1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 static struct kmem_cache
*extent_state_cache
;
22 static struct kmem_cache
*extent_buffer_cache
;
24 static LIST_HEAD(buffers
);
25 static LIST_HEAD(states
);
29 static DEFINE_SPINLOCK(leak_lock
);
32 #define BUFFER_LRU_MAX 64
37 struct rb_node rb_node
;
40 struct extent_page_data
{
42 struct extent_io_tree
*tree
;
43 get_extent_t
*get_extent
;
45 /* tells writepage not to lock the state bits for this range
46 * it still does the unlocking
48 unsigned int extent_locked
:1;
50 /* tells the submit_bio code to use a WRITE_SYNC */
51 unsigned int sync_io
:1;
54 int __init
extent_io_init(void)
56 extent_state_cache
= kmem_cache_create("extent_state",
57 sizeof(struct extent_state
), 0,
58 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
59 if (!extent_state_cache
)
62 extent_buffer_cache
= kmem_cache_create("extent_buffers",
63 sizeof(struct extent_buffer
), 0,
64 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
65 if (!extent_buffer_cache
)
66 goto free_state_cache
;
70 kmem_cache_destroy(extent_state_cache
);
74 void extent_io_exit(void)
76 struct extent_state
*state
;
77 struct extent_buffer
*eb
;
79 while (!list_empty(&states
)) {
80 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
81 printk(KERN_ERR
"btrfs state leak: start %llu end %llu "
82 "state %lu in tree %p refs %d\n",
83 (unsigned long long)state
->start
,
84 (unsigned long long)state
->end
,
85 state
->state
, state
->tree
, atomic_read(&state
->refs
));
86 list_del(&state
->leak_list
);
87 kmem_cache_free(extent_state_cache
, state
);
91 while (!list_empty(&buffers
)) {
92 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
93 printk(KERN_ERR
"btrfs buffer leak start %llu len %lu "
94 "refs %d\n", (unsigned long long)eb
->start
,
95 eb
->len
, atomic_read(&eb
->refs
));
96 list_del(&eb
->leak_list
);
97 kmem_cache_free(extent_buffer_cache
, eb
);
99 if (extent_state_cache
)
100 kmem_cache_destroy(extent_state_cache
);
101 if (extent_buffer_cache
)
102 kmem_cache_destroy(extent_buffer_cache
);
105 void extent_io_tree_init(struct extent_io_tree
*tree
,
106 struct address_space
*mapping
)
108 tree
->state
= RB_ROOT
;
109 INIT_RADIX_TREE(&tree
->buffer
, GFP_ATOMIC
);
111 tree
->dirty_bytes
= 0;
112 spin_lock_init(&tree
->lock
);
113 spin_lock_init(&tree
->buffer_lock
);
114 tree
->mapping
= mapping
;
117 static struct extent_state
*alloc_extent_state(gfp_t mask
)
119 struct extent_state
*state
;
124 state
= kmem_cache_alloc(extent_state_cache
, mask
);
131 spin_lock_irqsave(&leak_lock
, flags
);
132 list_add(&state
->leak_list
, &states
);
133 spin_unlock_irqrestore(&leak_lock
, flags
);
135 atomic_set(&state
->refs
, 1);
136 init_waitqueue_head(&state
->wq
);
140 void free_extent_state(struct extent_state
*state
)
144 if (atomic_dec_and_test(&state
->refs
)) {
148 WARN_ON(state
->tree
);
150 spin_lock_irqsave(&leak_lock
, flags
);
151 list_del(&state
->leak_list
);
152 spin_unlock_irqrestore(&leak_lock
, flags
);
154 kmem_cache_free(extent_state_cache
, state
);
158 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
159 struct rb_node
*node
)
161 struct rb_node
**p
= &root
->rb_node
;
162 struct rb_node
*parent
= NULL
;
163 struct tree_entry
*entry
;
167 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
169 if (offset
< entry
->start
)
171 else if (offset
> entry
->end
)
177 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
178 rb_link_node(node
, parent
, p
);
179 rb_insert_color(node
, root
);
183 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
184 struct rb_node
**prev_ret
,
185 struct rb_node
**next_ret
)
187 struct rb_root
*root
= &tree
->state
;
188 struct rb_node
*n
= root
->rb_node
;
189 struct rb_node
*prev
= NULL
;
190 struct rb_node
*orig_prev
= NULL
;
191 struct tree_entry
*entry
;
192 struct tree_entry
*prev_entry
= NULL
;
195 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
199 if (offset
< entry
->start
)
201 else if (offset
> entry
->end
)
209 while (prev
&& offset
> prev_entry
->end
) {
210 prev
= rb_next(prev
);
211 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
218 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
219 while (prev
&& offset
< prev_entry
->start
) {
220 prev
= rb_prev(prev
);
221 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
228 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
231 struct rb_node
*prev
= NULL
;
234 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
240 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
241 struct extent_state
*other
)
243 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
244 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
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 void 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
| EXTENT_BOUNDARY
))
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 merge_cb(tree
, state
, other
);
272 state
->start
= other
->start
;
274 rb_erase(&other
->rb_node
, &tree
->state
);
275 free_extent_state(other
);
278 other_node
= rb_next(&state
->rb_node
);
280 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
281 if (other
->start
== state
->end
+ 1 &&
282 other
->state
== state
->state
) {
283 merge_cb(tree
, state
, other
);
284 state
->end
= other
->end
;
286 rb_erase(&other
->rb_node
, &tree
->state
);
287 free_extent_state(other
);
292 static void set_state_cb(struct extent_io_tree
*tree
,
293 struct extent_state
*state
, int *bits
)
295 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
296 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
299 static void clear_state_cb(struct extent_io_tree
*tree
,
300 struct extent_state
*state
, int *bits
)
302 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
303 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
306 static void set_state_bits(struct extent_io_tree
*tree
,
307 struct extent_state
*state
, int *bits
);
310 * insert an extent_state struct into the tree. 'bits' are set on the
311 * struct before it is inserted.
313 * This may return -EEXIST if the extent is already there, in which case the
314 * state struct is freed.
316 * The tree lock is not taken internally. This is a utility function and
317 * probably isn't what you want to call (see set/clear_extent_bit).
319 static int insert_state(struct extent_io_tree
*tree
,
320 struct extent_state
*state
, u64 start
, u64 end
,
323 struct rb_node
*node
;
326 printk(KERN_ERR
"btrfs end < start %llu %llu\n",
327 (unsigned long long)end
,
328 (unsigned long long)start
);
331 state
->start
= start
;
334 set_state_bits(tree
, state
, bits
);
336 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
338 struct extent_state
*found
;
339 found
= rb_entry(node
, struct extent_state
, rb_node
);
340 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
341 "%llu %llu\n", (unsigned long long)found
->start
,
342 (unsigned long long)found
->end
,
343 (unsigned long long)start
, (unsigned long long)end
);
347 merge_state(tree
, state
);
351 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
354 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
355 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
359 * split a given extent state struct in two, inserting the preallocated
360 * struct 'prealloc' as the newly created second half. 'split' indicates an
361 * offset inside 'orig' where it should be split.
364 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
365 * are two extent state structs in the tree:
366 * prealloc: [orig->start, split - 1]
367 * orig: [ split, orig->end ]
369 * The tree locks are not taken by this function. They need to be held
372 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
373 struct extent_state
*prealloc
, u64 split
)
375 struct rb_node
*node
;
377 split_cb(tree
, orig
, split
);
379 prealloc
->start
= orig
->start
;
380 prealloc
->end
= split
- 1;
381 prealloc
->state
= orig
->state
;
384 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
386 free_extent_state(prealloc
);
389 prealloc
->tree
= tree
;
394 * utility function to clear some bits in an extent state struct.
395 * it will optionally wake up any one waiting on this state (wake == 1), or
396 * forcibly remove the state from the tree (delete == 1).
398 * If no bits are set on the state struct after clearing things, the
399 * struct is freed and removed from the tree
401 static int clear_state_bit(struct extent_io_tree
*tree
,
402 struct extent_state
*state
,
405 int bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
406 int ret
= state
->state
& bits_to_clear
;
408 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
409 u64 range
= state
->end
- state
->start
+ 1;
410 WARN_ON(range
> tree
->dirty_bytes
);
411 tree
->dirty_bytes
-= range
;
413 clear_state_cb(tree
, state
, bits
);
414 state
->state
&= ~bits_to_clear
;
417 if (state
->state
== 0) {
419 rb_erase(&state
->rb_node
, &tree
->state
);
421 free_extent_state(state
);
426 merge_state(tree
, state
);
431 static struct extent_state
*
432 alloc_extent_state_atomic(struct extent_state
*prealloc
)
435 prealloc
= alloc_extent_state(GFP_ATOMIC
);
441 * clear some bits on a range in the tree. This may require splitting
442 * or inserting elements in the tree, so the gfp mask is used to
443 * indicate which allocations or sleeping are allowed.
445 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
446 * the given range from the tree regardless of state (ie for truncate).
448 * the range [start, end] is inclusive.
450 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
451 * bits were already set, or zero if none of the bits were already set.
453 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
454 int bits
, int wake
, int delete,
455 struct extent_state
**cached_state
,
458 struct extent_state
*state
;
459 struct extent_state
*cached
;
460 struct extent_state
*prealloc
= NULL
;
461 struct rb_node
*next_node
;
462 struct rb_node
*node
;
469 bits
|= ~EXTENT_CTLBITS
;
470 bits
|= EXTENT_FIRST_DELALLOC
;
472 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
475 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
476 prealloc
= alloc_extent_state(mask
);
481 spin_lock(&tree
->lock
);
483 cached
= *cached_state
;
486 *cached_state
= NULL
;
490 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
491 cached
->end
> start
) {
493 atomic_dec(&cached
->refs
);
498 free_extent_state(cached
);
501 * this search will find the extents that end after
504 node
= tree_search(tree
, start
);
507 state
= rb_entry(node
, struct extent_state
, rb_node
);
509 if (state
->start
> end
)
511 WARN_ON(state
->end
< start
);
512 last_end
= state
->end
;
515 * | ---- desired range ---- |
517 * | ------------- state -------------- |
519 * We need to split the extent we found, and may flip
520 * bits on second half.
