1 #include <linux/bitops.h>
2 #include <linux/slab.h>
6 #include <linux/pagemap.h>
7 #include <linux/page-flags.h>
8 #include <linux/module.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/version.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 /* temporary define until extent_map moves out of btrfs */
22 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
23 unsigned long extra_flags
,
24 void (*ctor
)(void *, struct kmem_cache
*,
27 static struct kmem_cache
*extent_state_cache
;
28 static struct kmem_cache
*extent_buffer_cache
;
30 static LIST_HEAD(buffers
);
31 static LIST_HEAD(states
);
34 static spinlock_t leak_lock
= SPIN_LOCK_UNLOCKED
;
37 #define BUFFER_LRU_MAX 64
42 struct rb_node rb_node
;
45 struct extent_page_data
{
47 struct extent_io_tree
*tree
;
48 get_extent_t
*get_extent
;
51 int __init
extent_io_init(void)
53 extent_state_cache
= btrfs_cache_create("extent_state",
54 sizeof(struct extent_state
), 0,
56 if (!extent_state_cache
)
59 extent_buffer_cache
= btrfs_cache_create("extent_buffers",
60 sizeof(struct extent_buffer
), 0,
62 if (!extent_buffer_cache
)
63 goto free_state_cache
;
67 kmem_cache_destroy(extent_state_cache
);
71 void extent_io_exit(void)
73 struct extent_state
*state
;
74 struct extent_buffer
*eb
;
76 while (!list_empty(&states
)) {
77 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
78 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
));
79 list_del(&state
->leak_list
);
80 kmem_cache_free(extent_state_cache
, state
);
84 while (!list_empty(&buffers
)) {
85 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
86 printk("buffer leak start %Lu len %lu refs %d\n", eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
87 list_del(&eb
->leak_list
);
88 kmem_cache_free(extent_buffer_cache
, eb
);
90 if (extent_state_cache
)
91 kmem_cache_destroy(extent_state_cache
);
92 if (extent_buffer_cache
)
93 kmem_cache_destroy(extent_buffer_cache
);
96 void extent_io_tree_init(struct extent_io_tree
*tree
,
97 struct address_space
*mapping
, gfp_t mask
)
99 tree
->state
.rb_node
= NULL
;
100 tree
->buffer
.rb_node
= NULL
;
102 tree
->dirty_bytes
= 0;
103 spin_lock_init(&tree
->lock
);
104 spin_lock_init(&tree
->buffer_lock
);
105 tree
->mapping
= mapping
;
107 EXPORT_SYMBOL(extent_io_tree_init
);
109 struct extent_state
*alloc_extent_state(gfp_t mask
)
111 struct extent_state
*state
;
116 state
= kmem_cache_alloc(extent_state_cache
, mask
);
123 spin_lock_irqsave(&leak_lock
, flags
);
124 list_add(&state
->leak_list
, &states
);
125 spin_unlock_irqrestore(&leak_lock
, flags
);
127 atomic_set(&state
->refs
, 1);
128 init_waitqueue_head(&state
->wq
);
131 EXPORT_SYMBOL(alloc_extent_state
);
133 void free_extent_state(struct extent_state
*state
)
137 if (atomic_dec_and_test(&state
->refs
)) {
141 WARN_ON(state
->tree
);
143 spin_lock_irqsave(&leak_lock
, flags
);
144 list_del(&state
->leak_list
);
145 spin_unlock_irqrestore(&leak_lock
, flags
);
147 kmem_cache_free(extent_state_cache
, state
);
150 EXPORT_SYMBOL(free_extent_state
);
152 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
153 struct rb_node
*node
)
155 struct rb_node
** p
= &root
->rb_node
;
156 struct rb_node
* parent
= NULL
;
157 struct tree_entry
*entry
;
161 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
163 if (offset
< entry
->start
)
165 else if (offset
> entry
->end
)
171 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
172 rb_link_node(node
, parent
, p
);
173 rb_insert_color(node
, root
);
177 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
178 struct rb_node
**prev_ret
,
179 struct rb_node
**next_ret
)
181 struct rb_root
*root
= &tree
->state
;
182 struct rb_node
* n
= root
->rb_node
;
183 struct rb_node
*prev
= NULL
;
184 struct rb_node
*orig_prev
= NULL
;
185 struct tree_entry
*entry
;
186 struct tree_entry
*prev_entry
= NULL
;
189 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
193 if (offset
< entry
->start
)
195 else if (offset
> entry
->end
)
204 while(prev
&& offset
> prev_entry
->end
) {
205 prev
= rb_next(prev
);
206 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
213 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
214 while(prev
&& offset
< prev_entry
->start
) {
215 prev
= rb_prev(prev
);
216 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
223 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
226 struct rb_node
*prev
= NULL
;
229 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
236 static struct extent_buffer
*buffer_tree_insert(struct extent_io_tree
*tree
,
237 u64 offset
, struct rb_node
*node
)
239 struct rb_root
*root
= &tree
->buffer
;
240 struct rb_node
** p
= &root
->rb_node
;
241 struct rb_node
* parent
= NULL
;
242 struct extent_buffer
*eb
;
246 eb
= rb_entry(parent
, struct extent_buffer
, rb_node
);
248 if (offset
< eb
->start
)
250 else if (offset
> eb
->start
)
256 rb_link_node(node
, parent
, p
);
257 rb_insert_color(node
, root
);
261 static struct extent_buffer
*buffer_search(struct extent_io_tree
*tree
,
264 struct rb_root
*root
= &tree
->buffer
;
265 struct rb_node
* n
= root
->rb_node
;
266 struct extent_buffer
*eb
;
269 eb
= rb_entry(n
, struct extent_buffer
, rb_node
);
270 if (offset
< eb
->start
)
272 else if (offset
> eb
->start
)
281 * utility function to look for merge candidates inside a given range.
282 * Any extents with matching state are merged together into a single
283 * extent in the tree. Extents with EXTENT_IO in their state field
284 * are not merged because the end_io handlers need to be able to do
285 * operations on them without sleeping (or doing allocations/splits).
287 * This should be called with the tree lock held.
289 static int merge_state(struct extent_io_tree
*tree
,
290 struct extent_state
*state
)
292 struct extent_state
*other
;
293 struct rb_node
*other_node
;
295 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
298 other_node
= rb_prev(&state
->rb_node
);
300 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
301 if (other
->end
== state
->start
- 1 &&
302 other
->state
== state
->state
) {
303 state
->start
= other
->start
;
305 rb_erase(&other
->rb_node
, &tree
->state
);
306 free_extent_state(other
);
309 other_node
= rb_next(&state
->rb_node
);
311 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
312 if (other
->start
== state
->end
+ 1 &&
313 other
->state
== state
->state
) {
314 other
->start
= state
->start
;
316 rb_erase(&state
->rb_node
, &tree
->state
);
317 free_extent_state(state
);
323 static void set_state_cb(struct extent_io_tree
*tree
,
324 struct extent_state
*state
,
327 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
328 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
->start
,
329 state
->end
, state
->state
, bits
);
333 static void clear_state_cb(struct extent_io_tree
*tree
,
334 struct extent_state
*state
,
337 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
338 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
->start
,
339 state
->end
, state
->state
, bits
);
344 * insert an extent_state struct into the tree. 'bits' are set on the
345 * struct before it is inserted.
347 * This may return -EEXIST if the extent is already there, in which case the
348 * state struct is freed.
350 * The tree lock is not taken internally. This is a utility function and
351 * probably isn't what you want to call (see set/clear_extent_bit).
353 static int insert_state(struct extent_io_tree
*tree
,
354 struct extent_state
*state
, u64 start
, u64 end
,
357 struct rb_node
*node
;
360 printk("end < start %Lu %Lu\n", end
, start
);
363 if (bits
& EXTENT_DIRTY
)
364 tree
->dirty_bytes
+= end
- start
+ 1;
365 set_state_cb(tree
, state
, bits
);
366 state
->state
|= bits
;
367 state
->start
= start
;
369 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
371 struct extent_state
*found
;
372 found
= rb_entry(node
, struct extent_state
, rb_node
);
373 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, start
, end
);
374 free_extent_state(state
);
378 merge_state(tree
, state
);
383 * split a given extent state struct in two, inserting the preallocated
384 * struct 'prealloc' as the newly created second half. 'split' indicates an
385 * offset inside 'orig' where it should be split.
388 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
389 * are two extent state structs in the tree:
390 * prealloc: [orig->start, split - 1]
391 * orig: [ split, orig->end ]
393 * The tree locks are not taken by this function. They need to be held
396 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
397 struct extent_state
*prealloc
, u64 split
)
399 struct rb_node
*node
;
400 prealloc
->start
= orig
->start
;
401 prealloc
->end
= split
- 1;
402 prealloc
->state
= orig
->state
;
405 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
407 struct extent_state
*found
;
408 found
= rb_entry(node
, struct extent_state
, rb_node
);
409 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, prealloc
->start
, prealloc
->end
);
410 free_extent_state(prealloc
);
413 prealloc
->tree
= tree
;
418 * utility function to clear some bits in an extent state struct.
