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 "extent_io.h"
14 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 static struct kmem_cache
*extent_state_cache
;
20 static struct kmem_cache
*extent_buffer_cache
;
22 static LIST_HEAD(buffers
);
23 static LIST_HEAD(states
);
27 static DEFINE_SPINLOCK(leak_lock
);
30 #define BUFFER_LRU_MAX 64
35 struct rb_node rb_node
;
38 struct extent_page_data
{
40 struct extent_io_tree
*tree
;
41 get_extent_t
*get_extent
;
43 /* tells writepage not to lock the state bits for this range
44 * it still does the unlocking
46 unsigned int extent_locked
:1;
48 /* tells the submit_bio code to use a WRITE_SYNC */
49 unsigned int sync_io
:1;
52 int __init
extent_io_init(void)
54 extent_state_cache
= kmem_cache_create("extent_state",
55 sizeof(struct extent_state
), 0,
56 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
57 if (!extent_state_cache
)
60 extent_buffer_cache
= kmem_cache_create("extent_buffers",
61 sizeof(struct extent_buffer
), 0,
62 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
63 if (!extent_buffer_cache
)
64 goto free_state_cache
;
68 kmem_cache_destroy(extent_state_cache
);
72 void extent_io_exit(void)
74 struct extent_state
*state
;
75 struct extent_buffer
*eb
;
77 while (!list_empty(&states
)) {
78 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
79 printk(KERN_ERR
"btrfs state leak: start %llu end %llu "
80 "state %lu in tree %p refs %d\n",
81 (unsigned long long)state
->start
,
82 (unsigned long long)state
->end
,
83 state
->state
, state
->tree
, atomic_read(&state
->refs
));
84 list_del(&state
->leak_list
);
85 kmem_cache_free(extent_state_cache
, state
);
89 while (!list_empty(&buffers
)) {
90 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
91 printk(KERN_ERR
"btrfs buffer leak start %llu len %lu "
92 "refs %d\n", (unsigned long long)eb
->start
,
93 eb
->len
, atomic_read(&eb
->refs
));
94 list_del(&eb
->leak_list
);
95 kmem_cache_free(extent_buffer_cache
, eb
);
97 if (extent_state_cache
)
98 kmem_cache_destroy(extent_state_cache
);
99 if (extent_buffer_cache
)
100 kmem_cache_destroy(extent_buffer_cache
);
103 void extent_io_tree_init(struct extent_io_tree
*tree
,
104 struct address_space
*mapping
, gfp_t mask
)
106 tree
->state
= RB_ROOT
;
107 INIT_RADIX_TREE(&tree
->buffer
, GFP_ATOMIC
);
109 tree
->dirty_bytes
= 0;
110 spin_lock_init(&tree
->lock
);
111 spin_lock_init(&tree
->buffer_lock
);
112 tree
->mapping
= mapping
;
115 static struct extent_state
*alloc_extent_state(gfp_t mask
)
117 struct extent_state
*state
;
122 state
= kmem_cache_alloc(extent_state_cache
, mask
);
129 spin_lock_irqsave(&leak_lock
, flags
);
130 list_add(&state
->leak_list
, &states
);
131 spin_unlock_irqrestore(&leak_lock
, flags
);
133 atomic_set(&state
->refs
, 1);
134 init_waitqueue_head(&state
->wq
);
138 void free_extent_state(struct extent_state
*state
)
142 if (atomic_dec_and_test(&state
->refs
)) {
146 WARN_ON(state
->tree
);
148 spin_lock_irqsave(&leak_lock
, flags
);
149 list_del(&state
->leak_list
);
150 spin_unlock_irqrestore(&leak_lock
, flags
);
152 kmem_cache_free(extent_state_cache
, state
);
156 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
157 struct rb_node
*node
)
159 struct rb_node
**p
= &root
->rb_node
;
160 struct rb_node
*parent
= NULL
;
161 struct tree_entry
*entry
;
165 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
167 if (offset
< entry
->start
)
169 else if (offset
> entry
->end
)
175 entry
= rb_entry(node
, struct tree_entry
, rb_node
);
176 rb_link_node(node
, parent
, p
);
177 rb_insert_color(node
, root
);
181 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
182 struct rb_node
**prev_ret
,
183 struct rb_node
**next_ret
)
185 struct rb_root
*root
= &tree
->state
;
186 struct rb_node
*n
= root
->rb_node
;
187 struct rb_node
*prev
= NULL
;
188 struct rb_node
*orig_prev
= NULL
;
189 struct tree_entry
*entry
;
190 struct tree_entry
*prev_entry
= NULL
;
193 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
197 if (offset
< entry
->start
)
199 else if (offset
> entry
->end
)
207 while (prev
&& offset
> prev_entry
->end
) {
208 prev
= rb_next(prev
);
209 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
216 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
217 while (prev
&& offset
< prev_entry
->start
) {
218 prev
= rb_prev(prev
);
219 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
226 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
229 struct rb_node
*prev
= NULL
;
232 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
238 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
239 struct extent_state
*other
)
241 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
242 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
247 * utility function to look for merge candidates inside a given range.
248 * Any extents with matching state are merged together into a single
249 * extent in the tree. Extents with EXTENT_IO in their state field
250 * are not merged because the end_io handlers need to be able to do
251 * operations on them without sleeping (or doing allocations/splits).
253 * This should be called with the tree lock held.
255 static int merge_state(struct extent_io_tree
*tree
,
256 struct extent_state
*state
)
258 struct extent_state
*other
;
259 struct rb_node
*other_node
;
261 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
264 other_node
= rb_prev(&state
->rb_node
);
266 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
267 if (other
->end
== state
->start
- 1 &&
268 other
->state
== state
->state
) {
269 merge_cb(tree
, state
, other
);
270 state
->start
= other
->start
;
272 rb_erase(&other
->rb_node
, &tree
->state
);
273 free_extent_state(other
);
276 other_node
= rb_next(&state
->rb_node
);
278 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
279 if (other
->start
== state
->end
+ 1 &&
280 other
->state
== state
->state
) {
281 merge_cb(tree
, state
, other
);
282 other
->start
= state
->start
;
284 rb_erase(&state
->rb_node
, &tree
->state
);
285 free_extent_state(state
);
293 static int set_state_cb(struct extent_io_tree
*tree
,
294 struct extent_state
*state
, int *bits
)
296 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
297 return tree
->ops
->set_bit_hook(tree
->mapping
->host
,
304 static void clear_state_cb(struct extent_io_tree
*tree
,
305 struct extent_state
*state
, int *bits
)
307 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
308 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
312 * insert an extent_state struct into the tree. 'bits' are set on the
313 * struct before it is inserted.
315 * This may return -EEXIST if the extent is already there, in which case the
316 * state struct is freed.
318 * The tree lock is not taken internally. This is a utility function and
319 * probably isn't what you want to call (see set/clear_extent_bit).
321 static int insert_state(struct extent_io_tree
*tree
,
322 struct extent_state
*state
, u64 start
, u64 end
,
325 struct rb_node
*node
;
326 int bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
330 printk(KERN_ERR
"btrfs end < start %llu %llu\n",
331 (unsigned long long)end
,
332 (unsigned long long)start
);
335 state
->start
= start
;
337 ret
= set_state_cb(tree
, state
, bits
);
341 if (bits_to_set
& EXTENT_DIRTY
)
342 tree
->dirty_bytes
+= end
- start
+ 1;
343 state
->state
|= bits_to_set
;
344 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
346 struct extent_state
*found
;
347 found
= rb_entry(node
, struct extent_state
, rb_node
);
348 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
349 "%llu %llu\n", (unsigned long long)found
->start
,
350 (unsigned long long)found
->end
,
351 (unsigned long long)start
, (unsigned long long)end
);
352 free_extent_state(state
);
356 merge_state(tree
, state
);
360 static int split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
363 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
364 return tree
->ops
->split_extent_hook(tree
->mapping
->host
,
370 * split a given extent state struct in two, inserting the preallocated
371 * struct 'prealloc' as the newly created second half. 'split' indicates an
372 * offset inside 'orig' where it should be split.
375 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
376 * are two extent state structs in the tree:
377 * prealloc: [orig->start, split - 1]
378 * orig: [ split, orig->end ]
380 * The tree locks are not taken by this function. They need to be held
383 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
384 struct extent_state
*prealloc
, u64 split
)
386 struct rb_node
*node
;
388 split_cb(tree
, orig
, split
);
390 prealloc
->start
= orig
->start
;
391 prealloc
->end
= split
- 1;
392 prealloc
->state
= orig
->state
;
395 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
397 free_extent_state(prealloc
);
400 prealloc
->tree
= tree
;
405 * utility function to clear some bits in an extent state struct.
406 * it will optionally wake up any one waiting on this state (wake == 1), or
407 * forcibly remove the state from the tree (delete == 1).
409 * If no bits are set on the state struct after clearing things, the
410 * struct is freed and removed from the tree
412 static int clear_state_bit(struct extent_io_tree
*tree
,
413 struct extent_state
*state
,
416 int bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
417 int ret
= state
->state
& bits_to_clear
;
419 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
420 u64 range
= state
->end
- state
->start
+ 1;
421 WARN_ON(range
> tree
->dirty_bytes
);
422 tree
->dirty_bytes
-= range
;
424 clear_state_cb(tree
, state
, bits
);
425 state
->state
&= ~bits_to_clear
;
428 if (state
->state
== 0) {
430 rb_erase(&state
->rb_node
, &tree
->state
);
432 free_extent_state(state
);
437 merge_state(tree
, state
);
443 * clear some bits on a range in the tree. This may require splitting
444 * or inserting elements in the tree, so the gfp mask is used to
445 * indicate which allocations or sleeping are allowed.
447 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
448 * the given range from the tree regardless of state (ie for truncate).
450 * the range [start, end] is inclusive.
452 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
453 * bits were already set, or zero if none of the bits were already set.
