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 kmem_cache_destroy(extent_state_cache
);
210 kmem_cache_destroy(extent_buffer_cache
);
212 bioset_free(btrfs_bioset
);
215 void extent_io_tree_init(struct extent_io_tree
*tree
,
216 struct address_space
*mapping
)
218 tree
->state
= RB_ROOT
;
220 tree
->dirty_bytes
= 0;
221 spin_lock_init(&tree
->lock
);
222 tree
->mapping
= mapping
;
225 static struct extent_state
*alloc_extent_state(gfp_t mask
)
227 struct extent_state
*state
;
229 state
= kmem_cache_alloc(extent_state_cache
, mask
);
233 state
->failrec
= NULL
;
234 RB_CLEAR_NODE(&state
->rb_node
);
235 btrfs_leak_debug_add(&state
->leak_list
, &states
);
236 atomic_set(&state
->refs
, 1);
237 init_waitqueue_head(&state
->wq
);
238 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
242 void free_extent_state(struct extent_state
*state
)
246 if (atomic_dec_and_test(&state
->refs
)) {
247 WARN_ON(extent_state_in_tree(state
));
248 btrfs_leak_debug_del(&state
->leak_list
);
249 trace_free_extent_state(state
, _RET_IP_
);
250 kmem_cache_free(extent_state_cache
, state
);
254 static struct rb_node
*tree_insert(struct rb_root
*root
,
255 struct rb_node
*search_start
,
257 struct rb_node
*node
,
258 struct rb_node
***p_in
,
259 struct rb_node
**parent_in
)
262 struct rb_node
*parent
= NULL
;
263 struct tree_entry
*entry
;
265 if (p_in
&& parent_in
) {
271 p
= search_start
? &search_start
: &root
->rb_node
;
274 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
276 if (offset
< entry
->start
)
278 else if (offset
> entry
->end
)
285 rb_link_node(node
, parent
, p
);
286 rb_insert_color(node
, root
);
290 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
291 struct rb_node
**prev_ret
,
292 struct rb_node
**next_ret
,
293 struct rb_node
***p_ret
,
294 struct rb_node
**parent_ret
)
296 struct rb_root
*root
= &tree
->state
;
297 struct rb_node
**n
= &root
->rb_node
;
298 struct rb_node
*prev
= NULL
;
299 struct rb_node
*orig_prev
= NULL
;
300 struct tree_entry
*entry
;
301 struct tree_entry
*prev_entry
= NULL
;
305 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
308 if (offset
< entry
->start
)
310 else if (offset
> entry
->end
)
323 while (prev
&& offset
> prev_entry
->end
) {
324 prev
= rb_next(prev
);
325 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
332 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
333 while (prev
&& offset
< prev_entry
->start
) {
334 prev
= rb_prev(prev
);
335 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
342 static inline struct rb_node
*
343 tree_search_for_insert(struct extent_io_tree
*tree
,
345 struct rb_node
***p_ret
,
346 struct rb_node
**parent_ret
)
348 struct rb_node
*prev
= NULL
;
351 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
357 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
360 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
363 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
364 struct extent_state
*other
)
366 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
367 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
372 * utility function to look for merge candidates inside a given range.
373 * Any extents with matching state are merged together into a single
374 * extent in the tree. Extents with EXTENT_IO in their state field
375 * are not merged because the end_io handlers need to be able to do
376 * operations on them without sleeping (or doing allocations/splits).
378 * This should be called with the tree lock held.
380 static void merge_state(struct extent_io_tree
*tree
,
381 struct extent_state
*state
)
383 struct extent_state
*other
;
384 struct rb_node
*other_node
;
386 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
389 other_node
= rb_prev(&state
->rb_node
);
391 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
392 if (other
->end
== state
->start
- 1 &&
393 other
->state
== state
->state
) {
394 merge_cb(tree
, state
, other
);
395 state
->start
= other
->start
;
396 rb_erase(&other
->rb_node
, &tree
->state
);
397 RB_CLEAR_NODE(&other
->rb_node
);
398 free_extent_state(other
);
401 other_node
= rb_next(&state
->rb_node
);
403 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
404 if (other
->start
== state
->end
+ 1 &&
405 other
->state
== state
->state
) {
406 merge_cb(tree
, state
, other
);
407 state
->end
= other
->end
;
408 rb_erase(&other
->rb_node
, &tree
->state
);
409 RB_CLEAR_NODE(&other
->rb_node
);
410 free_extent_state(other
);
415 static void set_state_cb(struct extent_io_tree
*tree
,
416 struct extent_state
*state
, unsigned *bits
)
418 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
419 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
422 static void clear_state_cb(struct extent_io_tree
*tree
,
423 struct extent_state
*state
, unsigned *bits
)
425 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
426 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
429 static void set_state_bits(struct extent_io_tree
*tree
,
430 struct extent_state
*state
, unsigned *bits
,
431 struct extent_changeset
*changeset
);
434 * insert an extent_state struct into the tree. 'bits' are set on the
435 * struct before it is inserted.
437 * This may return -EEXIST if the extent is already there, in which case the
438 * state struct is freed.
440 * The tree lock is not taken internally. This is a utility function and
441 * probably isn't what you want to call (see set/clear_extent_bit).
443 static int insert_state(struct extent_io_tree
*tree
,
444 struct extent_state
*state
, u64 start
, u64 end
,
446 struct rb_node
**parent
,
447 unsigned *bits
, struct extent_changeset
*changeset
)
449 struct rb_node
*node
;
452 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
454 state
->start
= start
;
457 set_state_bits(tree
, state
, bits
, changeset
);
459 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
461 struct extent_state
*found
;
462 found
= rb_entry(node
, struct extent_state
, rb_node
);
463 printk(KERN_ERR
"BTRFS: found node %llu %llu on insert of "
465 found
->start
, found
->end
, start
, end
);
468 merge_state(tree
, state
);
472 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
475 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
476 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
480 * split a given extent state struct in two, inserting the preallocated
481 * struct 'prealloc' as the newly created second half. 'split' indicates an
482 * offset inside 'orig' where it should be split.
485 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
486 * are two extent state structs in the tree:
487 * prealloc: [orig->start, split - 1]
488 * orig: [ split, orig->end ]
490 * The tree locks are not taken by this function. They need to be held
493 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
494 struct extent_state
*prealloc
, u64 split
)
496 struct rb_node
*node
;
498 split_cb(tree
, orig
, split
);
500 prealloc
->start
= orig
->start
;
501 prealloc
->end
= split
- 1;
502 prealloc
->state
= orig
->state
;
505 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
506 &prealloc
->rb_node
, NULL
, NULL
);
508 free_extent_state(prealloc
);
514 static struct extent_state
*next_state(struct extent_state
*state
)
516 struct rb_node
*next
= rb_next(&state
->rb_node
);
518 return rb_entry(next
, struct extent_state
, rb_node
);
524 * utility function to clear some bits in an extent state struct.
525 * it will optionally wake up any one waiting on this state (wake == 1).
527 * If no bits are set on the state struct after clearing things, the
528 * struct is freed and removed from the tree
530 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
531 struct extent_state
*state
,
532 unsigned *bits
, int wake
,
533 struct extent_changeset
*changeset
)
535 struct extent_state
*next
;
536 unsigned bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
538 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
539 u64 range
= state
->end
- state
->start
+ 1;
540 WARN_ON(range
> tree
->dirty_bytes
);
541 tree
->dirty_bytes
-= range
;
543 clear_state_cb(tree
, state
, bits
);
544 add_extent_changeset(state
, bits_to_clear
, changeset
, 0);
545 state
->state
&= ~bits_to_clear
;
548 if (state
->state
== 0) {
549 next
= next_state(state
);
550 if (extent_state_in_tree(state
)) {
551 rb_erase(&state
->rb_node
, &tree
->state
);
552 RB_CLEAR_NODE(&state
->rb_node
);
553 free_extent_state(state
);
558 merge_state(tree
, state
);
559 next
= next_state(state
);
564 static struct extent_state
*
565 alloc_extent_state_atomic(struct extent_state
*prealloc
)
568 prealloc
= alloc_extent_state(GFP_ATOMIC
);
573 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
575 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
576 "Extent tree was modified by another "
577 "thread while locked.");
581 * clear some bits on a range in the tree. This may require splitting
582 * or inserting elements in the tree, so the gfp mask is used to
583 * indicate which allocations or sleeping are allowed.
585 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
586 * the given range from the tree regardless of state (ie for truncate).
588 * the range [start, end] is inclusive.
590 * This takes the tree lock, and returns 0 on success and < 0 on error.
592 static int __clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
593 unsigned bits
, int wake
, int delete,
594 struct extent_state
**cached_state
,
595 gfp_t mask
, struct extent_changeset
*changeset
)
597 struct extent_state
*state
;
598 struct extent_state
*cached
;
599 struct extent_state
*prealloc
= NULL
;
600 struct rb_node
*node
;
605 btrfs_debug_check_extent_io_range(tree
, start
, end
);
607 if (bits
& EXTENT_DELALLOC
)
608 bits
|= EXTENT_NORESERVE
;
611 bits
|= ~EXTENT_CTLBITS
;
612 bits
|= EXTENT_FIRST_DELALLOC
;
614 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
617 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
619 * Don't care for allocation failure here because we might end
620 * up not needing the pre-allocated extent state at all, which
621 * is the case if we only have in the tree extent states that
622 * cover our input range and don't cover too any other range.
623 * If we end up needing a new extent state we allocate it later.
625 prealloc
= alloc_extent_state(mask
);
628 spin_lock(&tree
->lock
);
630 cached
= *cached_state
;
633 *cached_state
= NULL
;
637 if (cached
&& extent_state_in_tree(cached
) &&
638 cached
->start
<= start
&& cached
->end
> start
) {
640 atomic_dec(&cached
->refs
);
645 free_extent_state(cached
);
648 * this search will find the extents that end after
651 node
= tree_search(tree
, start
);
654 state
= rb_entry(node
, struct extent_state
, rb_node
);
656 if (state
->start
> end
)
658 WARN_ON(state
->end
< start
);
659 last_end
= state
->end
;
661 /* the state doesn't have the wanted bits, go ahead */
662 if (!(state
->state
& bits
)) {
663 state
= next_state(state
);
668 * | ---- desired range ---- |
670 * | ------------- state -------------- |
672 * We need to split the extent we found, and may flip
673 * bits on second half.
675 * If the extent we found extends past our range, we
676 * just split and search again. It'll get split again
677 * the next time though.
679 * If the extent we found is inside our range, we clear
680 * the desired bit on it.