419 * it will optionally wake up any one waiting on this state (wake == 1), or
420 * forcibly remove the state from the tree (delete == 1).
422 * If no bits are set on the state struct after clearing things, the
423 * struct is freed and removed from the tree
425 static int clear_state_bit(struct extent_io_tree
*tree
,
426 struct extent_state
*state
, int bits
, int wake
,
429 int ret
= state
->state
& bits
;
431 if ((bits
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
432 u64 range
= state
->end
- state
->start
+ 1;
433 WARN_ON(range
> tree
->dirty_bytes
);
434 tree
->dirty_bytes
-= range
;
436 clear_state_cb(tree
, state
, bits
);
437 state
->state
&= ~bits
;
440 if (delete || state
->state
== 0) {
442 clear_state_cb(tree
, state
, state
->state
);
443 rb_erase(&state
->rb_node
, &tree
->state
);
445 free_extent_state(state
);
450 merge_state(tree
, state
);
456 * clear some bits on a range in the tree. This may require splitting
457 * or inserting elements in the tree, so the gfp mask is used to
458 * indicate which allocations or sleeping are allowed.
460 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
461 * the given range from the tree regardless of state (ie for truncate).
463 * the range [start, end] is inclusive.
465 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
466 * bits were already set, or zero if none of the bits were already set.
468 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
469 int bits
, int wake
, int delete, gfp_t mask
)
471 struct extent_state
*state
;
472 struct extent_state
*prealloc
= NULL
;
473 struct rb_node
*node
;
479 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
480 prealloc
= alloc_extent_state(mask
);
485 spin_lock_irqsave(&tree
->lock
, flags
);
487 * this search will find the extents that end after
490 node
= tree_search(tree
, start
);
493 state
= rb_entry(node
, struct extent_state
, rb_node
);
494 if (state
->start
> end
)
496 WARN_ON(state
->end
< start
);
499 * | ---- desired range ---- |
501 * | ------------- state -------------- |
503 * We need to split the extent we found, and may flip
504 * bits on second half.
506 * If the extent we found extends past our range, we
507 * just split and search again. It'll get split again
508 * the next time though.
510 * If the extent we found is inside our range, we clear
511 * the desired bit on it.
514 if (state
->start
< start
) {
516 prealloc
= alloc_extent_state(GFP_ATOMIC
);
517 err
= split_state(tree
, state
, prealloc
, start
);
518 BUG_ON(err
== -EEXIST
);
522 if (state
->end
<= end
) {
523 start
= state
->end
+ 1;
524 set
|= clear_state_bit(tree
, state
, bits
,
527 start
= state
->start
;
532 * | ---- desired range ---- |
534 * We need to split the extent, and clear the bit
537 if (state
->start
<= end
&& state
->end
> end
) {
539 prealloc
= alloc_extent_state(GFP_ATOMIC
);
540 err
= split_state(tree
, state
, prealloc
, end
+ 1);
541 BUG_ON(err
== -EEXIST
);
545 set
|= clear_state_bit(tree
, prealloc
, bits
,
551 start
= state
->end
+ 1;
552 set
|= clear_state_bit(tree
, state
, bits
, wake
, delete);
556 spin_unlock_irqrestore(&tree
->lock
, flags
);
558 free_extent_state(prealloc
);
565 spin_unlock_irqrestore(&tree
->lock
, flags
);
566 if (mask
& __GFP_WAIT
)
570 EXPORT_SYMBOL(clear_extent_bit
);
572 static int wait_on_state(struct extent_io_tree
*tree
,
573 struct extent_state
*state
)
576 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
577 spin_unlock_irq(&tree
->lock
);
579 spin_lock_irq(&tree
->lock
);
580 finish_wait(&state
->wq
, &wait
);
585 * waits for one or more bits to clear on a range in the state tree.
586 * The range [start, end] is inclusive.
587 * The tree lock is taken by this function
589 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
591 struct extent_state
*state
;
592 struct rb_node
*node
;
594 spin_lock_irq(&tree
->lock
);
598 * this search will find all the extents that end after
601 node
= tree_search(tree
, start
);
605 state
= rb_entry(node
, struct extent_state
, rb_node
);
607 if (state
->start
> end
)
610 if (state
->state
& bits
) {
611 start
= state
->start
;
612 atomic_inc(&state
->refs
);
613 wait_on_state(tree
, state
);
614 free_extent_state(state
);
617 start
= state
->end
+ 1;
622 if (need_resched()) {
623 spin_unlock_irq(&tree
->lock
);
625 spin_lock_irq(&tree
->lock
);
629 spin_unlock_irq(&tree
->lock
);
632 EXPORT_SYMBOL(wait_extent_bit
);
634 static void set_state_bits(struct extent_io_tree
*tree
,
635 struct extent_state
*state
,
638 if ((bits
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
639 u64 range
= state
->end
- state
->start
+ 1;
640 tree
->dirty_bytes
+= range
;
642 set_state_cb(tree
, state
, bits
);
643 state
->state
|= bits
;
647 * set some bits on a range in the tree. This may require allocations
648 * or sleeping, so the gfp mask is used to indicate what is allowed.
650 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
651 * range already has the desired bits set. The start of the existing
652 * range is returned in failed_start in this case.
654 * [start, end] is inclusive
655 * This takes the tree lock.
657 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
,
658 int exclusive
, u64
*failed_start
, gfp_t mask
)
660 struct extent_state
*state
;
661 struct extent_state
*prealloc
= NULL
;
662 struct rb_node
*node
;
669 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
670 prealloc
= alloc_extent_state(mask
);
675 spin_lock_irqsave(&tree
->lock
, flags
);
677 * this search will find all the extents that end after
680 node
= tree_search(tree
, start
);
682 err
= insert_state(tree
, prealloc
, start
, end
, bits
);
684 BUG_ON(err
== -EEXIST
);
688 state
= rb_entry(node
, struct extent_state
, rb_node
);
689 last_start
= state
->start
;
690 last_end
= state
->end
;
693 * | ---- desired range ---- |
696 * Just lock what we found and keep going
698 if (state
->start
== start
&& state
->end
<= end
) {
699 set
= state
->state
& bits
;
700 if (set
&& exclusive
) {
701 *failed_start
= state
->start
;
705 set_state_bits(tree
, state
, bits
);
706 start
= state
->end
+ 1;
707 merge_state(tree
, state
);
712 * | ---- desired range ---- |
715 * | ------------- state -------------- |
717 * We need to split the extent we found, and may flip bits on
720 * If the extent we found extends past our
721 * range, we just split and search again. It'll get split
722 * again the next time though.
724 * If the extent we found is inside our range, we set the
727 if (state
->start
< start
) {
728 set
= state
->state
& bits
;
729 if (exclusive
&& set
) {
730 *failed_start
= start
;
734 err
= split_state(tree
, state
, prealloc
, start
);
735 BUG_ON(err
== -EEXIST
);
739 if (state
->end
<= end
) {
740 set_state_bits(tree
, state
, bits
);
741 start
= state
->end
+ 1;
742 merge_state(tree
, state
);
744 start
= state
->start
;
749 * | ---- desired range ---- |
750 * | state | or | state |
752 * There's a hole, we need to insert something in it and
753 * ignore the extent we found.
