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 #include "check-integrity.h"
24 static struct kmem_cache
*extent_state_cache
;
25 static struct kmem_cache
*extent_buffer_cache
;
27 static LIST_HEAD(buffers
);
28 static LIST_HEAD(states
);
32 static DEFINE_SPINLOCK(leak_lock
);
35 #define BUFFER_LRU_MAX 64
40 struct rb_node rb_node
;
43 struct extent_page_data
{
45 struct extent_io_tree
*tree
;
46 get_extent_t
*get_extent
;
48 /* tells writepage not to lock the state bits for this range
49 * it still does the unlocking
51 unsigned int extent_locked
:1;
53 /* tells the submit_bio code to use a WRITE_SYNC */
54 unsigned int sync_io
:1;
57 static noinline
void flush_write_bio(void *data
);
58 static inline struct btrfs_fs_info
*
59 tree_fs_info(struct extent_io_tree
*tree
)
61 return btrfs_sb(tree
->mapping
->host
->i_sb
);
64 int __init
extent_io_init(void)
66 extent_state_cache
= kmem_cache_create("extent_state",
67 sizeof(struct extent_state
), 0,
68 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
69 if (!extent_state_cache
)
72 extent_buffer_cache
= kmem_cache_create("extent_buffers",
73 sizeof(struct extent_buffer
), 0,
74 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
75 if (!extent_buffer_cache
)
76 goto free_state_cache
;
80 kmem_cache_destroy(extent_state_cache
);
84 void extent_io_exit(void)
86 struct extent_state
*state
;
87 struct extent_buffer
*eb
;
89 while (!list_empty(&states
)) {
90 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
91 printk(KERN_ERR
"btrfs state leak: start %llu end %llu "
92 "state %lu in tree %p refs %d\n",
93 (unsigned long long)state
->start
,
94 (unsigned long long)state
->end
,
95 state
->state
, state
->tree
, atomic_read(&state
->refs
));
96 list_del(&state
->leak_list
);
97 kmem_cache_free(extent_state_cache
, state
);
101 while (!list_empty(&buffers
)) {
102 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
103 printk(KERN_ERR
"btrfs buffer leak start %llu len %lu "
104 "refs %d\n", (unsigned long long)eb
->start
,
105 eb
->len
, atomic_read(&eb
->refs
));
106 list_del(&eb
->leak_list
);
107 kmem_cache_free(extent_buffer_cache
, eb
);
109 if (extent_state_cache
)
110 kmem_cache_destroy(extent_state_cache
);
111 if (extent_buffer_cache
)
112 kmem_cache_destroy(extent_buffer_cache
);
115 void extent_io_tree_init(struct extent_io_tree
*tree
,
116 struct address_space
*mapping
)
118 tree
->state
= RB_ROOT
;
119 INIT_RADIX_TREE(&tree
->buffer
, GFP_ATOMIC
);
121 tree
->dirty_bytes
= 0;
122 spin_lock_init(&tree
->lock
);
123 spin_lock_init(&tree
->buffer_lock
);
124 tree
->mapping
= mapping
;
127 static struct extent_state
*alloc_extent_state(gfp_t mask
)
129 struct extent_state
*state
;
134 state
= kmem_cache_alloc(extent_state_cache
, mask
);
141 spin_lock_irqsave(&leak_lock
, flags
);
142 list_add(&state
->leak_list
, &states
);
143 spin_unlock_irqrestore(&leak_lock
, flags
);
145 atomic_set(&state
->refs
, 1);
146 init_waitqueue_head(&state
->wq
);
147 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
151 void free_extent_state(struct extent_state
*state
)
155 if (atomic_dec_and_test(&state
->refs
)) {
159 WARN_ON(state
->tree
);
161 spin_lock_irqsave(&leak_lock
, flags
);
162 list_del(&state
->leak_list
);
163 spin_unlock_irqrestore(&leak_lock
, flags
);
165 trace_free_extent_state(state
, _RET_IP_
);
166 kmem_cache_free(extent_state_cache
, state
);
170 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 offset
,
171 struct rb_node
*node
)
173 struct rb_node
**p
= &root
->rb_node
;
174 struct rb_node
*parent
= NULL
;
175 struct tree_entry
*entry
;
179 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
181 if (offset
< entry
->start
)
183 else if (offset
> entry
->end
)
189 rb_link_node(node
, parent
, p
);
190 rb_insert_color(node
, root
);
194 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
195 struct rb_node
**prev_ret
,
196 struct rb_node
**next_ret
)
198 struct rb_root
*root
= &tree
->state
;
199 struct rb_node
*n
= root
->rb_node
;
200 struct rb_node
*prev
= NULL
;
201 struct rb_node
*orig_prev
= NULL
;
202 struct tree_entry
*entry
;
203 struct tree_entry
*prev_entry
= NULL
;
206 entry
= rb_entry(n
, struct tree_entry
, rb_node
);
210 if (offset
< entry
->start
)
212 else if (offset
> entry
->end
)
220 while (prev
&& offset
> prev_entry
->end
) {
221 prev
= rb_next(prev
);
222 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
229 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
230 while (prev
&& offset
< prev_entry
->start
) {
231 prev
= rb_prev(prev
);
232 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
239 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
242 struct rb_node
*prev
= NULL
;
245 ret
= __etree_search(tree
, offset
, &prev
, NULL
);
251 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
252 struct extent_state
*other
)
254 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
255 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
260 * utility function to look for merge candidates inside a given range.
261 * Any extents with matching state are merged together into a single
262 * extent in the tree. Extents with EXTENT_IO in their state field
263 * are not merged because the end_io handlers need to be able to do
264 * operations on them without sleeping (or doing allocations/splits).
266 * This should be called with the tree lock held.
268 static void merge_state(struct extent_io_tree
*tree
,
269 struct extent_state
*state
)
271 struct extent_state
*other
;
272 struct rb_node
*other_node
;
274 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
277 other_node
= rb_prev(&state
->rb_node
);
279 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
280 if (other
->end
== state
->start
- 1 &&
281 other
->state
== state
->state
) {
282 merge_cb(tree
, state
, other
);
283 state
->start
= other
->start
;
285 rb_erase(&other
->rb_node
, &tree
->state
);
286 free_extent_state(other
);
289 other_node
= rb_next(&state
->rb_node
);
291 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
292 if (other
->start
== state
->end
+ 1 &&
293 other
->state
== state
->state
) {
294 merge_cb(tree
, state
, other
);
295 state
->end
= other
->end
;
297 rb_erase(&other
->rb_node
, &tree
->state
);
298 free_extent_state(other
);
303 static void set_state_cb(struct extent_io_tree
*tree
,
304 struct extent_state
*state
, int *bits
)
306 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
307 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
310 static void clear_state_cb(struct extent_io_tree
*tree
,
311 struct extent_state
*state
, int *bits
)
313 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
314 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
317 static void set_state_bits(struct extent_io_tree
*tree
,
318 struct extent_state
*state
, int *bits
);
321 * insert an extent_state struct into the tree. 'bits' are set on the
322 * struct before it is inserted.
324 * This may return -EEXIST if the extent is already there, in which case the
325 * state struct is freed.
327 * The tree lock is not taken internally. This is a utility function and
328 * probably isn't what you want to call (see set/clear_extent_bit).
330 static int insert_state(struct extent_io_tree
*tree
,
331 struct extent_state
*state
, u64 start
, u64 end
,
334 struct rb_node
*node
;
337 printk(KERN_ERR
"btrfs end < start %llu %llu\n",
338 (unsigned long long)end
,
339 (unsigned long long)start
);
342 state
->start
= start
;
345 set_state_bits(tree
, state
, bits
);
347 node
= tree_insert(&tree
->state
, end
, &state
->rb_node
);
349 struct extent_state
*found
;
350 found
= rb_entry(node
, struct extent_state
, rb_node
);
351 printk(KERN_ERR
"btrfs found node %llu %llu on insert of "
352 "%llu %llu\n", (unsigned long long)found
->start
,
353 (unsigned long long)found
->end
,
354 (unsigned long long)start
, (unsigned long long)end
);
358 merge_state(tree
, state
);
362 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
365 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
366 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
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
;
404 static struct extent_state
*next_state(struct extent_state
*state
)
406 struct rb_node
*next
= rb_next(&state
->rb_node
);
408 return rb_entry(next
, struct extent_state
, rb_node
);
414 * utility function to clear some bits in an extent state struct.
415 * it will optionally wake up any one waiting on this state (wake == 1).
417 * If no bits are set on the state struct after clearing things, the
418 * struct is freed and removed from the tree
420 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
421 struct extent_state
*state
,
424 struct extent_state
*next
;
425 int bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
427 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
428 u64 range
= state
->end
- state
->start
+ 1;
429 WARN_ON(range
> tree
->dirty_bytes
);
430 tree
->dirty_bytes
-= range
;
432 clear_state_cb(tree
, state
, bits
);
433 state
->state
&= ~bits_to_clear
;
436 if (state
->state
== 0) {
437 next
= next_state(state
);
439 rb_erase(&state
->rb_node
, &tree
->state
);
441 free_extent_state(state
);
446 merge_state(tree
, state
);
447 next
= next_state(state
);
452 static struct extent_state
*
453 alloc_extent_state_atomic(struct extent_state
*prealloc
)
456 prealloc
= alloc_extent_state(GFP_ATOMIC
);
461 void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
463 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
464 "Extent tree was modified by another "
465 "thread while locked.");
469 * clear some bits on a range in the tree. This may require splitting
470 * or inserting elements in the tree, so the gfp mask is used to
471 * indicate which allocations or sleeping are allowed.
473 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
474 * the given range from the tree regardless of state (ie for truncate).
476 * the range [start, end] is inclusive.
478 * This takes the tree lock, and returns 0 on success and < 0 on error.
480 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
481 int bits
, int wake
, int delete,
482 struct extent_state
**cached_state
,
485 struct extent_state
*state
;
486 struct extent_state
*cached
;
487 struct extent_state
*prealloc
= NULL
;
488 struct rb_node
*node
;
494 bits
|= ~EXTENT_CTLBITS
;
495 bits
|= EXTENT_FIRST_DELALLOC
;
497 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
500 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
501 prealloc
= alloc_extent_state(mask
);
506 spin_lock(&tree
->lock
);
508 cached
= *cached_state
;
511 *cached_state
= NULL
;
515 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
516 cached
->end
> start
) {
518 atomic_dec(&cached
->refs
);
523 free_extent_state(cached
);
526 * this search will find the extents that end after
529 node
= tree_search(tree
, start
);
532 state
= rb_entry(node
, struct extent_state
, rb_node
);
534 if (state
->start
> end
)
536 WARN_ON(state
->end
< start
);
537 last_end
= state
->end
;
539 /* the state doesn't have the wanted bits, go ahead */
540 if (!(state
->state
& bits
)) {
541 state
= next_state(state
);
546 * | ---- desired range ---- |
548 * | ------------- state -------------- |
550 * We need to split the extent we found, and may flip
551 * bits on second half.
553 * If the extent we found extends past our range, we
554 * just split and search again. It'll get split again
555 * the next time though.
557 * If the extent we found is inside our range, we clear
558 * the desired bit on it.
561 if (state
->start
< start
) {
562 prealloc
= alloc_extent_state_atomic(prealloc
);
564 err
= split_state(tree
, state
, prealloc
, start
);
566 extent_io_tree_panic(tree
, err
);
571 if (state
->end
<= end
) {
572 state
= clear_state_bit(tree
, state
, &bits
, wake
);
578 * | ---- desired range ---- |
580 * We need to split the extent, and clear the bit
583 if (state
->start
<= end
&& state
->end
> end
) {
584 prealloc
= alloc_extent_state_atomic(prealloc
);
586 err
= split_state(tree
, state
, prealloc
, end
+ 1);
588 extent_io_tree_panic(tree
, err
);
593 clear_state_bit(tree
, prealloc
, &bits
, wake
);
599 state
= clear_state_bit(tree
, state
, &bits
, wake
);
601 if (last_end
== (u64
)-1)
603 start
= last_end
+ 1;
604 if (start
<= end
&& state
&& !need_resched())
609 spin_unlock(&tree
->lock
);
611 free_extent_state(prealloc
);
618 spin_unlock(&tree
->lock
);
619 if (mask
& __GFP_WAIT
)
624 static void wait_on_state(struct extent_io_tree
*tree
,
625 struct extent_state
*state
)
626 __releases(tree
->lock
)
627 __acquires(tree
->lock
)
630 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
631 spin_unlock(&tree
->lock
);
633 spin_lock(&tree
->lock
);
634 finish_wait(&state
->wq
, &wait
);
638 * waits for one or more bits to clear on a range in the state tree.
639 * The range [start, end] is inclusive.