683 if (state
->start
< start
) {
684 prealloc
= alloc_extent_state_atomic(prealloc
);
686 err
= split_state(tree
, state
, prealloc
, start
);
688 extent_io_tree_panic(tree
, err
);
693 if (state
->end
<= end
) {
694 state
= clear_state_bit(tree
, state
, &bits
, wake
,
701 * | ---- desired range ---- |
703 * We need to split the extent, and clear the bit
706 if (state
->start
<= end
&& state
->end
> end
) {
707 prealloc
= alloc_extent_state_atomic(prealloc
);
709 err
= split_state(tree
, state
, prealloc
, end
+ 1);
711 extent_io_tree_panic(tree
, err
);
716 clear_state_bit(tree
, prealloc
, &bits
, wake
, changeset
);
722 state
= clear_state_bit(tree
, state
, &bits
, wake
, changeset
);
724 if (last_end
== (u64
)-1)
726 start
= last_end
+ 1;
727 if (start
<= end
&& state
&& !need_resched())
733 spin_unlock(&tree
->lock
);
734 if (gfpflags_allow_blocking(mask
))
739 spin_unlock(&tree
->lock
);
741 free_extent_state(prealloc
);
747 static void wait_on_state(struct extent_io_tree
*tree
,
748 struct extent_state
*state
)
749 __releases(tree
->lock
)
750 __acquires(tree
->lock
)
753 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
754 spin_unlock(&tree
->lock
);
756 spin_lock(&tree
->lock
);
757 finish_wait(&state
->wq
, &wait
);
761 * waits for one or more bits to clear on a range in the state tree.
762 * The range [start, end] is inclusive.
763 * The tree lock is taken by this function
765 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
768 struct extent_state
*state
;
769 struct rb_node
*node
;
771 btrfs_debug_check_extent_io_range(tree
, start
, end
);
773 spin_lock(&tree
->lock
);
777 * this search will find all the extents that end after
780 node
= tree_search(tree
, start
);
785 state
= rb_entry(node
, struct extent_state
, rb_node
);
787 if (state
->start
> end
)
790 if (state
->state
& bits
) {
791 start
= state
->start
;
792 atomic_inc(&state
->refs
);
793 wait_on_state(tree
, state
);
794 free_extent_state(state
);
797 start
= state
->end
+ 1;
802 if (!cond_resched_lock(&tree
->lock
)) {
803 node
= rb_next(node
);
808 spin_unlock(&tree
->lock
);
811 static void set_state_bits(struct extent_io_tree
*tree
,
812 struct extent_state
*state
,
813 unsigned *bits
, struct extent_changeset
*changeset
)
815 unsigned bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
817 set_state_cb(tree
, state
, bits
);
818 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
819 u64 range
= state
->end
- state
->start
+ 1;
820 tree
->dirty_bytes
+= range
;
822 add_extent_changeset(state
, bits_to_set
, changeset
, 1);
823 state
->state
|= bits_to_set
;
826 static void cache_state_if_flags(struct extent_state
*state
,
827 struct extent_state
**cached_ptr
,
830 if (cached_ptr
&& !(*cached_ptr
)) {
831 if (!flags
|| (state
->state
& flags
)) {
833 atomic_inc(&state
->refs
);
838 static void cache_state(struct extent_state
*state
,
839 struct extent_state
**cached_ptr
)
841 return cache_state_if_flags(state
, cached_ptr
,
842 EXTENT_IOBITS
| EXTENT_BOUNDARY
);
846 * set some bits on a range in the tree. This may require allocations or
847 * sleeping, so the gfp mask is used to indicate what is allowed.
849 * If any of the exclusive bits are set, this will fail with -EEXIST if some
850 * part of the range already has the desired bits set. The start of the
851 * existing range is returned in failed_start in this case.
853 * [start, end] is inclusive This takes the tree lock.
856 static int __must_check
857 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
858 unsigned bits
, unsigned exclusive_bits
,
859 u64
*failed_start
, struct extent_state
**cached_state
,
860 gfp_t mask
, struct extent_changeset
*changeset
)
862 struct extent_state
*state
;
863 struct extent_state
*prealloc
= NULL
;
864 struct rb_node
*node
;
866 struct rb_node
*parent
;
871 btrfs_debug_check_extent_io_range(tree
, start
, end
);
873 bits
|= EXTENT_FIRST_DELALLOC
;
875 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
877 * Don't care for allocation failure here because we might end
878 * up not needing the pre-allocated extent state at all, which
879 * is the case if we only have in the tree extent states that
880 * cover our input range and don't cover too any other range.
881 * If we end up needing a new extent state we allocate it later.
883 prealloc
= alloc_extent_state(mask
);
886 spin_lock(&tree
->lock
);
887 if (cached_state
&& *cached_state
) {
888 state
= *cached_state
;
889 if (state
->start
<= start
&& state
->end
> start
&&
890 extent_state_in_tree(state
)) {
891 node
= &state
->rb_node
;
896 * this search will find all the extents that end after
899 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
901 prealloc
= alloc_extent_state_atomic(prealloc
);
903 err
= insert_state(tree
, prealloc
, start
, end
,
904 &p
, &parent
, &bits
, changeset
);
906 extent_io_tree_panic(tree
, err
);
908 cache_state(prealloc
, cached_state
);
912 state
= rb_entry(node
, struct extent_state
, rb_node
);
914 last_start
= state
->start
;
915 last_end
= state
->end
;
918 * | ---- desired range ---- |
921 * Just lock what we found and keep going
923 if (state
->start
== start
&& state
->end
<= end
) {
924 if (state
->state
& exclusive_bits
) {
925 *failed_start
= state
->start
;
930 set_state_bits(tree
, state
, &bits
, changeset
);
931 cache_state(state
, cached_state
);
932 merge_state(tree
, state
);
933 if (last_end
== (u64
)-1)
935 start
= last_end
+ 1;
936 state
= next_state(state
);
937 if (start
< end
&& state
&& state
->start
== start
&&
944 * | ---- desired range ---- |
947 * | ------------- state -------------- |
949 * We need to split the extent we found, and may flip bits on
952 * If the extent we found extends past our
953 * range, we just split and search again. It'll get split
954 * again the next time though.
956 * If the extent we found is inside our range, we set the
959 if (state
->start
< start
) {
960 if (state
->state
& exclusive_bits
) {
961 *failed_start
= start
;
966 prealloc
= alloc_extent_state_atomic(prealloc
);
968 err
= split_state(tree
, state
, prealloc
, start
);
970 extent_io_tree_panic(tree
, err
);
975 if (state
->end
<= end
) {
976 set_state_bits(tree
, state
, &bits
, changeset
);
977 cache_state(state
, cached_state
);
978 merge_state(tree
, state
);
979 if (last_end
== (u64
)-1)
981 start
= last_end
+ 1;
982 state
= next_state(state
);
983 if (start
< end
&& state
&& state
->start
== start
&&
990 * | ---- desired range ---- |
991 * | state | or | state |
993 * There's a hole, we need to insert something in it and
994 * ignore the extent we found.
996 if (state
->start
> start
) {
998 if (end
< last_start
)
1001 this_end
= last_start
- 1;
1003 prealloc
= alloc_extent_state_atomic(prealloc
);
1007 * Avoid to free 'prealloc' if it can be merged with
1010 err
= insert_state(tree
, prealloc
, start
, this_end
,
1011 NULL
, NULL
, &bits
, changeset
);
1013 extent_io_tree_panic(tree
, err
);
1015 cache_state(prealloc
, cached_state
);
1017 start
= this_end
+ 1;
1021 * | ---- desired range ---- |
1023 * We need to split the extent, and set the bit
1026 if (state
->start
<= end
&& state
->end
> end
) {
1027 if (state
->state
& exclusive_bits
) {
1028 *failed_start
= start
;
1033 prealloc
= alloc_extent_state_atomic(prealloc
);
1035 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1037 extent_io_tree_panic(tree
, err
);
1039 set_state_bits(tree
, prealloc
, &bits
, changeset
);
1040 cache_state(prealloc
, cached_state
);
1041 merge_state(tree
, prealloc
);
1049 spin_unlock(&tree
->lock
);
1050 if (gfpflags_allow_blocking(mask
))
1055 spin_unlock(&tree
->lock
);
1057 free_extent_state(prealloc
);
1063 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1064 unsigned bits
, u64
* failed_start
,
1065 struct extent_state
**cached_state
, gfp_t mask
)
1067 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1068 cached_state
, mask
, NULL
);
1073 * convert_extent_bit - convert all bits in a given range from one bit to
1075 * @tree: the io tree to search
1076 * @start: the start offset in bytes
1077 * @end: the end offset in bytes (inclusive)
1078 * @bits: the bits to set in this range
1079 * @clear_bits: the bits to clear in this range
1080 * @cached_state: state that we're going to cache
1082 * This will go through and set bits for the given range. If any states exist
1083 * already in this range they are set with the given bit and cleared of the
1084 * clear_bits. This is only meant to be used by things that are mergeable, ie
1085 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1086 * boundary bits like LOCK.
1088 * All allocations are done with GFP_NOFS.
1090 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1091 unsigned bits
, unsigned clear_bits
,
1092 struct extent_state
**cached_state
)
1094 struct extent_state
*state
;
1095 struct extent_state
*prealloc
= NULL
;
1096 struct rb_node
*node
;
1098 struct rb_node
*parent
;
1102 bool first_iteration
= true;
1104 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1109 * Best effort, don't worry if extent state allocation fails
1110 * here for the first iteration. We might have a cached state
1111 * that matches exactly the target range, in which case no
1112 * extent state allocations are needed. We'll only know this
1113 * after locking the tree.
1115 prealloc
= alloc_extent_state(GFP_NOFS
);
1116 if (!prealloc
&& !first_iteration
)
1120 spin_lock(&tree
->lock
);
1121 if (cached_state
&& *cached_state
) {
1122 state
= *cached_state
;
1123 if (state
->start
<= start
&& state
->end
> start
&&
1124 extent_state_in_tree(state
)) {
1125 node
= &state
->rb_node
;
1131 * this search will find all the extents that end after
1134 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1136 prealloc
= alloc_extent_state_atomic(prealloc
);
1141 err
= insert_state(tree
, prealloc
, start
, end
,
1142 &p
, &parent
, &bits
, NULL
);
1144 extent_io_tree_panic(tree
, err
);
1145 cache_state(prealloc
, cached_state
);
1149 state
= rb_entry(node
, struct extent_state
, rb_node
);
1151 last_start
= state
->start
;
1152 last_end
= state
->end
;
1155 * | ---- desired range ---- |
1158 * Just lock what we found and keep going
1160 if (state
->start
== start
&& state
->end
<= end
) {
1161 set_state_bits(tree
, state
, &bits
, NULL
);
1162 cache_state(state
, cached_state
);
1163 state
= clear_state_bit(tree
, state
, &clear_bits
, 0, NULL
);
1164 if (last_end
== (u64
)-1)
1166 start
= last_end
+ 1;
1167 if (start
< end
&& state
&& state
->start
== start
&&
1174 * | ---- desired range ---- |
1177 * | ------------- state -------------- |
1179 * We need to split the extent we found, and may flip bits on
1182 * If the extent we found extends past our
1183 * range, we just split and search again. It'll get split
1184 * again the next time though.
