1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
24 static struct kmem_cache
*extent_state_cache
;
25 static struct kmem_cache
*extent_buffer_cache
;
26 static struct bio_set
*btrfs_bioset
;
28 static inline bool extent_state_in_tree(const struct extent_state
*state
)
30 return !RB_EMPTY_NODE(&state
->rb_node
);
33 #ifdef CONFIG_BTRFS_DEBUG
34 static LIST_HEAD(buffers
);
35 static LIST_HEAD(states
);
37 static DEFINE_SPINLOCK(leak_lock
);
40 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
44 spin_lock_irqsave(&leak_lock
, flags
);
46 spin_unlock_irqrestore(&leak_lock
, flags
);
50 void btrfs_leak_debug_del(struct list_head
*entry
)
54 spin_lock_irqsave(&leak_lock
, flags
);
56 spin_unlock_irqrestore(&leak_lock
, flags
);
60 void btrfs_leak_debug_check(void)
62 struct extent_state
*state
;
63 struct extent_buffer
*eb
;
65 while (!list_empty(&states
)) {
66 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
67 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
68 state
->start
, state
->end
, state
->state
,
69 extent_state_in_tree(state
),
70 atomic_read(&state
->refs
));
71 list_del(&state
->leak_list
);
72 kmem_cache_free(extent_state_cache
, state
);
75 while (!list_empty(&buffers
)) {
76 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
77 printk(KERN_ERR
"BTRFS: buffer leak start %llu len %lu "
79 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
80 list_del(&eb
->leak_list
);
81 kmem_cache_free(extent_buffer_cache
, eb
);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
88 struct extent_io_tree
*tree
, u64 start
, u64 end
)
96 inode
= tree
->mapping
->host
;
97 isize
= i_size_read(inode
);
98 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
99 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
100 "%s: ino %llu isize %llu odd range [%llu,%llu]",
101 caller
, btrfs_ino(inode
), isize
, start
, end
);
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry) do {} while (0)
107 #define btrfs_leak_debug_check() do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
111 #define BUFFER_LRU_MAX 64
116 struct rb_node rb_node
;
119 struct extent_page_data
{
121 struct extent_io_tree
*tree
;
122 get_extent_t
*get_extent
;
123 unsigned long bio_flags
;
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
128 unsigned int extent_locked
:1;
130 /* tells the submit_bio code to use a WRITE_SYNC */
131 unsigned int sync_io
:1;
134 static void add_extent_changeset(struct extent_state
*state
, unsigned bits
,
135 struct extent_changeset
*changeset
,
142 if (set
&& (state
->state
& bits
) == bits
)
144 if (!set
&& (state
->state
& bits
) == 0)
146 changeset
->bytes_changed
+= state
->end
- state
->start
+ 1;
147 ret
= ulist_add(changeset
->range_changed
, state
->start
, state
->end
,
153 static noinline
void flush_write_bio(void *data
);
154 static inline struct btrfs_fs_info
*
155 tree_fs_info(struct extent_io_tree
*tree
)
159 return btrfs_sb(tree
->mapping
->host
->i_sb
);
162 int __init
extent_io_init(void)
164 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
165 sizeof(struct extent_state
), 0,
166 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
167 if (!extent_state_cache
)
170 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
171 sizeof(struct extent_buffer
), 0,
172 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
173 if (!extent_buffer_cache
)
174 goto free_state_cache
;
176 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
177 offsetof(struct btrfs_io_bio
, bio
));
179 goto free_buffer_cache
;
181 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
187 bioset_free(btrfs_bioset
);
191 kmem_cache_destroy(extent_buffer_cache
);
192 extent_buffer_cache
= NULL
;
195 kmem_cache_destroy(extent_state_cache
);
196 extent_state_cache
= NULL
;
200 void extent_io_exit(void)
202 btrfs_leak_debug_check();
205 * Make sure all delayed rcu free are flushed before we
209 if (extent_state_cache
)
210 kmem_cache_destroy(extent_state_cache
);
211 if (extent_buffer_cache
)
212 kmem_cache_destroy(extent_buffer_cache
);
214 bioset_free(btrfs_bioset
);
217 void extent_io_tree_init(struct extent_io_tree
*tree
,
218 struct address_space
*mapping
)
220 tree
->state
= RB_ROOT
;
222 tree
->dirty_bytes
= 0;
223 spin_lock_init(&tree
->lock
);
224 tree
->mapping
= mapping
;
227 static struct extent_state
*alloc_extent_state(gfp_t mask
)
229 struct extent_state
*state
;
231 state
= kmem_cache_alloc(extent_state_cache
, mask
);
236 RB_CLEAR_NODE(&state
->rb_node
);
237 btrfs_leak_debug_add(&state
->leak_list
, &states
);
238 atomic_set(&state
->refs
, 1);
239 init_waitqueue_head(&state
->wq
);
240 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
244 void free_extent_state(struct extent_state
*state
)
248 if (atomic_dec_and_test(&state
->refs
)) {
249 WARN_ON(extent_state_in_tree(state
));
250 btrfs_leak_debug_del(&state
->leak_list
);
251 trace_free_extent_state(state
, _RET_IP_
);
252 kmem_cache_free(extent_state_cache
, state
);
256 static struct rb_node
*tree_insert(struct rb_root
*root
,
257 struct rb_node
*search_start
,
259 struct rb_node
*node
,
260 struct rb_node
***p_in
,
261 struct rb_node
**parent_in
)
264 struct rb_node
*parent
= NULL
;
265 struct tree_entry
*entry
;
267 if (p_in
&& parent_in
) {
273 p
= search_start
? &search_start
: &root
->rb_node
;
276 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
278 if (offset
< entry
->start
)
280 else if (offset
> entry
->end
)
287 rb_link_node(node
, parent
, p
);
288 rb_insert_color(node
, root
);
292 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
293 struct rb_node
**prev_ret
,
294 struct rb_node
**next_ret
,
295 struct rb_node
***p_ret
,
296 struct rb_node
**parent_ret
)
298 struct rb_root
*root
= &tree
->state
;
299 struct rb_node
**n
= &root
->rb_node
;
300 struct rb_node
*prev
= NULL
;
301 struct rb_node
*orig_prev
= NULL
;
302 struct tree_entry
*entry
;
303 struct tree_entry
*prev_entry
= NULL
;
307 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
310 if (offset
< entry
->start
)
312 else if (offset
> entry
->end
)
325 while (prev
&& offset
> prev_entry
->end
) {
326 prev
= rb_next(prev
);
327 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
334 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
335 while (prev
&& offset
< prev_entry
->start
) {
336 prev
= rb_prev(prev
);
337 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
344 static inline struct rb_node
*
345 tree_search_for_insert(struct extent_io_tree
*tree
,
347 struct rb_node
***p_ret
,
348 struct rb_node
**parent_ret
)
350 struct rb_node
*prev
= NULL
;
353 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
359 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
362 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
365 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
366 struct extent_state
*other
)
368 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
369 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
374 * utility function to look for merge candidates inside a given range.
375 * Any extents with matching state are merged together into a single
376 * extent in the tree. Extents with EXTENT_IO in their state field
377 * are not merged because the end_io handlers need to be able to do
378 * operations on them without sleeping (or doing allocations/splits).
380 * This should be called with the tree lock held.
382 static void merge_state(struct extent_io_tree
*tree
,
383 struct extent_state
*state
)
385 struct extent_state
*other
;
386 struct rb_node
*other_node
;
388 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
391 other_node
= rb_prev(&state
->rb_node
);
393 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
394 if (other
->end
== state
->start
- 1 &&
395 other
->state
== state
->state
) {
396 merge_cb(tree
, state
, other
);
397 state
->start
= other
->start
;
398 rb_erase(&other
->rb_node
, &tree
->state
);
399 RB_CLEAR_NODE(&other
->rb_node
);
400 free_extent_state(other
);
403 other_node
= rb_next(&state
->rb_node
);
405 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
406 if (other
->start
== state
->end
+ 1 &&
407 other
->state
== state
->state
) {
408 merge_cb(tree
, state
, other
);
409 state
->end
= other
->end
;
410 rb_erase(&other
->rb_node
, &tree
->state
);
411 RB_CLEAR_NODE(&other
->rb_node
);
412 free_extent_state(other
);
417 static void set_state_cb(struct extent_io_tree
*tree
,
418 struct extent_state
*state
, unsigned *bits
)
420 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
421 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
424 static void clear_state_cb(struct extent_io_tree
*tree
,
425 struct extent_state
*state
, unsigned *bits
)
427 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
428 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
431 static void set_state_bits(struct extent_io_tree
*tree
,
432 struct extent_state
*state
, unsigned *bits
,
433 struct extent_changeset
*changeset
);
436 * insert an extent_state struct into the tree. 'bits' are set on the
437 * struct before it is inserted.
439 * This may return -EEXIST if the extent is already there, in which case the
440 * state struct is freed.
442 * The tree lock is not taken internally. This is a utility function and
443 * probably isn't what you want to call (see set/clear_extent_bit).
445 static int insert_state(struct extent_io_tree
*tree
,
446 struct extent_state
*state
, u64 start
, u64 end
,
448 struct rb_node
**parent
,
449 unsigned *bits
, struct extent_changeset
*changeset
)
451 struct rb_node
*node
;
454 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
456 state
->start
= start
;
459 set_state_bits(tree
, state
, bits
, changeset
);
461 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
463 struct extent_state
*found
;
464 found
= rb_entry(node
, struct extent_state
, rb_node
);
465 printk(KERN_ERR
"BTRFS: found node %llu %llu on insert of "
467 found
->start
, found
->end
, start
, end
);
470 merge_state(tree
, state
);
474 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
477 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
478 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
482 * split a given extent state struct in two, inserting the preallocated
483 * struct 'prealloc' as the newly created second half. 'split' indicates an
484 * offset inside 'orig' where it should be split.
487 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
488 * are two extent state structs in the tree:
489 * prealloc: [orig->start, split - 1]
490 * orig: [ split, orig->end ]
492 * The tree locks are not taken by this function. They need to be held
495 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
496 struct extent_state
*prealloc
, u64 split
)
498 struct rb_node
*node
;
500 split_cb(tree
, orig
, split
);
502 prealloc
->start
= orig
->start
;
503 prealloc
->end
= split
- 1;
504 prealloc
->state
= orig
->state
;
507 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
508 &prealloc
->rb_node
, NULL
, NULL
);
510 free_extent_state(prealloc
);
516 static struct extent_state
*next_state(struct extent_state
*state
)
518 struct rb_node
*next
= rb_next(&state
->rb_node
);
520 return rb_entry(next
, struct extent_state
, rb_node
);
526 * utility function to clear some bits in an extent state struct.
527 * it will optionally wake up any one waiting on this state (wake == 1).
529 * If no bits are set on the state struct after clearing things, the
530 * struct is freed and removed from the tree
532 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
533 struct extent_state
*state
,
534 unsigned *bits
, int wake
,
535 struct extent_changeset
*changeset
)
537 struct extent_state
*next
;
538 unsigned bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
540 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
541 u64 range
= state
->end
- state
->start
+ 1;
542 WARN_ON(range
> tree
->dirty_bytes
);
543 tree
->dirty_bytes
-= range
;
545 clear_state_cb(tree
, state
, bits
);
546 add_extent_changeset(state
, bits_to_clear
, changeset
, 0);
547 state
->state
&= ~bits_to_clear
;
550 if (state
->state
== 0) {
551 next
= next_state(state
);
552 if (extent_state_in_tree(state
)) {
553 rb_erase(&state
->rb_node
, &tree
->state
);
554 RB_CLEAR_NODE(&state
->rb_node
);
555 free_extent_state(state
);
560 merge_state(tree
, state
);
561 next
= next_state(state
);
566 static struct extent_state
*
567 alloc_extent_state_atomic(struct extent_state
*prealloc
)
570 prealloc
= alloc_extent_state(GFP_ATOMIC
);
575 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
577 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
578 "Extent tree was modified by another "
579 "thread while locked.");
583 * clear some bits on a range in the tree. This may require splitting
584 * or inserting elements in the tree, so the gfp mask is used to
585 * indicate which allocations or sleeping are allowed.
587 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
588 * the given range from the tree regardless of state (ie for truncate).
590 * the range [start, end] is inclusive.
592 * This takes the tree lock, and returns 0 on success and < 0 on error.
594 static int __clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
595 unsigned bits
, int wake
, int delete,
596 struct extent_state
**cached_state
,
597 gfp_t mask
, struct extent_changeset
*changeset
)
599 struct extent_state
*state
;
600 struct extent_state
*cached
;
601 struct extent_state
*prealloc
= NULL
;
602 struct rb_node
*node
;
607 btrfs_debug_check_extent_io_range(tree
, start
, end
);
609 if (bits
& EXTENT_DELALLOC
)
610 bits
|= EXTENT_NORESERVE
;
613 bits
|= ~EXTENT_CTLBITS
;
614 bits
|= EXTENT_FIRST_DELALLOC
;
616 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
619 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
621 * Don't care for allocation failure here because we might end
622 * up not needing the pre-allocated extent state at all, which
623 * is the case if we only have in the tree extent states that
624 * cover our input range and don't cover too any other range.
625 * If we end up needing a new extent state we allocate it later.
627 prealloc
= alloc_extent_state(mask
);
630 spin_lock(&tree
->lock
);
632 cached
= *cached_state
;
635 *cached_state
= NULL
;
639 if (cached
&& extent_state_in_tree(cached
) &&
640 cached
->start
<= start
&& cached
->end
> start
) {
642 atomic_dec(&cached
->refs
);
647 free_extent_state(cached
);
650 * this search will find the extents that end after
653 node
= tree_search(tree
, start
);
656 state
= rb_entry(node
, struct extent_state
, rb_node
);
658 if (state
->start
> end
)
660 WARN_ON(state
->end
< start
);
661 last_end
= state
->end
;
663 /* the state doesn't have the wanted bits, go ahead */
664 if (!(state
->state
& bits
)) {
665 state
= next_state(state
);
670 * | ---- desired range ---- |
672 * | ------------- state -------------- |
674 * We need to split the extent we found, and may flip
675 * bits on second half.
677 * If the extent we found extends past our range, we
678 * just split and search again. It'll get split again
679 * the next time though.
681 * If the extent we found is inside our range, we clear
682 * the desired bit on it.
685 if (state
->start
< start
) {
686 prealloc
= alloc_extent_state_atomic(prealloc
);
688 err
= split_state(tree
, state
, prealloc
, start
);
690 extent_io_tree_panic(tree
, err
);
695 if (state
->end
<= end
) {
696 state
= clear_state_bit(tree
, state
, &bits
, wake
,
703 * | ---- desired range ---- |
705 * We need to split the extent, and clear the bit
708 if (state
->start
<= end
&& state
->end
> end
) {
709 prealloc
= alloc_extent_state_atomic(prealloc
);
711 err
= split_state(tree
, state
, prealloc
, end
+ 1);
713 extent_io_tree_panic(tree
, err
);
718 clear_state_bit(tree
, prealloc
, &bits
, wake
, changeset
);
724 state
= clear_state_bit(tree
, state
, &bits
, wake
, changeset
);
726 if (last_end
== (u64
)-1)
728 start
= last_end
+ 1;
729 if (start
<= end
&& state
&& !need_resched())
734 spin_unlock(&tree
->lock
);
736 free_extent_state(prealloc
);
743 spin_unlock(&tree
->lock
);
744 if (gfpflags_allow_blocking(mask
))
749 static void wait_on_state(struct extent_io_tree
*tree
,
750 struct extent_state
*state
)
751 __releases(tree
->lock
)
752 __acquires(tree
->lock
)
755 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
756 spin_unlock(&tree
->lock
);
758 spin_lock(&tree
->lock
);
759 finish_wait(&state
->wq
, &wait
);
763 * waits for one or more bits to clear on a range in the state tree.
