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
, gfp_t mask
)
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 int merge_state(struct extent_io_tree
*tree
,
258 struct extent_state
*state
)
260 struct extent_state
*other
;
261 struct rb_node
*other_node
;
263 if (state
->state
& (EXTENT_IOBITS
| 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 other
->start
= state
->start
;
286 rb_erase(&state
->rb_node
, &tree
->state
);
287 free_extent_state(state
);
295 static int set_state_cb(struct extent_io_tree
*tree
,
296 struct extent_state
*state
, int *bits
)
298 if (tree
->ops
&& tree
->ops
->set_bit_hook
) {
299 return tree
->ops
->set_bit_hook(tree
->mapping
->host
,
306 static void clear_state_cb(struct extent_io_tree
*tree
,
307 struct extent_state
*state
, int *bits
)
309 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
310 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
314 * insert an extent_state struct into the tree. 'bits' are set on the
315 * struct before it is inserted.
317 * This may return -EEXIST if the extent is already there, in which case the
318 * state struct is freed.
320 * The tree lock is not taken internally. This is a utility function and
321 * probably isn't what you want to call (see set/clear_extent_bit).
323 static int insert_state(struct extent_io_tree
*tree
,
324 struct extent_state
*state
, u64 start
, u64 end
,
327 struct rb_node
*node
;
328 int bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
332 printk(KERN_ERR
"btrfs end < start %llu %llu\n",
333 (unsigned long long)end
,
334 (unsigned long long)start
);
337 state
->start
= start
;
339 ret
= set_state_cb(tree
, state
, bits
);
343 if (bits_to_set
& EXTENT_DIRTY
)
344 tree
->dirty_bytes
+= end
- start
+ 1;
345 state
->state
|= bits_to_set
;
346 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
348 struct extent_state
*found
;
349 found
= rb_entry(node
, struct extent_state
, rb_node
);
350 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
351 "%llu %llu\n", (unsigned long long)found
->start
,
352 (unsigned long long)found
->end
,
353 (unsigned long long)start
, (unsigned long long)end
);
354 free_extent_state(state
);
358 merge_state(tree
, state
);
362 static int split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
365 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
366 return tree
->ops
->split_extent_hook(tree
->mapping
->host
,
372 * split a given extent state struct in two, inserting the preallocated
373 * struct 'prealloc' as the newly created second half. 'split' indicates an
374 * offset inside 'orig' where it should be split.
377 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
378 * are two extent state structs in the tree:
379 * prealloc: [orig->start, split - 1]
380 * orig: [ split, orig->end ]
382 * The tree locks are not taken by this function. They need to be held
385 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
386 struct extent_state
*prealloc
, u64 split
)
388 struct rb_node
*node
;
390 split_cb(tree
, orig
, split
);
392 prealloc
->start
= orig
->start
;
393 prealloc
->end
= split
- 1;
394 prealloc
->state
= orig
->state
;
397 node
= tree_insert(&tree
->state
, prealloc
->end
, &prealloc
->rb_node
);
399 free_extent_state(prealloc
);
402 prealloc
->tree
= tree
;
407 * utility function to clear some bits in an extent state struct.
408 * it will optionally wake up any one waiting on this state (wake == 1), or
409 * forcibly remove the state from the tree (delete == 1).
411 * If no bits are set on the state struct after clearing things, the
412 * struct is freed and removed from the tree
414 static int clear_state_bit(struct extent_io_tree
*tree
,
415 struct extent_state
*state
,
418 int bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
419 int ret
= state
->state
& bits_to_clear
;
421 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
422 u64 range
= state
->end
- state
->start
+ 1;
423 WARN_ON(range
> tree
->dirty_bytes
);
424 tree
->dirty_bytes
-= range
;
426 clear_state_cb(tree
, state
, bits
);
427 state
->state
&= ~bits_to_clear
;
430 if (state
->state
== 0) {
432 rb_erase(&state
->rb_node
, &tree
->state
);
434 free_extent_state(state
);
439 merge_state(tree
, state
);
445 * clear some bits on a range in the tree. This may require splitting
446 * or inserting elements in the tree, so the gfp mask is used to
447 * indicate which allocations or sleeping are allowed.
449 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
450 * the given range from the tree regardless of state (ie for truncate).
452 * the range [start, end] is inclusive.
454 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
455 * bits were already set, or zero if none of the bits were already set.
457 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
458 int bits
, int wake
, int delete,
459 struct extent_state
**cached_state
,
462 struct extent_state
*state
;
463 struct extent_state
*cached
;
464 struct extent_state
*prealloc
= NULL
;
465 struct rb_node
*next_node
;
466 struct rb_node
*node
;
473 bits
|= ~EXTENT_CTLBITS
;
474 bits
|= EXTENT_FIRST_DELALLOC
;
476 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
479 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
480 prealloc
= alloc_extent_state(mask
);
485 spin_lock(&tree
->lock
);
487 cached
= *cached_state
;
490 *cached_state
= NULL
;
494 if (cached
&& cached
->tree
&& cached
->start
== start
) {
496 atomic_dec(&cached
->refs
);
501 free_extent_state(cached
);
504 * this search will find the extents that end after
507 node
= tree_search(tree
, start
);
510 state
= rb_entry(node
, struct extent_state
, rb_node
);
512 if (state
->start
> end
)
514 WARN_ON(state
->end
< start
);
515 last_end
= state
->end
;
518 * | ---- desired range ---- |
520 * | ------------- state -------------- |
522 * We need to split the extent we found, and may flip
523 * bits on second half.
525 * If the extent we found extends past our range, we
526 * just split and search again. It'll get split again
527 * the next time though.
529 * If the extent we found is inside our range, we clear
530 * the desired bit on it.
533 if (state
->start
< start
) {
535 prealloc
= alloc_extent_state(GFP_ATOMIC
);
536 err
= split_state(tree
, state
, prealloc
, start
);
537 BUG_ON(err
== -EEXIST
);
541 if (state
->end
<= end
) {
542 set
|= clear_state_bit(tree
, state
, &bits
, wake
);
543 if (last_end
== (u64
)-1)
545 start
= last_end
+ 1;
550 * | ---- desired range ---- |
552 * We need to split the extent, and clear the bit
555 if (state
->start
<= end
&& state
->end
> end
) {
557 prealloc
= alloc_extent_state(GFP_ATOMIC
);
558 err
= split_state(tree
, state
, prealloc
, end
+ 1);
559 BUG_ON(err
== -EEXIST
);
563 set
|= clear_state_bit(tree
, prealloc
, &bits
, wake
);
569 if (state
->end
< end
&& prealloc
&& !need_resched())
570 next_node
= rb_next(&state
->rb_node
);
574 set
|= clear_state_bit(tree
, state
, &bits
, wake
);
575 if (last_end
== (u64
)-1)
577 start
= last_end
+ 1;
578 if (start
<= end
&& next_node
) {
579 state
= rb_entry(next_node
, struct extent_state
,
581 if (state
->start
== start
)
587 spin_unlock(&tree
->lock
);
589 free_extent_state(prealloc
);
596 spin_unlock(&tree
->lock
);
597 if (mask
& __GFP_WAIT
)
602 static int wait_on_state(struct extent_io_tree
*tree
,
603 struct extent_state
*state
)
604 __releases(tree
->lock
)
605 __acquires(tree
->lock
)
608 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
609 spin_unlock(&tree
->lock
);
611 spin_lock(&tree
->lock
);
612 finish_wait(&state
->wq
, &wait
);
617 * waits for one or more bits to clear on a range in the state tree.
618 * The range [start, end] is inclusive.
619 * The tree lock is taken by this function
621 int wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
623 struct extent_state
*state
;
624 struct rb_node
*node
;
626 spin_lock(&tree
->lock
);
630 * this search will find all the extents that end after
633 node
= tree_search(tree
, start
);
637 state
= rb_entry(node
, struct extent_state
, rb_node
);
639 if (state
->start
> end
)
642 if (state
->state
& bits
) {
643 start
= state
->start
;
644 atomic_inc(&state
->refs
);
645 wait_on_state(tree
, state
);
646 free_extent_state(state
);
649 start
= state
->end
+ 1;
654 if (need_resched()) {
655 spin_unlock(&tree
->lock
);
657 spin_lock(&tree
->lock
);
661 spin_unlock(&tree
->lock
);
665 static int set_state_bits(struct extent_io_tree
*tree
,
666 struct extent_state
*state
,
670 int bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
672 ret
= set_state_cb(tree
, state
, bits
);
675 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
676 u64 range
= state
->end
- state
->start
+ 1;
677 tree
->dirty_bytes
+= range
;
679 state
->state
|= bits_to_set
;
684 static void cache_state(struct extent_state
*state
,
685 struct extent_state
**cached_ptr
)
687 if (cached_ptr
&& !(*cached_ptr
)) {
688 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
690 atomic_inc(&state
->refs
);
695 static void uncache_state(struct extent_state
**cached_ptr
)
697 if (cached_ptr
&& (*cached_ptr
)) {
698 struct extent_state
*state
= *cached_ptr
;
700 free_extent_state(state
);
705 * set some bits on a range in the tree. This may require allocations or
706 * sleeping, so the gfp mask is used to indicate what is allowed.
708 * If any of the exclusive bits are set, this will fail with -EEXIST if some
709 * part of the range already has the desired bits set. The start of the
710 * existing range is returned in failed_start in this case.
712 * [start, end] is inclusive This takes the tree lock.
