1 #include <linux/bitops.h>
2 #include <linux/slab.h>
6 #include <linux/pagemap.h>
7 #include <linux/page-flags.h>
8 #include <linux/module.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/version.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
18 /* temporary define until extent_map moves out of btrfs */
19 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
20 unsigned long extra_flags
,
21 void (*ctor
)(void *, struct kmem_cache
*,
24 static struct kmem_cache
*extent_state_cache
;
25 static struct kmem_cache
*extent_buffer_cache
;
27 static LIST_HEAD(buffers
);
28 static LIST_HEAD(states
);
29 static spinlock_t leak_lock
= SPIN_LOCK_UNLOCKED
;
31 #define BUFFER_LRU_MAX 64
36 struct rb_node rb_node
;
39 struct extent_page_data
{
41 struct extent_io_tree
*tree
;
42 get_extent_t
*get_extent
;
45 int __init
extent_io_init(void)
47 extent_state_cache
= btrfs_cache_create("extent_state",
48 sizeof(struct extent_state
), 0,
50 if (!extent_state_cache
)
53 extent_buffer_cache
= btrfs_cache_create("extent_buffers",
54 sizeof(struct extent_buffer
), 0,
56 if (!extent_buffer_cache
)
57 goto free_state_cache
;
61 kmem_cache_destroy(extent_state_cache
);
65 void extent_io_exit(void)
67 struct extent_state
*state
;
68 struct extent_buffer
*eb
;
70 while (!list_empty(&states
)) {
71 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
72 printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state
->start
, state
->end
, state
->state
, state
->tree
, atomic_read(&state
->refs
));
73 list_del(&state
->leak_list
);
74 kmem_cache_free(extent_state_cache
, state
);
78 while (!list_empty(&buffers
)) {
79 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
80 printk("buffer leak start %Lu len %lu refs %d\n", eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
81 list_del(&eb
->leak_list
);
82 kmem_cache_free(extent_buffer_cache
, eb
);
84 if (extent_state_cache
)
85 kmem_cache_destroy(extent_state_cache
);
86 if (extent_buffer_cache
)
87 kmem_cache_destroy(extent_buffer_cache
);
90 void extent_io_tree_init(struct extent_io_tree
*tree
,
91 struct address_space
*mapping
, gfp_t mask
)
93 tree
->state
.rb_node
= NULL
;
95 tree
->dirty_bytes
= 0;
96 spin_lock_init(&tree
->lock
);
97 spin_lock_init(&tree
->lru_lock
);
98 tree
->mapping
= mapping
;
99 INIT_LIST_HEAD(&tree
->buffer_lru
);
103 EXPORT_SYMBOL(extent_io_tree_init
);
105 void extent_io_tree_empty_lru(struct extent_io_tree
*tree
)
107 struct extent_buffer
*eb
;
108 while(!list_empty(&tree
->buffer_lru
)) {
109 eb
= list_entry(tree
->buffer_lru
.next
, struct extent_buffer
,
111 list_del_init(&eb
->lru
);
112 free_extent_buffer(eb
);
115 EXPORT_SYMBOL(extent_io_tree_empty_lru
);
117 struct extent_state
*alloc_extent_state(gfp_t mask
)
119 struct extent_state
*state
;
122 state
= kmem_cache_alloc(extent_state_cache
, mask
);
128 spin_lock_irqsave(&leak_lock
, flags
);
129 list_add(&state
->leak_list
, &states
);
130 spin_unlock_irqrestore(&leak_lock
, flags
);
132 atomic_set(&state
->refs
, 1);
133 init_waitqueue_head(&state
->wq
);
136 EXPORT_SYMBOL(alloc_extent_state
);
138 void free_extent_state(struct extent_state
*state
)
142 if (atomic_dec_and_test(&state
->refs
)) {
144 WARN_ON(state
->tree
);
145 spin_lock_irqsave(&leak_lock
, flags
);
146 list_del(&state
->leak_list
);
147 spin_unlock_irqrestore(&leak_lock
, flags
);
148 kmem_cache_free(extent_state_cache
, state
);
151 EXPORT_SYMBOL(free_extent_state
);
153 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
154 struct rb_node
*node
)
156 struct rb_node
** p
= &root
->rb_node
;
157 struct rb_node
* parent
= NULL
;
158 struct tree_entry
*entry
;
162 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
164 if (offset
< entry
->start
)
166 else if (offset
> entry
->end
)
172 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
173 rb_link_node(node
, parent
, p
);
174 rb_insert_color(node
, root
);
178 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
179 struct rb_node
**prev_ret
,
180 struct rb_node
**next_ret
)
182 struct rb_root
*root
= &tree
->state
;
183 struct rb_node
* n
= root
->rb_node
;
184 struct rb_node
*prev
= NULL
;
185 struct rb_node
*orig_prev
= NULL
;
186 struct tree_entry
*entry
;
187 struct tree_entry
*prev_entry
= NULL
;
190 struct extent_state
*state
;
192 if (state
->start
<= offset
&& offset
<= state
->end
)
193 return &tree
->last
->rb_node
;
196 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
200 if (offset
< entry
->start
)
202 else if (offset
> entry
->end
)
205 tree
->last
= rb_entry(n
, struct extent_state
, rb_node
);
212 while(prev
&& offset
> prev_entry
->end
) {
213 prev
= rb_next(prev
);
214 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
221 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
222 while(prev
&& offset
< prev_entry
->start
) {
223 prev
= rb_prev(prev
);
224 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
231 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
234 struct rb_node
*prev
= NULL
;
237 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
240 tree
->last
= rb_entry(prev
, struct extent_state
,
249 * utility function to look for merge candidates inside a given range.
250 * Any extents with matching state are merged together into a single
251 * extent in the tree. Extents with EXTENT_IO in their state field
252 * are not merged because the end_io handlers need to be able to do
253 * operations on them without sleeping (or doing allocations/splits).
255 * This should be called with the tree lock held.
257 static int merge_state(struct extent_io_tree
*tree
,
258 struct extent_state
*state
)
260 struct extent_state
*other
;
261 struct rb_node
*other_node
;
263 if (state
->state
& EXTENT_IOBITS
)
266 other_node
= rb_prev(&state
->rb_node
);
268 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
269 if (other
->end
== state
->start
- 1 &&
270 other
->state
== state
->state
) {
271 state
->start
= other
->start
;
273 if (tree
->last
== other
)
275 rb_erase(&other
->rb_node
, &tree
->state
);
276 free_extent_state(other
);
279 other_node
= rb_next(&state
->rb_node
);
281 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
282 if (other
->start
== state
->end
+ 1 &&
283 other
->state
== state
->state
) {
284 other
->start
= state
->start
;
286 if (tree
->last
== state
)
288 rb_erase(&state
->rb_node
, &tree
->state
);
289 free_extent_state(state
);
295 static void set_state_cb(struct extent_io_tree
*tree
,
296 struct extent_state
*state
,
299 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
300 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
->start
,
301 state
->end
, state
->state
, bits
);
305 static void clear_state_cb(struct extent_io_tree
*tree
,
306 struct extent_state
*state
,
309 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
310 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
->start
,
311 state
->end
, state
->state
, bits
);
316 * insert an extent_state struct into the tree. 'bits' are set on the
317 * struct before it is inserted.
319 * This may return -EEXIST if the extent is already there, in which case the
320 * state struct is freed.
322 * The tree lock is not taken internally. This is a utility function and
323 * probably isn't what you want to call (see set/clear_extent_bit).
325 static int insert_state(struct extent_io_tree
*tree
,
326 struct extent_state
*state
, u64 start
, u64 end
,
329 struct rb_node
*node
;
332 printk("end < start %Lu %Lu\n", end
, start
);
335 if (bits
& EXTENT_DIRTY
)
336 tree
->dirty_bytes
+= end
- start
+ 1;
337 set_state_cb(tree
, state
, bits
);
338 state
->state
|= bits
;
339 state
->start
= start
;
341 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
343 struct extent_state
*found
;
344 found
= rb_entry(node
, struct extent_state
, rb_node
);
345 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, start
, end
);
346 free_extent_state(state
);
351 merge_state(tree
, state
);
356 * split a given extent state struct in two, inserting the preallocated
357 * struct 'prealloc' as the newly created second half. 'split' indicates an
358 * offset inside 'orig' where it should be split.
361 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
362 * are two extent state structs in the tree:
363 * prealloc: [orig->start, split - 1]
364 * orig: [ split, orig->end ]
366 * The tree locks are not taken by this function. They need to be held
369 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
370 struct extent_state
*prealloc
, u64 split
)
372 struct rb_node
*node
;
373 prealloc
->start
= orig
->start
;
374 prealloc
->end
= split
- 1;
375 prealloc
->state
= orig
->state
;
378 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
380 struct extent_state
*found
;
381 found
= rb_entry(node
, struct extent_state
, rb_node
);
382 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found
->start
, found
->end
, prealloc
->start
, prealloc
->end
);
383 free_extent_state(prealloc
);
386 prealloc
->tree
= tree
;
391 * utility function to clear some bits in an extent state struct.
392 * it will optionally wake up any one waiting on this state (wake == 1), or
393 * forcibly remove the state from the tree (delete == 1).