455 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
456 int bits
, int wake
, int delete,
457 struct extent_state
**cached_state
,
460 struct extent_state
*state
;
461 struct extent_state
*cached
;
462 struct extent_state
*prealloc
= NULL
;
463 struct rb_node
*next_node
;
464 struct rb_node
*node
;
471 bits
|= ~EXTENT_CTLBITS
;
472 bits
|= EXTENT_FIRST_DELALLOC
;
474 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
477 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
478 prealloc
= alloc_extent_state(mask
);
483 spin_lock(&tree
->lock
);
485 cached
= *cached_state
;
488 *cached_state
= NULL
;
492 if (cached
&& cached
->tree
&& cached
->start
== start
) {
494 atomic_dec(&cached
->refs
);
499 free_extent_state(cached
);
502 * this search will find the extents that end after
505 node
= tree_search(tree
, start
);
508 state
= rb_entry(node
, struct extent_state
, rb_node
);
510 if (state
->start
> end
)
512 WARN_ON(state
->end
< start
);
513 last_end
= state
->end
;
516 * | ---- desired range ---- |
518 * | ------------- state -------------- |
520 * We need to split the extent we found, and may flip
521 * bits on second half.
523 * If the extent we found extends past our range, we
524 * just split and search again. It'll get split again
525 * the next time though.
527 * If the extent we found is inside our range, we clear
528 * the desired bit on it.
531 if (state
->start
< start
) {
533 prealloc
= alloc_extent_state(GFP_ATOMIC
);
534 err
= split_state(tree
, state
, prealloc
, start
);
535 BUG_ON(err
== -EEXIST
);
539 if (state
->end
<= end
) {
540 set
|= clear_state_bit(tree
, state
, &bits
, wake
);
541 if (last_end
== (u64
)-1)
543 start
= last_end
+ 1;
548 * | ---- desired range ---- |
550 * We need to split the extent, and clear the bit
553 if (state
->start
<= end
&& state
->end
> end
) {
555 prealloc
= alloc_extent_state(GFP_ATOMIC
);
556 err
= split_state(tree
, state
, prealloc
, end
+ 1);
557 BUG_ON(err
== -EEXIST
);
561 set
|= clear_state_bit(tree
, prealloc
, &bits
, wake
);
567 if (state
->end
< end
&& prealloc
&& !need_resched())
568 next_node
= rb_next(&state
->rb_node
);
572 set
|= clear_state_bit(tree
, state
, &bits
, wake
);
573 if (last_end
== (u64
)-1)
575 start
= last_end
+ 1;
576 if (start
<= end
&& next_node
) {
577 state
= rb_entry(next_node
, struct extent_state
,
579 if (state
->start
== start
)
585 spin_unlock(&tree
->lock
);
587 free_extent_state(prealloc
);
594 spin_unlock(&tree
->lock
);
595 if (mask
& __GFP_WAIT
)
600 static int wait_on_state(struct extent_io_tree
*tree
,
601 struct extent_state
*state
)
602 __releases(tree
->lock
)
603 __acquires(tree
->lock
)
606 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
607 spin_unlock(&tree
->lock
);
609 spin_lock(&tree
->lock
);
610 finish_wait(&state
->wq
, &wait
);
615 * waits for one or more bits to clear on a range in the state tree.
616 * The range [start, end] is inclusive.
617 * The tree lock is taken by this function
619 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
621 struct extent_state
*state
;
622 struct rb_node
*node
;
624 spin_lock(&tree
->lock
);
628 * this search will find all the extents that end after
631 node
= tree_search(tree
, start
);
635 state
= rb_entry(node
, struct extent_state
, rb_node
);
637 if (state
->start
> end
)
640 if (state
->state
& bits
) {
641 start
= state
->start
;
642 atomic_inc(&state
->refs
);
643 wait_on_state(tree
, state
);
644 free_extent_state(state
);
647 start
= state
->end
+ 1;
652 if (need_resched()) {
653 spin_unlock(&tree
->lock
);
655 spin_lock(&tree
->lock
);
659 spin_unlock(&tree
->lock
);
663 static int set_state_bits(struct extent_io_tree
*tree
,
664 struct extent_state
*state
,
668 int bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
670 ret
= set_state_cb(tree
, state
, bits
);
673 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
674 u64 range
= state
->end
- state
->start
+ 1;
675 tree
->dirty_bytes
+= range
;
677 state
->state
|= bits_to_set
;
682 static void cache_state(struct extent_state
*state
,
683 struct extent_state
**cached_ptr
)
685 if (cached_ptr
&& !(*cached_ptr
)) {
686 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
688 atomic_inc(&state
->refs
);
693 static void uncache_state(struct extent_state
**cached_ptr
)
695 if (cached_ptr
&& (*cached_ptr
)) {
696 struct extent_state
*state
= *cached_ptr
;
698 free_extent_state(state
);
703 * set some bits on a range in the tree. This may require allocations or
704 * sleeping, so the gfp mask is used to indicate what is allowed.
706 * If any of the exclusive bits are set, this will fail with -EEXIST if some
707 * part of the range already has the desired bits set. The start of the
708 * existing range is returned in failed_start in this case.
710 * [start, end] is inclusive This takes the tree lock.
713 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
714 int bits
, int exclusive_bits
, u64
*failed_start
,
715 struct extent_state
**cached_state
, gfp_t mask
)
717 struct extent_state
*state
;
718 struct extent_state
*prealloc
= NULL
;
719 struct rb_node
*node
;
724 bits
|= EXTENT_FIRST_DELALLOC
;
726 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
727 prealloc
= alloc_extent_state(mask
);
732 spin_lock(&tree
->lock
);
733 if (cached_state
&& *cached_state
) {
734 state
= *cached_state
;
735 if (state
->start
== start
&& state
->tree
) {
736 node
= &state
->rb_node
;
741 * this search will find all the extents that end after
744 node
= tree_search(tree
, start
);
746 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
748 BUG_ON(err
== -EEXIST
);
751 state
= rb_entry(node
, struct extent_state
, rb_node
);
753 last_start
= state
->start
;
754 last_end
= state
->end
;
757 * | ---- desired range ---- |
760 * Just lock what we found and keep going
762 if (state
->start
== start
&& state
->end
<= end
) {
763 struct rb_node
*next_node
;
764 if (state
->state
& exclusive_bits
) {
765 *failed_start
= state
->start
;
770 err
= set_state_bits(tree
, state
, &bits
);
774 cache_state(state
, cached_state
);
775 merge_state(tree
, state
);
776 if (last_end
== (u64
)-1)
779 start
= last_end
+ 1;
780 if (start
< end
&& prealloc
&& !need_resched()) {
781 next_node
= rb_next(node
);
783 state
= rb_entry(next_node
, struct extent_state
,
785 if (state
->start
== start
)
793 * | ---- desired range ---- |
796 * | ------------- state -------------- |
798 * We need to split the extent we found, and may flip bits on
801 * If the extent we found extends past our
802 * range, we just split and search again. It'll get split
803 * again the next time though.
805 * If the extent we found is inside our range, we set the
808 if (state
->start
< start
) {
809 if (state
->state
& exclusive_bits
) {
810 *failed_start
= start
;
814 err
= split_state(tree
, state
, prealloc
, start
);
815 BUG_ON(err
== -EEXIST
);
819 if (state
->end
<= end
) {
820 err
= set_state_bits(tree
, state
, &bits
);
823 cache_state(state
, cached_state
);
824 merge_state(tree
, state
);
825 if (last_end
== (u64
)-1)
827 start
= last_end
+ 1;
832 * | ---- desired range ---- |
833 * | state | or | state |
835 * There's a hole, we need to insert something in it and
836 * ignore the extent we found.
838 if (state
->start
> start
) {
840 if (end
< last_start
)
843 this_end
= last_start
- 1;
844 err
= insert_state(tree
, prealloc
, start
, this_end
,
846 BUG_ON(err
== -EEXIST
);
851 cache_state(prealloc
, cached_state
);
853 start
= this_end
+ 1;
857 * | ---- desired range ---- |
859 * We need to split the extent, and set the bit
862 if (state
->start
<= end
&& state
->end
> end
) {
863 if (state
->state
& exclusive_bits
) {
864 *failed_start
= start
;
868 err
= split_state(tree
, state
, prealloc
, end
+ 1);
869 BUG_ON(err
== -EEXIST
);
871 err
= set_state_bits(tree
, prealloc
, &bits
);
876 cache_state(prealloc
, cached_state
);
877 merge_state(tree
, prealloc
);
885 spin_unlock(&tree
->lock
);
887 free_extent_state(prealloc
);
894 spin_unlock(&tree
->lock
);
895 if (mask
& __GFP_WAIT
)
900 /* wrappers around set/clear extent bit */
901 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
904 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
908 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
909 int bits
, gfp_t mask
)
911 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
915 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
916 int bits
, gfp_t mask
)
918 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
921 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
922 struct extent_state
**cached_state
, gfp_t mask
)
924 return set_extent_bit(tree
, start
, end
,
925 EXTENT_DELALLOC
| EXTENT_DIRTY
| EXTENT_UPTODATE
,
926 0, NULL
, cached_state
, mask
);
929 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
932 return clear_extent_bit(tree
, start
, end
,
933 EXTENT_DIRTY
| EXTENT_DELALLOC
|
934 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
937 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
940 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
944 static int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
947 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0,
951 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
952 struct extent_state
**cached_state
, gfp_t mask
)
954 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0,
955 NULL
, cached_state
, mask
);
958 static int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
,
959 u64 end
, struct extent_state
**cached_state
,
962 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
966 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
968 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
972 * either insert or lock state struct between start and end use mask to tell
973 * us if waiting is desired.
975 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
976 int bits
, struct extent_state
**cached_state
, gfp_t mask
)
981 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
982 EXTENT_LOCKED
, &failed_start
,
984 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
985 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
986 start
= failed_start
;
990 WARN_ON(start
> end
);
995 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
997 return lock_extent_bits(tree
, start
, end
, 0, NULL
, mask
);
1000 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1006 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1007 &failed_start
, NULL
, mask
);
1008 if (err
== -EEXIST
) {
1009 if (failed_start
> start
)
1010 clear_extent_bit(tree
, start
, failed_start
- 1,
1011 EXTENT_LOCKED
, 1, 0, NULL
, mask
);
1017 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1018 struct extent_state
**cached
, gfp_t mask
)
1020 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1024 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
1026 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1031 * helper function to set pages and extents in the tree dirty
1033 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1035 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1036 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1039 while (index
<= end_index
) {
1040 page
= find_get_page(tree
->mapping
, index
);
1042 __set_page_dirty_nobuffers(page
);
1043 page_cache_release(page
);
1050 * helper function to set both pages and extents in the tree writeback
1052 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1054 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1055 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1058 while (index
<= end_index
) {
1059 page
= find_get_page(tree
->mapping
, index
);
1061 set_page_writeback(page
);
1062 page_cache_release(page
);
1069 * find the first offset in the io tree with 'bits' set. zero is
1070 * returned if we find something, and *start_ret and *end_ret are
1071 * set to reflect the state struct that was found.