715 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
716 int bits
, int exclusive_bits
, u64
*failed_start
,
717 struct extent_state
**cached_state
, gfp_t mask
)
719 struct extent_state
*state
;
720 struct extent_state
*prealloc
= NULL
;
721 struct rb_node
*node
;
726 bits
|= EXTENT_FIRST_DELALLOC
;
728 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
729 prealloc
= alloc_extent_state(mask
);
734 spin_lock(&tree
->lock
);
735 if (cached_state
&& *cached_state
) {
736 state
= *cached_state
;
737 if (state
->start
== start
&& state
->tree
) {
738 node
= &state
->rb_node
;
743 * this search will find all the extents that end after
746 node
= tree_search(tree
, start
);
748 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
750 BUG_ON(err
== -EEXIST
);
753 state
= rb_entry(node
, struct extent_state
, rb_node
);
755 last_start
= state
->start
;
756 last_end
= state
->end
;
759 * | ---- desired range ---- |
762 * Just lock what we found and keep going
764 if (state
->start
== start
&& state
->end
<= end
) {
765 struct rb_node
*next_node
;
766 if (state
->state
& exclusive_bits
) {
767 *failed_start
= state
->start
;
772 err
= set_state_bits(tree
, state
, &bits
);
776 cache_state(state
, cached_state
);
777 merge_state(tree
, state
);
778 if (last_end
== (u64
)-1)
781 start
= last_end
+ 1;
782 if (start
< end
&& prealloc
&& !need_resched()) {
783 next_node
= rb_next(node
);
785 state
= rb_entry(next_node
, struct extent_state
,
787 if (state
->start
== start
)
795 * | ---- desired range ---- |
798 * | ------------- state -------------- |
800 * We need to split the extent we found, and may flip bits on
803 * If the extent we found extends past our
804 * range, we just split and search again. It'll get split
805 * again the next time though.
807 * If the extent we found is inside our range, we set the
810 if (state
->start
< start
) {
811 if (state
->state
& exclusive_bits
) {
812 *failed_start
= start
;
816 err
= split_state(tree
, state
, prealloc
, start
);
817 BUG_ON(err
== -EEXIST
);
821 if (state
->end
<= end
) {
822 err
= set_state_bits(tree
, state
, &bits
);
825 cache_state(state
, cached_state
);
826 merge_state(tree
, state
);
827 if (last_end
== (u64
)-1)
829 start
= last_end
+ 1;
834 * | ---- desired range ---- |
835 * | state | or | state |
837 * There's a hole, we need to insert something in it and
838 * ignore the extent we found.
840 if (state
->start
> start
) {
842 if (end
< last_start
)
845 this_end
= last_start
- 1;
846 err
= insert_state(tree
, prealloc
, start
, this_end
,
848 BUG_ON(err
== -EEXIST
);
853 cache_state(prealloc
, cached_state
);
855 start
= this_end
+ 1;
859 * | ---- desired range ---- |
861 * We need to split the extent, and set the bit
864 if (state
->start
<= end
&& state
->end
> end
) {
865 if (state
->state
& exclusive_bits
) {
866 *failed_start
= start
;
870 err
= split_state(tree
, state
, prealloc
, end
+ 1);
871 BUG_ON(err
== -EEXIST
);
873 err
= set_state_bits(tree
, prealloc
, &bits
);
878 cache_state(prealloc
, cached_state
);
879 merge_state(tree
, prealloc
);
887 spin_unlock(&tree
->lock
);
889 free_extent_state(prealloc
);
896 spin_unlock(&tree
->lock
);
897 if (mask
& __GFP_WAIT
)
902 /* wrappers around set/clear extent bit */
903 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
906 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, 0, NULL
,
910 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
911 int bits
, gfp_t mask
)
913 return set_extent_bit(tree
, start
, end
, bits
, 0, NULL
,
917 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
918 int bits
, gfp_t mask
)
920 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
923 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
924 struct extent_state
**cached_state
, gfp_t mask
)
926 return set_extent_bit(tree
, start
, end
,
927 EXTENT_DELALLOC
| EXTENT_DIRTY
| EXTENT_UPTODATE
,
928 0, NULL
, cached_state
, mask
);
931 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
934 return clear_extent_bit(tree
, start
, end
,
935 EXTENT_DIRTY
| EXTENT_DELALLOC
|
936 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
939 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
942 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, NULL
,
946 static int clear_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
949 return clear_extent_bit(tree
, start
, end
, EXTENT_NEW
, 0, 0,
953 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
954 struct extent_state
**cached_state
, gfp_t mask
)
956 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0,
957 NULL
, cached_state
, mask
);
960 static int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
,
961 u64 end
, struct extent_state
**cached_state
,
964 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
968 int wait_on_extent_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
970 return wait_extent_bit(tree
, start
, end
, EXTENT_WRITEBACK
);
974 * either insert or lock state struct between start and end use mask to tell
975 * us if waiting is desired.
977 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
978 int bits
, struct extent_state
**cached_state
, gfp_t mask
)
983 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
984 EXTENT_LOCKED
, &failed_start
,
986 if (err
== -EEXIST
&& (mask
& __GFP_WAIT
)) {
987 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
988 start
= failed_start
;
992 WARN_ON(start
> end
);
997 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
999 return lock_extent_bits(tree
, start
, end
, 0, NULL
, mask
);
1002 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1008 err
= set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1009 &failed_start
, NULL
, mask
);
1010 if (err
== -EEXIST
) {
1011 if (failed_start
> start
)
1012 clear_extent_bit(tree
, start
, failed_start
- 1,
1013 EXTENT_LOCKED
, 1, 0, NULL
, mask
);
1019 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1020 struct extent_state
**cached
, gfp_t mask
)
1022 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1026 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
, gfp_t mask
)
1028 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1033 * helper function to set pages and extents in the tree dirty
1035 int set_range_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1037 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1038 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1041 while (index
<= end_index
) {
1042 page
= find_get_page(tree
->mapping
, index
);
1044 __set_page_dirty_nobuffers(page
);
1045 page_cache_release(page
);
1052 * helper function to set both pages and extents in the tree writeback
1054 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1056 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1057 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1060 while (index
<= end_index
) {
1061 page
= find_get_page(tree
->mapping
, index
);
1063 set_page_writeback(page
);
1064 page_cache_release(page
);
1071 * find the first offset in the io tree with 'bits' set. zero is
1072 * returned if we find something, and *start_ret and *end_ret are
1073 * set to reflect the state struct that was found.
1075 * If nothing was found, 1 is returned, < 0 on error
1077 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1078 u64
*start_ret
, u64
*end_ret
, int bits
)
1080 struct rb_node
*node
;
1081 struct extent_state
*state
;
1084 spin_lock(&tree
->lock
);
1086 * this search will find all the extents that end after
1089 node
= tree_search(tree
, start
);
1094 state
= rb_entry(node
, struct extent_state
, rb_node
);
1095 if (state
->end
>= start
&& (state
->state
& bits
)) {
1096 *start_ret
= state
->start
;
1097 *end_ret
= state
->end
;
1101 node
= rb_next(node
);
1106 spin_unlock(&tree
->lock
);
1110 /* find the first state struct with 'bits' set after 'start', and
1111 * return it. tree->lock must be held. NULL will returned if
1112 * nothing was found after 'start'
1114 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
1115 u64 start
, int bits
)
1117 struct rb_node
*node
;
1118 struct extent_state
*state
;
1121 * this search will find all the extents that end after
1124 node
= tree_search(tree
, start
);
1129 state
= rb_entry(node
, struct extent_state
, rb_node
);
1130 if (state
->end
>= start
&& (state
->state
& bits
))
1133 node
= rb_next(node
);
1142 * find a contiguous range of bytes in the file marked as delalloc, not
1143 * more than 'max_bytes'. start and end are used to return the range,
1145 * 1 is returned if we find something, 0 if nothing was in the tree
1147 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1148 u64
*start
, u64
*end
, u64 max_bytes
,
1149 struct extent_state
**cached_state
)
1151 struct rb_node
*node
;
1152 struct extent_state
*state
;
1153 u64 cur_start
= *start
;
1155 u64 total_bytes
= 0;
1157 spin_lock(&tree
->lock
);
1160 * this search will find all the extents that end after
1163 node
= tree_search(tree
, cur_start
);
1171 state
= rb_entry(node
, struct extent_state
, rb_node
);
1172 if (found
&& (state
->start
!= cur_start
||
1173 (state
->state
& EXTENT_BOUNDARY
))) {
1176 if (!(state
->state
& EXTENT_DELALLOC
)) {
1182 *start
= state
->start
;
1183 *cached_state
= state
;
1184 atomic_inc(&state
->refs
);
1188 cur_start
= state
->end
+ 1;
1189 node
= rb_next(node
);
1192 total_bytes
+= state
->end
- state
->start
+ 1;
1193 if (total_bytes
>= max_bytes
)
1197 spin_unlock(&tree
->lock
);
1201 static noinline
int __unlock_for_delalloc(struct inode
*inode
,
1202 struct page
*locked_page
,
1206 struct page
*pages
[16];
1207 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1208 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1209 unsigned long nr_pages
= end_index
- index
+ 1;
1212 if (index
== locked_page
->index
&& end_index
== index
)
1215 while (nr_pages
> 0) {
1216 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1217 min_t(unsigned long, nr_pages
,
1218 ARRAY_SIZE(pages
)), pages
);
1219 for (i
= 0; i
< ret
; i
++) {
1220 if (pages
[i
] != locked_page
)
1221 unlock_page(pages
[i
]);
1222 page_cache_release(pages
[i
]);
1231 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1232 struct page
*locked_page
,
1236 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1237 unsigned long start_index
= index
;
1238 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1239 unsigned long pages_locked
= 0;
1240 struct page
*pages
[16];
1241 unsigned long nrpages
;
1245 /* the caller is responsible for locking the start index */