395 * If no bits are set on the state struct after clearing things, the
396 * struct is freed and removed from the tree
398 static int clear_state_bit(struct extent_io_tree
*tree
,
399 struct extent_state
*state
, int bits
, int wake
,
402 int ret
= state
->state
& bits
;
404 if ((bits
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
405 u64 range
= state
->end
- state
->start
+ 1;
406 WARN_ON(range
> tree
->dirty_bytes
);
407 tree
->dirty_bytes
-= range
;
409 clear_state_cb(tree
, state
, bits
);
410 state
->state
&= ~bits
;
413 if (delete || state
->state
== 0) {
415 clear_state_cb(tree
, state
, state
->state
);
416 if (tree
->last
== state
) {
417 tree
->last
= extent_state_next(state
);
419 rb_erase(&state
->rb_node
, &tree
->state
);
421 free_extent_state(state
);
426 merge_state(tree
, state
);
432 * clear some bits on a range in the tree. This may require splitting
433 * or inserting elements in the tree, so the gfp mask is used to
434 * indicate which allocations or sleeping are allowed.
436 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
437 * the given range from the tree regardless of state (ie for truncate).
439 * the range [start, end] is inclusive.
441 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
442 * bits were already set, or zero if none of the bits were already set.
444 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
445 int bits
, int wake
, int delete, gfp_t mask
)
447 struct extent_state
*state
;
448 struct extent_state
*prealloc
= NULL
;
449 struct rb_node
*node
;
455 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
456 prealloc
= alloc_extent_state(mask
);
461 spin_lock_irqsave(&tree
->lock
, flags
);
463 * this search will find the extents that end after
466 node
= tree_search(tree
, start
);
469 state
= rb_entry(node
, struct extent_state
, rb_node
);
470 if (state
->start
> end
)
472 WARN_ON(state
->end
< start
);
475 * | ---- desired range ---- |
477 * | ------------- state -------------- |
479 * We need to split the extent we found, and may flip
480 * bits on second half.
482 * If the extent we found extends past our range, we
483 * just split and search again. It'll get split again
484 * the next time though.
486 * If the extent we found is inside our range, we clear
487 * the desired bit on it.
490 if (state
->start
< start
) {
492 prealloc
= alloc_extent_state(GFP_ATOMIC
);
493 err
= split_state(tree
, state
, prealloc
, start
);
494 BUG_ON(err
== -EEXIST
);
498 if (state
->end
<= end
) {
499 start
= state
->end
+ 1;
500 set
|= clear_state_bit(tree
, state
, bits
,
503 start
= state
->start
;
508 * | ---- desired range ---- |
510 * We need to split the extent, and clear the bit
513 if (state
->start
<= end
&& state
->end
> end
) {
515 prealloc
= alloc_extent_state(GFP_ATOMIC
);
516 err
= split_state(tree
, state
, prealloc
, end
+ 1);
517 BUG_ON(err
== -EEXIST
);
521 set
|= clear_state_bit(tree
, prealloc
, bits
,
527 start
= state
->end
+ 1;
528 set
|= clear_state_bit(tree
, state
, bits
, wake
, delete);
532 spin_unlock_irqrestore(&tree
->lock
, flags
);
534 free_extent_state(prealloc
);
541 spin_unlock_irqrestore(&tree
->lock
, flags
);
542 if (mask
& __GFP_WAIT
)
546 EXPORT_SYMBOL(clear_extent_bit
);
548 static int wait_on_state(struct extent_io_tree
*tree
,
549 struct extent_state
*state
)
552 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
553 spin_unlock_irq(&tree
->lock
);
555 spin_lock_irq(&tree
->lock
);
556 finish_wait(&state
->wq
, &wait
);
561 * waits for one or more bits to clear on a range in the state tree.
562 * The range [start, end] is inclusive.
563 * The tree lock is taken by this function
565 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
567 struct extent_state
*state
;
568 struct rb_node
*node
;
570 spin_lock_irq(&tree
->lock
);
574 * this search will find all the extents that end after
577 node
= tree_search(tree
, start
);
581 state
= rb_entry(node
, struct extent_state
, rb_node
);
583 if (state
->start
> end
)
586 if (state
->state
& bits
) {
587 start
= state
->start
;
588 atomic_inc(&state
->refs
);
589 wait_on_state(tree
, state
);
590 free_extent_state(state
);
593 start
= state
->end
+ 1;
598 if (need_resched()) {
599 spin_unlock_irq(&tree
->lock
);
601 spin_lock_irq(&tree
->lock
);
605 spin_unlock_irq(&tree
->lock
);
608 EXPORT_SYMBOL(wait_extent_bit
);
610 static void set_state_bits(struct extent_io_tree
*tree
,
611 struct extent_state
*state
,
614 if ((bits
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
615 u64 range
= state
->end
- state
->start
+ 1;
616 tree
->dirty_bytes
+= range
;
618 set_state_cb(tree
, state
, bits
);
619 state
->state
|= bits
;
623 * set some bits on a range in the tree. This may require allocations
624 * or sleeping, so the gfp mask is used to indicate what is allowed.
626 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
627 * range already has the desired bits set. The start of the existing
628 * range is returned in failed_start in this case.
630 * [start, end] is inclusive
631 * This takes the tree lock.
633 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
,
634 int exclusive
, u64
*failed_start
, gfp_t mask
)
636 struct extent_state
*state
;
637 struct extent_state
*prealloc
= NULL
;
638 struct rb_node
*node
;
645 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
646 prealloc
= alloc_extent_state(mask
);
651 spin_lock_irqsave(&tree
->lock
, flags
);
653 * this search will find all the extents that end after
656 node
= tree_search(tree
, start
);
658 err
= insert_state(tree
, prealloc
, start
, end
, bits
);
660 BUG_ON(err
== -EEXIST
);
664 state
= rb_entry(node
, struct extent_state
, rb_node
);
665 last_start
= state
->start
;
666 last_end
= state
->end
;
669 * | ---- desired range ---- |
672 * Just lock what we found and keep going
674 if (state
->start
== start
&& state
->end
<= end
) {
675 set
= state
->state
& bits
;
676 if (set
&& exclusive
) {
677 *failed_start
= state
->start
;
681 set_state_bits(tree
, state
, bits
);
682 start
= state
->end
+ 1;
683 merge_state(tree
, state
);
688 * | ---- desired range ---- |
691 * | ------------- state -------------- |
693 * We need to split the extent we found, and may flip bits on
696 * If the extent we found extends past our
697 * range, we just split and search again. It'll get split
698 * again the next time though.
700 * If the extent we found is inside our range, we set the
703 if (state
->start
< start
) {
704 set
= state
->state
& bits
;
705 if (exclusive
&& set
) {
706 *failed_start
= start
;
710 err
= split_state(tree
, state
, prealloc
, start
);
711 BUG_ON(err
== -EEXIST
);
715 if (state
->end
<= end
) {
716 set_state_bits(tree
, state
, bits
);
717 start
= state
->end
+ 1;
718 merge_state(tree
, state
);
720 start
= state
->start
;
725 * | ---- desired range ---- |
726 * | state | or | state |
728 * There's a hole, we need to insert something in it and
729 * ignore the extent we found.