1186 * If the extent we found is inside our range, we set the
1187 * desired bit on it.
1189 if (state
->start
< start
) {
1190 prealloc
= alloc_extent_state_atomic(prealloc
);
1195 err
= split_state(tree
, state
, prealloc
, start
);
1197 extent_io_tree_panic(tree
, err
);
1201 if (state
->end
<= end
) {
1202 set_state_bits(tree
, state
, &bits
, NULL
);
1203 cache_state(state
, cached_state
);
1204 state
= clear_state_bit(tree
, state
, &clear_bits
, 0,
1206 if (last_end
== (u64
)-1)
1208 start
= last_end
+ 1;
1209 if (start
< end
&& state
&& state
->start
== start
&&
1216 * | ---- desired range ---- |
1217 * | state | or | state |
1219 * There's a hole, we need to insert something in it and
1220 * ignore the extent we found.
1222 if (state
->start
> start
) {
1224 if (end
< last_start
)
1227 this_end
= last_start
- 1;
1229 prealloc
= alloc_extent_state_atomic(prealloc
);
1236 * Avoid to free 'prealloc' if it can be merged with
1239 err
= insert_state(tree
, prealloc
, start
, this_end
,
1240 NULL
, NULL
, &bits
, NULL
);
1242 extent_io_tree_panic(tree
, err
);
1243 cache_state(prealloc
, cached_state
);
1245 start
= this_end
+ 1;
1249 * | ---- desired range ---- |
1251 * We need to split the extent, and set the bit
1254 if (state
->start
<= end
&& state
->end
> end
) {
1255 prealloc
= alloc_extent_state_atomic(prealloc
);
1261 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1263 extent_io_tree_panic(tree
, err
);
1265 set_state_bits(tree
, prealloc
, &bits
, NULL
);
1266 cache_state(prealloc
, cached_state
);
1267 clear_state_bit(tree
, prealloc
, &clear_bits
, 0, NULL
);
1275 spin_unlock(&tree
->lock
);
1277 first_iteration
= false;
1281 spin_unlock(&tree
->lock
);
1283 free_extent_state(prealloc
);
1288 /* wrappers around set/clear extent bit */
1289 int set_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1290 unsigned bits
, 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
, GFP_NOFS
,
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
, struct extent_changeset
*changeset
)
1316 * Don't support EXTENT_LOCKED case, same reason as
1317 * set_record_extent_bits().
1319 BUG_ON(bits
& EXTENT_LOCKED
);
1321 return __clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, GFP_NOFS
,
1326 * either insert or lock state struct between start and end use mask to tell
1327 * us if waiting is desired.
1329 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1330 struct extent_state
**cached_state
)
1336 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
,
1337 EXTENT_LOCKED
, &failed_start
,
1338 cached_state
, GFP_NOFS
, NULL
);
1339 if (err
== -EEXIST
) {
1340 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1341 start
= failed_start
;
1344 WARN_ON(start
> end
);
1349 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1354 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1355 &failed_start
, NULL
, GFP_NOFS
, NULL
);
1356 if (err
== -EEXIST
) {
1357 if (failed_start
> start
)
1358 clear_extent_bit(tree
, start
, failed_start
- 1,
1359 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1365 void extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1367 unsigned long index
= start
>> PAGE_SHIFT
;
1368 unsigned long end_index
= end
>> PAGE_SHIFT
;
1371 while (index
<= end_index
) {
1372 page
= find_get_page(inode
->i_mapping
, index
);
1373 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1374 clear_page_dirty_for_io(page
);
1380 void extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1382 unsigned long index
= start
>> PAGE_SHIFT
;
1383 unsigned long end_index
= end
>> PAGE_SHIFT
;
1386 while (index
<= end_index
) {
1387 page
= find_get_page(inode
->i_mapping
, index
);
1388 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1389 __set_page_dirty_nobuffers(page
);
1390 account_page_redirty(page
);
1397 * helper function to set both pages and extents in the tree writeback
1399 static void set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1401 unsigned long index
= start
>> PAGE_SHIFT
;
1402 unsigned long end_index
= end
>> PAGE_SHIFT
;
1405 while (index
<= end_index
) {
1406 page
= find_get_page(tree
->mapping
, index
);
1407 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1408 set_page_writeback(page
);
1414 /* find the first state struct with 'bits' set after 'start', and
1415 * return it. tree->lock must be held. NULL will returned if
1416 * nothing was found after 'start'
1418 static struct extent_state
*
1419 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1420 u64 start
, unsigned bits
)
1422 struct rb_node
*node
;
1423 struct extent_state
*state
;
1426 * this search will find all the extents that end after
1429 node
= tree_search(tree
, start
);
1434 state
= rb_entry(node
, struct extent_state
, rb_node
);
1435 if (state
->end
>= start
&& (state
->state
& bits
))
1438 node
= rb_next(node
);
1447 * find the first offset in the io tree with 'bits' set. zero is
1448 * returned if we find something, and *start_ret and *end_ret are
1449 * set to reflect the state struct that was found.
1451 * If nothing was found, 1 is returned. If found something, return 0.
1453 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1454 u64
*start_ret
, u64
*end_ret
, unsigned bits
,
1455 struct extent_state
**cached_state
)
1457 struct extent_state
*state
;
1461 spin_lock(&tree
->lock
);
1462 if (cached_state
&& *cached_state
) {
1463 state
= *cached_state
;
1464 if (state
->end
== start
- 1 && extent_state_in_tree(state
)) {
1465 n
= rb_next(&state
->rb_node
);
1467 state
= rb_entry(n
, struct extent_state
,
1469 if (state
->state
& bits
)
1473 free_extent_state(*cached_state
);
1474 *cached_state
= NULL
;
1477 free_extent_state(*cached_state
);
1478 *cached_state
= NULL
;
1481 state
= find_first_extent_bit_state(tree
, start
, bits
);
1484 cache_state_if_flags(state
, cached_state
, 0);
1485 *start_ret
= state
->start
;
1486 *end_ret
= state
->end
;
1490 spin_unlock(&tree
->lock
);
1495 * find a contiguous range of bytes in the file marked as delalloc, not
1496 * more than 'max_bytes'. start and end are used to return the range,
1498 * 1 is returned if we find something, 0 if nothing was in the tree
1500 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1501 u64
*start
, u64
*end
, u64 max_bytes
,
1502 struct extent_state
**cached_state
)
1504 struct rb_node
*node
;
1505 struct extent_state
*state
;
1506 u64 cur_start
= *start
;
1508 u64 total_bytes
= 0;
1510 spin_lock(&tree
->lock
);
1513 * this search will find all the extents that end after
1516 node
= tree_search(tree
, cur_start
);
1524 state
= rb_entry(node
, struct extent_state
, rb_node
);
1525 if (found
&& (state
->start
!= cur_start
||
1526 (state
->state
& EXTENT_BOUNDARY
))) {
1529 if (!(state
->state
& EXTENT_DELALLOC
)) {
1535 *start
= state
->start
;
1536 *cached_state
= state
;
1537 atomic_inc(&state
->refs
);
1541 cur_start
= state
->end
+ 1;
1542 node
= rb_next(node
);
1543 total_bytes
+= state
->end
- state
->start
+ 1;
1544 if (total_bytes
>= max_bytes
)
1550 spin_unlock(&tree
->lock
);
1554 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1555 struct page
*locked_page
,
1559 struct page
*pages
[16];
1560 unsigned long index
= start
>> PAGE_SHIFT
;
1561 unsigned long end_index
= end
>> PAGE_SHIFT
;
1562 unsigned long nr_pages
= end_index
- index
+ 1;
1565 if (index
== locked_page
->index
&& end_index
== index
)
1568 while (nr_pages
> 0) {
1569 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1570 min_t(unsigned long, nr_pages
,
1571 ARRAY_SIZE(pages
)), pages
);
1572 for (i
= 0; i
< ret
; i
++) {
1573 if (pages
[i
] != locked_page
)
1574 unlock_page(pages
[i
]);
1583 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1584 struct page
*locked_page
,
1588 unsigned long index
= delalloc_start
>> PAGE_SHIFT
;
1589 unsigned long start_index
= index
;
1590 unsigned long end_index
= delalloc_end
>> PAGE_SHIFT
;
1591 unsigned long pages_locked
= 0;
1592 struct page
*pages
[16];
1593 unsigned long nrpages
;
1597 /* the caller is responsible for locking the start index */
1598 if (index
== locked_page
->index
&& index
== end_index
)
1601 /* skip the page at the start index */
1602 nrpages
= end_index
- index
+ 1;
1603 while (nrpages
> 0) {
1604 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1605 min_t(unsigned long,
1606 nrpages
, ARRAY_SIZE(pages
)), pages
);
1611 /* now we have an array of pages, lock them all */
1612 for (i
= 0; i
< ret
; i
++) {
1614 * the caller is taking responsibility for
1617 if (pages
[i
] != locked_page
) {
1618 lock_page(pages
[i
]);
1619 if (!PageDirty(pages
[i
]) ||
1620 pages
[i
]->mapping
!= inode
->i_mapping
) {
1622 unlock_page(pages
[i
]);
1636 if (ret
&& pages_locked
) {
1637 __unlock_for_delalloc(inode
, locked_page
,
1639 ((u64
)(start_index
+ pages_locked
- 1)) <<
1646 * find a contiguous range of bytes in the file marked as delalloc, not
1647 * more than 'max_bytes'. start and end are used to return the range,
1649 * 1 is returned if we find something, 0 if nothing was in the tree
1651 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1652 struct extent_io_tree
*tree
,
1653 struct page
*locked_page
, u64
*start
,
1654 u64
*end
, u64 max_bytes
)
1659 struct extent_state
*cached_state
= NULL
;
1664 /* step one, find a bunch of delalloc bytes starting at start */
1665 delalloc_start
= *start
;
1667 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1668 max_bytes
, &cached_state
);
1669 if (!found
|| delalloc_end
<= *start
) {
1670 *start
= delalloc_start
;
1671 *end
= delalloc_end
;
1672 free_extent_state(cached_state
);
1677 * start comes from the offset of locked_page. We have to lock
1678 * pages in order, so we can't process delalloc bytes before
1681 if (delalloc_start
< *start
)
1682 delalloc_start
= *start
;
1685 * make sure to limit the number of pages we try to lock down
1687 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1688 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1690 /* step two, lock all the pages after the page that has start */
1691 ret
= lock_delalloc_pages(inode
, locked_page
,
1692 delalloc_start
, delalloc_end
);
1693 if (ret
== -EAGAIN
) {
1694 /* some of the pages are gone, lets avoid looping by
1695 * shortening the size of the delalloc range we're searching
1697 free_extent_state(cached_state
);
1698 cached_state
= NULL
;
1700 max_bytes
= PAGE_SIZE
;
1708 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1710 /* step three, lock the state bits for the whole range */
1711 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, &cached_state
);
1713 /* then test to make sure it is all still delalloc */
1714 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1715 EXTENT_DELALLOC
, 1, cached_state
);
1717 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1718 &cached_state
, GFP_NOFS
);
1719 __unlock_for_delalloc(inode
, locked_page
,
1720 delalloc_start
, delalloc_end
);
1724 free_extent_state(cached_state
);
1725 *start
= delalloc_start
;
1726 *end
= delalloc_end
;
1731 void extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1732 struct page
*locked_page
,
1733 unsigned clear_bits
,
1734 unsigned long page_ops
)
1736 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1738 struct page
*pages
[16];
1739 unsigned long index
= start
>> PAGE_SHIFT
;
1740 unsigned long end_index
= end
>> PAGE_SHIFT
;
1741 unsigned long nr_pages
= end_index
- index
+ 1;
1744 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1748 if ((page_ops
& PAGE_SET_ERROR
) && nr_pages
> 0)
1749 mapping_set_error(inode
->i_mapping
, -EIO
);
1751 while (nr_pages
> 0) {
1752 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1753 min_t(unsigned long,
1754 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1755 for (i
= 0; i
< ret
; i
++) {
1757 if (page_ops
& PAGE_SET_PRIVATE2
)
1758 SetPagePrivate2(pages
[i
]);
1760 if (pages
[i
] == locked_page
) {
1764 if (page_ops
& PAGE_CLEAR_DIRTY
)
1765 clear_page_dirty_for_io(pages
[i
]);
1766 if (page_ops
& PAGE_SET_WRITEBACK
)
1767 set_page_writeback(pages
[i
]);
1768 if (page_ops
& PAGE_SET_ERROR
)
1769 SetPageError(pages
[i
]);
1770 if (page_ops
& PAGE_END_WRITEBACK
)
1771 end_page_writeback(pages
[i
]);
1772 if (page_ops
& PAGE_UNLOCK
)
1773 unlock_page(pages
[i
]);
1783 * count the number of bytes in the tree that have a given bit(s)
1784 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1785 * cached. The total number found is returned.