755 if (state
->start
> start
) {
757 if (end
< last_start
)
760 this_end
= last_start
-1;
761 err
= insert_state(tree
, prealloc
, start
, this_end
,
764 BUG_ON(err
== -EEXIST
);
767 start
= this_end
+ 1;
771 * | ---- desired range ---- |
773 * We need to split the extent, and set the bit
776 if (state
->start
<= end
&& state
->end
> end
) {
777 set
= state
->state
& bits
;
778 if (exclusive
&& set
) {
779 *failed_start
= start
;
783 err
= split_state(tree
, state
, prealloc
, end
+ 1);
784 BUG_ON(err
== -EEXIST
);
786 set_state_bits(tree
, prealloc
, bits
);
787 merge_state(tree
, prealloc
);
795 spin_unlock_irqrestore(&tree
->lock
, flags
);
797 free_extent_state(prealloc
);
804 spin_unlock_irqrestore(&tree
->lock
, flags
);
805 if (mask
& __GFP_WAIT
)
809 EXPORT_SYMBOL(set_extent_bit
);
811 /* wrappers around set/clear extent bit */
812 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
815 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
818 EXPORT_SYMBOL(set_extent_dirty
);
820 int set_extent_ordered(struct extent_io_tree
*tree
, u64 start
, u64 end
,
823 return set_extent_bit(tree
, start
, end
, EXTENT_ORDERED
, 0, NULL
, mask
);
825 EXPORT_SYMBOL(set_extent_ordered
);
827 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
828 int bits
, gfp_t mask
)
830 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
833 EXPORT_SYMBOL(set_extent_bits
);
835 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
836 int bits
, gfp_t mask
)
838 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, mask
);
840 EXPORT_SYMBOL(clear_extent_bits
);
842 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
845 return set_extent_bit(tree
, start
, end
,
846 EXTENT_DELALLOC
| EXTENT_DIRTY
,
849 EXPORT_SYMBOL(set_extent_delalloc
);
851 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
854 return clear_extent_bit(tree
, start
, end
,
855 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, mask
);
857 EXPORT_SYMBOL(clear_extent_dirty
);
859 int clear_extent_ordered(struct extent_io_tree
*tree
, u64 start
, u64 end
,
862 return clear_extent_bit(tree
, start
, end
, EXTENT_ORDERED
, 1, 0, mask
);
864 EXPORT_SYMBOL(clear_extent_ordered
);
866 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
869 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
872 EXPORT_SYMBOL(set_extent_new
);
874 int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
877 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0, mask
);
879 EXPORT_SYMBOL(clear_extent_new
);
881 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
884 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, NULL
,
887 EXPORT_SYMBOL(set_extent_uptodate
);
889 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
892 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0, mask
);
894 EXPORT_SYMBOL(clear_extent_uptodate
);
896 int set_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
899 return set_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
,
902 EXPORT_SYMBOL(set_extent_writeback
);
904 int clear_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
907 return clear_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 1, 0, mask
);
909 EXPORT_SYMBOL(clear_extent_writeback
);
911 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
913 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
915 EXPORT_SYMBOL(wait_on_extent_writeback
);
917 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
922 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
923 &failed_start
, mask
);
924 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
925 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
926 start
= failed_start
;
930 WARN_ON(start
> end
);
934 EXPORT_SYMBOL(lock_extent
);
936 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
939 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, mask
);
941 EXPORT_SYMBOL(unlock_extent
);
944 * helper function to set pages and extents in the tree dirty
946 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
948 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
949 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
952 while (index
<= end_index
) {
953 page
= find_get_page(tree
->mapping
, index
);
955 __set_page_dirty_nobuffers(page
);
956 page_cache_release(page
);
959 set_extent_dirty(tree
, start
, end
, GFP_NOFS
);
962 EXPORT_SYMBOL(set_range_dirty
);
965 * helper function to set both pages and extents in the tree writeback
967 int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
969 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
970 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
973 while (index
<= end_index
) {
974 page
= find_get_page(tree
->mapping
, index
);
976 set_page_writeback(page
);
977 page_cache_release(page
);
980 set_extent_writeback(tree
, start
, end
, GFP_NOFS
);
983 EXPORT_SYMBOL(set_range_writeback
);
985 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
986 u64
*start_ret
, u64
*end_ret
, int bits
)
988 struct rb_node
*node
;
989 struct extent_state
*state
;
992 spin_lock_irq(&tree
->lock
);
994 * this search will find all the extents that end after
997 node
= tree_search(tree
, start
);
1003 state
= rb_entry(node
, struct extent_state
, rb_node
);
1004 if (state
->end
>= start
&& (state
->state
& bits
)) {
1005 *start_ret
= state
->start
;
1006 *end_ret
= state
->end
;
1010 node
= rb_next(node
);
1015 spin_unlock_irq(&tree
->lock
);
1018 EXPORT_SYMBOL(find_first_extent_bit
);
1020 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
1021 u64 start
, int bits
)
1023 struct rb_node
*node
;
1024 struct extent_state
*state
;
1027 * this search will find all the extents that end after
1030 node
= tree_search(tree
, start
);
1036 state
= rb_entry(node
, struct extent_state
, rb_node
);
1037 if (state
->end
>= start
&& (state
->state
& bits
)) {
1040 node
= rb_next(node
);
1047 EXPORT_SYMBOL(find_first_extent_bit_state
);
1049 u64
find_lock_delalloc_range(struct extent_io_tree
*tree
,
1050 u64
*start
, u64
*end
, u64 max_bytes
)
1052 struct rb_node
*node
;
1053 struct extent_state
*state
;
1054 u64 cur_start
= *start
;
1056 u64 total_bytes
= 0;
1058 spin_lock_irq(&tree
->lock
);
1060 * this search will find all the extents that end after
1064 node
= tree_search(tree
, cur_start
);
1072 state
= rb_entry(node
, struct extent_state
, rb_node
);
1073 if (found
&& (state
->start
!= cur_start
||
1074 (state
->state
& EXTENT_BOUNDARY
))) {
1077 if (!(state
->state
& EXTENT_DELALLOC
)) {
1082 if (!found
&& !(state
->state
& EXTENT_BOUNDARY
)) {
1083 struct extent_state
*prev_state
;
1084 struct rb_node
*prev_node
= node
;
1086 prev_node
= rb_prev(prev_node
);
1089 prev_state
= rb_entry(prev_node
,
1090 struct extent_state
,
1092 if ((prev_state
->end
+ 1 != state
->start
) ||
1093 !(prev_state
->state
& EXTENT_DELALLOC
))
1095 if ((cur_start
- prev_state
->start
) * 2 >
1102 if (state
->state
& EXTENT_LOCKED
) {
1104 atomic_inc(&state
->refs
);
1105 prepare_to_wait(&state
->wq
, &wait
,
1106 TASK_UNINTERRUPTIBLE
);
1107 spin_unlock_irq(&tree
->lock
);
1109 spin_lock_irq(&tree
->lock
);
1110 finish_wait(&state
->wq
, &wait
);
1111 free_extent_state(state
);
1114 set_state_cb(tree
, state
, EXTENT_LOCKED
);
1115 state
->state
|= EXTENT_LOCKED
;
1117 *start
= state
->start
;
1120 cur_start
= state
->end
+ 1;
1121 node
= rb_next(node
);
1124 total_bytes
+= state
->end
- state
->start
+ 1;
1125 if (total_bytes
>= max_bytes
)
1129 spin_unlock_irq(&tree
->lock
);
1133 u64
count_range_bits(struct extent_io_tree
*tree
,
1134 u64
*start
, u64 search_end
, u64 max_bytes
,
1137 struct rb_node
*node
;
1138 struct extent_state
*state
;
1139 u64 cur_start
= *start
;
1140 u64 total_bytes
= 0;
1143 if (search_end
<= cur_start
) {
1144 printk("search_end %Lu start %Lu\n", search_end
, cur_start
);
1149 spin_lock_irq(&tree
->lock
);
1150 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1151 total_bytes
= tree
->dirty_bytes
;
1155 * this search will find all the extents that end after
1158 node
= tree_search(tree
, cur_start
);
1164 state
= rb_entry(node
, struct extent_state
, rb_node
);
1165 if (state
->start
> search_end
)
1167 if (state
->end
>= cur_start
&& (state
->state
& bits
)) {
1168 total_bytes
+= min(search_end
, state
->end
) + 1 -
1169 max(cur_start
, state
->start
);
1170 if (total_bytes
>= max_bytes
)
1173 *start
= state
->start
;
1177 node
= rb_next(node
);
1182 spin_unlock_irq(&tree
->lock
);
1186 * helper function to lock both pages and extents in the tree.
1187 * pages must be locked first.
1189 int lock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1191 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1192 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1196 while (index
<= end_index
) {
1197 page
= grab_cache_page(tree
->mapping
, index
);
1203 err
= PTR_ERR(page
);
1208 lock_extent(tree
, start
, end
, GFP_NOFS
);
1213 * we failed above in getting the page at 'index', so we undo here
1214 * up to but not including the page at 'index'
1217 index
= start
>> PAGE_CACHE_SHIFT
;
1218 while (index
< end_index
) {
1219 page
= find_get_page(tree
->mapping
, index
);
1221 page_cache_release(page
);
1226 EXPORT_SYMBOL(lock_range
);
1229 * helper function to unlock both pages and extents in the tree.
1231 int unlock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1233 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1234 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1237 while (index
<= end_index
) {
1238 page
= find_get_page(tree
->mapping
, index
);
1240 page_cache_release(page
);
1243 unlock_extent(tree
, start
, end
, GFP_NOFS
);
1246 EXPORT_SYMBOL(unlock_range
);
1248 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1250 struct rb_node
*node
;
1251 struct extent_state
*state
;
1254 spin_lock_irq(&tree
->lock
);
1256 * this search will find all the extents that end after
1259 node
= tree_search(tree
, start
);
1264 state
= rb_entry(node
, struct extent_state
, rb_node
);
1265 if (state
->start
!= start
) {
1269 state
->private = private;
1271 spin_unlock_irq(&tree
->lock
);
1275 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1277 struct rb_node
*node
;
1278 struct extent_state
*state
;
1281 spin_lock_irq(&tree
->lock
);
1283 * this search will find all the extents that end after
1286 node
= tree_search(tree
, start
);
1291 state
= rb_entry(node
, struct extent_state
, rb_node
);
1292 if (state
->start
!= start
) {
1296 *private = state
->private;
1298 spin_unlock_irq(&tree
->lock
);
1303 * searches a range in the state tree for a given mask.
1304 * If 'filled' == 1, this returns 1 only if every extent in the tree
1305 * has the bits set. Otherwise, 1 is returned if any bit in the
1306 * range is found set.