1246 if (index
== locked_page
->index
&& index
== end_index
)
1249 /* skip the page at the start index */
1250 nrpages
= end_index
- index
+ 1;
1251 while (nrpages
> 0) {
1252 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1253 min_t(unsigned long,
1254 nrpages
, ARRAY_SIZE(pages
)), pages
);
1259 /* now we have an array of pages, lock them all */
1260 for (i
= 0; i
< ret
; i
++) {
1262 * the caller is taking responsibility for
1265 if (pages
[i
] != locked_page
) {
1266 lock_page(pages
[i
]);
1267 if (!PageDirty(pages
[i
]) ||
1268 pages
[i
]->mapping
!= inode
->i_mapping
) {
1270 unlock_page(pages
[i
]);
1271 page_cache_release(pages
[i
]);
1275 page_cache_release(pages
[i
]);
1284 if (ret
&& pages_locked
) {
1285 __unlock_for_delalloc(inode
, locked_page
,
1287 ((u64
)(start_index
+ pages_locked
- 1)) <<
1294 * find a contiguous range of bytes in the file marked as delalloc, not
1295 * more than 'max_bytes'. start and end are used to return the range,
1297 * 1 is returned if we find something, 0 if nothing was in the tree
1299 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1300 struct extent_io_tree
*tree
,
1301 struct page
*locked_page
,
1302 u64
*start
, u64
*end
,
1308 struct extent_state
*cached_state
= NULL
;
1313 /* step one, find a bunch of delalloc bytes starting at start */
1314 delalloc_start
= *start
;
1316 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1317 max_bytes
, &cached_state
);
1318 if (!found
|| delalloc_end
<= *start
) {
1319 *start
= delalloc_start
;
1320 *end
= delalloc_end
;
1321 free_extent_state(cached_state
);
1326 * start comes from the offset of locked_page. We have to lock
1327 * pages in order, so we can't process delalloc bytes before
1330 if (delalloc_start
< *start
)
1331 delalloc_start
= *start
;
1334 * make sure to limit the number of pages we try to lock down
1337 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
)
1338 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1340 /* step two, lock all the pages after the page that has start */
1341 ret
= lock_delalloc_pages(inode
, locked_page
,
1342 delalloc_start
, delalloc_end
);
1343 if (ret
== -EAGAIN
) {
1344 /* some of the pages are gone, lets avoid looping by
1345 * shortening the size of the delalloc range we're searching
1347 free_extent_state(cached_state
);
1349 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1350 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1360 /* step three, lock the state bits for the whole range */
1361 lock_extent_bits(tree
, delalloc_start
, delalloc_end
,
1362 0, &cached_state
, GFP_NOFS
);
1364 /* then test to make sure it is all still delalloc */
1365 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1366 EXTENT_DELALLOC
, 1, cached_state
);
1368 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1369 &cached_state
, GFP_NOFS
);
1370 __unlock_for_delalloc(inode
, locked_page
,
1371 delalloc_start
, delalloc_end
);
1375 free_extent_state(cached_state
);
1376 *start
= delalloc_start
;
1377 *end
= delalloc_end
;
1382 int extent_clear_unlock_delalloc(struct inode
*inode
,
1383 struct extent_io_tree
*tree
,
1384 u64 start
, u64 end
, struct page
*locked_page
,
1388 struct page
*pages
[16];
1389 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1390 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1391 unsigned long nr_pages
= end_index
- index
+ 1;
1395 if (op
& EXTENT_CLEAR_UNLOCK
)
1396 clear_bits
|= EXTENT_LOCKED
;
1397 if (op
& EXTENT_CLEAR_DIRTY
)
1398 clear_bits
|= EXTENT_DIRTY
;
1400 if (op
& EXTENT_CLEAR_DELALLOC
)
1401 clear_bits
|= EXTENT_DELALLOC
;
1403 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1404 if (!(op
& (EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
1405 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
|
1406 EXTENT_SET_PRIVATE2
)))
1409 while (nr_pages
> 0) {
1410 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1411 min_t(unsigned long,
1412 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1413 for (i
= 0; i
< ret
; i
++) {
1415 if (op
& EXTENT_SET_PRIVATE2
)
1416 SetPagePrivate2(pages
[i
]);
1418 if (pages
[i
] == locked_page
) {
1419 page_cache_release(pages
[i
]);
1422 if (op
& EXTENT_CLEAR_DIRTY
)
1423 clear_page_dirty_for_io(pages
[i
]);
1424 if (op
& EXTENT_SET_WRITEBACK
)
1425 set_page_writeback(pages
[i
]);
1426 if (op
& EXTENT_END_WRITEBACK
)
1427 end_page_writeback(pages
[i
]);
1428 if (op
& EXTENT_CLEAR_UNLOCK_PAGE
)
1429 unlock_page(pages
[i
]);
1430 page_cache_release(pages
[i
]);
1440 * count the number of bytes in the tree that have a given bit(s)
1441 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1442 * cached. The total number found is returned.
1444 u64
count_range_bits(struct extent_io_tree
*tree
,
1445 u64
*start
, u64 search_end
, u64 max_bytes
,
1446 unsigned long bits
, int contig
)
1448 struct rb_node
*node
;
1449 struct extent_state
*state
;
1450 u64 cur_start
= *start
;
1451 u64 total_bytes
= 0;
1455 if (search_end
<= cur_start
) {
1460 spin_lock(&tree
->lock
);
1461 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1462 total_bytes
= tree
->dirty_bytes
;
1466 * this search will find all the extents that end after
1469 node
= tree_search(tree
, cur_start
);
1474 state
= rb_entry(node
, struct extent_state
, rb_node
);
1475 if (state
->start
> search_end
)
1477 if (contig
&& found
&& state
->start
> last
+ 1)
1479 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1480 total_bytes
+= min(search_end
, state
->end
) + 1 -
1481 max(cur_start
, state
->start
);
1482 if (total_bytes
>= max_bytes
)
1485 *start
= state
->start
;
1489 } else if (contig
&& found
) {
1492 node
= rb_next(node
);
1497 spin_unlock(&tree
->lock
);
1502 * set the private field for a given byte offset in the tree. If there isn't
1503 * an extent_state there already, this does nothing.
1505 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1507 struct rb_node
*node
;
1508 struct extent_state
*state
;
1511 spin_lock(&tree
->lock
);
1513 * this search will find all the extents that end after
1516 node
= tree_search(tree
, start
);
1521 state
= rb_entry(node
, struct extent_state
, rb_node
);
1522 if (state
->start
!= start
) {
1526 state
->private = private;
1528 spin_unlock(&tree
->lock
);
1532 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1534 struct rb_node
*node
;
1535 struct extent_state
*state
;
1538 spin_lock(&tree
->lock
);
1540 * this search will find all the extents that end after
1543 node
= tree_search(tree
, start
);
1548 state
= rb_entry(node
, struct extent_state
, rb_node
);
1549 if (state
->start
!= start
) {
1553 *private = state
->private;
1555 spin_unlock(&tree
->lock
);
1560 * searches a range in the state tree for a given mask.
1561 * If 'filled' == 1, this returns 1 only if every extent in the tree
1562 * has the bits set. Otherwise, 1 is returned if any bit in the
1563 * range is found set.
1565 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1566 int bits
, int filled
, struct extent_state
*cached
)
1568 struct extent_state
*state
= NULL
;
1569 struct rb_node
*node
;
1572 spin_lock(&tree
->lock
);
1573 if (cached
&& cached
->tree
&& cached
->start
== start
)
1574 node
= &cached
->rb_node
;
1576 node
= tree_search(tree
, start
);
1577 while (node
&& start
<= end
) {
1578 state
= rb_entry(node
, struct extent_state
, rb_node
);
1580 if (filled
&& state
->start
> start
) {
1585 if (state
->start
> end
)
1588 if (state
->state
& bits
) {
1592 } else if (filled
) {
1597 if (state
->end
== (u64
)-1)
1600 start
= state
->end
+ 1;
1603 node
= rb_next(node
);
1610 spin_unlock(&tree
->lock
);
1615 * helper function to set a given page up to date if all the
1616 * extents in the tree for that page are up to date
1618 static int check_page_uptodate(struct extent_io_tree
*tree
,
1621 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1622 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1623 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1624 SetPageUptodate(page
);
1629 * helper function to unlock a page if all the extents in the tree
1630 * for that page are unlocked
1632 static int check_page_locked(struct extent_io_tree
*tree
,
1635 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1636 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1637 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0, NULL
))
1643 * helper function to end page writeback if all the extents
1644 * in the tree for that page are done with writeback
1646 static int check_page_writeback(struct extent_io_tree
*tree
,
1649 end_page_writeback(page
);
1653 /* lots and lots of room for performance fixes in the end_bio funcs */
1656 * after a writepage IO is done, we need to:
1657 * clear the uptodate bits on error
1658 * clear the writeback bits in the extent tree for this IO
1659 * end_page_writeback if the page has no more pending IO
1661 * Scheduling is not allowed, so the extent state tree is expected
1662 * to have one and only one object corresponding to this IO.
1664 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
1666 int uptodate
= err
== 0;
1667 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1668 struct extent_io_tree
*tree
;
1675 struct page
*page
= bvec
->bv_page
;
1676 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1678 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1680 end
= start
+ bvec
->bv_len
- 1;
1682 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1687 if (--bvec
>= bio
->bi_io_vec
)
1688 prefetchw(&bvec
->bv_page
->flags
);
1689 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
1690 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
1691 end
, NULL
, uptodate
);
1696 if (!uptodate
&& tree
->ops
&&
1697 tree
->ops
->writepage_io_failed_hook
) {
1698 ret
= tree
->ops
->writepage_io_failed_hook(bio
, page
,
1701 uptodate
= (err
== 0);
1707 clear_extent_uptodate(tree
, start
, end
, NULL
, GFP_NOFS
);
1708 ClearPageUptodate(page
);
1713 end_page_writeback(page
);
1715 check_page_writeback(tree
, page
);
1716 } while (bvec
>= bio
->bi_io_vec
);
1722 * after a readpage IO is done, we need to:
1723 * clear the uptodate bits on error
1724 * set the uptodate bits if things worked
1725 * set the page up to date if all extents in the tree are uptodate
1726 * clear the lock bit in the extent tree
1727 * unlock the page if there are no other extents locked for it
1729 * Scheduling is not allowed, so the extent state tree is expected
1730 * to have one and only one object corresponding to this IO.