522 * If the extent we found extends past our range, we
523 * just split and search again. It'll get split again
524 * the next time though.
526 * If the extent we found is inside our range, we clear
527 * the desired bit on it.
530 if (state
->start
< start
) {
531 prealloc
= alloc_extent_state_atomic(prealloc
);
533 err
= split_state(tree
, state
, prealloc
, start
);
534 BUG_ON(err
== -EEXIST
);
538 if (state
->end
<= end
) {
539 set
|= clear_state_bit(tree
, state
, &bits
, wake
);
540 if (last_end
== (u64
)-1)
542 start
= last_end
+ 1;
547 * | ---- desired range ---- |
549 * We need to split the extent, and clear the bit
552 if (state
->start
<= end
&& state
->end
> end
) {
553 prealloc
= alloc_extent_state_atomic(prealloc
);
555 err
= split_state(tree
, state
, prealloc
, end
+ 1);
556 BUG_ON(err
== -EEXIST
);
560 set
|= clear_state_bit(tree
, prealloc
, &bits
, wake
);
566 if (state
->end
< end
&& prealloc
&& !need_resched())
567 next_node
= rb_next(&state
->rb_node
);
571 set
|= clear_state_bit(tree
, state
, &bits
, wake
);
572 if (last_end
== (u64
)-1)
574 start
= last_end
+ 1;
575 if (start
<= end
&& next_node
) {
576 state
= rb_entry(next_node
, struct extent_state
,
578 if (state
->start
== start
)
584 spin_unlock(&tree
->lock
);
586 free_extent_state(prealloc
);
593 spin_unlock(&tree
->lock
);
594 if (mask
& __GFP_WAIT
)
599 static int wait_on_state(struct extent_io_tree
*tree
,
600 struct extent_state
*state
)
601 __releases(tree
->lock
)
602 __acquires(tree
->lock
)
605 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
606 spin_unlock(&tree
->lock
);
608 spin_lock(&tree
->lock
);
609 finish_wait(&state
->wq
, &wait
);
614 * waits for one or more bits to clear on a range in the state tree.
615 * The range [start, end] is inclusive.
616 * The tree lock is taken by this function
618 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
620 struct extent_state
*state
;
621 struct rb_node
*node
;
623 spin_lock(&tree
->lock
);
627 * this search will find all the extents that end after
630 node
= tree_search(tree
, start
);
634 state
= rb_entry(node
, struct extent_state
, rb_node
);
636 if (state
->start
> end
)
639 if (state
->state
& bits
) {
640 start
= state
->start
;
641 atomic_inc(&state
->refs
);
642 wait_on_state(tree
, state
);
643 free_extent_state(state
);
646 start
= state
->end
+ 1;
651 cond_resched_lock(&tree
->lock
);
654 spin_unlock(&tree
->lock
);
658 static void set_state_bits(struct extent_io_tree
*tree
,
659 struct extent_state
*state
,
662 int bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
664 set_state_cb(tree
, state
, bits
);
665 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
666 u64 range
= state
->end
- state
->start
+ 1;
667 tree
->dirty_bytes
+= range
;
669 state
->state
|= bits_to_set
;
672 static void cache_state(struct extent_state
*state
,
673 struct extent_state
**cached_ptr
)
675 if (cached_ptr
&& !(*cached_ptr
)) {
676 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
678 atomic_inc(&state
->refs
);
683 static void uncache_state(struct extent_state
**cached_ptr
)
685 if (cached_ptr
&& (*cached_ptr
)) {
686 struct extent_state
*state
= *cached_ptr
;
688 free_extent_state(state
);
693 * set some bits on a range in the tree. This may require allocations or
694 * sleeping, so the gfp mask is used to indicate what is allowed.
696 * If any of the exclusive bits are set, this will fail with -EEXIST if some
697 * part of the range already has the desired bits set. The start of the
698 * existing range is returned in failed_start in this case.
700 * [start, end] is inclusive This takes the tree lock.
703 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
704 int bits
, int exclusive_bits
, u64
*failed_start
,
705 struct extent_state
**cached_state
, gfp_t mask
)
707 struct extent_state
*state
;
708 struct extent_state
*prealloc
= NULL
;
709 struct rb_node
*node
;
714 bits
|= EXTENT_FIRST_DELALLOC
;
716 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
717 prealloc
= alloc_extent_state(mask
);
721 spin_lock(&tree
->lock
);
722 if (cached_state
&& *cached_state
) {
723 state
= *cached_state
;
724 if (state
->start
<= start
&& state
->end
> start
&&
726 node
= &state
->rb_node
;
731 * this search will find all the extents that end after
734 node
= tree_search(tree
, start
);
736 prealloc
= alloc_extent_state_atomic(prealloc
);
738 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
740 BUG_ON(err
== -EEXIST
);
743 state
= rb_entry(node
, struct extent_state
, rb_node
);
745 last_start
= state
->start
;
746 last_end
= state
->end
;
749 * | ---- desired range ---- |
752 * Just lock what we found and keep going
754 if (state
->start
== start
&& state
->end
<= end
) {
755 struct rb_node
*next_node
;
756 if (state
->state
& exclusive_bits
) {
757 *failed_start
= state
->start
;
762 set_state_bits(tree
, state
, &bits
);
764 cache_state(state
, cached_state
);
765 merge_state(tree
, state
);
766 if (last_end
== (u64
)-1)
769 start
= last_end
+ 1;
770 next_node
= rb_next(&state
->rb_node
);
771 if (next_node
&& start
< end
&& prealloc
&& !need_resched()) {
772 state
= rb_entry(next_node
, struct extent_state
,
774 if (state
->start
== start
)
781 * | ---- desired range ---- |
784 * | ------------- state -------------- |
786 * We need to split the extent we found, and may flip bits on
789 * If the extent we found extends past our
790 * range, we just split and search again. It'll get split
791 * again the next time though.
793 * If the extent we found is inside our range, we set the
796 if (state
->start
< start
) {
797 if (state
->state
& exclusive_bits
) {
798 *failed_start
= start
;
803 prealloc
= alloc_extent_state_atomic(prealloc
);
805 err
= split_state(tree
, state
, prealloc
, start
);
806 BUG_ON(err
== -EEXIST
);
810 if (state
->end
<= end
) {
811 set_state_bits(tree
, state
, &bits
);
812 cache_state(state
, cached_state
);
813 merge_state(tree
, state
);
814 if (last_end
== (u64
)-1)
816 start
= last_end
+ 1;
821 * | ---- desired range ---- |
822 * | state | or | state |
824 * There's a hole, we need to insert something in it and
825 * ignore the extent we found.
827 if (state
->start
> start
) {
829 if (end
< last_start
)
832 this_end
= last_start
- 1;
834 prealloc
= alloc_extent_state_atomic(prealloc
);
838 * Avoid to free 'prealloc' if it can be merged with
841 err
= insert_state(tree
, prealloc
, start
, this_end
,
843 BUG_ON(err
== -EEXIST
);
845 free_extent_state(prealloc
);
849 cache_state(prealloc
, cached_state
);
851 start
= this_end
+ 1;
855 * | ---- desired range ---- |
857 * We need to split the extent, and set the bit
860 if (state
->start
<= end
&& state
->end
> end
) {
861 if (state
->state
& exclusive_bits
) {
862 *failed_start
= start
;
867 prealloc
= alloc_extent_state_atomic(prealloc
);
869 err
= split_state(tree
, state
, prealloc
, end
+ 1);
870 BUG_ON(err
== -EEXIST
);
872 set_state_bits(tree
, prealloc
, &bits
);
873 cache_state(prealloc
, cached_state
);
874 merge_state(tree
, prealloc
);
882 spin_unlock(&tree
->lock
);
884 free_extent_state(prealloc
);
891 spin_unlock(&tree
->lock
);
892 if (mask
& __GFP_WAIT
)
898 * convert_extent - convert all bits in a given range from one bit to another
899 * @tree: the io tree to search
900 * @start: the start offset in bytes
901 * @end: the end offset in bytes (inclusive)
902 * @bits: the bits to set in this range
903 * @clear_bits: the bits to clear in this range
904 * @mask: the allocation mask
906 * This will go through and set bits for the given range. If any states exist
907 * already in this range they are set with the given bit and cleared of the
908 * clear_bits. This is only meant to be used by things that are mergeable, ie
909 * converting from say DELALLOC to DIRTY. This is not meant to be used with
910 * boundary bits like LOCK.
912 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
913 int bits
, int clear_bits
, gfp_t mask
)
915 struct extent_state
*state
;
916 struct extent_state
*prealloc
= NULL
;
917 struct rb_node
*node
;
923 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
924 prealloc
= alloc_extent_state(mask
);
929 spin_lock(&tree
->lock
);
931 * this search will find all the extents that end after
934 node
= tree_search(tree
, start
);
936 prealloc
= alloc_extent_state_atomic(prealloc
);
939 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
941 BUG_ON(err
== -EEXIST
);
944 state
= rb_entry(node
, struct extent_state
, rb_node
);
946 last_start
= state
->start
;
947 last_end
= state
->end
;
950 * | ---- desired range ---- |
953 * Just lock what we found and keep going
955 if (state
->start
== start
&& state
->end
<= end
) {
956 struct rb_node
*next_node
;
958 set_state_bits(tree
, state
, &bits
);
959 clear_state_bit(tree
, state
, &clear_bits
, 0);
961 merge_state(tree
, state
);
962 if (last_end
== (u64
)-1)
965 start
= last_end
+ 1;
966 next_node
= rb_next(&state
->rb_node
);
967 if (next_node
&& start
< end
&& prealloc
&& !need_resched()) {
968 state
= rb_entry(next_node
, struct extent_state
,
970 if (state
->start
== start
)
977 * | ---- desired range ---- |
980 * | ------------- state -------------- |
982 * We need to split the extent we found, and may flip bits on
985 * If the extent we found extends past our
986 * range, we just split and search again. It'll get split
987 * again the next time though.