1073 * If nothing was found, 1 is returned, < 0 on error
1075 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1076 u64
*start_ret
, u64
*end_ret
, int bits
)
1078 struct rb_node
*node
;
1079 struct extent_state
*state
;
1082 spin_lock(&tree
->lock
);
1084 * this search will find all the extents that end after
1087 node
= tree_search(tree
, start
);
1092 state
= rb_entry(node
, struct extent_state
, rb_node
);
1093 if (state
->end
>= start
&& (state
->state
& bits
)) {
1094 *start_ret
= state
->start
;
1095 *end_ret
= state
->end
;
1099 node
= rb_next(node
);
1104 spin_unlock(&tree
->lock
);
1108 /* find the first state struct with 'bits' set after 'start', and
1109 * return it. tree->lock must be held. NULL will returned if
1110 * nothing was found after 'start'
1112 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
1113 u64 start
, int bits
)
1115 struct rb_node
*node
;
1116 struct extent_state
*state
;
1119 * this search will find all the extents that end after
1122 node
= tree_search(tree
, start
);
1127 state
= rb_entry(node
, struct extent_state
, rb_node
);
1128 if (state
->end
>= start
&& (state
->state
& bits
))
1131 node
= rb_next(node
);
1140 * find a contiguous range of bytes in the file marked as delalloc, not
1141 * more than 'max_bytes'. start and end are used to return the range,
1143 * 1 is returned if we find something, 0 if nothing was in the tree
1145 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1146 u64
*start
, u64
*end
, u64 max_bytes
,
1147 struct extent_state
**cached_state
)
1149 struct rb_node
*node
;
1150 struct extent_state
*state
;
1151 u64 cur_start
= *start
;
1153 u64 total_bytes
= 0;
1155 spin_lock(&tree
->lock
);
1158 * this search will find all the extents that end after
1161 node
= tree_search(tree
, cur_start
);
1169 state
= rb_entry(node
, struct extent_state
, rb_node
);
1170 if (found
&& (state
->start
!= cur_start
||
1171 (state
->state
& EXTENT_BOUNDARY
))) {
1174 if (!(state
->state
& EXTENT_DELALLOC
)) {
1180 *start
= state
->start
;
1181 *cached_state
= state
;
1182 atomic_inc(&state
->refs
);
1186 cur_start
= state
->end
+ 1;
1187 node
= rb_next(node
);
1190 total_bytes
+= state
->end
- state
->start
+ 1;
1191 if (total_bytes
>= max_bytes
)
1195 spin_unlock(&tree
->lock
);
1199 static noinline
int __unlock_for_delalloc(struct inode
*inode
,
1200 struct page
*locked_page
,
1204 struct page
*pages
[16];
1205 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1206 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1207 unsigned long nr_pages
= end_index
- index
+ 1;
1210 if (index
== locked_page
->index
&& end_index
== index
)
1213 while (nr_pages
> 0) {
1214 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1215 min_t(unsigned long, nr_pages
,
1216 ARRAY_SIZE(pages
)), pages
);
1217 for (i
= 0; i
< ret
; i
++) {
1218 if (pages
[i
] != locked_page
)
1219 unlock_page(pages
[i
]);
1220 page_cache_release(pages
[i
]);
1229 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1230 struct page
*locked_page
,
1234 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1235 unsigned long start_index
= index
;
1236 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1237 unsigned long pages_locked
= 0;
1238 struct page
*pages
[16];
1239 unsigned long nrpages
;
1243 /* the caller is responsible for locking the start index */
1244 if (index
== locked_page
->index
&& index
== end_index
)
1247 /* skip the page at the start index */
1248 nrpages
= end_index
- index
+ 1;
1249 while (nrpages
> 0) {
1250 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1251 min_t(unsigned long,
1252 nrpages
, ARRAY_SIZE(pages
)), pages
);
1257 /* now we have an array of pages, lock them all */
1258 for (i
= 0; i
< ret
; i
++) {
1260 * the caller is taking responsibility for
1263 if (pages
[i
] != locked_page
) {
1264 lock_page(pages
[i
]);
1265 if (!PageDirty(pages
[i
]) ||
1266 pages
[i
]->mapping
!= inode
->i_mapping
) {
1268 unlock_page(pages
[i
]);
1269 page_cache_release(pages
[i
]);
1273 page_cache_release(pages
[i
]);
1282 if (ret
&& pages_locked
) {
1283 __unlock_for_delalloc(inode
, locked_page
,
1285 ((u64
)(start_index
+ pages_locked
- 1)) <<
1292 * find a contiguous range of bytes in the file marked as delalloc, not
1293 * more than 'max_bytes'. start and end are used to return the range,
1295 * 1 is returned if we find something, 0 if nothing was in the tree
1297 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1298 struct extent_io_tree
*tree
,
1299 struct page
*locked_page
,
1300 u64
*start
, u64
*end
,
1306 struct extent_state
*cached_state
= NULL
;
1311 /* step one, find a bunch of delalloc bytes starting at start */
1312 delalloc_start
= *start
;
1314 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1315 max_bytes
, &cached_state
);
1316 if (!found
|| delalloc_end
<= *start
) {
1317 *start
= delalloc_start
;
1318 *end
= delalloc_end
;
1319 free_extent_state(cached_state
);
1324 * start comes from the offset of locked_page. We have to lock
1325 * pages in order, so we can't process delalloc bytes before
1328 if (delalloc_start
< *start
)
1329 delalloc_start
= *start
;
1332 * make sure to limit the number of pages we try to lock down
1335 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
)
1336 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1338 /* step two, lock all the pages after the page that has start */
1339 ret
= lock_delalloc_pages(inode
, locked_page
,
1340 delalloc_start
, delalloc_end
);
1341 if (ret
== -EAGAIN
) {
1342 /* some of the pages are gone, lets avoid looping by
1343 * shortening the size of the delalloc range we're searching
1345 free_extent_state(cached_state
);
1347 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1348 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1358 /* step three, lock the state bits for the whole range */
1359 lock_extent_bits(tree
, delalloc_start
, delalloc_end
,
1360 0, &cached_state
, GFP_NOFS
);
1362 /* then test to make sure it is all still delalloc */
1363 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1364 EXTENT_DELALLOC
, 1, cached_state
);
1366 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1367 &cached_state
, GFP_NOFS
);
1368 __unlock_for_delalloc(inode
, locked_page
,
1369 delalloc_start
, delalloc_end
);
1373 free_extent_state(cached_state
);
1374 *start
= delalloc_start
;
1375 *end
= delalloc_end
;
1380 int extent_clear_unlock_delalloc(struct inode
*inode
,
1381 struct extent_io_tree
*tree
,
1382 u64 start
, u64 end
, struct page
*locked_page
,
1386 struct page
*pages
[16];
1387 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1388 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1389 unsigned long nr_pages
= end_index
- index
+ 1;
1393 if (op
& EXTENT_CLEAR_UNLOCK
)
1394 clear_bits
|= EXTENT_LOCKED
;
1395 if (op
& EXTENT_CLEAR_DIRTY
)
1396 clear_bits
|= EXTENT_DIRTY
;
1398 if (op
& EXTENT_CLEAR_DELALLOC
)
1399 clear_bits
|= EXTENT_DELALLOC
;
1401 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1402 if (!(op
& (EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
1403 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
|
1404 EXTENT_SET_PRIVATE2
)))
1407 while (nr_pages
> 0) {
1408 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1409 min_t(unsigned long,
1410 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1411 for (i
= 0; i
< ret
; i
++) {
1413 if (op
& EXTENT_SET_PRIVATE2
)
1414 SetPagePrivate2(pages
[i
]);
1416 if (pages
[i
] == locked_page
) {
1417 page_cache_release(pages
[i
]);
1420 if (op
& EXTENT_CLEAR_DIRTY
)
1421 clear_page_dirty_for_io(pages
[i
]);
1422 if (op
& EXTENT_SET_WRITEBACK
)
1423 set_page_writeback(pages
[i
]);
1424 if (op
& EXTENT_END_WRITEBACK
)
1425 end_page_writeback(pages
[i
]);
1426 if (op
& EXTENT_CLEAR_UNLOCK_PAGE
)
1427 unlock_page(pages
[i
]);
1428 page_cache_release(pages
[i
]);
1438 * count the number of bytes in the tree that have a given bit(s)
1439 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1440 * cached. The total number found is returned.
1442 u64
count_range_bits(struct extent_io_tree
*tree
,
1443 u64
*start
, u64 search_end
, u64 max_bytes
,
1444 unsigned long bits
, int contig
)
1446 struct rb_node
*node
;
1447 struct extent_state
*state
;
1448 u64 cur_start
= *start
;
1449 u64 total_bytes
= 0;
1453 if (search_end
<= cur_start
) {
1458 spin_lock(&tree
->lock
);
1459 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1460 total_bytes
= tree
->dirty_bytes
;
1464 * this search will find all the extents that end after
1467 node
= tree_search(tree
, cur_start
);
1472 state
= rb_entry(node
, struct extent_state
, rb_node
);
1473 if (state
->start
> search_end
)
1475 if (contig
&& found
&& state
->start
> last
+ 1)
1477 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1478 total_bytes
+= min(search_end
, state
->end
) + 1 -
1479 max(cur_start
, state
->start
);
1480 if (total_bytes
>= max_bytes
)
1483 *start
= state
->start
;
1487 } else if (contig
&& found
) {
1490 node
= rb_next(node
);
1495 spin_unlock(&tree
->lock
);
1500 * set the private field for a given byte offset in the tree. If there isn't
1501 * an extent_state there already, this does nothing.