1732 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
1734 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1735 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1736 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1737 struct extent_io_tree
*tree
;
1747 struct page
*page
= bvec
->bv_page
;
1748 struct extent_state
*cached
= NULL
;
1749 struct extent_state
*state
;
1751 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1753 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1755 end
= start
+ bvec
->bv_len
- 1;
1757 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
1762 if (++bvec
<= bvec_end
)
1763 prefetchw(&bvec
->bv_page
->flags
);
1765 spin_lock(&tree
->lock
);
1766 state
= find_first_extent_bit_state(tree
, start
, EXTENT_LOCKED
);
1767 if (state
&& state
->start
== start
) {
1769 * take a reference on the state, unlock will drop
1772 cache_state(state
, &cached
);
1774 spin_unlock(&tree
->lock
);
1776 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
1777 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
1782 if (!uptodate
&& tree
->ops
&&
1783 tree
->ops
->readpage_io_failed_hook
) {
1784 ret
= tree
->ops
->readpage_io_failed_hook(bio
, page
,
1788 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1791 uncache_state(&cached
);
1797 set_extent_uptodate(tree
, start
, end
, &cached
,
1800 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
1804 SetPageUptodate(page
);
1806 ClearPageUptodate(page
);
1812 check_page_uptodate(tree
, page
);
1814 ClearPageUptodate(page
);
1817 check_page_locked(tree
, page
);
1819 } while (bvec
<= bvec_end
);
1825 * IO done from prepare_write is pretty simple, we just unlock
1826 * the structs in the extent tree when done, and set the uptodate bits
1829 static void end_bio_extent_preparewrite(struct bio
*bio
, int err
)
1831 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1832 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1833 struct extent_io_tree
*tree
;
1838 struct page
*page
= bvec
->bv_page
;
1839 struct extent_state
*cached
= NULL
;
1840 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1842 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
1844 end
= start
+ bvec
->bv_len
- 1;
1846 if (--bvec
>= bio
->bi_io_vec
)
1847 prefetchw(&bvec
->bv_page
->flags
);
1850 set_extent_uptodate(tree
, start
, end
, &cached
,
1853 ClearPageUptodate(page
);
1857 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
1859 } while (bvec
>= bio
->bi_io_vec
);
1865 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
1870 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1872 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
1873 while (!bio
&& (nr_vecs
/= 2))
1874 bio
= bio_alloc(gfp_flags
, nr_vecs
);
1879 bio
->bi_bdev
= bdev
;
1880 bio
->bi_sector
= first_sector
;
1885 static int submit_one_bio(int rw
, struct bio
*bio
, int mirror_num
,
1886 unsigned long bio_flags
)
1889 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1890 struct page
*page
= bvec
->bv_page
;
1891 struct extent_io_tree
*tree
= bio
->bi_private
;
1894 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
1896 bio
->bi_private
= NULL
;
1900 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
1901 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
1902 mirror_num
, bio_flags
, start
);
1904 submit_bio(rw
, bio
);
1905 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
1911 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
1912 struct page
*page
, sector_t sector
,
1913 size_t size
, unsigned long offset
,
1914 struct block_device
*bdev
,
1915 struct bio
**bio_ret
,
1916 unsigned long max_pages
,
1917 bio_end_io_t end_io_func
,
1919 unsigned long prev_bio_flags
,
1920 unsigned long bio_flags
)
1926 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
1927 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
1928 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
1930 if (bio_ret
&& *bio_ret
) {
1933 contig
= bio
->bi_sector
== sector
;
1935 contig
= bio
->bi_sector
+ (bio
->bi_size
>> 9) ==
1938 if (prev_bio_flags
!= bio_flags
|| !contig
||
1939 (tree
->ops
&& tree
->ops
->merge_bio_hook
&&
1940 tree
->ops
->merge_bio_hook(page
, offset
, page_size
, bio
,
1942 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
1943 ret
= submit_one_bio(rw
, bio
, mirror_num
,
1950 if (this_compressed
)
1953 nr
= bio_get_nr_vecs(bdev
);
1955 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
1959 bio_add_page(bio
, page
, page_size
, offset
);
1960 bio
->bi_end_io
= end_io_func
;
1961 bio
->bi_private
= tree
;
1966 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
1971 void set_page_extent_mapped(struct page
*page
)
1973 if (!PagePrivate(page
)) {
1974 SetPagePrivate(page
);
1975 page_cache_get(page
);
1976 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
1980 static void set_page_extent_head(struct page
*page
, unsigned long len
)
1982 WARN_ON(!PagePrivate(page
));
1983 set_page_private(page
, EXTENT_PAGE_PRIVATE_FIRST_PAGE
| len
<< 2);
1987 * basic readpage implementation. Locked extent state structs are inserted
1988 * into the tree that are removed when the IO is done (by the end_io
1991 static int __extent_read_full_page(struct extent_io_tree
*tree
,
1993 get_extent_t
*get_extent
,
1994 struct bio
**bio
, int mirror_num
,
1995 unsigned long *bio_flags
)
1997 struct inode
*inode
= page
->mapping
->host
;
1998 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1999 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2003 u64 last_byte
= i_size_read(inode
);
2007 struct extent_map
*em
;
2008 struct block_device
*bdev
;
2009 struct btrfs_ordered_extent
*ordered
;
2012 size_t page_offset
= 0;
2014 size_t disk_io_size
;
2015 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2016 unsigned long this_bio_flag
= 0;
2018 set_page_extent_mapped(page
);
2020 if (!PageUptodate(page
)) {
2021 if (cleancache_get_page(page
) == 0) {
2022 BUG_ON(blocksize
!= PAGE_SIZE
);
2029 lock_extent(tree
, start
, end
, GFP_NOFS
);
2030 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
2033 unlock_extent(tree
, start
, end
, GFP_NOFS
);
2034 btrfs_start_ordered_extent(inode
, ordered
, 1);
2035 btrfs_put_ordered_extent(ordered
);
2038 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2040 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2043 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2044 userpage
= kmap_atomic(page
, KM_USER0
);
2045 memset(userpage
+ zero_offset
, 0, iosize
);
2046 flush_dcache_page(page
);
2047 kunmap_atomic(userpage
, KM_USER0
);
2050 while (cur
<= end
) {
2051 if (cur
>= last_byte
) {
2053 struct extent_state
*cached
= NULL
;
2055 iosize
= PAGE_CACHE_SIZE
- page_offset
;
2056 userpage
= kmap_atomic(page
, KM_USER0
);
2057 memset(userpage
+ page_offset
, 0, iosize
);
2058 flush_dcache_page(page
);
2059 kunmap_atomic(userpage
, KM_USER0
);
2060 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2062 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2066 em
= get_extent(inode
, page
, page_offset
, cur
,
2068 if (IS_ERR(em
) || !em
) {
2070 unlock_extent(tree
, cur
, end
, GFP_NOFS
);
2073 extent_offset
= cur
- em
->start
;
2074 BUG_ON(extent_map_end(em
) <= cur
);
2077 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2078 this_bio_flag
= EXTENT_BIO_COMPRESSED
;
2079 extent_set_compress_type(&this_bio_flag
,
2083 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2084 cur_end
= min(extent_map_end(em
) - 1, end
);
2085 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2086 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2087 disk_io_size
= em
->block_len
;
2088 sector
= em
->block_start
>> 9;
2090 sector
= (em
->block_start
+ extent_offset
) >> 9;
2091 disk_io_size
= iosize
;
2094 block_start
= em
->block_start
;
2095 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2096 block_start
= EXTENT_MAP_HOLE
;
2097 free_extent_map(em
);
2100 /* we've found a hole, just zero and go on */
2101 if (block_start
== EXTENT_MAP_HOLE
) {
2103 struct extent_state
*cached
= NULL
;
2105 userpage
= kmap_atomic(page
, KM_USER0
);
2106 memset(userpage
+ page_offset
, 0, iosize
);
2107 flush_dcache_page(page
);
2108 kunmap_atomic(userpage
, KM_USER0
);
2110 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2112 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2115 page_offset
+= iosize
;
2118 /* the get_extent function already copied into the page */
2119 if (test_range_bit(tree
, cur
, cur_end
,
2120 EXTENT_UPTODATE
, 1, NULL
)) {
2121 check_page_uptodate(tree
, page
);
2122 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2124 page_offset
+= iosize
;
2127 /* we have an inline extent but it didn't get marked up
2128 * to date. Error out
2130 if (block_start
== EXTENT_MAP_INLINE
) {
2132 unlock_extent(tree
, cur
, cur
+ iosize
- 1, GFP_NOFS
);
2134 page_offset
+= iosize
;
2139 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
2140 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
2144 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2146 ret
= submit_extent_page(READ
, tree
, page
,
2147 sector
, disk_io_size
, page_offset
,
2149 end_bio_extent_readpage
, mirror_num
,
2153 *bio_flags
= this_bio_flag
;
2158 page_offset
+= iosize
;
2162 if (!PageError(page
))
2163 SetPageUptodate(page
);
2169 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2170 get_extent_t
*get_extent
)
2172 struct bio
*bio
= NULL
;
2173 unsigned long bio_flags