731 if (state
->start
> start
) {
733 if (end
< last_start
)
736 this_end
= last_start
-1;
737 err
= insert_state(tree
, prealloc
, start
, this_end
,
740 BUG_ON(err
== -EEXIST
);
743 start
= this_end
+ 1;
747 * | ---- desired range ---- |
749 * We need to split the extent, and set the bit
752 if (state
->start
<= end
&& state
->end
> end
) {
753 set
= state
->state
& bits
;
754 if (exclusive
&& set
) {
755 *failed_start
= start
;
759 err
= split_state(tree
, state
, prealloc
, end
+ 1);
760 BUG_ON(err
== -EEXIST
);
762 set_state_bits(tree
, prealloc
, bits
);
763 merge_state(tree
, prealloc
);
771 spin_unlock_irqrestore(&tree
->lock
, flags
);
773 free_extent_state(prealloc
);
780 spin_unlock_irqrestore(&tree
->lock
, flags
);
781 if (mask
& __GFP_WAIT
)
785 EXPORT_SYMBOL(set_extent_bit
);
787 /* wrappers around set/clear extent bit */
788 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
791 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
794 EXPORT_SYMBOL(set_extent_dirty
);
796 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
797 int bits
, gfp_t mask
)
799 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
802 EXPORT_SYMBOL(set_extent_bits
);
804 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
805 int bits
, gfp_t mask
)
807 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, mask
);
809 EXPORT_SYMBOL(clear_extent_bits
);
811 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
814 return set_extent_bit(tree
, start
, end
,
815 EXTENT_DELALLOC
| EXTENT_DIRTY
, 0, NULL
,
818 EXPORT_SYMBOL(set_extent_delalloc
);
820 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
823 return clear_extent_bit(tree
, start
, end
,
824 EXTENT_DIRTY
| EXTENT_DELALLOC
, 0, 0, mask
);
826 EXPORT_SYMBOL(clear_extent_dirty
);
828 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
831 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
834 EXPORT_SYMBOL(set_extent_new
);
836 int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
839 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0, mask
);
841 EXPORT_SYMBOL(clear_extent_new
);
843 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
846 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, NULL
,
849 EXPORT_SYMBOL(set_extent_uptodate
);
851 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
854 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0, mask
);
856 EXPORT_SYMBOL(clear_extent_uptodate
);
858 int set_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
861 return set_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
,
864 EXPORT_SYMBOL(set_extent_writeback
);
866 int clear_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
,
869 return clear_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 1, 0, mask
);
871 EXPORT_SYMBOL(clear_extent_writeback
);
873 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
875 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
877 EXPORT_SYMBOL(wait_on_extent_writeback
);
879 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
884 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1,
885 &failed_start
, mask
);
886 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
887 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
888 start
= failed_start
;
892 WARN_ON(start
> end
);
896 EXPORT_SYMBOL(lock_extent
);
898 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
901 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, mask
);
903 EXPORT_SYMBOL(unlock_extent
);
906 * helper function to set pages and extents in the tree dirty
908 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
910 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
911 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
914 while (index
<= end_index
) {
915 page
= find_get_page(tree
->mapping
, index
);
917 __set_page_dirty_nobuffers(page
);
918 page_cache_release(page
);
921 set_extent_dirty(tree
, start
, end
, GFP_NOFS
);
924 EXPORT_SYMBOL(set_range_dirty
);
927 * helper function to set both pages and extents in the tree writeback
929 int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
931 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
932 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
935 while (index
<= end_index
) {
936 page
= find_get_page(tree
->mapping
, index
);
938 set_page_writeback(page
);
939 page_cache_release(page
);
942 set_extent_writeback(tree
, start
, end
, GFP_NOFS
);
945 EXPORT_SYMBOL(set_range_writeback
);
947 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
948 u64
*start_ret
, u64
*end_ret
, int bits
)
950 struct rb_node
*node
;
951 struct extent_state
*state
;
954 spin_lock_irq(&tree
->lock
);
956 * this search will find all the extents that end after
959 node
= tree_search(tree
, start
);
965 state
= rb_entry(node
, struct extent_state
, rb_node
);
966 if (state
->end
>= start
&& (state
->state
& bits
)) {
967 *start_ret
= state
->start
;
968 *end_ret
= state
->end
;
972 node
= rb_next(node
);
977 spin_unlock_irq(&tree
->lock
);
980 EXPORT_SYMBOL(find_first_extent_bit
);
982 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
985 struct rb_node
*node
;
986 struct extent_state
*state
;
989 * this search will find all the extents that end after
992 node
= tree_search(tree
, start
);
998 state
= rb_entry(node
, struct extent_state
, rb_node
);
999 if (state
->end
>= start
&& (state
->state
& bits
)) {
1002 node
= rb_next(node
);
1009 EXPORT_SYMBOL(find_first_extent_bit_state
);
1011 u64
find_lock_delalloc_range(struct extent_io_tree
*tree
,
1012 u64
*start
, u64
*end
, u64 max_bytes
)
1014 struct rb_node
*node
;
1015 struct extent_state
*state
;
1016 u64 cur_start
= *start
;
1018 u64 total_bytes
= 0;
1020 spin_lock_irq(&tree
->lock
);
1022 * this search will find all the extents that end after
1026 node
= tree_search(tree
, cur_start
);
1033 state
= rb_entry(node
, struct extent_state
, rb_node
);
1034 if (found
&& state
->start
!= cur_start
) {
1037 if (!(state
->state
& EXTENT_DELALLOC
)) {
1043 struct extent_state
*prev_state
;
1044 struct rb_node
*prev_node
= node
;
1046 prev_node
= rb_prev(prev_node
);
1049 prev_state
= rb_entry(prev_node
,
1050 struct extent_state
,
1052 if (!(prev_state
->state
& EXTENT_DELALLOC
))
1058 if (state
->state
& EXTENT_LOCKED
) {
1060 atomic_inc(&state
->refs
);
1061 prepare_to_wait(&state
->wq
, &wait
,
1062 TASK_UNINTERRUPTIBLE
);
1063 spin_unlock_irq(&tree
->lock
);
1065 spin_lock_irq(&tree
->lock
);
1066 finish_wait(&state
->wq
, &wait
);
1067 free_extent_state(state
);
1070 set_state_cb(tree
, state
, EXTENT_LOCKED
);
1071 state
->state
|= EXTENT_LOCKED
;
1073 *start
= state
->start
;
1076 cur_start
= state
->end
+ 1;
1077 node
= rb_next(node
);
1080 total_bytes
+= state
->end
- state
->start
+ 1;
1081 if (total_bytes
>= max_bytes
)
1085 spin_unlock_irq(&tree
->lock
);
1089 u64
count_range_bits(struct extent_io_tree
*tree
,
1090 u64
*start
, u64 search_end
, u64 max_bytes
,
1093 struct rb_node
*node
;
1094 struct extent_state
*state
;
1095 u64 cur_start
= *start
;
1096 u64 total_bytes
= 0;
1099 if (search_end
<= cur_start
) {
1100 printk("search_end %Lu start %Lu\n", search_end
, cur_start
);
1105 spin_lock_irq(&tree
->lock
);
1106 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1107 total_bytes
= tree
->dirty_bytes
;
1111 * this search will find all the extents that end after
1114 node
= tree_search(tree
, cur_start
);
1120 state
= rb_entry(node
, struct extent_state
, rb_node
);
1121 if (state
->start
> search_end
)
1123 if (state
->end
>= cur_start
&& (state
->state
& bits
)) {
1124 total_bytes
+= min(search_end
, state
->end
) + 1 -
1125 max(cur_start
, state
->start
);
1126 if (total_bytes
>= max_bytes
)
1129 *start
= state
->start
;
1133 node
= rb_next(node
);
1138 spin_unlock_irq(&tree
->lock
);
1142 * helper function to lock both pages and extents in the tree.
1143 * pages must be locked first.
1145 int lock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1147 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1148 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1152 while (index
<= end_index
) {
1153 page
= grab_cache_page(tree
->mapping
, index
);
1159 err
= PTR_ERR(page
);
1164 lock_extent(tree
, start
, end
, GFP_NOFS
);
1169 * we failed above in getting the page at 'index', so we undo here
1170 * up to but not including the page at 'index'
1173 index
= start
>> PAGE_CACHE_SHIFT
;
1174 while (index
< end_index
) {
1175 page
= find_get_page(tree
->mapping
, index
);
1177 page_cache_release(page
);
1182 EXPORT_SYMBOL(lock_range
);
1185 * helper function to unlock both pages and extents in the tree.
1187 int unlock_range(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1189 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1190 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1193 while (index
<= end_index
) {
1194 page
= find_get_page(tree
->mapping
, index
);
1196 page_cache_release(page
);
1199 unlock_extent(tree
, start
, end
, GFP_NOFS
);
1202 EXPORT_SYMBOL(unlock_range
);
1204 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1206 struct rb_node
*node
;
1207 struct extent_state
*state
;
1210 spin_lock_irq(&tree
->lock
);
1212 * this search will find all the extents that end after
1215 node
= tree_search(tree
, start
);
1220 state
= rb_entry(node
, struct extent_state
, rb_node
);
1221 if (state
->start
!= start
) {
1225 state
->private = private;
1227 spin_unlock_irq(&tree
->lock
);
1231 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1233 struct rb_node
*node
;
1234 struct extent_state
*state
;
1237 spin_lock_irq(&tree
->lock
);
1239 * this search will find all the extents that end after
1242 node
= tree_search(tree
, start
);
1247 state
= rb_entry(node
, struct extent_state
, rb_node
);
1248 if (state
->start
!= start
) {
1252 *private = state
->private;
1254 spin_unlock_irq(&tree
->lock
);
1259 * searches a range in the state tree for a given mask.
1260 * If 'filled' == 1, this returns 1 only if every extent in the tree
1261 * has the bits set. Otherwise, 1 is returned if any bit in the
1262 * range is found set.
1264 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1265 int bits
, int filled
)
1267 struct extent_state
*state
= NULL
;
1268 struct rb_node
*node
;
1270 unsigned long flags
;
1272 spin_lock_irqsave(&tree
->lock
, flags
);
1273 node
= tree_search(tree
, start
);
1274 while (node
&& start
<= end
) {
1275 state
= rb_entry(node
, struct extent_state
, rb_node
);
1277 if (filled
&& state
->start
> start
) {
1282 if (state
->start
> end
)
1285 if (state
->state
& bits
) {
1289 } else if (filled
) {
1293 start
= state
->end
+ 1;
1296 node
= rb_next(node
);
1303 spin_unlock_irqrestore(&tree
->lock
, flags
);
1306 EXPORT_SYMBOL(test_range_bit
);
1309 * helper function to set a given page up to date if all the
1310 * extents in the tree for that page are up to date
1312 static int check_page_uptodate(struct extent_io_tree
*tree
,
1315 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1316 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1317 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1))
1318 SetPageUptodate(page
);
1323 * helper function to unlock a page if all the extents in the tree
1324 * for that page are unlocked
1326 static int check_page_locked(struct extent_io_tree
*tree
,
1329 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1330 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1331 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0))
1337 * helper function to end page writeback if all the extents
1338 * in the tree for that page are done with writeback
1340 static int check_page_writeback(struct extent_io_tree
*tree
,
1343 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1344 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1345 if (!test_range_bit(tree
, start
, end
, EXTENT_WRITEBACK
, 0))
1346 end_page_writeback(page
);
1350 /* lots and lots of room for performance fixes in the end_bio funcs */
1353 * after a writepage IO is done, we need to:
1354 * clear the uptodate bits on error
1355 * clear the writeback bits in the extent tree for this IO
1356 * end_page_writeback if the page has no more pending IO
1358 * Scheduling is not allowed, so the extent state tree is expected
1359 * to have one and only one object corresponding to this IO.