640 * The tree lock is taken by this function
642 void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
)
644 struct extent_state
*state
;
645 struct rb_node
*node
;
647 spin_lock(&tree
->lock
);
651 * this search will find all the extents that end after
654 node
= tree_search(tree
, start
);
658 state
= rb_entry(node
, struct extent_state
, rb_node
);
660 if (state
->start
> end
)
663 if (state
->state
& bits
) {
664 start
= state
->start
;
665 atomic_inc(&state
->refs
);
666 wait_on_state(tree
, state
);
667 free_extent_state(state
);
670 start
= state
->end
+ 1;
675 cond_resched_lock(&tree
->lock
);
678 spin_unlock(&tree
->lock
);
681 static void set_state_bits(struct extent_io_tree
*tree
,
682 struct extent_state
*state
,
685 int bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
687 set_state_cb(tree
, state
, bits
);
688 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
689 u64 range
= state
->end
- state
->start
+ 1;
690 tree
->dirty_bytes
+= range
;
692 state
->state
|= bits_to_set
;
695 static void cache_state(struct extent_state
*state
,
696 struct extent_state
**cached_ptr
)
698 if (cached_ptr
&& !(*cached_ptr
)) {
699 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
)) {
701 atomic_inc(&state
->refs
);
706 static void uncache_state(struct extent_state
**cached_ptr
)
708 if (cached_ptr
&& (*cached_ptr
)) {
709 struct extent_state
*state
= *cached_ptr
;
711 free_extent_state(state
);
716 * set some bits on a range in the tree. This may require allocations or
717 * sleeping, so the gfp mask is used to indicate what is allowed.
719 * If any of the exclusive bits are set, this will fail with -EEXIST if some
720 * part of the range already has the desired bits set. The start of the
721 * existing range is returned in failed_start in this case.
723 * [start, end] is inclusive This takes the tree lock.
726 static int __must_check
727 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
728 int bits
, int exclusive_bits
, u64
*failed_start
,
729 struct extent_state
**cached_state
, gfp_t mask
)
731 struct extent_state
*state
;
732 struct extent_state
*prealloc
= NULL
;
733 struct rb_node
*node
;
738 bits
|= EXTENT_FIRST_DELALLOC
;
740 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
741 prealloc
= alloc_extent_state(mask
);
745 spin_lock(&tree
->lock
);
746 if (cached_state
&& *cached_state
) {
747 state
= *cached_state
;
748 if (state
->start
<= start
&& state
->end
> start
&&
750 node
= &state
->rb_node
;
755 * this search will find all the extents that end after
758 node
= tree_search(tree
, start
);
760 prealloc
= alloc_extent_state_atomic(prealloc
);
762 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
764 extent_io_tree_panic(tree
, err
);
769 state
= rb_entry(node
, struct extent_state
, rb_node
);
771 last_start
= state
->start
;
772 last_end
= state
->end
;
775 * | ---- desired range ---- |
778 * Just lock what we found and keep going
780 if (state
->start
== start
&& state
->end
<= end
) {
781 if (state
->state
& exclusive_bits
) {
782 *failed_start
= state
->start
;
787 set_state_bits(tree
, state
, &bits
);
788 cache_state(state
, cached_state
);
789 merge_state(tree
, state
);
790 if (last_end
== (u64
)-1)
792 start
= last_end
+ 1;
793 state
= next_state(state
);
794 if (start
< end
&& state
&& state
->start
== start
&&
801 * | ---- desired range ---- |
804 * | ------------- state -------------- |
806 * We need to split the extent we found, and may flip bits on
809 * If the extent we found extends past our
810 * range, we just split and search again. It'll get split
811 * again the next time though.
813 * If the extent we found is inside our range, we set the
816 if (state
->start
< start
) {
817 if (state
->state
& exclusive_bits
) {
818 *failed_start
= start
;
823 prealloc
= alloc_extent_state_atomic(prealloc
);
825 err
= split_state(tree
, state
, prealloc
, start
);
827 extent_io_tree_panic(tree
, err
);
832 if (state
->end
<= end
) {
833 set_state_bits(tree
, state
, &bits
);
834 cache_state(state
, cached_state
);
835 merge_state(tree
, state
);
836 if (last_end
== (u64
)-1)
838 start
= last_end
+ 1;
839 state
= next_state(state
);
840 if (start
< end
&& state
&& state
->start
== start
&&
847 * | ---- desired range ---- |
848 * | state | or | state |
850 * There's a hole, we need to insert something in it and
851 * ignore the extent we found.
853 if (state
->start
> start
) {
855 if (end
< last_start
)
858 this_end
= last_start
- 1;
860 prealloc
= alloc_extent_state_atomic(prealloc
);
864 * Avoid to free 'prealloc' if it can be merged with
867 err
= insert_state(tree
, prealloc
, start
, this_end
,
870 extent_io_tree_panic(tree
, err
);
872 cache_state(prealloc
, cached_state
);
874 start
= this_end
+ 1;
878 * | ---- desired range ---- |
880 * We need to split the extent, and set the bit
883 if (state
->start
<= end
&& state
->end
> end
) {
884 if (state
->state
& exclusive_bits
) {
885 *failed_start
= start
;
890 prealloc
= alloc_extent_state_atomic(prealloc
);
892 err
= split_state(tree
, state
, prealloc
, end
+ 1);
894 extent_io_tree_panic(tree
, err
);
896 set_state_bits(tree
, prealloc
, &bits
);
897 cache_state(prealloc
, cached_state
);
898 merge_state(tree
, prealloc
);
906 spin_unlock(&tree
->lock
);
908 free_extent_state(prealloc
);
915 spin_unlock(&tree
->lock
);
916 if (mask
& __GFP_WAIT
)
921 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
, int bits
,
922 u64
*failed_start
, struct extent_state
**cached_state
,
925 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
931 * convert_extent - convert all bits in a given range from one bit to another
932 * @tree: the io tree to search
933 * @start: the start offset in bytes
934 * @end: the end offset in bytes (inclusive)
935 * @bits: the bits to set in this range
936 * @clear_bits: the bits to clear in this range
937 * @mask: the allocation mask
939 * This will go through and set bits for the given range. If any states exist
940 * already in this range they are set with the given bit and cleared of the
941 * clear_bits. This is only meant to be used by things that are mergeable, ie
942 * converting from say DELALLOC to DIRTY. This is not meant to be used with
943 * boundary bits like LOCK.
945 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
946 int bits
, int clear_bits
, gfp_t mask
)
948 struct extent_state
*state
;
949 struct extent_state
*prealloc
= NULL
;
950 struct rb_node
*node
;
956 if (!prealloc
&& (mask
& __GFP_WAIT
)) {
957 prealloc
= alloc_extent_state(mask
);
962 spin_lock(&tree
->lock
);
964 * this search will find all the extents that end after
967 node
= tree_search(tree
, start
);
969 prealloc
= alloc_extent_state_atomic(prealloc
);
974 err
= insert_state(tree
, prealloc
, start
, end
, &bits
);
977 extent_io_tree_panic(tree
, err
);
980 state
= rb_entry(node
, struct extent_state
, rb_node
);
982 last_start
= state
->start
;
983 last_end
= state
->end
;
986 * | ---- desired range ---- |
989 * Just lock what we found and keep going
991 if (state
->start
== start
&& state
->end
<= end
) {
992 set_state_bits(tree
, state
, &bits
);
993 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
994 if (last_end
== (u64
)-1)
996 start
= last_end
+ 1;
997 if (start
< end
&& state
&& state
->start
== start
&&
1004 * | ---- desired range ---- |
1007 * | ------------- state -------------- |
1009 * We need to split the extent we found, and may flip bits on
1012 * If the extent we found extends past our
1013 * range, we just split and search again. It'll get split
1014 * again the next time though.
1016 * If the extent we found is inside our range, we set the
1017 * desired bit on it.
1019 if (state
->start
< start
) {
1020 prealloc
= alloc_extent_state_atomic(prealloc
);
1025 err
= split_state(tree
, state
, prealloc
, start
);
1027 extent_io_tree_panic(tree
, err
);
1031 if (state
->end
<= end
) {
1032 set_state_bits(tree
, state
, &bits
);
1033 state
= clear_state_bit(tree
, state
, &clear_bits
, 0);
1034 if (last_end
== (u64
)-1)
1036 start
= last_end
+ 1;
1037 if (start
< end
&& state
&& state
->start
== start
&&
1044 * | ---- desired range ---- |
1045 * | state | or | state |
1047 * There's a hole, we need to insert something in it and
1048 * ignore the extent we found.
1050 if (state
->start
> start
) {
1052 if (end
< last_start
)
1055 this_end
= last_start
- 1;
1057 prealloc
= alloc_extent_state_atomic(prealloc
);
1064 * Avoid to free 'prealloc' if it can be merged with
1067 err
= insert_state(tree
, prealloc
, start
, this_end
,
1070 extent_io_tree_panic(tree
, err
);
1072 start
= this_end
+ 1;
1076 * | ---- desired range ---- |
1078 * We need to split the extent, and set the bit
1081 if (state
->start
<= end
&& state
->end
> end
) {
1082 prealloc
= alloc_extent_state_atomic(prealloc
);
1088 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1090 extent_io_tree_panic(tree
, err
);
1092 set_state_bits(tree
, prealloc
, &bits
);
1093 clear_state_bit(tree
, prealloc
, &clear_bits
, 0);
1101 spin_unlock(&tree
->lock
);
1103 free_extent_state(prealloc
);
1110 spin_unlock(&tree
->lock
);
1111 if (mask
& __GFP_WAIT
)
1116 /* wrappers around set/clear extent bit */
1117 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1120 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, NULL
,
1124 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1125 int bits
, gfp_t mask
)
1127 return set_extent_bit(tree
, start
, end
, bits
, NULL
,
1131 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1132 int bits
, gfp_t mask
)
1134 return clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
);
1137 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1138 struct extent_state
**cached_state
, gfp_t mask
)
1140 return set_extent_bit(tree
, start
, end
,
1141 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1142 NULL
, cached_state
, mask
);
1145 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1148 return clear_extent_bit(tree
, start
, end
,
1149 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1150 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1153 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1156 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, NULL
,
1160 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1161 struct extent_state
**cached_state
, gfp_t mask
)
1163 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0,
1164 cached_state
, mask
);
1167 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1168 struct extent_state
**cached_state
, gfp_t mask
)
1170 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1171 cached_state
, mask
);
1175 * either insert or lock state struct between start and end use mask to tell
1176 * us if waiting is desired.
1178 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1179 int bits
, struct extent_state
**cached_state
)
1184 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1185 EXTENT_LOCKED
, &failed_start
,
1186 cached_state
, GFP_NOFS
);
1187 if (err
== -EEXIST
) {
1188 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1189 start
= failed_start
;
1192 WARN_ON(start
> end
);
1197 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1199 return lock_extent_bits(tree
, start
, end
, 0, NULL
);
1202 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1207 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1208 &failed_start
, NULL
, GFP_NOFS
);
1209 if (err
== -EEXIST
) {
1210 if (failed_start
> start
)
1211 clear_extent_bit(tree
, start
, failed_start
- 1,
1212 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1218 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1219 struct extent_state
**cached
, gfp_t mask
)
1221 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1225 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1227 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1232 * helper function to set both pages and extents in the tree writeback
1234 static int set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1236 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1237 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1240 while (index
<= end_index
) {
1241 page
= find_get_page(tree
->mapping
, index
);
1242 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1243 set_page_writeback(page
);
1244 page_cache_release(page
);
1250 /* find the first state struct with 'bits' set after 'start', and
1251 * return it. tree->lock must be held. NULL will returned if
1252 * nothing was found after 'start'
1254 struct extent_state
*find_first_extent_bit_state(struct extent_io_tree
*tree
,
1255 u64 start
, int bits
)
1257 struct rb_node
*node
;
1258 struct extent_state
*state
;
1261 * this search will find all the extents that end after
1264 node
= tree_search(tree
, start
);
1269 state
= rb_entry(node
, struct extent_state
, rb_node
);
1270 if (state
->end
>= start
&& (state
->state
& bits
))
1273 node
= rb_next(node
);
1282 * find the first offset in the io tree with 'bits' set. zero is
1283 * returned if we find something, and *start_ret and *end_ret are
1284 * set to reflect the state struct that was found.
1286 * If nothing was found, 1 is returned. If found something, return 0.