= 0;
2176 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, 0,
2179 ret
= submit_one_bio(READ
, bio
, 0, bio_flags
);
2183 static noinline
void update_nr_written(struct page
*page
,
2184 struct writeback_control
*wbc
,
2185 unsigned long nr_written
)
2187 wbc
->nr_to_write
-= nr_written
;
2188 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
2189 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
2190 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
2194 * the writepage semantics are similar to regular writepage. extent
2195 * records are inserted to lock ranges in the tree, and as dirty areas
2196 * are found, they are marked writeback. Then the lock bits are removed
2197 * and the end_io handler clears the writeback ranges
2199 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
2202 struct inode
*inode
= page
->mapping
->host
;
2203 struct extent_page_data
*epd
= data
;
2204 struct extent_io_tree
*tree
= epd
->tree
;
2205 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2207 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2211 u64 last_byte
= i_size_read(inode
);
2215 struct extent_state
*cached_state
= NULL
;
2216 struct extent_map
*em
;
2217 struct block_device
*bdev
;
2220 size_t pg_offset
= 0;
2222 loff_t i_size
= i_size_read(inode
);
2223 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2229 unsigned long nr_written
= 0;
2231 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2232 write_flags
= WRITE_SYNC
;
2234 write_flags
= WRITE
;
2236 trace___extent_writepage(page
, inode
, wbc
);
2238 WARN_ON(!PageLocked(page
));
2239 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
2240 if (page
->index
> end_index
||
2241 (page
->index
== end_index
&& !pg_offset
)) {
2242 page
->mapping
->a_ops
->invalidatepage(page
, 0);
2247 if (page
->index
== end_index
) {
2250 userpage
= kmap_atomic(page
, KM_USER0
);
2251 memset(userpage
+ pg_offset
, 0,
2252 PAGE_CACHE_SIZE
- pg_offset
);
2253 kunmap_atomic(userpage
, KM_USER0
);
2254 flush_dcache_page(page
);
2258 set_page_extent_mapped(page
);
2260 delalloc_start
= start
;
2263 if (!epd
->extent_locked
) {
2264 u64 delalloc_to_write
= 0;
2266 * make sure the wbc mapping index is at least updated
2269 update_nr_written(page
, wbc
, 0);
2271 while (delalloc_end
< page_end
) {
2272 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
2277 if (nr_delalloc
== 0) {
2278 delalloc_start
= delalloc_end
+ 1;
2281 tree
->ops
->fill_delalloc(inode
, page
, delalloc_start
,
2282 delalloc_end
, &page_started
,
2285 * delalloc_end is already one less than the total
2286 * length, so we don't subtract one from
2289 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
2292 delalloc_start
= delalloc_end
+ 1;
2294 if (wbc
->nr_to_write
< delalloc_to_write
) {
2297 if (delalloc_to_write
< thresh
* 2)
2298 thresh
= delalloc_to_write
;
2299 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
2303 /* did the fill delalloc function already unlock and start
2309 * we've unlocked the page, so we can't update
2310 * the mapping's writeback index, just update
2313 wbc
->nr_to_write
-= nr_written
;
2317 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
2318 ret
= tree
->ops
->writepage_start_hook(page
, start
,
2320 if (ret
== -EAGAIN
) {
2321 redirty_page_for_writepage(wbc
, page
);
2322 update_nr_written(page
, wbc
, nr_written
);
2330 * we don't want to touch the inode after unlocking the page,
2331 * so we update the mapping writeback index now
2333 update_nr_written(page
, wbc
, nr_written
+ 1);
2336 if (last_byte
<= start
) {
2337 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2338 tree
->ops
->writepage_end_io_hook(page
, start
,
2343 blocksize
= inode
->i_sb
->s_blocksize
;
2345 while (cur
<= end
) {
2346 if (cur
>= last_byte
) {
2347 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2348 tree
->ops
->writepage_end_io_hook(page
, cur
,
2352 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
2354 if (IS_ERR(em
) || !em
) {
2359 extent_offset
= cur
- em
->start
;
2360 BUG_ON(extent_map_end(em
) <= cur
);
2362 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2363 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2364 sector
= (em
->block_start
+ extent_offset
) >> 9;
2366 block_start
= em
->block_start
;
2367 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
2368 free_extent_map(em
);
2372 * compressed and inline extents are written through other
2375 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
2376 block_start
== EXTENT_MAP_INLINE
) {
2378 * end_io notification does not happen here for
2379 * compressed extents
2381 if (!compressed
&& tree
->ops
&&
2382 tree
->ops
->writepage_end_io_hook
)
2383 tree
->ops
->writepage_end_io_hook(page
, cur
,
2386 else if (compressed
) {
2387 /* we don't want to end_page_writeback on
2388 * a compressed extent. this happens
2395 pg_offset
+= iosize
;
2398 /* leave this out until we have a page_mkwrite call */
2399 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2400 EXTENT_DIRTY
, 0, NULL
)) {
2402 pg_offset
+= iosize
;
2406 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2407 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2415 unsigned long max_nr
= end_index
+ 1;
2417 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
2418 if (!PageWriteback(page
)) {
2419 printk(KERN_ERR
"btrfs warning page %lu not "
2420 "writeback, cur %llu end %llu\n",
2421 page
->index
, (unsigned long long)cur
,
2422 (unsigned long long)end
);
2425 ret
= submit_extent_page(write_flags
, tree
, page
,
2426 sector
, iosize
, pg_offset
,
2427 bdev
, &epd
->bio
, max_nr
,
2428 end_bio_extent_writepage
,
2434 pg_offset
+= iosize
;
2439 /* make sure the mapping tag for page dirty gets cleared */
2440 set_page_writeback(page
);
2441 end_page_writeback(page
);
2447 /* drop our reference on any cached states */
2448 free_extent_state(cached_state
);
2453 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2454 * @mapping: address space structure to write
2455 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2456 * @writepage: function called for each page
2457 * @data: data passed to writepage function
2459 * If a page is already under I/O, write_cache_pages() skips it, even
2460 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2461 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2462 * and msync() need to guarantee that all the data which was dirty at the time
2463 * the call was made get new I/O started against them. If wbc->sync_mode is
2464 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2465 * existing IO to complete.
2467 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
2468 struct address_space
*mapping
,
2469 struct writeback_control
*wbc
,
2470 writepage_t writepage
, void *data
,
2471 void (*flush_fn
)(void *))
2475 int nr_to_write_done
= 0;
2476 struct pagevec pvec
;
2479 pgoff_t end
; /* Inclusive */
2482 pagevec_init(&pvec
, 0);
2483 if (wbc
->range_cyclic
) {
2484 index
= mapping
->writeback_index
; /* Start from prev offset */
2487 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2488 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2492 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
2493 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2494 PAGECACHE_TAG_DIRTY
, min(end
- index
,
2495 (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
2499 for (i
= 0; i
< nr_pages
; i
++) {
2500 struct page
*page
= pvec
.pages
[i
];
2503 * At this point we hold neither mapping->tree_lock nor
2504 * lock on the page itself: the page may be truncated or
2505 * invalidated (changing page->mapping to NULL), or even
2506 * swizzled back from swapper_space to tmpfs file
2509 if (tree
->ops
&& tree
->ops
->write_cache_pages_lock_hook
)
2510 tree
->ops
->write_cache_pages_lock_hook(page
);
2514 if (unlikely(page
->mapping
!= mapping
)) {
2519 if (!wbc
->range_cyclic
&& page
->index
> end
) {
2525 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
2526 if (PageWriteback(page
))
2528 wait_on_page_writeback(page
);
2531 if (PageWriteback(page
) ||
2532 !clear_page_dirty_for_io(page
)) {
2537 ret
= (*writepage
)(page
, wbc
, data
);
2539 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
2547 * the filesystem may choose to bump up nr_to_write.
2548 * We have to make sure to honor the new nr_to_write
2551 nr_to_write_done
= wbc
->nr_to_write
<= 0;
2553 pagevec_release(&pvec
);
2556 if (!scanned
&& !done
) {
2558 * We hit the last page and there is more work to be done: wrap
2559 * back to the start of the file
2568 static void flush_epd_write_bio(struct extent_page_data
*epd
)
2572 submit_one_bio(WRITE_SYNC
, epd
->bio
, 0, 0);
2574 submit_one_bio(WRITE
, epd
->bio
, 0, 0);
2579 static noinline
void flush_write_bio(void *data
)
2581 struct extent_page_data
*epd
= data
;
2582 flush_epd_write_bio(epd
);
2585 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2586 get_extent_t
*get_extent
,
2587 struct writeback_control
*wbc
)
2590 struct address_space
*mapping
= page
->mapping
;
2591 struct extent_page_data epd
= {
2594 .get_extent
= get_extent
,
2596 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
2598 struct writeback_control wbc_writepages
= {
2599 .sync_mode
= wbc
->sync_mode
,
2600 .older_than_this
= NULL
,
2602 .range_start
= page_offset(page
) + PAGE_CACHE_SIZE
,
2603 .range_end
= (loff_t
)-1,
2606 ret
= __extent_writepage(page
, wbc
, &epd
);
2608 extent_write_cache_pages(tree