1787 u64
count_range_bits(struct extent_io_tree
*tree
,
1788 u64
*start
, u64 search_end
, u64 max_bytes
,
1789 unsigned bits
, int contig
)
1791 struct rb_node
*node
;
1792 struct extent_state
*state
;
1793 u64 cur_start
= *start
;
1794 u64 total_bytes
= 0;
1798 if (WARN_ON(search_end
<= cur_start
))
1801 spin_lock(&tree
->lock
);
1802 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1803 total_bytes
= tree
->dirty_bytes
;
1807 * this search will find all the extents that end after
1810 node
= tree_search(tree
, cur_start
);
1815 state
= rb_entry(node
, struct extent_state
, rb_node
);
1816 if (state
->start
> search_end
)
1818 if (contig
&& found
&& state
->start
> last
+ 1)
1820 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1821 total_bytes
+= min(search_end
, state
->end
) + 1 -
1822 max(cur_start
, state
->start
);
1823 if (total_bytes
>= max_bytes
)
1826 *start
= max(cur_start
, state
->start
);
1830 } else if (contig
&& found
) {
1833 node
= rb_next(node
);
1838 spin_unlock(&tree
->lock
);
1843 * set the private field for a given byte offset in the tree. If there isn't
1844 * an extent_state there already, this does nothing.
1846 static noinline
int set_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1847 struct io_failure_record
*failrec
)
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
->failrec
= failrec
;
1870 spin_unlock(&tree
->lock
);
1874 static noinline
int get_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1875 struct io_failure_record
**failrec
)
1877 struct rb_node
*node
;
1878 struct extent_state
*state
;
1881 spin_lock(&tree
->lock
);
1883 * this search will find all the extents that end after
1886 node
= tree_search(tree
, start
);
1891 state
= rb_entry(node
, struct extent_state
, rb_node
);
1892 if (state
->start
!= start
) {
1896 *failrec
= state
->failrec
;
1898 spin_unlock(&tree
->lock
);
1903 * searches a range in the state tree for a given mask.
1904 * If 'filled' == 1, this returns 1 only if every extent in the tree
1905 * has the bits set. Otherwise, 1 is returned if any bit in the
1906 * range is found set.
1908 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1909 unsigned bits
, int filled
, struct extent_state
*cached
)
1911 struct extent_state
*state
= NULL
;
1912 struct rb_node
*node
;
1915 spin_lock(&tree
->lock
);
1916 if (cached
&& extent_state_in_tree(cached
) && cached
->start
<= start
&&
1917 cached
->end
> start
)
1918 node
= &cached
->rb_node
;
1920 node
= tree_search(tree
, start
);
1921 while (node
&& start
<= end
) {
1922 state
= rb_entry(node
, struct extent_state
, rb_node
);
1924 if (filled
&& state
->start
> start
) {
1929 if (state
->start
> end
)
1932 if (state
->state
& bits
) {
1936 } else if (filled
) {
1941 if (state
->end
== (u64
)-1)
1944 start
= state
->end
+ 1;
1947 node
= rb_next(node
);
1954 spin_unlock(&tree
->lock
);
1959 * helper function to set a given page up to date if all the
1960 * extents in the tree for that page are up to date
1962 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1964 u64 start
= page_offset(page
);
1965 u64 end
= start
+ PAGE_SIZE
- 1;
1966 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1967 SetPageUptodate(page
);
1970 int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
)
1974 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1976 set_state_failrec(failure_tree
, rec
->start
, NULL
);
1977 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1978 rec
->start
+ rec
->len
- 1,
1979 EXTENT_LOCKED
| EXTENT_DIRTY
);
1983 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1984 rec
->start
+ rec
->len
- 1,
1994 * this bypasses the standard btrfs submit functions deliberately, as
1995 * the standard behavior is to write all copies in a raid setup. here we only
1996 * want to write the one bad copy. so we do the mapping for ourselves and issue
1997 * submit_bio directly.
1998 * to avoid any synchronization issues, wait for the data after writing, which
1999 * actually prevents the read that triggered the error from finishing.
2000 * currently, there can be no more than two copies of every data bit. thus,
2001 * exactly one rewrite is required.
2003 int repair_io_failure(struct inode
*inode
, u64 start
, u64 length
, u64 logical
,
2004 struct page
*page
, unsigned int pg_offset
, int mirror_num
)
2006 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2008 struct btrfs_device
*dev
;
2011 struct btrfs_bio
*bbio
= NULL
;
2012 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2015 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2016 BUG_ON(!mirror_num
);
2018 /* we can't repair anything in raid56 yet */
2019 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2022 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2025 bio
->bi_iter
.bi_size
= 0;
2026 map_length
= length
;
2029 * Avoid races with device replace and make sure our bbio has devices
2030 * associated to its stripes that don't go away while we are doing the
2031 * read repair operation.
2033 btrfs_bio_counter_inc_blocked(fs_info
);
2034 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2035 &map_length
, &bbio
, mirror_num
);
2037 btrfs_bio_counter_dec(fs_info
);
2041 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2042 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2043 bio
->bi_iter
.bi_sector
= sector
;
2044 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2045 btrfs_put_bbio(bbio
);
2046 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2047 btrfs_bio_counter_dec(fs_info
);
2051 bio
->bi_bdev
= dev
->bdev
;
2052 bio_add_page(bio
, page
, length
, pg_offset
);
2054 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
)) {
2055 /* try to remap that extent elsewhere? */
2056 btrfs_bio_counter_dec(fs_info
);
2058 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2062 btrfs_info_rl_in_rcu(fs_info
,
2063 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2064 btrfs_ino(inode
), start
,
2065 rcu_str_deref(dev
->name
), sector
);
2066 btrfs_bio_counter_dec(fs_info
);
2071 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2074 u64 start
= eb
->start
;
2075 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2078 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2081 for (i
= 0; i
< num_pages
; i
++) {
2082 struct page
*p
= eb
->pages
[i
];
2084 ret
= repair_io_failure(root
->fs_info
->btree_inode
, start
,
2085 PAGE_SIZE
, start
, p
,
2086 start
- page_offset(p
), mirror_num
);
2096 * each time an IO finishes, we do a fast check in the IO failure tree
2097 * to see if we need to process or clean up an io_failure_record
2099 int clean_io_failure(struct inode
*inode
, u64 start
, struct page
*page
,
2100 unsigned int pg_offset
)
2103 struct io_failure_record
*failrec
;
2104 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2105 struct extent_state
*state
;
2110 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2111 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2115 ret
= get_state_failrec(&BTRFS_I(inode
)->io_failure_tree
, start
,
2120 BUG_ON(!failrec
->this_mirror
);
2122 if (failrec
->in_validation
) {
2123 /* there was no real error, just free the record */
2124 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2128 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2131 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2132 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2135 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2137 if (state
&& state
->start
<= failrec
->start
&&
2138 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2139 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2141 if (num_copies
> 1) {
2142 repair_io_failure(inode
, start
, failrec
->len
,
2143 failrec
->logical
, page
,
2144 pg_offset
, failrec
->failed_mirror
);
2149 free_io_failure(inode
, failrec
);
2155 * Can be called when
2156 * - hold extent lock
2157 * - under ordered extent
2158 * - the inode is freeing
2160 void btrfs_free_io_failure_record(struct inode
*inode
, u64 start
, u64 end
)
2162 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2163 struct io_failure_record
*failrec
;
2164 struct extent_state
*state
, *next
;
2166 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2169 spin_lock(&failure_tree
->lock
);
2170 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2172 if (state
->start
> end
)
2175 ASSERT(state
->end
<= end
);
2177 next
= next_state(state
);
2179 failrec
= state
->failrec
;
2180 free_extent_state(state
);
2185 spin_unlock(&failure_tree
->lock
);
2188 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2189 struct io_failure_record
**failrec_ret
)
2191 struct io_failure_record
*failrec
;
2192 struct extent_map
*em
;
2193 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2194 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2195 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2199 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2201 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2205 failrec
->start
= start
;
2206 failrec
->len
= end
- start
+ 1;
2207 failrec
->this_mirror
= 0;
2208 failrec
->bio_flags
= 0;
2209 failrec
->in_validation
= 0;
2211 read_lock(&em_tree
->lock
);
2212 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2214 read_unlock(&em_tree
->lock
);
2219 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2220 free_extent_map(em
);
2223 read_unlock(&em_tree
->lock
);
2229 logical
= start
- em
->start
;
2230 logical
= em
->block_start
+ logical
;
2231 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2232 logical
= em
->block_start
;
2233 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2234 extent_set_compress_type(&failrec
->bio_flags
,
2238 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2239 logical
, start
, failrec
->len
);
2241 failrec
->logical
= logical
;
2242 free_extent_map(em
);
2244 /* set the bits in the private failure tree */
2245 ret
= set_extent_bits(failure_tree
, start
, end
,
2246 EXTENT_LOCKED
| EXTENT_DIRTY
);
2248 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2249 /* set the bits in the inode's tree */
2251 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2257 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2258 failrec
->logical
, failrec
->start
, failrec
->len
,
2259 failrec
->in_validation
);
2261 * when data can be on disk more than twice, add to failrec here
2262 * (e.g. with a list for failed_mirror) to make
2263 * clean_io_failure() clean all those errors at once.