989 * If the extent we found is inside our range, we set the
992 if (state
->start
< start
) {
993 prealloc
= alloc_extent_state_atomic(prealloc
);
996 err
= split_state(tree
, state
, prealloc
, start
);
997 BUG_ON(err
== -EEXIST
);
1001 if (state
->end
<= end
) {
1002 set_state_bits(tree
, state
, &bits
);
1003 clear_state_bit(tree
, state
, &clear_bits
, 0);
1004 merge_state(tree
, state
);
1005 if (last_end
== (u64
)-1)
1007 start
= last_end
+ 1;
1012 * | ---- desired range ---- |
1013 * | state | or | state |
1015 * There's a hole, we need to insert something in it and
1016 * ignore the extent we found.
1018 if (state
->start
> start
) {
1020 if (end
< last_start
)
1023 this_end
= last_start
- 1;
1025 prealloc
= alloc_extent_state_atomic(prealloc
);
1030 * Avoid to free 'prealloc' if it can be merged with
1033 err
= insert_state(tree
, prealloc
, start
, this_end
,
1035 BUG_ON(err
== -EEXIST
);
1037 free_extent_state(prealloc
);
1042 start
= this_end
+ 1;
1046 * | ---- desired range ---- |
1048 * We need to split the extent, and set the bit
1051 if (state
->start
<= end
&& state
->end
> end
) {
1052 prealloc
= alloc_extent_state_atomic(prealloc
);
1056 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1057 BUG_ON(err
== -EEXIST
);
1059 set_state_bits(tree
, prealloc
, &bits
);
1060 clear_state_bit(tree
, prealloc
, &clear_bits
, 0);
1062 merge_state(tree
, prealloc
);
1070 spin_unlock(&tree
->lock
);
1072 free_extent_state(prealloc
);
1079 spin_unlock(&tree
->lock
);
1080 if (mask
& __GFP_WAIT
)
1085 /* wrappers around set/clear extent bit */
1086 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1089 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
1093 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1094 int bits
, gfp_t mask
)
1096 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
1100 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1101 int bits
, gfp_t mask
)
1103 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
1106 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1107 struct extent_state
**cached_state
, gfp_t mask
)
1109 return set_extent_bit(tree
, start
, end
,
1110 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1111 0, NULL
, cached_state
, mask
);
1114 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1117 return clear_extent_bit(tree
, start
, end
,
1118 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1119 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1122 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1125 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
1129 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1130 struct extent_state
**cached_state
, gfp_t mask
)
1132 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0,
1133 NULL
, cached_state
, mask
);
1136 static int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
,
1137 u64 end
, struct extent_state
**cached_state
,
1140 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1141 cached_state
, mask
);
1145 * either insert or lock state struct between start and end use mask to tell
1146 * us if waiting is desired.
1148 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1149 int bits
, struct extent_state
**cached_state
, gfp_t mask
)
1154 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1155 EXTENT_LOCKED
, &failed_start
,
1156 cached_state
, mask
);
1157 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
1158 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1159 start
= failed_start
;
1163 WARN_ON(start
> end
);
1168 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
1170 return lock_extent_bits(tree
, start
, end
, 0, NULL
, mask
);
1173 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1179 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1180 &failed_start
, NULL
, mask
);
1181 if (err
== -EEXIST
) {
1182 if (failed_start
> start
)
1183 clear_extent_bit(tree
, start
, failed_start
- 1,
1184 EXTENT_LOCKED
, 1, 0, NULL
, mask
);
1190 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1191 struct extent_state
**cached
, gfp_t mask
)
1193 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1197 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
1199 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1204 * helper function to set both pages and extents in the tree writeback
1206 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1208 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1209 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1212 while (index
<= end_index
) {
1213 page
= find_get_page(tree
->mapping
, index
);
1215 set_page_writeback(page
);
1216 page_cache_release(page
);
1222 /* find the first state struct with 'bits' set after 'start', and
1223 * return it. tree->lock must be held. NULL will returned if
1224 * nothing was found after 'start'
1226 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
1227 u64 start
, int bits
)
1229 struct rb_node
*node
;
1230 struct extent_state
*state
;
1233 * this search will find all the extents that end after
1236 node
= tree_search(tree
, start
);
1241 state
= rb_entry(node
, struct extent_state
, rb_node
);
1242 if (state
->end
>= start
&& (state
->state
& bits
))
1245 node
= rb_next(node
);
1254 * find the first offset in the io tree with 'bits' set. zero is
1255 * returned if we find something, and *start_ret and *end_ret are
1256 * set to reflect the state struct that was found.
1258 * If nothing was found, 1 is returned, < 0 on error
1260 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1261 u64
*start_ret
, u64
*end_ret
, int bits
)
1263 struct extent_state
*state
;
1266 spin_lock(&tree
->lock
);
1267 state
= find_first_extent_bit_state(tree
, start
, bits
);
1269 *start_ret
= state
->start
;
1270 *end_ret
= state
->end
;
1273 spin_unlock(&tree
->lock
);
1278 * find a contiguous range of bytes in the file marked as delalloc, not
1279 * more than 'max_bytes'. start and end are used to return the range,
1281 * 1 is returned if we find something, 0 if nothing was in the tree
1283 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1284 u64
*start
, u64
*end
, u64 max_bytes
,
1285 struct extent_state
**cached_state
)
1287 struct rb_node
*node
;
1288 struct extent_state
*state
;
1289 u64 cur_start
= *start
;
1291 u64 total_bytes
= 0;
1293 spin_lock(&tree
->lock
);
1296 * this search will find all the extents that end after
1299 node
= tree_search(tree
, cur_start
);
1307 state
= rb_entry(node
, struct extent_state
, rb_node
);
1308 if (found
&& (state
->start
!= cur_start
||
1309 (state
->state
& EXTENT_BOUNDARY
))) {
1312 if (!(state
->state
& EXTENT_DELALLOC
)) {
1318 *start
= state
->start
;
1319 *cached_state
= state
;
1320 atomic_inc(&state
->refs
);
1324 cur_start
= state
->end
+ 1;
1325 node
= rb_next(node
);
1328 total_bytes
+= state
->end
- state
->start
+ 1;
1329 if (total_bytes
>= max_bytes
)
1333 spin_unlock(&tree
->lock
);
1337 static noinline
int __unlock_for_delalloc(struct inode
*inode
,
1338 struct page
*locked_page
,
1342 struct page
*pages
[16];
1343 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1344 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1345 unsigned long nr_pages
= end_index
- index
+ 1;
1348 if (index
== locked_page
->index
&& end_index
== index
)
1351 while (nr_pages
> 0) {
1352 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1353 min_t(unsigned long, nr_pages
,
1354 ARRAY_SIZE(pages
)), pages
);
1355 for (i
= 0; i
< ret
; i
++) {
1356 if (pages
[i
] != locked_page
)
1357 unlock_page(pages
[i
]);
1358 page_cache_release(pages
[i
]);
1367 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1368 struct page
*locked_page
,
1372 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1373 unsigned long start_index
= index
;
1374 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1375 unsigned long pages_locked
= 0;
1376 struct page
*pages
[16];
1377 unsigned long nrpages
;
1381 /* the caller is responsible for locking the start index */
1382 if (index
== locked_page
->index
&& index
== end_index
)
1385 /* skip the page at the start index */
1386 nrpages
= end_index
- index
+ 1;
1387 while (nrpages
> 0) {
1388 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1389 min_t(unsigned long,
1390 nrpages
, ARRAY_SIZE(pages
)), pages
);
1395 /* now we have an array of pages, lock them all */
1396 for (i
= 0; i
< ret
; i
++) {
1398 * the caller is taking responsibility for
1401 if (pages
[i
] != locked_page
) {
1402 lock_page(pages
[i
]);
1403 if (!PageDirty(pages
[i
]) ||
1404 pages
[i
]->mapping
!= inode
->i_mapping
) {
1406 unlock_page(pages
[i
]);
1407 page_cache_release(pages
[i
]);
1411 page_cache_release(pages
[i
]);
1420 if (ret
&& pages_locked
) {
1421 __unlock_for_delalloc(inode
, locked_page
,
1423 ((u64
)(start_index
+ pages_locked
- 1)) <<
1430 * find a contiguous range of bytes in the file marked as delalloc, not
1431 * more than 'max_bytes'. start and end are used to return the range,
1433 * 1 is returned if we find something, 0 if nothing was in the tree
1435 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1436 struct extent_io_tree
*tree
,
1437 struct page
*locked_page
,
1438 u64
*start
, u64
*end
,
1444 struct extent_state
*cached_state
= NULL
;
1449 /* step one, find a bunch of delalloc bytes starting at start */
1450 delalloc_start
= *start
;
1452 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1453 max_bytes
, &cached_state
);
1454 if (!found
|| delalloc_end
<= *start
) {
1455 *start
= delalloc_start
;
1456 *end
= delalloc_end
;
1457 free_extent_state(cached_state
);
1462 * start comes from the offset of locked_page. We have to lock
1463 * pages in order, so we can't process delalloc bytes before
1466 if (delalloc_start
< *start
)
1467 delalloc_start
= *start
;
1470 * make sure to limit the number of pages we try to lock down
1473 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
)
1474 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1476 /* step two, lock all the pages after the page that has start */
1477 ret
= lock_delalloc_pages(inode
, locked_page
,
1478 delalloc_start
, delalloc_end
);
1479 if (ret
== -EAGAIN
) {
1480 /* some of the pages are gone, lets avoid looping by
1481 * shortening the size of the delalloc range we're searching
1483 free_extent_state(cached_state
);
1485 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1486 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1496 /* step three, lock the state bits for the whole range */
1497 lock_extent_bits(tree
, delalloc_start
, delalloc_end
,
1498 0, &cached_state
, GFP_NOFS
);
1500 /* then test to make sure it is all still delalloc */
1501 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1502 EXTENT_DELALLOC
, 1, cached_state
);
1504 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1505 &cached_state
, GFP_NOFS
);
1506 __unlock_for_delalloc(inode
, locked_page
,
1507 delalloc_start
, delalloc_end
);
1511 free_extent_state(cached_state
);
1512 *start
= delalloc_start
;
1513 *end
= delalloc_end
;
1518 int extent_clear_unlock_delalloc(struct inode
*inode
,
1519 struct extent_io_tree
*tree
,
1520 u64 start
, u64 end
, struct page
*locked_page
,
1524 struct page
*pages
[16];
1525 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1526 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1527 unsigned long nr_pages
= end_index
- index
+ 1;
1531 if (op
& EXTENT_CLEAR_UNLOCK
)
1532 clear_bits
|= EXTENT_LOCKED
;
1533 if (op
& EXTENT_CLEAR_DIRTY
)
1534 clear_bits
|= EXTENT_DIRTY
;
1536 if (op
& EXTENT_CLEAR_DELALLOC
)
1537 clear_bits
|= EXTENT_DELALLOC
;
1539 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1540 if (!(op
& (EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
1541 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
|
1542 EXTENT_SET_PRIVATE2
)))
1545 while (nr_pages
> 0) {
1546 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1547 min_t(unsigned long,
1548 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1549 for (i
= 0; i
< ret
; i
++) {
1551 if (op
& EXTENT_SET_PRIVATE2
)
1552 SetPagePrivate2(pages
[i
]);
1554 if (pages
[i
] == locked_page
) {
1555 page_cache_release(pages
[i
]);
1558 if (op
& EXTENT_CLEAR_DIRTY
)
1559 clear_page_dirty_for_io(pages
[i
]);
1560 if (op
& EXTENT_SET_WRITEBACK
)
1561 set_page_writeback(pages
[i
]);
1562 if (op
& EXTENT_END_WRITEBACK
)
1563 end_page_writeback(pages
[i
]);
1564 if (op
& EXTENT_CLEAR_UNLOCK_PAGE
)
1565 unlock_page(pages
[i
]);
1566 page_cache_release(pages
[i
]);
1576 * count the number of bytes in the tree that have a given bit(s)
1577 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1578 * cached. The total number found is returned.