, mapping
, &wbc_writepages
,
2609 __extent_writepage
, &epd
, flush_write_bio
);
2610 flush_epd_write_bio(&epd
);
2614 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
2615 u64 start
, u64 end
, get_extent_t
*get_extent
,
2619 struct address_space
*mapping
= inode
->i_mapping
;
2621 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
2624 struct extent_page_data epd
= {
2627 .get_extent
= get_extent
,
2629 .sync_io
= mode
== WB_SYNC_ALL
,
2631 struct writeback_control wbc_writepages
= {
2633 .older_than_this
= NULL
,
2634 .nr_to_write
= nr_pages
* 2,
2635 .range_start
= start
,
2636 .range_end
= end
+ 1,
2639 while (start
<= end
) {
2640 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
2641 if (clear_page_dirty_for_io(page
))
2642 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
2644 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2645 tree
->ops
->writepage_end_io_hook(page
, start
,
2646 start
+ PAGE_CACHE_SIZE
- 1,
2650 page_cache_release(page
);
2651 start
+= PAGE_CACHE_SIZE
;
2654 flush_epd_write_bio(&epd
);
2658 int extent_writepages(struct extent_io_tree
*tree
,
2659 struct address_space
*mapping
,
2660 get_extent_t
*get_extent
,
2661 struct writeback_control
*wbc
)
2664 struct extent_page_data epd
= {
2667 .get_extent
= get_extent
,
2669 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
2672 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
2673 __extent_writepage
, &epd
,
2675 flush_epd_write_bio(&epd
);
2679 int extent_readpages(struct extent_io_tree
*tree
,
2680 struct address_space
*mapping
,
2681 struct list_head
*pages
, unsigned nr_pages
,
2682 get_extent_t get_extent
)
2684 struct bio
*bio
= NULL
;
2686 unsigned long bio_flags
= 0;
2688 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
2689 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
2691 prefetchw(&page
->flags
);
2692 list_del(&page
->lru
);
2693 if (!add_to_page_cache_lru(page
, mapping
,
2694 page
->index
, GFP_NOFS
)) {
2695 __extent_read_full_page(tree
, page
, get_extent
,
2696 &bio
, 0, &bio_flags
);
2698 page_cache_release(page
);
2700 BUG_ON(!list_empty(pages
));
2702 submit_one_bio(READ
, bio
, 0, bio_flags
);
2707 * basic invalidatepage code, this waits on any locked or writeback
2708 * ranges corresponding to the page, and then deletes any extent state
2709 * records from the tree
2711 int extent_invalidatepage(struct extent_io_tree
*tree
,
2712 struct page
*page
, unsigned long offset
)
2714 struct extent_state
*cached_state
= NULL
;
2715 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2716 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2717 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
2719 start
+= (offset
+ blocksize
- 1) & ~(blocksize
- 1);
2723 lock_extent_bits(tree
, start
, end
, 0, &cached_state
, GFP_NOFS
);
2724 wait_on_page_writeback(page
);
2725 clear_extent_bit(tree
, start
, end
,
2726 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
2727 EXTENT_DO_ACCOUNTING
,
2728 1, 1, &cached_state
, GFP_NOFS
);
2733 * simple commit_write call, set_range_dirty is used to mark both
2734 * the pages and the extent records as dirty
2736 int extent_commit_write(struct extent_io_tree
*tree
,
2737 struct inode
*inode
, struct page
*page
,
2738 unsigned from
, unsigned to
)
2740 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
2742 set_page_extent_mapped(page
);
2743 set_page_dirty(page
);
2745 if (pos
> inode
->i_size
) {
2746 i_size_write(inode
, pos
);
2747 mark_inode_dirty(inode
);
2752 int extent_prepare_write(struct extent_io_tree
*tree
,
2753 struct inode
*inode
, struct page
*page
,
2754 unsigned from
, unsigned to
, get_extent_t
*get_extent
)
2756 u64 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2757 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2759 u64 orig_block_start
;
2762 struct extent_map
*em
;
2763 unsigned blocksize
= 1 << inode
->i_blkbits
;
2764 size_t page_offset
= 0;
2765 size_t block_off_start
;
2766 size_t block_off_end
;
2772 set_page_extent_mapped(page
);
2774 block_start
= (page_start
+ from
) & ~((u64
)blocksize
- 1);
2775 block_end
= (page_start
+ to
- 1) | (blocksize
- 1);
2776 orig_block_start
= block_start
;
2778 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
2779 while (block_start
<= block_end
) {
2780 em
= get_extent(inode
, page
, page_offset
, block_start
,
2781 block_end
- block_start
+ 1, 1);
2782 if (IS_ERR(em
) || !em
)
2785 cur_end
= min(block_end
, extent_map_end(em
) - 1);
2786 block_off_start
= block_start
& (PAGE_CACHE_SIZE
- 1);
2787 block_off_end
= block_off_start
+ blocksize
;
2788 isnew
= clear_extent_new(tree
, block_start
, cur_end
, GFP_NOFS
);
2790 if (!PageUptodate(page
) && isnew
&&
2791 (block_off_end
> to
|| block_off_start
< from
)) {
2794 kaddr
= kmap_atomic(page
, KM_USER0
);
2795 if (block_off_end
> to
)
2796 memset(kaddr
+ to
, 0, block_off_end
- to
);
2797 if (block_off_start
< from
)
2798 memset(kaddr
+ block_off_start
, 0,
2799 from
- block_off_start
);
2800 flush_dcache_page(page
);
2801 kunmap_atomic(kaddr
, KM_USER0
);
2803 if ((em
->block_start
!= EXTENT_MAP_HOLE
&&
2804 em
->block_start
!= EXTENT_MAP_INLINE
) &&
2805 !isnew
&& !PageUptodate(page
) &&
2806 (block_off_end
> to
|| block_off_start
< from
) &&
2807 !test_range_bit(tree
, block_start
, cur_end
,
2808 EXTENT_UPTODATE
, 1, NULL
)) {
2810 u64 extent_offset
= block_start
- em
->start
;
2812 sector
= (em
->block_start
+ extent_offset
) >> 9;
2813 iosize
= (cur_end
- block_start
+ blocksize
) &
2814 ~((u64
)blocksize
- 1);
2816 * we've already got the extent locked, but we
2817 * need to split the state such that our end_bio
2818 * handler can clear the lock.
2820 set_extent_bit(tree
, block_start
,
2821 block_start
+ iosize
- 1,
2822 EXTENT_LOCKED
, 0, NULL
, NULL
, GFP_NOFS
);
2823 ret
= submit_extent_page(READ
, tree
, page
,
2824 sector
, iosize
, page_offset
, em
->bdev
,
2826 end_bio_extent_preparewrite
, 0,
2831 block_start
= block_start
+ iosize
;
2833 struct extent_state
*cached
= NULL
;
2835 set_extent_uptodate(tree
, block_start
, cur_end
, &cached
,
2837 unlock_extent_cached(tree
, block_start
, cur_end
,
2839 block_start
= cur_end
+ 1;
2841 page_offset
= block_start
& (PAGE_CACHE_SIZE
- 1);
2842 free_extent_map(em
);
2845 wait_extent_bit(tree
, orig_block_start
,
2846 block_end
, EXTENT_LOCKED
);
2848 check_page_uptodate(tree
, page
);
2850 /* FIXME, zero out newly allocated blocks on error */
2855 * a helper for releasepage, this tests for areas of the page that
2856 * are locked or under IO and drops the related state bits if it is safe
2859 int try_release_extent_state(struct extent_map_tree
*map
,
2860 struct extent_io_tree
*tree
, struct page
*page
,
2863 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2864 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2867 if (test_range_bit(tree
, start
, end
,
2868 EXTENT_IOBITS
, 0, NULL
))
2871 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
2874 * at this point we can safely clear everything except the
2875 * locked bit and the nodatasum bit
2877 ret
= clear_extent_bit(tree
, start
, end
,
2878 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
2881 /* if clear_extent_bit failed for enomem reasons,
2882 * we can't allow the release to continue.
2893 * a helper for releasepage. As long as there are no locked extents
2894 * in the range corresponding to the page, both state records and extent
2895 * map records are removed
2897 int try_release_extent_mapping(struct extent_map_tree
*map
,
2898 struct extent_io_tree
*tree
, struct page
*page
,
2901 struct extent_map
*em
;
2902 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2903 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2905 if ((mask
& __GFP_WAIT
) &&
2906 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
2908 while (start
<= end
) {
2909 len
= end
- start
+ 1;
2910 write_lock(&map
->lock
);
2911 em
= lookup_extent_mapping(map
, start
, len
);
2912 if (!em
|| IS_ERR(em
)) {
2913 write_unlock(&map
->lock
);
2916 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
2917 em
->start
!= start
) {
2918 write_unlock(&map
->lock
);
2919 free_extent_map(em
);
2922 if (!test_range_bit(tree
, em
->start
,
2923 extent_map_end(em
) - 1,
2924 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
2926 remove_extent_mapping(map
, em
);
2927 /* once for the rb tree */
2928 free_extent_map(em
);
2930 start
= extent_map_end(em
);
2931 write_unlock(&map
->lock
);
2934 free_extent_map(em
);
2937 return try_release_extent_state(map
, tree
, page
, mask
);
2940 sector_t
extent_bmap(struct address_space
*mapping
, sector_t iblock
,
2941 get_extent_t
*get_extent
)
2943 struct inode
*inode
= mapping
->host
;
2944 struct extent_state
*cached_state
= NULL
;
2945 u64 start
= iblock
<< inode
->i_blkbits
;
2946 sector_t sector
= 0;
2947 size_t blksize
= (1 << inode
->i_blkbits
);
2948 struct extent_map
*em
;
2950 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ blksize
- 1,
2951 0, &cached_state
, GFP_NOFS
);
2952 em
= get_extent(inode
, NULL
, 0, start
, blksize
, 0);
2953 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
,
2954 start
+ blksize
- 1, &cached_state
, GFP_NOFS
);
2955 if (!em
|| IS_ERR(em
))
2958 if (em
->block_start
> EXTENT_MAP_LAST_BYTE
)
2961 sector
= (em
->block_start
+ start
- em
->start
) >> inode
->i_blkbits
;
2963 free_extent_map(em
);
2968 * helper function for fiemap, which doesn't want to see any holes.