1361 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1362 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1364 static int end_bio_extent_writepage(struct bio
*bio
,
1365 unsigned int bytes_done
, int err
)
1368 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1369 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1370 struct extent_state
*state
= bio
->bi_private
;
1371 struct extent_io_tree
*tree
= state
->tree
;
1372 struct rb_node
*node
;
1377 unsigned long flags
;
1379 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1384 struct page
*page
= bvec
->bv_page
;
1385 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1387 end
= start
+ bvec
->bv_len
- 1;
1389 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1394 if (--bvec
>= bio
->bi_io_vec
)
1395 prefetchw(&bvec
->bv_page
->flags
);
1398 clear_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1399 ClearPageUptodate(page
);
1403 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1404 tree
->ops
->writepage_end_io_hook(page
, start
, end
,
1409 * bios can get merged in funny ways, and so we need to
1410 * be careful with the state variable. We know the
1411 * state won't be merged with others because it has
1412 * WRITEBACK set, but we can't be sure each biovec is
1413 * sequential in the file. So, if our cached state
1414 * doesn't match the expected end, search the tree
1415 * for the correct one.
1418 spin_lock_irqsave(&tree
->lock
, flags
);
1419 if (!state
|| state
->end
!= end
) {
1421 node
= __etree_search(tree
, start
, NULL
, NULL
);
1423 state
= rb_entry(node
, struct extent_state
,
1425 if (state
->end
!= end
||
1426 !(state
->state
& EXTENT_WRITEBACK
))
1430 spin_unlock_irqrestore(&tree
->lock
, flags
);
1431 clear_extent_writeback(tree
, start
,
1438 struct extent_state
*clear
= state
;
1440 node
= rb_prev(&state
->rb_node
);
1442 state
= rb_entry(node
,
1443 struct extent_state
,
1449 clear_state_bit(tree
, clear
, EXTENT_WRITEBACK
,
1460 /* before releasing the lock, make sure the next state
1461 * variable has the expected bits set and corresponds
1462 * to the correct offsets in the file
1464 if (state
&& (state
->end
+ 1 != start
||
1465 !(state
->state
& EXTENT_WRITEBACK
))) {
1468 spin_unlock_irqrestore(&tree
->lock
, flags
);
1472 end_page_writeback(page
);
1474 check_page_writeback(tree
, page
);
1475 } while (bvec
>= bio
->bi_io_vec
);
1477 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1483 * after a readpage IO is done, we need to:
1484 * clear the uptodate bits on error
1485 * set the uptodate bits if things worked
1486 * set the page up to date if all extents in the tree are uptodate
1487 * clear the lock bit in the extent tree
1488 * unlock the page if there are no other extents locked for it
1490 * Scheduling is not allowed, so the extent state tree is expected
1491 * to have one and only one object corresponding to this IO.
1493 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1494 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1496 static int end_bio_extent_readpage(struct bio
*bio
,
1497 unsigned int bytes_done
, int err
)
1500 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1501 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1502 struct extent_state
*state
= bio
->bi_private
;
1503 struct extent_io_tree
*tree
= state
->tree
;
1504 struct rb_node
*node
;
1508 unsigned long flags
;
1512 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1518 struct page
*page
= bvec
->bv_page
;
1519 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1521 end
= start
+ bvec
->bv_len
- 1;
1523 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1528 if (--bvec
>= bio
->bi_io_vec
)
1529 prefetchw(&bvec
->bv_page
->flags
);
1531 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1532 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1538 spin_lock_irqsave(&tree
->lock
, flags
);
1539 if (!state
|| state
->end
!= end
) {
1541 node
= __etree_search(tree
, start
, NULL
, NULL
);
1543 state
= rb_entry(node
, struct extent_state
,
1545 if (state
->end
!= end
||
1546 !(state
->state
& EXTENT_LOCKED
))
1550 spin_unlock_irqrestore(&tree
->lock
, flags
);
1551 set_extent_uptodate(tree
, start
, end
,
1553 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1560 struct extent_state
*clear
= state
;
1562 node
= rb_prev(&state
->rb_node
);
1564 state
= rb_entry(node
,
1565 struct extent_state
,
1570 set_state_cb(tree
, clear
, EXTENT_UPTODATE
);
1571 clear
->state
|= EXTENT_UPTODATE
;
1572 clear_state_bit(tree
, clear
, EXTENT_LOCKED
,
1583 /* before releasing the lock, make sure the next state
1584 * variable has the expected bits set and corresponds
1585 * to the correct offsets in the file
1587 if (state
&& (state
->end
+ 1 != start
||
1588 !(state
->state
& EXTENT_LOCKED
))) {
1591 spin_unlock_irqrestore(&tree
->lock
, flags
);
1595 SetPageUptodate(page
);
1597 ClearPageUptodate(page
);
1603 check_page_uptodate(tree
, page
);
1605 ClearPageUptodate(page
);
1608 check_page_locked(tree
, page
);
1610 } while (bvec
>= bio
->bi_io_vec
);
1613 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1619 * IO done from prepare_write is pretty simple, we just unlock
1620 * the structs in the extent tree when done, and set the uptodate bits
1623 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1624 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1626 static int end_bio_extent_preparewrite(struct bio
*bio
,
1627 unsigned int bytes_done
, int err
)
1630 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1631 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1632 struct extent_state
*state
= bio
->bi_private
;
1633 struct extent_io_tree
*tree
= state
->tree
;
1637 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1643 struct page
*page
= bvec
->bv_page
;
1644 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1646 end
= start
+ bvec
->bv_len
- 1;
1648 if (--bvec
>= bio
->bi_io_vec
)
1649 prefetchw(&bvec
->bv_page
->flags
);
1652 set_extent_uptodate(tree
, start
, end
, GFP_ATOMIC
);
1654 ClearPageUptodate(page
);
1658 unlock_extent(tree
, start
, end
, GFP_ATOMIC
);
1660 } while (bvec
>= bio
->bi_io_vec
);
1663 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1669 extent_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1674 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1676 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1677 while (!bio
&& (nr_vecs
/= 2))
1678 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1682 bio
->bi_bdev
= bdev
;
1683 bio
->bi_sector
= first_sector
;
1688 static int submit_one_bio(int rw
, struct bio
*bio
)
1692 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1693 struct page
*page
= bvec
->bv_page
;
1694 struct extent_io_tree
*tree
= bio
->bi_private
;
1695 struct rb_node
*node
;
1696 struct extent_state
*state
;
1700 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1701 end
= start
+ bvec
->bv_len
- 1;
1703 spin_lock_irq(&tree
->lock
);
1704 node
= __etree_search(tree
, start
, NULL
, NULL
);
1706 state
= rb_entry(node
, struct extent_state
, rb_node
);
1707 while(state
->end
< end
) {
1708 node
= rb_next(node
);
1709 state
= rb_entry(node
, struct extent_state
, rb_node
);
1711 BUG_ON(state
->end
!= end
);
1712 spin_unlock_irq(&tree
->lock
);
1714 bio
->bi_private
= state
;
1718 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1719 if (maxsector
< bio
->bi_sector
) {
1720 printk("sector too large max %Lu got %llu\n", maxsector
,
1721 (unsigned long long)bio
->bi_sector
);
1724 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
1725 tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
);
1727 submit_bio(rw
, bio
);
1728 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1734 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1735 struct page
*page
, sector_t sector
,
1736 size_t size
, unsigned long offset
,
1737 struct block_device
*bdev
,
1738 struct bio
**bio_ret
,
1739 unsigned long max_pages
,
1740 bio_end_io_t end_io_func
)
1746 if (bio_ret
&& *bio_ret
) {
1748 if (bio
->bi_sector
+ (bio
->bi_size
>> 9) != sector
||
1749 (tree
->ops
&& tree
->ops
->merge_bio_hook
&&
1750 tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
)) ||
1751 bio_add_page(bio
, page
, size
, offset
) < size
) {
1752 ret
= submit_one_bio(rw
, bio
);
1758 nr
= bio_get_nr_vecs(bdev
);
1759 bio
= extent_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1761 printk("failed to allocate bio nr %d\n", nr
