(find_extent_buffer
);
2710 void free_extent_buffer(struct extent_buffer
*eb
)
2713 unsigned long num_pages
;
2718 if (!atomic_dec_and_test(&eb
->refs
))
2721 WARN_ON(!list_empty(&eb
->lru
));
2722 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2724 for (i
= 1; i
< num_pages
; i
++) {
2725 page_cache_release(extent_buffer_page(eb
, i
));
2727 page_cache_release(extent_buffer_page(eb
, 0));
2728 __free_extent_buffer(eb
);
2730 EXPORT_SYMBOL(free_extent_buffer
);
2732 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
2733 struct extent_buffer
*eb
)
2737 unsigned long num_pages
;
2740 u64 start
= eb
->start
;
2741 u64 end
= start
+ eb
->len
- 1;
2743 set
= clear_extent_dirty(tree
, start
, end
, GFP_NOFS
);
2744 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2746 for (i
= 0; i
< num_pages
; i
++) {
2747 page
= extent_buffer_page(eb
, i
);
2750 set_page_extent_head(page
, eb
->len
);
2752 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2755 * if we're on the last page or the first page and the
2756 * block isn't aligned on a page boundary, do extra checks
2757 * to make sure we don't clean page that is partially dirty
2759 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2760 ((i
== num_pages
- 1) &&
2761 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2762 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2763 end
= start
+ PAGE_CACHE_SIZE
- 1;
2764 if (test_range_bit(tree
, start
, end
,
2770 clear_page_dirty_for_io(page
);
2771 read_lock_irq(&page
->mapping
->tree_lock
);
2772 if (!PageDirty(page
)) {
2773 radix_tree_tag_clear(&page
->mapping
->page_tree
,
2775 PAGECACHE_TAG_DIRTY
);
2777 read_unlock_irq(&page
->mapping
->tree_lock
);
2782 EXPORT_SYMBOL(clear_extent_buffer_dirty
);
2784 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
2785 struct extent_buffer
*eb
)
2787 return wait_on_extent_writeback(tree
, eb
->start
,
2788 eb
->start
+ eb
->len
- 1);
2790 EXPORT_SYMBOL(wait_on_extent_buffer_writeback
);
2792 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
2793 struct extent_buffer
*eb
)
2796 unsigned long num_pages
;
2798 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2799 for (i
= 0; i
< num_pages
; i
++) {
2800 struct page
*page
= extent_buffer_page(eb
, i
);
2801 /* writepage may need to do something special for the
2802 * first page, we have to make sure page->private is
2803 * properly set. releasepage may drop page->private
2804 * on us if the page isn't already dirty.
2808 set_page_extent_head(page
, eb
->len
);
2809 } else if (PagePrivate(page
) &&
2810 page
->private != EXTENT_PAGE_PRIVATE
) {
2812 set_page_extent_mapped(page
);
2815 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
2819 return set_extent_dirty(tree
, eb
->start
,
2820 eb
->start
+ eb
->len
- 1, GFP_NOFS
);
2822 EXPORT_SYMBOL(set_extent_buffer_dirty
);
2824 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
2825 struct extent_buffer
*eb
)
2829 unsigned long num_pages
;
2831 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2833 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2835 for (i
= 0; i
< num_pages
; i
++) {
2836 page
= extent_buffer_page(eb
, i
);
2837 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2838 ((i
== num_pages
- 1) &&
2839 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2840 check_page_uptodate(tree
, page
);
2843 SetPageUptodate(page
);
2847 EXPORT_SYMBOL(set_extent_buffer_uptodate
);
2849 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
2850 struct extent_buffer
*eb
)
2852 if (eb
->flags
& EXTENT_UPTODATE
)
2854 return test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2855 EXTENT_UPTODATE
, 1);
2857 EXPORT_SYMBOL(extent_buffer_uptodate
);
2859 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
2860 struct extent_buffer
*eb
,
2865 unsigned long start_i
;
2869 unsigned long num_pages
;
2871 if (eb
->flags
& EXTENT_UPTODATE
)
2874 if (0 && test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2875 EXTENT_UPTODATE
, 1)) {
2880 WARN_ON(start
< eb
->start
);
2881 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
2882 (eb
->start
>> PAGE_CACHE_SHIFT
);
2887 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2888 for (i
= start_i
; i
< num_pages
; i
++) {
2889 page
= extent_buffer_page(eb
, i
);
2890 if (PageUptodate(page
)) {
2894 if (TestSetPageLocked(page
)) {
2900 if (!PageUptodate(page
)) {
2901 err
= page
->mapping
->a_ops
->readpage(NULL
, page
);
2913 for (i
= start_i
; i
< num_pages
; i
++) {
2914 page
= extent_buffer_page(eb