2969 * This maps until we find something past 'last'
2971 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
2974 get_extent_t
*get_extent
)
2976 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
2977 struct extent_map
*em
;
2984 len
= last
- offset
;
2987 len
= (len
+ sectorsize
- 1) & ~(sectorsize
- 1);
2988 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
2989 if (!em
|| IS_ERR(em
))
2992 /* if this isn't a hole return it */
2993 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
2994 em
->block_start
!= EXTENT_MAP_HOLE
) {
2998 /* this is a hole, advance to the next extent */
2999 offset
= extent_map_end(em
);
3000 free_extent_map(em
);
3007 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
3008 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
3012 u64 max
= start
+ len
;
3016 u64 last_for_get_extent
= 0;
3018 u64 isize
= i_size_read(inode
);
3019 struct btrfs_key found_key
;
3020 struct extent_map
*em
= NULL
;
3021 struct extent_state
*cached_state
= NULL
;
3022 struct btrfs_path
*path
;
3023 struct btrfs_file_extent_item
*item
;
3028 unsigned long emflags
;
3033 path
= btrfs_alloc_path();
3036 path
->leave_spinning
= 1;
3039 * lookup the last file extent. We're not using i_size here
3040 * because there might be preallocation past i_size
3042 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
3043 path
, inode
->i_ino
, -1, 0);
3045 btrfs_free_path(path
);
3050 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3051 struct btrfs_file_extent_item
);
3052 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
3053 found_type
= btrfs_key_type(&found_key
);
3055 /* No extents, but there might be delalloc bits */
3056 if (found_key
.objectid
!= inode
->i_ino
||
3057 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3058 /* have to trust i_size as the end */
3060 last_for_get_extent
= isize
;
3063 * remember the start of the last extent. There are a
3064 * bunch of different factors that go into the length of the
3065 * extent, so its much less complex to remember where it started
3067 last
= found_key
.offset
;
3068 last_for_get_extent
= last
+ 1;
3070 btrfs_free_path(path
);
3073 * we might have some extents allocated but more delalloc past those
3074 * extents. so, we trust isize unless the start of the last extent is
3079 last_for_get_extent
= isize
;
3082 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
, 0,
3083 &cached_state
, GFP_NOFS
);
3085 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
3095 u64 offset_in_extent
;
3097 /* break if the extent we found is outside the range */
3098 if (em
->start
>= max
|| extent_map_end(em
) < off
)
3102 * get_extent may return an extent that starts before our
3103 * requested range. We have to make sure the ranges
3104 * we return to fiemap always move forward and don't
3105 * overlap, so adjust the offsets here
3107 em_start
= max(em
->start
, off
);
3110 * record the offset from the start of the extent
3111 * for adjusting the disk offset below
3113 offset_in_extent
= em_start
- em
->start
;
3114 em_end
= extent_map_end(em
);
3115 em_len
= em_end
- em_start
;
3116 emflags
= em
->flags
;
3121 * bump off for our next call to get_extent
3123 off
= extent_map_end(em
);
3127 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
3129 flags
|= FIEMAP_EXTENT_LAST
;
3130 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
3131 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
3132 FIEMAP_EXTENT_NOT_ALIGNED
);
3133 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
3134 flags
|= (FIEMAP_EXTENT_DELALLOC
|
3135 FIEMAP_EXTENT_UNKNOWN
);
3137 disko
= em
->block_start
+ offset_in_extent
;
3139 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
3140 flags
|= FIEMAP_EXTENT_ENCODED
;
3142 free_extent_map(em
);
3144 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
3145 (last
== (u64
)-1 && isize
<= em_end
)) {
3146 flags
|= FIEMAP_EXTENT_LAST
;
3150 /* now scan forward to see if this is really the last extent. */
3151 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
3158 flags
|= FIEMAP_EXTENT_LAST
;
3161 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
3167 free_extent_map(em
);
3169 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
,
3170 &cached_state
, GFP_NOFS
);
3174 static inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
3178 struct address_space
*mapping
;
3181 return eb
->first_page
;
3182 i
+= eb
->start
>> PAGE_CACHE_SHIFT
;
3183 mapping
= eb
->first_page
->mapping
;
3188 * extent_buffer_page is only called after pinning the page
3189 * by increasing the reference count. So we know the page must
3190 * be in the radix tree.
3193 p
= radix_tree_lookup(&mapping
->page_tree
, i
);
3199 static inline unsigned long num_extent_pages(u64 start
, u64 len
)
3201 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
3202 (start
>> PAGE_CACHE_SHIFT
);
3205 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
3210 struct extent_buffer
*eb
= NULL
;
3212 unsigned long flags
;
3215 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
3220 spin_lock_init(&eb
->lock
);
3221 init_waitqueue_head(&eb
->lock_wq
);
3224 spin_lock_irqsave(&leak_lock
, flags
);
3225 list_add(&eb
->leak_list
, &buffers
);
3226 spin_unlock_irqrestore(&leak_lock
, flags
);
3228 atomic_set(&eb
->refs
, 1);
3233 static void __free_extent_buffer(struct extent_buffer
*eb
)
3236 unsigned long flags
;
3237 spin_lock_irqsave(&leak_lock
, flags
);
3238 list_del(&eb
->leak_list
);
3239 spin_unlock_irqrestore(&leak_lock
, flags
);
3241 kmem_cache_free(extent_buffer_cache
, eb
);
3245 * Helper for releasing extent buffer page.
3247 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
3248 unsigned long start_idx
)
3250 unsigned long index
;
3253 if (!eb
->first_page
)
3256 index
= num_extent_pages(eb
->start
, eb
->len
);
3257 if (start_idx
>= index
)
3262 page
= extent_buffer_page(eb
, index
);
3264 page_cache_release(page
);
3265 } while (index
!= start_idx
);
3269 * Helper for releasing the extent buffer.
3271 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
3273 btrfs_release_extent_buffer_page(eb
, 0);
3274 __free_extent_buffer(eb
);
3277 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
3278 u64 start
, unsigned long len
,
3282 unsigned long num_pages
= num_extent_pages(start
, len
);
3284 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
3285 struct extent_buffer
*eb
;
3286 struct extent_buffer
*exists
= NULL
;
3288 struct address_space
*mapping
= tree
->mapping
;
3293 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
3294 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
3296 mark_page_accessed(eb
->first_page
);
3301 eb
= __alloc_extent_buffer(tree
, start
, len
, mask
);
3306 eb
->first_page
= page0
;
3309 page_cache_get(page0
);
3310 mark_page_accessed(page0
);
3311 set_page_extent_mapped(page0
);
3312 set_page_extent_head(page0
, len
);
3313 uptodate
= PageUptodate(page0
);
3317 for (; i
< num_pages
; i
++, index
++) {
3318 p
= find_or_create_page(mapping
, index
, mask
| __GFP_HIGHMEM
);
3323 set_page_extent_mapped(p
);
3324 mark_page_accessed(p
);
3327 set_page_extent_head(p
, len
);
3329 set_page_private(p
, EXTENT_PAGE_PRIVATE
);
3331 if (!PageUptodate(p
))
3335 * see below about how we avoid a nasty race with release page
3336 * and why we unlock later
3342 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3344 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
3348 spin_lock(&tree
->buffer_lock
);
3349 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
3350 if (ret
== -EEXIST
) {
3351 exists
= radix_tree_lookup(&tree
->buffer
,
3352 start
>> PAGE_CACHE_SHIFT
);
3353 /* add one reference for the caller */
3354 atomic_inc(&exists
->refs
);
3355 spin_unlock(&tree
->buffer_lock
);
3356 radix_tree_preload_end();
3359 /* add one reference for the tree */
3360 atomic_inc(&eb
->refs
);
3361 spin_unlock(&tree
->buffer_lock
);
3362 radix_tree_preload_end();
3365 * there is a race where release page may have
3366 * tried to find this extent buffer in the radix
3367 * but failed. It will tell the VM it is safe to
3368 * reclaim the, and it will clear the page private bit.
3369 * We must make sure to set the page private bit properly
3370 * after the extent buffer is in the radix tree so
3371 * it doesn't get lost
3373 set_page_extent_mapped(eb
->first_page
);
3374 set_page_extent_head(eb
->first_page
, eb
->len
);
3376 unlock_page(eb
->first_page
);
3380 if (eb
->first_page
&& !page0
)
3381 unlock_page(eb
->first_page
);
3383 if (!atomic_dec_and_test(&eb
->refs
))
3385 btrfs_release_extent_buffer(eb
);
3389 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
3390 u64 start
, unsigned long len
,
3393 struct extent_buffer
*eb
;
3396 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
3397 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
3399 mark_page_accessed(eb
->first_page
);
3407 void free_extent_buffer(struct extent_buffer
*eb
)
3412 if (!atomic_dec_and_test(&eb
->refs
))
3418 int clear_extent_buffer_dirty(struct extent_io_tree
*tree
,
3419 struct extent_buffer
*eb
)
3422 unsigned long num_pages
;
3425 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3427 for (i
= 0; i
< num_pages
; i
++) {
3428 page
= extent_buffer_page(eb
, i
);
3429 if (!PageDirty(page
))
3433 WARN_ON(!PagePrivate(page
));
3435 set_page_extent_mapped(page
);
3437 set_page_extent_head(page
, eb
->len
);
3439 clear_page_dirty_for_io(page
);
3440 spin_lock_irq(&page
->mapping
->tree_lock
);
3441 if (!PageDirty(page
)) {
3442 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3444 PAGECACHE_TAG_DIRTY
);
3446 spin_unlock_irq(&page
->mapping
->tree_lock
);
3452 int wait_on_extent_buffer_writeback(struct extent_io_tree
*tree
,
3453 struct extent_buffer
*eb
)
3455 return wait_on_extent_writeback(tree
, eb
->start
,
3456 eb
->start
+ eb
->len
- 1);
3459 int set_extent_buffer_dirty(struct extent_io_tree
*tree
,
3460 struct extent_buffer
*eb
)
3463 unsigned long num_pages
;
3466 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
3467 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3468 for (i
= 0; i
< num_pages
; i
++)
3469 __set_page_dirty_nobuffers(extent_buffer_page(eb
, i
));
3473 int clear_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3474 struct extent_buffer
*eb
,
3475 struct extent_state
**cached_state
)
3479 unsigned long num_pages
;
3481 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3482 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3484 clear_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3485 cached_state
, GFP_NOFS
);
3486 for (i
= 0; i
< num_pages
; i
++) {
3487 page
= extent_buffer_page(eb
, i
);
3489 ClearPageUptodate(page
);
3494 int set_extent_buffer_uptodate(struct extent_io_tree
*tree
,
3495 struct extent_buffer
*eb
)
3499 unsigned long num_pages
;
3501 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3503 set_extent_uptodate(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3505 for (i
= 0; i
< num_pages
; i
++) {
3506 page
= extent_buffer_page(eb
, i
);
3507 if ((i
== 0 && (eb
->start
& (PAGE_CACHE_SIZE
- 1))) ||
3508 ((i
== num_pages
- 1) &&
3509 ((eb
->start
+ eb
->len
) & (PAGE_CACHE_SIZE
- 1)))) {
3510 check_page_uptodate(tree
, page
);
3513 SetPageUptodate(page
);
3518 int extent_range_uptodate(struct extent_io_tree
*tree
,
3523 int pg_uptodate
= 1;
3525 unsigned long index
;
3527 ret
= test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
);
3530 while (start
<= end
) {
3531 index
= start
>> PAGE_CACHE_SHIFT
;
3532 page
= find_get_page(tree
->mapping
, index
);
3533 uptodate
= PageUptodate(page
);
3534 page_cache_release(page
);
3539 start
+= PAGE_CACHE_SIZE
;