);
1765 bio_add_page(bio
, page
, size
, offset
);
1766 bio
->bi_end_io
= end_io_func
;
1767 bio
->bi_private
= tree
;
1772 ret
= submit_one_bio(rw
, bio
);
1778 void set_page_extent_mapped(struct page
*page
)
1780 if (!PagePrivate(page
)) {
1781 SetPagePrivate(page
);
1782 WARN_ON(!page
->mapping
->a_ops
->invalidatepage
);
1783 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1784 page_cache_get(page
);
1788 void set_page_extent_head(struct page
*page
, unsigned long len
)
1790 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1794 * basic readpage implementation. Locked extent state structs are inserted
1795 * into the tree that are removed when the IO is done (by the end_io
1798 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1800 get_extent_t
*get_extent
,
1803 struct inode
*inode
= page
->mapping
->host
;
1804 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1805 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1809 u64 last_byte
= i_size_read(inode
);
1813 struct extent_map
*em
;
1814 struct block_device
*bdev
;
1817 size_t page_offset
= 0;
1819 size_t blocksize
= inode
->i_sb
->s_blocksize
;
1821 set_page_extent_mapped(page
);
1824 lock_extent(tree
, start
, end
, GFP_NOFS
);
1826 while (cur
<= end
) {
1827 if (cur
>= last_byte
) {
1829 iosize
= PAGE_CACHE_SIZE
- page_offset
;
1830 userpage
= kmap_atomic(page
, KM_USER0
);
1831 memset(userpage
+ page_offset
, 0, iosize
);
1832 flush_dcache_page(page
);
1833 kunmap_atomic(userpage
, KM_USER0
);
1834 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1836 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1839 em
= get_extent(inode
, page
, page_offset
, cur
,
1841 if (IS_ERR(em
) || !em
) {
1843 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
1847 extent_offset
= cur
- em
->start
;
1848 BUG_ON(extent_map_end(em
) <= cur
);
1851 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
1852 cur_end
= min(extent_map_end(em
) - 1, end
);
1853 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
1854 sector
= (em
->block_start
+ extent_offset
) >> 9;
1856 block_start
= em
->block_start
;
1857 free_extent_map(em
);
1860 /* we've found a hole, just zero and go on */
1861 if (block_start
== EXTENT_MAP_HOLE
) {
1863 userpage
= kmap_atomic(page
, KM_USER0
);
1864 memset(userpage
+ page_offset
, 0, iosize
);
1865 flush_dcache_page(page
);
1866 kunmap_atomic(userpage
, KM_USER0
);
1868 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
1870 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1872 page_offset
+= iosize
;
1875 /* the get_extent function already copied into the page */
1876 if (test_range_bit(tree
, cur
, cur_end
, EXTENT_UPTODATE
, 1)) {
1877 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1879 page_offset
+= iosize
;
1882 /* we have an inline extent but it didn't get marked up
1883 * to date. Error out
1885 if (block_start
== EXTENT_MAP_INLINE
) {
1887 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
1889 page_offset
+= iosize
;
1894 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
1895 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
1899 unsigned long nr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
1901 ret
= submit_extent_page(READ
, tree
, page
,
1902 sector
, iosize
, page_offset
,
1904 end_bio_extent_readpage
);
1909 page_offset
+= iosize
;
1913 if (!PageError(page
))
1914 SetPageUptodate(page
);
1920 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
1921 get_extent_t
*get_extent
)
1923 struct bio
*bio
= NULL
;
1926 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
);
1928 submit_one_bio(READ
, bio
);
1931 EXPORT_SYMBOL(extent_read_full_page
);
1934 * the writepage semantics are similar to regular writepage. extent
1935 * records are inserted to lock ranges in the tree, and as dirty areas
1936 * are found, they are marked writeback. Then the lock bits are removed
1937 * and the end_io handler clears the writeback ranges
1939 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
1942 struct inode
*inode
= page
->mapping
->host
;
1943 struct extent_page_data
*epd
= data
;
1944 struct extent_io_tree
*tree
= epd
->tree
;
1945 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1947 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
1951 u64 last_byte
= i_size_read(inode
);
1955 struct extent_map
*em
;
1956 struct block_device
*bdev
;
1959 size_t page_offset
= 0;
1961 loff_t i_size
= i_size_read(inode
);
1962 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1966 WARN_ON(!PageLocked(page
));
1967 if (page
->index
> end_index
) {
1968 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
1973 if (page
->index
== end_index
) {
1976 size_t offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
1978 userpage
= kmap_atomic(page
, KM_USER0
);
1979 memset(userpage
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
1980 flush_dcache_page(page
);
1981 kunmap_atomic(userpage
, KM_USER0
);
1984 set_page_extent_mapped(page
);
1986 delalloc_start
= start
;
1988 while(delalloc_end
< page_end
) {
1989 nr_delalloc
= find_lock_delalloc_range(tree
, &delalloc_start
,
1992 if (nr_delalloc
== 0) {
1993 delalloc_start
= delalloc_end
+ 1;
1996 tree
->ops
->fill_delalloc(inode
, delalloc_start
,
1998 clear_extent_bit(tree
, delalloc_start
,
2000 EXTENT_LOCKED
| EXTENT_DELALLOC
,
2002 delalloc_start
= delalloc_end
+ 1;
2004 lock_extent(tree
, start
, page_end
, GFP_NOFS
);
2007 if (test_range_bit(tree
, start
, page_end
, EXTENT_DELALLOC
, 0)) {
2008 printk("found delalloc bits after lock_extent\n");
2011 if (last_byte
<= start
) {
2012 clear_extent_dirty(tree
, start
, page_end
, GFP_NOFS
);
2016 set_extent_uptodate(tree
, start
, page_end
, GFP_NOFS
);
2017 blocksize
= inode
->i_sb
->s_blocksize
;
2019 while (cur
<= end
) {
2020 if (cur
>= last_byte
) {
2021 clear_extent_dirty(tree
, cur
, page_end
, GFP_NOFS
);
2024 em
= epd
->get_extent(inode
, page
, page_offset
, cur
,
2026 if (IS_ERR(em
) || !em
) {
2031 extent_offset
= cur
- em
->start
;
2032 BUG_ON(extent_map_end(em
) <= cur
);
2034 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2035 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2036 sector
= (em
->block_start
+ extent_offset
) >> 9;
2038 block_start
= em
->block_start
;
2039 free_extent_map(em
);
2042 if (block_start
== EXTENT_MAP_HOLE
||
2043 block_start
== EXTENT_MAP_INLINE
) {
2044 clear_extent_dirty(tree
, cur
,
2045 cur
+ iosize
- 1, GFP_NOFS
);
2047 page_offset
+= iosize
;
2051 /* leave this out until we have a page_mkwrite call */
2052 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2055 page_offset
+= iosize
;
2058 clear_extent_dirty(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2059 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2060 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2068 unsigned long max_nr
= end_index
+ 1;
2069 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2070 if (!PageWriteback(page
)) {
2071 printk("warning page %lu not writeback, "
2072 "cur %llu end %llu\n", page
->index
,
2073 (unsigned long long)cur
,
2074 (unsigned long long)end
);
2077 ret
= submit_extent_page(WRITE
, tree
, page
, sector
,
2078 iosize
, page_offset
, bdev
,
2080 end_bio_extent_writepage
);
2085 page_offset
+= iosize
;
2090 /* make sure the mapping tag for page dirty gets cleared */
2091 set_page_writeback(page
);
2092 end_page_writeback(page
);
2094 unlock_extent(tree
, start
, page_end
, GFP_NOFS
);
2099 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2101 /* Taken directly from 2.6.23 for 2.6.18 back port */
2102 typedef int (*writepage_t
)(struct page
*page
, struct writeback_control
*wbc
,
2106 * write_cache_pages - walk the list of dirty pages of the given address space
2107 * and write all of them.
2108 * @mapping: address space structure to write
2109 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2110 * @writepage: function called for each page
2111 * @data: data passed to writepage function
2113 * If a page is already under I/O, write_cache_pages() skips it, even
2114 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2115 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2116 * and msync() need to guarantee that all the data which was dirty at the time
2117 * the call was made get new I/O started against them. If wbc->sync_mode is
2118 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2119 * existing IO to complete.