764 * The range [start, end] is inclusive.
765 * The tree lock is taken by this function
767 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
770 struct extent_state
*state
;
771 struct rb_node
*node
;
773 btrfs_debug_check_extent_io_range(tree
, start
, end
);
775 spin_lock(&tree
->lock
);
779 * this search will find all the extents that end after
782 node
= tree_search(tree
, start
);
787 state
= rb_entry(node
, struct extent_state
, rb_node
);
789 if (state
->start
> end
)
792 if (state
->state
& bits
) {
793 start
= state
->start
;
794 atomic_inc(&state
->refs
);
795 wait_on_state(tree
, state
);
796 free_extent_state(state
);
799 start
= state
->end
+ 1;
804 if (!cond_resched_lock(&tree
->lock
)) {
805 node
= rb_next(node
);
810 spin_unlock(&tree
->lock
);
813 static void set_state_bits(struct extent_io_tree
*tree
,
814 struct extent_state
*state
,
815 unsigned *bits
, struct extent_changeset
*changeset
)
817 unsigned bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
819 set_state_cb(tree
, state
, bits
);
820 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
821 u64 range
= state
->end
- state
->start
+ 1;
822 tree
->dirty_bytes
+= range
;
824 add_extent_changeset(state
, bits_to_set
, changeset
, 1);
825 state
->state
|= bits_to_set
;
828 static void cache_state_if_flags(struct extent_state
*state
,
829 struct extent_state
**cached_ptr
,
832 if (cached_ptr
&& !(*cached_ptr
)) {
833 if (!flags
|| (state
->state
& flags
)) {
835 atomic_inc(&state
->refs
);
840 static void cache_state(struct extent_state
*state
,
841 struct extent_state
**cached_ptr
)
843 return cache_state_if_flags(state
, cached_ptr
,
844 EXTENT_IOBITS
| EXTENT_BOUNDARY
);
848 * set some bits on a range in the tree. This may require allocations or
849 * sleeping, so the gfp mask is used to indicate what is allowed.
851 * If any of the exclusive bits are set, this will fail with -EEXIST if some
852 * part of the range already has the desired bits set. The start of the
853 * existing range is returned in failed_start in this case.
855 * [start, end] is inclusive This takes the tree lock.
858 static int __must_check
859 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
860 unsigned bits
, unsigned exclusive_bits
,
861 u64
*failed_start
, struct extent_state
**cached_state
,
862 gfp_t mask
, struct extent_changeset
*changeset
)
864 struct extent_state
*state
;
865 struct extent_state
*prealloc
= NULL
;
866 struct rb_node
*node
;
868 struct rb_node
*parent
;
873 btrfs_debug_check_extent_io_range(tree
, start
, end
);
875 bits
|= EXTENT_FIRST_DELALLOC
;
877 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
878 prealloc
= alloc_extent_state(mask
);
882 spin_lock(&tree
->lock
);
883 if (cached_state
&& *cached_state
) {
884 state
= *cached_state
;
885 if (state
->start
<= start
&& state
->end
> start
&&
886 extent_state_in_tree(state
)) {
887 node
= &state
->rb_node
;
892 * this search will find all the extents that end after
895 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
897 prealloc
= alloc_extent_state_atomic(prealloc
);
899 err
= insert_state(tree
, prealloc
, start
, end
,
900 &p
, &parent
, &bits
, changeset
);
902 extent_io_tree_panic(tree
, err
);
904 cache_state(prealloc
, cached_state
);
908 state
= rb_entry(node
, struct extent_state
, rb_node
);
910 last_start
= state
->start
;
911 last_end
= state
->end
;
914 * | ---- desired range ---- |
917 * Just lock what we found and keep going
919 if (state
->start
== start
&& state
->end
<= end
) {
920 if (state
->state
& exclusive_bits
) {
921 *failed_start
= state
->start
;
926 set_state_bits(tree
, state
, &bits
, changeset
);
927 cache_state(state
, cached_state
);
928 merge_state(tree
, state
);
929 if (last_end
== (u64
)-1)
931 start
= last_end
+ 1;
932 state
= next_state(state
);
933 if (start
< end
&& state
&& state
->start
== start
&&
940 * | ---- desired range ---- |
943 * | ------------- state -------------- |
945 * We need to split the extent we found, and may flip bits on
948 * If the extent we found extends past our
949 * range, we just split and search again. It'll get split
950 * again the next time though.
952 * If the extent we found is inside our range, we set the
955 if (state
->start
< start
) {
956 if (state
->state
& exclusive_bits
) {
957 *failed_start
= start
;
962 prealloc
= alloc_extent_state_atomic(prealloc
);
964 err
= split_state(tree
, state
, prealloc
, start
);
966 extent_io_tree_panic(tree
, err
);
971 if (state
->end
<= end
) {
972 set_state_bits(tree
, state
, &bits
, changeset
);
973 cache_state(state
, cached_state
);
974 merge_state(tree
, state
);
975 if (last_end
== (u64
)-1)
977 start
= last_end
+ 1;
978 state
= next_state(state
);
979 if (start
< end
&& state
&& state
->start
== start
&&
986 * | ---- desired range ---- |
987 * | state | or | state |
989 * There's a hole, we need to insert something in it and
990 * ignore the extent we found.
992 if (state
->start
> start
) {
994 if (end
< last_start
)
997 this_end
= last_start
- 1;
999 prealloc
= alloc_extent_state_atomic(prealloc
);
1003 * Avoid to free 'prealloc' if it can be merged with
1006 err
= insert_state(tree
, prealloc
, start
, this_end
,
1007 NULL
, NULL
, &bits
, changeset
);
1009 extent_io_tree_panic(tree
, err
);
1011 cache_state(prealloc
, cached_state
);
1013 start
= this_end
+ 1;
1017 * | ---- desired range ---- |
1019 * We need to split the extent, and set the bit
1022 if (state
->start
<= end
&& state
->end
> end
) {
1023 if (state
->state
& exclusive_bits
) {
1024 *failed_start
= start
;
1029 prealloc
= alloc_extent_state_atomic(prealloc
);
1031 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1033 extent_io_tree_panic(tree
, err
);
1035 set_state_bits(tree
, prealloc
, &bits
, changeset
);
1036 cache_state(prealloc
, cached_state
);
1037 merge_state(tree
, prealloc
);
1045 spin_unlock(&tree
->lock
);
1047 free_extent_state(prealloc
);
1054 spin_unlock(&tree
->lock
);
1055 if (gfpflags_allow_blocking(mask
))
1060 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1061 unsigned bits
, u64
* failed_start
,
1062 struct extent_state
**cached_state
, gfp_t mask
)
1064 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1065 cached_state
, mask
, NULL
);
1070 * convert_extent_bit - convert all bits in a given range from one bit to
1072 * @tree: the io tree to search
1073 * @start: the start offset in bytes
1074 * @end: the end offset in bytes (inclusive)
1075 * @bits: the bits to set in this range
1076 * @clear_bits: the bits to clear in this range
1077 * @cached_state: state that we're going to cache
1078 * @mask: the allocation mask
1080 * This will go through and set bits for the given range. If any states exist
1081 * already in this range they are set with the given bit and cleared of the
1082 * clear_bits. This is only meant to be used by things that are mergeable, ie
1083 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1084 * boundary bits like LOCK.
1086 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1087 unsigned bits
, unsigned clear_bits
,
1088 struct extent_state
**cached_state
, gfp_t mask
)
1090 struct extent_state
*state
;
1091 struct extent_state
*prealloc
= NULL
;
1092 struct rb_node
*node
;
1094 struct rb_node
*parent
;
1098 bool first_iteration
= true;
1100 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1103 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
1105 * Best effort, don't worry if extent state allocation fails
1106 * here for the first iteration. We might have a cached state
1107 * that matches exactly the target range, in which case no
1108 * extent state allocations are needed. We'll only know this
1109 * after locking the tree.
1111 prealloc
= alloc_extent_state(mask
);
1112 if (!prealloc
&& !first_iteration
)
1116 spin_lock(&tree
->lock
);
1117 if (cached_state
&& *cached_state
) {
1118 state
= *cached_state
;
1119 if (state
->start
<= start
&& state
->end
> start
&&
1120 extent_state_in_tree(state
)) {
1121 node
= &state
->rb_node
;
1127 * this search will find all the extents that end after
1130 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1132 prealloc
= alloc_extent_state_atomic(prealloc
);
1137 err
= insert_state(tree
, prealloc
, start
, end
,
1138 &p
, &parent
, &bits
, NULL
);
1140 extent_io_tree_panic(tree
, err
);
1141 cache_state(prealloc
, cached_state
);
1145 state
= rb_entry(node
, struct extent_state
, rb_node
);
1147 last_start
= state
->start
;
1148 last_end
= state
->end
;
1151 * | ---- desired range ---- |
1154 * Just lock what we found and keep going
1156 if (state
->start
== start
&& state
->end
<= end
) {
1157 set_state_bits(tree
, state
, &bits
, NULL
);
1158 cache_state(state
, cached_state
);
1159 state
= clear_state_bit(tree
, state
, &clear_bits
, 0, NULL
);
1160 if (last_end
== (u64
)-1)
1162 start
= last_end
+ 1;
1163 if (start
< end
&& state
&& state
->start
== start
&&
1170 * | ---- desired range ---- |
1173 * | ------------- state -------------- |
1175 * We need to split the extent we found, and may flip bits on
1178 * If the extent we found extends past our
1179 * range, we just split and search again. It'll get split
1180 * again the next time though.
1182 * If the extent we found is inside our range, we set the
1183 * desired bit on it.
1185 if (state
->start
< start
) {
1186 prealloc
= alloc_extent_state_atomic(prealloc
);
1191 err
= split_state(tree
, state
, prealloc
, start
);
1193 extent_io_tree_panic(tree
, err
);
1197 if (state
->end
<= end
) {
1198 set_state_bits(tree
, state
, &bits
, NULL
);
1199 cache_state(state
, cached_state
);
1200 state
= clear_state_bit(tree
, state
, &clear_bits
, 0,
1202 if (last_end
== (u64
)-1)
1204 start
= last_end
+ 1;
1205 if (start
< end
&& state
&& state
->start
== start
&&
1212 * | ---- desired range ---- |
1213 * | state | or | state |
1215 * There's a hole, we need to insert something in it and
1216 * ignore the extent we found.
1218 if (state
->start
> start
) {
1220 if (end
< last_start
)
1223 this_end
= last_start
- 1;
1225 prealloc
= alloc_extent_state_atomic(prealloc
);
1232 * Avoid to free 'prealloc' if it can be merged with
1235 err
= insert_state(tree
, prealloc
, start
, this_end
,
1236 NULL
, NULL
, &bits
, NULL
);
1238 extent_io_tree_panic(tree
, err
);
1239 cache_state(prealloc
, cached_state
);
1241 start
= this_end
+ 1;
1245 * | ---- desired range ---- |
1247 * We need to split the extent, and set the bit
1250 if (state
->start
<= end
&& state
->end
> end
) {
1251 prealloc
= alloc_extent_state_atomic(prealloc
);
1257 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1259 extent_io_tree_panic(tree
, err
);
1261 set_state_bits(tree
, prealloc
, &bits
, NULL
);
1262 cache_state(prealloc
, cached_state
);
1263 clear_state_bit(tree
, prealloc
, &clear_bits
, 0, NULL
);
1271 spin_unlock(&tree
->lock
);
1273 free_extent_state(prealloc
);
1280 spin_unlock(&tree
->lock
);
1281 if (gfpflags_allow_blocking(mask
))
1283 first_iteration
= false;
1287 /* wrappers around set/clear extent bit */
1288 int set_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1289 unsigned bits
, gfp_t mask
,
1290 struct extent_changeset
*changeset
)
1293 * We don't support EXTENT_LOCKED yet, as current changeset will
1294 * record any bits changed, so for EXTENT_LOCKED case, it will
1295 * either fail with -EEXIST or changeset will record the whole
1298 BUG_ON(bits
& EXTENT_LOCKED
);
1300 return __set_extent_bit(tree
, start
, end
, bits
, 0, NULL
, NULL
, mask
,
1304 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1305 unsigned bits
, int wake
, int delete,
1306 struct extent_state
**cached
, gfp_t mask
)
1308 return __clear_extent_bit(tree
, start
, end
, bits
, wake
, delete,
1309 cached
, mask
, NULL
);
1312 int clear_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1313 unsigned bits
, gfp_t mask
,
1314 struct extent_changeset
*changeset
)
1317 * Don't support EXTENT_LOCKED case, same reason as
1318 * set_record_extent_bits().
1320 BUG_ON(bits
& EXTENT_LOCKED
);
1322 return __clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
,
1327 * either insert or lock state struct between start and end use mask to tell
1328 * us if waiting is desired.
1330 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1331 struct extent_state
**cached_state
)
1337 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
,
1338 EXTENT_LOCKED
, &failed_start
,
1339 cached_state
, GFP_NOFS
, NULL
);
1340 if (err
== -EEXIST
) {
1341 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1342 start
= failed_start
;
1345 WARN_ON(start
> end
);
1350 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1355 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1356 &failed_start
, NULL
, GFP_NOFS
, NULL
);
1357 if (err
== -EEXIST
) {
1358 if (failed_start
> start
)
1359 clear_extent_bit(tree
, start
, failed_start
- 1,
1360 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1366 void extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1368 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1369 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1372 while (index
<= end_index
) {
1373 page
= find_get_page(inode
->i_mapping
, index
);
1374 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1375 clear_page_dirty_for_io(page
);
1376 page_cache_release(page
);
1381 void extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1383 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1384 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1387 while (index
<= end_index
) {
1388 page
= find_get_page(inode
->i_mapping
, index
);
1389 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1390 __set_page_dirty_nobuffers(page
);
1391 account_page_redirty(page
);
1392 page_cache_release(page
);
1398 * helper function to set both pages and extents in the tree writeback
1400 static void set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1402 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1403 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1406 while (index
<= end_index
) {
1407 page
= find_get_page(tree
->mapping
, index
);
1408 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1409 set_page_writeback(page
);
1410 page_cache_release(page
);
1415 /* find the first state struct with 'bits' set after 'start', and
1416 * return it. tree->lock must be held. NULL will returned if
1417 * nothing was found after 'start'
1419 static struct extent_state
*
1420 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1421 u64 start
, unsigned bits
)
1423 struct rb_node
*node
;
1424 struct extent_state
*state
;
1427 * this search will find all the extents that end after
1430 node
= tree_search(tree
, start
);
1435 state
= rb_entry(node
, struct extent_state
, rb_node
);
1436 if (state
->end
>= start
&& (state
->state
& bits
))
1439 node
= rb_next(node
);
1448 * find the first offset in the io tree with 'bits' set. zero is
1449 * returned if we find something, and *start_ret and *end_ret are
1450 * set to reflect the state struct that was found.