2267 *failrec_ret
= failrec
;
2272 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2273 struct io_failure_record
*failrec
, int failed_mirror
)
2277 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2278 failrec
->logical
, failrec
->len
);
2279 if (num_copies
== 1) {
2281 * we only have a single copy of the data, so don't bother with
2282 * all the retry and error correction code that follows. no
2283 * matter what the error is, it is very likely to persist.
2285 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2286 num_copies
, failrec
->this_mirror
, failed_mirror
);
2291 * there are two premises:
2292 * a) deliver good data to the caller
2293 * b) correct the bad sectors on disk
2295 if (failed_bio
->bi_vcnt
> 1) {
2297 * to fulfill b), we need to know the exact failing sectors, as
2298 * we don't want to rewrite any more than the failed ones. thus,
2299 * we need separate read requests for the failed bio
2301 * if the following BUG_ON triggers, our validation request got
2302 * merged. we need separate requests for our algorithm to work.
2304 BUG_ON(failrec
->in_validation
);
2305 failrec
->in_validation
= 1;
2306 failrec
->this_mirror
= failed_mirror
;
2309 * we're ready to fulfill a) and b) alongside. get a good copy
2310 * of the failed sector and if we succeed, we have setup
2311 * everything for repair_io_failure to do the rest for us.
2313 if (failrec
->in_validation
) {
2314 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2315 failrec
->in_validation
= 0;
2316 failrec
->this_mirror
= 0;
2318 failrec
->failed_mirror
= failed_mirror
;
2319 failrec
->this_mirror
++;
2320 if (failrec
->this_mirror
== failed_mirror
)
2321 failrec
->this_mirror
++;
2324 if (failrec
->this_mirror
> num_copies
) {
2325 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2326 num_copies
, failrec
->this_mirror
, failed_mirror
);
2334 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2335 struct io_failure_record
*failrec
,
2336 struct page
*page
, int pg_offset
, int icsum
,
2337 bio_end_io_t
*endio_func
, void *data
)
2340 struct btrfs_io_bio
*btrfs_failed_bio
;
2341 struct btrfs_io_bio
*btrfs_bio
;
2343 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2347 bio
->bi_end_io
= endio_func
;
2348 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2349 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2350 bio
->bi_iter
.bi_size
= 0;
2351 bio
->bi_private
= data
;
2353 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2354 if (btrfs_failed_bio
->csum
) {
2355 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2356 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2358 btrfs_bio
= btrfs_io_bio(bio
);
2359 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2361 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2365 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2371 * this is a generic handler for readpage errors (default
2372 * readpage_io_failed_hook). if other copies exist, read those and write back
2373 * good data to the failed position. does not investigate in remapping the
2374 * failed extent elsewhere, hoping the device will be smart enough to do this as
2378 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2379 struct page
*page
, u64 start
, u64 end
,
2382 struct io_failure_record
*failrec
;
2383 struct inode
*inode
= page
->mapping
->host
;
2384 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2389 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2391 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2395 ret
= btrfs_check_repairable(inode
, failed_bio
, failrec
, failed_mirror
);
2397 free_io_failure(inode
, failrec
);
2401 if (failed_bio
->bi_vcnt
> 1)
2402 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2404 read_mode
= READ_SYNC
;
2406 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2407 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2408 start
- page_offset(page
),
2409 (int)phy_offset
, failed_bio
->bi_end_io
,
2412 free_io_failure(inode
, failrec
);
2416 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2417 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2419 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2420 failrec
->this_mirror
,
2421 failrec
->bio_flags
, 0);
2423 free_io_failure(inode
, failrec
);
2430 /* lots and lots of room for performance fixes in the end_bio funcs */
2432 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2434 int uptodate
= (err
== 0);
2435 struct extent_io_tree
*tree
;
2438 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2440 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2441 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2442 end
, NULL
, uptodate
);
2448 ClearPageUptodate(page
);
2450 ret
= ret
< 0 ? ret
: -EIO
;
2451 mapping_set_error(page
->mapping
, ret
);
2456 * after a writepage IO is done, we need to:
2457 * clear the uptodate bits on error
2458 * clear the writeback bits in the extent tree for this IO
2459 * end_page_writeback if the page has no more pending IO
2461 * Scheduling is not allowed, so the extent state tree is expected
2462 * to have one and only one object corresponding to this IO.
2464 static void end_bio_extent_writepage(struct bio
*bio
)
2466 struct bio_vec
*bvec
;
2471 bio_for_each_segment_all(bvec
, bio
, i
) {
2472 struct page
*page
= bvec
->bv_page
;
2474 /* We always issue full-page reads, but if some block
2475 * in a page fails to read, blk_update_request() will
2476 * advance bv_offset and adjust bv_len to compensate.
2477 * Print a warning for nonzero offsets, and an error
2478 * if they don't add up to a full page. */
2479 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2480 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2481 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2482 "partial page write in btrfs with offset %u and length %u",
2483 bvec
->bv_offset
, bvec
->bv_len
);
2485 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2486 "incomplete page write in btrfs with offset %u and "
2488 bvec
->bv_offset
, bvec
->bv_len
);
2491 start
= page_offset(page
);
2492 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2494 end_extent_writepage(page
, bio
->bi_error
, start
, end
);
2495 end_page_writeback(page
);
2502 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2505 struct extent_state
*cached
= NULL
;
2506 u64 end
= start
+ len
- 1;
2508 if (uptodate
&& tree
->track_uptodate
)
2509 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2510 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2514 * after a readpage IO is done, we need to:
2515 * clear the uptodate bits on error
2516 * set the uptodate bits if things worked
2517 * set the page up to date if all extents in the tree are uptodate
2518 * clear the lock bit in the extent tree
2519 * unlock the page if there are no other extents locked for it
2521 * Scheduling is not allowed, so the extent state tree is expected
2522 * to have one and only one object corresponding to this IO.
2524 static void end_bio_extent_readpage(struct bio
*bio
)
2526 struct bio_vec
*bvec
;
2527 int uptodate
= !bio
->bi_error
;
2528 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2529 struct extent_io_tree
*tree
;
2534 u64 extent_start
= 0;
2540 bio_for_each_segment_all(bvec
, bio
, i
) {
2541 struct page
*page
= bvec
->bv_page
;
2542 struct inode
*inode
= page
->mapping
->host
;
2544 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2545 "mirror=%u\n", (u64
)bio
->bi_iter
.bi_sector
,
2546 bio
->bi_error
, io_bio
->mirror_num
);
2547 tree
= &BTRFS_I(inode
)->io_tree
;
2549 /* We always issue full-page reads, but if some block
2550 * in a page fails to read, blk_update_request() will
2551 * advance bv_offset and adjust bv_len to compensate.
2552 * Print a warning for nonzero offsets, and an error
2553 * if they don't add up to a full page. */
2554 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2555 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2556 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2557 "partial page read in btrfs with offset %u and length %u",
2558 bvec
->bv_offset
, bvec
->bv_len
);
2560 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2561 "incomplete page read in btrfs with offset %u and "
2563 bvec
->bv_offset
, bvec
->bv_len
);
2566 start
= page_offset(page
);
2567 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2570 mirror
= io_bio
->mirror_num
;
2571 if (likely(uptodate
&& tree
->ops
&&
2572 tree
->ops
->readpage_end_io_hook
)) {
2573 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2579 clean_io_failure(inode
, start
, page
, 0);
2582 if (likely(uptodate
))
2585 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2586 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2587 if (!ret
&& !bio
->bi_error
)
2591 * The generic bio_readpage_error handles errors the
2592 * following way: If possible, new read requests are
2593 * created and submitted and will end up in
2594 * end_bio_extent_readpage as well (if we're lucky, not
2595 * in the !uptodate case). In that case it returns 0 and
2596 * we just go on with the next page in our bio. If it
2597 * can't handle the error it will return -EIO and we
2598 * remain responsible for that page.