1288 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1289 u64
*start_ret
, u64
*end_ret
, int bits
)
1291 struct extent_state
*state
;
1294 spin_lock(&tree
->lock
);
1295 state
= find_first_extent_bit_state(tree
, start
, bits
);
1297 *start_ret
= state
->start
;
1298 *end_ret
= state
->end
;
1301 spin_unlock(&tree
->lock
);
1306 * find a contiguous range of bytes in the file marked as delalloc, not
1307 * more than 'max_bytes'. start and end are used to return the range,
1309 * 1 is returned if we find something, 0 if nothing was in the tree
1311 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1312 u64
*start
, u64
*end
, u64 max_bytes
,
1313 struct extent_state
**cached_state
)
1315 struct rb_node
*node
;
1316 struct extent_state
*state
;
1317 u64 cur_start
= *start
;
1319 u64 total_bytes
= 0;
1321 spin_lock(&tree
->lock
);
1324 * this search will find all the extents that end after
1327 node
= tree_search(tree
, cur_start
);
1335 state
= rb_entry(node
, struct extent_state
, rb_node
);
1336 if (found
&& (state
->start
!= cur_start
||
1337 (state
->state
& EXTENT_BOUNDARY
))) {
1340 if (!(state
->state
& EXTENT_DELALLOC
)) {
1346 *start
= state
->start
;
1347 *cached_state
= state
;
1348 atomic_inc(&state
->refs
);
1352 cur_start
= state
->end
+ 1;
1353 node
= rb_next(node
);
1356 total_bytes
+= state
->end
- state
->start
+ 1;
1357 if (total_bytes
>= max_bytes
)
1361 spin_unlock(&tree
->lock
);
1365 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1366 struct page
*locked_page
,
1370 struct page
*pages
[16];
1371 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1372 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1373 unsigned long nr_pages
= end_index
- index
+ 1;
1376 if (index
== locked_page
->index
&& end_index
== index
)
1379 while (nr_pages
> 0) {
1380 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1381 min_t(unsigned long, nr_pages
,
1382 ARRAY_SIZE(pages
)), pages
);
1383 for (i
= 0; i
< ret
; i
++) {
1384 if (pages
[i
] != locked_page
)
1385 unlock_page(pages
[i
]);
1386 page_cache_release(pages
[i
]);
1394 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1395 struct page
*locked_page
,
1399 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1400 unsigned long start_index
= index
;
1401 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1402 unsigned long pages_locked
= 0;
1403 struct page
*pages
[16];
1404 unsigned long nrpages
;
1408 /* the caller is responsible for locking the start index */
1409 if (index
== locked_page
->index
&& index
== end_index
)
1412 /* skip the page at the start index */
1413 nrpages
= end_index
- index
+ 1;
1414 while (nrpages
> 0) {
1415 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1416 min_t(unsigned long,
1417 nrpages
, ARRAY_SIZE(pages
)), pages
);
1422 /* now we have an array of pages, lock them all */
1423 for (i
= 0; i
< ret
; i
++) {
1425 * the caller is taking responsibility for
1428 if (pages
[i
] != locked_page
) {
1429 lock_page(pages
[i
]);
1430 if (!PageDirty(pages
[i
]) ||
1431 pages
[i
]->mapping
!= inode
->i_mapping
) {
1433 unlock_page(pages
[i
]);
1434 page_cache_release(pages
[i
]);
1438 page_cache_release(pages
[i
]);
1447 if (ret
&& pages_locked
) {
1448 __unlock_for_delalloc(inode
, locked_page
,
1450 ((u64
)(start_index
+ pages_locked
- 1)) <<
1457 * find a contiguous range of bytes in the file marked as delalloc, not
1458 * more than 'max_bytes'. start and end are used to return the range,
1460 * 1 is returned if we find something, 0 if nothing was in the tree
1462 static noinline u64
find_lock_delalloc_range(struct inode
*inode
,
1463 struct extent_io_tree
*tree
,
1464 struct page
*locked_page
,
1465 u64
*start
, u64
*end
,
1471 struct extent_state
*cached_state
= NULL
;
1476 /* step one, find a bunch of delalloc bytes starting at start */
1477 delalloc_start
= *start
;
1479 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1480 max_bytes
, &cached_state
);
1481 if (!found
|| delalloc_end
<= *start
) {
1482 *start
= delalloc_start
;
1483 *end
= delalloc_end
;
1484 free_extent_state(cached_state
);
1489 * start comes from the offset of locked_page. We have to lock
1490 * pages in order, so we can't process delalloc bytes before
1493 if (delalloc_start
< *start
)
1494 delalloc_start
= *start
;
1497 * make sure to limit the number of pages we try to lock down
1500 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
&& loops
)
1501 delalloc_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
1503 /* step two, lock all the pages after the page that has start */
1504 ret
= lock_delalloc_pages(inode
, locked_page
,
1505 delalloc_start
, delalloc_end
);
1506 if (ret
== -EAGAIN
) {
1507 /* some of the pages are gone, lets avoid looping by
1508 * shortening the size of the delalloc range we're searching
1510 free_extent_state(cached_state
);
1512 unsigned long offset
= (*start
) & (PAGE_CACHE_SIZE
- 1);
1513 max_bytes
= PAGE_CACHE_SIZE
- offset
;
1521 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1523 /* step three, lock the state bits for the whole range */
1524 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1526 /* then test to make sure it is all still delalloc */
1527 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1528 EXTENT_DELALLOC
, 1, cached_state
);
1530 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1531 &cached_state
, GFP_NOFS
);
1532 __unlock_for_delalloc(inode
, locked_page
,
1533 delalloc_start
, delalloc_end
);
1537 free_extent_state(cached_state
);
1538 *start
= delalloc_start
;
1539 *end
= delalloc_end
;
1544 int extent_clear_unlock_delalloc(struct inode
*inode
,
1545 struct extent_io_tree
*tree
,
1546 u64 start
, u64 end
, struct page
*locked_page
,
1550 struct page
*pages
[16];
1551 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1552 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1553 unsigned long nr_pages
= end_index
- index
+ 1;
1557 if (op
& EXTENT_CLEAR_UNLOCK
)
1558 clear_bits
|= EXTENT_LOCKED
;
1559 if (op
& EXTENT_CLEAR_DIRTY
)
1560 clear_bits
|= EXTENT_DIRTY
;
1562 if (op
& EXTENT_CLEAR_DELALLOC
)
1563 clear_bits
|= EXTENT_DELALLOC
;
1565 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1566 if (!(op
& (EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
1567 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
|
1568 EXTENT_SET_PRIVATE2
)))
1571 while (nr_pages
> 0) {
1572 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1573 min_t(unsigned long,
1574 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1575 for (i
= 0; i
< ret
; i
++) {
1577 if (op
& EXTENT_SET_PRIVATE2
)
1578 SetPagePrivate2(pages
[i
]);
1580 if (pages
[i
] == locked_page
) {
1581 page_cache_release(pages
[i
]);
1584 if (op
& EXTENT_CLEAR_DIRTY
)
1585 clear_page_dirty_for_io(pages
[i
]);
1586 if (op
& EXTENT_SET_WRITEBACK
)
1587 set_page_writeback(pages
[i
]);
1588 if (op
& EXTENT_END_WRITEBACK
)
1589 end_page_writeback(pages
[i
]);
1590 if (op
& EXTENT_CLEAR_UNLOCK_PAGE
)
1591 unlock_page(pages
[i
]);
1592 page_cache_release(pages
[i
]);
1602 * count the number of bytes in the tree that have a given bit(s)
1603 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1604 * cached. The total number found is returned.
1606 u64
count_range_bits(struct extent_io_tree
*tree
,
1607 u64
*start
, u64 search_end
, u64 max_bytes
,
1608 unsigned long bits
, int contig
)
1610 struct rb_node
*node
;
1611 struct extent_state
*state
;
1612 u64 cur_start
= *start
;
1613 u64 total_bytes
= 0;
1617 if (search_end
<= cur_start
) {
1622 spin_lock(&tree
->lock
);
1623 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1624 total_bytes
= tree
->dirty_bytes
;
1628 * this search will find all the extents that end after
1631 node
= tree_search(tree
, cur_start
);
1636 state
= rb_entry(node
, struct extent_state
, rb_node
);
1637 if (state
->start
> search_end
)
1639 if (contig
&& found
&& state
->start
> last
+ 1)
1641 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1642 total_bytes
+= min(search_end
, state
->end
) + 1 -
1643 max(cur_start
, state
->start
);
1644 if (total_bytes
>= max_bytes
)
1647 *start
= max(cur_start
, state
->start
);
1651 } else if (contig
&& found
) {
1654 node
= rb_next(node
);
1659 spin_unlock(&tree
->lock
);
1664 * set the private field for a given byte offset in the tree. If there isn't
1665 * an extent_state there already, this does nothing.
1667 int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1669 struct rb_node
*node
;
1670 struct extent_state
*state
;
1673 spin_lock(&tree
->lock
);
1675 * this search will find all the extents that end after
1678 node
= tree_search(tree
, start
);
1683 state
= rb_entry(node
, struct extent_state
, rb_node
);
1684 if (state
->start
!= start
) {
1688 state
->private = private;
1690 spin_unlock(&tree
->lock
);
1694 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1696 struct rb_node
*node
;
1697 struct extent_state
*state
;
1700 spin_lock(&tree
->lock
);
1702 * this search will find all the extents that end after
1705 node
= tree_search(tree
, start
);
1710 state
= rb_entry(node
, struct extent_state
, rb_node
);
1711 if (state
->start
!= start
) {
1715 *private = state
->private;
1717 spin_unlock(&tree
->lock
);
1722 * searches a range in the state tree for a given mask.
1723 * If 'filled' == 1, this returns 1 only if every extent in the tree
1724 * has the bits set. Otherwise, 1 is returned if any bit in the
1725 * range is found set.
1727 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1728 int bits
, int filled
, struct extent_state
*cached
)
1730 struct extent_state
*state
= NULL
;
1731 struct rb_node
*node
;
1734 spin_lock(&tree
->lock
);
1735 if (cached
&& cached
->tree
&& cached
->start
<= start
&&
1736 cached
->end
> start
)
1737 node
= &cached
->rb_node
;
1739 node
= tree_search(tree
, start
);
1740 while (node
&& start
<= end
) {
1741 state
= rb_entry(node
, struct extent_state
, rb_node
);
1743 if (filled
&& state
->start
> start
) {
1748 if (state
->start
> end
)
1751 if (state
->state
& bits
) {
1755 } else if (filled
) {
1760 if (state
->end
== (u64
)-1)
1763 start
= state
->end
+ 1;
1766 node
= rb_next(node
);
1773 spin_unlock(&tree
->lock
);
1778 * helper function to set a given page up to date if all the
1779 * extents in the tree for that page are up to date
1781 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1783 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1784 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1785 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1786 SetPageUptodate(page
);
1790 * helper function to unlock a page if all the extents in the tree
1791 * for that page are unlocked
1793 static void check_page_locked(struct extent_io_tree
*tree
, struct page
*page
)
1795 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
1796 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1797 if (!test_range_bit(tree
, start
, end
, EXTENT_LOCKED
, 0, NULL
))
1802 * helper function to end page writeback if all the extents
1803 * in the tree for that page are done with writeback
1805 static void check_page_writeback(struct extent_io_tree
*tree
,
1808 end_page_writeback(page
);
1812 * When IO fails, either with EIO or csum verification fails, we
1813 * try other mirrors that might have a good copy of the data. This
1814 * io_failure_record is used to record state as we go through all the
1815 * mirrors. If another mirror has good data, the page is set up to date
1816 * and things continue. If a good mirror can't be found, the original
1817 * bio end_io callback is called to indicate things have failed.
1819 struct io_failure_record
{
1824 unsigned long bio_flags
;
1830 static int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
,
1835 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1837 set_state_private(failure_tree
, rec
->start
, 0);
1838 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1839 rec
->start
+ rec
->len
- 1,
1840 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1845 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1846 rec
->start
+ rec
->len
- 1,
1847 EXTENT_DAMAGED
, GFP_NOFS
);
1856 static void repair_io_failure_callback(struct bio
*bio
, int err
)
1858 complete(bio
->bi_private
);
1862 * this bypasses the standard btrfs submit functions deliberately, as
1863 * the standard behavior is to write all copies in a raid setup. here we only
1864 * want to write the one bad copy. so we do the mapping for ourselves and issue
1865 * submit_bio directly.
1866 * to avoid any synchonization issues, wait for the data after writing, which
1867 * actually prevents the read that triggered the error from finishing.
1868 * currently, there can be no more than two copies of every data bit. thus,
1869 * exactly one rewrite is required.