1308 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1309 int bits
, int filled
)
1311 struct extent_state
*state
= NULL
;
1312 struct rb_node
*node
;
1314 unsigned long flags
;
1316 spin_lock_irqsave(&tree
->lock
, flags
);
1317 node
= tree_search(tree
, start
);
1318 while (node
&& start
<= end
) {
1319 state
= rb_entry(node
, struct extent_state
, rb_node
);
1321 if (filled
&& state
->start
> start
) {
1326 if (state
->start
> end
)
1329 if (state
->state
& bits
) {
1333 } else if (filled
) {
1337 start
= state
->end
+ 1;
1340 node
= rb_next(node
);
1347 spin_unlock_irqrestore(&tree
->lock
, flags
);
1350 EXPORT_SYMBOL(test_range_bit
);
1353 * helper function to set a given page up to date if all the
1354 * extents in the tree for that page are up to date
1356 static int check_page_uptodate(struct extent_io_tree
*tree
,
1359 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1360 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1361 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1))
1362 SetPageUptodate(page
);
1367 * helper function to unlock a page if all the extents in the tree
1368 * for that page are unlocked
1370 static int check_page_locked(struct extent_io_tree
*tree
,
1373 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1374 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1375 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0))
1381 * helper function to end page writeback if all the extents
1382 * in the tree for that page are done with writeback
1384 static int check_page_writeback(struct extent_io_tree
*tree
,
1387 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1388 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1389 if (!test_range_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 0))
1390 end_page_writeback(page
);
1394 /* lots and lots of room for performance fixes in the end_bio funcs */
1397 * after a writepage IO is done, we need to:
1398 * clear the uptodate bits on error
1399 * clear the writeback bits in the extent tree for this IO
1400 * end_page_writeback if the page has no more pending IO
1402 * Scheduling is not allowed, so the extent state tree is expected
1403 * to have one and only one object corresponding to this IO.
1405 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1407 int uptodate
= err
== 0;
1408 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1409 struct extent_io_tree
*tree
;
1416 struct page
*page
= bvec
->bv_page
;
1417 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1419 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1421 end
= start
+ bvec
->bv_len
- 1;
1423 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1428 if (--bvec
>= bio
->bi_io_vec
)
1429 prefetchw(&bvec
->bv_page
->flags
);
1430 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1431 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
1432 end
, NULL
, uptodate
);
1437 if (!uptodate
&& tree
->ops
&&
1438 tree
->ops
->writepage_io_failed_hook
) {
1439 ret
= tree
->ops
->writepage_io_failed_hook(bio
, page
,
1442 uptodate
= (err
== 0);
1448 clear_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1449 ClearPageUptodate(page
);
1453 clear_extent_writeback(tree
, start
, end
, GFP_ATOMIC
);
1456 end_page_writeback(page
);
1458 check_page_writeback(tree
, page
);
1459 } while (bvec
>= bio
->bi_io_vec
);
1465 * after a readpage IO is done, we need to:
1466 * clear the uptodate bits on error
1467 * set the uptodate bits if things worked
1468 * set the page up to date if all extents in the tree are uptodate
1469 * clear the lock bit in the extent tree
1470 * unlock the page if there are no other extents locked for it
1472 * Scheduling is not allowed, so the extent state tree is expected
1473 * to have one and only one object corresponding to this IO.
1475 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1477 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1478 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1479 struct extent_io_tree
*tree
;
1486 struct page
*page
= bvec
->bv_page
;
1487 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1489 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1491 end
= start
+ bvec
->bv_len
- 1;
1493 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1498 if (--bvec
>= bio
->bi_io_vec
)
1499 prefetchw(&bvec
->bv_page
->flags
);
1501 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1502 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1507 if (!uptodate
&& tree
->ops
&&
1508 tree
->ops
->readpage_io_failed_hook
) {
1509 ret
= tree
->ops
->readpage_io_failed_hook(bio
, page
,
1513 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1519 set_extent_uptodate(tree
, start
, end
,
1521 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1525 SetPageUptodate(page
);
1527 ClearPageUptodate(page
);
1533 check_page_uptodate(tree
, page
);
1535 ClearPageUptodate(page
);
1538 check_page_locked(tree
, page
);
1540 } while (bvec
>= bio
->bi_io_vec
);
1546 * IO done from prepare_write is pretty simple, we just unlock
1547 * the structs in the extent tree when done, and set the uptodate bits
1550 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1552 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1553 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1554 struct extent_io_tree
*tree
;
1559 struct page
*page
= bvec
->bv_page
;
1560 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1562 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1564 end
= start
+ bvec
->bv_len
- 1;
1566 if (--bvec
>= bio
->bi_io_vec
)
1567 prefetchw(&bvec
->bv_page
->flags
);
1570 set_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1572 ClearPageUptodate(page
);
1576 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1578 } while (bvec
>= bio
->bi_io_vec
);
1584 extent_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1589 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1591 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1592 while (!bio
&& (nr_vecs
/= 2))
1593 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1598 bio
->bi_bdev
= bdev
;
1599 bio
->bi_sector
= first_sector
;
1604 static int submit_one_bio(int rw
, struct bio
*bio
, int mirror_num
)
1607 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1608 struct page
*page
= bvec
->bv_page
;
1609 struct extent_io_tree
*tree
= bio
->bi_private
;
1610 struct rb_node
*node
;
1611 struct extent_state
*state
;
1615 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1616 end
= start
+ bvec
->bv_len
- 1;
1618 spin_lock_irq(&tree
->lock
);
1619 node
= __etree_search(tree
, start
, NULL
, NULL
);
1621 state
= rb_entry(node
, struct extent_state
, rb_node
);
1622 while(state
->end
< end
) {
1623 node
= rb_next(node
);
1624 state
= rb_entry(node
, struct extent_state
, rb_node
);
1626 BUG_ON(state
->end
!= end
);
1627 spin_unlock_irq(&tree
->lock
);
1629 bio
->bi_private
= NULL
;
1633 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
1634 tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
1637 submit_bio(rw
, bio
);
1638 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1644 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1645 struct page
*page
, sector_t sector
,
1646 size_t size
, unsigned long offset
,
1647 struct block_device
*bdev
,
1648 struct bio
**bio_ret
,
1649 unsigned long max_pages
,
1650 bio_end_io_t end_io_func
,
1657 if (bio_ret
&& *bio_ret
) {
1659 if (bio
->bi_sector
+ (bio
->bi_size
>> 9) != sector
||
1660 (tree
->ops
&& tree
->ops
->merge_bio_hook
&&
1661 tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
)) ||
1662 bio_add_page(bio
, page
, size
, offset
) < size
) {
1663 ret
= submit_one_bio(rw
, bio
, mirror_num
);
1669 nr
= bio_get_nr_vecs(bdev
);
1670 bio
= extent_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1672 printk("failed to allocate bio nr %d\n", nr
);
1676 bio_add_page(bio
, page
, size
, offset
);
1677 bio
->bi_end_io
= end_io_func
;
1678 bio
->bi_private
= tree
;
1683 ret
= submit_one_bio(rw
, bio
, mirror_num
);
1689 void set_page_extent_mapped(struct page
*page
)
1691 if (!PagePrivate(page
)) {
1692 SetPagePrivate(page
);
1693 page_cache_get(page
);
1694 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1698 void set_page_extent_head(struct page
*page
, unsigned long len
)
1700 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1704 * basic readpage implementation. Locked extent state structs are inserted