2600 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2603 uptodate
= !bio
->bi_error
;
2609 if (likely(uptodate
)) {
2610 loff_t i_size
= i_size_read(inode
);
2611 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2614 /* Zero out the end if this page straddles i_size */
2615 off
= i_size
& (PAGE_SIZE
-1);
2616 if (page
->index
== end_index
&& off
)
2617 zero_user_segment(page
, off
, PAGE_SIZE
);
2618 SetPageUptodate(page
);
2620 ClearPageUptodate(page
);
2626 if (unlikely(!uptodate
)) {
2628 endio_readpage_release_extent(tree
,
2634 endio_readpage_release_extent(tree
, start
,
2635 end
- start
+ 1, 0);
2636 } else if (!extent_len
) {
2637 extent_start
= start
;
2638 extent_len
= end
+ 1 - start
;
2639 } else if (extent_start
+ extent_len
== start
) {
2640 extent_len
+= end
+ 1 - start
;
2642 endio_readpage_release_extent(tree
, extent_start
,
2643 extent_len
, uptodate
);
2644 extent_start
= start
;
2645 extent_len
= end
+ 1 - start
;
2650 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2653 io_bio
->end_io(io_bio
, bio
->bi_error
);
2658 * this allocates from the btrfs_bioset. We're returning a bio right now
2659 * but you can call btrfs_io_bio for the appropriate container_of magic
2662 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2665 struct btrfs_io_bio
*btrfs_bio
;
2668 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2670 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2671 while (!bio
&& (nr_vecs
/= 2)) {
2672 bio
= bio_alloc_bioset(gfp_flags
,
2673 nr_vecs
, btrfs_bioset
);
2678 bio
->bi_bdev
= bdev
;
2679 bio
->bi_iter
.bi_sector
= first_sector
;
2680 btrfs_bio
= btrfs_io_bio(bio
);
2681 btrfs_bio
->csum
= NULL
;
2682 btrfs_bio
->csum_allocated
= NULL
;
2683 btrfs_bio
->end_io
= NULL
;
2688 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2690 struct btrfs_io_bio
*btrfs_bio
;
2693 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2695 btrfs_bio
= btrfs_io_bio(new);
2696 btrfs_bio
->csum
= NULL
;
2697 btrfs_bio
->csum_allocated
= NULL
;
2698 btrfs_bio
->end_io
= NULL
;
2700 #ifdef CONFIG_BLK_CGROUP
2701 /* FIXME, put this into bio_clone_bioset */
2703 bio_associate_blkcg(new, bio
->bi_css
);
2709 /* this also allocates from the btrfs_bioset */
2710 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2712 struct btrfs_io_bio
*btrfs_bio
;
2715 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2717 btrfs_bio
= btrfs_io_bio(bio
);
2718 btrfs_bio
->csum
= NULL
;
2719 btrfs_bio
->csum_allocated
= NULL
;
2720 btrfs_bio
->end_io
= NULL
;
2726 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2727 int mirror_num
, unsigned long bio_flags
)
2730 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2731 struct page
*page
= bvec
->bv_page
;
2732 struct extent_io_tree
*tree
= bio
->bi_private
;
2735 start
= page_offset(page
) + bvec
->bv_offset
;
2737 bio
->bi_private
= NULL
;
2741 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2742 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2743 mirror_num
, bio_flags
, start
);
2745 btrfsic_submit_bio(rw
, bio
);
2751 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2752 unsigned long offset
, size_t size
, struct bio
*bio
,
2753 unsigned long bio_flags
)
2756 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2757 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2764 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2765 struct writeback_control
*wbc
,
2766 struct page
*page
, sector_t sector
,
2767 size_t size
, unsigned long offset
,
2768 struct block_device
*bdev
,
2769 struct bio
**bio_ret
,
2770 unsigned long max_pages
,
2771 bio_end_io_t end_io_func
,
2773 unsigned long prev_bio_flags
,
2774 unsigned long bio_flags
,
2775 bool force_bio_submit
)
2780 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2781 size_t page_size
= min_t(size_t, size
, PAGE_SIZE
);
2783 if (bio_ret
&& *bio_ret
) {
2786 contig
= bio
->bi_iter
.bi_sector
== sector
;
2788 contig
= bio_end_sector(bio
) == sector
;
2790 if (prev_bio_flags
!= bio_flags
|| !contig
||
2792 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2793 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2794 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2803 wbc_account_io(wbc
, page
, page_size
);
2808 bio
= btrfs_bio_alloc(bdev
, sector
, BIO_MAX_PAGES
,
2809 GFP_NOFS
| __GFP_HIGH
);
2813 bio_add_page(bio
, page
, page_size
, offset
);
2814 bio
->bi_end_io
= end_io_func
;
2815 bio
->bi_private
= tree
;
2817 wbc_init_bio(wbc
, bio
);
2818 wbc_account_io(wbc
, page
, page_size
);
2824 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2829 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2832 if (!PagePrivate(page
)) {
2833 SetPagePrivate(page
);
2835 set_page_private(page
, (unsigned long)eb
);
2837 WARN_ON(page
->private != (unsigned long)eb
);
2841 void set_page_extent_mapped(struct page
*page
)
2843 if (!PagePrivate(page
)) {
2844 SetPagePrivate(page
);
2846 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2850 static struct extent_map
*
2851 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2852 u64 start
, u64 len
, get_extent_t
*get_extent
,
2853 struct extent_map
**em_cached
)
2855 struct extent_map
*em
;
2857 if (em_cached
&& *em_cached
) {
2859 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2860 start
< extent_map_end(em
)) {
2861 atomic_inc(&em
->refs
);
2865 free_extent_map(em
);
2869 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2870 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2872 atomic_inc(&em
->refs
);
2878 * basic readpage implementation. Locked extent state structs are inserted
2879 * into the tree that are removed when the IO is done (by the end_io
2881 * XXX JDM: This needs looking at to ensure proper page locking
2883 static int __do_readpage(struct extent_io_tree
*tree
,
2885 get_extent_t
*get_extent
,
2886 struct extent_map
**em_cached
,
2887 struct bio
**bio
, int mirror_num
,
2888 unsigned long *bio_flags
, int rw
,
2891 struct inode
*inode
= page
->mapping
->host
;
2892 u64 start
= page_offset(page
);
2893 u64 page_end
= start
+ PAGE_SIZE
- 1;
2897 u64 last_byte
= i_size_read(inode
);
2901 struct extent_map
*em
;
2902 struct block_device
*bdev
;
2905 size_t pg_offset
= 0;
2907 size_t disk_io_size
;
2908 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2909 unsigned long this_bio_flag
= 0;
2911 set_page_extent_mapped(page
);
2914 if (!PageUptodate(page
)) {
2915 if (cleancache_get_page(page
) == 0) {
2916 BUG_ON(blocksize
!= PAGE_SIZE
);
2917 unlock_extent(tree
, start
, end
);
2922 if (page
->index
== last_byte
>> PAGE_SHIFT
) {
2924 size_t zero_offset
= last_byte
& (PAGE_SIZE
- 1);
2927 iosize
= PAGE_SIZE
- zero_offset
;
2928 userpage
= kmap_atomic(page
);
2929 memset(userpage
+ zero_offset
, 0, iosize
);
2930 flush_dcache_page(page
);
2931 kunmap_atomic(userpage
);
2934 while (cur
<= end
) {
2935 unsigned long pnr
= (last_byte
>> PAGE_SHIFT
) + 1;
2936 bool force_bio_submit
= false;
2938 if (cur
>= last_byte
) {
2940 struct extent_state
*cached
= NULL
;
2942 iosize
= PAGE_SIZE
- pg_offset
;
2943 userpage
= kmap_atomic(page
);
2944 memset(userpage
+ pg_offset
, 0, iosize
);
2945 flush_dcache_page(page
);
2946 kunmap_atomic(userpage
);
2947 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2949 unlock_extent_cached(tree
, cur
,
2954 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2955 end
- cur
+ 1, get_extent
, em_cached
);
2956 if (IS_ERR_OR_NULL(em
)) {
2958 unlock_extent(tree
, cur
, end
);
2961 extent_offset
= cur
- em
->start
;
2962 BUG_ON(extent_map_end(em
) <= cur
);
2965 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2966 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2967 extent_set_compress_type(&this_bio_flag
,
2971 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2972 cur_end
= min(extent_map_end(em
) - 1, end
);
2973 iosize
= ALIGN(iosize
, blocksize
);
2974 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2975 disk_io_size
= em
->block_len
;
2976 sector
= em
->block_start
>> 9;
2978 sector
= (em
->block_start
+ extent_offset
) >> 9;
2979 disk_io_size
= iosize
;
2982 block_start
= em
->block_start
;
2983 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2984 block_start
= EXTENT_MAP_HOLE
;
2987 * If we have a file range that points to a compressed extent
2988 * and it's followed by a consecutive file range that points to
2989 * to the same compressed extent (possibly with a different
2990 * offset and/or length, so it either points to the whole extent
2991 * or only part of it), we must make sure we do not submit a
2992 * single bio to populate the pages for the 2 ranges because
2993 * this makes the compressed extent read zero out the pages
2994 * belonging to the 2nd range. Imagine the following scenario:
2997 * [0 - 8K] [8K - 24K]
3000 * points to extent X, points to extent X,
3001 * offset 4K, length of 8K offset 0, length 16K
3003 * [extent X, compressed length = 4K uncompressed length = 16K]
3005 * If the bio to read the compressed extent covers both ranges,
3006 * it will decompress extent X into the pages belonging to the
3007 * first range and then it will stop, zeroing out the remaining
3008 * pages that belong to the other range that points to extent X.
3009 * So here we make sure we submit 2 bios, one for the first
3010 * range and another one for the third range. Both will target
3011 * the same physical extent from disk, but we can't currently
3012 * make the compressed bio endio callback populate the pages
3013 * for both ranges because each compressed bio is tightly
3014 * coupled with a single extent map, and each range can have
3015 * an extent map with a different offset value relative to the
3016 * uncompressed data of our extent and different lengths. This
3017 * is a corner case so we prioritize correctness over
3018 * non-optimal behavior (submitting 2 bios for the same extent).