1871 int repair_io_failure(struct btrfs_mapping_tree
*map_tree
, u64 start
,
1872 u64 length
, u64 logical
, struct page
*page
,
1876 struct btrfs_device
*dev
;
1877 DECLARE_COMPLETION_ONSTACK(compl);
1880 struct btrfs_bio
*bbio
= NULL
;
1883 BUG_ON(!mirror_num
);
1885 bio
= bio_alloc(GFP_NOFS
, 1);
1888 bio
->bi_private
= &compl;
1889 bio
->bi_end_io
= repair_io_failure_callback
;
1891 map_length
= length
;
1893 ret
= btrfs_map_block(map_tree
, WRITE
, logical
,
1894 &map_length
, &bbio
, mirror_num
);
1899 BUG_ON(mirror_num
!= bbio
->mirror_num
);
1900 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
1901 bio
->bi_sector
= sector
;
1902 dev
= bbio
->stripes
[mirror_num
-1].dev
;
1904 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
1908 bio
->bi_bdev
= dev
->bdev
;
1909 bio_add_page(bio
, page
, length
, start
-page_offset(page
));
1910 btrfsic_submit_bio(WRITE_SYNC
, bio
);
1911 wait_for_completion(&compl);
1913 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
1914 /* try to remap that extent elsewhere? */
1916 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
1920 printk(KERN_INFO
"btrfs read error corrected: ino %lu off %llu (dev %s "
1921 "sector %llu)\n", page
->mapping
->host
->i_ino
, start
,
1928 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
1931 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
1932 u64 start
= eb
->start
;
1933 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
1936 for (i
= 0; i
< num_pages
; i
++) {
1937 struct page
*p
= extent_buffer_page(eb
, i
);
1938 ret
= repair_io_failure(map_tree
, start
, PAGE_CACHE_SIZE
,
1939 start
, p
, mirror_num
);
1942 start
+= PAGE_CACHE_SIZE
;
1949 * each time an IO finishes, we do a fast check in the IO failure tree
1950 * to see if we need to process or clean up an io_failure_record
1952 static int clean_io_failure(u64 start
, struct page
*page
)
1955 u64 private_failure
;
1956 struct io_failure_record
*failrec
;
1957 struct btrfs_mapping_tree
*map_tree
;
1958 struct extent_state
*state
;
1962 struct inode
*inode
= page
->mapping
->host
;
1965 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1966 (u64
)-1, 1, EXTENT_DIRTY
, 0);
1970 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
1975 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
1976 BUG_ON(!failrec
->this_mirror
);
1978 if (failrec
->in_validation
) {
1979 /* there was no real error, just free the record */
1980 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1986 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1987 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1990 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1992 if (state
&& state
->start
== failrec
->start
) {
1993 map_tree
= &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
;
1994 num_copies
= btrfs_num_copies(map_tree
, failrec
->logical
,
1996 if (num_copies
> 1) {
1997 ret
= repair_io_failure(map_tree
, start
, failrec
->len
,
1998 failrec
->logical
, page
,
1999 failrec
->failed_mirror
);
2006 ret
= free_io_failure(inode
, failrec
, did_repair
);
2012 * this is a generic handler for readpage errors (default
2013 * readpage_io_failed_hook). if other copies exist, read those and write back
2014 * good data to the failed position. does not investigate in remapping the
2015 * failed extent elsewhere, hoping the device will be smart enough to do this as
2019 static int bio_readpage_error(struct bio
*failed_bio
, struct page
*page
,
2020 u64 start
, u64 end
, int failed_mirror
,
2021 struct extent_state
*state
)
2023 struct io_failure_record
*failrec
= NULL
;
2025 struct extent_map
*em
;
2026 struct inode
*inode
= page
->mapping
->host
;
2027 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2028 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2029 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2036 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2038 ret
= get_state_private(failure_tree
, start
, &private);
2040 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2043 failrec
->start
= start
;
2044 failrec
->len
= end
- start
+ 1;
2045 failrec
->this_mirror
= 0;
2046 failrec
->bio_flags
= 0;
2047 failrec
->in_validation
= 0;
2049 read_lock(&em_tree
->lock
);
2050 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2052 read_unlock(&em_tree
->lock
);
2057 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
2058 free_extent_map(em
);
2061 read_unlock(&em_tree
->lock
);
2063 if (!em
|| IS_ERR(em
)) {
2067 logical
= start
- em
->start
;
2068 logical
= em
->block_start
+ logical
;
2069 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2070 logical
= em
->block_start
;
2071 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2072 extent_set_compress_type(&failrec
->bio_flags
,
2075 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2076 "len=%llu\n", logical
, start
, failrec
->len
);
2077 failrec
->logical
= logical
;
2078 free_extent_map(em
);
2080 /* set the bits in the private failure tree */
2081 ret
= set_extent_bits(failure_tree
, start
, end
,
2082 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2084 ret
= set_state_private(failure_tree
, start
,
2085 (u64
)(unsigned long)failrec
);
2086 /* set the bits in the inode's tree */
2088 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2095 failrec
= (struct io_failure_record
*)(unsigned long)private;
2096 pr_debug("bio_readpage_error: (found) logical=%llu, "
2097 "start=%llu, len=%llu, validation=%d\n",
2098 failrec
->logical
, failrec
->start
, failrec
->len
,
2099 failrec
->in_validation
);
2101 * when data can be on disk more than twice, add to failrec here
2102 * (e.g. with a list for failed_mirror) to make
2103 * clean_io_failure() clean all those errors at once.
2106 num_copies
= btrfs_num_copies(
2107 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
2108 failrec
->logical
, failrec
->len
);
2109 if (num_copies
== 1) {
2111 * we only have a single copy of the data, so don't bother with
2112 * all the retry and error correction code that follows. no
2113 * matter what the error is, it is very likely to persist.
2115 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2116 "state=%p, num_copies=%d, next_mirror %d, "
2117 "failed_mirror %d\n", state
, num_copies
,
2118 failrec
->this_mirror
, failed_mirror
);
2119 free_io_failure(inode
, failrec
, 0);
2124 spin_lock(&tree
->lock
);
2125 state
= find_first_extent_bit_state(tree
, failrec
->start
,
2127 if (state
&& state
->start
!= failrec
->start
)
2129 spin_unlock(&tree
->lock
);
2133 * there are two premises:
2134 * a) deliver good data to the caller
2135 * b) correct the bad sectors on disk
2137 if (failed_bio
->bi_vcnt
> 1) {
2139 * to fulfill b), we need to know the exact failing sectors, as
2140 * we don't want to rewrite any more than the failed ones. thus,
2141 * we need separate read requests for the failed bio
2143 * if the following BUG_ON triggers, our validation request got
2144 * merged. we need separate requests for our algorithm to work.
2146 BUG_ON(failrec
->in_validation
);
2147 failrec
->in_validation
= 1;
2148 failrec
->this_mirror
= failed_mirror
;
2149 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2152 * we're ready to fulfill a) and b) alongside. get a good copy
2153 * of the failed sector and if we succeed, we have setup
2154 * everything for repair_io_failure to do the rest for us.
2156 if (failrec
->in_validation
) {
2157 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2158 failrec
->in_validation
= 0;
2159 failrec
->this_mirror
= 0;
2161 failrec
->failed_mirror
= failed_mirror
;
2162 failrec
->this_mirror
++;
2163 if (failrec
->this_mirror
== failed_mirror
)
2164 failrec
->this_mirror
++;
2165 read_mode
= READ_SYNC
;
2168 if (!state
|| failrec
->this_mirror
> num_copies
) {
2169 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2170 "next_mirror %d, failed_mirror %d\n", state
,
2171 num_copies
, failrec
->this_mirror
, failed_mirror
);
2172 free_io_failure(inode
, failrec
, 0);
2176 bio
= bio_alloc(GFP_NOFS
, 1);
2178 free_io_failure(inode
, failrec
, 0);
2181 bio
->bi_private
= state
;
2182 bio
->bi_end_io
= failed_bio
->bi_end_io
;
2183 bio
->bi_sector
= failrec
->logical
>> 9;
2184 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2187 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
2189 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2190 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode
,
2191 failrec
->this_mirror
, num_copies
, failrec
->in_validation
);
2193 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2194 failrec
->this_mirror
,
2195 failrec
->bio_flags
, 0);
2199 /* lots and lots of room for performance fixes in the end_bio funcs */
2201 int end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2203 int uptodate
= (err
== 0);
2204 struct extent_io_tree
*tree
;
2207 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2209 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2210 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2211 end
, NULL
, uptodate
);
2217 ClearPageUptodate(page
);
2224 * after a writepage IO is done, we need to:
2225 * clear the uptodate bits on error
2226 * clear the writeback bits in the extent tree for this IO
2227 * end_page_writeback if the page has no more pending IO
2229 * Scheduling is not allowed, so the extent state tree is expected
2230 * to have one and only one object corresponding to this IO.
2232 static void end_bio_extent_writepage(struct bio
*bio
, int err
)
2234 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2235 struct extent_io_tree
*tree
;
2241 struct page
*page
= bvec
->bv_page
;
2242 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2244 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
2246 end
= start
+ bvec
->bv_len
- 1;
2248 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
2253 if (--bvec
>= bio
->bi_io_vec
)
2254 prefetchw(&bvec
->bv_page
->flags
);
2256 if (end_extent_writepage(page
, err
, start
, end
))
2260 end_page_writeback(page
);
2262 check_page_writeback(tree
, page
);
2263 } while (bvec
>= bio
->bi_io_vec
);
2269 * after a readpage IO is done, we need to:
2270 * clear the uptodate bits on error
2271 * set the uptodate bits if things worked
2272 * set the page up to date if all extents in the tree are uptodate
2273 * clear the lock bit in the extent tree
2274 * unlock the page if there are no other extents locked for it
2276 * Scheduling is not allowed, so the extent state tree is expected
2277 * to have one and only one object corresponding to this IO.
2279 static void end_bio_extent_readpage(struct bio
*bio
, int err
)
2281 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2282 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2283 struct bio_vec
*bvec
= bio
->bi_io_vec
;
2284 struct extent_io_tree
*tree
;
2295 struct page
*page
= bvec
->bv_page
;
2296 struct extent_state
*cached
= NULL
;
2297 struct extent_state
*state
;
2299 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2300 "mirror=%ld\n", bio
->bi_vcnt
, bio
->bi_idx
, err
,
2301 (long int)bio
->bi_bdev
);
2302 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2304 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) +
2306 end
= start
+ bvec
->bv_len
- 1;
2308 if (bvec
->bv_offset
== 0 && bvec
->bv_len
== PAGE_CACHE_SIZE
)
2313 if (++bvec
<= bvec_end
)
2314 prefetchw(&bvec
->bv_page
->flags
);
2316 spin_lock(&tree
->lock
);
2317 state
= find_first_extent_bit_state(tree
, start
, EXTENT_LOCKED
);
2318 if (state
&& state
->start
== start
) {
2320 * take a reference on the state, unlock will drop
2323 cache_state(state
, &cached
);
2325 spin_unlock(&tree
->lock
);
2327 mirror
= (int)(unsigned long)bio
->bi_bdev
;
2328 if (uptodate
&& tree
->ops
&& tree
->ops
->readpage_end_io_hook
) {
2329 ret
= tree
->ops
->readpage_end_io_hook(page
, start
, end
,
2332 /* no IO indicated but software detected errors
2333 * in the block, either checksum errors or
2334 * issues with the contents */
2335 struct btrfs_root
*root
=
2336 BTRFS_I(page
->mapping
->host
)->root
;
2337 struct btrfs_device
*device
;
2340 device
= btrfs_find_device_for_logical(
2341 root
, start
, mirror
);
2343 btrfs_dev_stat_inc_and_print(device
,
2344 BTRFS_DEV_STAT_CORRUPTION_ERRS
);
2346 clean_io_failure(start
, page
);
2350 if (!uptodate
&& tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2351 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2353 test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2355 } else if (!uptodate
) {
2357 * The generic bio_readpage_error handles errors the
2358 * following way: If possible, new read requests are
2359 * created and submitted and will end up in
2360 * end_bio_extent_readpage as well (if we're lucky, not
2361 * in the !uptodate case). In that case it returns 0 and
2362 * we just go on with the next page in our bio. If it
2363 * can't handle the error it will return -EIO and we
2364 * remain responsible for that page.
2366 ret
= bio_readpage_error(bio
, page
, start
, end
, mirror
, NULL
);
2369 test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2372 uncache_state(&cached
);
2377 if (uptodate
&& tree
->track_uptodate
) {
2378 set_extent_uptodate(tree
, start
, end
, &cached
,
2381 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2385 SetPageUptodate(page
);
2387 ClearPageUptodate(page
);
2393 check_page_uptodate(tree
, page
);
2395 ClearPageUptodate(page
);
2398 check_page_locked(tree
, page
);
2400 } while (bvec
<= bvec_end
);
2406 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2411 bio
= bio_alloc(gfp_flags
, nr_vecs
);
2413 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2414 while (!bio
&& (nr_vecs
/= 2))
2415 bio
= bio_alloc(gfp_flags
, nr_vecs
);
2420 bio
->bi_bdev
= bdev
;
2421 bio
->bi_sector
= first_sector
;
2427 * Since writes are async, they will only return -ENOMEM.
2428 * Reads can return the full range of I/O error conditions.