1452 * If nothing was found, 1 is returned. If found something, return 0.
1454 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1455 u64
*start_ret
, u64
*end_ret
, unsigned bits
,
1456 struct extent_state
**cached_state
)
1458 struct extent_state
*state
;
1462 spin_lock(&tree
->lock
);
1463 if (cached_state
&& *cached_state
) {
1464 state
= *cached_state
;
1465 if (state
->end
== start
- 1 && extent_state_in_tree(state
)) {
1466 n
= rb_next(&state
->rb_node
);
1468 state
= rb_entry(n
, struct extent_state
,
1470 if (state
->state
& bits
)
1474 free_extent_state(*cached_state
);
1475 *cached_state
= NULL
;
1478 free_extent_state(*cached_state
);
1479 *cached_state
= NULL
;
1482 state
= find_first_extent_bit_state(tree
, start
, bits
);
1485 cache_state_if_flags(state
, cached_state
, 0);
1486 *start_ret
= state
->start
;
1487 *end_ret
= state
->end
;
1491 spin_unlock(&tree
->lock
);
1496 * find a contiguous range of bytes in the file marked as delalloc, not
1497 * more than 'max_bytes'. start and end are used to return the range,
1499 * 1 is returned if we find something, 0 if nothing was in the tree
1501 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1502 u64
*start
, u64
*end
, u64 max_bytes
,
1503 struct extent_state
**cached_state
)
1505 struct rb_node
*node
;
1506 struct extent_state
*state
;
1507 u64 cur_start
= *start
;
1509 u64 total_bytes
= 0;
1511 spin_lock(&tree
->lock
);
1514 * this search will find all the extents that end after
1517 node
= tree_search(tree
, cur_start
);
1525 state
= rb_entry(node
, struct extent_state
, rb_node
);
1526 if (found
&& (state
->start
!= cur_start
||
1527 (state
->state
& EXTENT_BOUNDARY
))) {
1530 if (!(state
->state
& EXTENT_DELALLOC
)) {
1536 *start
= state
->start
;
1537 *cached_state
= state
;
1538 atomic_inc(&state
->refs
);
1542 cur_start
= state
->end
+ 1;
1543 node
= rb_next(node
);
1544 total_bytes
+= state
->end
- state
->start
+ 1;
1545 if (total_bytes
>= max_bytes
)
1551 spin_unlock(&tree
->lock
);
1555 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1556 struct page
*locked_page
,
1560 struct page
*pages
[16];
1561 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1562 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1563 unsigned long nr_pages
= end_index
- index
+ 1;
1566 if (index
== locked_page
->index
&& end_index
== index
)
1569 while (nr_pages
> 0) {
1570 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1571 min_t(unsigned long, nr_pages
,
1572 ARRAY_SIZE(pages
)), pages
);
1573 for (i
= 0; i
< ret
; i
++) {
1574 if (pages
[i
] != locked_page
)
1575 unlock_page(pages
[i
]);
1576 page_cache_release(pages
[i
]);
1584 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1585 struct page
*locked_page
,
1589 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1590 unsigned long start_index
= index
;
1591 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1592 unsigned long pages_locked
= 0;
1593 struct page
*pages
[16];
1594 unsigned long nrpages
;
1598 /* the caller is responsible for locking the start index */
1599 if (index
== locked_page
->index
&& index
== end_index
)
1602 /* skip the page at the start index */
1603 nrpages
= end_index
- index
+ 1;
1604 while (nrpages
> 0) {
1605 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1606 min_t(unsigned long,
1607 nrpages
, ARRAY_SIZE(pages
)), pages
);
1612 /* now we have an array of pages, lock them all */
1613 for (i
= 0; i
< ret
; i
++) {
1615 * the caller is taking responsibility for
1618 if (pages
[i
] != locked_page
) {
1619 lock_page(pages
[i
]);
1620 if (!PageDirty(pages
[i
]) ||
1621 pages
[i
]->mapping
!= inode
->i_mapping
) {
1623 unlock_page(pages
[i
]);
1624 page_cache_release(pages
[i
]);
1628 page_cache_release(pages
[i
]);
1637 if (ret
&& pages_locked
) {
1638 __unlock_for_delalloc(inode
, locked_page
,
1640 ((u64
)(start_index
+ pages_locked
- 1)) <<
1647 * find a contiguous range of bytes in the file marked as delalloc, not
1648 * more than 'max_bytes'. start and end are used to return the range,
1650 * 1 is returned if we find something, 0 if nothing was in the tree
1652 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1653 struct extent_io_tree
*tree
,
1654 struct page
*locked_page
, u64
*start
,
1655 u64
*end
, u64 max_bytes
)
1660 struct extent_state
*cached_state
= NULL
;
1665 /* step one, find a bunch of delalloc bytes starting at start */
1666 delalloc_start
= *start
;
1668 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1669 max_bytes
, &cached_state
);
1670 if (!found
|| delalloc_end
<= *start
) {
1671 *start
= delalloc_start
;
1672 *end
= delalloc_end
;
1673 free_extent_state(cached_state
);
1678 * start comes from the offset of locked_page. We have to lock
1679 * pages in order, so we can't process delalloc bytes before
1682 if (delalloc_start
< *start
)
1683 delalloc_start
= *start
;
1686 * make sure to limit the number of pages we try to lock down
1688 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1689 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1691 /* step two, lock all the pages after the page that has start */
1692 ret
= lock_delalloc_pages(inode
, locked_page
,
1693 delalloc_start
, delalloc_end
);
1694 if (ret
== -EAGAIN
) {
1695 /* some of the pages are gone, lets avoid looping by
1696 * shortening the size of the delalloc range we're searching
1698 free_extent_state(cached_state
);
1699 cached_state
= NULL
;
1701 max_bytes
= PAGE_CACHE_SIZE
;
1709 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1711 /* step three, lock the state bits for the whole range */
1712 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, &cached_state
);
1714 /* then test to make sure it is all still delalloc */
1715 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1716 EXTENT_DELALLOC
, 1, cached_state
);
1718 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1719 &cached_state
, GFP_NOFS
);
1720 __unlock_for_delalloc(inode
, locked_page
,
1721 delalloc_start
, delalloc_end
);
1725 free_extent_state(cached_state
);
1726 *start
= delalloc_start
;
1727 *end
= delalloc_end
;
1732 void extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1733 struct page
*locked_page
,
1734 unsigned clear_bits
,
1735 unsigned long page_ops
)
1737 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1739 struct page
*pages
[16];
1740 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1741 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1742 unsigned long nr_pages
= end_index
- index
+ 1;
1745 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1749 if ((page_ops
& PAGE_SET_ERROR
) && nr_pages
> 0)
1750 mapping_set_error(inode
->i_mapping
, -EIO
);
1752 while (nr_pages
> 0) {
1753 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1754 min_t(unsigned long,
1755 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1756 for (i
= 0; i
< ret
; i
++) {
1758 if (page_ops
& PAGE_SET_PRIVATE2
)
1759 SetPagePrivate2(pages
[i
]);
1761 if (pages
[i
] == locked_page
) {
1762 page_cache_release(pages
[i
]);
1765 if (page_ops
& PAGE_CLEAR_DIRTY
)
1766 clear_page_dirty_for_io(pages
[i
]);
1767 if (page_ops
& PAGE_SET_WRITEBACK
)
1768 set_page_writeback(pages
[i
]);
1769 if (page_ops
& PAGE_SET_ERROR
)
1770 SetPageError(pages
[i
]);
1771 if (page_ops
& PAGE_END_WRITEBACK
)
1772 end_page_writeback(pages
[i
]);
1773 if (page_ops
& PAGE_UNLOCK
)
1774 unlock_page(pages
[i
]);
1775 page_cache_release(pages
[i
]);
1784 * count the number of bytes in the tree that have a given bit(s)
1785 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1786 * cached. The total number found is returned.
1788 u64
count_range_bits(struct extent_io_tree
*tree
,
1789 u64
*start
, u64 search_end
, u64 max_bytes
,
1790 unsigned bits
, int contig
)
1792 struct rb_node
*node
;
1793 struct extent_state
*state
;
1794 u64 cur_start
= *start
;
1795 u64 total_bytes
= 0;
1799 if (WARN_ON(search_end
<= cur_start
))
1802 spin_lock(&tree
->lock
);
1803 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1804 total_bytes
= tree
->dirty_bytes
;
1808 * this search will find all the extents that end after
1811 node
= tree_search(tree
, cur_start
);
1816 state
= rb_entry(node
, struct extent_state
, rb_node
);
1817 if (state
->start
> search_end
)
1819 if (contig
&& found
&& state
->start
> last
+ 1)
1821 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1822 total_bytes
+= min(search_end
, state
->end
) + 1 -
1823 max(cur_start
, state
->start
);
1824 if (total_bytes
>= max_bytes
)
1827 *start
= max(cur_start
, state
->start
);
1831 } else if (contig
&& found
) {
1834 node
= rb_next(node
);
1839 spin_unlock(&tree
->lock
);
1844 * set the private field for a given byte offset in the tree. If there isn't
1845 * an extent_state there already, this does nothing.
1847 static int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1849 struct rb_node
*node
;
1850 struct extent_state
*state
;
1853 spin_lock(&tree
->lock
);
1855 * this search will find all the extents that end after
1858 node
= tree_search(tree
, start
);
1863 state
= rb_entry(node
, struct extent_state
, rb_node
);
1864 if (state
->start
!= start
) {
1868 state
->private = private;
1870 spin_unlock(&tree
->lock
);
1874 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1876 struct rb_node
*node
;
1877 struct extent_state
*state
;
1880 spin_lock(&tree
->lock
);
1882 * this search will find all the extents that end after
1885 node
= tree_search(tree
, start
);
1890 state
= rb_entry(node
, struct extent_state
, rb_node
);
1891 if (state
->start
!= start
) {
1895 *private = state
->private;
1897 spin_unlock(&tree
->lock
);
1902 * searches a range in the state tree for a given mask.
1903 * If 'filled' == 1, this returns 1 only if every extent in the tree
1904 * has the bits set. Otherwise, 1 is returned if any bit in the
1905 * range is found set.
1907 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1908 unsigned bits
, int filled
, struct extent_state
*cached
)
1910 struct extent_state
*state
= NULL
;
1911 struct rb_node
*node
;
1914 spin_lock(&tree
->lock
);
1915 if (cached
&& extent_state_in_tree(cached
) && cached
->start
<= start
&&
1916 cached
->end
> start
)
1917 node
= &cached
->rb_node
;
1919 node
= tree_search(tree
, start
);
1920 while (node
&& start
<= end
) {
1921 state
= rb_entry(node
, struct extent_state
, rb_node
);
1923 if (filled
&& state
->start
> start
) {
1928 if (state
->start
> end
)
1931 if (state
->state
& bits
) {
1935 } else if (filled
) {
1940 if (state
->end
== (u64
)-1)
1943 start
= state
->end
+ 1;
1946 node
= rb_next(node
);
1953 spin_unlock(&tree
->lock
);
1958 * helper function to set a given page up to date if all the
1959 * extents in the tree for that page are up to date
1961 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1963 u64 start
= page_offset(page
);
1964 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
1965 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1966 SetPageUptodate(page
);
1969 int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
)
1973 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1975 set_state_private(failure_tree
, rec
->start
, 0);
1976 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1977 rec
->start
+ rec
->len
- 1,
1978 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1982 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1983 rec
->start
+ rec
->len
- 1,
1984 EXTENT_DAMAGED
, GFP_NOFS
);
1993 * this bypasses the standard btrfs submit functions deliberately, as
1994 * the standard behavior is to write all copies in a raid setup. here we only
1995 * want to write the one bad copy. so we do the mapping for ourselves and issue
1996 * submit_bio directly.
1997 * to avoid any synchronization issues, wait for the data after writing, which
1998 * actually prevents the read that triggered the error from finishing.
1999 * currently, there can be no more than two copies of every data bit. thus,
2000 * exactly one rewrite is required.
2002 int repair_io_failure(struct inode
*inode
, u64 start
, u64 length
, u64 logical
,
2003 struct page
*page
, unsigned int pg_offset
, int mirror_num
)
2005 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2007 struct btrfs_device
*dev
;
2010 struct btrfs_bio
*bbio
= NULL
;
2011 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2014 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2015 BUG_ON(!mirror_num
);
2017 /* we can't repair anything in raid56 yet */
2018 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2021 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2024 bio
->bi_iter
.bi_size
= 0;
2025 map_length
= length
;
2027 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2028 &map_length
, &bbio
, mirror_num
);
2033 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2034 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2035 bio
->bi_iter
.bi_sector
= sector
;
2036 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2037 btrfs_put_bbio(bbio
);
2038 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2042 bio
->bi_bdev
= dev
->bdev
;
2043 bio_add_page(bio
, page
, length
, pg_offset
);
2045 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
)) {
2046 /* try to remap that extent elsewhere? */
2048 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2052 btrfs_info_rl_in_rcu(fs_info
,
2053 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2054 btrfs_ino(inode
), start
,
2055 rcu_str_deref(dev
->name
), sector
);
2060 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2063 u64 start
= eb
->start
;
2064 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2067 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2070 for (i
= 0; i
< num_pages
; i
++) {
2071 struct page
*p
= eb
->pages
[i
];
2073 ret
= repair_io_failure(root
->fs_info
->btree_inode
, start
,
2074 PAGE_CACHE_SIZE
, start
, p
,
2075 start
- page_offset(p
), mirror_num
);
2078 start
+= PAGE_CACHE_SIZE
;
2085 * each time an IO finishes, we do a fast check in the IO failure tree
2086 * to see if we need to process or clean up an io_failure_record
2088 int clean_io_failure(struct inode
*inode
, u64 start
, struct page
*page
,
2089 unsigned int pg_offset
)
2092 u64 private_failure
;
2093 struct io_failure_record
*failrec
;
2094 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2095 struct extent_state
*state
;
2100 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2101 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2105 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2110 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2111 BUG_ON(!failrec
->this_mirror
);
2113 if (failrec
->in_validation
) {
2114 /* there was no real error, just free the record */
2115 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2119 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2122 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2123 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2126 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2128 if (state
&& state
->start
<= failrec
->start
&&
2129 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2130 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2132 if (num_copies
> 1) {
2133 repair_io_failure(inode
, start
, failrec
->len
,
2134 failrec
->logical
, page
,
2135 pg_offset
, failrec
->failed_mirror
);
2140 free_io_failure(inode
, failrec
);
2146 * Can be called when
2147 * - hold extent lock
2148 * - under ordered extent
2149 * - the inode is freeing
2151 void btrfs_free_io_failure_record(struct inode
*inode
, u64 start
, u64 end
)
2153 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2154 struct io_failure_record
*failrec
;
2155 struct extent_state
*state
, *next
;
2157 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2160 spin_lock(&failure_tree
->lock
);
2161 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2163 if (state
->start
> end
)
2166 ASSERT(state
->end
<= end
);
2168 next
= next_state(state
);
2170 failrec
= (struct io_failure_record
*)(unsigned long)state
->private;
2171 free_extent_state(state
);
2176 spin_unlock(&failure_tree
->lock
);
2179 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2180 struct io_failure_record
**failrec_ret
)
2182 struct io_failure_record
*failrec
;
2184 struct extent_map
*em
;
2185 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2186 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2187 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2191 ret
= get_state_private(failure_tree
, start
, &private);
2193 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2197 failrec
->start
= start
;
2198 failrec
->len
= end
- start
+ 1;
2199 failrec
->this_mirror
= 0;
2200 failrec
->bio_flags
= 0;
2201 failrec
->in_validation
= 0;
2203 read_lock(&em_tree
->lock
);
2204 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2206 read_unlock(&em_tree
->lock
);
2211 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2212 free_extent_map(em
);
2215 read_unlock(&em_tree
->lock
);
2221 logical
= start
- em
->start
;
2222 logical
= em
->block_start
+ logical
;
2223 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2224 logical
= em
->block_start
;
2225 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2226 extent_set_compress_type(&failrec
->bio_flags
,
2230 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2231 logical
, start
, failrec
->len
);
2233 failrec
->logical
= logical
;
2234 free_extent_map(em
);
2236 /* set the bits in the private failure tree */
2237 ret
= set_extent_bits(failure_tree
, start
, end
,
2238 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2240 ret
= set_state_private(failure_tree
, start
,
2241 (u64
)(unsigned long)failrec
);
2242 /* set the bits in the inode's tree */
2244 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2251 failrec
= (struct io_failure_record
*)(unsigned long)private;
2252 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2253 failrec
->logical
, failrec
->start
, failrec
->len
,
2254 failrec
->in_validation
);
2256 * when data can be on disk more than twice, add to failrec here
2257 * (e.g. with a list for failed_mirror) to make
2258 * clean_io_failure() clean all those errors at once.