1503 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1505 struct rb_node
*node
;
1506 struct extent_state
*state
;
1509 spin_lock(&tree
->lock
);
1511 * this search will find all the extents that end after
1514 node
= tree_search(tree
, start
);
1519 state
= rb_entry(node
, struct extent_state
, rb_node
);
1520 if (state
->start
!= start
) {
1524 state
->private = private;
1526 spin_unlock(&tree
->lock
);
1530 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1532 struct rb_node
*node
;
1533 struct extent_state
*state
;
1536 spin_lock(&tree
->lock
);
1538 * this search will find all the extents that end after
1541 node
= tree_search(tree
, start
);
1546 state
= rb_entry(node
, struct extent_state
, rb_node
);
1547 if (state
->start
!= start
) {
1551 *private = state
->private;
1553 spin_unlock(&tree
->lock
);
1558 * searches a range in the state tree for a given mask.
1559 * If 'filled' == 1, this returns 1 only if every extent in the tree
1560 * has the bits set. Otherwise, 1 is returned if any bit in the
1561 * range is found set.
1563 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1564 int bits
, int filled
, struct extent_state
*cached
)
1566 struct extent_state
*state
= NULL
;
1567 struct rb_node
*node
;
1570 spin_lock(&tree
->lock
);
1571 if (cached
&& cached
->tree
&& cached
->start
== start
)
1572 node
= &cached
->rb_node
;
1574 node
= tree_search(tree
, start
);
1575 while (node
&& start
<= end
) {
1576 state
= rb_entry(node
, struct extent_state
, rb_node
);
1578 if (filled
&& state
->start
> start
) {
1583 if (state
->start
> end
)
1586 if (state
->state
& bits
) {
1590 } else if (filled
) {
1595 if (state
->end
== (u64
)-1)
1598 start
= state
->end
+ 1;
1601 node
= rb_next(node
);
1608 spin_unlock(&tree
->lock
);
1613 * helper function to set a given page up to date if all the
1614 * extents in the tree for that page are up to date
1616 static int check_page_uptodate(struct extent_io_tree
*tree
,
1619 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1620 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1621 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1622 SetPageUptodate(page
);
1627 * helper function to unlock a page if all the extents in the tree
1628 * for that page are unlocked
1630 static int check_page_locked(struct extent_io_tree
*tree
,
1633 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1634 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1635 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0, NULL
))
1641 * helper function to end page writeback if all the extents
1642 * in the tree for that page are done with writeback
1644 static int check_page_writeback(struct extent_io_tree
*tree
,
1647 end_page_writeback(page
);
1651 /* lots and lots of room for performance fixes in the end_bio funcs */
1654 * after a writepage IO is done, we need to:
1655 * clear the uptodate bits on error
1656 * clear the writeback bits in the extent tree for this IO
1657 * end_page_writeback if the page has no more pending IO
1659 * Scheduling is not allowed, so the extent state tree is expected
1660 * to have one and only one object corresponding to this IO.
1662 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1664 int uptodate
= err
== 0;
1665 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1666 struct extent_io_tree
*tree
;
1673 struct page
*page
= bvec
->bv_page
;
1674 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1676 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1678 end
= start
+ bvec
->bv_len
- 1;
1680 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1685 if (--bvec
>= bio
->bi_io_vec
)
1686 prefetchw(&bvec
->bv_page
->flags
);
1687 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1688 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
1689 end
, NULL
, uptodate
);
1694 if (!uptodate
&& tree
->ops
&&
1695 tree
->ops
->writepage_io_failed_hook
) {
1696 ret
= tree
->ops
->writepage_io_failed_hook(bio
, page
,
1699 uptodate
= (err
== 0);
1705 clear_extent_uptodate(tree
, start
, end
, NULL
, GFP_NOFS
);
1706 ClearPageUptodate(page
);
1711 end_page_writeback(page
);
1713 check_page_writeback(tree
, page
);
1714 } while (bvec
>= bio
->bi_io_vec
);
1720 * after a readpage IO is done, we need to:
1721 * clear the uptodate bits on error
1722 * set the uptodate bits if things worked
1723 * set the page up to date if all extents in the tree are uptodate
1724 * clear the lock bit in the extent tree
1725 * unlock the page if there are no other extents locked for it
1727 * Scheduling is not allowed, so the extent state tree is expected
1728 * to have one and only one object corresponding to this IO.
1730 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1732 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1733 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1734 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1735 struct extent_io_tree
*tree
;
1745 struct page
*page
= bvec
->bv_page
;
1746 struct extent_state
*cached
= NULL
;
1747 struct extent_state
*state
;
1749 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1751 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1753 end
= start
+ bvec
->bv_len
- 1;
1755 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1760 if (++bvec
<= bvec_end
)
1761 prefetchw(&bvec
->bv_page
->flags
);
1763 spin_lock(&tree
->lock
);
1764 state
= find_first_extent_bit_state(tree
, start
, EXTENT_LOCKED
);
1765 if (state
&& state
->start
== start
) {
1767 * take a reference on the state, unlock will drop
1770 cache_state(state
, &cached
);
1772 spin_unlock(&tree
->lock
);
1774 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1775 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1780 if (!uptodate
&& tree
->ops
&&
1781 tree
->ops
->readpage_io_failed_hook
) {
1782 ret
= tree
->ops
->readpage_io_failed_hook(bio
, page
,
1786 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1789 uncache_state(&cached
);
1795 set_extent_uptodate(tree
, start
, end
, &cached
,
1798 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
1802 SetPageUptodate(page
);
1804 ClearPageUptodate(page
);
1810 check_page_uptodate(tree
, page
);
1812 ClearPageUptodate(page
);
1815 check_page_locked(tree
, page
);
1817 } while (bvec
<= bvec_end
);
1823 * IO done from prepare_write is pretty simple, we just unlock
1824 * the structs in the extent tree when done, and set the uptodate bits
1827 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1829 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1830 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1831 struct extent_io_tree
*tree
;
1836 struct page
*page
= bvec
->bv_page
;
1837 struct extent_state
*cached
= NULL
;
1838 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1840 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1842 end
= start
+ bvec
->bv_len
- 1;
1844 if (--bvec
>= bio
->bi_io_vec
)
1845 prefetchw(&bvec
->bv_page
->flags
);
1848 set_extent_uptodate(tree
, start
, end
, &cached
,
1851 ClearPageUptodate(page
);
1855 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
1857 } while (bvec
>= bio
->bi_io_vec
);
1863 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1868 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1870 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1871 while (!bio
&& (nr_vecs
/= 2))
1872 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1877 bio
->bi_bdev
= bdev
;
1878 bio
->bi_sector
= first_sector
;
1883 static int submit_one_bio(int rw
, struct bio
*bio
, int mirror_num
,
1884 unsigned long bio_flags
)
1887 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1888 struct page
*page
= bvec
->bv_page
;
1889 struct extent_io_tree
*tree
= bio
->bi_private
;
1892 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1894 bio
->bi_private
= NULL
;
1898 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
1899 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
1900 mirror_num
, bio_flags
, start
);
1902 submit_bio(rw
, bio
);
1903 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1909 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1910 struct page
*page
, sector_t sector
,
1911 size_t size
, unsigned long offset
,
1912 struct block_device
*bdev
,
1913 struct bio
**bio_ret
,
1914 unsigned long max_pages
,
1915 bio_end_io_t end_io_func
,
1917 unsigned long prev_bio_flags
,
1918 unsigned long bio_flags
)
1924 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
1925 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
1926 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
1928 if (bio_ret
&& *bio_ret
) {
1931 contig
= bio
->bi_sector
== sector
;
1933 contig
= bio
->bi_sector
+ (bio
->bi_size
>> 9) ==
1936 if (prev_bio_flags
!= bio_flags
|| !contig
||
1937 (tree
->ops
&& tree
->ops
->merge_bio_hook
&&
1938 tree
->ops
->merge_bio_hook(page
, offset
, page_size
, bio
,
1940 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
1941 ret
= submit_one_bio(rw
, bio
, mirror_num
,
1948 if (this_compressed
)
1951 nr
= bio_get_nr_vecs(bdev
);
1953 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1957 bio_add_page(bio
, page
, page_size
, offset
);
1958 bio
->bi_end_io
= end_io_func
;
1959 bio
->bi_private
= tree
;
1964 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
1969 void set_page_extent_mapped(struct page
*page
)
1971 if (!PagePrivate(page
)) {
1972 SetPagePrivate(page
);
1973 page_cache_get(page
);
1974 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1978 static void set_page_extent_head(struct page
*page
, unsigned long len
)
1980 WARN_ON(!PagePrivate(page
));
1981 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1985 * basic readpage implementation. Locked extent state structs are inserted
1986 * into the tree that are removed when the IO is done (by the end_io
1989 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1991 get_extent_t
*get_extent
,
1992 struct bio
**bio
, int mirror_num
,
1993 unsigned long *bio_flags
)
1995 struct inode
*inode
= page
->mapping
->host
;
1996 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1997 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2001 u64 last_byte
= i_size_read(inode
);
2005 struct extent_map
*em
;
2006 struct block_device
*bdev
;
2007 struct btrfs_ordered_extent
*ordered
;
2010 size_t page_offset
= 0;
2012 size_t disk_io_size
;
2013 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2014 unsigned long this_bio_flag
= 0;
2016 set_page_extent_mapped(page
);
2020 lock_extent(tree
, start
, end
, GFP_NOFS
);
2021 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
2024 unlock_extent(tree
, start
, end
, GFP_NOFS
);
2025 btrfs_start_ordered_extent(inode
, ordered
, 1);
2026 btrfs_put_ordered_extent(ordered
);
2029 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2031 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2034 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2035 userpage
= kmap_atomic(page
, KM_USER0
);
2036 memset(userpage
+ zero_offset
, 0, iosize
);
2037 flush_dcache_page(page
);
2038 kunmap_atomic(userpage
, KM_USER0
);
2041 while (cur
<= end
) {
2042 if (cur