1705 * into the tree that are removed when the IO is done (by the end_io
1708 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1710 get_extent_t
*get_extent
,
1711 struct bio
**bio
, int mirror_num
)
1713 struct inode
*inode
= page
->mapping
->host
;
1714 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1715 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1719 u64 last_byte
= i_size_read(inode
);
1723 struct extent_map
*em
;
1724 struct block_device
*bdev
;
1727 size_t page_offset
= 0;
1729 size_t blocksize
= inode
->i_sb
->s_blocksize
;
1731 set_page_extent_mapped(page
);
1734 lock_extent(tree
, start
, end
, GFP_NOFS
);
1736 while (cur
<= end
) {
1737 if (cur
>= last_byte
) {
1739 iosize
= PAGE_CACHE_SIZE
- page_offset
;
1740 userpage
= kmap_atomic(page
, KM_USER0
);
1741 memset(userpage
+ page_offset
, 0, iosize
);
1742 flush_dcache_page(page
);
1743 kunmap_atomic(userpage
, KM_USER0
);
1744 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1746 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1749 em
= get_extent(inode
, page
, page_offset
, cur
,
1751 if (IS_ERR(em
) || !em
) {
1753 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
1756 extent_offset
= cur
- em
->start
;
1757 if (extent_map_end(em
) <= cur
) {
1758 printk("bad mapping em [%Lu %Lu] cur %Lu\n", em
->start
, extent_map_end(em
), cur
);
1760 BUG_ON(extent_map_end(em
) <= cur
);
1762 printk("2bad mapping end %Lu cur %Lu\n", end
, cur
);
1766 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
1767 cur_end
= min(extent_map_end(em
) - 1, end
);
1768 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
1769 sector
= (em
->block_start
+ extent_offset
) >> 9;
1771 block_start
= em
->block_start
;
1772 free_extent_map(em
);
1775 /* we've found a hole, just zero and go on */
1776 if (block_start
== EXTENT_MAP_HOLE
) {
1778 userpage
= kmap_atomic(page
, KM_USER0
);
1779 memset(userpage
+ page_offset
, 0, iosize
);
1780 flush_dcache_page(page
);
1781 kunmap_atomic(userpage
, KM_USER0
);
1783 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1785 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1787 page_offset
+= iosize
;
1790 /* the get_extent function already copied into the page */
1791 if (test_range_bit(tree
, cur
, cur_end
, EXTENT_UPTODATE
, 1)) {
1792 check_page_uptodate(tree
, page
);
1793 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1795 page_offset
+= iosize
;
1798 /* we have an inline extent but it didn't get marked up
1799 * to date. Error out
1801 if (block_start
== EXTENT_MAP_INLINE
) {
1803 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1805 page_offset
+= iosize
;
1810 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
1811 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
1815 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
1817 ret
= submit_extent_page(READ
, tree
, page
,
1818 sector
, iosize
, page_offset
,
1820 end_bio_extent_readpage
, mirror_num
);
1826 page_offset
+= iosize
;
1829 if (!PageError(page
))
1830 SetPageUptodate(page
);
1836 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
1837 get_extent_t
*get_extent
)
1839 struct bio
*bio
= NULL
;
1842 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, 0);
1844 submit_one_bio(READ
, bio
, 0);
1847 EXPORT_SYMBOL(extent_read_full_page
);
1850 * the writepage semantics are similar to regular writepage. extent
1851 * records are inserted to lock ranges in the tree, and as dirty areas
1852 * are found, they are marked writeback. Then the lock bits are removed
1853 * and the end_io handler clears the writeback ranges
1855 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
1858 struct inode
*inode
= page
->mapping
->host
;
1859 struct extent_page_data
*epd
= data
;
1860 struct extent_io_tree
*tree
= epd
->tree
;
1861 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1863 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1867 u64 last_byte
= i_size_read(inode
);
1872 struct extent_map
*em
;
1873 struct block_device
*bdev
;
1876 size_t pg_offset
= 0;
1878 loff_t i_size
= i_size_read(inode
);
1879 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1883 WARN_ON(!PageLocked(page
));
1884 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
1885 if (page
->index
> end_index
||
1886 (page
->index
== end_index
&& !pg_offset
)) {
1887 page
->mapping
->a_ops
->invalidatepage(page
, 0);
1892 if (page
->index
== end_index
) {
1895 userpage
= kmap_atomic(page
, KM_USER0
);
1896 memset(userpage
+ pg_offset
, 0,
1897 PAGE_CACHE_SIZE
- pg_offset
);
1898 kunmap_atomic(userpage
, KM_USER0
);
1899 flush_dcache_page(page
);
1903 set_page_extent_mapped(page
);
1905 delalloc_start
= start
;
1907 while(delalloc_end
< page_end
) {
1908 nr_delalloc
= find_lock_delalloc_range(tree
, &delalloc_start
,
1911 if (nr_delalloc
== 0) {
1912 delalloc_start
= delalloc_end
+ 1;
1915 tree
->ops
->fill_delalloc(inode
, delalloc_start
,
1917 clear_extent_bit(tree
, delalloc_start
,
1919 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1921 delalloc_start
= delalloc_end
+ 1;
1923 lock_extent(tree
, start
, page_end
, GFP_NOFS
);
1924 unlock_start
= start
;
1926 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
1927 ret
= tree
->ops
->writepage_start_hook(page
, start
, page_end
);
1928 if (ret
== -EAGAIN
) {
1929 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
1930 redirty_page_for_writepage(wbc
, page
);
1937 if (test_range_bit(tree
, start
, page_end
, EXTENT_DELALLOC
, 0)) {
1938 printk("found delalloc bits after lock_extent\n");
1941 if (last_byte
<= start
) {
1942 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
1943 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
1944 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
1945 tree
->ops
->writepage_end_io_hook(page
, start
,
1947 unlock_start
= page_end
+ 1;
1951 set_extent_uptodate(tree
, start
, page_end
, GFP_NOFS
);
1952 blocksize
= inode
->i_sb
->s_blocksize
;
1954 while (cur
<= end
) {
1955 if (cur
>= last_byte
) {
1956 clear_extent_dirty(tree
, cur
, page_end
, GFP_NOFS
);
1957 unlock_extent(tree
, unlock_start
, page_end
, GFP_NOFS
);
1958 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
1959 tree
->ops
->writepage_end_io_hook(page
, cur
,
1961 unlock_start
= page_end
+ 1;
1964 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
1966 if (IS_ERR(em
) || !em
) {
1971 extent_offset
= cur
- em
->start
;
1972 BUG_ON(extent_map_end(em
) <= cur
);
1974 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
1975 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
1976 sector
= (em
->block_start
+ extent_offset
) >> 9;
1978 block_start
= em
->block_start
;
1979 free_extent_map(em
);
1982 if (block_start
== EXTENT_MAP_HOLE
||
1983 block_start
== EXTENT_MAP_INLINE
) {
1984 clear_extent_dirty(tree
, cur
,
1985 cur
+ iosize
- 1, GFP_NOFS
);
1987 unlock_extent(tree
, unlock_start
, cur
+ iosize
-1,
1990 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
1991 tree
->ops
->writepage_end_io_hook(page
, cur
,
1995 pg_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 pg_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;
2019 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2020 if (!PageWriteback(page
)) {
2021 printk("warning page %lu not writeback, "
2022 "cur %llu end %llu\n", page
->index
,
2023 (unsigned long long)cur
,
2024 (unsigned long long)end
);
2027 ret
= submit_extent_page(WRITE
, tree
, page
, sector
,
2028 iosize
, pg_offset
, bdev
,
2030 end_bio_extent_writepage
, 0);
2035 pg_offset
+= iosize
;
2040 /* make sure the mapping tag for page dirty gets cleared */
2041 set_page_writeback(page
);
2042 end_page_writeback(page
);
2044 if (unlock_start
<= page_end
)
2045 unlock_extent(tree
, unlock_start
, page_end
, GFP_NOFS
);
2051 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2052 * @mapping: address space structure to write
2053 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2054 * @writepage: function called for each page
2055 * @data: data passed to writepage function
2057 * If a page is already under I/O, write_cache_pages() skips it, even
2058 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2059 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2060 * and msync() need to guarantee that all the data which was dirty at the time
2061 * the call was made get new I/O started against them. If wbc->sync_mode is
2062 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2063 * existing IO to complete.