2430 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2431 int mirror_num
, unsigned long bio_flags
)
2434 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2435 struct page
*page
= bvec
->bv_page
;
2436 struct extent_io_tree
*tree
= bio
->bi_private
;
2439 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + bvec
->bv_offset
;
2441 bio
->bi_private
= NULL
;
2445 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2446 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2447 mirror_num
, bio_flags
, start
);
2449 btrfsic_submit_bio(rw
, bio
);
2451 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
2457 static int merge_bio(struct extent_io_tree
*tree
, struct page
*page
,
2458 unsigned long offset
, size_t size
, struct bio
*bio
,
2459 unsigned long bio_flags
)
2462 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2463 ret
= tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
,
2470 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2471 struct page
*page
, sector_t sector
,
2472 size_t size
, unsigned long offset
,
2473 struct block_device
*bdev
,
2474 struct bio
**bio_ret
,
2475 unsigned long max_pages
,
2476 bio_end_io_t end_io_func
,
2478 unsigned long prev_bio_flags
,
2479 unsigned long bio_flags
)
2485 int this_compressed
= bio_flags
& EXTENT_BIO_COMPRESSED
;
2486 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2487 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2489 if (bio_ret
&& *bio_ret
) {
2492 contig
= bio
->bi_sector
== sector
;
2494 contig
= bio
->bi_sector
+ (bio
->bi_size
>> 9) ==
2497 if (prev_bio_flags
!= bio_flags
|| !contig
||
2498 merge_bio(tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2499 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2500 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2509 if (this_compressed
)
2512 nr
= bio_get_nr_vecs(bdev
);
2514 bio
= btrfs_bio_alloc(bdev
, sector
, nr
, GFP_NOFS
| __GFP_HIGH
);
2518 bio_add_page(bio
, page
, page_size
, offset
);
2519 bio
->bi_end_io
= end_io_func
;
2520 bio
->bi_private
= tree
;
2525 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2530 void attach_extent_buffer_page(struct extent_buffer
*eb
, struct page
*page
)
2532 if (!PagePrivate(page
)) {
2533 SetPagePrivate(page
);
2534 page_cache_get(page
);
2535 set_page_private(page
, (unsigned long)eb
);
2537 WARN_ON(page
->private != (unsigned long)eb
);
2541 void set_page_extent_mapped(struct page
*page
)
2543 if (!PagePrivate(page
)) {
2544 SetPagePrivate(page
);
2545 page_cache_get(page
);
2546 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2551 * basic readpage implementation. Locked extent state structs are inserted
2552 * into the tree that are removed when the IO is done (by the end_io
2554 * XXX JDM: This needs looking at to ensure proper page locking
2556 static int __extent_read_full_page(struct extent_io_tree
*tree
,
2558 get_extent_t
*get_extent
,
2559 struct bio
**bio
, int mirror_num
,
2560 unsigned long *bio_flags
)
2562 struct inode
*inode
= page
->mapping
->host
;
2563 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2564 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2568 u64 last_byte
= i_size_read(inode
);
2572 struct extent_map
*em
;
2573 struct block_device
*bdev
;
2574 struct btrfs_ordered_extent
*ordered
;
2577 size_t pg_offset
= 0;
2579 size_t disk_io_size
;
2580 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2581 unsigned long this_bio_flag
= 0;
2583 set_page_extent_mapped(page
);
2585 if (!PageUptodate(page
)) {
2586 if (cleancache_get_page(page
) == 0) {
2587 BUG_ON(blocksize
!= PAGE_SIZE
);
2594 lock_extent(tree
, start
, end
);
2595 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
2598 unlock_extent(tree
, start
, end
);
2599 btrfs_start_ordered_extent(inode
, ordered
, 1);
2600 btrfs_put_ordered_extent(ordered
);
2603 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2605 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2608 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2609 userpage
= kmap_atomic(page
);
2610 memset(userpage
+ zero_offset
, 0, iosize
);
2611 flush_dcache_page(page
);
2612 kunmap_atomic(userpage
);
2615 while (cur
<= end
) {
2616 if (cur
>= last_byte
) {
2618 struct extent_state
*cached
= NULL
;
2620 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2621 userpage
= kmap_atomic(page
);
2622 memset(userpage
+ pg_offset
, 0, iosize
);
2623 flush_dcache_page(page
);
2624 kunmap_atomic(userpage
);
2625 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2627 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2631 em
= get_extent(inode
, page
, pg_offset
, cur
,
2633 if (IS_ERR_OR_NULL(em
)) {
2635 unlock_extent(tree
, cur
, end
);
2638 extent_offset
= cur
- em
->start
;
2639 BUG_ON(extent_map_end(em
) <= cur
);
2642 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2643 this_bio_flag
= EXTENT_BIO_COMPRESSED
;
2644 extent_set_compress_type(&this_bio_flag
,
2648 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2649 cur_end
= min(extent_map_end(em
) - 1, end
);
2650 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2651 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2652 disk_io_size
= em
->block_len
;
2653 sector
= em
->block_start
>> 9;
2655 sector
= (em
->block_start
+ extent_offset
) >> 9;
2656 disk_io_size
= iosize
;
2659 block_start
= em
->block_start
;
2660 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2661 block_start
= EXTENT_MAP_HOLE
;
2662 free_extent_map(em
);
2665 /* we've found a hole, just zero and go on */
2666 if (block_start
== EXTENT_MAP_HOLE
) {
2668 struct extent_state
*cached
= NULL
;
2670 userpage
= kmap_atomic(page
);
2671 memset(userpage
+ pg_offset
, 0, iosize
);
2672 flush_dcache_page(page
);
2673 kunmap_atomic(userpage
);
2675 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2677 unlock_extent_cached(tree
, cur
, cur
+ iosize
- 1,
2680 pg_offset
+= iosize
;
2683 /* the get_extent function already copied into the page */
2684 if (test_range_bit(tree
, cur
, cur_end
,
2685 EXTENT_UPTODATE
, 1, NULL
)) {
2686 check_page_uptodate(tree
, page
);
2687 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2689 pg_offset
+= iosize
;
2692 /* we have an inline extent but it didn't get marked up
2693 * to date. Error out
2695 if (block_start
== EXTENT_MAP_INLINE
) {
2697 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
2699 pg_offset
+= iosize
;
2704 if (tree
->ops
&& tree
->ops
->readpage_io_hook
) {
2705 ret
= tree
->ops
->readpage_io_hook(page
, cur
,
2709 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2711 ret
= submit_extent_page(READ
, tree
, page
,
2712 sector
, disk_io_size
, pg_offset
,
2714 end_bio_extent_readpage
, mirror_num
,
2717 BUG_ON(ret
== -ENOMEM
);
2719 *bio_flags
= this_bio_flag
;
2724 pg_offset
+= iosize
;
2728 if (!PageError(page
))
2729 SetPageUptodate(page
);
2735 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
2736 get_extent_t
*get_extent
, int mirror_num
)
2738 struct bio
*bio
= NULL
;
2739 unsigned long bio_flags
= 0;
2742 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
2745 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
2749 static noinline
void update_nr_written(struct page
*page
,
2750 struct writeback_control
*wbc
,
2751 unsigned long nr_written
)
2753 wbc
->nr_to_write
-= nr_written
;
2754 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
2755 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
2756 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
2760 * the writepage semantics are similar to regular writepage. extent
2761 * records are inserted to lock ranges in the tree, and as dirty areas
2762 * are found, they are marked writeback. Then the lock bits are removed
2763 * and the end_io handler clears the writeback ranges
2765 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
2768 struct inode
*inode
= page
->mapping
->host
;
2769 struct extent_page_data
*epd
= data
;
2770 struct extent_io_tree
*tree
= epd
->tree
;
2771 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2773 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2777 u64 last_byte
= i_size_read(inode
);
2781 struct extent_state
*cached_state
= NULL
;
2782 struct extent_map
*em
;
2783 struct block_device
*bdev
;
2786 size_t pg_offset
= 0;
2788 loff_t i_size
= i_size_read(inode
);
2789 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2795 unsigned long nr_written
= 0;
2796 bool fill_delalloc
= true;
2798 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2799 write_flags
= WRITE_SYNC
;
2801 write_flags
= WRITE
;
2803 trace___extent_writepage(page
, inode
, wbc
);
2805 WARN_ON(!PageLocked(page
));
2807 ClearPageError(page
);
2809 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
2810 if (page
->index
> end_index
||
2811 (page
->index
== end_index
&& !pg_offset
)) {
2812 page
->mapping
->a_ops
->invalidatepage(page
, 0);
2817 if (page
->index
== end_index
) {
2820 userpage
= kmap_atomic(page
);
2821 memset(userpage
+ pg_offset
, 0,
2822 PAGE_CACHE_SIZE
- pg_offset
);
2823 kunmap_atomic(userpage
);
2824 flush_dcache_page(page
);
2828 set_page_extent_mapped(page
);
2830 if (!tree
->ops
|| !tree
->ops
->fill_delalloc
)
2831 fill_delalloc
= false;
2833 delalloc_start
= start
;
2836 if (!epd
->extent_locked
&& fill_delalloc
) {
2837 u64 delalloc_to_write
= 0;
2839 * make sure the wbc mapping index is at least updated
2842 update_nr_written(page
, wbc
, 0);
2844 while (delalloc_end
< page_end
) {
2845 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
2850 if (nr_delalloc
== 0) {
2851 delalloc_start
= delalloc_end
+ 1;
2854 ret
= tree
->ops
->fill_delalloc(inode
, page
,
2859 /* File system has been set read-only */
2865 * delalloc_end is already one less than the total
2866 * length, so we don't subtract one from
2869 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
2872 delalloc_start
= delalloc_end
+ 1;
2874 if (wbc
->nr_to_write
< delalloc_to_write
) {
2877 if (delalloc_to_write
< thresh
* 2)
2878 thresh
= delalloc_to_write
;
2879 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
2883 /* did the fill delalloc function already unlock and start
2889 * we've unlocked the page, so we can't update
2890 * the mapping's writeback index, just update
2893 wbc
->nr_to_write
-= nr_written
;
2897 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
2898 ret
= tree
->ops
->writepage_start_hook(page
, start
,
2901 /* Fixup worker will requeue */
2903 wbc
->pages_skipped
++;
2905 redirty_page_for_writepage(wbc
, page
);
2906 update_nr_written(page
, wbc
, nr_written
);
2914 * we don't want to touch the inode after unlocking the page,
2915 * so we update the mapping writeback index now
2917 update_nr_written(page
, wbc
, nr_written
+ 1);
2920 if (last_byte
<= start
) {
2921 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2922 tree
->ops
->writepage_end_io_hook(page
, start
,
2927 blocksize
= inode
->i_sb
->s_blocksize
;
2929 while (cur
<= end
) {
2930 if (cur
>= last_byte
) {
2931 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2932 tree
->ops
->writepage_end_io_hook(page
, cur
,
2936 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
2938 if (IS_ERR_OR_NULL(em
)) {
2943 extent_offset
= cur
- em
->start
;
2944 BUG_ON(extent_map_end(em
) <= cur
);
2946 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2947 iosize
= (iosize
+ blocksize
- 1) & ~((u64
)blocksize
- 1);
2948 sector
= (em
->block_start
+ extent_offset
) >> 9;
2950 block_start
= em
->block_start
;
2951 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
2952 free_extent_map(em
);
2956 * compressed and inline extents are written through other
2959 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
2960 block_start
== EXTENT_MAP_INLINE
) {
2962 * end_io notification does not happen here for
2963 * compressed extents
2965 if (!compressed
&& tree
->ops
&&
2966 tree
->ops
->writepage_end_io_hook
)
2967 tree
->ops
->writepage_end_io_hook(page
, cur
,
2970 else if (compressed
) {
2971 /* we don't want to end_page_writeback on
2972 * a compressed extent. this happens
2979 pg_offset
+= iosize
;
2982 /* leave this out until we have a page_mkwrite call */
2983 if (0 && !test_range_bit(tree
, cur
, cur
+ iosize
- 1,
2984 EXTENT_DIRTY
, 0, NULL
)) {
2986 pg_offset
+= iosize
;
2990 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
2991 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
2999 unsigned long max_nr
= end_index
+ 1;
3001 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3002 if (!PageWriteback(page
)) {
3003 printk(KERN_ERR
"btrfs warning page %lu not "
3004 "writeback, cur %llu end %llu\n",
3005 page
->index
, (unsigned long long)cur
,
3006 (unsigned long long)end
);
3009 ret
= submit_extent_page(write_flags
, tree
, page
,
3010 sector
, iosize
, pg_offset
,
3011 bdev
, &epd
->bio
, max_nr
,
3012 end_bio_extent_writepage
,
3018 pg_offset
+= iosize
;
3023 /* make sure the mapping tag for page dirty gets cleared */
3024 set_page_writeback(page
);
3025 end_page_writeback(page
);
3031 /* drop our reference on any cached states */
3032 free_extent_state(cached_state
);
3036 static int eb_wait(void *word
)
3042 static void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3044 wait_on_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
, eb_wait
,
3045 TASK_UNINTERRUPTIBLE
);
3048 static int lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3049 struct btrfs_fs_info
*fs_info
,
3050 struct extent_page_data
*epd
)
3052 unsigned long i
, num_pages
;
3056 if (!btrfs_try_tree_write_lock(eb
)) {
3058 flush_write_bio(epd
);
3059 btrfs_tree_lock(eb
);
3062 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3063 btrfs_tree_unlock(eb
);
3067 flush_write_bio(epd
);
3071 wait_on_extent_buffer_writeback(eb
);
3072 btrfs_tree_lock(eb
);
3073 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3075 btrfs_tree_unlock(eb
);
3079 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3080 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3081 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3082 spin_lock(&fs_info
->delalloc_lock
);
3083 if (fs_info
->dirty_metadata_bytes
>= eb
->len
)
3084 fs_info
->dirty_metadata_bytes
-= eb
->len
;
3087 spin_unlock(&fs_info
->delalloc_lock
);
3091 btrfs_tree_unlock(eb
);
3096 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3097 for (i
= 0; i
< num_pages
; i
++) {
3098 struct page
*p
= extent_buffer_page(eb
, i
);
3100 if (!trylock_page(p
)) {
3102 flush_write_bio(epd
);
3112 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3114 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3115 smp_mb__after_clear_bit();
3116 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3119 static void end_bio_extent_buffer_writepage(struct bio
*bio
, int err
)
3121 int uptodate
= err
== 0;
3122 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
3123 struct extent_buffer
*eb
;
3127 struct page
*page
= bvec
->bv_page
;
3130 eb
= (struct extent_buffer
*)page
->private;
3132 done
= atomic_dec_and_test(&eb
->io_pages
);
3134 if (!uptodate
|| test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
3135 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3136 ClearPageUptodate(page
);
3140 end_page_writeback(page
);
3145 end_extent_buffer_writeback(eb
);
3146 } while (bvec
>= bio
->bi_io_vec
);
3152 static int write_one_eb(struct extent_buffer
*eb
,
3153 struct btrfs_fs_info
*fs_info
,
3154 struct writeback_control
*wbc
,
3155 struct extent_page_data
*epd
)
3157 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3158 u64 offset
= eb
->start
;
3159 unsigned long i
, num_pages
;
3160 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
);
3163 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3164 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3165 atomic_set(&eb
->io_pages
, num_pages
);
3166 for (i
= 0; i
< num_pages
; i
++) {
3167 struct page
*p
= extent_buffer_page(eb
, i
);
3169 clear_page_dirty_for_io(p
);
3170 set_page_writeback(p
);
3171 ret
= submit_extent_page(rw
, eb
->tree
, p
, offset
>> 9,
3172 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3173 -1, end_bio_extent_buffer_writepage
,
3176 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
3178 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3179 end_extent_buffer_writeback(eb
);
3183 offset
+= PAGE_CACHE_SIZE
;
3184 update_nr_written(p
, wbc
, 1);
3188 if (unlikely(ret
)) {
3189 for (; i
< num_pages
; i
++) {
3190 struct page
*p
= extent_buffer_page(eb
, i
);
3198 int btree_write_cache_pages(struct address_space
*mapping
,
3199 struct writeback_control
*wbc
)
3201 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3202 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3203 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3204 struct extent_page_data epd
= {
3208 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3212 int nr_to_write_done
= 0;
3213 struct pagevec pvec
;
3216 pgoff_t end
; /* Inclusive */
3220 pagevec_init(&pvec
, 0);
3221 if (wbc
->range_cyclic
) {
3222 index
= mapping
->writeback_index
; /* Start from prev offset */
3225 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3226 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3229 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3230 tag
= PAGECACHE_TAG_TOWRITE
;
3232 tag
= PAGECACHE_TAG_DIRTY
;
3234 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3235 tag_pages_for_writeback(mapping
, index
, end
);
3236 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3237 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3238 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3242 for (i
= 0; i
< nr_pages
; i
++) {
3243 struct page
*page
= pvec
.pages
[i
];
3245 if (!PagePrivate(page
))
3248 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3253 eb
= (struct extent_buffer
*)page
->private;
3262 if (!atomic_inc_not_zero(&eb
->refs
)) {
3268 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3270 free_extent_buffer(eb
);
3274 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3277 free_extent_buffer(eb
);
3280 free_extent_buffer(eb
);
3283 * the filesystem may choose to bump up nr_to_write.