, i
);
2915 wait_on_page_locked(page
);
2916 if (!PageUptodate(page
)) {
2921 eb
->flags
|= EXTENT_UPTODATE
;
2924 EXPORT_SYMBOL(read_extent_buffer_pages
);
2926 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
2927 unsigned long start
,
2934 char *dst
= (char *)dstv
;
2935 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
2936 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
2937 unsigned long num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2939 WARN_ON(start
> eb
->len
);
2940 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
2942 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
2945 page
= extent_buffer_page(eb
, i
);
2946 if (!PageUptodate(page
)) {
2947 printk("page %lu not up to date i %lu, total %lu, len %lu\n", page
->index
, i
, num_pages
, eb
->len
);
2950 WARN_ON(!PageUptodate(page
));
2952 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
2953 kaddr
= kmap_atomic(page
, KM_USER1
);
2954 memcpy(dst
, kaddr
+ offset
, cur
);
2955 kunmap_atomic(kaddr
, KM_USER1
);
2963 EXPORT_SYMBOL(read_extent_buffer
);
2965 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
2966 unsigned long min_len
, char **token
, char **map
,
2967 unsigned long *map_start
,
2968 unsigned long *map_len
, int km
)
2970 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
2973 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
2974 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
2975 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
2982 offset
= start_offset
;
2986 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
2988 if (start
+ min_len
> eb
->len
) {
2989 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb
->start
, eb
->len
, start
, min_len
);
2993 p
= extent_buffer_page(eb
, i
);
2994 WARN_ON(!PageUptodate(p
));
2995 kaddr
= kmap_atomic(p
, km
);
2997 *map
= kaddr
+ offset
;
2998 *map_len
= PAGE_CACHE_SIZE
- offset
;
3001 EXPORT_SYMBOL(map_private_extent_buffer
);
3003 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3004 unsigned long min_len
,
3005 char **token
, char **map
,
3006 unsigned long *map_start
,
3007 unsigned long *map_len
, int km
)
3011 if (eb
->map_token
) {
3012 unmap_extent_buffer(eb
, eb
->map_token
, km
);
3013 eb
->map_token
= NULL
;
3016 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
3017 map_start
, map_len
, km
);
3019 eb
->map_token
= *token
;
3021 eb
->map_start
= *map_start
;
3022 eb
->map_len
= *map_len
;
3026 EXPORT_SYMBOL(map_extent_buffer
);
3028 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
3030 kunmap_atomic(token
, km
);
3032 EXPORT_SYMBOL(unmap_extent_buffer
);
3034 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3035 unsigned long start
,
3042 char *ptr
= (char *)ptrv
;
3043 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3044 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3047 WARN_ON(start
> eb
->len
);
3048 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3050 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3053 page
= extent_buffer_page(eb
, i
);
3054 WARN_ON(!PageUptodate(page
));
3056 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3058 kaddr
= kmap_atomic(page
, KM_USER0
);
3059 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3060 kunmap_atomic(kaddr
, KM_USER0
);
3071 EXPORT_SYMBOL(memcmp_extent_buffer
);
3073 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3074 unsigned long start
, unsigned long len
)
3080 char *src
= (char *)srcv
;
3081 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3082 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3084 WARN_ON(start
> eb
->len
);
3085 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3087 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3090 page
= extent_buffer_page(eb
, i
);
3091 WARN_ON(!PageUptodate(page
));
3093 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3094 kaddr
= kmap_atomic(page
, KM_USER1
);
3095 memcpy(kaddr
+ offset
, src
, cur
);
3096 kunmap_atomic(kaddr
, KM_USER1
);
3104 EXPORT_SYMBOL(write_extent_buffer
);
3106 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3107 unsigned long start
, unsigned long len
)
3113 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3114 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3116 WARN_ON(start
> eb
->len
);
3117 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3119 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3122 page
= extent_buffer_page(eb
, i
);
3123 WARN_ON(!PageUptodate(page
));