2065 int extent_write_cache_pages(struct extent_io_tree
*tree
,
2066 struct address_space
*mapping
,
2067 struct writeback_control
*wbc
,
2068 writepage_t writepage
, void *data
)
2070 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
2073 struct pagevec pvec
;
2076 pgoff_t end
; /* Inclusive */
2078 int range_whole
= 0;
2080 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2081 wbc
->encountered_congestion
= 1;
2085 pagevec_init(&pvec
, 0);
2086 if (wbc
->range_cyclic
) {
2087 index
= mapping
->writeback_index
; /* Start from prev offset */
2090 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2091 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2092 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2097 while (!done
&& (index
<= end
) &&
2098 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2099 PAGECACHE_TAG_DIRTY
,
2100 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2104 for (i
= 0; i
< nr_pages
; i
++) {
2105 struct page
*page
= pvec
.pages
[i
];
2108 * At this point we hold neither mapping->tree_lock nor
2109 * lock on the page itself: the page may be truncated or
2110 * invalidated (changing page->mapping to NULL), or even
2111 * swizzled back from swapper_space to tmpfs file
2114 if (tree
->ops
&& tree
->ops
->write_cache_pages_lock_hook
)
2115 tree
->ops
->write_cache_pages_lock_hook(page
);
2119 if (unlikely(page
->mapping
!= mapping
)) {
2124 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2130 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2131 wait_on_page_writeback(page
);
2133 if (PageWriteback(page
) ||
2134 !clear_page_dirty_for_io(page
)) {
2139 ret
= (*writepage
)(page
, wbc
, data
);
2141 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2145 if (ret
|| (--(wbc
->nr_to_write
) <= 0))
2147 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2148 wbc
->encountered_congestion
= 1;
2152 pagevec_release(&pvec
);
2155 if (!scanned
&& !done
) {
2157 * We hit the last page and there is more work to be done: wrap
2158 * back to the start of the file
2164 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2165 mapping
->writeback_index
= index
;
2167 if (wbc
->range_cont
)
2168 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2171 EXPORT_SYMBOL(extent_write_cache_pages
);
2173 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2174 get_extent_t
*get_extent
,
2175 struct writeback_control
*wbc
)
2178 struct address_space
*mapping
= page
->mapping
;
2179 struct extent_page_data epd
= {
2182 .get_extent
= get_extent
,
2184 struct writeback_control wbc_writepages
= {
2186 .sync_mode
= WB_SYNC_NONE
,
2187 .older_than_this
= NULL
,
2189 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2190 .range_end
= (loff_t
)-1,
2194 ret
= __extent_writepage(page
, wbc
, &epd
);
2196 extent_write_cache_pages(tree
, mapping
, &wbc_writepages
,
2197 __extent_writepage
, &epd
);
2199 submit_one_bio(WRITE
, epd
.bio
, 0);
2203 EXPORT_SYMBOL(extent_write_full_page
);
2206 int extent_writepages(struct extent_io_tree
*tree
,
2207 struct address_space
*mapping
,
2208 get_extent_t
*get_extent
,
2209 struct writeback_control
*wbc
)
2212 struct extent_page_data epd
= {
2215 .get_extent
= get_extent
,
2218 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
2219 __extent_writepage
, &epd
);
2221 submit_one_bio(WRITE
, epd
.bio
, 0);
2225 EXPORT_SYMBOL(extent_writepages
);
2227 int extent_readpages(struct extent_io_tree
*tree
,
2228 struct address_space
*mapping
,
2229 struct list_head
*pages
, unsigned nr_pages
,
2230 get_extent_t get_extent
)
2232 struct bio
*bio
= NULL
;
2234 struct pagevec pvec
;
2236 pagevec_init(&pvec
, 0);
2237 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2238 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2240 prefetchw(&page
->flags
);
2241 list_del(&page
->lru
);
2243 * what we want to do here is call add_to_page_cache_lru,
2244 * but that isn't exported, so we reproduce it here
2246 if (!add_to_page_cache(page
, mapping
,
2247 page
->index
, GFP_KERNEL
)) {
2249 /* open coding of lru_cache_add, also not exported */
2250 page_cache_get(page
);
2251 if (!pagevec_add(&pvec
, page
))
2252 __pagevec_lru_add(&pvec
);
2253 __extent_read_full_page(tree
, page
, get_extent
,
2256 page_cache_release(page
);
2258 if (pagevec_count(&pvec
))
2259 __pagevec_lru_add(&pvec
);
2260 BUG_ON(!list_empty(pages
));
2262 submit_one_bio(READ
, bio
, 0);
2265 EXPORT_SYMBOL(extent_readpages
);
2268 * basic invalidatepage code, this waits on any locked or writeback
2269 * ranges corresponding to the page, and then deletes any extent state
2270 * records from the tree
2272 int extent_invalidatepage(struct extent_io_tree
*tree
,
2273 struct page
*page
, unsigned long offset
)
2275 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2276 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2277 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2279 start
+= (offset
+ blocksize
-1) & ~(blocksize
- 1);
2283 lock_extent(tree
, start
, end
, GFP_NOFS
);
2284 wait_on_extent_writeback(tree
, start
, end
);
2285 clear_extent_bit(tree
, start
, end
,
2286 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
,
2290 EXPORT_SYMBOL(extent_invalidatepage
);
2293 * simple commit_write call, set_range_dirty is used to mark both
2294 * the pages and the extent records as dirty
2296 int extent_commit_write(struct extent_io_tree
*tree
,
2297 struct inode
*inode
, struct page
*page
,
2298 unsigned from
, unsigned to
)
2300 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2302 set_page_extent_mapped(page
);
2303 set_page_dirty(page
);
2305 if (pos
> inode
->i_size
) {
2306 i_size_write(inode
, pos
);
2307 mark_inode_dirty(inode
);
2311 EXPORT_SYMBOL(extent_commit_write
);
2313 int extent_prepare_write(struct extent_io_tree
*tree
,
2314 struct inode
*inode
, struct page
*page
,
2315 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2317 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2318 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2320 u64 orig_block_start
;
2323 struct extent_map
*em
;
2324 unsigned blocksize
= 1 << inode
->i_blkbits
;
2325 size_t page_offset
= 0;
2326 size_t block_off_start
;
2327 size_t block_off_end
;
2333 set_page_extent_mapped(page
);
2335 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2336 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2337 orig_block_start
= block_start
;
2339 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2340 while(block_start
<= block_end
) {
2341 em
= get_extent(inode
, page
, page_offset
, block_start
,
2342 block_end
- block_start
+ 1, 1);
2343 if (IS_ERR(em
) || !em
) {
2346 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2347 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2348 block_off_end
= block_off_start
+ blocksize
;
2349 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2351 if (!PageUptodate(page
) && isnew
&&
2352 (block_off_end
> to
|| block_off_start
< from
)) {
2355 kaddr
= kmap_atomic(page
, KM_USER0
);
2356 if (block_off_end
> to
)
2357 memset(kaddr
+ to
, 0, block_off_end
- to
);
2358 if (block_off_start
< from
)
2359 memset(kaddr
+ block_off_start
, 0,
2360 from
- block_off_start
);
2361 flush_dcache_page(page
);
2362 kunmap_atomic(kaddr
, KM_USER0
);
2364 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2365 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2366 !isnew
&& !PageUptodate(page
) &&
2367 (block_off_end
> to
|| block_off_start
< from
) &&
2368 !test_range_bit(tree
, block_start
, cur_end
,
2369 EXTENT_UPTODATE
, 1)) {
2371 u64 extent_offset
= block_start
- em
->start
;
2373 sector
= (em
->block_start
+ extent_offset
) >> 9;
2374 iosize
= (cur_end
- block_start
+ blocksize
) &
2375 ~((u64
)blocksize
- 1);
2377 * we've already got the extent locked, but we
2378 * need to split the state such that our end_bio
2379 * handler can clear the lock.
2381 set_extent_bit(tree
, block_start
,
2382 block_start
+ iosize
- 1,
2383 EXTENT_LOCKED
, 0, NULL
, GFP_NOFS
);
2384 ret
= submit_extent_page(READ
, tree
, page
,
2385 sector
, iosize
, page_offset
, em
->bdev
,
2387 end_bio_extent_preparewrite
, 0);
2389 block_start
= block_start
+ iosize
;
2391 set_extent_uptodate(tree
, block_start
, cur_end
,
2393 unlock_extent(tree
, block_start
, cur_end
, GFP_NOFS
);
2394 block_start
= cur_end
+ 1;
2396 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2397 free_extent_map(em
);
2400 wait_extent_bit(tree
, orig_block_start
,
2401 block_end
, EXTENT_LOCKED
);
2403 check_page_uptodate(tree
, page
);
2405 /* FIXME, zero out newly allocated blocks on error */
2408 EXPORT_SYMBOL(extent_prepare_write
);
2411 * a helper for releasepage, this tests for areas of the page that
2412 * are locked or under IO and drops the related state bits if it is safe
2415 int try_release_extent_state(struct extent_map_tree
*map
,
2416 struct extent_io_tree
*tree
, struct page
*page
,
2419 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2420 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2423 if (test_range_bit(tree
, start
, end
,
2424 EXTENT_IOBITS
| EXTENT_ORDERED
, 0))
2427 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2429 clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
,
2434 EXPORT_SYMBOL(try_release_extent_state
);
2437 * a helper for releasepage. As long as there are no locked extents
2438 * in the range corresponding to the page, both state records and extent
2439 * map records are removed
2441 int try_release_extent_mapping(struct extent_map_tree
*map
,
2442 struct extent_io_tree
*tree
, struct page
*page
,
2445 struct extent_map
*em
;
2446 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2447 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2449 if ((mask
& __GFP_WAIT
) &&
2450 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2452 while (start
<= end
) {
2453 len
= end
- start
+ 1;
2454 spin_lock(&map
->lock
);
2455 em
= lookup_extent_mapping(map
, start
, len
);
2456 if (!em
|| IS_ERR(em
)) {
2457 spin_unlock(&map
->lock
);
2460 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
2461 em
->start
!= start
) {
2462 spin_unlock(&map
->lock
);
2463 free_extent_map(em
);
2466 if (!test_range_bit(tree
, em
->start
,
2467 extent_map_end(em
) - 1,
2468 EXTENT_LOCKED
, 0)) {
2469 remove_extent_mapping(map
, em
);
2470 /* once for the rb tree */
2471 free_extent_map(em
);
2473 start
= extent_map_end(em
);
2474 spin_unlock(&map
->lock
);
2477 free_extent_map(em
);
2480 return try_release_extent_state(map
, tree
, page
, mask
);
2482 EXPORT_SYMBOL(try_release_extent_mapping
);
2484 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2485 get_extent_t
*get_extent
)
2487 struct inode
*inode
= mapping
->host
;
2488 u64 start
= iblock
<< inode
->i_blkbits
;
2489 sector_t sector
= 0;
2490 struct extent_map
*em
;
2492 em
= get_extent(inode
, NULL
, 0, start
, (1 << inode
->i_blkbits
), 0);
2493 if (!em
|| IS_ERR(em
))
2496 if (em
->block_start
== EXTENT_MAP_INLINE
||
2497 em
->block_start
== EXTENT_MAP_HOLE
)
2500 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2502 free_extent_map(em
);
2506 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
2510 struct address_space
*mapping
;
2513 return eb
->first_page
;
2514 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
2515 mapping
= eb
->first_page
->mapping
;
2520 * extent_buffer_page is only called after pinning the page
2521 * by increasing the reference count. So we know the page must
2522 * be in the radix tree.