3020 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3021 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3022 *prev_em_start
!= em
->orig_start
)
3023 force_bio_submit
= true;
3026 *prev_em_start
= em
->orig_start
;
3028 free_extent_map(em
);
3031 /* we've found a hole, just zero and go on */
3032 if (block_start
== EXTENT_MAP_HOLE
) {
3034 struct extent_state
*cached
= NULL
;
3036 userpage
= kmap_atomic(page
);
3037 memset(userpage
+ pg_offset
, 0, iosize
);
3038 flush_dcache_page(page
);
3039 kunmap_atomic(userpage
);
3041 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3043 unlock_extent_cached(tree
, cur
,
3047 pg_offset
+= iosize
;
3050 /* the get_extent function already copied into the page */
3051 if (test_range_bit(tree
, cur
, cur_end
,
3052 EXTENT_UPTODATE
, 1, NULL
)) {
3053 check_page_uptodate(tree
, page
);
3054 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3056 pg_offset
+= iosize
;
3059 /* we have an inline extent but it didn't get marked up
3060 * to date. Error out
3062 if (block_start
== EXTENT_MAP_INLINE
) {
3064 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3066 pg_offset
+= iosize
;
3071 ret
= submit_extent_page(rw
, tree
, NULL
, page
,
3072 sector
, disk_io_size
, pg_offset
,
3074 end_bio_extent_readpage
, mirror_num
,
3080 *bio_flags
= this_bio_flag
;
3083 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3086 pg_offset
+= iosize
;
3090 if (!PageError(page
))
3091 SetPageUptodate(page
);
3097 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3098 struct page
*pages
[], int nr_pages
,
3100 get_extent_t
*get_extent
,
3101 struct extent_map
**em_cached
,
3102 struct bio
**bio
, int mirror_num
,
3103 unsigned long *bio_flags
, int rw
,
3106 struct inode
*inode
;
3107 struct btrfs_ordered_extent
*ordered
;
3110 inode
= pages
[0]->mapping
->host
;
3112 lock_extent(tree
, start
, end
);
3113 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3117 unlock_extent(tree
, start
, end
);
3118 btrfs_start_ordered_extent(inode
, ordered
, 1);
3119 btrfs_put_ordered_extent(ordered
);
3122 for (index
= 0; index
< nr_pages
; index
++) {
3123 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3124 mirror_num
, bio_flags
, rw
, prev_em_start
);
3125 put_page(pages
[index
]);
3129 static void __extent_readpages(struct extent_io_tree
*tree
,
3130 struct page
*pages
[],
3131 int nr_pages
, get_extent_t
*get_extent
,
3132 struct extent_map
**em_cached
,
3133 struct bio
**bio
, int mirror_num
,
3134 unsigned long *bio_flags
, int rw
,
3141 int first_index
= 0;
3143 for (index
= 0; index
< nr_pages
; index
++) {
3144 page_start
= page_offset(pages
[index
]);
3147 end
= start
+ PAGE_SIZE
- 1;
3148 first_index
= index
;
3149 } else if (end
+ 1 == page_start
) {
3152 __do_contiguous_readpages(tree
, &pages
[first_index
],
3153 index
- first_index
, start
,
3154 end
, get_extent
, em_cached
,
3155 bio
, mirror_num
, bio_flags
,
3158 end
= start
+ PAGE_SIZE
- 1;
3159 first_index
= index
;
3164 __do_contiguous_readpages(tree
, &pages
[first_index
],
3165 index
- first_index
, start
,
3166 end
, get_extent
, em_cached
, bio
,
3167 mirror_num
, bio_flags
, rw
,
3171 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3173 get_extent_t
*get_extent
,
3174 struct bio
**bio
, int mirror_num
,
3175 unsigned long *bio_flags
, int rw
)
3177 struct inode
*inode
= page
->mapping
->host
;
3178 struct btrfs_ordered_extent
*ordered
;
3179 u64 start
= page_offset(page
);
3180 u64 end
= start
+ PAGE_SIZE
- 1;
3184 lock_extent(tree
, start
, end
);
3185 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3189 unlock_extent(tree
, start
, end
);
3190 btrfs_start_ordered_extent(inode
, ordered
, 1);
3191 btrfs_put_ordered_extent(ordered
);
3194 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3195 bio_flags
, rw
, NULL
);
3199 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3200 get_extent_t
*get_extent
, int mirror_num
)
3202 struct bio
*bio
= NULL
;
3203 unsigned long bio_flags
= 0;
3206 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3209 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3213 static void update_nr_written(struct page
*page
, struct writeback_control
*wbc
,
3214 unsigned long nr_written
)
3216 wbc
->nr_to_write
-= nr_written
;
3220 * helper for __extent_writepage, doing all of the delayed allocation setup.
3222 * This returns 1 if our fill_delalloc function did all the work required
3223 * to write the page (copy into inline extent). In this case the IO has
3224 * been started and the page is already unlocked.
3226 * This returns 0 if all went well (page still locked)
3227 * This returns < 0 if there were errors (page still locked)
3229 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3230 struct page
*page
, struct writeback_control
*wbc
,
3231 struct extent_page_data
*epd
,
3233 unsigned long *nr_written
)
3235 struct extent_io_tree
*tree
= epd
->tree
;
3236 u64 page_end
= delalloc_start
+ PAGE_SIZE
- 1;
3238 u64 delalloc_to_write
= 0;
3239 u64 delalloc_end
= 0;
3241 int page_started
= 0;
3243 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3246 while (delalloc_end
< page_end
) {
3247 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3251 BTRFS_MAX_EXTENT_SIZE
);
3252 if (nr_delalloc
== 0) {
3253 delalloc_start
= delalloc_end
+ 1;
3256 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3261 /* File system has been set read-only */
3264 /* fill_delalloc should be return < 0 for error
3265 * but just in case, we use > 0 here meaning the
3266 * IO is started, so we don't want to return > 0
3267 * unless things are going well.
3269 ret
= ret
< 0 ? ret
: -EIO
;
3273 * delalloc_end is already one less than the total length, so
3274 * we don't subtract one from PAGE_SIZE
3276 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3277 PAGE_SIZE
) >> PAGE_SHIFT
;
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_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 unsigned long max_nr
;
3379 if (cur
>= i_size
) {
3380 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3381 tree
->ops
->writepage_end_io_hook(page
, cur
,
3385 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3387 if (IS_ERR_OR_NULL(em
)) {
3389 ret
= PTR_ERR_OR_ZERO(em
);
3393 extent_offset
= cur
- em
->start
;
3394 em_end
= extent_map_end(em
);
3395 BUG_ON(em_end
<= cur
);
3397 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3398 iosize
= ALIGN(iosize
, blocksize
);
3399 sector
= (em
->block_start
+ extent_offset
) >> 9;
3401 block_start
= em
->block_start
;
3402 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3403 free_extent_map(em
);
3407 * compressed and inline extents are written through other
3410 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3411 block_start
== EXTENT_MAP_INLINE
) {
3413 * end_io notification does not happen here for
3414 * compressed extents
3416 if (!compressed
&& tree
->ops
&&
3417 tree
->ops
->writepage_end_io_hook
)
3418 tree
->ops
->writepage_end_io_hook(page
, cur
,
3421 else if (compressed
) {
3422 /* we don't want to end_page_writeback on
3423 * a compressed extent. this happens
3430 pg_offset
+= iosize
;
3434 max_nr
= (i_size
>> PAGE_SHIFT
) + 1;
3436 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3437 if (!PageWriteback(page
)) {
3438 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3439 "page %lu not writeback, cur %llu end %llu",
3440 page
->index
, cur
, end
);
3443 ret
= submit_extent_page(write_flags
, tree
, wbc
, page
,
3444 sector
, iosize
, pg_offset
,
3445 bdev
, &epd
->bio
, max_nr
,
3446 end_bio_extent_writepage
,
3452 pg_offset
+= iosize
;
3460 /* drop our reference on any cached states */
3461 free_extent_state(cached_state
);
3466 * the writepage semantics are similar to regular writepage. extent
3467 * records are inserted to lock ranges in the tree, and as dirty areas
3468 * are found, they are marked writeback. Then the lock bits are removed
3469 * and the end_io handler clears the writeback ranges
3471 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3474 struct inode
*inode
= page
->mapping
->host
;
3475 struct extent_page_data
*epd
= data
;
3476 u64 start
= page_offset(page
);
3477 u64 page_end
= start
+ PAGE_SIZE
- 1;
3480 size_t pg_offset
= 0;
3481 loff_t i_size
= i_size_read(inode
);
3482 unsigned long end_index
= i_size
>> PAGE_SHIFT
;
3484 unsigned long nr_written
= 0;
3486 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3487 write_flags
= WRITE_SYNC
;
3489 write_flags
= WRITE
;
3491 trace___extent_writepage(page
, inode
, wbc
);
3493 WARN_ON(!PageLocked(page
));
3495 ClearPageError(page
);
3497 pg_offset
= i_size
& (PAGE_SIZE
- 1);
3498 if (page
->index
> end_index
||
3499 (page
->index
== end_index
&& !pg_offset
)) {
3500 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3505 if (page
->index
== end_index
) {
3508 userpage
= kmap_atomic(page
);
3509 memset(userpage
+ pg_offset
, 0,
3510 PAGE_SIZE
- pg_offset
);
3511 kunmap_atomic(userpage
);
3512 flush_dcache_page(page
);
3517 set_page_extent_mapped(page
);
3519 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3525 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3526 i_size
, nr_written
, write_flags
, &nr
);
3532 /* make sure the mapping tag for page dirty gets cleared */
3533 set_page_writeback(page
);
3534 end_page_writeback(page
);
3536 if (PageError(page
)) {
3537 ret
= ret
< 0 ? ret
: -EIO
;
3538 end_extent_writepage(page
, ret
, start
, page_end
);
3547 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3549 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3550 TASK_UNINTERRUPTIBLE
);
3553 static noinline_for_stack
int
3554 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3555 struct btrfs_fs_info
*fs_info
,
3556 struct extent_page_data
*epd
)
3558 unsigned long i
, num_pages
;
3562 if (!btrfs_try_tree_write_lock(eb
)) {
3564 flush_write_bio(epd
);
3565 btrfs_tree_lock(eb
);
3568 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3569 btrfs_tree_unlock(eb
);
3573 flush_write_bio(epd
);
3577 wait_on_extent_buffer_writeback(eb
);
3578 btrfs_tree_lock(eb
);
3579 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3581 btrfs_tree_unlock(eb
);
3586 * We need to do this to prevent races in people who check if the eb is
3587 * under IO since we can end up having no IO bits set for a short period
3590 spin_lock(&eb
->refs_lock
);
3591 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3592 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3593 spin_unlock(&eb
->refs_lock
);
3594 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3595 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3597 fs_info
->dirty_metadata_batch
);
3600 spin_unlock(&eb
->refs_lock
);
3603 btrfs_tree_unlock(eb
);
3608 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3609 for (i
= 0; i
< num_pages
; i
++) {
3610 struct page
*p
= eb
->pages
[i
];
3612 if (!trylock_page(p
)) {
3614 flush_write_bio(epd
);
3624 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3626 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3627 smp_mb__after_atomic();
3628 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3631 static void set_btree_ioerr(struct page
*page
)
3633 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3634 struct btrfs_inode
*btree_ino
= BTRFS_I(eb
->fs_info
->btree_inode
);
3637 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3641 * If writeback for a btree extent that doesn't belong to a log tree
3642 * failed, increment the counter transaction->eb_write_errors.
3643 * We do this because while the transaction is running and before it's
3644 * committing (when we call filemap_fdata[write|wait]_range against
3645 * the btree inode), we might have
3646 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3647 * returns an error or an error happens during writeback, when we're
3648 * committing the transaction we wouldn't know about it, since the pages
3649 * can be no longer dirty nor marked anymore for writeback (if a
3650 * subsequent modification to the extent buffer didn't happen before the
3651 * transaction commit), which makes filemap_fdata[write|wait]_range not
3652 * able to find the pages tagged with SetPageError at transaction
3653 * commit time. So if this happens we must abort the transaction,
3654 * otherwise we commit a super block with btree roots that point to
3655 * btree nodes/leafs whose content on disk is invalid - either garbage
3656 * or the content of some node/leaf from a past generation that got
3657 * cowed or deleted and is no longer valid.