2121 static int write_cache_pages(struct address_space
*mapping
,
2122 struct writeback_control
*wbc
, writepage_t writepage
,
2125 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
2128 struct pagevec pvec
;
2131 pgoff_t end
; /* Inclusive */
2133 int range_whole
= 0;
2135 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2136 wbc
->encountered_congestion
= 1;
2140 pagevec_init(&pvec
, 0);
2141 if (wbc
->range_cyclic
) {
2142 index
= mapping
->writeback_index
; /* Start from prev offset */
2145 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2146 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2147 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2152 while (!done
&& (index
<= end
) &&
2153 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2154 PAGECACHE_TAG_DIRTY
,
2155 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2159 for (i
= 0; i
< nr_pages
; i
++) {
2160 struct page
*page
= pvec
.pages
[i
];
2163 * At this point we hold neither mapping->tree_lock nor
2164 * lock on the page itself: the page may be truncated or
2165 * invalidated (changing page->mapping to NULL), or even
2166 * swizzled back from swapper_space to tmpfs file
2171 if (unlikely(page
->mapping
!= mapping
)) {
2176 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2182 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2183 wait_on_page_writeback(page
);
2185 if (PageWriteback(page
) ||
2186 !clear_page_dirty_for_io(page
)) {
2191 ret
= (*writepage
)(page
, wbc
, data
);
2193 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2197 if (ret
|| (--(wbc
->nr_to_write
) <= 0))
2199 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
2200 wbc
->encountered_congestion
= 1;
2204 pagevec_release(&pvec
);
2207 if (!scanned
&& !done
) {
2209 * We hit the last page and there is more work to be done: wrap
2210 * back to the start of the file
2216 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2217 mapping
->writeback_index
= index
;
2222 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2223 get_extent_t
*get_extent
,
2224 struct writeback_control
*wbc
)
2227 struct address_space
*mapping
= page
->mapping
;
2228 struct extent_page_data epd
= {
2231 .get_extent
= get_extent
,
2233 struct writeback_control wbc_writepages
= {
2235 .sync_mode
= WB_SYNC_NONE
,
2236 .older_than_this
= NULL
,
2238 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2239 .range_end
= (loff_t
)-1,
2243 ret
= __extent_writepage(page
, wbc
, &epd
);
2245 write_cache_pages(mapping
, &wbc_writepages
, __extent_writepage
, &epd
);
2247 submit_one_bio(WRITE
, epd
.bio
);
2251 EXPORT_SYMBOL(extent_write_full_page
);
2254 int extent_writepages(struct extent_io_tree
*tree
,
2255 struct address_space
*mapping
,
2256 get_extent_t
*get_extent
,
2257 struct writeback_control
*wbc
)
2260 struct extent_page_data epd
= {
2263 .get_extent
= get_extent
,
2266 ret
= write_cache_pages(mapping
, wbc
, __extent_writepage
, &epd
);
2268 submit_one_bio(WRITE
, epd
.bio
);
2272 EXPORT_SYMBOL(extent_writepages
);
2274 int extent_readpages(struct extent_io_tree
*tree
,
2275 struct address_space
*mapping
,
2276 struct list_head
*pages
, unsigned nr_pages
,
2277 get_extent_t get_extent
)
2279 struct bio
*bio
= NULL
;
2281 struct pagevec pvec
;
2283 pagevec_init(&pvec
, 0);
2284 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2285 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2287 prefetchw(&page
->flags
);
2288 list_del(&page
->lru
);
2290 * what we want to do here is call add_to_page_cache_lru,
2291 * but that isn't exported, so we reproduce it here
2293 if (!add_to_page_cache(page
, mapping
,
2294 page
->index
, GFP_KERNEL
)) {
2296 /* open coding of lru_cache_add, also not exported */
2297 page_cache_get(page
);
2298 if (!pagevec_add(&pvec
, page
))
2299 __pagevec_lru_add(&pvec
);
2300 __extent_read_full_page(tree
, page
, get_extent
, &bio
);
2302 page_cache_release(page
);
2304 if (pagevec_count(&pvec
))
2305 __pagevec_lru_add(&pvec
);
2306 BUG_ON(!list_empty(pages
));
2308 submit_one_bio(READ
, bio
);
2311 EXPORT_SYMBOL(extent_readpages
);
2314 * basic invalidatepage code, this waits on any locked or writeback
2315 * ranges corresponding to the page, and then deletes any extent state
2316 * records from the tree
2318 int extent_invalidatepage(struct extent_io_tree
*tree
,
2319 struct page
*page
, unsigned long offset
)
2321 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2322 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2323 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2325 start
+= (offset
+ blocksize
-1) & ~(blocksize
- 1);
2329 lock_extent(tree
, start
, end
, GFP_NOFS
);
2330 wait_on_extent_writeback(tree
, start
, end
);
2331 clear_extent_bit(tree
, start
, end
,
2332 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
,
2336 EXPORT_SYMBOL(extent_invalidatepage
);
2339 * simple commit_write call, set_range_dirty is used to mark both
2340 * the pages and the extent records as dirty
2342 int extent_commit_write(struct extent_io_tree
*tree
,
2343 struct inode
*inode
, struct page
*page
,
2344 unsigned from
, unsigned to
)
2346 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2348 set_page_extent_mapped(page
);
2349 set_page_dirty(page
);
2351 if (pos
> inode
->i_size
) {
2352 i_size_write(inode
, pos
);
2353 mark_inode_dirty(inode
);
2357 EXPORT_SYMBOL(extent_commit_write
);
2359 int extent_prepare_write(struct extent_io_tree
*tree
,
2360 struct inode
*inode
, struct page
*page
,
2361 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2363 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2364 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2366 u64 orig_block_start
;
2369 struct extent_map
*em
;
2370 unsigned blocksize
= 1 << inode
->i_blkbits
;
2371 size_t page_offset
= 0;
2372 size_t block_off_start
;
2373 size_t block_off_end
;
2379 set_page_extent_mapped(page
);
2381 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2382 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2383 orig_block_start
= block_start
;
2385 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2386 while(block_start
<= block_end
) {
2387 em
= get_extent(inode
, page
, page_offset
, block_start
,
2388 block_end
- block_start
+ 1, 1);
2389 if (IS_ERR(em
) || !em
) {
2392 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2393 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2394 block_off_end
= block_off_start
+ blocksize
;
2395 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2397 if (!PageUptodate(page
) && isnew
&&
2398 (block_off_end
> to
|| block_off_start
< from
)) {
2401 kaddr
= kmap_atomic(page
, KM_USER0
);
2402 if (block_off_end
> to
)
2403 memset(kaddr
+ to
, 0, block_off_end
- to
);
2404 if (block_off_start
< from
)
2405 memset(kaddr
+ block_off_start
, 0,
2406 from
- block_off_start
);
2407 flush_dcache_page(page
);
2408 kunmap_atomic(kaddr
, KM_USER0
);
2410 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2411 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2412 !isnew
&& !PageUptodate(page
) &&
2413 (block_off_end
> to
|| block_off_start
< from
) &&
2414 !test_range_bit(tree
, block_start
, cur_end
,
2415 EXTENT_UPTODATE
, 1)) {
2417 u64 extent_offset
= block_start
- em
->start
;
2419 sector
= (em
->block_start
+ extent_offset
) >> 9;
2420 iosize
= (cur_end
- block_start
+ blocksize
) &
2421 ~((u64
)blocksize
- 1);
2423 * we've already got the extent locked, but we
2424 * need to split the state such that our end_bio
2425 * handler can clear the lock.
2427 set_extent_bit(tree
, block_start
,
2428 block_start
+ iosize
- 1,
2429 EXTENT_LOCKED
, 0, NULL
, GFP_NOFS
);
2430 ret
= submit_extent_page(READ
, tree
, page
,
2431 sector
, iosize
, page_offset
, em
->bdev
,
2433 end_bio_extent_preparewrite
);
2435 block_start
= block_start
+ iosize
;
2437 set_extent_uptodate(tree
, block_start
, cur_end
,
2439 unlock_extent(tree
, block_start
, cur_end
, GFP_NOFS
);
2440 block_start
= cur_end
+ 1;
2442 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2443 free_extent_map(em
);
2446 wait_extent_bit(tree
, orig_block_start
,
2447 block_end
, EXTENT_LOCKED
);
2449 check_page_uptodate(tree
, page
);
2451 /* FIXME, zero out newly allocated blocks on error */
2454 EXPORT_SYMBOL(extent_prepare_write
);
2457 * a helper for releasepage. As long as there are no locked extents
2458 * in the range corresponding to the page, both state records and extent
2459 * map records are removed
2461 int try_release_extent_mapping(struct extent_map_tree
*map
,
2462 struct extent_io_tree
*tree
, struct page
*page
,
2465 struct extent_map
*em
;
2466 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2467 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2468 u64 orig_start
= start
;