>= last_byte
) {
2044 struct extent_state
*cached
= NULL
;
2046 iosize
= PAGE_CACHE_SIZE
- page_offset
;
2047 userpage
= kmap_atomic(page
, KM_USER0
);
2048 memset(userpage
+ page_offset
, 0, iosize
);
2049 flush_dcache_page(page
);
2050 kunmap_atomic(userpage
, KM_USER0
);
2051 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2053 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2057 em
= get_extent(inode
, page
, page_offset
, cur
,
2059 if (IS_ERR(em
) || !em
) {
2061 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
2064 extent_offset
= cur
- em
->start
;
2065 BUG_ON(extent_map_end(em
) <= cur
);
2068 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2069 this_bio_flag
= EXTENT_BIO_COMPRESSED
;
2070 extent_set_compress_type(&this_bio_flag
,
2074 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2075 cur_end
= min(extent_map_end(em
) - 1, end
);
2076 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2077 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2078 disk_io_size
= em
->block_len
;
2079 sector
= em
->block_start
>> 9;
2081 sector
= (em
->block_start
+ extent_offset
) >> 9;
2082 disk_io_size
= iosize
;
2085 block_start
= em
->block_start
;
2086 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2087 block_start
= EXTENT_MAP_HOLE
;
2088 free_extent_map(em
);
2091 /* we've found a hole, just zero and go on */
2092 if (block_start
== EXTENT_MAP_HOLE
) {
2094 struct extent_state
*cached
= NULL
;
2096 userpage
= kmap_atomic(page
, KM_USER0
);
2097 memset(userpage
+ page_offset
, 0, iosize
);
2098 flush_dcache_page(page
);
2099 kunmap_atomic(userpage
, KM_USER0
);
2101 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2103 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2106 page_offset
+= iosize
;
2109 /* the get_extent function already copied into the page */
2110 if (test_range_bit(tree
, cur
, cur_end
,
2111 EXTENT_UPTODATE
, 1, NULL
)) {
2112 check_page_uptodate(tree
, page
);
2113 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2115 page_offset
+= iosize
;
2118 /* we have an inline extent but it didn't get marked up
2119 * to date. Error out
2121 if (block_start
== EXTENT_MAP_INLINE
) {
2123 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2125 page_offset
+= iosize
;
2130 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
2131 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
2135 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2137 ret
= submit_extent_page(READ
, tree
, page
,
2138 sector
, disk_io_size
, page_offset
,
2140 end_bio_extent_readpage
, mirror_num
,
2144 *bio_flags
= this_bio_flag
;
2149 page_offset
+= iosize
;
2152 if (!PageError(page
))
2153 SetPageUptodate(page
);
2159 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2160 get_extent_t
*get_extent
)
2162 struct bio
*bio
= NULL
;
2163 unsigned long bio_flags
= 0;
2166 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, 0,
2169 ret
= submit_one_bio(READ
, bio
, 0, bio_flags
);
2173 static noinline
void update_nr_written(struct page
*page
,
2174 struct writeback_control
*wbc
,
2175 unsigned long nr_written
)
2177 wbc
->nr_to_write
-= nr_written
;
2178 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
2179 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
2180 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
2184 * the writepage semantics are similar to regular writepage. extent
2185 * records are inserted to lock ranges in the tree, and as dirty areas
2186 * are found, they are marked writeback. Then the lock bits are removed
2187 * and the end_io handler clears the writeback ranges
2189 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
2192 struct inode
*inode
= page
->mapping
->host
;
2193 struct extent_page_data
*epd
= data
;
2194 struct extent_io_tree
*tree
= epd
->tree
;
2195 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2197 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2201 u64 last_byte
= i_size_read(inode
);
2205 struct extent_state
*cached_state
= NULL
;
2206 struct extent_map
*em
;
2207 struct block_device
*bdev
;
2210 size_t pg_offset
= 0;
2212 loff_t i_size
= i_size_read(inode
);
2213 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2219 unsigned long nr_written
= 0;
2221 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2222 write_flags
= WRITE_SYNC
;
2224 write_flags
= WRITE
;
2226 trace___extent_writepage(page
, inode
, wbc
);
2228 WARN_ON(!PageLocked(page
));
2229 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
2230 if (page
->index
> end_index
||
2231 (page
->index
== end_index
&& !pg_offset
)) {
2232 page
->mapping
->a_ops
->invalidatepage(page
, 0);
2237 if (page
->index
== end_index
) {
2240 userpage
= kmap_atomic(page
, KM_USER0
);
2241 memset(userpage
+ pg_offset
, 0,
2242 PAGE_CACHE_SIZE
- pg_offset
);
2243 kunmap_atomic(userpage
, KM_USER0
);
2244 flush_dcache_page(page
);
2248 set_page_extent_mapped(page
);
2250 delalloc_start
= start
;
2253 if (!epd
->extent_locked
) {
2254 u64 delalloc_to_write
= 0;
2256 * make sure the wbc mapping index is at least updated
2259 update_nr_written(page
, wbc
, 0);
2261 while (delalloc_end
< page_end
) {
2262 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
2267 if (nr_delalloc
== 0) {
2268 delalloc_start
= delalloc_end
+ 1;
2271 tree
->ops
->fill_delalloc(inode
, page
, delalloc_start
,
2272 delalloc_end
, &page_started
,
2275 * delalloc_end is already one less than the total
2276 * length, so we don't subtract one from
2279 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
2282 delalloc_start
= delalloc_end
+ 1;
2284 if (wbc
->nr_to_write
< delalloc_to_write
) {
2287 if (delalloc_to_write
< thresh
* 2)
2288 thresh
= delalloc_to_write
;
2289 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
2293 /* did the fill delalloc function already unlock and start
2299 * we've unlocked the page, so we can't update
2300 * the mapping's writeback index, just update
2303 wbc
->nr_to_write
-= nr_written
;
2307 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
2308 ret
= tree
->ops
->writepage_start_hook(page
, start
,
2310 if (ret
== -EAGAIN
) {
2311 redirty_page_for_writepage(wbc
, page
);
2312 update_nr_written(page
, wbc
, nr_written
);
2320 * we don't want to touch the inode after unlocking the page,
2321 * so we update the mapping writeback index now
2323 update_nr_written(page
, wbc
, nr_written
+ 1);
2326 if (last_byte
<= start
) {
2327 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2328 tree
->ops
->writepage_end_io_hook(page
, start
,
2333 blocksize
= inode
->i_sb
->s_blocksize
;
2335 while (cur
<= end
) {
2336 if (cur
>= last_byte
) {
2337 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2338 tree
->ops
->writepage_end_io_hook(page
, cur
,
2342 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
2344 if (IS_ERR(em
) || !em
) {
2349 extent_offset
= cur
- em
->start
;
2350 BUG_ON(extent_map_end(em
) <= cur
);
2352 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2353 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2354 sector
= (em
->block_start
+ extent_offset
) >> 9;
2356 block_start
= em
->block_start
;
2357 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
2358 free_extent_map(em
);
2362 * compressed and inline extents are written through other
2365 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
2366 block_start
== EXTENT_MAP_INLINE
) {
2368 * end_io notification does not happen here for
2369 * compressed extents
2371 if (!compressed
&& tree
->ops
&&
2372 tree
->ops
->writepage_end_io_hook
)
2373 tree
->ops
->writepage_end_io_hook(page
, cur
,
2376 else if (compressed
) {
2377 /* we don't want to end_page_writeback on
2378 * a compressed extent. this happens
2385 pg_offset
+= iosize
;
2388 /* leave this out until we have a page_mkwrite call */
2389 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2390 EXTENT_DIRTY
, 0, NULL
)) {
2392 pg_offset
+= iosize
;
2396 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2397 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2405 unsigned long max_nr
= end_index
+ 1;
2407 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2408 if (!PageWriteback(page
)) {
2409 printk(KERN_ERR
"btrfs warning page %lu not "
2410 "writeback, cur %llu end %llu\n",
2411 page
->index
, (unsigned long long)cur
,
2412 (unsigned long long)end
);
2415 ret
= submit_extent_page(write_flags
, tree
, page
,
2416 sector
, iosize
, pg_offset
,
2417 bdev
, &epd
->bio
, max_nr
,
2418 end_bio_extent_writepage
,
2424 pg_offset
+= iosize
;
2429 /* make sure the mapping tag for page dirty gets cleared */
2430 set_page_writeback(page
);
2431 end_page_writeback(page
);
2437 /* drop our reference on any cached states */
2438 free_extent_state(cached_state
);
2443 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2444 * @mapping: address space structure to write
2445 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2446 * @writepage: function called for each page
2447 * @data: data passed to writepage function
2449 * If a page is already under I/O, write_cache_pages() skips it, even
2450 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2451 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2452 * and msync() need to guarantee that all the data which was dirty at the time
2453 * the call was made get new I/O started against them. If wbc->sync_mode is
2454 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2455 * existing IO to complete.
2457 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
2458 struct address_space
*mapping
,
2459 struct writeback_control
*wbc
,
2460 writepage_t writepage
, void *data
,
2461 void (*flush_fn
)(void *))
2465 int nr_to_write_done
= 0;
2466 struct pagevec pvec
;
2469 pgoff_t end
; /* Inclusive */
2472 pagevec_init(&pvec
, 0);
2473 if (wbc
->range_cyclic
) {
2474 index
= mapping
->writeback_index
; /* Start from prev offset */
2477 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2478 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2482 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
2483 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2484 PAGECACHE_TAG_DIRTY
, min(end
- index
,
2485 (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2489 for (i
= 0; i
< nr_pages
; i
++) {
2490 struct page
*page
= pvec
.pages
[i
];
2493 * At this point we hold neither mapping->tree_lock nor
2494 * lock on the page itself: the page may be truncated or
2495 * invalidated (changing page->mapping to NULL), or even
2496 * swizzled back from swapper_space to tmpfs file
2499 if (tree
->ops
&& tree
->ops
->write_cache_pages_lock_hook
)
2500 tree
->ops
->write_cache_pages_lock_hook(page
);
2504 if (unlikely(page
->mapping
!= mapping
)) {
2509 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2515 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
2516 if (PageWriteback(page
))
2518 wait_on_page_writeback(page
);
2521 if (PageWriteback(page
) ||
2522 !clear_page_dirty_for_io(page
)) {
2527 ret
= (*writepage
)(page
, wbc
, data
);
2529 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2537 * the filesystem may choose to bump up nr_to_write.