1580 u64
count_range_bits(struct extent_io_tree
*tree
,
1581 u64
*start
, u64 search_end
, u64 max_bytes
,
1582 unsigned long bits
, int contig
)
1584 struct rb_node
*node
;
1585 struct extent_state
*state
;
1586 u64 cur_start
= *start
;
1587 u64 total_bytes
= 0;
1591 if (search_end
<= cur_start
) {
1596 spin_lock(&tree
->lock
);
1597 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1598 total_bytes
= tree
->dirty_bytes
;
1602 * this search will find all the extents that end after
1605 node
= tree_search(tree
, cur_start
);
1610 state
= rb_entry(node
, struct extent_state
, rb_node
);
1611 if (state
->start
> search_end
)
1613 if (contig
&& found
&& state
->start
> last
+ 1)
1615 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1616 total_bytes
+= min(search_end
, state
->end
) + 1 -
1617 max(cur_start
, state
->start
);
1618 if (total_bytes
>= max_bytes
)
1621 *start
= max(cur_start
, state
->start
);
1625 } else if (contig
&& found
) {
1628 node
= rb_next(node
);
1633 spin_unlock(&tree
->lock
);
1638 * set the private field for a given byte offset in the tree. If there isn't
1639 * an extent_state there already, this does nothing.
1641 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1643 struct rb_node
*node
;
1644 struct extent_state
*state
;
1647 spin_lock(&tree
->lock
);
1649 * this search will find all the extents that end after
1652 node
= tree_search(tree
, start
);
1657 state
= rb_entry(node
, struct extent_state
, rb_node
);
1658 if (state
->start
!= start
) {
1662 state
->private = private;
1664 spin_unlock(&tree
->lock
);
1668 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1670 struct rb_node
*node
;
1671 struct extent_state
*state
;
1674 spin_lock(&tree
->lock
);
1676 * this search will find all the extents that end after
1679 node
= tree_search(tree
, start
);
1684 state
= rb_entry(node
, struct extent_state
, rb_node
);
1685 if (state
->start
!= start
) {
1689 *private = state
->private;
1691 spin_unlock(&tree
->lock
);
1696 * searches a range in the state tree for a given mask.
1697 * If 'filled' == 1, this returns 1 only if every extent in the tree
1698 * has the bits set. Otherwise, 1 is returned if any bit in the
1699 * range is found set.
1701 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1702 int bits
, int filled
, struct extent_state
*cached
)
1704 struct extent_state
*state
= NULL
;
1705 struct rb_node
*node
;
1708 spin_lock(&tree
->lock
);
1709 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1710 cached
->end
> start
)
1711 node
= &cached
->rb_node
;
1713 node
= tree_search(tree
, start
);
1714 while (node
&& start
<= end
) {
1715 state
= rb_entry(node
, struct extent_state
, rb_node
);
1717 if (filled
&& state
->start
> start
) {
1722 if (state
->start
> end
)
1725 if (state
->state
& bits
) {
1729 } else if (filled
) {
1734 if (state
->end
== (u64
)-1)
1737 start
= state
->end
+ 1;
1740 node
= rb_next(node
);
1747 spin_unlock(&tree
->lock
);
1752 * helper function to set a given page up to date if all the
1753 * extents in the tree for that page are up to date
1755 static int check_page_uptodate(struct extent_io_tree
*tree
,
1758 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1759 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1760 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1761 SetPageUptodate(page
);
1766 * helper function to unlock a page if all the extents in the tree
1767 * for that page are unlocked
1769 static int check_page_locked(struct extent_io_tree
*tree
,
1772 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1773 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1774 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0, NULL
))
1780 * helper function to end page writeback if all the extents
1781 * in the tree for that page are done with writeback
1783 static int check_page_writeback(struct extent_io_tree
*tree
,
1786 end_page_writeback(page
);
1790 /* lots and lots of room for performance fixes in the end_bio funcs */
1793 * after a writepage IO is done, we need to:
1794 * clear the uptodate bits on error
1795 * clear the writeback bits in the extent tree for this IO
1796 * end_page_writeback if the page has no more pending IO
1798 * Scheduling is not allowed, so the extent state tree is expected
1799 * to have one and only one object corresponding to this IO.
1801 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1803 int uptodate
= err
== 0;
1804 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1805 struct extent_io_tree
*tree
;
1812 struct page
*page
= bvec
->bv_page
;
1813 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1815 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1817 end
= start
+ bvec
->bv_len
- 1;
1819 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1824 if (--bvec
>= bio
->bi_io_vec
)
1825 prefetchw(&bvec
->bv_page
->flags
);
1826 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1827 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
1828 end
, NULL
, uptodate
);
1833 if (!uptodate
&& tree
->ops
&&
1834 tree
->ops
->writepage_io_failed_hook
) {
1835 ret
= tree
->ops
->writepage_io_failed_hook(bio
, page
,
1838 uptodate
= (err
== 0);
1844 clear_extent_uptodate(tree
, start
, end
, NULL
, GFP_NOFS
);
1845 ClearPageUptodate(page
);
1850 end_page_writeback(page
);
1852 check_page_writeback(tree
, page
);
1853 } while (bvec
>= bio
->bi_io_vec
);
1859 * after a readpage IO is done, we need to:
1860 * clear the uptodate bits on error
1861 * set the uptodate bits if things worked
1862 * set the page up to date if all extents in the tree are uptodate
1863 * clear the lock bit in the extent tree
1864 * unlock the page if there are no other extents locked for it
1866 * Scheduling is not allowed, so the extent state tree is expected
1867 * to have one and only one object corresponding to this IO.