3544 int extent_buffer_uptodate(struct extent_io_tree
*tree
,
3545 struct extent_buffer
*eb
,
3546 struct extent_state
*cached_state
)
3549 unsigned long num_pages
;
3552 int pg_uptodate
= 1;
3554 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
3557 ret
= test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3558 EXTENT_UPTODATE
, 1, cached_state
);
3562 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3563 for (i
= 0; i
< num_pages
; i
++) {
3564 page
= extent_buffer_page(eb
, i
);
3565 if (!PageUptodate(page
)) {
3573 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
3574 struct extent_buffer
*eb
,
3575 u64 start
, int wait
,
3576 get_extent_t
*get_extent
, int mirror_num
)
3579 unsigned long start_i
;
3583 int locked_pages
= 0;
3584 int all_uptodate
= 1;
3585 int inc_all_pages
= 0;
3586 unsigned long num_pages
;
3587 struct bio
*bio
= NULL
;
3588 unsigned long bio_flags
= 0;
3590 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
3593 if (test_range_bit(tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
3594 EXTENT_UPTODATE
, 1, NULL
)) {
3599 WARN_ON(start
< eb
->start
);
3600 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
3601 (eb
->start
>> PAGE_CACHE_SHIFT
);
3606 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3607 for (i
= start_i
; i
< num_pages
; i
++) {
3608 page
= extent_buffer_page(eb
, i
);
3610 if (!trylock_page(page
))
3616 if (!PageUptodate(page
))
3621 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3625 for (i
= start_i
; i
< num_pages
; i
++) {
3626 page
= extent_buffer_page(eb
, i
);
3628 WARN_ON(!PagePrivate(page
));
3630 set_page_extent_mapped(page
);
3632 set_page_extent_head(page
, eb
->len
);
3635 page_cache_get(page
);
3636 if (!PageUptodate(page
)) {
3639 ClearPageError(page
);
3640 err
= __extent_read_full_page(tree
, page
,
3642 mirror_num
, &bio_flags
);
3651 submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3656 for (i
= start_i
; i
< num_pages
; i
++) {
3657 page
= extent_buffer_page(eb
, i
);
3658 wait_on_page_locked(page
);
3659 if (!PageUptodate(page
))
3664 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
3669 while (locked_pages
> 0) {
3670 page
= extent_buffer_page(eb
, i
);
3678 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
3679 unsigned long start
,
3686 char *dst
= (char *)dstv
;
3687 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3688 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3690 WARN_ON(start
> eb
->len
);
3691 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3693 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3696 page
= extent_buffer_page(eb
, i
);
3698 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3699 kaddr
= kmap_atomic(page
, KM_USER1
);
3700 memcpy(dst
, kaddr
+ offset
, cur
);
3701 kunmap_atomic(kaddr
, KM_USER1
);
3710 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3711 unsigned long min_len
, char **token
, char **map
,
3712 unsigned long *map_start
,
3713 unsigned long *map_len
, int km
)
3715 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
3718 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3719 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3720 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
3727 offset
= start_offset
;
3731 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
3734 if (start
+ min_len
> eb
->len
) {
3735 printk(KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
3736 "wanted %lu %lu\n", (unsigned long long)eb
->start
,
3737 eb
->len
, start
, min_len
);
3742 p
= extent_buffer_page(eb
, i
);
3743 kaddr
= kmap_atomic(p
, km
);
3745 *map
= kaddr
+ offset
;
3746 *map_len
= PAGE_CACHE_SIZE
- offset
;
3750 int map_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
3751 unsigned long min_len
,
3752 char **token
, char **map
,
3753 unsigned long *map_start
,
3754 unsigned long *map_len
, int km
)
3758 if (eb
->map_token
) {
3759 unmap_extent_buffer(eb
, eb
->map_token
, km
);
3760 eb
->map_token
= NULL
;
3763 err
= map_private_extent_buffer(eb
, start
, min_len
, token
, map
,
3764 map_start
, map_len
, km
);
3766 eb
->map_token
= *token
;
3768 eb
->map_start
= *map_start
;
3769 eb
->map_len
= *map_len
;
3774 void unmap_extent_buffer(struct extent_buffer
*eb
, char *token
, int km
)
3776 kunmap_atomic(token
, km
);
3779 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
3780 unsigned long start
,
3787 char *ptr
= (char *)ptrv
;
3788 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3789 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3792 WARN_ON(start
> eb
->len
);
3793 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3795 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3798 page
= extent_buffer_page(eb
, i
);
3800 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
3802 kaddr
= kmap_atomic(page
, KM_USER0
);
3803 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
3804 kunmap_atomic(kaddr
, KM_USER0
);
3816 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
3817 unsigned long start
, unsigned long len
)
3823 char *src
= (char *)srcv
;
3824 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3825 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3827 WARN_ON(start
> eb
->len
);
3828 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3830 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3833 page
= extent_buffer_page(eb
, i
);
3834 WARN_ON(!PageUptodate(page
));
3836 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3837 kaddr
= kmap_atomic(page
, KM_USER1
);
3838 memcpy(kaddr
+ offset
, src
, cur
);
3839 kunmap_atomic(kaddr
, KM_USER1
);
3848 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
3849 unsigned long start
, unsigned long len
)
3855 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3856 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
3858 WARN_ON(start
> eb
->len
);
3859 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
3861 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
3864 page
= extent_buffer_page(eb
, i
);
3865 WARN_ON(!PageUptodate(page
));
3867 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
3868 kaddr
= kmap_atomic(page
, KM_USER0
);
3869 memset(kaddr
+ offset
, c
, cur
);
3870 kunmap_atomic(kaddr
, KM_USER0
);
3878 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
3879 unsigned long dst_offset
, unsigned long src_offset
,
3882 u64 dst_len
= dst
->len
;
3887 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3888 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3890 WARN_ON(src
->len
!= dst_len
);
3892 offset
= (start_offset
+ dst_offset
) &
3893 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3896 page
= extent_buffer_page(dst
, i
);
3897 WARN_ON(!PageUptodate(page
));
3899 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
3901 kaddr
= kmap_atomic(page
, KM_USER0
);
3902 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
3903 kunmap_atomic(kaddr
, KM_USER0
);
3912 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
3913 unsigned long dst_off
, unsigned long src_off
,
3916 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3917 if (dst_page
== src_page
) {
3918 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
3920 char *src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3921 char *p
= dst_kaddr
+ dst_off
+ len
;
3922 char *s
= src_kaddr
+ src_off
+ len
;
3927 kunmap_atomic(src_kaddr
, KM_USER1
);
3929 kunmap_atomic(dst_kaddr
, KM_USER0
);
3932 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
3934 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
3935 return distance
< len
;
3938 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
3939 unsigned long dst_off
, unsigned long src_off
,
3942 char *dst_kaddr
= kmap_atomic(dst_page
, KM_USER0
);
3945 if (dst_page
!= src_page
) {
3946 src_kaddr
= kmap_atomic(src_page
, KM_USER1
);
3948 src_kaddr
= dst_kaddr
;
3949 BUG_ON(areas_overlap(src_off
, dst_off
, len
));
3952 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
3953 kunmap_atomic(dst_kaddr
, KM_USER0
);
3954 if (dst_page
!= src_page
)
3955 kunmap_atomic(src_kaddr
, KM_USER1
);
3958 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
3959 unsigned long src_offset
, unsigned long len
)
3962 size_t dst_off_in_page
;
3963 size_t src_off_in_page
;
3964 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
3965 unsigned long dst_i
;
3966 unsigned long src_i
;
3968 if (src_offset
+ len
> dst
->len
) {
3969 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
3970 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
3973 if (dst_offset
+ len
> dst
->len
) {
3974 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
3975 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
3980 dst_off_in_page
= (start_offset
+ dst_offset
) &
3981 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3982 src_off_in_page
= (start_offset
+ src_offset
) &
3983 ((unsigned long)PAGE_CACHE_SIZE
- 1);
3985 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
3986 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
3988 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
3990 cur
= min_t(unsigned long, cur
,
3991 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
3993 copy_pages(extent_buffer_page(dst
, dst_i
),
3994 extent_buffer_page(dst
, src_i
),
3995 dst_off_in_page
, src_off_in_page
, cur
);
4003 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
4004 unsigned long src_offset
, unsigned long len
)
4007 size_t dst_off_in_page
;
4008 size_t src_off_in_page
;
4009 unsigned long dst_end
= dst_offset
+ len
- 1;
4010 unsigned long src_end
= src_offset
+ len
- 1;
4011 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4012 unsigned long dst_i
;
4013 unsigned long src_i
;
4015 if (src_offset
+ len
> dst
->len
) {
4016 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
4017 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
4020 if (dst_offset
+ len
> dst
->len
) {
4021 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
4022 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
4025 if (!areas_overlap(src_offset
, dst_offset
, len
)) {
4026 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
4030 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
4031 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
4033 dst_off_in_page
= (start_offset
+ dst_end
) &
4034 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4035 src_off_in_page
= (start_offset
+ src_end
) &
4036 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4038 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
4039 cur
= min(cur
, dst_off_in_page
+ 1);
4040 move_pages(extent_buffer_page(dst
, dst_i
),
4041 extent_buffer_page(dst
, src_i
),
4042 dst_off_in_page
- cur
+ 1,
4043 src_off_in_page
- cur
+ 1, cur
);
4051 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4053 struct extent_buffer
*eb
=
4054 container_of(head
, struct extent_buffer
, rcu_head
);
4056 btrfs_release_extent_buffer(eb
);
4059 int try_release_extent_buffer(struct extent_io_tree
*tree
, struct page
*page
)
4061 u64 start
= page_offset(page
);
4062 struct extent_buffer
*eb
;
4065 spin_lock(&tree
->buffer_lock
);
4066 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4068 spin_unlock(&tree
->buffer_lock
);
4072 if (test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
4078 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4081 if (atomic_cmpxchg(&eb
->refs
, 1, 0) != 1) {
4086 radix_tree_delete(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4088 spin_unlock(&tree
->buffer_lock
);
4090 /* at this point we can safely release the extent buffer */
4091 if (atomic_read(&eb
->refs
) == 0)
4092 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);