2262 *failrec_ret
= failrec
;
2267 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2268 struct io_failure_record
*failrec
, int failed_mirror
)
2272 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2273 failrec
->logical
, failrec
->len
);
2274 if (num_copies
== 1) {
2276 * we only have a single copy of the data, so don't bother with
2277 * all the retry and error correction code that follows. no
2278 * matter what the error is, it is very likely to persist.
2280 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2281 num_copies
, failrec
->this_mirror
, failed_mirror
);
2286 * there are two premises:
2287 * a) deliver good data to the caller
2288 * b) correct the bad sectors on disk
2290 if (failed_bio
->bi_vcnt
> 1) {
2292 * to fulfill b), we need to know the exact failing sectors, as
2293 * we don't want to rewrite any more than the failed ones. thus,
2294 * we need separate read requests for the failed bio
2296 * if the following BUG_ON triggers, our validation request got
2297 * merged. we need separate requests for our algorithm to work.
2299 BUG_ON(failrec
->in_validation
);
2300 failrec
->in_validation
= 1;
2301 failrec
->this_mirror
= failed_mirror
;
2304 * we're ready to fulfill a) and b) alongside. get a good copy
2305 * of the failed sector and if we succeed, we have setup
2306 * everything for repair_io_failure to do the rest for us.
2308 if (failrec
->in_validation
) {
2309 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2310 failrec
->in_validation
= 0;
2311 failrec
->this_mirror
= 0;
2313 failrec
->failed_mirror
= failed_mirror
;
2314 failrec
->this_mirror
++;
2315 if (failrec
->this_mirror
== failed_mirror
)
2316 failrec
->this_mirror
++;
2319 if (failrec
->this_mirror
> num_copies
) {
2320 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2321 num_copies
, failrec
->this_mirror
, failed_mirror
);
2329 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2330 struct io_failure_record
*failrec
,
2331 struct page
*page
, int pg_offset
, int icsum
,
2332 bio_end_io_t
*endio_func
, void *data
)
2335 struct btrfs_io_bio
*btrfs_failed_bio
;
2336 struct btrfs_io_bio
*btrfs_bio
;
2338 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2342 bio
->bi_end_io
= endio_func
;
2343 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2344 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2345 bio
->bi_iter
.bi_size
= 0;
2346 bio
->bi_private
= data
;
2348 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2349 if (btrfs_failed_bio
->csum
) {
2350 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2351 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2353 btrfs_bio
= btrfs_io_bio(bio
);
2354 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2356 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2360 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2366 * this is a generic handler for readpage errors (default
2367 * readpage_io_failed_hook). if other copies exist, read those and write back
2368 * good data to the failed position. does not investigate in remapping the
2369 * failed extent elsewhere, hoping the device will be smart enough to do this as
2373 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2374 struct page
*page
, u64 start
, u64 end
,
2377 struct io_failure_record
*failrec
;
2378 struct inode
*inode
= page
->mapping
->host
;
2379 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2384 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2386 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2390 ret
= btrfs_check_repairable(inode
, failed_bio
, failrec
, failed_mirror
);
2392 free_io_failure(inode
, failrec
);
2396 if (failed_bio
->bi_vcnt
> 1)
2397 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2399 read_mode
= READ_SYNC
;
2401 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2402 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2403 start
- page_offset(page
),
2404 (int)phy_offset
, failed_bio
->bi_end_io
,
2407 free_io_failure(inode
, failrec
);
2411 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2412 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2414 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2415 failrec
->this_mirror
,
2416 failrec
->bio_flags
, 0);
2418 free_io_failure(inode
, failrec
);
2425 /* lots and lots of room for performance fixes in the end_bio funcs */
2427 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2429 int uptodate
= (err
== 0);
2430 struct extent_io_tree
*tree
;
2433 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2435 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2436 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2437 end
, NULL
, uptodate
);
2443 ClearPageUptodate(page
);
2445 ret
= ret
< 0 ? ret
: -EIO
;
2446 mapping_set_error(page
->mapping
, ret
);
2451 * after a writepage IO is done, we need to:
2452 * clear the uptodate bits on error
2453 * clear the writeback bits in the extent tree for this IO
2454 * end_page_writeback if the page has no more pending IO
2456 * Scheduling is not allowed, so the extent state tree is expected
2457 * to have one and only one object corresponding to this IO.
2459 static void end_bio_extent_writepage(struct bio
*bio
)
2461 struct bio_vec
*bvec
;
2466 bio_for_each_segment_all(bvec
, bio
, i
) {
2467 struct page
*page
= bvec
->bv_page
;
2469 /* We always issue full-page reads, but if some block
2470 * in a page fails to read, blk_update_request() will
2471 * advance bv_offset and adjust bv_len to compensate.
2472 * Print a warning for nonzero offsets, and an error
2473 * if they don't add up to a full page. */
2474 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2475 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2476 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2477 "partial page write in btrfs with offset %u and length %u",
2478 bvec
->bv_offset
, bvec
->bv_len
);
2480 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2481 "incomplete page write in btrfs with offset %u and "
2483 bvec
->bv_offset
, bvec
->bv_len
);
2486 start
= page_offset(page
);
2487 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2489 end_extent_writepage(page
, bio
->bi_error
, start
, end
);
2490 end_page_writeback(page
);
2497 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2500 struct extent_state
*cached
= NULL
;
2501 u64 end
= start
+ len
- 1;
2503 if (uptodate
&& tree
->track_uptodate
)
2504 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2505 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2509 * after a readpage IO is done, we need to:
2510 * clear the uptodate bits on error
2511 * set the uptodate bits if things worked
2512 * set the page up to date if all extents in the tree are uptodate
2513 * clear the lock bit in the extent tree
2514 * unlock the page if there are no other extents locked for it
2516 * Scheduling is not allowed, so the extent state tree is expected
2517 * to have one and only one object corresponding to this IO.
2519 static void end_bio_extent_readpage(struct bio
*bio
)
2521 struct bio_vec
*bvec
;
2522 int uptodate
= !bio
->bi_error
;
2523 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2524 struct extent_io_tree
*tree
;
2529 u64 extent_start
= 0;
2535 bio_for_each_segment_all(bvec
, bio
, i
) {
2536 struct page
*page
= bvec
->bv_page
;
2537 struct inode
*inode
= page
->mapping
->host
;
2539 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2540 "mirror=%u\n", (u64
)bio
->bi_iter
.bi_sector
,
2541 bio
->bi_error
, io_bio
->mirror_num
);
2542 tree
= &BTRFS_I(inode
)->io_tree
;
2544 /* We always issue full-page reads, but if some block
2545 * in a page fails to read, blk_update_request() will
2546 * advance bv_offset and adjust bv_len to compensate.
2547 * Print a warning for nonzero offsets, and an error
2548 * if they don't add up to a full page. */
2549 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2550 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2551 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2552 "partial page read in btrfs with offset %u and length %u",
2553 bvec
->bv_offset
, bvec
->bv_len
);
2555 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2556 "incomplete page read in btrfs with offset %u and "
2558 bvec
->bv_offset
, bvec
->bv_len
);
2561 start
= page_offset(page
);
2562 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2565 mirror
= io_bio
->mirror_num
;
2566 if (likely(uptodate
&& tree
->ops
&&
2567 tree
->ops
->readpage_end_io_hook
)) {
2568 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2574 clean_io_failure(inode
, start
, page
, 0);
2577 if (likely(uptodate
))
2580 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2581 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2582 if (!ret
&& !bio
->bi_error
)
2586 * The generic bio_readpage_error handles errors the
2587 * following way: If possible, new read requests are
2588 * created and submitted and will end up in
2589 * end_bio_extent_readpage as well (if we're lucky, not
2590 * in the !uptodate case). In that case it returns 0 and
2591 * we just go on with the next page in our bio. If it
2592 * can't handle the error it will return -EIO and we
2593 * remain responsible for that page.
2595 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2598 uptodate
= !bio
->bi_error
;
2604 if (likely(uptodate
)) {
2605 loff_t i_size
= i_size_read(inode
);
2606 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2609 /* Zero out the end if this page straddles i_size */
2610 off
= i_size
& (PAGE_CACHE_SIZE
-1);
2611 if (page
->index
== end_index
&& off
)
2612 zero_user_segment(page
, off
, PAGE_CACHE_SIZE
);
2613 SetPageUptodate(page
);
2615 ClearPageUptodate(page
);
2621 if (unlikely(!uptodate
)) {
2623 endio_readpage_release_extent(tree
,
2629 endio_readpage_release_extent(tree
, start
,
2630 end
- start
+ 1, 0);
2631 } else if (!extent_len
) {
2632 extent_start
= start
;
2633 extent_len
= end
+ 1 - start
;
2634 } else if (extent_start
+ extent_len
== start
) {
2635 extent_len
+= end
+ 1 - start
;
2637 endio_readpage_release_extent(tree
, extent_start
,
2638 extent_len
, uptodate
);
2639 extent_start
= start
;
2640 extent_len
= end
+ 1 - start
;
2645 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2648 io_bio
->end_io(io_bio
, bio
->bi_error
);
2653 * this allocates from the btrfs_bioset. We're returning a bio right now
2654 * but you can call btrfs_io_bio for the appropriate container_of magic
2657 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2660 struct btrfs_io_bio
*btrfs_bio
;
2663 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2665 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2666 while (!bio
&& (nr_vecs
/= 2)) {
2667 bio
= bio_alloc_bioset(gfp_flags
,
2668 nr_vecs
, btrfs_bioset
);
2673 bio
->bi_bdev
= bdev
;
2674 bio
->bi_iter
.bi_sector
= first_sector
;
2675 btrfs_bio
= btrfs_io_bio(bio
);
2676 btrfs_bio
->csum
= NULL
;
2677 btrfs_bio
->csum_allocated
= NULL
;
2678 btrfs_bio
->end_io
= NULL
;
2683 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2685 struct btrfs_io_bio
*btrfs_bio
;
2688 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2690 btrfs_bio
= btrfs_io_bio(new);
2691 btrfs_bio
->csum
= NULL
;
2692 btrfs_bio
->csum_allocated
= NULL
;
2693 btrfs_bio
->end_io
= NULL
;
2695 #ifdef CONFIG_BLK_CGROUP
2696 /* FIXME, put this into bio_clone_bioset */
2698 bio_associate_blkcg(new, bio
->bi_css
);
2704 /* this also allocates from the btrfs_bioset */
2705 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2707 struct btrfs_io_bio
*btrfs_bio
;
2710 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2712 btrfs_bio
= btrfs_io_bio(bio
);
2713 btrfs_bio
->csum
= NULL
;
2714 btrfs_bio
->csum_allocated
= NULL
;
2715 btrfs_bio
->end_io
= NULL
;
2721 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2722 int mirror_num
, unsigned long bio_flags
)
2725 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2726 struct page
*page
= bvec
->bv_page
;
2727 struct extent_io_tree
*tree
= bio
->bi_private
;
2730 start
= page_offset(page
) + bvec
->bv_offset
;
2732 bio
->bi_private
= NULL
;
2736 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2737 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2738 mirror_num
, bio_flags
, start
);
2740 btrfsic_submit_bio(rw
, bio
);
2746 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2747 unsigned long offset
, size_t size
, struct bio
*bio
,
2748 unsigned long bio_flags
)
2751 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2752 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2759 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2760 struct writeback_control
*wbc
,
2761 struct page
*page
, sector_t sector
,
2762 size_t size
, unsigned long offset
,
2763 struct block_device
*bdev
,
2764 struct bio
**bio_ret
,
2765 unsigned long max_pages
,
2766 bio_end_io_t end_io_func
,
2768 unsigned long prev_bio_flags
,
2769 unsigned long bio_flags
,
2770 bool force_bio_submit
)
2775 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2776 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2778 if (bio_ret
&& *bio_ret
) {
2781 contig
= bio
->bi_iter
.bi_sector
== sector
;
2783 contig
= bio_end_sector(bio
) == sector
;
2785 if (prev_bio_flags
!= bio_flags
|| !contig
||
2787 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2788 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2789 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2798 wbc_account_io(wbc
, page
, page_size
);
2803 bio
= btrfs_bio_alloc(bdev
, sector
, BIO_MAX_PAGES
,
2804 GFP_NOFS
| __GFP_HIGH
);
2808 bio_add_page(bio
, page
, page_size
, offset
);
2809 bio
->bi_end_io
= end_io_func
;
2810 bio
->bi_private
= tree
;
2812 wbc_init_bio(wbc
, bio
);
2813 wbc_account_io(wbc
, page
, page_size
);
2819 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2824 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2827 if (!PagePrivate(page
)) {
2828 SetPagePrivate(page
);
2829 page_cache_get(page
);
2830 set_page_private(page
, (unsigned long)eb
);
2832 WARN_ON(page
->private != (unsigned long)eb
);
2836 void set_page_extent_mapped(struct page
*page
)
2838 if (!PagePrivate(page
)) {
2839 SetPagePrivate(page
);
2840 page_cache_get(page
);
2841 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2845 static struct extent_map
*
2846 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2847 u64 start
, u64 len
, get_extent_t
*get_extent
,
2848 struct extent_map
**em_cached
)
2850 struct extent_map
*em
;
2852 if (em_cached
&& *em_cached
) {
2854 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2855 start
< extent_map_end(em
)) {
2856 atomic_inc(&em
->refs
);
2860 free_extent_map(em
);
2864 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2865 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2867 atomic_inc(&em
->refs
);
2873 * basic readpage implementation. Locked extent state structs are inserted
2874 * into the tree that are removed when the IO is done (by the end_io
2876 * XXX JDM: This needs looking at to ensure proper page locking
2878 static int __do_readpage(struct extent_io_tree
*tree
,
2880 get_extent_t
*get_extent
,
2881 struct extent_map
**em_cached
,
2882 struct bio
**bio
, int mirror_num
,
2883 unsigned long *bio_flags
, int rw
,
2886 struct inode
*inode
= page
->mapping
->host
;
2887 u64 start
= page_offset(page
);
2888 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2892 u64 last_byte
= i_size_read(inode
);
2896 struct extent_map
*em
;
2897 struct block_device
*bdev
;
2900 size_t pg_offset
= 0;
2902 size_t disk_io_size
;
2903 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2904 unsigned long this_bio_flag
= 0;
2906 set_page_extent_mapped(page
);
2909 if (!PageUptodate(page
)) {
2910 if (cleancache_get_page(page
) == 0) {
2911 BUG_ON(blocksize
!= PAGE_SIZE
);
2912 unlock_extent(tree
, start
, end
);
2917 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
2919 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
2922 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
2923 userpage
= kmap_atomic(page
);
2924 memset(userpage
+ zero_offset
, 0, iosize
);
2925 flush_dcache_page(page
);
2926 kunmap_atomic(userpage
);
2929 while (cur
<= end
) {
2930 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
2931 bool force_bio_submit
= false;
2933 if (cur
>= last_byte
) {
2935 struct extent_state
*cached
= NULL
;
2937 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
2938 userpage
= kmap_atomic(page
);
2939 memset(userpage
+ pg_offset
, 0, iosize
);
2940 flush_dcache_page(page
);
2941 kunmap_atomic(userpage
);
2942 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2944 unlock_extent_cached(tree
, cur
,
2949 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2950 end
- cur
+ 1, get_extent
, em_cached
);
2951 if (IS_ERR_OR_NULL(em
)) {
2953 unlock_extent(tree
, cur
, end
);
2956 extent_offset
= cur
- em
->start
;
2957 BUG_ON(extent_map_end(em
) <= cur
);
2960 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2961 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2962 extent_set_compress_type(&this_bio_flag
,
2966 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2967 cur_end
= min(extent_map_end(em
) - 1, end
);
2968 iosize
= ALIGN(iosize
, blocksize
);
2969 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2970 disk_io_size
= em
->block_len
;
2971 sector
= em
->block_start
>> 9;
2973 sector
= (em
->block_start
+ extent_offset
) >> 9;
2974 disk_io_size
= iosize
;
2977 block_start
= em
->block_start
;
2978 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2979 block_start
= EXTENT_MAP_HOLE
;
2982 * If we have a file range that points to a compressed extent
2983 * and it's followed by a consecutive file range that points to
2984 * to the same compressed extent (possibly with a different
2985 * offset and/or length, so it either points to the whole extent
2986 * or only part of it), we must make sure we do not submit a
2987 * single bio to populate the pages for the 2 ranges because
2988 * this makes the compressed extent read zero out the pages
2989 * belonging to the 2nd range. Imagine the following scenario:
2992 * [0 - 8K] [8K - 24K]
2995 * points to extent X, points to extent X,
2996 * offset 4K, length of 8K offset 0, length 16K
2998 * [extent X, compressed length = 4K uncompressed length = 16K]
3000 * If the bio to read the compressed extent covers both ranges,
3001 * it will decompress extent X into the pages belonging to the
3002 * first range and then it will stop, zeroing out the remaining
3003 * pages that belong to the other range that points to extent X.