2525 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
2531 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
2533 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
2534 (start
>> PAGE_CACHE_SHIFT
);
2537 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
2542 struct extent_buffer
*eb
= NULL
;
2544 unsigned long flags
;
2547 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
2550 mutex_init(&eb
->mutex
);
2552 spin_lock_irqsave(&leak_lock
, flags
);
2553 list_add(&eb
->leak_list
, &buffers
);
2554 spin_unlock_irqrestore(&leak_lock
, flags
);
2556 atomic_set(&eb
->refs
, 1);
2561 static void __free_extent_buffer(struct extent_buffer
*eb
)
2564 unsigned long flags
;
2565 spin_lock_irqsave(&leak_lock
, flags
);
2566 list_del(&eb
->leak_list
);
2567 spin_unlock_irqrestore(&leak_lock
, flags
);
2569 kmem_cache_free(extent_buffer_cache
, eb
);
2572 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
2573 u64 start
, unsigned long len
,
2577 unsigned long num_pages
= num_extent_pages(start
, len
);
2579 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2580 struct extent_buffer
*eb
;
2581 struct extent_buffer
*exists
= NULL
;
2583 struct address_space
*mapping
= tree
->mapping
;
2586 spin_lock(&tree
->buffer_lock
);
2587 eb
= buffer_search(tree
, start
);
2589 atomic_inc(&eb
->refs
);
2590 spin_unlock(&tree
->buffer_lock
);
2591 mark_page_accessed(eb
->first_page
);
2594 spin_unlock(&tree
->buffer_lock
);
2596 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2601 eb
->first_page
= page0
;
2604 page_cache_get(page0
);
2605 mark_page_accessed(page0
);
2606 set_page_extent_mapped(page0
);
2607 set_page_extent_head(page0
, len
);
2608 uptodate
= PageUptodate(page0
);
2612 for (; i
< num_pages
; i
++, index
++) {
2613 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
2618 set_page_extent_mapped(p
);
2619 mark_page_accessed(p
);
2622 set_page_extent_head(p
, len
);
2624 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2626 if (!PageUptodate(p
))
2631 eb
->flags
|= EXTENT_UPTODATE
;
2632 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2634 spin_lock(&tree
->buffer_lock
);
2635 exists
= buffer_tree_insert(tree
, start
, &eb
->rb_node
);
2637 /* add one reference for the caller */
2638 atomic_inc(&exists
->refs
);
2639 spin_unlock(&tree
->buffer_lock
);
2642 spin_unlock(&tree
->buffer_lock
);
2644 /* add one reference for the tree */
2645 atomic_inc(&eb
->refs
);
2649 if (!atomic_dec_and_test(&eb
->refs
))
2651 for (index
= 1; index
< i
; index
++)
2652 page_cache_release(extent_buffer_page(eb
, index
));
2653 page_cache_release(extent_buffer_page(eb
, 0));
2654 __free_extent_buffer(eb
);
2657 EXPORT_SYMBOL(alloc_extent_buffer
);
2659 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
2660 u64 start
, unsigned long len
,
2663 struct extent_buffer
*eb
;
2665 spin_lock(&tree
->buffer_lock
);
2666 eb
= buffer_search(tree
, start
);
2668 atomic_inc(&eb
->refs
);
2669 spin_unlock(&tree
->buffer_lock
);
2672 mark_page_accessed(eb
->first_page
);
2676 EXPORT_SYMBOL(find_extent_buffer
);
2678 void free_extent_buffer(struct extent_buffer
*eb
)
2683 if (!atomic_dec_and_test(&eb
->refs
))
2688 EXPORT_SYMBOL(free_extent_buffer
);
2690 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
2691 struct extent_buffer
*eb
)
2695 unsigned long num_pages
;
2698 u64 start
= eb
->start
;
2699 u64 end
= start
+ eb
->len
- 1;
2701 set
= clear_extent_dirty(tree
, start
, end
, GFP_NOFS
);
2702 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2704 for (i
= 0; i
< num_pages
; i
++) {
2705 page
= extent_buffer_page(eb
, i
);
2708 set_page_extent_head(page
, eb
->len
);
2710 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2713 * if we're on the last page or the first page and the
2714 * block isn't aligned on a page boundary, do extra checks
2715 * to make sure we don't clean page that is partially dirty
2717 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2718 ((i
== num_pages
- 1) &&
2719 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2720 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2721 end
= start
+ PAGE_CACHE_SIZE
- 1;
2722 if (test_range_bit(tree
, start
, end
,
2728 clear_page_dirty_for_io(page
);
2729 spin_lock_irq(&page
->mapping
->tree_lock
);
2730 if (!PageDirty(page
)) {
2731 radix_tree_tag_clear(&page
->mapping
->page_tree
,
2733 PAGECACHE_TAG_DIRTY
);
2735 spin_unlock_irq(&page
->mapping
->tree_lock
);
2740 EXPORT_SYMBOL(clear_extent_buffer_dirty
);
2742 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
2743 struct extent_buffer
*eb
)
2745 return wait_on_extent_writeback(tree
, eb
->start
,
2746 eb
->start
+ eb
->len
- 1);
2748 EXPORT_SYMBOL(wait_on_extent_buffer_writeback
);
2750 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
2751 struct extent_buffer
*eb
)
2754 unsigned long num_pages
;
2756 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2757 for (i
= 0; i
< num_pages
; i
++) {
2758 struct page
*page
= extent_buffer_page(eb
, i
);
2759 /* writepage may need to do something special for the
2760 * first page, we have to make sure page->private is
2761 * properly set. releasepage may drop page->private
2762 * on us if the page isn't already dirty.
2766 set_page_extent_head(page
, eb
->len
);
2767 } else if (PagePrivate(page
) &&
2768 page
->private != EXTENT_PAGE_PRIVATE
) {
2769 set_page_extent_mapped(page
);
2771 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
2772 set_extent_dirty(tree
, page_offset(page
),
2773 page_offset(page
) + PAGE_CACHE_SIZE
-1,
2779 EXPORT_SYMBOL(set_extent_buffer_dirty
);
2781 int clear_extent_buffer_uptodate(struct extent_io_tree
*tree
,
2782 struct extent_buffer
*eb
)
2786 unsigned long num_pages
;
2788 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2789 eb
->flags
&= ~EXTENT_UPTODATE
;
2791 clear_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2793 for (i
= 0; i
< num_pages
; i
++) {
2794 page
= extent_buffer_page(eb
, i
);
2796 ClearPageUptodate(page
);
2801 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
2802 struct extent_buffer
*eb
)
2806 unsigned long num_pages
;
2808 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2810 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2812 for (i
= 0; i
< num_pages
; i
++) {
2813 page
= extent_buffer_page(eb
, i
);
2814 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2815 ((i
== num_pages
- 1) &&
2816 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2817 check_page_uptodate(tree
, page
);
2820 SetPageUptodate(page
);
2824 EXPORT_SYMBOL(set_extent_buffer_uptodate
);
2826 int extent_range_uptodate(struct extent_io_tree
*tree
,
2831 int pg_uptodate
= 1;
2833 unsigned long index
;
2835 ret
= test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1);
2838 while(start
<= end
) {
2839 index
= start
>> PAGE_CACHE_SHIFT
;
2840 page
= find_get_page(tree
->mapping
, index
);
2841 uptodate
= PageUptodate(page
);
2842 page_cache_release(page
);
2847 start
+= PAGE_CACHE_SIZE
;
2852 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
2853 struct extent_buffer
*eb
)
2856 unsigned long num_pages
;
2859 int pg_uptodate
= 1;
2861 if (eb
->flags
& EXTENT_UPTODATE
)
2864 ret
= test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2865 EXTENT_UPTODATE
, 1);
2869 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2870 for (i
= 0; i
< num_pages
; i
++) {
2871 page
= extent_buffer_page(eb
, i
);
2872 if (!PageUptodate(page
)) {
2879 EXPORT_SYMBOL(extent_buffer_uptodate
);
2881 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
2882 struct extent_buffer
*eb
,
2883 u64 start
, int wait
,
2884 get_extent_t
*get_extent
, int mirror_num
)
2887 unsigned long start_i
;
2891 int locked_pages
= 0;
2892 int all_uptodate
= 1;
2893 int inc_all_pages
= 0;
2894 unsigned long num_pages
;
2895 struct bio
*bio
= NULL
;
2897 if (eb
->flags
& EXTENT_UPTODATE
)
2900 if (test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2901 EXTENT_UPTODATE
, 1)) {
2906 WARN_ON(start
< eb
->start
);
2907 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
2908 (eb
->start
>> PAGE_CACHE_SHIFT
);
2913 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2914 for (i
= start_i
; i
< num_pages
; i
++) {
2915 page
= extent_buffer_page(eb
, i
);
2917 if (!trylock_page(page
))
2923 if (!PageUptodate(page
)) {
2929 eb
->flags
|= EXTENT_UPTODATE
;
2931 printk("all up to date but ret is %d\n", ret
);
2936 for (i
= start_i
; i
< num_pages
; i
++) {
2937 page
= extent_buffer_page(eb
, i
);
2939 page_cache_get(page
);
2940 if (!PageUptodate(page
)) {
2943 ClearPageError(page
);
2944 err
= __extent_read_full_page(tree
, page
,
2949 printk("err %d from __extent_read_full_page\n", ret
);
2957 submit_one_bio(READ
, bio
, mirror_num
);
2961 printk("ret %d wait %d returning\n", ret
, wait
);
2964 for (i
= start_i
; i
< num_pages
; i
++) {
2965 page
= extent_buffer_page(eb
, i
);
2966 wait_on_page_locked(page
);
2967 if (!PageUptodate(page
)) {
2968 printk("page not uptodate after wait_on_page_locked\n");
2973 eb
->flags
|= EXTENT_UPTODATE
;
2978 while(locked_pages
> 0) {
2979 page