2538 * We have to make sure to honor the new nr_to_write
2541 nr_to_write_done
= wbc
->nr_to_write
<= 0;
2543 pagevec_release(&pvec
);
2546 if (!scanned
&& !done
) {
2548 * We hit the last page and there is more work to be done: wrap
2549 * back to the start of the file
2558 static void flush_epd_write_bio(struct extent_page_data
*epd
)
2562 submit_one_bio(WRITE_SYNC
, epd
->bio
, 0, 0);
2564 submit_one_bio(WRITE
, epd
->bio
, 0, 0);
2569 static noinline
void flush_write_bio(void *data
)
2571 struct extent_page_data
*epd
= data
;
2572 flush_epd_write_bio(epd
);
2575 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2576 get_extent_t
*get_extent
,
2577 struct writeback_control
*wbc
)
2580 struct address_space
*mapping
= page
->mapping
;
2581 struct extent_page_data epd
= {
2584 .get_extent
= get_extent
,
2586 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
2588 struct writeback_control wbc_writepages
= {
2589 .sync_mode
= wbc
->sync_mode
,
2590 .older_than_this
= NULL
,
2592 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2593 .range_end
= (loff_t
)-1,
2596 ret
= __extent_writepage(page
, wbc
, &epd
);
2598 extent_write_cache_pages(tree
, mapping
, &wbc_writepages
,
2599 __extent_writepage
, &epd
, flush_write_bio
);
2600 flush_epd_write_bio(&epd
);
2604 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
2605 u64 start
, u64 end
, get_extent_t
*get_extent
,
2609 struct address_space
*mapping
= inode
->i_mapping
;
2611 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
2614 struct extent_page_data epd
= {
2617 .get_extent
= get_extent
,
2619 .sync_io
= mode
== WB_SYNC_ALL
,
2621 struct writeback_control wbc_writepages
= {
2623 .older_than_this
= NULL
,
2624 .nr_to_write
= nr_pages
* 2,
2625 .range_start
= start
,
2626 .range_end
= end
+ 1,
2629 while (start
<= end
) {
2630 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
2631 if (clear_page_dirty_for_io(page
))
2632 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
2634 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2635 tree
->ops
->writepage_end_io_hook(page
, start
,
2636 start
+ PAGE_CACHE_SIZE
- 1,
2640 page_cache_release(page
);
2641 start
+= PAGE_CACHE_SIZE
;
2644 flush_epd_write_bio(&epd
);
2648 int extent_writepages(struct extent_io_tree
*tree
,
2649 struct address_space
*mapping
,
2650 get_extent_t
*get_extent
,
2651 struct writeback_control
*wbc
)
2654 struct extent_page_data epd
= {
2657 .get_extent
= get_extent
,
2659 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
2662 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
2663 __extent_writepage
, &epd
,
2665 flush_epd_write_bio(&epd
);
2669 int extent_readpages(struct extent_io_tree
*tree
,
2670 struct address_space
*mapping
,
2671 struct list_head
*pages
, unsigned nr_pages
,
2672 get_extent_t get_extent
)
2674 struct bio
*bio
= NULL
;
2676 unsigned long bio_flags
= 0;
2678 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2679 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2681 prefetchw(&page
->flags
);
2682 list_del(&page
->lru
);
2683 if (!add_to_page_cache_lru(page
, mapping
,
2684 page
->index
, GFP_NOFS
)) {
2685 __extent_read_full_page(tree
, page
, get_extent
,
2686 &bio
, 0, &bio_flags
);
2688 page_cache_release(page
);
2690 BUG_ON(!list_empty(pages
));
2692 submit_one_bio(READ
, bio
, 0, bio_flags
);
2697 * basic invalidatepage code, this waits on any locked or writeback
2698 * ranges corresponding to the page, and then deletes any extent state
2699 * records from the tree
2701 int extent_invalidatepage(struct extent_io_tree
*tree
,
2702 struct page
*page
, unsigned long offset
)
2704 struct extent_state
*cached_state
= NULL
;
2705 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2706 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2707 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2709 start
+= (offset
+ blocksize
- 1) & ~(blocksize
- 1);
2713 lock_extent_bits(tree
, start
, end
, 0, &cached_state
, GFP_NOFS
);
2714 wait_on_page_writeback(page
);
2715 clear_extent_bit(tree
, start
, end
,
2716 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
2717 EXTENT_DO_ACCOUNTING
,
2718 1, 1, &cached_state
, GFP_NOFS
);
2723 * simple commit_write call, set_range_dirty is used to mark both
2724 * the pages and the extent records as dirty
2726 int extent_commit_write(struct extent_io_tree
*tree
,
2727 struct inode
*inode
, struct page
*page
,
2728 unsigned from
, unsigned to
)
2730 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2732 set_page_extent_mapped(page
);
2733 set_page_dirty(page
);
2735 if (pos
> inode
->i_size
) {
2736 i_size_write(inode
, pos
);
2737 mark_inode_dirty(inode
);
2742 int extent_prepare_write(struct extent_io_tree
*tree
,
2743 struct inode
*inode
, struct page
*page
,
2744 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2746 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2747 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2749 u64 orig_block_start
;
2752 struct extent_map
*em
;
2753 unsigned blocksize
= 1 << inode
->i_blkbits
;
2754 size_t page_offset
= 0;
2755 size_t block_off_start
;
2756 size_t block_off_end
;
2762 set_page_extent_mapped(page
);
2764 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2765 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2766 orig_block_start
= block_start
;
2768 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2769 while (block_start
<= block_end
) {
2770 em
= get_extent(inode
, page
, page_offset
, block_start
,
2771 block_end
- block_start
+ 1, 1);
2772 if (IS_ERR(em
) || !em
)
2775 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2776 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2777 block_off_end
= block_off_start
+ blocksize
;
2778 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2780 if (!PageUptodate(page
) && isnew
&&
2781 (block_off_end
> to
|| block_off_start
< from
)) {
2784 kaddr
= kmap_atomic(page
, KM_USER0
);
2785 if (block_off_end
> to
)
2786 memset(kaddr
+ to
, 0, block_off_end
- to
);
2787 if (block_off_start
< from
)
2788 memset(kaddr
+ block_off_start
, 0,
2789 from
- block_off_start
);
2790 flush_dcache_page(page
);
2791 kunmap_atomic(kaddr
, KM_USER0
);
2793 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2794 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2795 !isnew
&& !PageUptodate(page
) &&
2796 (block_off_end
> to
|| block_off_start
< from
) &&
2797 !test_range_bit(tree
, block_start
, cur_end
,
2798 EXTENT_UPTODATE
, 1, NULL
)) {
2800 u64 extent_offset
= block_start
- em
->start
;
2802 sector
= (em
->block_start
+ extent_offset
) >> 9;
2803 iosize
= (cur_end
- block_start
+ blocksize
) &
2804 ~((u64
)blocksize
- 1);
2806 * we've already got the extent locked, but we
2807 * need to split the state such that our end_bio
2808 * handler can clear the lock.
2810 set_extent_bit(tree
, block_start
,
2811 block_start
+ iosize
- 1,
2812 EXTENT_LOCKED
, 0, NULL
, NULL
, GFP_NOFS
);
2813 ret
= submit_extent_page(READ
, tree
, page
,
2814 sector
, iosize
, page_offset
, em
->bdev
,
2816 end_bio_extent_preparewrite
, 0,
2821 block_start
= block_start
+ iosize
;
2823 struct extent_state
*cached
= NULL
;
2825 set_extent_uptodate(tree
, block_start
, cur_end
, &cached
,
2827 unlock_extent_cached(tree
, block_start
, cur_end
,
2829 block_start
= cur_end
+ 1;
2831 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2832 free_extent_map(em
);
2835 wait_extent_bit(tree
, orig_block_start
,
2836 block_end
, EXTENT_LOCKED
);
2838 check_page_uptodate(tree
, page
);
2840 /* FIXME, zero out newly allocated blocks on error */
2845 * a helper for releasepage, this tests for areas of the page that
2846 * are locked or under IO and drops the related state bits if it is safe
2849 int try_release_extent_state(struct extent_map_tree
*map
,
2850 struct extent_io_tree
*tree
, struct page
*page
,
2853 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2854 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2857 if (test_range_bit(tree
, start
, end
,
2858 EXTENT_IOBITS
, 0, NULL
))
2861 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2864 * at this point we can safely clear everything except the
2865 * locked bit and the nodatasum bit
2867 ret
= clear_extent_bit(tree
, start
, end
,
2868 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
2871 /* if clear_extent_bit failed for enomem reasons,
2872 * we can't allow the release to continue.
2883 * a helper for releasepage. As long as there are no locked extents
2884 * in the range corresponding to the page, both state records and extent
2885 * map records are removed
2887 int try_release_extent_mapping(struct extent_map_tree
*map
,
2888 struct extent_io_tree
*tree
, struct page
*page
,
2891 struct extent_map
*em
;
2892 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2893 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2895 if ((mask
& __GFP_WAIT
) &&
2896 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2898 while (start
<= end
) {
2899 len
= end
- start
+ 1;
2900 write_lock(&map
->lock
);
2901 em
= lookup_extent_mapping(map
, start
, len
);
2902 if (!em
|| IS_ERR(em
)) {
2903 write_unlock(&map
->lock
);
2906 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
2907 em
->start
!= start
) {
2908 write_unlock(&map
->lock
);
2909 free_extent_map(em
);
2912 if (!test_range_bit(tree
, em
->start
,
2913 extent_map_end(em
) - 1,
2914 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
2916 remove_extent_mapping(map
, em
);
2917 /* once for the rb tree */
2918 free_extent_map(em
);
2920 start
= extent_map_end(em
);
2921 write_unlock(&map
->lock
);
2924 free_extent_map(em
);
2927 return try_release_extent_state(map
, tree
, page
, mask
);
2930 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2931 get_extent_t
*get_extent
)
2933 struct inode
*inode
= mapping
->host
;
2934 struct extent_state
*cached_state
= NULL
;
2935 u64 start
= iblock
<< inode
->i_blkbits
;
2936 sector_t sector
= 0;
2937 size_t blksize
= (1 << inode
->i_blkbits
);
2938 struct extent_map
*em
;
2940 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ blksize
- 1,
2941 0, &cached_state
, GFP_NOFS
);
2942 em
= get_extent(inode
, NULL
, 0, start
, blksize
, 0);
2943 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
,
2944 start
+ blksize
- 1, &cached_state
, GFP_NOFS
);
2945 if (!em
|| IS_ERR(em
))
2948 if (em
->block_start
> EXTENT_MAP_LAST_BYTE
)
2951 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2953 free_extent_map(em
);
2958 * helper function for fiemap, which doesn't want to see any holes.