3004 * So here we make sure we submit 2 bios, one for the first
3005 * range and another one for the third range. Both will target
3006 * the same physical extent from disk, but we can't currently
3007 * make the compressed bio endio callback populate the pages
3008 * for both ranges because each compressed bio is tightly
3009 * coupled with a single extent map, and each range can have
3010 * an extent map with a different offset value relative to the
3011 * uncompressed data of our extent and different lengths. This
3012 * is a corner case so we prioritize correctness over
3013 * non-optimal behavior (submitting 2 bios for the same extent).
3015 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3016 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3017 *prev_em_start
!= em
->orig_start
)
3018 force_bio_submit
= true;
3021 *prev_em_start
= em
->orig_start
;
3023 free_extent_map(em
);
3026 /* we've found a hole, just zero and go on */
3027 if (block_start
== EXTENT_MAP_HOLE
) {
3029 struct extent_state
*cached
= NULL
;
3031 userpage
= kmap_atomic(page
);
3032 memset(userpage
+ pg_offset
, 0, iosize
);
3033 flush_dcache_page(page
);
3034 kunmap_atomic(userpage
);
3036 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3038 unlock_extent_cached(tree
, cur
,
3042 pg_offset
+= iosize
;
3045 /* the get_extent function already copied into the page */
3046 if (test_range_bit(tree
, cur
, cur_end
,
3047 EXTENT_UPTODATE
, 1, NULL
)) {
3048 check_page_uptodate(tree
, page
);
3049 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3051 pg_offset
+= iosize
;
3054 /* we have an inline extent but it didn't get marked up
3055 * to date. Error out
3057 if (block_start
== EXTENT_MAP_INLINE
) {
3059 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3061 pg_offset
+= iosize
;
3066 ret
= submit_extent_page(rw
, tree
, NULL
, page
,
3067 sector
, disk_io_size
, pg_offset
,
3069 end_bio_extent_readpage
, mirror_num
,
3075 *bio_flags
= this_bio_flag
;
3078 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3081 pg_offset
+= iosize
;
3085 if (!PageError(page
))
3086 SetPageUptodate(page
);
3092 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3093 struct page
*pages
[], int nr_pages
,
3095 get_extent_t
*get_extent
,
3096 struct extent_map
**em_cached
,
3097 struct bio
**bio
, int mirror_num
,
3098 unsigned long *bio_flags
, int rw
,
3101 struct inode
*inode
;
3102 struct btrfs_ordered_extent
*ordered
;
3105 inode
= pages
[0]->mapping
->host
;
3107 lock_extent(tree
, start
, end
);
3108 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3112 unlock_extent(tree
, start
, end
);
3113 btrfs_start_ordered_extent(inode
, ordered
, 1);
3114 btrfs_put_ordered_extent(ordered
);
3117 for (index
= 0; index
< nr_pages
; index
++) {
3118 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3119 mirror_num
, bio_flags
, rw
, prev_em_start
);
3120 page_cache_release(pages
[index
]);
3124 static void __extent_readpages(struct extent_io_tree
*tree
,
3125 struct page
*pages
[],
3126 int nr_pages
, get_extent_t
*get_extent
,
3127 struct extent_map
**em_cached
,
3128 struct bio
**bio
, int mirror_num
,
3129 unsigned long *bio_flags
, int rw
,
3136 int first_index
= 0;
3138 for (index
= 0; index
< nr_pages
; index
++) {
3139 page_start
= page_offset(pages
[index
]);
3142 end
= start
+ PAGE_CACHE_SIZE
- 1;
3143 first_index
= index
;
3144 } else if (end
+ 1 == page_start
) {
3145 end
+= PAGE_CACHE_SIZE
;
3147 __do_contiguous_readpages(tree
, &pages
[first_index
],
3148 index
- first_index
, start
,
3149 end
, get_extent
, em_cached
,
3150 bio
, mirror_num
, bio_flags
,
3153 end
= start
+ PAGE_CACHE_SIZE
- 1;
3154 first_index
= index
;
3159 __do_contiguous_readpages(tree
, &pages
[first_index
],
3160 index
- first_index
, start
,
3161 end
, get_extent
, em_cached
, bio
,
3162 mirror_num
, bio_flags
, rw
,
3166 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3168 get_extent_t
*get_extent
,
3169 struct bio
**bio
, int mirror_num
,
3170 unsigned long *bio_flags
, int rw
)
3172 struct inode
*inode
= page
->mapping
->host
;
3173 struct btrfs_ordered_extent
*ordered
;
3174 u64 start
= page_offset(page
);
3175 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3179 lock_extent(tree
, start
, end
);
3180 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
3183 unlock_extent(tree
, start
, end
);
3184 btrfs_start_ordered_extent(inode
, ordered
, 1);
3185 btrfs_put_ordered_extent(ordered
);
3188 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3189 bio_flags
, rw
, NULL
);
3193 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3194 get_extent_t
*get_extent
, int mirror_num
)
3196 struct bio
*bio
= NULL
;
3197 unsigned long bio_flags
= 0;
3200 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3203 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3207 static noinline
void update_nr_written(struct page
*page
,
3208 struct writeback_control
*wbc
,
3209 unsigned long nr_written
)
3211 wbc
->nr_to_write
-= nr_written
;
3212 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
3213 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
3214 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
3218 * helper for __extent_writepage, doing all of the delayed allocation setup.
3220 * This returns 1 if our fill_delalloc function did all the work required
3221 * to write the page (copy into inline extent). In this case the IO has
3222 * been started and the page is already unlocked.
3224 * This returns 0 if all went well (page still locked)
3225 * This returns < 0 if there were errors (page still locked)
3227 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3228 struct page
*page
, struct writeback_control
*wbc
,
3229 struct extent_page_data
*epd
,
3231 unsigned long *nr_written
)
3233 struct extent_io_tree
*tree
= epd
->tree
;
3234 u64 page_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
3236 u64 delalloc_to_write
= 0;
3237 u64 delalloc_end
= 0;
3239 int page_started
= 0;
3241 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3244 while (delalloc_end
< page_end
) {
3245 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3249 BTRFS_MAX_EXTENT_SIZE
);
3250 if (nr_delalloc
== 0) {
3251 delalloc_start
= delalloc_end
+ 1;
3254 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3259 /* File system has been set read-only */
3262 /* fill_delalloc should be return < 0 for error
3263 * but just in case, we use > 0 here meaning the
3264 * IO is started, so we don't want to return > 0
3265 * unless things are going well.
3267 ret
= ret
< 0 ? ret
: -EIO
;
3271 * delalloc_end is already one less than the total
3272 * length, so we don't subtract one from
3275 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3278 delalloc_start
= delalloc_end
+ 1;
3280 if (wbc
->nr_to_write
< delalloc_to_write
) {
3283 if (delalloc_to_write
< thresh
* 2)
3284 thresh
= delalloc_to_write
;
3285 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3289 /* did the fill delalloc function already unlock and start
3294 * we've unlocked the page, so we can't update
3295 * the mapping's writeback index, just update
3298 wbc
->nr_to_write
-= *nr_written
;
3309 * helper for __extent_writepage. This calls the writepage start hooks,
3310 * and does the loop to map the page into extents and bios.
3312 * We return 1 if the IO is started and the page is unlocked,
3313 * 0 if all went well (page still locked)
3314 * < 0 if there were errors (page still locked)
3316 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3318 struct writeback_control
*wbc
,
3319 struct extent_page_data
*epd
,
3321 unsigned long nr_written
,
3322 int write_flags
, int *nr_ret
)
3324 struct extent_io_tree
*tree
= epd
->tree
;
3325 u64 start
= page_offset(page
);
3326 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3333 struct extent_state
*cached_state
= NULL
;
3334 struct extent_map
*em
;
3335 struct block_device
*bdev
;
3336 size_t pg_offset
= 0;
3342 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3343 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3346 /* Fixup worker will requeue */
3348 wbc
->pages_skipped
++;
3350 redirty_page_for_writepage(wbc
, page
);
3352 update_nr_written(page
, wbc
, nr_written
);
3360 * we don't want to touch the inode after unlocking the page,
3361 * so we update the mapping writeback index now
3363 update_nr_written(page
, wbc
, nr_written
+ 1);
3366 if (i_size
<= start
) {
3367 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3368 tree
->ops
->writepage_end_io_hook(page
, start
,
3373 blocksize
= inode
->i_sb
->s_blocksize
;
3375 while (cur
<= end
) {
3377 if (cur
>= i_size
) {
3378 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3379 tree
->ops
->writepage_end_io_hook(page
, cur
,
3383 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3385 if (IS_ERR_OR_NULL(em
)) {
3387 ret
= PTR_ERR_OR_ZERO(em
);
3391 extent_offset
= cur
- em
->start
;
3392 em_end
= extent_map_end(em
);
3393 BUG_ON(em_end
<= cur
);
3395 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3396 iosize
= ALIGN(iosize
, blocksize
);
3397 sector
= (em
->block_start
+ extent_offset
) >> 9;
3399 block_start
= em
->block_start
;
3400 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3401 free_extent_map(em
);
3405 * compressed and inline extents are written through other
3408 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3409 block_start
== EXTENT_MAP_INLINE
) {
3411 * end_io notification does not happen here for
3412 * compressed extents
3414 if (!compressed
&& tree
->ops
&&
3415 tree
->ops
->writepage_end_io_hook
)
3416 tree
->ops
->writepage_end_io_hook(page
, cur
,
3419 else if (compressed
) {
3420 /* we don't want to end_page_writeback on
3421 * a compressed extent. this happens
3428 pg_offset
+= iosize
;
3432 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3433 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3441 unsigned long max_nr
= (i_size
>> PAGE_CACHE_SHIFT
) + 1;
3443 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3444 if (!PageWriteback(page
)) {
3445 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3446 "page %lu not writeback, cur %llu end %llu",
3447 page
->index
, cur
, end
);
3450 ret
= submit_extent_page(write_flags
, tree
, wbc
, page
,
3451 sector
, iosize
, pg_offset
,
3452 bdev
, &epd
->bio
, max_nr
,
3453 end_bio_extent_writepage
,
3459 pg_offset
+= iosize
;
3467 /* drop our reference on any cached states */
3468 free_extent_state(cached_state
);
3473 * the writepage semantics are similar to regular writepage. extent
3474 * records are inserted to lock ranges in the tree, and as dirty areas
3475 * are found, they are marked writeback. Then the lock bits are removed
3476 * and the end_io handler clears the writeback ranges
3478 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3481 struct inode
*inode
= page
->mapping
->host
;
3482 struct extent_page_data
*epd
= data
;
3483 u64 start
= page_offset(page
);
3484 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3487 size_t pg_offset
= 0;
3488 loff_t i_size
= i_size_read(inode
);
3489 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
3491 unsigned long nr_written
= 0;
3493 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3494 write_flags
= WRITE_SYNC
;
3496 write_flags
= WRITE
;
3498 trace___extent_writepage(page
, inode
, wbc
);
3500 WARN_ON(!PageLocked(page
));
3502 ClearPageError(page
);
3504 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
3505 if (page
->index
> end_index
||
3506 (page
->index
== end_index
&& !pg_offset
)) {
3507 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
3512 if (page
->index
== end_index
) {
3515 userpage
= kmap_atomic(page
);
3516 memset(userpage
+ pg_offset
, 0,
3517 PAGE_CACHE_SIZE
- pg_offset
);
3518 kunmap_atomic(userpage
);
3519 flush_dcache_page(page
);
3524 set_page_extent_mapped(page
);
3526 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3532 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3533 i_size
, nr_written
, write_flags
, &nr
);
3539 /* make sure the mapping tag for page dirty gets cleared */
3540 set_page_writeback(page
);
3541 end_page_writeback(page
);
3543 if (PageError(page
)) {
3544 ret
= ret
< 0 ? ret
: -EIO
;
3545 end_extent_writepage(page
, ret
, start
, page_end
);
3554 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3556 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3557 TASK_UNINTERRUPTIBLE
);
3560 static noinline_for_stack
int
3561 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3562 struct btrfs_fs_info
*fs_info
,
3563 struct extent_page_data
*epd
)
3565 unsigned long i
, num_pages
;
3569 if (!btrfs_try_tree_write_lock(eb
)) {
3571 flush_write_bio(epd
);
3572 btrfs_tree_lock(eb
);
3575 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3576 btrfs_tree_unlock(eb
);
3580 flush_write_bio(epd
);
3584 wait_on_extent_buffer_writeback(eb
);
3585 btrfs_tree_lock(eb
);
3586 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3588 btrfs_tree_unlock(eb
);
3593 * We need to do this to prevent races in people who check if the eb is
3594 * under IO since we can end up having no IO bits set for a short period
3597 spin_lock(&eb
->refs_lock
);
3598 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3599 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3600 spin_unlock(&eb
->refs_lock
);
3601 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3602 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3604 fs_info
->dirty_metadata_batch
);
3607 spin_unlock(&eb
->refs_lock
);
3610 btrfs_tree_unlock(eb
);
3615 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3616 for (i
= 0; i
< num_pages
; i
++) {
3617 struct page
*p
= eb
->pages
[i
];
3619 if (!trylock_page(p
)) {
3621 flush_write_bio(epd
);
3631 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3633 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3634 smp_mb__after_atomic();
3635 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3638 static void set_btree_ioerr(struct page
*page
)
3640 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3641 struct btrfs_inode
*btree_ino
= BTRFS_I(eb
->fs_info
->btree_inode
);
3644 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3648 * If writeback for a btree extent that doesn't belong to a log tree
3649 * failed, increment the counter transaction->eb_write_errors.