1869 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1871 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1872 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1873 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1874 struct extent_io_tree
*tree
;
1884 struct page
*page
= bvec
->bv_page
;
1885 struct extent_state
*cached
= NULL
;
1886 struct extent_state
*state
;
1888 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1890 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1892 end
= start
+ bvec
->bv_len
- 1;
1894 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1899 if (++bvec
<= bvec_end
)
1900 prefetchw(&bvec
->bv_page
->flags
);
1902 spin_lock(&tree
->lock
);
1903 state
= find_first_extent_bit_state(tree
, start
, EXTENT_LOCKED
);
1904 if (state
&& state
->start
== start
) {
1906 * take a reference on the state, unlock will drop
1909 cache_state(state
, &cached
);
1911 spin_unlock(&tree
->lock
);
1913 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1914 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1919 if (!uptodate
&& tree
->ops
&&
1920 tree
->ops
->readpage_io_failed_hook
) {
1921 ret
= tree
->ops
->readpage_io_failed_hook(bio
, page
,
1925 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1928 uncache_state(&cached
);
1934 set_extent_uptodate(tree
, start
, end
, &cached
,
1937 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
1941 SetPageUptodate(page
);
1943 ClearPageUptodate(page
);
1949 check_page_uptodate(tree
, page
);
1951 ClearPageUptodate(page
);
1954 check_page_locked(tree
, page
);
1956 } while (bvec
<= bvec_end
);
1962 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1967 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1969 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1970 while (!bio
&& (nr_vecs
/= 2))
1971 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1976 bio
->bi_bdev
= bdev
;
1977 bio
->bi_sector
= first_sector
;
1982 static int submit_one_bio(int rw
, struct bio
*bio
, int mirror_num
,
1983 unsigned long bio_flags
)
1986 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1987 struct page
*page
= bvec
->bv_page
;
1988 struct extent_io_tree
*tree
= bio
->bi_private
;
1991 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1993 bio
->bi_private
= NULL
;
1997 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
1998 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
1999 mirror_num
, bio_flags
, start
);
2001 submit_bio(rw
, bio
);
2002 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2008 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2009 struct page
*page
, sector_t sector
,
2010 size_t size
, unsigned long offset
,
2011 struct block_device
*bdev
,
2012 struct bio
**bio_ret
,
2013 unsigned long max_pages
,
2014 bio_end_io_t end_io_func
,
2016 unsigned long prev_bio_flags
,
2017 unsigned long bio_flags
)
2023 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2024 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2025 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2027 if (bio_ret
&& *bio_ret
) {
2030 contig
= bio
->bi_sector
== sector
;
2032 contig
= bio
->bi_sector
+ (bio
->bi_size
>> 9) ==
2035 if (prev_bio_flags
!= bio_flags
|| !contig
||
2036 (tree
->ops
&& tree
->ops
->merge_bio_hook
&&
2037 tree
->ops
->merge_bio_hook(page
, offset
, page_size
, bio
,
2039 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2040 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2047 if (this_compressed
)
2050 nr
= bio_get_nr_vecs(bdev
);
2052 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2056 bio_add_page(bio
, page
, page_size
, offset
);
2057 bio
->bi_end_io
= end_io_func
;
2058 bio
->bi_private
= tree
;
2063 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2068 void set_page_extent_mapped(struct page
*page
)
2070 if (!PagePrivate(page
)) {
2071 SetPagePrivate(page
);
2072 page_cache_get(page
);
2073 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2077 static void set_page_extent_head(struct page
*page
, unsigned long len
)
2079 WARN_ON(!PagePrivate(page
));
2080 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
2084 * basic readpage implementation. Locked extent state structs are inserted
2085 * into the tree that are removed when the IO is done (by the end_io
2088 static int __extent_read_full_page(struct extent_io_tree
*tree
,
2090 get_extent_t
*get_extent
,
2091 struct bio
**bio
, int mirror_num
,
2092 unsigned long *bio_flags
)
2094 struct inode
*inode
= page
->mapping
->host
;
2095 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2096 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2100 u64 last_byte
= i_size_read(inode
);
2104 struct extent_map
*em
;
2105 struct block_device
*bdev
;
2106 struct btrfs_ordered_extent
*ordered
;
2109 size_t pg_offset
= 0;
2111 size_t disk_io_size
;
2112 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2113 unsigned long this_bio_flag
= 0;
2115 set_page_extent_mapped(page
);
2117 if (!PageUptodate(page
)) {
2118 if (cleancache_get_page(page
) == 0) {
2119 BUG_ON(blocksize
!= PAGE_SIZE
);
2126 lock_extent(tree
, start
, end
, GFP_NOFS
);
2127 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
2130 unlock_extent(tree
, start
, end
, GFP_NOFS
);
2131 btrfs_start_ordered_extent(inode
, ordered
, 1);
2132 btrfs_put_ordered_extent(ordered
);
2135 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2137 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2140 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2141 userpage
= kmap_atomic(page
, KM_USER0
);
2142 memset(userpage
+ zero_offset
, 0, iosize
);
2143 flush_dcache_page(page
);
2144 kunmap_atomic(userpage
, KM_USER0
);
2147 while (cur
<= end
) {
2148 if (cur
>= last_byte
) {
2150 struct extent_state
*cached
= NULL
;
2152 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2153 userpage
= kmap_atomic(page
, KM_USER0
);
2154 memset(userpage
+ pg_offset
, 0, iosize
);
2155 flush_dcache_page(page
);
2156 kunmap_atomic(userpage
, KM_USER0
);
2157 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2159 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2163 em
= get_extent(inode
, page
, pg_offset
, cur
,
2165 if (IS_ERR_OR_NULL(em
)) {
2167 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
2170 extent_offset
= cur
- em
->start
;
2171 BUG_ON(extent_map_end(em
) <= cur
);
2174 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2175 this_bio_flag
= EXTENT_BIO_COMPRESSED
;
2176 extent_set_compress_type(&this_bio_flag
,
2180 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2181 cur_end
= min(extent_map_end(em
) - 1, end
);
2182 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2183 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2184 disk_io_size
= em
->block_len
;
2185 sector
= em
->block_start
>> 9;
2187 sector
= (em
->block_start
+ extent_offset
) >> 9;
2188 disk_io_size
= iosize
;
2191 block_start
= em
->block_start
;
2192 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2193 block_start
= EXTENT_MAP_HOLE
;
2194 free_extent_map(em
);
2197 /* we've found a hole, just zero and go on */
2198 if (block_start
== EXTENT_MAP_HOLE
) {
2200 struct extent_state
*cached
= NULL
;
2202 userpage
= kmap_atomic(page
, KM_USER0
);
2203 memset(userpage
+ pg_offset
, 0, iosize
);
2204 flush_dcache_page(page
);
2205 kunmap_atomic(userpage
, KM_USER0
);
2207 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2209 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2212 pg_offset
+= iosize
;
2215 /* the get_extent function already copied into the page */
2216 if (test_range_bit(tree
, cur
, cur_end
,
2217 EXTENT_UPTODATE
, 1, NULL
)) {
2218 check_page_uptodate(tree
, page
);
2219 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2221 pg_offset
+= iosize
;
2224 /* we have an inline extent but it didn't get marked up
2225 * to date. Error out
2227 if (block_start
== EXTENT_MAP_INLINE
) {
2229 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2231 pg_offset
+= iosize
;
2236 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
2237 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
2241 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2243 ret
= submit_extent_page(READ
, tree
, page
,
2244 sector
, disk_io_size
, pg_offset
,
2246 end_bio_extent_readpage
, mirror_num
,
2250 *bio_flags
= this_bio_flag
;
2255 pg_offset
+= iosize
;
2259 if (!PageError(page
))
2260 SetPageUptodate(page
);
2266 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2267 get_extent_t
*get_extent
)
2269 struct bio
*bio
= NULL
;
2270 unsigned long bio_flags
= 0;
2273 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, 0,
2276 ret
= submit_one_bio(READ
, bio
, 0, bio_flags
);
2280 static noinline
void update_nr_written(struct page
*page
,
2281 struct writeback_control
*wbc
,
2282 unsigned long nr_written
)
2284 wbc
->nr_to_write
-= nr_written
;
2285 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
2286 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
2287 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
2291 * the writepage semantics are similar to regular writepage. extent
2292 * records are inserted to lock ranges in the tree, and as dirty areas
2293 * are found, they are marked writeback. Then the lock bits are removed
2294 * and the end_io handler clears the writeback ranges
2296 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
2299 struct inode
*inode
= page
->mapping
->host
;
2300 struct extent_page_data
*epd
= data
;
2301 struct extent_io_tree
*tree
= epd
->tree
;
2302 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2304 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2308 u64 last_byte
= i_size_read(inode
);
2312 struct extent_state
*cached_state
= NULL
;
2313 struct extent_map
*em
;
2314 struct block_device
*bdev
;
2317 size_t pg_offset
= 0;
2319 loff_t i_size
= i_size_read(inode
);
2320 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2326 unsigned long nr_written
= 0;
2327 bool fill_delalloc
= true;
2329 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2330 write_flags
= WRITE_SYNC
;
2332 write_flags
= WRITE
;
2334 trace___extent_writepage(page
, inode
, wbc
);
2336 WARN_ON(!PageLocked(page
));
2337 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
2338 if (page
->index
> end_index
||
2339 (page
->index
== end_index
&& !pg_offset
)) {
2340 page
->mapping
->a_ops
->invalidatepage(page
, 0);
2345 if (page
->index
== end_index
) {
2348 userpage
= kmap_atomic(page
, KM_USER0
);
2349 memset(userpage
+ pg_offset
, 0,
2350 PAGE_CACHE_SIZE
- pg_offset
);
2351 kunmap_atomic(userpage
, KM_USER0
);
2352 flush_dcache_page(page
);
2356 set_page_extent_mapped(page
);
2358 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
2359 fill_delalloc
= false;
2361 delalloc_start
= start
;
2364 if (!epd
->extent_locked
&& fill_delalloc
) {
2365 u64 delalloc_to_write
= 0;
2367 * make sure the wbc mapping index is at least updated
2370 update_nr_written(page
, wbc
, 0);
2372 while (delalloc_end
< page_end
) {
2373 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
2378 if (nr_delalloc
== 0) {
2379 delalloc_start
= delalloc_end
+ 1;
2382 tree
->ops
->fill_delalloc(inode
, page
, delalloc_start
,
2383 delalloc_end
, &page_started
,
2386 * delalloc_end is already one less than the total
2387 * length, so we don't subtract one from
2390 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
2393 delalloc_start
= delalloc_end
+ 1;
2395 if (wbc
->nr_to_write
< delalloc_to_write
) {
2398 if (delalloc_to_write
< thresh
* 2)
2399 thresh
= delalloc_to_write
;
2400 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
2404 /* did the fill delalloc function already unlock and start
2410 * we've unlocked the page, so we can't update
2411 * the mapping's writeback index, just update
2414 wbc
->nr_to_write
-= nr_written
;
2418 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
2419 ret
= tree
->ops
->writepage_start_hook(page
, start
,
2421 if (ret
== -EAGAIN
) {
2422 redirty_page_for_writepage(wbc
, page
);
2423 update_nr_written(page
, wbc
, nr_written
);
2431 * we don't want to touch the inode after unlocking the page,
2432 * so we update the mapping writeback index now
2434 update_nr_written(page
, wbc
, nr_written
+ 1);
2437 if (last_byte
<= start
) {
2438 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2439 tree
->ops
->writepage_end_io_hook(page
, start
,
2444 blocksize
= inode
->i_sb
->s_blocksize
;
2446 while (cur
<= end
) {
2447 if (cur
>= last_byte
) {
2448 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2449 tree
->ops
->writepage_end_io_hook(page
, cur
,
2453 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
2455 if (IS_ERR_OR_NULL(em
)) {
2460 extent_offset
= cur
- em
->start
;
2461 BUG_ON(extent_map_end(em
) <= cur
);
2463 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2464 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2465 sector
= (em
->block_start
+ extent_offset
) >> 9;
2467 block_start
= em
->block_start
;
2468 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
2469 free_extent_map(em
);
2473 * compressed and inline extents are written through other
2476 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
2477 block_start
== EXTENT_MAP_INLINE
) {
2479 * end_io notification does not happen here for
2480 * compressed extents
2482 if (!compressed
&& tree
->ops
&&
2483 tree
->ops
->writepage_end_io_hook
)
2484 tree
->ops
->writepage_end_io_hook(page
, cur
,
2487 else if (compressed
) {
2488 /* we don't want to end_page_writeback on
2489 * a compressed extent. this happens
2496 pg_offset
+= iosize
;
2499 /* leave this out until we have a page_mkwrite call */
2500 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2501 EXTENT_DIRTY
, 0, NULL
)) {
2503 pg_offset
+= iosize
;
2507 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2508 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2516 unsigned long max_nr
= end_index
+ 1;
2518 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2519 if (!PageWriteback(page
)) {
2520 printk(KERN_ERR
"btrfs warning page %lu not "
2521 "writeback, cur %llu end %llu\n",
2522 page
->index
, (unsigned long long)cur
,
2523 (unsigned long long)end
);
2526 ret
= submit_extent_page(write_flags
, tree
, page
,
2527 sector
, iosize
, pg_offset
,
2528 bdev
, &epd
->bio
, max_nr
,
2529 end_bio_extent_writepage
,
2535 pg_offset
+= iosize
;
2540 /* make sure the mapping tag for page dirty gets cleared */
2541 set_page_writeback(page
);
2542 end_page_writeback(page
);
2548 /* drop our reference on any cached states */
2549 free_extent_state(cached_state
);
2554 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2555 * @mapping: address space structure to write
2556 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2557 * @writepage: function called for each page
2558 * @data: data passed to writepage function
2560 * If a page is already under I/O, write_cache_pages() skips it, even
2561 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2562 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2563 * and msync() need to guarantee that all the data which was dirty at the time
2564 * the call was made get new I/O started against them. If wbc->sync_mode is
2565 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2566 * existing IO to complete.