2959 * This maps until we find something past 'last'
2961 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
2964 get_extent_t
*get_extent
)
2966 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
2967 struct extent_map
*em
;
2974 len
= last
- offset
;
2977 len
= (len
+ sectorsize
- 1) & ~(sectorsize
- 1);
2978 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
2979 if (!em
|| IS_ERR(em
))
2982 /* if this isn't a hole return it */
2983 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
2984 em
->block_start
!= EXTENT_MAP_HOLE
) {
2988 /* this is a hole, advance to the next extent */
2989 offset
= extent_map_end(em
);
2990 free_extent_map(em
);
2997 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
2998 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
3002 u64 max
= start
+ len
;
3006 u64 last_for_get_extent
= 0;
3008 u64 isize
= i_size_read(inode
);
3009 struct btrfs_key found_key
;
3010 struct extent_map
*em
= NULL
;
3011 struct extent_state
*cached_state
= NULL
;
3012 struct btrfs_path
*path
;
3013 struct btrfs_file_extent_item
*item
;
3018 unsigned long emflags
;
3023 path
= btrfs_alloc_path();
3026 path
->leave_spinning
= 1;
3029 * lookup the last file extent. We're not using i_size here
3030 * because there might be preallocation past i_size
3032 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
3033 path
, inode
->i_ino
, -1, 0);
3035 btrfs_free_path(path
);
3040 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3041 struct btrfs_file_extent_item
);
3042 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
3043 found_type
= btrfs_key_type(&found_key
);
3045 /* No extents, but there might be delalloc bits */
3046 if (found_key
.objectid
!= inode
->i_ino
||
3047 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3048 /* have to trust i_size as the end */
3050 last_for_get_extent
= isize
;
3053 * remember the start of the last extent. There are a
3054 * bunch of different factors that go into the length of the
3055 * extent, so its much less complex to remember where it started
3057 last
= found_key
.offset
;
3058 last_for_get_extent
= last
+ 1;
3060 btrfs_free_path(path
);
3063 * we might have some extents allocated but more delalloc past those
3064 * extents. so, we trust isize unless the start of the last extent is
3069 last_for_get_extent
= isize
;
3072 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
, 0,
3073 &cached_state
, GFP_NOFS
);
3075 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
3085 u64 offset_in_extent
;
3087 /* break if the extent we found is outside the range */
3088 if (em
->start
>= max
|| extent_map_end(em
) < off
)
3092 * get_extent may return an extent that starts before our
3093 * requested range. We have to make sure the ranges
3094 * we return to fiemap always move forward and don't
3095 * overlap, so adjust the offsets here
3097 em_start
= max(em
->start
, off
);
3100 * record the offset from the start of the extent
3101 * for adjusting the disk offset below
3103 offset_in_extent
= em_start
- em
->start
;
3104 em_end
= extent_map_end(em
);
3105 em_len
= em_end
- em_start
;
3106 emflags
= em
->flags
;
3111 * bump off for our next call to get_extent
3113 off
= extent_map_end(em
);
3117 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
3119 flags
|= FIEMAP_EXTENT_LAST
;
3120 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
3121 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
3122 FIEMAP_EXTENT_NOT_ALIGNED
);
3123 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
3124 flags
|= (FIEMAP_EXTENT_DELALLOC
|
3125 FIEMAP_EXTENT_UNKNOWN
);
3127 disko
= em
->block_start
+ offset_in_extent
;
3129 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
3130 flags
|= FIEMAP_EXTENT_ENCODED
;
3132 free_extent_map(em
);
3134 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
3135 (last
== (u64
)-1 && isize
<= em_end
)) {
3136 flags
|= FIEMAP_EXTENT_LAST
;
3140 /* now scan forward to see if this is really the last extent. */
3141 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
3148 flags
|= FIEMAP_EXTENT_LAST
;
3151 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
3157 free_extent_map(em
);
3159 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
,
3160 &cached_state
, GFP_NOFS
);
3164 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
3168 struct address_space
*mapping
;
3171 return eb
->first_page
;
3172 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
3173 mapping
= eb
->first_page
->mapping
;
3178 * extent_buffer_page is only called after pinning the page
3179 * by increasing the reference count. So we know the page must
3180 * be in the radix tree.
3183 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
3189 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
3191 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
3192 (start
>> PAGE_CACHE_SHIFT
);
3195 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
3200 struct extent_buffer
*eb
= NULL
;
3202 unsigned long flags
;
3205 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
3210 spin_lock_init(&eb
->lock
);
3211 init_waitqueue_head(&eb
->lock_wq
);
3214 spin_lock_irqsave(&leak_lock
, flags
);
3215 list_add(&eb
->leak_list
, &buffers
);
3216 spin_unlock_irqrestore(&leak_lock
, flags
);
3218 atomic_set(&eb
->refs
, 1);
3223 static void __free_extent_buffer(struct extent_buffer
*eb
)
3226 unsigned long flags
;
3227 spin_lock_irqsave(&leak_lock
, flags
);
3228 list_del(&eb
->leak_list
);
3229 spin_unlock_irqrestore(&leak_lock
, flags
);
3231 kmem_cache_free(extent_buffer_cache
, eb
);
3235 * Helper for releasing extent buffer page.
3237 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
3238 unsigned long start_idx
)
3240 unsigned long index
;
3243 if (!eb
->first_page
)
3246 index
= num_extent_pages(eb
->start
, eb
->len
);
3247 if (start_idx
>= index
)
3252 page
= extent_buffer_page(eb
, index
);
3254 page_cache_release(page
);
3255 } while (index
!= start_idx
);
3259 * Helper for releasing the extent buffer.
3261 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
3263 btrfs_release_extent_buffer_page(eb
, 0);
3264 __free_extent_buffer(eb
);
3267 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
3268 u64 start
, unsigned long len
,
3272 unsigned long num_pages
= num_extent_pages(start
, len
);
3274 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
3275 struct extent_buffer
*eb
;
3276 struct extent_buffer
*exists
= NULL
;
3278 struct address_space
*mapping
= tree
->mapping
;
3283 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
3284 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
3286 mark_page_accessed(eb
->first_page
);
3291 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
3296 eb
->first_page
= page0
;
3299 page_cache_get(page0
);
3300 mark_page_accessed(page0
);
3301 set_page_extent_mapped(page0
);
3302 set_page_extent_head(page0
, len
);
3303 uptodate
= PageUptodate(page0
);
3307 for (; i
< num_pages
; i
++, index
++) {
3308 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
3313 set_page_extent_mapped(p
);
3314 mark_page_accessed(p
);
3317 set_page_extent_head(p
, len
);
3319 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
3321 if (!PageUptodate(p
))
3325 * see below about how we avoid a nasty race with release page
3326 * and why we unlock later
3332 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3334 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
3338 spin_lock(&tree
->buffer_lock
);
3339 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
3340 if (ret
== -EEXIST
) {
3341 exists
= radix_tree_lookup(&tree
->buffer
,
3342 start
>> PAGE_CACHE_SHIFT
);
3343 /* add one reference for the caller */
3344 atomic_inc(&exists
->refs
);
3345 spin_unlock(&tree
->buffer_lock
);
3346 radix_tree_preload_end();
3349 /* add one reference for the tree */
3350 atomic_inc(&eb
->refs
);
3351 spin_unlock(&tree
->buffer_lock
);
3352 radix_tree_preload_end();
3355 * there is a race where release page may have
3356 * tried to find this extent buffer in the radix
3357 * but failed. It will tell the VM it is safe to
3358 * reclaim the, and it will clear the page private bit.