3650 * We do this because while the transaction is running and before it's
3651 * committing (when we call filemap_fdata[write|wait]_range against
3652 * the btree inode), we might have
3653 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3654 * returns an error or an error happens during writeback, when we're
3655 * committing the transaction we wouldn't know about it, since the pages
3656 * can be no longer dirty nor marked anymore for writeback (if a
3657 * subsequent modification to the extent buffer didn't happen before the
3658 * transaction commit), which makes filemap_fdata[write|wait]_range not
3659 * able to find the pages tagged with SetPageError at transaction
3660 * commit time. So if this happens we must abort the transaction,
3661 * otherwise we commit a super block with btree roots that point to
3662 * btree nodes/leafs whose content on disk is invalid - either garbage
3663 * or the content of some node/leaf from a past generation that got
3664 * cowed or deleted and is no longer valid.
3666 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3667 * not be enough - we need to distinguish between log tree extents vs
3668 * non-log tree extents, and the next filemap_fdatawait_range() call
3669 * will catch and clear such errors in the mapping - and that call might
3670 * be from a log sync and not from a transaction commit. Also, checking
3671 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3672 * not done and would not be reliable - the eb might have been released
3673 * from memory and reading it back again means that flag would not be
3674 * set (since it's a runtime flag, not persisted on disk).
3676 * Using the flags below in the btree inode also makes us achieve the
3677 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3678 * writeback for all dirty pages and before filemap_fdatawait_range()
3679 * is called, the writeback for all dirty pages had already finished
3680 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3681 * filemap_fdatawait_range() would return success, as it could not know
3682 * that writeback errors happened (the pages were no longer tagged for
3685 switch (eb
->log_index
) {
3687 set_bit(BTRFS_INODE_BTREE_ERR
, &btree_ino
->runtime_flags
);
3690 set_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
3693 set_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
3696 BUG(); /* unexpected, logic error */
3700 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3702 struct bio_vec
*bvec
;
3703 struct extent_buffer
*eb
;
3706 bio_for_each_segment_all(bvec
, bio
, i
) {
3707 struct page
*page
= bvec
->bv_page
;
3709 eb
= (struct extent_buffer
*)page
->private;
3711 done
= atomic_dec_and_test(&eb
->io_pages
);
3713 if (bio
->bi_error
||
3714 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3715 ClearPageUptodate(page
);
3716 set_btree_ioerr(page
);
3719 end_page_writeback(page
);
3724 end_extent_buffer_writeback(eb
);
3730 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3731 struct btrfs_fs_info
*fs_info
,
3732 struct writeback_control
*wbc
,
3733 struct extent_page_data
*epd
)
3735 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3736 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3737 u64 offset
= eb
->start
;
3738 unsigned long i
, num_pages
;
3739 unsigned long bio_flags
= 0;
3740 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3743 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3744 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3745 atomic_set(&eb
->io_pages
, num_pages
);
3746 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3747 bio_flags
= EXTENT_BIO_TREE_LOG
;
3749 for (i
= 0; i
< num_pages
; i
++) {
3750 struct page
*p
= eb
->pages
[i
];
3752 clear_page_dirty_for_io(p
);
3753 set_page_writeback(p
);
3754 ret
= submit_extent_page(rw
, tree
, wbc
, p
, offset
>> 9,
3755 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3756 -1, end_bio_extent_buffer_writepage
,
3757 0, epd
->bio_flags
, bio_flags
, false);
3758 epd
->bio_flags
= bio_flags
;
3761 end_page_writeback(p
);
3762 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3763 end_extent_buffer_writeback(eb
);
3767 offset
+= PAGE_CACHE_SIZE
;
3768 update_nr_written(p
, wbc
, 1);
3772 if (unlikely(ret
)) {
3773 for (; i
< num_pages
; i
++) {
3774 struct page
*p
= eb
->pages
[i
];
3775 clear_page_dirty_for_io(p
);
3783 int btree_write_cache_pages(struct address_space
*mapping
,
3784 struct writeback_control
*wbc
)
3786 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3787 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3788 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3789 struct extent_page_data epd
= {
3793 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3798 int nr_to_write_done
= 0;
3799 struct pagevec pvec
;
3802 pgoff_t end
; /* Inclusive */
3806 pagevec_init(&pvec
, 0);
3807 if (wbc
->range_cyclic
) {
3808 index
= mapping
->writeback_index
; /* Start from prev offset */
3811 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3812 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3815 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3816 tag
= PAGECACHE_TAG_TOWRITE
;
3818 tag
= PAGECACHE_TAG_DIRTY
;
3820 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3821 tag_pages_for_writeback(mapping
, index
, end
);
3822 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3823 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3824 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3828 for (i
= 0; i
< nr_pages
; i
++) {
3829 struct page
*page
= pvec
.pages
[i
];
3831 if (!PagePrivate(page
))
3834 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3839 spin_lock(&mapping
->private_lock
);
3840 if (!PagePrivate(page
)) {
3841 spin_unlock(&mapping
->private_lock
);
3845 eb
= (struct extent_buffer
*)page
->private;
3848 * Shouldn't happen and normally this would be a BUG_ON
3849 * but no sense in crashing the users box for something
3850 * we can survive anyway.
3853 spin_unlock(&mapping
->private_lock
);
3857 if (eb
== prev_eb
) {
3858 spin_unlock(&mapping
->private_lock
);
3862 ret
= atomic_inc_not_zero(&eb
->refs
);
3863 spin_unlock(&mapping
->private_lock
);
3868 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3870 free_extent_buffer(eb
);
3874 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3877 free_extent_buffer(eb
);
3880 free_extent_buffer(eb
);
3883 * the filesystem may choose to bump up nr_to_write.
3884 * We have to make sure to honor the new nr_to_write
3887 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3889 pagevec_release(&pvec
);
3892 if (!scanned
&& !done
) {
3894 * We hit the last page and there is more work to be done: wrap
3895 * back to the start of the file
3901 flush_write_bio(&epd
);
3906 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3907 * @mapping: address space structure to write
3908 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3909 * @writepage: function called for each page
3910 * @data: data passed to writepage function
3912 * If a page is already under I/O, write_cache_pages() skips it, even
3913 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3914 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3915 * and msync() need to guarantee that all the data which was dirty at the time
3916 * the call was made get new I/O started against them. If wbc->sync_mode is
3917 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3918 * existing IO to complete.
3920 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3921 struct address_space
*mapping
,
3922 struct writeback_control
*wbc
,
3923 writepage_t writepage
, void *data
,
3924 void (*flush_fn
)(void *))
3926 struct inode
*inode
= mapping
->host
;
3930 int nr_to_write_done
= 0;
3931 struct pagevec pvec
;
3934 pgoff_t end
; /* Inclusive */
3939 * We have to hold onto the inode so that ordered extents can do their
3940 * work when the IO finishes. The alternative to this is failing to add
3941 * an ordered extent if the igrab() fails there and that is a huge pain
3942 * to deal with, so instead just hold onto the inode throughout the
3943 * writepages operation. If it fails here we are freeing up the inode
3944 * anyway and we'd rather not waste our time writing out stuff that is
3945 * going to be truncated anyway.
3950 pagevec_init(&pvec
, 0);
3951 if (wbc
->range_cyclic
) {
3952 index
= mapping
->writeback_index
; /* Start from prev offset */
3955 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3956 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3959 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3960 tag
= PAGECACHE_TAG_TOWRITE
;
3962 tag
= PAGECACHE_TAG_DIRTY
;
3964 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3965 tag_pages_for_writeback(mapping
, index
, end
);
3966 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3967 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3968 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3972 for (i
= 0; i
< nr_pages
; i
++) {
3973 struct page
*page
= pvec
.pages
[i
];
3976 * At this point we hold neither mapping->tree_lock nor
3977 * lock on the page itself: the page may be truncated or
3978 * invalidated (changing page->mapping to NULL), or even
3979 * swizzled back from swapper_space to tmpfs file
3982 if (!trylock_page(page
)) {
3987 if (unlikely(page
->mapping
!= mapping
)) {
3992 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3998 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3999 if (PageWriteback(page
))
4001 wait_on_page_writeback(page
);
4004 if (PageWriteback(page
) ||
4005 !clear_page_dirty_for_io(page
)) {
4010 ret
= (*writepage
)(page
, wbc
, data
);
4012 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4016 if (!err
&& ret
< 0)
4020 * the filesystem may choose to bump up nr_to_write.
4021 * We have to make sure to honor the new nr_to_write
4024 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4026 pagevec_release(&pvec
);
4029 if (!scanned
&& !done
&& !err
) {
4031 * We hit the last page and there is more work to be done: wrap
4032 * back to the start of the file
4038 btrfs_add_delayed_iput(inode
);
4042 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4051 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
4052 BUG_ON(ret
< 0); /* -ENOMEM */
4057 static noinline
void flush_write_bio(void *data
)
4059 struct extent_page_data
*epd
= data
;
4060 flush_epd_write_bio(epd
);
4063 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4064 get_extent_t
*get_extent
,
4065 struct writeback_control
*wbc
)
4068 struct extent_page_data epd
= {
4071 .get_extent
= get_extent
,
4073 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4077 ret
= __extent_writepage(page
, wbc
, &epd
);
4079 flush_epd_write_bio(&epd
);
4083 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4084 u64 start
, u64 end
, get_extent_t
*get_extent
,
4088 struct address_space
*mapping
= inode
->i_mapping
;
4090 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
4093 struct extent_page_data epd
= {
4096 .get_extent
= get_extent
,
4098 .sync_io
= mode
== WB_SYNC_ALL
,
4101 struct writeback_control wbc_writepages
= {
4103 .nr_to_write
= nr_pages
* 2,
4104 .range_start
= start
,
4105 .range_end
= end
+ 1,
4108 while (start
<= end
) {
4109 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
4110 if (clear_page_dirty_for_io(page
))
4111 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4113 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4114 tree
->ops
->writepage_end_io_hook(page
, start
,
4115 start
+ PAGE_CACHE_SIZE
- 1,
4119 page_cache_release(page
);
4120 start
+= PAGE_CACHE_SIZE
;
4123 flush_epd_write_bio(&epd
);
4127 int extent_writepages(struct extent_io_tree
*tree
,
4128 struct address_space
*mapping
,
4129 get_extent_t
*get_extent
,
4130 struct writeback_control
*wbc
)
4133 struct extent_page_data epd
= {
4136 .get_extent
= get_extent
,
4138 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4142 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
4143 __extent_writepage
, &epd
,
4145 flush_epd_write_bio(&epd
);
4149 int extent_readpages(struct extent_io_tree
*tree
,
4150 struct address_space
*mapping
,
4151 struct list_head
*pages
, unsigned nr_pages
,
4152 get_extent_t get_extent
)
4154 struct bio
*bio
= NULL
;
4156 unsigned long bio_flags
= 0;
4157 struct page
*pagepool
[16];
4159 struct extent_map
*em_cached
= NULL
;
4161 u64 prev_em_start
= (u64
)-1;
4163 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4164 page
= list_entry(pages
->prev
, struct page
, lru
);
4166 prefetchw(&page
->flags
);
4167 list_del(&page
->lru
);
4168 if (add_to_page_cache_lru(page
, mapping
,
4169 page
->index
, GFP_NOFS
)) {
4170 page_cache_release(page
);
4174 pagepool
[nr
++] = page
;
4175 if (nr
< ARRAY_SIZE(pagepool
))
4177 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4178 &bio
, 0, &bio_flags
, READ
, &prev_em_start
);
4182 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4183 &bio
, 0, &bio_flags
, READ
, &prev_em_start
);
4186 free_extent_map(em_cached
);
4188 BUG_ON(!list_empty(pages
));
4190 return submit_one_bio(READ
, bio
, 0, bio_flags
);
4195 * basic invalidatepage code, this waits on any locked or writeback
4196 * ranges corresponding to the page, and then deletes any extent state
4197 * records from the tree
4199 int extent_invalidatepage(struct extent_io_tree
*tree
,
4200 struct page
*page
, unsigned long offset
)
4202 struct extent_state
*cached_state
= NULL
;
4203 u64 start
= page_offset(page
);
4204 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4205 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4207 start
+= ALIGN(offset
, blocksize
);
4211 lock_extent_bits(tree
, start
, end
, &cached_state
);
4212 wait_on_page_writeback(page
);
4213 clear_extent_bit(tree
, start
, end
,
4214 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4215 EXTENT_DO_ACCOUNTING
,
4216 1, 1, &cached_state
, GFP_NOFS
);
4221 * a helper for releasepage, this tests for areas of the page that
4222 * are locked or under IO and drops the related state bits if it is safe
4225 static int try_release_extent_state(struct extent_map_tree
*map
,
4226 struct extent_io_tree
*tree
,
4227 struct page
*page
, gfp_t mask
)
4229 u64 start
= page_offset(page
);
4230 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4233 if (test_range_bit(tree
, start
, end
,
4234 EXTENT_IOBITS
, 0, NULL
))
4237 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4240 * at this point we can safely clear everything except the
4241 * locked bit and the nodatasum bit
4243 ret
= clear_extent_bit(tree
, start
, end
,
4244 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4247 /* if clear_extent_bit failed for enomem reasons,
4248 * we can't allow the release to continue.