2568 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
2569 struct address_space
*mapping
,
2570 struct writeback_control
*wbc
,
2571 writepage_t writepage
, void *data
,
2572 void (*flush_fn
)(void *))
2576 int nr_to_write_done
= 0;
2577 struct pagevec pvec
;
2580 pgoff_t end
; /* Inclusive */
2584 pagevec_init(&pvec
, 0);
2585 if (wbc
->range_cyclic
) {
2586 index
= mapping
->writeback_index
; /* Start from prev offset */
2589 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2590 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2593 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2594 tag
= PAGECACHE_TAG_TOWRITE
;
2596 tag
= PAGECACHE_TAG_DIRTY
;
2598 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2599 tag_pages_for_writeback(mapping
, index
, end
);
2600 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
2601 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2602 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2606 for (i
= 0; i
< nr_pages
; i
++) {
2607 struct page
*page
= pvec
.pages
[i
];
2610 * At this point we hold neither mapping->tree_lock nor
2611 * lock on the page itself: the page may be truncated or
2612 * invalidated (changing page->mapping to NULL), or even
2613 * swizzled back from swapper_space to tmpfs file
2617 tree
->ops
->write_cache_pages_lock_hook
) {
2618 tree
->ops
->write_cache_pages_lock_hook(page
,
2621 if (!trylock_page(page
)) {
2627 if (unlikely(page
->mapping
!= mapping
)) {
2632 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2638 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
2639 if (PageWriteback(page
))
2641 wait_on_page_writeback(page
);
2644 if (PageWriteback(page
) ||
2645 !clear_page_dirty_for_io(page
)) {
2650 ret
= (*writepage
)(page
, wbc
, data
);
2652 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2660 * the filesystem may choose to bump up nr_to_write.
2661 * We have to make sure to honor the new nr_to_write
2664 nr_to_write_done
= wbc
->nr_to_write
<= 0;
2666 pagevec_release(&pvec
);
2669 if (!scanned
&& !done
) {
2671 * We hit the last page and there is more work to be done: wrap
2672 * back to the start of the file
2681 static void flush_epd_write_bio(struct extent_page_data
*epd
)
2685 submit_one_bio(WRITE_SYNC
, epd
->bio
, 0, 0);
2687 submit_one_bio(WRITE
, epd
->bio
, 0, 0);
2692 static noinline
void flush_write_bio(void *data
)
2694 struct extent_page_data
*epd
= data
;
2695 flush_epd_write_bio(epd
);
2698 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2699 get_extent_t
*get_extent
,
2700 struct writeback_control
*wbc
)
2703 struct extent_page_data epd
= {
2706 .get_extent
= get_extent
,
2708 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
2711 ret
= __extent_writepage(page
, wbc
, &epd
);
2713 flush_epd_write_bio(&epd
);
2717 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
2718 u64 start
, u64 end
, get_extent_t
*get_extent
,
2722 struct address_space
*mapping
= inode
->i_mapping
;
2724 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
2727 struct extent_page_data epd
= {
2730 .get_extent
= get_extent
,
2732 .sync_io
= mode
== WB_SYNC_ALL
,
2734 struct writeback_control wbc_writepages
= {
2736 .nr_to_write
= nr_pages
* 2,
2737 .range_start
= start
,
2738 .range_end
= end
+ 1,
2741 while (start
<= end
) {
2742 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
2743 if (clear_page_dirty_for_io(page
))
2744 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
2746 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2747 tree
->ops
->writepage_end_io_hook(page
, start
,
2748 start
+ PAGE_CACHE_SIZE
- 1,
2752 page_cache_release(page
);
2753 start
+= PAGE_CACHE_SIZE
;
2756 flush_epd_write_bio(&epd
);
2760 int extent_writepages(struct extent_io_tree
*tree
,
2761 struct address_space
*mapping
,
2762 get_extent_t
*get_extent
,
2763 struct writeback_control
*wbc
)
2766 struct extent_page_data epd
= {
2769 .get_extent
= get_extent
,
2771 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
2774 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
2775 __extent_writepage
, &epd
,
2777 flush_epd_write_bio(&epd
);
2781 int extent_readpages(struct extent_io_tree
*tree
,
2782 struct address_space
*mapping
,
2783 struct list_head
*pages
, unsigned nr_pages
,
2784 get_extent_t get_extent
)
2786 struct bio
*bio
= NULL
;
2788 unsigned long bio_flags
= 0;
2790 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2791 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2793 prefetchw(&page
->flags
);
2794 list_del(&page
->lru
);
2795 if (!add_to_page_cache_lru(page
, mapping
,
2796 page
->index
, GFP_NOFS
)) {
2797 __extent_read_full_page(tree
, page
, get_extent
,
2798 &bio
, 0, &bio_flags
);
2800 page_cache_release(page
);
2802 BUG_ON(!list_empty(pages
));
2804 submit_one_bio(READ
, bio
, 0, bio_flags
);
2809 * basic invalidatepage code, this waits on any locked or writeback
2810 * ranges corresponding to the page, and then deletes any extent state
2811 * records from the tree
2813 int extent_invalidatepage(struct extent_io_tree
*tree
,
2814 struct page
*page
, unsigned long offset
)
2816 struct extent_state
*cached_state
= NULL
;
2817 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2818 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2819 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2821 start
+= (offset
+ blocksize
- 1) & ~(blocksize
- 1);
2825 lock_extent_bits(tree
, start
, end
, 0, &cached_state
, GFP_NOFS
);
2826 wait_on_page_writeback(page
);
2827 clear_extent_bit(tree
, start
, end
,
2828 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
2829 EXTENT_DO_ACCOUNTING
,
2830 1, 1, &cached_state
, GFP_NOFS
);
2835 * a helper for releasepage, this tests for areas of the page that
2836 * are locked or under IO and drops the related state bits if it is safe
2839 int try_release_extent_state(struct extent_map_tree
*map
,
2840 struct extent_io_tree
*tree
, struct page
*page
,
2843 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2844 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2847 if (test_range_bit(tree
, start
, end
,
2848 EXTENT_IOBITS
, 0, NULL
))
2851 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2854 * at this point we can safely clear everything except the
2855 * locked bit and the nodatasum bit
2857 ret
= clear_extent_bit(tree
, start
, end
,
2858 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
2861 /* if clear_extent_bit failed for enomem reasons,
2862 * we can't allow the release to continue.
2873 * a helper for releasepage. As long as there are no locked extents
2874 * in the range corresponding to the page, both state records and extent
2875 * map records are removed
2877 int try_release_extent_mapping(struct extent_map_tree
*map
,
2878 struct extent_io_tree
*tree
, struct page
*page
,
2881 struct extent_map
*em
;
2882 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2883 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2885 if ((mask
& __GFP_WAIT
) &&
2886 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2888 while (start
<= end
) {
2889 len
= end
- start
+ 1;
2890 write_lock(&map
->lock
);
2891 em
= lookup_extent_mapping(map
, start
, len
);
2892 if (IS_ERR_OR_NULL(em
)) {
2893 write_unlock(&map
->lock
);
2896 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
2897 em
->start
!= start
) {
2898 write_unlock(&map
->lock
);
2899 free_extent_map(em
);
2902 if (!test_range_bit(tree
, em
->start
,
2903 extent_map_end(em
) - 1,
2904 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
2906 remove_extent_mapping(map
, em
);
2907 /* once for the rb tree */
2908 free_extent_map(em
);
2910 start
= extent_map_end(em
);
2911 write_unlock(&map
->lock
);
2914 free_extent_map(em
);
2917 return try_release_extent_state(map
, tree
, page
, mask
);
2921 * helper function for fiemap, which doesn't want to see any holes.