3359 * We must make sure to set the page private bit properly
3360 * after the extent buffer is in the radix tree so
3361 * it doesn't get lost
3363 set_page_extent_mapped(eb
->first_page
);
3364 set_page_extent_head(eb
->first_page
, eb
->len
);
3366 unlock_page(eb
->first_page
);
3370 if (eb
->first_page
&& !page0
)
3371 unlock_page(eb
->first_page
);
3373 if (!atomic_dec_and_test(&eb
->refs
))
3375 btrfs_release_extent_buffer(eb
);
3379 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
3380 u64 start
, unsigned long len
,
3383 struct extent_buffer
*eb
;
3386 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
3387 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
3389 mark_page_accessed(eb
->first_page
);
3397 void free_extent_buffer(struct extent_buffer
*eb
)
3402 if (!atomic_dec_and_test(&eb
->refs
))
3408 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
3409 struct extent_buffer
*eb
)
3412 unsigned long num_pages
;
3415 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3417 for (i
= 0; i
< num_pages
; i
++) {
3418 page
= extent_buffer_page(eb
, i
);
3419 if (!PageDirty(page
))
3423 WARN_ON(!PagePrivate(page
));
3425 set_page_extent_mapped(page
);
3427 set_page_extent_head(page
, eb
->len
);
3429 clear_page_dirty_for_io(page
);
3430 spin_lock_irq(&page
->mapping
->tree_lock
);
3431 if (!PageDirty(page
)) {
3432 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3434 PAGECACHE_TAG_DIRTY
);
3436 spin_unlock_irq(&page
->mapping
->tree_lock
);
3442 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
3443 struct extent_buffer
*eb
)
3445 return wait_on_extent_writeback(tree
, eb
->start
,
3446 eb
->start
+ eb
->len
- 1);
3449 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
3450 struct extent_buffer
*eb
)
3453 unsigned long num_pages
;
3456 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
3457 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3458 for (i
= 0; i
< num_pages
; i
++)
3459 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
3463 int clear_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3464 struct extent_buffer
*eb
,
3465 struct extent_state
**cached_state
)
3469 unsigned long num_pages
;
3471 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3472 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3474 clear_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3475 cached_state
, GFP_NOFS
);
3476 for (i
= 0; i
< num_pages
; i
++) {
3477 page
= extent_buffer_page(eb
, i
);
3479 ClearPageUptodate(page
);
3484 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3485 struct extent_buffer
*eb
)
3489 unsigned long num_pages
;
3491 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3493 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3495 for (i
= 0; i
< num_pages
; i
++) {
3496 page
= extent_buffer_page(eb
, i
);
3497 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
3498 ((i
== num_pages
- 1) &&
3499 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
3500 check_page_uptodate(tree
, page
);
3503 SetPageUptodate(page
);
3508 int extent_range_uptodate(struct extent_io_tree
*tree
,
3513 int pg_uptodate
= 1;
3515 unsigned long index
;
3517 ret
= test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
);
3520 while (start
<= end
) {
3521 index
= start
>> PAGE_CACHE_SHIFT
;
3522 page
= find_get_page(tree
->mapping
, index
);
3523 uptodate
= PageUptodate(page
);
3524 page_cache_release(page
);
3529 start
+= PAGE_CACHE_SIZE
;
3534 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
3535 struct extent_buffer
*eb
,
3536 struct extent_state
*cached_state
)
3539 unsigned long num_pages
;
3542 int pg_uptodate
= 1;
3544 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
3547 ret
= test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3548 EXTENT_UPTODATE
, 1, cached_state
);
3552 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3553 for (i
= 0; i
< num_pages
; i
++) {
3554 page
= extent_buffer_page(eb
, i
);
3555 if (!PageUptodate(page
)) {
3563 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
3564 struct extent_buffer
*eb
,
3565 u64 start
, int wait
,
3566 get_extent_t
*get_extent
, int mirror_num
)
3569 unsigned long start_i
;
3573 int locked_pages
= 0;
3574 int all_uptodate
= 1;
3575 int inc_all_pages
= 0;
3576 unsigned long num_pages
;
3577 struct bio
*bio
= NULL
;
3578 unsigned long bio_flags
= 0;
3580 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
3583 if (test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3584 EXTENT_UPTODATE
, 1, NULL
)) {
3589 WARN_ON(start
< eb
->start
);
3590 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
3591 (eb
->start
>> PAGE_CACHE_SHIFT
);
3596 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3597 for (i
= start_i
; i
< num_pages
; i
++) {
3598 page
= extent_buffer_page(eb
, i
);
3600 if (!trylock_page(page
))
3606 if (!PageUptodate(page
))
3611 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3615 for (i
= start_i
; i
< num_pages
; i
++) {
3616 page
= extent_buffer_page(eb
, i
);
3618 WARN_ON(!PagePrivate(page
));
3620 set_page_extent_mapped(page
);
3622 set_page_extent_head(page
, eb
->len
);
3625 page_cache_get(page
);
3626 if (!PageUptodate(page
)) {
3629 ClearPageError(page
);
3630 err
= __extent_read_full_page(tree
, page
,
3632 mirror_num
, &bio_flags
);
3641 submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3646 for (i
= start_i
; i
< num_pages
; i
++) {
3647 page
= extent_buffer_page(eb
, i
);
3648 wait_on_page_locked(page
);
3649 if (!PageUptodate(page
))
3654 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3659 while (locked_pages
> 0) {
3660 page
= extent_buffer_page(eb
, i
);
3668 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
3669 unsigned long start
,
3676 char *dst
= (char *)dstv
;
3677 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3678 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3680 WARN_ON(start
> eb
->len
);
3681 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3683 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3686 page
= extent_buffer_page(eb
, i
);
3688 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3689 kaddr
= kmap_atomic(page
, KM_USER1
);
3690 memcpy(dst
, kaddr
+ offset
, cur
);
3691 kunmap_atomic(kaddr
, KM_USER1
);
3700 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3701 unsigned long min_len
, char **token
, char **map
,
3702 unsigned long *map_start
,
3703 unsigned long *map_len
, int km
)
3705 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
3708 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3709 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3710 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
3717 offset
= start_offset
;
3721 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
3724 if (start
+ min_len
> eb
->len
) {
3725 printk(KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
3726 "wanted %lu %lu\n", (unsigned long long)eb
->start
,
3727 eb
->len
, start
, min_len
);
3732 p
= extent_buffer_page(eb
, i
);
3733 kaddr
= kmap_atomic(p
, km
);
3735 *map
= kaddr
+ offset
;
3736 *map_len
= PAGE_CACHE_SIZE
- offset
;
3740 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3741 unsigned long min_len
,
3742 char **token
, char **map
,
3743 unsigned long *map_start
,
3744 unsigned long *map_len
, int km
)
3748 if (eb
->map_token
) {
3749 unmap_extent_buffer(eb
, eb
->map_token
, km
);
3750 eb
->map_token
= NULL
;
3753 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
3754 map_start
, map_len
, km
);
3756 eb
->map_token
= *token
;
3758 eb
->map_start
= *map_start
;
3759 eb
->map_len
= *map_len
;
3764 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
3766 kunmap_atomic(token
, km
);
3769 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3770 unsigned long start
,
3777 char *ptr
= (char *)ptrv
;
3778 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3779 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3782 WARN_ON(start
> eb
->len
);
3783 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3785 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3788 page
= extent_buffer_page(eb
, i
);
3790 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3792 kaddr
= kmap_atomic(page
, KM_USER0
);
3793 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3794 kunmap_atomic(kaddr
, KM_USER0
);
3806 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3807 unsigned long start
, unsigned long len
)
3813 char *src
= (char *)srcv
;
3814 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3815 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3817 WARN_ON(start
> eb
->len
);
3818 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3820 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3823 page
= extent_buffer_page(eb
, i
);
3824 WARN_ON(!PageUptodate(page
));
3826 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3827 kaddr
= kmap_atomic(page
, KM_USER1
);
3828 memcpy(kaddr
+ offset
, src
, cur
);
3829 kunmap_atomic(kaddr
, KM_USER1
);
3838 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3839 unsigned long start
, unsigned long len
)
3845 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3846 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3848 WARN_ON(start
> eb
->len
);
3849 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3851 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3854 page
= extent_buffer_page(eb
, i
);
3855 WARN_ON(!PageUptodate(page
));
3857 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3858 kaddr
= kmap_atomic(page
, KM_USER0
);
3859 memset(kaddr
+ offset
, c
, cur
);
3860 kunmap_atomic(kaddr
, KM_USER0
);
3868 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3869 unsigned long dst_offset
, unsigned long src_offset
,
3872 u64 dst_len
= dst
->len
;
3877 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3878 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3880 WARN_ON(src
->len
!= dst_len
);
3882 offset
= (start_offset
+ dst_offset
) &
3883 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3886 page
= extent_buffer_page(dst
, i
);
3887 WARN_ON(!PageUptodate(page
));
3889 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3891 kaddr
= kmap_atomic(page
, KM_USER0
);
3892 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3893 kunmap_atomic(kaddr
, KM_USER0
);
3902 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3903 unsigned long dst_off
, unsigned long src_off
,
3906 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3907 if (dst_page
== src_page
) {
3908 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3910 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3911 char *p
= dst_kaddr
+ dst_off
+ len
;
3912 char *s
= src_kaddr
+ src_off
+ len
;
3917 kunmap_atomic(src_kaddr
, KM_USER1
);
3919 kunmap_atomic(dst_kaddr
, KM_USER0
);
3922 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
3924 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
3925 return distance
< len
;
3928 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3929 unsigned long dst_off
, unsigned long src_off
,
3932 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3935 if (dst_page
!= src_page
) {
3936 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3938 src_kaddr
= dst_kaddr
;
3939 BUG_ON(areas_overlap(src_off
, dst_off
, len
));
3942 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3943 kunmap_atomic(dst_kaddr
, KM_USER0
);
3944 if (dst_page
!= src_page
)
3945 kunmap_atomic(src_kaddr
, KM_USER1
);
3948 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3949 unsigned long src_offset
, unsigned long len
)
3952 size_t dst_off_in_page
;
3953 size_t src_off_in_page
;
3954 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3955 unsigned long dst_i
;
3956 unsigned long src_i
;
3958 if (src_offset
+ len
> dst
->len
) {
3959 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
3960 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
3963 if (dst_offset
+ len
> dst
->len
) {
3964 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
3965 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
3970 dst_off_in_page
= (start_offset
+ dst_offset
) &
3971 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3972 src_off_in_page
= (start_offset
+ src_offset
) &
3973 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3975 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3976 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3978 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3980 cur
= min_t(unsigned long, cur
,
3981 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3983 copy_pages(extent_buffer_page(dst
, dst_i
),
3984 extent_buffer_page(dst
, src_i
),
3985 dst_off_in_page
, src_off_in_page
, cur
);
3993 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3994 unsigned long src_offset
, unsigned long len
)
3997 size_t dst_off_in_page
;
3998 size_t src_off_in_page
;
3999 unsigned long dst_end
= dst_offset
+ len
- 1;
4000 unsigned long src_end
= src_offset
+ len
- 1;
4001 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4002 unsigned long dst_i
;
4003 unsigned long src_i
;
4005 if (src_offset
+ len
> dst
->len
) {
4006 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
4007 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
4010 if (dst_offset
+ len
> dst
->len
) {
4011 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
4012 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
4015 if (!areas_overlap(src_offset
, dst_offset
, len
)) {
4016 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
4020 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
4021 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
4023 dst_off_in_page
= (start_offset
+ dst_end
) &
4024 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4025 src_off_in_page
= (start_offset
+ src_end
) &
4026 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4028 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
4029 cur
= min(cur
, dst_off_in_page
+ 1);
4030 move_pages(extent_buffer_page(dst
, dst_i
),
4031 extent_buffer_page(dst
, src_i
),
4032 dst_off_in_page
- cur
+ 1,
4033 src_off_in_page
- cur
+ 1, cur
);
4041 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4043 struct extent_buffer
*eb
=
4044 container_of(head
, struct extent_buffer
, rcu_head
);
4046 btrfs_release_extent_buffer(eb
);
4049 int try_release_extent_buffer(struct extent_io_tree
*tree
, struct page
*page
)
4051 u64 start
= page_offset(page
);
4052 struct extent_buffer
*eb
;
4055 spin_lock(&tree
->buffer_lock
);
4056 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4058 spin_unlock(&tree
->buffer_lock
);
4062 if (test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
4068 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4071 if (atomic_cmpxchg(&eb
->refs
, 1, 0) != 1) {
4076 radix_tree_delete(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4078 spin_unlock(&tree
->buffer_lock
);
4080 /* at this point we can safely release the extent buffer */
4081 if (atomic_read(&eb
->refs
) == 0)
4082 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);