= extent_buffer_page(eb
, i
);
2986 EXPORT_SYMBOL(read_extent_buffer_pages
);
2988 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
2989 unsigned long start
,
2996 char *dst
= (char *)dstv
;
2997 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
2998 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
);
3008 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3009 kaddr
= kmap_atomic(page
, KM_USER1
);
3010 memcpy(dst
, kaddr
+ offset
, cur
);
3011 kunmap_atomic(kaddr
, KM_USER1
);
3019 EXPORT_SYMBOL(read_extent_buffer
);
3021 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3022 unsigned long min_len
, char **token
, char **map
,
3023 unsigned long *map_start
,
3024 unsigned long *map_len
, int km
)
3026 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
3029 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3030 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3031 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
3038 offset
= start_offset
;
3042 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
3044 if (start
+ min_len
> eb
->len
) {
3045 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb
->start
, eb
->len
, start
, min_len
);
3049 p
= extent_buffer_page(eb
, i
);
3050 kaddr
= kmap_atomic(p
, km
);
3052 *map
= kaddr
+ offset
;
3053 *map_len
= PAGE_CACHE_SIZE
- offset
;
3056 EXPORT_SYMBOL(map_private_extent_buffer
);
3058 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3059 unsigned long min_len
,
3060 char **token
, char **map
,
3061 unsigned long *map_start
,
3062 unsigned long *map_len
, int km
)
3066 if (eb
->map_token
) {
3067 unmap_extent_buffer(eb
, eb
->map_token
, km
);
3068 eb
->map_token
= NULL
;
3071 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
3072 map_start
, map_len
, km
);
3074 eb
->map_token
= *token
;
3076 eb
->map_start
= *map_start
;
3077 eb
->map_len
= *map_len
;
3081 EXPORT_SYMBOL(map_extent_buffer
);
3083 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
3085 kunmap_atomic(token
, km
);
3087 EXPORT_SYMBOL(unmap_extent_buffer
);
3089 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3090 unsigned long start
,
3097 char *ptr
= (char *)ptrv
;
3098 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3099 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3102 WARN_ON(start
> eb
->len
);
3103 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3105 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3108 page
= extent_buffer_page(eb
, i
);
3110 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3112 kaddr
= kmap_atomic(page
, KM_USER0
);
3113 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3114 kunmap_atomic(kaddr
, KM_USER0
);
3125 EXPORT_SYMBOL(memcmp_extent_buffer
);
3127 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3128 unsigned long start
, unsigned long len
)
3134 char *src
= (char *)srcv
;
3135 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3136 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3138 WARN_ON(start
> eb
->len
);
3139 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3141 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3144 page
= extent_buffer_page(eb
, i
);
3145 WARN_ON(!PageUptodate(page
));
3147 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3148 kaddr
= kmap_atomic(page
, KM_USER1
);
3149 memcpy(kaddr
+ offset
, src
, cur
);
3150 kunmap_atomic(kaddr
, KM_USER1
);
3158 EXPORT_SYMBOL(write_extent_buffer
);
3160 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3161 unsigned long start
, unsigned long len
)
3167 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3168 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3170 WARN_ON(start
> eb
->len
);
3171 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3173 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3176 page
= extent_buffer_page(eb
, i
);
3177 WARN_ON(!PageUptodate(page
));
3179 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3180 kaddr
= kmap_atomic(page
, KM_USER0
);
3181 memset(kaddr
+ offset
, c
, cur
);
3182 kunmap_atomic(kaddr
, KM_USER0
);
3189 EXPORT_SYMBOL(memset_extent_buffer
);
3191 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3192 unsigned long dst_offset
, unsigned long src_offset
,
3195 u64 dst_len
= dst
->len
;
3200 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3201 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3203 WARN_ON(src
->len
!= dst_len
);
3205 offset
= (start_offset
+ dst_offset
) &
3206 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3209 page
= extent_buffer_page(dst
, i
);
3210 WARN_ON(!PageUptodate(page
));
3212 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3214 kaddr
= kmap_atomic(page
, KM_USER0
);
3215 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3216 kunmap_atomic(kaddr
, KM_USER0
);
3224 EXPORT_SYMBOL(copy_extent_buffer
);
3226 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3227 unsigned long dst_off
, unsigned long src_off
,
3230 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3231 if (dst_page
== src_page
) {
3232 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3234 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3235 char *p
= dst_kaddr
+ dst_off
+ len
;
3236 char *s
= src_kaddr
+ src_off
+ len
;
3241 kunmap_atomic(src_kaddr
, KM_USER1
);
3243 kunmap_atomic(dst_kaddr
, KM_USER0
);
3246 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3247 unsigned long dst_off
, unsigned long src_off
,
3250 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3253 if (dst_page
!= src_page
)
3254 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3256 src_kaddr
= dst_kaddr
;
3258 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3259 kunmap_atomic(dst_kaddr
, KM_USER0
);
3260 if (dst_page
!= src_page
)
3261 kunmap_atomic(src_kaddr
, KM_USER1
);
3264 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3265 unsigned long src_offset
, unsigned long len
)
3268 size_t dst_off_in_page
;
3269 size_t src_off_in_page
;
3270 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3271 unsigned long dst_i
;
3272 unsigned long src_i
;
3274 if (src_offset
+ len
> dst
->len
) {
3275 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3276 src_offset
, len
, dst
->len
);
3279 if (dst_offset
+ len
> dst
->len
) {
3280 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3281 dst_offset
, len
, dst
->len
);
3286 dst_off_in_page
= (start_offset
+ dst_offset
) &
3287 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3288 src_off_in_page
= (start_offset
+ src_offset
) &
3289 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3291 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3292 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3294 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3296 cur
= min_t(unsigned long, cur
,
3297 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3299 copy_pages(extent_buffer_page(dst
, dst_i
),
3300 extent_buffer_page(dst
, src_i
),
3301 dst_off_in_page
, src_off_in_page
, cur
);
3308 EXPORT_SYMBOL(memcpy_extent_buffer
);
3310 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3311 unsigned long src_offset
, unsigned long len
)
3314 size_t dst_off_in_page
;
3315 size_t src_off_in_page
;
3316 unsigned long dst_end
= dst_offset
+ len
- 1;
3317 unsigned long src_end
= src_offset
+ len
- 1;
3318 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3319 unsigned long dst_i
;
3320 unsigned long src_i
;
3322 if (src_offset
+ len
> dst
->len
) {
3323 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3324 src_offset
, len
, dst
->len
);
3327 if (dst_offset
+ len
> dst
->len
) {
3328 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3329 dst_offset
, len
, dst
->len
);
3332 if (dst_offset
< src_offset
) {
3333 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
3337 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
3338 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
3340 dst_off_in_page
= (start_offset
+ dst_end
) &
3341 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3342 src_off_in_page
= (start_offset
+ src_end
) &
3343 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3345 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
3346 cur
= min(cur
, dst_off_in_page
+ 1);
3347 move_pages(extent_buffer_page(dst
, dst_i
),
3348 extent_buffer_page(dst
, src_i
),
3349 dst_off_in_page
- cur
+ 1,
3350 src_off_in_page
- cur
+ 1, cur
);
3357 EXPORT_SYMBOL(memmove_extent_buffer
);
3359 int try_release_extent_buffer(struct extent_io_tree
*tree
, struct page
*page
)
3361 u64 start
= page_offset(page
);
3362 struct extent_buffer
*eb
;
3365 unsigned long num_pages
;
3367 spin_lock(&tree
->buffer_lock
);
3368 eb
= buffer_search(tree
, start
);
3372 if (atomic_read(&eb
->refs
) > 1) {
3376 /* at this point we can safely release the extent buffer */
3377 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3378 for (i
= 0; i
< num_pages
; i
++)
3379 page_cache_release(extent_buffer_page(eb
, i
));
3380 rb_erase(&eb
->rb_node
, &tree
->buffer
);
3381 __free_extent_buffer(eb
);
3383 spin_unlock(&tree
->buffer_lock
);
3386 EXPORT_SYMBOL(try_release_extent_buffer
);