2922 * This maps until we find something past 'last'
2924 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
2927 get_extent_t
*get_extent
)
2929 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
2930 struct extent_map
*em
;
2937 len
= last
- offset
;
2940 len
= (len
+ sectorsize
- 1) & ~(sectorsize
- 1);
2941 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
2942 if (IS_ERR_OR_NULL(em
))
2945 /* if this isn't a hole return it */
2946 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
2947 em
->block_start
!= EXTENT_MAP_HOLE
) {
2951 /* this is a hole, advance to the next extent */
2952 offset
= extent_map_end(em
);
2953 free_extent_map(em
);
2960 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
2961 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
2965 u64 max
= start
+ len
;
2969 u64 last_for_get_extent
= 0;
2971 u64 isize
= i_size_read(inode
);
2972 struct btrfs_key found_key
;
2973 struct extent_map
*em
= NULL
;
2974 struct extent_state
*cached_state
= NULL
;
2975 struct btrfs_path
*path
;
2976 struct btrfs_file_extent_item
*item
;
2981 unsigned long emflags
;
2986 path
= btrfs_alloc_path();
2989 path
->leave_spinning
= 1;
2992 * lookup the last file extent. We're not using i_size here
2993 * because there might be preallocation past i_size
2995 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
2996 path
, btrfs_ino(inode
), -1, 0);
2998 btrfs_free_path(path
);
3003 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3004 struct btrfs_file_extent_item
);
3005 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
3006 found_type
= btrfs_key_type(&found_key
);
3008 /* No extents, but there might be delalloc bits */
3009 if (found_key
.objectid
!= btrfs_ino(inode
) ||
3010 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3011 /* have to trust i_size as the end */
3013 last_for_get_extent
= isize
;
3016 * remember the start of the last extent. There are a
3017 * bunch of different factors that go into the length of the
3018 * extent, so its much less complex to remember where it started
3020 last
= found_key
.offset
;
3021 last_for_get_extent
= last
+ 1;
3023 btrfs_free_path(path
);
3026 * we might have some extents allocated but more delalloc past those
3027 * extents. so, we trust isize unless the start of the last extent is
3032 last_for_get_extent
= isize
;
3035 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
, 0,
3036 &cached_state
, GFP_NOFS
);
3038 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
3048 u64 offset_in_extent
;
3050 /* break if the extent we found is outside the range */
3051 if (em
->start
>= max
|| extent_map_end(em
) < off
)
3055 * get_extent may return an extent that starts before our
3056 * requested range. We have to make sure the ranges
3057 * we return to fiemap always move forward and don't
3058 * overlap, so adjust the offsets here
3060 em_start
= max(em
->start
, off
);
3063 * record the offset from the start of the extent
3064 * for adjusting the disk offset below
3066 offset_in_extent
= em_start
- em
->start
;
3067 em_end
= extent_map_end(em
);
3068 em_len
= em_end
- em_start
;
3069 emflags
= em
->flags
;
3074 * bump off for our next call to get_extent
3076 off
= extent_map_end(em
);
3080 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
3082 flags
|= FIEMAP_EXTENT_LAST
;
3083 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
3084 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
3085 FIEMAP_EXTENT_NOT_ALIGNED
);
3086 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
3087 flags
|= (FIEMAP_EXTENT_DELALLOC
|
3088 FIEMAP_EXTENT_UNKNOWN
);
3090 disko
= em
->block_start
+ offset_in_extent
;
3092 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
3093 flags
|= FIEMAP_EXTENT_ENCODED
;
3095 free_extent_map(em
);
3097 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
3098 (last
== (u64
)-1 && isize
<= em_end
)) {
3099 flags
|= FIEMAP_EXTENT_LAST
;
3103 /* now scan forward to see if this is really the last extent. */
3104 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
3111 flags
|= FIEMAP_EXTENT_LAST
;
3114 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
3120 free_extent_map(em
);
3122 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
,
3123 &cached_state
, GFP_NOFS
);
3127 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
3131 struct address_space
*mapping
;
3134 return eb
->first_page
;
3135 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
3136 mapping
= eb
->first_page
->mapping
;
3141 * extent_buffer_page is only called after pinning the page
3142 * by increasing the reference count. So we know the page must
3143 * be in the radix tree.
3146 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
3152 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
3154 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
3155 (start
>> PAGE_CACHE_SHIFT
);
3158 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
3163 struct extent_buffer
*eb
= NULL
;
3165 unsigned long flags
;
3168 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
3173 rwlock_init(&eb
->lock
);
3174 atomic_set(&eb
->write_locks
, 0);
3175 atomic_set(&eb
->read_locks
, 0);
3176 atomic_set(&eb
->blocking_readers
, 0);
3177 atomic_set(&eb
->blocking_writers
, 0);
3178 atomic_set(&eb
->spinning_readers
, 0);
3179 atomic_set(&eb
->spinning_writers
, 0);
3180 init_waitqueue_head(&eb
->write_lock_wq
);
3181 init_waitqueue_head(&eb
->read_lock_wq
);
3184 spin_lock_irqsave(&leak_lock
, flags
);
3185 list_add(&eb
->leak_list
, &buffers
);
3186 spin_unlock_irqrestore(&leak_lock
, flags
);
3188 atomic_set(&eb
->refs
, 1);
3193 static void __free_extent_buffer(struct extent_buffer
*eb
)
3196 unsigned long flags
;
3197 spin_lock_irqsave(&leak_lock
, flags
);
3198 list_del(&eb
->leak_list
);
3199 spin_unlock_irqrestore(&leak_lock
, flags
);
3201 kmem_cache_free(extent_buffer_cache
, eb
);
3205 * Helper for releasing extent buffer page.
3207 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
3208 unsigned long start_idx
)
3210 unsigned long index
;
3213 if (!eb
->first_page
)
3216 index
= num_extent_pages(eb
->start
, eb
->len
);
3217 if (start_idx
>= index
)
3222 page
= extent_buffer_page(eb
, index
);
3224 page_cache_release(page
);
3225 } while (index
!= start_idx
);
3229 * Helper for releasing the extent buffer.
3231 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
3233 btrfs_release_extent_buffer_page(eb
, 0);
3234 __free_extent_buffer(eb
);
3237 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
3238 u64 start
, unsigned long len
,
3241 unsigned long num_pages
= num_extent_pages(start
, len
);
3243 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
3244 struct extent_buffer
*eb
;
3245 struct extent_buffer
*exists
= NULL
;
3247 struct address_space
*mapping
= tree
->mapping
;
3252 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
3253 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
3255 mark_page_accessed(eb
->first_page
);
3260 eb
= __alloc_extent_buffer(tree
, start
, len
, GFP_NOFS
);
3265 eb
->first_page
= page0
;
3268 page_cache_get(page0
);
3269 mark_page_accessed(page0
);
3270 set_page_extent_mapped(page0
);
3271 set_page_extent_head(page0
, len
);
3272 uptodate
= PageUptodate(page0
);
3276 for (; i
< num_pages
; i
++, index
++) {
3277 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
3282 set_page_extent_mapped(p
);
3283 mark_page_accessed(p
);
3286 set_page_extent_head(p
, len
);
3288 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
3290 if (!PageUptodate(p
))
3294 * see below about how we avoid a nasty race with release page
3295 * and why we unlock later
3301 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3303 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
3307 spin_lock(&tree
->buffer_lock
);
3308 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
3309 if (ret
== -EEXIST
) {
3310 exists
= radix_tree_lookup(&tree
->buffer
,
3311 start
>> PAGE_CACHE_SHIFT
);
3312 /* add one reference for the caller */
3313 atomic_inc(&exists
->refs
);
3314 spin_unlock(&tree
->buffer_lock
);
3315 radix_tree_preload_end();
3318 /* add one reference for the tree */
3319 atomic_inc(&eb
->refs
);
3320 spin_unlock(&tree
->buffer_lock
);
3321 radix_tree_preload_end();
3324 * there is a race where release page may have
3325 * tried to find this extent buffer in the radix
3326 * but failed. It will tell the VM it is safe to
3327 * reclaim the, and it will clear the page private bit.