4259 * a helper for releasepage. As long as there are no locked extents
4260 * in the range corresponding to the page, both state records and extent
4261 * map records are removed
4263 int try_release_extent_mapping(struct extent_map_tree
*map
,
4264 struct extent_io_tree
*tree
, struct page
*page
,
4267 struct extent_map
*em
;
4268 u64 start
= page_offset(page
);
4269 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4271 if (gfpflags_allow_blocking(mask
) &&
4272 page
->mapping
->host
->i_size
> SZ_16M
) {
4274 while (start
<= end
) {
4275 len
= end
- start
+ 1;
4276 write_lock(&map
->lock
);
4277 em
= lookup_extent_mapping(map
, start
, len
);
4279 write_unlock(&map
->lock
);
4282 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4283 em
->start
!= start
) {
4284 write_unlock(&map
->lock
);
4285 free_extent_map(em
);
4288 if (!test_range_bit(tree
, em
->start
,
4289 extent_map_end(em
) - 1,
4290 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4292 remove_extent_mapping(map
, em
);
4293 /* once for the rb tree */
4294 free_extent_map(em
);
4296 start
= extent_map_end(em
);
4297 write_unlock(&map
->lock
);
4300 free_extent_map(em
);
4303 return try_release_extent_state(map
, tree
, page
, mask
);
4307 * helper function for fiemap, which doesn't want to see any holes.
4308 * This maps until we find something past 'last'
4310 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4313 get_extent_t
*get_extent
)
4315 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4316 struct extent_map
*em
;
4323 len
= last
- offset
;
4326 len
= ALIGN(len
, sectorsize
);
4327 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4328 if (IS_ERR_OR_NULL(em
))
4331 /* if this isn't a hole return it */
4332 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4333 em
->block_start
!= EXTENT_MAP_HOLE
) {
4337 /* this is a hole, advance to the next extent */
4338 offset
= extent_map_end(em
);
4339 free_extent_map(em
);
4346 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4347 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4351 u64 max
= start
+ len
;
4355 u64 last_for_get_extent
= 0;
4357 u64 isize
= i_size_read(inode
);
4358 struct btrfs_key found_key
;
4359 struct extent_map
*em
= NULL
;
4360 struct extent_state
*cached_state
= NULL
;
4361 struct btrfs_path
*path
;
4362 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4371 path
= btrfs_alloc_path();
4374 path
->leave_spinning
= 1;
4376 start
= round_down(start
, BTRFS_I(inode
)->root
->sectorsize
);
4377 len
= round_up(max
, BTRFS_I(inode
)->root
->sectorsize
) - start
;
4380 * lookup the last file extent. We're not using i_size here
4381 * because there might be preallocation past i_size
4383 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
), -1,
4386 btrfs_free_path(path
);
4391 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4392 found_type
= found_key
.type
;
4394 /* No extents, but there might be delalloc bits */
4395 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4396 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4397 /* have to trust i_size as the end */
4399 last_for_get_extent
= isize
;
4402 * remember the start of the last extent. There are a
4403 * bunch of different factors that go into the length of the
4404 * extent, so its much less complex to remember where it started
4406 last
= found_key
.offset
;
4407 last_for_get_extent
= last
+ 1;
4409 btrfs_release_path(path
);
4412 * we might have some extents allocated but more delalloc past those
4413 * extents. so, we trust isize unless the start of the last extent is
4418 last_for_get_extent
= isize
;
4421 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4424 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4434 u64 offset_in_extent
= 0;
4436 /* break if the extent we found is outside the range */
4437 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4441 * get_extent may return an extent that starts before our
4442 * requested range. We have to make sure the ranges
4443 * we return to fiemap always move forward and don't
4444 * overlap, so adjust the offsets here
4446 em_start
= max(em
->start
, off
);
4449 * record the offset from the start of the extent
4450 * for adjusting the disk offset below. Only do this if the
4451 * extent isn't compressed since our in ram offset may be past
4452 * what we have actually allocated on disk.
4454 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4455 offset_in_extent
= em_start
- em
->start
;
4456 em_end
= extent_map_end(em
);
4457 em_len
= em_end
- em_start
;
4462 * bump off for our next call to get_extent
4464 off
= extent_map_end(em
);
4468 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4470 flags
|= FIEMAP_EXTENT_LAST
;
4471 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4472 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4473 FIEMAP_EXTENT_NOT_ALIGNED
);
4474 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4475 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4476 FIEMAP_EXTENT_UNKNOWN
);
4477 } else if (fieinfo
->fi_extents_max
) {
4478 u64 bytenr
= em
->block_start
-
4479 (em
->start
- em
->orig_start
);
4481 disko
= em
->block_start
+ offset_in_extent
;
4484 * As btrfs supports shared space, this information
4485 * can be exported to userspace tools via
4486 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4487 * then we're just getting a count and we can skip the
4490 ret
= btrfs_check_shared(NULL
, root
->fs_info
,
4492 btrfs_ino(inode
), bytenr
);
4496 flags
|= FIEMAP_EXTENT_SHARED
;
4499 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4500 flags
|= FIEMAP_EXTENT_ENCODED
;
4501 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4502 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4504 free_extent_map(em
);
4506 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4507 (last
== (u64
)-1 && isize
<= em_end
)) {
4508 flags
|= FIEMAP_EXTENT_LAST
;
4512 /* now scan forward to see if this is really the last extent. */
4513 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4520 flags
|= FIEMAP_EXTENT_LAST
;
4523 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4532 free_extent_map(em
);
4534 btrfs_free_path(path
);
4535 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4536 &cached_state
, GFP_NOFS
);
4540 static void __free_extent_buffer(struct extent_buffer
*eb
)
4542 btrfs_leak_debug_del(&eb
->leak_list
);
4543 kmem_cache_free(extent_buffer_cache
, eb
);
4546 int extent_buffer_under_io(struct extent_buffer
*eb
)
4548 return (atomic_read(&eb
->io_pages
) ||
4549 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4550 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4554 * Helper for releasing extent buffer page.
4556 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4558 unsigned long index
;
4560 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4562 BUG_ON(extent_buffer_under_io(eb
));
4564 index
= num_extent_pages(eb
->start
, eb
->len
);
4570 page
= eb
->pages
[index
];
4574 spin_lock(&page
->mapping
->private_lock
);
4576 * We do this since we'll remove the pages after we've
4577 * removed the eb from the radix tree, so we could race
4578 * and have this page now attached to the new eb. So
4579 * only clear page_private if it's still connected to
4582 if (PagePrivate(page
) &&
4583 page
->private == (unsigned long)eb
) {
4584 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4585 BUG_ON(PageDirty(page
));
4586 BUG_ON(PageWriteback(page
));
4588 * We need to make sure we haven't be attached
4591 ClearPagePrivate(page
);
4592 set_page_private(page
, 0);
4593 /* One for the page private */
4594 page_cache_release(page
);
4598 spin_unlock(&page
->mapping
->private_lock
);
4600 /* One for when we alloced the page */
4601 page_cache_release(page
);
4602 } while (index
!= 0);
4606 * Helper for releasing the extent buffer.
4608 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4610 btrfs_release_extent_buffer_page(eb
);
4611 __free_extent_buffer(eb
);
4614 static struct extent_buffer
*
4615 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4618 struct extent_buffer
*eb
= NULL
;
4620 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4623 eb
->fs_info
= fs_info
;
4625 rwlock_init(&eb
->lock
);
4626 atomic_set(&eb
->write_locks
, 0);
4627 atomic_set(&eb
->read_locks
, 0);
4628 atomic_set(&eb
->blocking_readers
, 0);
4629 atomic_set(&eb
->blocking_writers
, 0);
4630 atomic_set(&eb
->spinning_readers
, 0);
4631 atomic_set(&eb
->spinning_writers
, 0);
4632 eb
->lock_nested
= 0;
4633 init_waitqueue_head(&eb
->write_lock_wq
);
4634 init_waitqueue_head(&eb
->read_lock_wq
);
4636 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4638 spin_lock_init(&eb
->refs_lock
);
4639 atomic_set(&eb
->refs
, 1);
4640 atomic_set(&eb
->io_pages
, 0);
4643 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4645 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4646 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4647 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4652 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4656 struct extent_buffer
*new;
4657 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4659 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4663 for (i
= 0; i
< num_pages
; i
++) {
4664 p
= alloc_page(GFP_NOFS
);
4666 btrfs_release_extent_buffer(new);
4669 attach_extent_buffer_page(new, p
);
4670 WARN_ON(PageDirty(p
));
4675 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4676 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4677 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4682 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4683 u64 start
, unsigned long len
)
4685 struct extent_buffer
*eb
;
4686 unsigned long num_pages
;
4689 num_pages
= num_extent_pages(start
, len
);
4691 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4695 for (i
= 0; i
< num_pages
; i
++) {
4696 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4700 set_extent_buffer_uptodate(eb
);
4701 btrfs_set_header_nritems(eb
, 0);
4702 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4707 __free_page(eb
->pages
[i
- 1]);
4708 __free_extent_buffer(eb
);
4712 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4719 * Called only from tests that don't always have a fs_info
4720 * available, but we know that nodesize is 4096
4724 len
= fs_info
->tree_root
->nodesize
;
4727 return __alloc_dummy_extent_buffer(fs_info
, start
, len
);
4730 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4733 /* the ref bit is tricky. We have to make sure it is set
4734 * if we have the buffer dirty. Otherwise the
4735 * code to free a buffer can end up dropping a dirty
4738 * Once the ref bit is set, it won't go away while the
4739 * buffer is dirty or in writeback, and it also won't
4740 * go away while we have the reference count on the
4743 * We can't just set the ref bit without bumping the
4744 * ref on the eb because free_extent_buffer might
4745 * see the ref bit and try to clear it. If this happens
4746 * free_extent_buffer might end up dropping our original
4747 * ref by mistake and freeing the page before we are able
4748 * to add one more ref.
4750 * So bump the ref count first, then set the bit. If someone
4751 * beat us to it, drop the ref we added.
4753 refs
= atomic_read(&eb
->refs
);
4754 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4757 spin_lock(&eb
->refs_lock
);
4758 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4759 atomic_inc(&eb
->refs
);
4760 spin_unlock(&eb
->refs_lock
);
4763 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4764 struct page
*accessed
)
4766 unsigned long num_pages
, i
;
4768 check_buffer_tree_ref(eb
);
4770 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4771 for (i
= 0; i
< num_pages
; i
++) {
4772 struct page
*p
= eb
->pages
[i
];
4775 mark_page_accessed(p
);
4779 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4782 struct extent_buffer
*eb
;
4785 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4786 start
>> PAGE_CACHE_SHIFT
);
4787 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4790 * Lock our eb's refs_lock to avoid races with
4791 * free_extent_buffer. When we get our eb it might be flagged
4792 * with EXTENT_BUFFER_STALE and another task running
4793 * free_extent_buffer might have seen that flag set,
4794 * eb->refs == 2, that the buffer isn't under IO (dirty and
4795 * writeback flags not set) and it's still in the tree (flag
4796 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4797 * of decrementing the extent buffer's reference count twice.
4798 * So here we could race and increment the eb's reference count,
4799 * clear its stale flag, mark it as dirty and drop our reference
4800 * before the other task finishes executing free_extent_buffer,
4801 * which would later result in an attempt to free an extent
4802 * buffer that is dirty.
4804 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4805 spin_lock(&eb
->refs_lock
);
4806 spin_unlock(&eb
->refs_lock
);
4808 mark_extent_buffer_accessed(eb
, NULL
);
4816 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4817 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4820 struct extent_buffer
*eb
, *exists
= NULL
;
4823 eb
= find_extent_buffer(fs_info
, start
);
4826 eb
= alloc_dummy_extent_buffer(fs_info
, start
);
4829 eb
->fs_info
= fs_info
;
4831 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4834 spin_lock(&fs_info
->buffer_lock
);
4835 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4836 start
>> PAGE_CACHE_SHIFT
, eb
);
4837 spin_unlock(&fs_info
->buffer_lock
);
4838 radix_tree_preload_end();
4839 if (ret
== -EEXIST
) {
4840 exists
= find_extent_buffer(fs_info
, start
);
4846 check_buffer_tree_ref(eb
);
4847 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4850 * We will free dummy extent buffer's if they come into
4851 * free_extent_buffer with a ref count of 2, but if we are using this we
4852 * want the buffers to stay in memory until we're done with them, so
4853 * bump the ref count again.
4855 atomic_inc(&eb
->refs
);
4858 btrfs_release_extent_buffer(eb
);
4863 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4866 unsigned long len
= fs_info
->tree_root
->nodesize
;
4867 unsigned long num_pages
= num_extent_pages(start
, len
);
4869 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4870 struct extent_buffer
*eb
;
4871 struct extent_buffer
*exists
= NULL
;
4873 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4877 eb
= find_extent_buffer(fs_info
, start
);
4881 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4885 for (i
= 0; i
< num_pages
; i
++, index
++) {
4886 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4890 spin_lock(&mapping
->private_lock
);
4891 if (PagePrivate(p
)) {
4893 * We could have already allocated an eb for this page
4894 * and attached one so lets see if we can get a ref on
4895 * the existing eb, and if we can we know it's good and
4896 * we can just return that one, else we know we can just
4897 * overwrite page->private.
4899 exists
= (struct extent_buffer
*)p
->private;
4900 if (atomic_inc_not_zero(&exists
->refs
)) {
4901 spin_unlock(&mapping
->private_lock
);
4903 page_cache_release(p
);
4904 mark_extent_buffer_accessed(exists
, p
);
4910 * Do this so attach doesn't complain and we need to
4911 * drop the ref the old guy had.
4913 ClearPagePrivate(p
);
4914 WARN_ON(PageDirty(p
));
4915 page_cache_release(p
);
4917 attach_extent_buffer_page(eb
, p
);
4918 spin_unlock(&mapping
->private_lock
);
4919 WARN_ON(PageDirty(p
));
4921 if (!PageUptodate(p
))
4925 * see below about how we avoid a nasty race with release page
4926 * and why we unlock later
4930 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4932 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4936 spin_lock(&fs_info
->buffer_lock
);
4937 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4938 start
>> PAGE_CACHE_SHIFT
, eb
);
4939 spin_unlock(&fs_info
->buffer_lock
);
4940 radix_tree_preload_end();
4941 if (ret
== -EEXIST
) {
4942 exists
= find_extent_buffer(fs_info
, start
);
4948 /* add one reference for the tree */
4949 check_buffer_tree_ref(eb
);
4950 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4953 * there is a race where release page may have
4954 * tried to find this extent buffer in the radix
4955 * but failed. It will tell the VM it is safe to
4956 * reclaim the, and it will clear the page private bit.