3284 * We have to make sure to honor the new nr_to_write
3287 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3289 pagevec_release(&pvec
);
3292 if (!scanned
&& !done
) {
3294 * We hit the last page and there is more work to be done: wrap
3295 * back to the start of the file
3301 flush_write_bio(&epd
);
3306 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3307 * @mapping: address space structure to write
3308 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3309 * @writepage: function called for each page
3310 * @data: data passed to writepage function
3312 * If a page is already under I/O, write_cache_pages() skips it, even
3313 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3314 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3315 * and msync() need to guarantee that all the data which was dirty at the time
3316 * the call was made get new I/O started against them. If wbc->sync_mode is
3317 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3318 * existing IO to complete.
3320 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3321 struct address_space
*mapping
,
3322 struct writeback_control
*wbc
,
3323 writepage_t writepage
, void *data
,
3324 void (*flush_fn
)(void *))
3328 int nr_to_write_done
= 0;
3329 struct pagevec pvec
;
3332 pgoff_t end
; /* Inclusive */
3336 pagevec_init(&pvec
, 0);
3337 if (wbc
->range_cyclic
) {
3338 index
= mapping
->writeback_index
; /* Start from prev offset */
3341 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3342 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3345 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3346 tag
= PAGECACHE_TAG_TOWRITE
;
3348 tag
= PAGECACHE_TAG_DIRTY
;
3350 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3351 tag_pages_for_writeback(mapping
, index
, end
);
3352 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3353 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3354 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3358 for (i
= 0; i
< nr_pages
; i
++) {
3359 struct page
*page
= pvec
.pages
[i
];
3362 * At this point we hold neither mapping->tree_lock nor
3363 * lock on the page itself: the page may be truncated or
3364 * invalidated (changing page->mapping to NULL), or even
3365 * swizzled back from swapper_space to tmpfs file
3369 tree
->ops
->write_cache_pages_lock_hook
) {
3370 tree
->ops
->write_cache_pages_lock_hook(page
,
3373 if (!trylock_page(page
)) {
3379 if (unlikely(page
->mapping
!= mapping
)) {
3384 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3390 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3391 if (PageWriteback(page
))
3393 wait_on_page_writeback(page
);
3396 if (PageWriteback(page
) ||
3397 !clear_page_dirty_for_io(page
)) {
3402 ret
= (*writepage
)(page
, wbc
, data
);
3404 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
3412 * the filesystem may choose to bump up nr_to_write.
3413 * We have to make sure to honor the new nr_to_write
3416 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3418 pagevec_release(&pvec
);
3421 if (!scanned
&& !done
) {
3423 * We hit the last page and there is more work to be done: wrap
3424 * back to the start of the file
3433 static void flush_epd_write_bio(struct extent_page_data
*epd
)
3442 ret
= submit_one_bio(rw
, epd
->bio
, 0, 0);
3443 BUG_ON(ret
< 0); /* -ENOMEM */
3448 static noinline
void flush_write_bio(void *data
)
3450 struct extent_page_data
*epd
= data
;
3451 flush_epd_write_bio(epd
);
3454 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3455 get_extent_t
*get_extent
,
3456 struct writeback_control
*wbc
)
3459 struct extent_page_data epd
= {
3462 .get_extent
= get_extent
,
3464 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3467 ret
= __extent_writepage(page
, wbc
, &epd
);
3469 flush_epd_write_bio(&epd
);
3473 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
3474 u64 start
, u64 end
, get_extent_t
*get_extent
,
3478 struct address_space
*mapping
= inode
->i_mapping
;
3480 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
3483 struct extent_page_data epd
= {
3486 .get_extent
= get_extent
,
3488 .sync_io
= mode
== WB_SYNC_ALL
,
3490 struct writeback_control wbc_writepages
= {
3492 .nr_to_write
= nr_pages
* 2,
3493 .range_start
= start
,
3494 .range_end
= end
+ 1,
3497 while (start
<= end
) {
3498 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
3499 if (clear_page_dirty_for_io(page
))
3500 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
3502 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3503 tree
->ops
->writepage_end_io_hook(page
, start
,
3504 start
+ PAGE_CACHE_SIZE
- 1,
3508 page_cache_release(page
);
3509 start
+= PAGE_CACHE_SIZE
;
3512 flush_epd_write_bio(&epd
);
3516 int extent_writepages(struct extent_io_tree
*tree
,
3517 struct address_space
*mapping
,
3518 get_extent_t
*get_extent
,
3519 struct writeback_control
*wbc
)
3522 struct extent_page_data epd
= {
3525 .get_extent
= get_extent
,
3527 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3530 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
3531 __extent_writepage
, &epd
,
3533 flush_epd_write_bio(&epd
);
3537 int extent_readpages(struct extent_io_tree
*tree
,
3538 struct address_space
*mapping
,
3539 struct list_head
*pages
, unsigned nr_pages
,
3540 get_extent_t get_extent
)
3542 struct bio
*bio
= NULL
;
3544 unsigned long bio_flags
= 0;
3546 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
3547 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
3549 prefetchw(&page
->flags
);
3550 list_del(&page
->lru
);
3551 if (!add_to_page_cache_lru(page
, mapping
,
3552 page
->index
, GFP_NOFS
)) {
3553 __extent_read_full_page(tree
, page
, get_extent
,
3554 &bio
, 0, &bio_flags
);
3556 page_cache_release(page
);
3558 BUG_ON(!list_empty(pages
));
3560 return submit_one_bio(READ
, bio
, 0, bio_flags
);
3565 * basic invalidatepage code, this waits on any locked or writeback
3566 * ranges corresponding to the page, and then deletes any extent state
3567 * records from the tree
3569 int extent_invalidatepage(struct extent_io_tree
*tree
,
3570 struct page
*page
, unsigned long offset
)
3572 struct extent_state
*cached_state
= NULL
;
3573 u64 start
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
3574 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3575 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
3577 start
+= (offset
+ blocksize
- 1) & ~(blocksize
- 1);
3581 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
3582 wait_on_page_writeback(page
);
3583 clear_extent_bit(tree
, start
, end
,
3584 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3585 EXTENT_DO_ACCOUNTING
,
3586 1, 1, &cached_state
, GFP_NOFS
);
3591 * a helper for releasepage, this tests for areas of the page that
3592 * are locked or under IO and drops the related state bits if it is safe
3595 int try_release_extent_state(struct extent_map_tree
*map
,
3596 struct extent_io_tree
*tree
, struct page
*page
,
3599 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3600 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3603 if (test_range_bit(tree
, start
, end
,
3604 EXTENT_IOBITS
, 0, NULL
))
3607 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
3610 * at this point we can safely clear everything except the
3611 * locked bit and the nodatasum bit
3613 ret
= clear_extent_bit(tree
, start
, end
,
3614 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
3617 /* if clear_extent_bit failed for enomem reasons,
3618 * we can't allow the release to continue.
3629 * a helper for releasepage. As long as there are no locked extents
3630 * in the range corresponding to the page, both state records and extent
3631 * map records are removed
3633 int try_release_extent_mapping(struct extent_map_tree
*map
,
3634 struct extent_io_tree
*tree
, struct page
*page
,
3637 struct extent_map
*em
;
3638 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3639 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3641 if ((mask
& __GFP_WAIT
) &&
3642 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
3644 while (start
<= end
) {
3645 len
= end
- start
+ 1;
3646 write_lock(&map
->lock
);
3647 em
= lookup_extent_mapping(map
, start
, len
);
3649 write_unlock(&map
->lock
);
3652 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
3653 em
->start
!= start
) {
3654 write_unlock(&map
->lock
);
3655 free_extent_map(em
);
3658 if (!test_range_bit(tree
, em
->start
,
3659 extent_map_end(em
) - 1,
3660 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
3662 remove_extent_mapping(map
, em
);
3663 /* once for the rb tree */
3664 free_extent_map(em
);
3666 start
= extent_map_end(em
);
3667 write_unlock(&map
->lock
);
3670 free_extent_map(em
);
3673 return try_release_extent_state(map
, tree
, page
, mask
);
3677 * helper function for fiemap, which doesn't want to see any holes.