3328 * We must make sure to set the page private bit properly
3329 * after the extent buffer is in the radix tree so
3330 * it doesn't get lost
3332 set_page_extent_mapped(eb
->first_page
);
3333 set_page_extent_head(eb
->first_page
, eb
->len
);
3335 unlock_page(eb
->first_page
);
3339 if (eb
->first_page
&& !page0
)
3340 unlock_page(eb
->first_page
);
3342 if (!atomic_dec_and_test(&eb
->refs
))
3344 btrfs_release_extent_buffer(eb
);
3348 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
3349 u64 start
, unsigned long len
)
3351 struct extent_buffer
*eb
;
3354 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
3355 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
3357 mark_page_accessed(eb
->first_page
);
3365 void free_extent_buffer(struct extent_buffer
*eb
)
3370 if (!atomic_dec_and_test(&eb
->refs
))
3376 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
3377 struct extent_buffer
*eb
)
3380 unsigned long num_pages
;
3383 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3385 for (i
= 0; i
< num_pages
; i
++) {
3386 page
= extent_buffer_page(eb
, i
);
3387 if (!PageDirty(page
))
3391 WARN_ON(!PagePrivate(page
));
3393 set_page_extent_mapped(page
);
3395 set_page_extent_head(page
, eb
->len
);
3397 clear_page_dirty_for_io(page
);
3398 spin_lock_irq(&page
->mapping
->tree_lock
);
3399 if (!PageDirty(page
)) {
3400 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3402 PAGECACHE_TAG_DIRTY
);
3404 spin_unlock_irq(&page
->mapping
->tree_lock
);
3410 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
3411 struct extent_buffer
*eb
)
3414 unsigned long num_pages
;
3417 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
3418 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3419 for (i
= 0; i
< num_pages
; i
++)
3420 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
3424 static int __eb_straddles_pages(u64 start
, u64 len
)
3426 if (len
< PAGE_CACHE_SIZE
)
3428 if (start
& (PAGE_CACHE_SIZE
- 1))
3430 if ((start
+ len
) & (PAGE_CACHE_SIZE
- 1))
3435 static int eb_straddles_pages(struct extent_buffer
*eb
)
3437 return __eb_straddles_pages(eb
->start
, eb
->len
);
3440 int clear_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3441 struct extent_buffer
*eb
,
3442 struct extent_state
**cached_state
)
3446 unsigned long num_pages
;
3448 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3449 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3451 if (eb_straddles_pages(eb
)) {
3452 clear_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3453 cached_state
, GFP_NOFS
);
3455 for (i
= 0; i
< num_pages
; i
++) {
3456 page
= extent_buffer_page(eb
, i
);
3458 ClearPageUptodate(page
);
3463 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3464 struct extent_buffer
*eb
)
3468 unsigned long num_pages
;
3470 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3472 if (eb_straddles_pages(eb
)) {
3473 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3476 for (i
= 0; i
< num_pages
; i
++) {
3477 page
= extent_buffer_page(eb
, i
);
3478 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
3479 ((i
== num_pages
- 1) &&
3480 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
3481 check_page_uptodate(tree
, page
);
3484 SetPageUptodate(page
);
3489 int extent_range_uptodate(struct extent_io_tree
*tree
,
3494 int pg_uptodate
= 1;
3496 unsigned long index
;
3498 if (__eb_straddles_pages(start
, end
- start
+ 1)) {
3499 ret
= test_range_bit(tree
, start
, end
,
3500 EXTENT_UPTODATE
, 1, NULL
);
3504 while (start
<= end
) {
3505 index
= start
>> PAGE_CACHE_SHIFT
;
3506 page
= find_get_page(tree
->mapping
, index
);
3507 uptodate
= PageUptodate(page
);
3508 page_cache_release(page
);
3513 start
+= PAGE_CACHE_SIZE
;
3518 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
3519 struct extent_buffer
*eb
,
3520 struct extent_state
*cached_state
)
3523 unsigned long num_pages
;
3526 int pg_uptodate
= 1;
3528 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
3531 if (eb_straddles_pages(eb
)) {
3532 ret
= test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3533 EXTENT_UPTODATE
, 1, cached_state
);
3538 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3539 for (i
= 0; i
< num_pages
; i
++) {
3540 page
= extent_buffer_page(eb
, i
);
3541 if (!PageUptodate(page
)) {
3549 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
3550 struct extent_buffer
*eb
,
3551 u64 start
, int wait
,
3552 get_extent_t
*get_extent
, int mirror_num
)
3555 unsigned long start_i
;
3559 int locked_pages
= 0;
3560 int all_uptodate
= 1;
3561 int inc_all_pages
= 0;
3562 unsigned long num_pages
;
3563 struct bio
*bio
= NULL
;
3564 unsigned long bio_flags
= 0;
3566 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
3569 if (eb_straddles_pages(eb
)) {
3570 if (test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3571 EXTENT_UPTODATE
, 1, NULL
)) {
3577 WARN_ON(start
< eb
->start
);
3578 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
3579 (eb
->start
>> PAGE_CACHE_SHIFT
);
3584 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3585 for (i
= start_i
; i
< num_pages
; i
++) {
3586 page
= extent_buffer_page(eb
, i
);
3588 if (!trylock_page(page
))
3594 if (!PageUptodate(page
))
3599 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3603 for (i
= start_i
; i
< num_pages
; i
++) {
3604 page
= extent_buffer_page(eb
, i
);
3606 WARN_ON(!PagePrivate(page
));
3608 set_page_extent_mapped(page
);
3610 set_page_extent_head(page
, eb
->len
);
3613 page_cache_get(page
);
3614 if (!PageUptodate(page
)) {
3617 ClearPageError(page
);
3618 err
= __extent_read_full_page(tree
, page
,
3620 mirror_num
, &bio_flags
);
3629 submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3634 for (i
= start_i
; i
< num_pages
; i
++) {
3635 page
= extent_buffer_page(eb
, i
);
3636 wait_on_page_locked(page
);
3637 if (!PageUptodate(page
))
3642 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3647 while (locked_pages
> 0) {
3648 page
= extent_buffer_page(eb
, i
);
3656 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
3657 unsigned long start
,
3664 char *dst
= (char *)dstv
;
3665 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3666 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3668 WARN_ON(start
> eb
->len
);
3669 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3671 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3674 page
= extent_buffer_page(eb
, i
);
3676 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3677 kaddr
= page_address(page
);
3678 memcpy(dst
, kaddr
+ offset
, cur
);
3687 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3688 unsigned long min_len
, char **map
,
3689 unsigned long *map_start
,
3690 unsigned long *map_len
)
3692 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
3695 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3696 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3697 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
3704 offset
= start_offset
;
3708 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
3711 if (start
+ min_len
> eb
->len
) {
3712 printk(KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
3713 "wanted %lu %lu\n", (unsigned long long)eb
->start
,
3714 eb
->len
, start
, min_len
);
3719 p
= extent_buffer_page(eb
, i
);
3720 kaddr
= page_address(p
);
3721 *map
= kaddr
+ offset
;
3722 *map_len
= PAGE_CACHE_SIZE
- offset
;
3726 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3727 unsigned long start
,
3734 char *ptr
= (char *)ptrv
;
3735 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3736 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3739 WARN_ON(start
> eb
->len
);
3740 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3742 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3745 page
= extent_buffer_page(eb
, i
);
3747 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3749 kaddr
= page_address(page
);
3750 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3762 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3763 unsigned long start
, unsigned long len
)
3769 char *src
= (char *)srcv
;
3770 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3771 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3773 WARN_ON(start
> eb
->len
);
3774 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3776 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3779 page
= extent_buffer_page(eb
, i
);
3780 WARN_ON(!PageUptodate(page
));
3782 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3783 kaddr
= page_address(page
);
3784 memcpy(kaddr
+ offset
, src
, cur
);
3793 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3794 unsigned long start
, unsigned long len
)
3800 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3801 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3803 WARN_ON(start
> eb
->len
);
3804 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3806 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3809 page
= extent_buffer_page(eb
, i
);
3810 WARN_ON(!PageUptodate(page
));
3812 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3813 kaddr
= page_address(page
);
3814 memset(kaddr
+ offset
, c
, cur
);
3822 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3823 unsigned long dst_offset
, unsigned long src_offset
,
3826 u64 dst_len
= dst
->len
;
3831 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3832 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3834 WARN_ON(src
->len
!= dst_len
);
3836 offset
= (start_offset
+ dst_offset
) &
3837 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3840 page
= extent_buffer_page(dst
, i
);
3841 WARN_ON(!PageUptodate(page
));
3843 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3845 kaddr
= page_address(page
);
3846 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3855 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3856 unsigned long dst_off
, unsigned long src_off
,
3859 char *dst_kaddr
= page_address(dst_page
);
3860 if (dst_page
== src_page
) {
3861 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3863 char *src_kaddr
= page_address(src_page
);
3864 char *p
= dst_kaddr
+ dst_off
+ len
;
3865 char *s
= src_kaddr
+ src_off
+ len
;
3872 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
3874 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
3875 return distance
< len
;
3878 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3879 unsigned long dst_off
, unsigned long src_off
,
3882 char *dst_kaddr
= page_address(dst_page
);
3885 if (dst_page
!= src_page
) {
3886 src_kaddr
= page_address(src_page
);
3888 src_kaddr
= dst_kaddr
;
3889 BUG_ON(areas_overlap(src_off
, dst_off
, len
));
3892 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3895 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3896 unsigned long src_offset
, unsigned long len
)
3899 size_t dst_off_in_page
;
3900 size_t src_off_in_page
;
3901 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3902 unsigned long dst_i
;
3903 unsigned long src_i
;
3905 if (src_offset
+ len
> dst
->len
) {
3906 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
3907 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
3910 if (dst_offset
+ len
> dst
->len
) {
3911 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
3912 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
3917 dst_off_in_page
= (start_offset
+ dst_offset
) &
3918 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3919 src_off_in_page
= (start_offset
+ src_offset
) &
3920 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3922 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3923 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3925 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3927 cur
= min_t(unsigned long, cur
,
3928 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3930 copy_pages(extent_buffer_page(dst
, dst_i
),
3931 extent_buffer_page(dst
, src_i
),
3932 dst_off_in_page
, src_off_in_page
, cur
);
3940 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3941 unsigned long src_offset
, unsigned long len
)
3944 size_t dst_off_in_page
;
3945 size_t src_off_in_page
;
3946 unsigned long dst_end
= dst_offset
+ len
- 1;
3947 unsigned long src_end
= src_offset
+ len
- 1;
3948 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3949 unsigned long dst_i
;
3950 unsigned long src_i
;
3952 if (src_offset
+ len
> dst
->len
) {
3953 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
3954 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
3957 if (dst_offset
+ len
> dst
->len
) {
3958 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
3959 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
3962 if (!areas_overlap(src_offset
, dst_offset
, len
)) {
3963 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
3967 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
3968 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
3970 dst_off_in_page
= (start_offset
+ dst_end
) &
3971 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3972 src_off_in_page
= (start_offset
+ src_end
) &
3973 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3975 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
3976 cur
= min(cur
, dst_off_in_page
+ 1);
3977 move_pages(extent_buffer_page(dst
, dst_i
),
3978 extent_buffer_page(dst
, src_i
),
3979 dst_off_in_page
- cur
+ 1,
3980 src_off_in_page
- cur
+ 1, cur
);
3988 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
3990 struct extent_buffer
*eb
=
3991 container_of(head
, struct extent_buffer
, rcu_head
);
3993 btrfs_release_extent_buffer(eb
);
3996 int try_release_extent_buffer(struct extent_io_tree
*tree
, struct page
*page
)
3998 u64 start
= page_offset(page
);
3999 struct extent_buffer
*eb
;
4002 spin_lock(&tree
->buffer_lock
);
4003 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4005 spin_unlock(&tree
->buffer_lock
);
4009 if (test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
4015 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4018 if (atomic_cmpxchg(&eb
->refs
, 1, 0) != 1) {
4023 radix_tree_delete(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4025 spin_unlock(&tree
->buffer_lock
);
4027 /* at this point we can safely release the extent buffer */
4028 if (atomic_read(&eb
->refs
) == 0)
4029 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);