3125 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3126 kaddr
= kmap_atomic(page
, KM_USER0
);
3127 memset(kaddr
+ offset
, c
, cur
);
3128 kunmap_atomic(kaddr
, KM_USER0
);
3135 EXPORT_SYMBOL(memset_extent_buffer
);
3137 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3138 unsigned long dst_offset
, unsigned long src_offset
,
3141 u64 dst_len
= dst
->len
;
3146 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3147 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3149 WARN_ON(src
->len
!= dst_len
);
3151 offset
= (start_offset
+ dst_offset
) &
3152 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3155 page
= extent_buffer_page(dst
, i
);
3156 WARN_ON(!PageUptodate(page
));
3158 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3160 kaddr
= kmap_atomic(page
, KM_USER0
);
3161 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3162 kunmap_atomic(kaddr
, KM_USER0
);
3170 EXPORT_SYMBOL(copy_extent_buffer
);
3172 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3173 unsigned long dst_off
, unsigned long src_off
,
3176 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3177 if (dst_page
== src_page
) {
3178 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3180 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3181 char *p
= dst_kaddr
+ dst_off
+ len
;
3182 char *s
= src_kaddr
+ src_off
+ len
;
3187 kunmap_atomic(src_kaddr
, KM_USER1
);
3189 kunmap_atomic(dst_kaddr
, KM_USER0
);
3192 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3193 unsigned long dst_off
, unsigned long src_off
,
3196 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3199 if (dst_page
!= src_page
)
3200 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3202 src_kaddr
= dst_kaddr
;
3204 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3205 kunmap_atomic(dst_kaddr
, KM_USER0
);
3206 if (dst_page
!= src_page
)
3207 kunmap_atomic(src_kaddr
, KM_USER1
);
3210 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3211 unsigned long src_offset
, unsigned long len
)
3214 size_t dst_off_in_page
;
3215 size_t src_off_in_page
;
3216 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3217 unsigned long dst_i
;
3218 unsigned long src_i
;
3220 if (src_offset
+ len
> dst
->len
) {
3221 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3222 src_offset
, len
, dst
->len
);
3225 if (dst_offset
+ len
> dst
->len
) {
3226 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3227 dst_offset
, len
, dst
->len
);
3232 dst_off_in_page
= (start_offset
+ dst_offset
) &
3233 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3234 src_off_in_page
= (start_offset
+ src_offset
) &
3235 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3237 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3238 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3240 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3242 cur
= min_t(unsigned long, cur
,
3243 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3245 copy_pages(extent_buffer_page(dst
, dst_i
),
3246 extent_buffer_page(dst
, src_i
),
3247 dst_off_in_page
, src_off_in_page
, cur
);
3254 EXPORT_SYMBOL(memcpy_extent_buffer
);
3256 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3257 unsigned long src_offset
, unsigned long len
)
3260 size_t dst_off_in_page
;
3261 size_t src_off_in_page
;
3262 unsigned long dst_end
= dst_offset
+ len
- 1;
3263 unsigned long src_end
= src_offset
+ len
- 1;
3264 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3265 unsigned long dst_i
;
3266 unsigned long src_i
;
3268 if (src_offset
+ len
> dst
->len
) {
3269 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3270 src_offset
, len
, dst
->len
);
3273 if (dst_offset
+ len
> dst
->len
) {
3274 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3275 dst_offset
, len
, dst
->len
);
3278 if (dst_offset
< src_offset
) {
3279 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
3283 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
3284 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
3286 dst_off_in_page
= (start_offset
+ dst_end
) &
3287 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3288 src_off_in_page
= (start_offset
+ src_end
) &
3289 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3291 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
3292 cur
= min(cur
, dst_off_in_page
+ 1);
3293 move_pages(extent_buffer_page(dst
, dst_i
),
3294 extent_buffer_page(dst
, src_i
),
3295 dst_off_in_page
- cur
+ 1,
3296 src_off_in_page
- cur
+ 1, cur
);
3303 EXPORT_SYMBOL(memmove_extent_buffer
);