3678 * This maps until we find something past 'last'
3680 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
3683 get_extent_t
*get_extent
)
3685 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
3686 struct extent_map
*em
;
3693 len
= last
- offset
;
3696 len
= (len
+ sectorsize
- 1) & ~(sectorsize
- 1);
3697 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
3698 if (IS_ERR_OR_NULL(em
))
3701 /* if this isn't a hole return it */
3702 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
3703 em
->block_start
!= EXTENT_MAP_HOLE
) {
3707 /* this is a hole, advance to the next extent */
3708 offset
= extent_map_end(em
);
3709 free_extent_map(em
);
3716 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
3717 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
3721 u64 max
= start
+ len
;
3725 u64 last_for_get_extent
= 0;
3727 u64 isize
= i_size_read(inode
);
3728 struct btrfs_key found_key
;
3729 struct extent_map
*em
= NULL
;
3730 struct extent_state
*cached_state
= NULL
;
3731 struct btrfs_path
*path
;
3732 struct btrfs_file_extent_item
*item
;
3737 unsigned long emflags
;
3742 path
= btrfs_alloc_path();
3745 path
->leave_spinning
= 1;
3747 start
= ALIGN(start
, BTRFS_I(inode
)->root
->sectorsize
);
3748 len
= ALIGN(len
, BTRFS_I(inode
)->root
->sectorsize
);
3751 * lookup the last file extent. We're not using i_size here
3752 * because there might be preallocation past i_size
3754 ret
= btrfs_lookup_file_extent(NULL
, BTRFS_I(inode
)->root
,
3755 path
, btrfs_ino(inode
), -1, 0);
3757 btrfs_free_path(path
);
3762 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3763 struct btrfs_file_extent_item
);
3764 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
3765 found_type
= btrfs_key_type(&found_key
);
3767 /* No extents, but there might be delalloc bits */
3768 if (found_key
.objectid
!= btrfs_ino(inode
) ||
3769 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3770 /* have to trust i_size as the end */
3772 last_for_get_extent
= isize
;
3775 * remember the start of the last extent. There are a
3776 * bunch of different factors that go into the length of the
3777 * extent, so its much less complex to remember where it started
3779 last
= found_key
.offset
;
3780 last_for_get_extent
= last
+ 1;
3782 btrfs_free_path(path
);
3785 * we might have some extents allocated but more delalloc past those
3786 * extents. so, we trust isize unless the start of the last extent is
3791 last_for_get_extent
= isize
;
3794 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
, 0,
3797 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
3807 u64 offset_in_extent
;
3809 /* break if the extent we found is outside the range */
3810 if (em
->start
>= max
|| extent_map_end(em
) < off
)
3814 * get_extent may return an extent that starts before our
3815 * requested range. We have to make sure the ranges
3816 * we return to fiemap always move forward and don't
3817 * overlap, so adjust the offsets here
3819 em_start
= max(em
->start
, off
);
3822 * record the offset from the start of the extent
3823 * for adjusting the disk offset below
3825 offset_in_extent
= em_start
- em
->start
;
3826 em_end
= extent_map_end(em
);
3827 em_len
= em_end
- em_start
;
3828 emflags
= em
->flags
;
3833 * bump off for our next call to get_extent
3835 off
= extent_map_end(em
);
3839 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
3841 flags
|= FIEMAP_EXTENT_LAST
;
3842 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
3843 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
3844 FIEMAP_EXTENT_NOT_ALIGNED
);
3845 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
3846 flags
|= (FIEMAP_EXTENT_DELALLOC
|
3847 FIEMAP_EXTENT_UNKNOWN
);
3849 disko
= em
->block_start
+ offset_in_extent
;
3851 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
3852 flags
|= FIEMAP_EXTENT_ENCODED
;
3854 free_extent_map(em
);
3856 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
3857 (last
== (u64
)-1 && isize
<= em_end
)) {
3858 flags
|= FIEMAP_EXTENT_LAST
;
3862 /* now scan forward to see if this is really the last extent. */
3863 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
3870 flags
|= FIEMAP_EXTENT_LAST
;
3873 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
3879 free_extent_map(em
);
3881 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
,
3882 &cached_state
, GFP_NOFS
);
3886 inline struct page
*extent_buffer_page(struct extent_buffer
*eb
,
3889 return eb
->pages
[i
];
3892 inline unsigned long num_extent_pages(u64 start
, u64 len
)
3894 return ((start
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
) -
3895 (start
>> PAGE_CACHE_SHIFT
);
3898 static void __free_extent_buffer(struct extent_buffer
*eb
)
3901 unsigned long flags
;
3902 spin_lock_irqsave(&leak_lock
, flags
);
3903 list_del(&eb
->leak_list
);
3904 spin_unlock_irqrestore(&leak_lock
, flags
);
3906 if (eb
->pages
&& eb
->pages
!= eb
->inline_pages
)
3908 kmem_cache_free(extent_buffer_cache
, eb
);
3911 static struct extent_buffer
*__alloc_extent_buffer(struct extent_io_tree
*tree
,
3916 struct extent_buffer
*eb
= NULL
;
3918 unsigned long flags
;
3921 eb
= kmem_cache_zalloc(extent_buffer_cache
, mask
);
3928 rwlock_init(&eb
->lock
);
3929 atomic_set(&eb
->write_locks
, 0);
3930 atomic_set(&eb
->read_locks
, 0);
3931 atomic_set(&eb
->blocking_readers
, 0);
3932 atomic_set(&eb
->blocking_writers
, 0);
3933 atomic_set(&eb
->spinning_readers
, 0);
3934 atomic_set(&eb
->spinning_writers
, 0);
3935 eb
->lock_nested
= 0;
3936 init_waitqueue_head(&eb
->write_lock_wq
);
3937 init_waitqueue_head(&eb
->read_lock_wq
);
3940 spin_lock_irqsave(&leak_lock
, flags
);
3941 list_add(&eb
->leak_list
, &buffers
);
3942 spin_unlock_irqrestore(&leak_lock
, flags
);
3944 spin_lock_init(&eb
->refs_lock
);
3945 atomic_set(&eb
->refs
, 1);
3946 atomic_set(&eb
->io_pages
, 0);
3948 if (len
> MAX_INLINE_EXTENT_BUFFER_SIZE
) {
3949 struct page
**pages
;
3950 int num_pages
= (len
+ PAGE_CACHE_SIZE
- 1) >>
3952 pages
= kzalloc(num_pages
, mask
);
3954 __free_extent_buffer(eb
);
3959 eb
->pages
= eb
->inline_pages
;
3965 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
3969 struct extent_buffer
*new;
3970 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
3972 new = __alloc_extent_buffer(NULL
, src
->start
, src
->len
, GFP_ATOMIC
);
3976 for (i
= 0; i
< num_pages
; i
++) {
3977 p
= alloc_page(GFP_ATOMIC
);
3979 attach_extent_buffer_page(new, p
);
3980 WARN_ON(PageDirty(p
));
3985 copy_extent_buffer(new, src
, 0, 0, src
->len
);
3986 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
3987 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
3992 struct extent_buffer
*alloc_dummy_extent_buffer(u64 start
, unsigned long len
)
3994 struct extent_buffer
*eb
;
3995 unsigned long num_pages
= num_extent_pages(0, len
);
3998 eb
= __alloc_extent_buffer(NULL
, start
, len
, GFP_ATOMIC
);
4002 for (i
= 0; i
< num_pages
; i
++) {
4003 eb
->pages
[i
] = alloc_page(GFP_ATOMIC
);
4007 set_extent_buffer_uptodate(eb
);
4008 btrfs_set_header_nritems(eb
, 0);
4009 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4013 for (i
--; i
> 0; i
--)
4014 __free_page(eb
->pages
[i
]);
4015 __free_extent_buffer(eb
);
4019 static int extent_buffer_under_io(struct extent_buffer
*eb
)
4021 return (atomic_read(&eb
->io_pages
) ||
4022 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4023 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4027 * Helper for releasing extent buffer page.
4029 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
,
4030 unsigned long start_idx
)
4032 unsigned long index
;
4033 unsigned long num_pages
;
4035 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4037 BUG_ON(extent_buffer_under_io(eb
));
4039 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4040 index
= start_idx
+ num_pages
;
4041 if (start_idx
>= index
)
4046 page
= extent_buffer_page(eb
, index
);
4047 if (page
&& mapped
) {
4048 spin_lock(&page
->mapping
->private_lock
);
4050 * We do this since we'll remove the pages after we've
4051 * removed the eb from the radix tree, so we could race
4052 * and have this page now attached to the new eb. So
4053 * only clear page_private if it's still connected to
4056 if (PagePrivate(page
) &&
4057 page
->private == (unsigned long)eb
) {
4058 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4059 BUG_ON(PageDirty(page
));
4060 BUG_ON(PageWriteback(page
));
4062 * We need to make sure we haven't be attached
4065 ClearPagePrivate(page
);
4066 set_page_private(page
, 0);
4067 /* One for the page private */
4068 page_cache_release(page
);
4070 spin_unlock(&page
->mapping
->private_lock
);
4074 /* One for when we alloced the page */
4075 page_cache_release(page
);
4077 } while (index
!= start_idx
);
4081 * Helper for releasing the extent buffer.
4083 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4085 btrfs_release_extent_buffer_page(eb
, 0);
4086 __free_extent_buffer(eb
);
4089 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4091 /* the ref bit is tricky. We have to make sure it is set
4092 * if we have the buffer dirty. Otherwise the
4093 * code to free a buffer can end up dropping a dirty
4096 * Once the ref bit is set, it won't go away while the
4097 * buffer is dirty or in writeback, and it also won't
4098 * go away while we have the reference count on the
4101 * We can't just set the ref bit without bumping the
4102 * ref on the eb because free_extent_buffer might
4103 * see the ref bit and try to clear it. If this happens
4104 * free_extent_buffer might end up dropping our original
4105 * ref by mistake and freeing the page before we are able
4106 * to add one more ref.
4108 * So bump the ref count first, then set the bit. If someone
4109 * beat us to it, drop the ref we added.
4111 if (!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
4112 atomic_inc(&eb
->refs
);
4113 if (test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4114 atomic_dec(&eb
->refs
);
4118 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
)
4120 unsigned long num_pages
, i
;
4122 check_buffer_tree_ref(eb
);
4124 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4125 for (i
= 0; i
< num_pages
; i
++) {
4126 struct page
*p
= extent_buffer_page(eb
, i
);
4127 mark_page_accessed(p
);
4131 struct extent_buffer
*alloc_extent_buffer(struct extent_io_tree
*tree
,
4132 u64 start
, unsigned long len
)
4134 unsigned long num_pages
= num_extent_pages(start
, len
);
4136 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4137 struct extent_buffer
*eb
;
4138 struct extent_buffer
*exists
= NULL
;
4140 struct address_space
*mapping
= tree
->mapping
;
4145 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4146 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4148 mark_extent_buffer_accessed(eb
);
4153 eb
= __alloc_extent_buffer(tree
, start
, len
, GFP_NOFS
);
4157 for (i
= 0; i
< num_pages
; i
++, index
++) {
4158 p
= find_or_create_page(mapping
, index
, GFP_NOFS
);
4164 spin_lock(&mapping
->private_lock
);
4165 if (PagePrivate(p
)) {
4167 * We could have already allocated an eb for this page
4168 * and attached one so lets see if we can get a ref on
4169 * the existing eb, and if we can we know it's good and
4170 * we can just return that one, else we know we can just
4171 * overwrite page->private.
4173 exists
= (struct extent_buffer
*)p
->private;
4174 if (atomic_inc_not_zero(&exists
->refs
)) {
4175 spin_unlock(&mapping
->private_lock
);
4177 page_cache_release(p
);
4178 mark_extent_buffer_accessed(exists
);
4183 * Do this so attach doesn't complain and we need to
4184 * drop the ref the old guy had.
4186 ClearPagePrivate(p
);
4187 WARN_ON(PageDirty(p
));
4188 page_cache_release(p
);
4190 attach_extent_buffer_page(eb
, p
);
4191 spin_unlock(&mapping
->private_lock
);
4192 WARN_ON(PageDirty(p
));
4193 mark_page_accessed(p
);
4195 if (!PageUptodate(p
))
4199 * see below about how we avoid a nasty race with release page
4200 * and why we unlock later
4204 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4206 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4210 spin_lock(&tree
->buffer_lock
);
4211 ret
= radix_tree_insert(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
, eb
);
4212 if (ret
== -EEXIST
) {
4213 exists
= radix_tree_lookup(&tree
->buffer
,
4214 start
>> PAGE_CACHE_SHIFT
);
4215 if (!atomic_inc_not_zero(&exists
->refs
)) {
4216 spin_unlock(&tree
->buffer_lock
);
4217 radix_tree_preload_end();
4221 spin_unlock(&tree
->buffer_lock
);
4222 radix_tree_preload_end();
4223 mark_extent_buffer_accessed(exists
);
4226 /* add one reference for the tree */
4227 spin_lock(&eb
->refs_lock
);
4228 check_buffer_tree_ref(eb
);
4229 spin_unlock(&eb
->refs_lock
);
4230 spin_unlock(&tree
->buffer_lock
);
4231 radix_tree_preload_end();
4234 * there is a race where release page may have
4235 * tried to find this extent buffer in the radix
4236 * but failed. It will tell the VM it is safe to
4237 * reclaim the, and it will clear the page private bit.
4238 * We must make sure to set the page private bit properly
4239 * after the extent buffer is in the radix tree so
4240 * it doesn't get lost
4242 SetPageChecked(eb
->pages
[0]);
4243 for (i
= 1; i
< num_pages
; i
++) {
4244 p
= extent_buffer_page(eb
, i
);
4245 ClearPageChecked(p
);
4248 unlock_page(eb
->pages
[0]);
4252 for (i
= 0; i
< num_pages
; i
++) {
4254 unlock_page(eb
->pages
[i
]);
4257 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4258 btrfs_release_extent_buffer(eb
);
4262 struct extent_buffer
*find_extent_buffer(struct extent_io_tree
*tree
,
4263 u64 start
, unsigned long len
)
4265 struct extent_buffer
*eb
;
4268 eb
= radix_tree_lookup(&tree
->buffer
, start
>> PAGE_CACHE_SHIFT
);
4269 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4271 mark_extent_buffer_accessed(eb
);
4279 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4281 struct extent_buffer
*eb
=
4282 container_of(head
, struct extent_buffer
, rcu_head
);
4284 __free_extent_buffer(eb
);
4287 /* Expects to have eb->eb_lock already held */
4288 static void release_extent_buffer(struct extent_buffer
*eb
, gfp_t mask
)
4290 WARN_ON(atomic_read(&eb
->refs
) == 0);
4291 if (atomic_dec_and_test(&eb
->refs
)) {
4292 if (test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
)) {
4293 spin_unlock(&eb
->refs_lock
);
4295 struct extent_io_tree
*tree
= eb
->tree
;
4297 spin_unlock(&eb
->refs_lock
);
4299 spin_lock(&tree
->buffer_lock
);
4300 radix_tree_delete(&tree
->buffer
,
4301 eb
->start
>> PAGE_CACHE_SHIFT
);
4302 spin_unlock(&tree
->buffer_lock
);
4305 /* Should be safe to release our pages at this point */
4306 btrfs_release_extent_buffer_page(eb
, 0);
4308 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
4311 spin_unlock(&eb
->refs_lock
);
4314 void free_extent_buffer(struct extent_buffer
*eb
)
4319 spin_lock(&eb
->refs_lock
);
4320 if (atomic_read(&eb
->refs
) == 2 &&
4321 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
4322 atomic_dec(&eb
->refs
);
4324 if (atomic_read(&eb
->refs
) == 2 &&
4325 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
4326 !extent_buffer_under_io(eb
) &&
4327 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4328 atomic_dec(&eb
->refs
);
4331 * I know this is terrible, but it's temporary until we stop tracking
4332 * the uptodate bits and such for the extent buffers.