2470 if ((mask
& __GFP_WAIT
) &&
2471 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2473 while (start
<= end
) {
2474 len
= end
- start
+ 1;
2475 spin_lock(&map
->lock
);
2476 em
= lookup_extent_mapping(map
, start
, len
);
2477 if (!em
|| IS_ERR(em
)) {
2478 spin_unlock(&map
->lock
);
2481 if (em
->start
!= start
) {
2482 spin_unlock(&map
->lock
);
2483 free_extent_map(em
);
2486 if (!test_range_bit(tree
, em
->start
,
2487 extent_map_end(em
) - 1,
2488 EXTENT_LOCKED
, 0)) {
2489 remove_extent_mapping(map
, em
);
2490 /* once for the rb tree */
2491 free_extent_map(em
);
2493 start
= extent_map_end(em
);
2494 spin_unlock(&map
->lock
);
2497 free_extent_map(em
);
2500 if (test_range_bit(tree
, orig_start
, end
, EXTENT_IOBITS
, 0))
2503 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2505 clear_extent_bit(tree
, orig_start
, end
, EXTENT_UPTODATE
,
2510 EXPORT_SYMBOL(try_release_extent_mapping
);
2512 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2513 get_extent_t
*get_extent
)
2515 struct inode
*inode
= mapping
->host
;
2516 u64 start
= iblock
<< inode
->i_blkbits
;
2517 sector_t sector
= 0;
2518 struct extent_map
*em
;
2520 em
= get_extent(inode
, NULL
, 0, start
, (1 << inode
->i_blkbits
), 0);
2521 if (!em
|| IS_ERR(em
))
2524 if (em
->block_start
== EXTENT_MAP_INLINE
||
2525 em
->block_start
== EXTENT_MAP_HOLE
)
2528 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2530 free_extent_map(em
);
2534 static int add_lru(struct extent_io_tree
*tree
, struct extent_buffer
*eb
)
2536 if (list_empty(&eb
->lru
)) {
2537 extent_buffer_get(eb
);
2538 list_add(&eb
->lru
, &tree
->buffer_lru
);
2540 if (tree
->lru_size
>= BUFFER_LRU_MAX
) {
2541 struct extent_buffer
*rm
;
2542 rm
= list_entry(tree
->buffer_lru
.prev
,
2543 struct extent_buffer
, lru
);
2545 list_del_init(&rm
->lru
);
2546 free_extent_buffer(rm
);
2549 list_move(&eb
->lru
, &tree
->buffer_lru
);
2552 static struct extent_buffer
*find_lru(struct extent_io_tree
*tree
,
2553 u64 start
, unsigned long len
)
2555 struct list_head
*lru
= &tree
->buffer_lru
;
2556 struct list_head
*cur
= lru
->next
;
2557 struct extent_buffer
*eb
;
2559 if (list_empty(lru
))
2563 eb
= list_entry(cur
, struct extent_buffer
, lru
);
2564 if (eb
->start
== start
&& eb
->len
== len
) {
2565 extent_buffer_get(eb
);
2569 } while (cur
!= lru
);
2573 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
2575 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
2576 (start
>> PAGE_CACHE_SHIFT
);
2579 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
2583 struct address_space
*mapping
;
2586 return eb
->first_page
;
2587 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
2588 mapping
= eb
->first_page
->mapping
;
2589 read_lock_irq(&mapping
->tree_lock
);
2590 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
2591 read_unlock_irq(&mapping
->tree_lock
);
2595 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
2600 struct extent_buffer
*eb
= NULL
;
2601 unsigned long flags
;
2603 spin_lock(&tree
->lru_lock
);
2604 eb
= find_lru(tree
, start
, len
);
2605 spin_unlock(&tree
->lru_lock
);
2610 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
2611 INIT_LIST_HEAD(&eb
->lru
);
2614 spin_lock_irqsave(&leak_lock
, flags
);
2615 list_add(&eb
->leak_list
, &buffers
);
2616 spin_unlock_irqrestore(&leak_lock
, flags
);
2617 atomic_set(&eb
->refs
, 1);
2622 static void __free_extent_buffer(struct extent_buffer
*eb
)
2624 unsigned long flags
;
2625 spin_lock_irqsave(&leak_lock
, flags
);
2626 list_del(&eb
->leak_list
);
2627 spin_unlock_irqrestore(&leak_lock
, flags
);
2628 kmem_cache_free(extent_buffer_cache
, eb
);
2631 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
2632 u64 start
, unsigned long len
,
2636 unsigned long num_pages
= num_extent_pages(start
, len
);
2638 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2639 struct extent_buffer
*eb
;
2641 struct address_space
*mapping
= tree
->mapping
;
2644 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2648 if (eb
->flags
& EXTENT_BUFFER_FILLED
)
2652 eb
->first_page
= page0
;
2655 page_cache_get(page0
);
2656 mark_page_accessed(page0
);
2657 set_page_extent_mapped(page0
);
2658 WARN_ON(!PageUptodate(page0
));
2659 set_page_extent_head(page0
, len
);
2663 for (; i
< num_pages
; i
++, index
++) {
2664 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
2669 set_page_extent_mapped(p
);
2670 mark_page_accessed(p
);
2673 set_page_extent_head(p
, len
);
2675 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2677 if (!PageUptodate(p
))
2682 eb
->flags
|= EXTENT_UPTODATE
;
2683 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2686 spin_lock(&tree
->lru_lock
);
2688 spin_unlock(&tree
->lru_lock
);
2692 spin_lock(&tree
->lru_lock
);
2693 list_del_init(&eb
->lru
);
2694 spin_unlock(&tree
->lru_lock
);
2695 if (!atomic_dec_and_test(&eb
->refs
))
2697 for (index
= 1; index
< i
; index
++) {
2698 page_cache_release(extent_buffer_page(eb
, index
));
2701 page_cache_release(extent_buffer_page(eb
, 0));
2702 __free_extent_buffer(eb
);
2705 EXPORT_SYMBOL(alloc_extent_buffer
);
2707 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
2708 u64 start
, unsigned long len
,
2711 unsigned long num_pages
= num_extent_pages(start
, len
);
2713 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
2714 struct extent_buffer
*eb
;
2716 struct address_space
*mapping
= tree
->mapping
;
2719 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
2723 if (eb
->flags
& EXTENT_BUFFER_FILLED
)
2726 for (i
= 0; i
< num_pages
; i
++, index
++) {
2727 p
= find_lock_page(mapping
, index
);
2731 set_page_extent_mapped(p
);
2732 mark_page_accessed(p
);
2736 set_page_extent_head(p
, len
);
2738 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
2741 if (!PageUptodate(p
))
2746 eb
->flags
|= EXTENT_UPTODATE
;
2747 eb
->flags
|= EXTENT_BUFFER_FILLED
;
2750 spin_lock(&tree
->lru_lock
);
2752 spin_unlock(&tree
->lru_lock
);
2755 spin_lock(&tree
->lru_lock
);
2756 list_del_init(&eb
->lru
);
2757 spin_unlock(&tree
->lru_lock
);
2758 if (!atomic_dec_and_test(&eb
->refs
))
2760 for (index
= 1; index
< i
; index
++) {
2761 page_cache_release(extent_buffer_page(eb
, index
));
2764 page_cache_release(extent_buffer_page(eb
, 0));
2765 __free_extent_buffer(eb
);
2768 EXPORT_SYMBOL(find_extent_buffer
);
2770 void free_extent_buffer(struct extent_buffer
*eb
)
2773 unsigned long num_pages
;
2778 if (!atomic_dec_and_test(&eb
->refs
))
2781 WARN_ON(!list_empty(&eb
->lru
));
2782 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2784 for (i
= 1; i
< num_pages
; i
++) {
2785 page_cache_release(extent_buffer_page(eb
, i
));
2787 page_cache_release(extent_buffer_page(eb
, 0));
2788 __free_extent_buffer(eb
);
2790 EXPORT_SYMBOL(free_extent_buffer
);
2792 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
2793 struct extent_buffer
*eb
)
2797 unsigned long num_pages
;
2800 u64 start
= eb
->start
;
2801 u64 end
= start
+ eb
->len
- 1;
2803 set
= clear_extent_dirty(tree
, start
, end
, GFP_NOFS
);
2804 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2806 for (i
= 0; i
< num_pages
; i
++) {
2807 page
= extent_buffer_page(eb
, i
);
2810 set_page_extent_head(page
, eb
->len
);
2812 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2815 * if we're on the last page or the first page and the
2816 * block isn't aligned on a page boundary, do extra checks
2817 * to make sure we don't clean page that is partially dirty
2819 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2820 ((i
== num_pages
- 1) &&
2821 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2822 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2823 end
= start
+ PAGE_CACHE_SIZE
- 1;
2824 if (test_range_bit(tree
, start
, end
,
2830 clear_page_dirty_for_io(page
);
2831 read_lock_irq(&page
->mapping
->tree_lock
);
2832 if (!PageDirty(page
)) {
2833 radix_tree_tag_clear(&page
->mapping
->page_tree
,
2835 PAGECACHE_TAG_DIRTY
);
2837 read_unlock_irq(&page
->mapping
->tree_lock
);
2842 EXPORT_SYMBOL(clear_extent_buffer_dirty
);
2844 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
2845 struct extent_buffer
*eb
)
2847 return wait_on_extent_writeback(tree
, eb
->start
,
2848 eb
->start
+ eb
->len
- 1);
2850 EXPORT_SYMBOL(wait_on_extent_buffer_writeback
);
2852 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
2853 struct extent_buffer
*eb
)
2856 unsigned long num_pages
;
2858 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2859 for (i
= 0; i
< num_pages
; i
++) {
2860 struct page
*page
= extent_buffer_page(eb
, i
);
2861 /* writepage may need to do something special for the
2862 * first page, we have to make sure page->private is
2863 * properly set. releasepage may drop page->private
2864 * on us if the page isn't already dirty.