4957 * We must make sure to set the page private bit properly
4958 * after the extent buffer is in the radix tree so
4959 * it doesn't get lost
4961 SetPageChecked(eb
->pages
[0]);
4962 for (i
= 1; i
< num_pages
; i
++) {
4964 ClearPageChecked(p
);
4967 unlock_page(eb
->pages
[0]);
4971 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4972 for (i
= 0; i
< num_pages
; i
++) {
4974 unlock_page(eb
->pages
[i
]);
4977 btrfs_release_extent_buffer(eb
);
4981 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4983 struct extent_buffer
*eb
=
4984 container_of(head
, struct extent_buffer
, rcu_head
);
4986 __free_extent_buffer(eb
);
4989 /* Expects to have eb->eb_lock already held */
4990 static int release_extent_buffer(struct extent_buffer
*eb
)
4992 WARN_ON(atomic_read(&eb
->refs
) == 0);
4993 if (atomic_dec_and_test(&eb
->refs
)) {
4994 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
4995 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
4997 spin_unlock(&eb
->refs_lock
);
4999 spin_lock(&fs_info
->buffer_lock
);
5000 radix_tree_delete(&fs_info
->buffer_radix
,
5001 eb
->start
>> PAGE_CACHE_SHIFT
);
5002 spin_unlock(&fs_info
->buffer_lock
);
5004 spin_unlock(&eb
->refs_lock
);
5007 /* Should be safe to release our pages at this point */
5008 btrfs_release_extent_buffer_page(eb
);
5009 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5010 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5011 __free_extent_buffer(eb
);
5015 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5018 spin_unlock(&eb
->refs_lock
);
5023 void free_extent_buffer(struct extent_buffer
*eb
)
5031 refs
= atomic_read(&eb
->refs
);
5034 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5039 spin_lock(&eb
->refs_lock
);
5040 if (atomic_read(&eb
->refs
) == 2 &&
5041 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5042 atomic_dec(&eb
->refs
);
5044 if (atomic_read(&eb
->refs
) == 2 &&
5045 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5046 !extent_buffer_under_io(eb
) &&
5047 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5048 atomic_dec(&eb
->refs
);
5051 * I know this is terrible, but it's temporary until we stop tracking
5052 * the uptodate bits and such for the extent buffers.
5054 release_extent_buffer(eb
);
5057 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5062 spin_lock(&eb
->refs_lock
);
5063 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5065 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5066 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5067 atomic_dec(&eb
->refs
);
5068 release_extent_buffer(eb
);
5071 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5074 unsigned long num_pages
;
5077 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5079 for (i
= 0; i
< num_pages
; i
++) {
5080 page
= eb
->pages
[i
];
5081 if (!PageDirty(page
))
5085 WARN_ON(!PagePrivate(page
));
5087 clear_page_dirty_for_io(page
);
5088 spin_lock_irq(&page
->mapping
->tree_lock
);
5089 if (!PageDirty(page
)) {
5090 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5092 PAGECACHE_TAG_DIRTY
);
5094 spin_unlock_irq(&page
->mapping
->tree_lock
);
5095 ClearPageError(page
);
5098 WARN_ON(atomic_read(&eb
->refs
) == 0);
5101 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5104 unsigned long num_pages
;
5107 check_buffer_tree_ref(eb
);
5109 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5111 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5112 WARN_ON(atomic_read(&eb
->refs
) == 0);
5113 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5115 for (i
= 0; i
< num_pages
; i
++)
5116 set_page_dirty(eb
->pages
[i
]);
5120 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5124 unsigned long num_pages
;
5126 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5127 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5128 for (i
= 0; i
< num_pages
; i
++) {
5129 page
= eb
->pages
[i
];
5131 ClearPageUptodate(page
);
5135 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5139 unsigned long num_pages
;
5141 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5142 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5143 for (i
= 0; i
< num_pages
; i
++) {
5144 page
= eb
->pages
[i
];
5145 SetPageUptodate(page
);
5149 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5151 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5154 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5155 struct extent_buffer
*eb
, u64 start
, int wait
,
5156 get_extent_t
*get_extent
, int mirror_num
)
5159 unsigned long start_i
;
5163 int locked_pages
= 0;
5164 int all_uptodate
= 1;
5165 unsigned long num_pages
;
5166 unsigned long num_reads
= 0;
5167 struct bio
*bio
= NULL
;
5168 unsigned long bio_flags
= 0;
5170 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5174 WARN_ON(start
< eb
->start
);
5175 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
5176 (eb
->start
>> PAGE_CACHE_SHIFT
);
5181 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5182 for (i
= start_i
; i
< num_pages
; i
++) {
5183 page
= eb
->pages
[i
];
5184 if (wait
== WAIT_NONE
) {
5185 if (!trylock_page(page
))
5191 if (!PageUptodate(page
)) {
5198 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5202 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5203 eb
->read_mirror
= 0;
5204 atomic_set(&eb
->io_pages
, num_reads
);
5205 for (i
= start_i
; i
< num_pages
; i
++) {
5206 page
= eb
->pages
[i
];
5207 if (!PageUptodate(page
)) {
5208 ClearPageError(page
);
5209 err
= __extent_read_full_page(tree
, page
,
5211 mirror_num
, &bio_flags
,
5221 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
5227 if (ret
|| wait
!= WAIT_COMPLETE
)
5230 for (i
= start_i
; i
< num_pages
; i
++) {
5231 page
= eb
->pages
[i
];
5232 wait_on_page_locked(page
);
5233 if (!PageUptodate(page
))
5241 while (locked_pages
> 0) {
5242 page
= eb
->pages
[i
];
5250 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5251 unsigned long start
,
5258 char *dst
= (char *)dstv
;
5259 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5260 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5262 WARN_ON(start
> eb
->len
);
5263 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5265 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5268 page
= eb
->pages
[i
];
5270 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5271 kaddr
= page_address(page
);
5272 memcpy(dst
, kaddr
+ offset
, cur
);
5281 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5282 unsigned long start
,
5289 char __user
*dst
= (char __user
*)dstv
;
5290 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5291 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5294 WARN_ON(start
> eb
->len
);
5295 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5297 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5300 page
= eb
->pages
[i
];
5302 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5303 kaddr
= page_address(page
);
5304 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5318 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5319 unsigned long min_len
, char **map
,
5320 unsigned long *map_start
,
5321 unsigned long *map_len
)
5323 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
5326 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5327 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5328 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5335 offset
= start_offset
;
5339 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
5342 if (start
+ min_len
> eb
->len
) {
5343 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
5345 eb
->start
, eb
->len
, start
, min_len
);
5350 kaddr
= page_address(p
);
5351 *map
= kaddr
+ offset
;
5352 *map_len
= PAGE_CACHE_SIZE
- offset
;
5356 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5357 unsigned long start
,
5364 char *ptr
= (char *)ptrv
;
5365 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5366 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5369 WARN_ON(start
> eb
->len
);
5370 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5372 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5375 page
= eb
->pages
[i
];
5377 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5379 kaddr
= page_address(page
);
5380 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5392 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5393 unsigned long start
, unsigned long len
)
5399 char *src
= (char *)srcv
;
5400 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5401 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5403 WARN_ON(start
> eb
->len
);
5404 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5406 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5409 page
= eb
->pages
[i
];
5410 WARN_ON(!PageUptodate(page
));
5412 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5413 kaddr
= page_address(page
);
5414 memcpy(kaddr
+ offset
, src
, cur
);
5423 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5424 unsigned long start
, unsigned long len
)
5430 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5431 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5433 WARN_ON(start
> eb
->len
);
5434 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5436 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5439 page
= eb
->pages
[i
];
5440 WARN_ON(!PageUptodate(page
));
5442 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5443 kaddr
= page_address(page
);
5444 memset(kaddr
+ offset
, c
, cur
);
5452 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5453 unsigned long dst_offset
, unsigned long src_offset
,
5456 u64 dst_len
= dst
->len
;
5461 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5462 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5464 WARN_ON(src
->len
!= dst_len
);
5466 offset
= (start_offset
+ dst_offset
) &
5467 (PAGE_CACHE_SIZE
- 1);
5470 page
= dst
->pages
[i
];
5471 WARN_ON(!PageUptodate(page
));
5473 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
5475 kaddr
= page_address(page
);
5476 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5486 * The extent buffer bitmap operations are done with byte granularity because
5487 * bitmap items are not guaranteed to be aligned to a word and therefore a
5488 * single word in a bitmap may straddle two pages in the extent buffer.
5490 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
5491 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
5492 #define BITMAP_FIRST_BYTE_MASK(start) \
5493 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
5494 #define BITMAP_LAST_BYTE_MASK(nbits) \
5495 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
5498 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5500 * @eb: the extent buffer
5501 * @start: offset of the bitmap item in the extent buffer
5503 * @page_index: return index of the page in the extent buffer that contains the
5505 * @page_offset: return offset into the page given by page_index
5507 * This helper hides the ugliness of finding the byte in an extent buffer which
5508 * contains a given bit.
5510 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5511 unsigned long start
, unsigned long nr
,
5512 unsigned long *page_index
,
5513 size_t *page_offset
)
5515 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5516 size_t byte_offset
= BIT_BYTE(nr
);
5520 * The byte we want is the offset of the extent buffer + the offset of
5521 * the bitmap item in the extent buffer + the offset of the byte in the
5524 offset
= start_offset
+ start
+ byte_offset
;
5526 *page_index
= offset
>> PAGE_CACHE_SHIFT
;
5527 *page_offset
= offset
& (PAGE_CACHE_SIZE
- 1);
5531 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5532 * @eb: the extent buffer
5533 * @start: offset of the bitmap item in the extent buffer
5534 * @nr: bit number to test
5536 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5544 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5545 page
= eb
->pages
[i
];
5546 WARN_ON(!PageUptodate(page
));
5547 kaddr
= page_address(page
);
5548 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5552 * extent_buffer_bitmap_set - set an area of a bitmap
5553 * @eb: the extent buffer
5554 * @start: offset of the bitmap item in the extent buffer
5555 * @pos: bit number of the first bit
5556 * @len: number of bits to set
5558 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5559 unsigned long pos
, unsigned long len
)
5565 const unsigned int size
= pos
+ len
;
5566 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5567 unsigned int mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5569 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5570 page
= eb
->pages
[i
];
5571 WARN_ON(!PageUptodate(page
));
5572 kaddr
= page_address(page
);
5574 while (len
>= bits_to_set
) {
5575 kaddr
[offset
] |= mask_to_set
;
5577 bits_to_set
= BITS_PER_BYTE
;
5579 if (++offset
>= PAGE_CACHE_SIZE
&& len
> 0) {
5581 page
= eb
->pages
[++i
];
5582 WARN_ON(!PageUptodate(page
));
5583 kaddr
= page_address(page
);
5587 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5588 kaddr
[offset
] |= mask_to_set
;
5594 * extent_buffer_bitmap_clear - clear an area of a bitmap
5595 * @eb: the extent buffer
5596 * @start: offset of the bitmap item in the extent buffer
5597 * @pos: bit number of the first bit
5598 * @len: number of bits to clear
5600 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5601 unsigned long pos
, unsigned long len
)
5607 const unsigned int size
= pos
+ len
;
5608 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5609 unsigned int mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5611 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5612 page
= eb
->pages
[i
];
5613 WARN_ON(!PageUptodate(page
));
5614 kaddr
= page_address(page
);
5616 while (len
>= bits_to_clear
) {
5617 kaddr
[offset
] &= ~mask_to_clear
;
5618 len
-= bits_to_clear
;
5619 bits_to_clear
= BITS_PER_BYTE
;
5620 mask_to_clear
= ~0U;
5621 if (++offset
>= PAGE_CACHE_SIZE
&& len
> 0) {
5623 page
= eb
->pages
[++i
];
5624 WARN_ON(!PageUptodate(page
));
5625 kaddr
= page_address(page
);
5629 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5630 kaddr
[offset
] &= ~mask_to_clear
;
5634 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5636 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5637 return distance
< len
;
5640 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5641 unsigned long dst_off
, unsigned long src_off
,
5644 char *dst_kaddr
= page_address(dst_page
);
5646 int must_memmove
= 0;
5648 if (dst_page
!= src_page
) {
5649 src_kaddr
= page_address(src_page
);
5651 src_kaddr
= dst_kaddr
;
5652 if (areas_overlap(src_off
, dst_off
, len
))
5657 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5659 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5662 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5663 unsigned long src_offset
, unsigned long len
)
5666 size_t dst_off_in_page
;
5667 size_t src_off_in_page
;
5668 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5669 unsigned long dst_i
;
5670 unsigned long src_i
;
5672 if (src_offset
+ len
> dst
->len
) {
5673 btrfs_err(dst
->fs_info
,
5674 "memmove bogus src_offset %lu move "
5675 "len %lu dst len %lu", src_offset
, len
, dst
->len
);
5678 if (dst_offset
+ len
> dst
->len
) {
5679 btrfs_err(dst
->fs_info
,
5680 "memmove bogus dst_offset %lu move "
5681 "len %lu dst len %lu", dst_offset
, len
, dst
->len
);
5686 dst_off_in_page
= (start_offset
+ dst_offset
) &
5687 (PAGE_CACHE_SIZE
- 1);
5688 src_off_in_page
= (start_offset
+ src_offset
) &
5689 (PAGE_CACHE_SIZE
- 1);
5691 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5692 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5694 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5696 cur
= min_t(unsigned long, cur
,
5697 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5699 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5700 dst_off_in_page
, src_off_in_page
, cur
);
5708 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5709 unsigned long src_offset
, unsigned long len
)
5712 size_t dst_off_in_page
;
5713 size_t src_off_in_page
;
5714 unsigned long dst_end
= dst_offset
+ len
- 1;
5715 unsigned long src_end
= src_offset
+ len
- 1;
5716 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5717 unsigned long dst_i
;
5718 unsigned long src_i
;
5720 if (src_offset
+ len
> dst
->len
) {
5721 btrfs_err(dst
->fs_info
, "memmove bogus src_offset %lu move "
5722 "len %lu len %lu", src_offset
, len
, dst
->len
);
5725 if (dst_offset
+ len
> dst
->len
) {
5726 btrfs_err(dst
->fs_info
, "memmove bogus dst_offset %lu move "
5727 "len %lu len %lu", dst_offset
, len
, dst
->len
);
5730 if (dst_offset
< src_offset
) {
5731 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5735 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5736 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5738 dst_off_in_page
= (start_offset
+ dst_end
) &
5739 (PAGE_CACHE_SIZE
- 1);
5740 src_off_in_page
= (start_offset
+ src_end
) &
5741 (PAGE_CACHE_SIZE
- 1);
5743 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5744 cur
= min(cur
, dst_off_in_page
+ 1);
5745 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5746 dst_off_in_page
- cur
+ 1,
5747 src_off_in_page
- cur
+ 1, cur
);
5755 int try_release_extent_buffer(struct page
*page
)
5757 struct extent_buffer
*eb
;
5760 * We need to make sure noboody is attaching this page to an eb right
5763 spin_lock(&page
->mapping
->private_lock
);
5764 if (!PagePrivate(page
)) {
5765 spin_unlock(&page
->mapping
->private_lock
);
5769 eb
= (struct extent_buffer
*)page
->private;
5773 * This is a little awful but should be ok, we need to make sure that
5774 * the eb doesn't disappear out from under us while we're looking at
5777 spin_lock(&eb
->refs_lock
);
5778 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5779 spin_unlock(&eb
->refs_lock
);
5780 spin_unlock(&page
->mapping
->private_lock
);
5783 spin_unlock(&page
->mapping
->private_lock
);
5786 * If tree ref isn't set then we know the ref on this eb is a real ref,
5787 * so just return, this page will likely be freed soon anyway.
5789 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5790 spin_unlock(&eb
->refs_lock
);
5794 return release_extent_buffer(eb
);