4334 release_extent_buffer(eb
, GFP_ATOMIC
);
4337 void free_extent_buffer_stale(struct extent_buffer
*eb
)
4342 spin_lock(&eb
->refs_lock
);
4343 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
4345 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
4346 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4347 atomic_dec(&eb
->refs
);
4348 release_extent_buffer(eb
, GFP_NOFS
);
4351 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
4354 unsigned long num_pages
;
4357 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4359 for (i
= 0; i
< num_pages
; i
++) {
4360 page
= extent_buffer_page(eb
, i
);
4361 if (!PageDirty(page
))
4365 WARN_ON(!PagePrivate(page
));
4367 clear_page_dirty_for_io(page
);
4368 spin_lock_irq(&page
->mapping
->tree_lock
);
4369 if (!PageDirty(page
)) {
4370 radix_tree_tag_clear(&page
->mapping
->page_tree
,
4372 PAGECACHE_TAG_DIRTY
);
4374 spin_unlock_irq(&page
->mapping
->tree_lock
);
4375 ClearPageError(page
);
4378 WARN_ON(atomic_read(&eb
->refs
) == 0);
4381 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
4384 unsigned long num_pages
;
4387 check_buffer_tree_ref(eb
);
4389 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
4391 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4392 WARN_ON(atomic_read(&eb
->refs
) == 0);
4393 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
4395 for (i
= 0; i
< num_pages
; i
++)
4396 set_page_dirty(extent_buffer_page(eb
, i
));
4400 static int range_straddles_pages(u64 start
, u64 len
)
4402 if (len
< PAGE_CACHE_SIZE
)
4404 if (start
& (PAGE_CACHE_SIZE
- 1))
4406 if ((start
+ len
) & (PAGE_CACHE_SIZE
- 1))
4411 int clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
4415 unsigned long num_pages
;
4417 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4418 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4419 for (i
= 0; i
< num_pages
; i
++) {
4420 page
= extent_buffer_page(eb
, i
);
4422 ClearPageUptodate(page
);
4427 int set_extent_buffer_uptodate(struct extent_buffer
*eb
)
4431 unsigned long num_pages
;
4433 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4434 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4435 for (i
= 0; i
< num_pages
; i
++) {
4436 page
= extent_buffer_page(eb
, i
);
4437 SetPageUptodate(page
);
4442 int extent_range_uptodate(struct extent_io_tree
*tree
,
4447 int pg_uptodate
= 1;
4449 unsigned long index
;
4451 if (range_straddles_pages(start
, end
- start
+ 1)) {
4452 ret
= test_range_bit(tree
, start
, end
,
4453 EXTENT_UPTODATE
, 1, NULL
);
4457 while (start
<= end
) {
4458 index
= start
>> PAGE_CACHE_SHIFT
;
4459 page
= find_get_page(tree
->mapping
, index
);
4462 uptodate
= PageUptodate(page
);
4463 page_cache_release(page
);
4468 start
+= PAGE_CACHE_SIZE
;
4473 int extent_buffer_uptodate(struct extent_buffer
*eb
)
4475 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4478 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
4479 struct extent_buffer
*eb
, u64 start
, int wait
,
4480 get_extent_t
*get_extent
, int mirror_num
)
4483 unsigned long start_i
;
4487 int locked_pages
= 0;
4488 int all_uptodate
= 1;
4489 unsigned long num_pages
;
4490 unsigned long num_reads
= 0;
4491 struct bio
*bio
= NULL
;
4492 unsigned long bio_flags
= 0;
4494 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
4498 WARN_ON(start
< eb
->start
);
4499 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
4500 (eb
->start
>> PAGE_CACHE_SHIFT
);
4505 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4506 for (i
= start_i
; i
< num_pages
; i
++) {
4507 page
= extent_buffer_page(eb
, i
);
4508 if (wait
== WAIT_NONE
) {
4509 if (!trylock_page(page
))
4515 if (!PageUptodate(page
)) {
4522 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4526 clear_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
4527 eb
->read_mirror
= 0;
4528 atomic_set(&eb
->io_pages
, num_reads
);
4529 for (i
= start_i
; i
< num_pages
; i
++) {
4530 page
= extent_buffer_page(eb
, i
);
4531 if (!PageUptodate(page
)) {
4532 ClearPageError(page
);
4533 err
= __extent_read_full_page(tree
, page
,
4535 mirror_num
, &bio_flags
);
4544 err
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
4549 if (ret
|| wait
!= WAIT_COMPLETE
)
4552 for (i
= start_i
; i
< num_pages
; i
++) {
4553 page
= extent_buffer_page(eb
, i
);
4554 wait_on_page_locked(page
);
4555 if (!PageUptodate(page
))
4563 while (locked_pages
> 0) {
4564 page
= extent_buffer_page(eb
, i
);
4572 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
4573 unsigned long start
,
4580 char *dst
= (char *)dstv
;
4581 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4582 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4584 WARN_ON(start
> eb
->len
);
4585 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4587 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4590 page
= extent_buffer_page(eb
, i
);
4592 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4593 kaddr
= page_address(page
);
4594 memcpy(dst
, kaddr
+ offset
, cur
);
4603 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
4604 unsigned long min_len
, char **map
,
4605 unsigned long *map_start
,
4606 unsigned long *map_len
)
4608 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
4611 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4612 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4613 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
4620 offset
= start_offset
;
4624 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
4627 if (start
+ min_len
> eb
->len
) {
4628 printk(KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
4629 "wanted %lu %lu\n", (unsigned long long)eb
->start
,
4630 eb
->len
, start
, min_len
);
4635 p
= extent_buffer_page(eb
, i
);
4636 kaddr
= page_address(p
);
4637 *map
= kaddr
+ offset
;
4638 *map_len
= PAGE_CACHE_SIZE
- offset
;
4642 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
4643 unsigned long start
,
4650 char *ptr
= (char *)ptrv
;
4651 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4652 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4655 WARN_ON(start
> eb
->len
);
4656 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4658 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4661 page
= extent_buffer_page(eb
, i
);
4663 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
4665 kaddr
= page_address(page
);
4666 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
4678 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
4679 unsigned long start
, unsigned long len
)
4685 char *src
= (char *)srcv
;
4686 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4687 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4689 WARN_ON(start
> eb
->len
);
4690 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4692 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4695 page
= extent_buffer_page(eb
, i
);
4696 WARN_ON(!PageUptodate(page
));
4698 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4699 kaddr
= page_address(page
);
4700 memcpy(kaddr
+ offset
, src
, cur
);
4709 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
4710 unsigned long start
, unsigned long len
)
4716 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4717 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
4719 WARN_ON(start
> eb
->len
);
4720 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
4722 offset
= (start_offset
+ start
) & ((unsigned long)PAGE_CACHE_SIZE
- 1);
4725 page
= extent_buffer_page(eb
, i
);
4726 WARN_ON(!PageUptodate(page
));
4728 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
4729 kaddr
= page_address(page
);
4730 memset(kaddr
+ offset
, c
, cur
);
4738 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
4739 unsigned long dst_offset
, unsigned long src_offset
,
4742 u64 dst_len
= dst
->len
;
4747 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4748 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
4750 WARN_ON(src
->len
!= dst_len
);
4752 offset
= (start_offset
+ dst_offset
) &
4753 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4756 page
= extent_buffer_page(dst
, i
);
4757 WARN_ON(!PageUptodate(page
));
4759 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
4761 kaddr
= page_address(page
);
4762 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
4771 static void move_pages(struct page
*dst_page
, struct page
*src_page
,
4772 unsigned long dst_off
, unsigned long src_off
,
4775 char *dst_kaddr
= page_address(dst_page
);
4776 if (dst_page
== src_page
) {
4777 memmove(dst_kaddr
+ dst_off
, dst_kaddr
+ src_off
, len
);
4779 char *src_kaddr
= page_address(src_page
);
4780 char *p
= dst_kaddr
+ dst_off
+ len
;
4781 char *s
= src_kaddr
+ src_off
+ len
;
4788 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
4790 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
4791 return distance
< len
;
4794 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
4795 unsigned long dst_off
, unsigned long src_off
,
4798 char *dst_kaddr
= page_address(dst_page
);
4800 int must_memmove
= 0;
4802 if (dst_page
!= src_page
) {
4803 src_kaddr
= page_address(src_page
);
4805 src_kaddr
= dst_kaddr
;
4806 if (areas_overlap(src_off
, dst_off
, len
))
4811 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
4813 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
4816 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
4817 unsigned long src_offset
, unsigned long len
)
4820 size_t dst_off_in_page
;
4821 size_t src_off_in_page
;
4822 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4823 unsigned long dst_i
;
4824 unsigned long src_i
;
4826 if (src_offset
+ len
> dst
->len
) {
4827 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
4828 "len %lu dst len %lu\n", src_offset
, len
, dst
->len
);
4831 if (dst_offset
+ len
> dst
->len
) {
4832 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
4833 "len %lu dst len %lu\n", dst_offset
, len
, dst
->len
);
4838 dst_off_in_page
= (start_offset
+ dst_offset
) &
4839 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4840 src_off_in_page
= (start_offset
+ src_offset
) &
4841 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4843 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
4844 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
4846 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
4848 cur
= min_t(unsigned long, cur
,
4849 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
4851 copy_pages(extent_buffer_page(dst
, dst_i
),
4852 extent_buffer_page(dst
, src_i
),
4853 dst_off_in_page
, src_off_in_page
, cur
);
4861 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
4862 unsigned long src_offset
, unsigned long len
)
4865 size_t dst_off_in_page
;
4866 size_t src_off_in_page
;
4867 unsigned long dst_end
= dst_offset
+ len
- 1;
4868 unsigned long src_end
= src_offset
+ len
- 1;
4869 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
4870 unsigned long dst_i
;
4871 unsigned long src_i
;
4873 if (src_offset
+ len
> dst
->len
) {
4874 printk(KERN_ERR
"btrfs memmove bogus src_offset %lu move "
4875 "len %lu len %lu\n", src_offset
, len
, dst
->len
);
4878 if (dst_offset
+ len
> dst
->len
) {
4879 printk(KERN_ERR
"btrfs memmove bogus dst_offset %lu move "
4880 "len %lu len %lu\n", dst_offset
, len
, dst
->len
);
4883 if (dst_offset
< src_offset
) {
4884 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
4888 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
4889 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
4891 dst_off_in_page
= (start_offset
+ dst_end
) &
4892 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4893 src_off_in_page
= (start_offset
+ src_end
) &
4894 ((unsigned long)PAGE_CACHE_SIZE
- 1);
4896 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
4897 cur
= min(cur
, dst_off_in_page
+ 1);
4898 move_pages(extent_buffer_page(dst
, dst_i
),
4899 extent_buffer_page(dst
, src_i
),
4900 dst_off_in_page
- cur
+ 1,
4901 src_off_in_page
- cur
+ 1, cur
);
4909 int try_release_extent_buffer(struct page
*page
, gfp_t mask
)
4911 struct extent_buffer
*eb
;
4914 * We need to make sure noboody is attaching this page to an eb right
4917 spin_lock(&page
->mapping
->private_lock
);
4918 if (!PagePrivate(page
)) {
4919 spin_unlock(&page
->mapping
->private_lock
);
4923 eb
= (struct extent_buffer
*)page
->private;
4927 * This is a little awful but should be ok, we need to make sure that
4928 * the eb doesn't disappear out from under us while we're looking at
4931 spin_lock(&eb
->refs_lock
);
4932 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
4933 spin_unlock(&eb
->refs_lock
);
4934 spin_unlock(&page
->mapping
->private_lock
);
4937 spin_unlock(&page
->mapping
->private_lock
);
4939 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4943 * If tree ref isn't set then we know the ref on this eb is a real ref,
4944 * so just return, this page will likely be freed soon anyway.
4946 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
4947 spin_unlock(&eb
->refs_lock
);
4950 release_extent_buffer(eb
, mask
);