2868 set_page_extent_head(page
, eb
->len
);
2869 } else if (PagePrivate(page
) &&
2870 page
->private != EXTENT_PAGE_PRIVATE
) {
2872 set_page_extent_mapped(page
);
2875 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
2879 return set_extent_dirty(tree
, eb
->start
,
2880 eb
->start
+ eb
->len
- 1, GFP_NOFS
);
2882 EXPORT_SYMBOL(set_extent_buffer_dirty
);
2884 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
2885 struct extent_buffer
*eb
)
2889 unsigned long num_pages
;
2891 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2893 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2895 for (i
= 0; i
< num_pages
; i
++) {
2896 page
= extent_buffer_page(eb
, i
);
2897 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
2898 ((i
== num_pages
- 1) &&
2899 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
2900 check_page_uptodate(tree
, page
);
2903 SetPageUptodate(page
);
2907 EXPORT_SYMBOL(set_extent_buffer_uptodate
);
2909 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
2910 struct extent_buffer
*eb
)
2912 if (eb
->flags
& EXTENT_UPTODATE
)
2914 return test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2915 EXTENT_UPTODATE
, 1);
2917 EXPORT_SYMBOL(extent_buffer_uptodate
);
2919 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
2920 struct extent_buffer
*eb
,
2921 u64 start
, int wait
,
2922 get_extent_t
*get_extent
)
2925 unsigned long start_i
;
2929 unsigned long num_pages
;
2930 struct bio
*bio
= NULL
;
2933 if (eb
->flags
& EXTENT_UPTODATE
)
2936 if (0 && test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
2937 EXTENT_UPTODATE
, 1)) {
2942 WARN_ON(start
< eb
->start
);
2943 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
2944 (eb
->start
>> PAGE_CACHE_SHIFT
);
2949 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2950 for (i
= start_i
; i
< num_pages
; i
++) {
2951 page
= extent_buffer_page(eb
, i
);
2952 if (PageUptodate(page
)) {
2956 if (TestSetPageLocked(page
)) {
2962 if (!PageUptodate(page
)) {
2963 err
= __extent_read_full_page(tree
, page
,
2974 submit_one_bio(READ
, bio
);
2979 for (i
= start_i
; i
< num_pages
; i
++) {
2980 page
= extent_buffer_page(eb
, i
);
2981 wait_on_page_locked(page
);
2982 if (!PageUptodate(page
)) {
2987 eb
->flags
|= EXTENT_UPTODATE
;
2990 EXPORT_SYMBOL(read_extent_buffer_pages
);
2992 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
2993 unsigned long start
,
3000 char *dst
= (char *)dstv
;
3001 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3002 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3003 unsigned long num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3005 WARN_ON(start
> eb
->len
);
3006 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3008 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3011 page
= extent_buffer_page(eb
, i
);
3012 if (!PageUptodate(page
)) {
3013 printk("page %lu not up to date i %lu, total %lu, len %lu\n", page
->index
, i
, num_pages
, eb
->len
);
3016 WARN_ON(!PageUptodate(page
));
3018 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3019 kaddr
= kmap_atomic(page
, KM_USER1
);
3020 memcpy(dst
, kaddr
+ offset
, cur
);
3021 kunmap_atomic(kaddr
, KM_USER1
);
3029 EXPORT_SYMBOL(read_extent_buffer
);
3031 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3032 unsigned long min_len
, char **token
, char **map
,
3033 unsigned long *map_start
,
3034 unsigned long *map_len
, int km
)
3036 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
3039 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3040 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3041 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
3048 offset
= start_offset
;
3052 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
3054 if (start
+ min_len
> eb
->len
) {
3055 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb
->start
, eb
->len
, start
, min_len
);
3059 p
= extent_buffer_page(eb
, i
);
3060 WARN_ON(!PageUptodate(p
));
3061 kaddr
= kmap_atomic(p
, km
);
3063 *map
= kaddr
+ offset
;
3064 *map_len
= PAGE_CACHE_SIZE
- offset
;
3067 EXPORT_SYMBOL(map_private_extent_buffer
);
3069 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3070 unsigned long min_len
,
3071 char **token
, char **map
,
3072 unsigned long *map_start
,
3073 unsigned long *map_len
, int km
)
3077 if (eb
->map_token
) {
3078 unmap_extent_buffer(eb
, eb
->map_token
, km
);
3079 eb
->map_token
= NULL
;
3082 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
3083 map_start
, map_len
, km
);
3085 eb
->map_token
= *token
;
3087 eb
->map_start
= *map_start
;
3088 eb
->map_len
= *map_len
;
3092 EXPORT_SYMBOL(map_extent_buffer
);
3094 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
3096 kunmap_atomic(token
, km
);
3098 EXPORT_SYMBOL(unmap_extent_buffer
);
3100 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3101 unsigned long start
,
3108 char *ptr
= (char *)ptrv
;
3109 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3110 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3113 WARN_ON(start
> eb
->len
);
3114 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3116 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3119 page
= extent_buffer_page(eb
, i
);
3120 WARN_ON(!PageUptodate(page
));
3122 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3124 kaddr
= kmap_atomic(page
, KM_USER0
);
3125 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3126 kunmap_atomic(kaddr
, KM_USER0
);
3137 EXPORT_SYMBOL(memcmp_extent_buffer
);
3139 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3140 unsigned long start
, unsigned long len
)
3146 char *src
= (char *)srcv
;
3147 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3148 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3150 WARN_ON(start
> eb
->len
);
3151 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3153 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3156 page
= extent_buffer_page(eb
, i
);
3157 WARN_ON(!PageUptodate(page
));
3159 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3160 kaddr
= kmap_atomic(page
, KM_USER1
);
3161 memcpy(kaddr
+ offset
, src
, cur
);
3162 kunmap_atomic(kaddr
, KM_USER1
);
3170 EXPORT_SYMBOL(write_extent_buffer
);
3172 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3173 unsigned long start
, unsigned long len
)
3179 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3180 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3182 WARN_ON(start
> eb
->len
);
3183 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3185 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3188 page
= extent_buffer_page(eb
, i
);
3189 WARN_ON(!PageUptodate(page
));
3191 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3192 kaddr
= kmap_atomic(page
, KM_USER0
);
3193 memset(kaddr
+ offset
, c
, cur
);
3194 kunmap_atomic(kaddr
, KM_USER0
);
3201 EXPORT_SYMBOL(memset_extent_buffer
);
3203 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3204 unsigned long dst_offset
, unsigned long src_offset
,
3207 u64 dst_len
= dst
->len
;
3212 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3213 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3215 WARN_ON(src
->len
!= dst_len
);
3217 offset
= (start_offset
+ dst_offset
) &
3218 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3221 page
= extent_buffer_page(dst
, i
);
3222 WARN_ON(!PageUptodate(page
));
3224 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3226 kaddr
= kmap_atomic(page
, KM_USER0
);
3227 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3228 kunmap_atomic(kaddr
, KM_USER0
);
3236 EXPORT_SYMBOL(copy_extent_buffer
);
3238 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3239 unsigned long dst_off
, unsigned long src_off
,
3242 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3243 if (dst_page
== src_page
) {
3244 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3246 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3247 char *p
= dst_kaddr
+ dst_off
+ len
;
3248 char *s
= src_kaddr
+ src_off
+ len
;
3253 kunmap_atomic(src_kaddr
, KM_USER1
);
3255 kunmap_atomic(dst_kaddr
, KM_USER0
);
3258 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3259 unsigned long dst_off
, unsigned long src_off
,
3262 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3265 if (dst_page
!= src_page
)
3266 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3268 src_kaddr
= dst_kaddr
;
3270 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3271 kunmap_atomic(dst_kaddr
, KM_USER0
);
3272 if (dst_page
!= src_page
)
3273 kunmap_atomic(src_kaddr
, KM_USER1
);
3276 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3277 unsigned long src_offset
, unsigned long len
)
3280 size_t dst_off_in_page
;
3281 size_t src_off_in_page
;
3282 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3283 unsigned long dst_i
;
3284 unsigned long src_i
;
3286 if (src_offset
+ len
> dst
->len
) {
3287 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3288 src_offset
, len
, dst
->len
);
3291 if (dst_offset
+ len
> dst
->len
) {
3292 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3293 dst_offset
, len
, dst
->len
);
3298 dst_off_in_page
= (start_offset
+ dst_offset
) &
3299 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3300 src_off_in_page
= (start_offset
+ src_offset
) &
3301 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3303 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3304 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3306 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3308 cur
= min_t(unsigned long, cur
,
3309 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3311 copy_pages(extent_buffer_page(dst
, dst_i
),
3312 extent_buffer_page(dst
, src_i
),
3313 dst_off_in_page
, src_off_in_page
, cur
);
3320 EXPORT_SYMBOL(memcpy_extent_buffer
);
3322 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3323 unsigned long src_offset
, unsigned long len
)
3326 size_t dst_off_in_page
;
3327 size_t src_off_in_page
;
3328 unsigned long dst_end
= dst_offset
+ len
- 1;
3329 unsigned long src_end
= src_offset
+ len
- 1;
3330 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3331 unsigned long dst_i
;
3332 unsigned long src_i
;
3334 if (src_offset
+ len
> dst
->len
) {
3335 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3336 src_offset
, len
, dst
->len
);
3339 if (dst_offset
+ len
> dst
->len
) {
3340 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3341 dst_offset
, len
, dst
->len
);
3344 if (dst_offset
< src_offset
) {
3345 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
3349 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
3350 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
3352 dst_off_in_page
= (start_offset
+ dst_end
) &
3353 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3354 src_off_in_page
= (start_offset
+ src_end
) &
3355 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3357 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
3358 cur
= min(cur
, dst_off_in_page
+ 1);
3359 move_pages(extent_buffer_page(dst
, dst_i
),
3360 extent_buffer_page(dst
, src_i
),
3361 dst_off_in_page
- cur
+ 1,
3362 src_off_in_page
- cur
+ 1, cur
);
3369 EXPORT_SYMBOL(memmove_extent_buffer
);