3659 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3660 * not be enough - we need to distinguish between log tree extents vs
3661 * non-log tree extents, and the next filemap_fdatawait_range() call
3662 * will catch and clear such errors in the mapping - and that call might
3663 * be from a log sync and not from a transaction commit. Also, checking
3664 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3665 * not done and would not be reliable - the eb might have been released
3666 * from memory and reading it back again means that flag would not be
3667 * set (since it's a runtime flag, not persisted on disk).
3669 * Using the flags below in the btree inode also makes us achieve the
3670 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3671 * writeback for all dirty pages and before filemap_fdatawait_range()
3672 * is called, the writeback for all dirty pages had already finished
3673 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3674 * filemap_fdatawait_range() would return success, as it could not know
3675 * that writeback errors happened (the pages were no longer tagged for
3678 switch (eb
->log_index
) {
3680 set_bit(BTRFS_INODE_BTREE_ERR
, &btree_ino
->runtime_flags
);
3683 set_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
3686 set_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
3689 BUG(); /* unexpected, logic error */
3693 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3695 struct bio_vec
*bvec
;
3696 struct extent_buffer
*eb
;
3699 bio_for_each_segment_all(bvec
, bio
, i
) {
3700 struct page
*page
= bvec
->bv_page
;
3702 eb
= (struct extent_buffer
*)page
->private;
3704 done
= atomic_dec_and_test(&eb
->io_pages
);
3706 if (bio
->bi_error
||
3707 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3708 ClearPageUptodate(page
);
3709 set_btree_ioerr(page
);
3712 end_page_writeback(page
);
3717 end_extent_buffer_writeback(eb
);
3723 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3724 struct btrfs_fs_info
*fs_info
,
3725 struct writeback_control
*wbc
,
3726 struct extent_page_data
*epd
)
3728 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3729 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3730 u64 offset
= eb
->start
;
3731 unsigned long i
, num_pages
;
3732 unsigned long bio_flags
= 0;
3733 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3736 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3737 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3738 atomic_set(&eb
->io_pages
, num_pages
);
3739 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3740 bio_flags
= EXTENT_BIO_TREE_LOG
;
3742 for (i
= 0; i
< num_pages
; i
++) {
3743 struct page
*p
= eb
->pages
[i
];
3745 clear_page_dirty_for_io(p
);
3746 set_page_writeback(p
);
3747 ret
= submit_extent_page(rw
, tree
, wbc
, p
, offset
>> 9,
3748 PAGE_SIZE
, 0, bdev
, &epd
->bio
,
3749 -1, end_bio_extent_buffer_writepage
,
3750 0, epd
->bio_flags
, bio_flags
, false);
3751 epd
->bio_flags
= bio_flags
;
3754 end_page_writeback(p
);
3755 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3756 end_extent_buffer_writeback(eb
);
3760 offset
+= PAGE_SIZE
;
3761 update_nr_written(p
, wbc
, 1);
3765 if (unlikely(ret
)) {
3766 for (; i
< num_pages
; i
++) {
3767 struct page
*p
= eb
->pages
[i
];
3768 clear_page_dirty_for_io(p
);
3776 int btree_write_cache_pages(struct address_space
*mapping
,
3777 struct writeback_control
*wbc
)
3779 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3780 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3781 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3782 struct extent_page_data epd
= {
3786 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3791 int nr_to_write_done
= 0;
3792 struct pagevec pvec
;
3795 pgoff_t end
; /* Inclusive */
3799 pagevec_init(&pvec
, 0);
3800 if (wbc
->range_cyclic
) {
3801 index
= mapping
->writeback_index
; /* Start from prev offset */
3804 index
= wbc
->range_start
>> PAGE_SHIFT
;
3805 end
= wbc
->range_end
>> PAGE_SHIFT
;
3808 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3809 tag
= PAGECACHE_TAG_TOWRITE
;
3811 tag
= PAGECACHE_TAG_DIRTY
;
3813 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3814 tag_pages_for_writeback(mapping
, index
, end
);
3815 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3816 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3817 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3821 for (i
= 0; i
< nr_pages
; i
++) {
3822 struct page
*page
= pvec
.pages
[i
];
3824 if (!PagePrivate(page
))
3827 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3832 spin_lock(&mapping
->private_lock
);
3833 if (!PagePrivate(page
)) {
3834 spin_unlock(&mapping
->private_lock
);
3838 eb
= (struct extent_buffer
*)page
->private;
3841 * Shouldn't happen and normally this would be a BUG_ON
3842 * but no sense in crashing the users box for something
3843 * we can survive anyway.
3846 spin_unlock(&mapping
->private_lock
);
3850 if (eb
== prev_eb
) {
3851 spin_unlock(&mapping
->private_lock
);
3855 ret
= atomic_inc_not_zero(&eb
->refs
);
3856 spin_unlock(&mapping
->private_lock
);
3861 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3863 free_extent_buffer(eb
);
3867 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3870 free_extent_buffer(eb
);
3873 free_extent_buffer(eb
);
3876 * the filesystem may choose to bump up nr_to_write.
3877 * We have to make sure to honor the new nr_to_write
3880 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3882 pagevec_release(&pvec
);
3885 if (!scanned
&& !done
) {
3887 * We hit the last page and there is more work to be done: wrap
3888 * back to the start of the file
3894 flush_write_bio(&epd
);
3899 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3900 * @mapping: address space structure to write
3901 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3902 * @writepage: function called for each page
3903 * @data: data passed to writepage function
3905 * If a page is already under I/O, write_cache_pages() skips it, even
3906 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3907 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3908 * and msync() need to guarantee that all the data which was dirty at the time
3909 * the call was made get new I/O started against them. If wbc->sync_mode is
3910 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3911 * existing IO to complete.
3913 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3914 struct address_space
*mapping
,
3915 struct writeback_control
*wbc
,
3916 writepage_t writepage
, void *data
,
3917 void (*flush_fn
)(void *))
3919 struct inode
*inode
= mapping
->host
;
3922 int nr_to_write_done
= 0;
3923 struct pagevec pvec
;
3926 pgoff_t end
; /* Inclusive */
3928 int range_whole
= 0;
3933 * We have to hold onto the inode so that ordered extents can do their
3934 * work when the IO finishes. The alternative to this is failing to add
3935 * an ordered extent if the igrab() fails there and that is a huge pain
3936 * to deal with, so instead just hold onto the inode throughout the
3937 * writepages operation. If it fails here we are freeing up the inode
3938 * anyway and we'd rather not waste our time writing out stuff that is
3939 * going to be truncated anyway.
3944 pagevec_init(&pvec
, 0);
3945 if (wbc
->range_cyclic
) {
3946 index
= mapping
->writeback_index
; /* Start from prev offset */
3949 index
= wbc
->range_start
>> PAGE_SHIFT
;
3950 end
= wbc
->range_end
>> PAGE_SHIFT
;
3951 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3955 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3956 tag
= PAGECACHE_TAG_TOWRITE
;
3958 tag
= PAGECACHE_TAG_DIRTY
;
3960 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3961 tag_pages_for_writeback(mapping
, index
, end
);
3963 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3964 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3965 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3969 for (i
= 0; i
< nr_pages
; i
++) {
3970 struct page
*page
= pvec
.pages
[i
];
3972 done_index
= page
->index
;
3974 * At this point we hold neither mapping->tree_lock nor
3975 * lock on the page itself: the page may be truncated or
3976 * invalidated (changing page->mapping to NULL), or even
3977 * swizzled back from swapper_space to tmpfs file
3980 if (!trylock_page(page
)) {
3985 if (unlikely(page
->mapping
!= mapping
)) {
3990 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3996 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3997 if (PageWriteback(page
))
3999 wait_on_page_writeback(page
);
4002 if (PageWriteback(page
) ||
4003 !clear_page_dirty_for_io(page
)) {
4008 ret
= (*writepage
)(page
, wbc
, data
);
4010 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4016 * done_index is set past this page,
4017 * so media errors will not choke
4018 * background writeout for the entire
4019 * file. This has consequences for
4020 * range_cyclic semantics (ie. it may
4021 * not be suitable for data integrity
4024 done_index
= page
->index
+ 1;
4030 * the filesystem may choose to bump up nr_to_write.
4031 * We have to make sure to honor the new nr_to_write
4034 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4036 pagevec_release(&pvec
);
4039 if (!scanned
&& !done
) {
4041 * We hit the last page and there is more work to be done: wrap
4042 * back to the start of the file
4049 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4050 mapping
->writeback_index
= done_index
;
4052 btrfs_add_delayed_iput(inode
);
4056 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4065 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
4066 BUG_ON(ret
< 0); /* -ENOMEM */
4071 static noinline
void flush_write_bio(void *data
)
4073 struct extent_page_data
*epd
= data
;
4074 flush_epd_write_bio(epd
);
4077 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4078 get_extent_t
*get_extent
,
4079 struct writeback_control
*wbc
)
4082 struct extent_page_data epd
= {
4085 .get_extent
= get_extent
,
4087 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4091 ret
= __extent_writepage(page
, wbc
, &epd
);
4093 flush_epd_write_bio(&epd
);
4097 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4098 u64 start
, u64 end
, get_extent_t
*get_extent
,
4102 struct address_space
*mapping
= inode
->i_mapping
;
4104 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4107 struct extent_page_data epd
= {
4110 .get_extent
= get_extent
,
4112 .sync_io
= mode
== WB_SYNC_ALL
,
4115 struct writeback_control wbc_writepages
= {
4117 .nr_to_write
= nr_pages
* 2,
4118 .range_start
= start
,
4119 .range_end
= end
+ 1,
4122 while (start
<= end
) {
4123 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4124 if (clear_page_dirty_for_io(page
))
4125 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4127 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4128 tree
->ops
->writepage_end_io_hook(page
, start
,
4129 start
+ PAGE_SIZE
- 1,
4137 flush_epd_write_bio(&epd
);
4141 int extent_writepages(struct extent_io_tree
*tree
,
4142 struct address_space
*mapping
,
4143 get_extent_t
*get_extent
,
4144 struct writeback_control
*wbc
)
4147 struct extent_page_data epd
= {
4150 .get_extent
= get_extent
,
4152 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4156 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
4157 __extent_writepage
, &epd
,
4159 flush_epd_write_bio(&epd
);
4163 int extent_readpages(struct extent_io_tree
*tree
,
4164 struct address_space
*mapping
,
4165 struct list_head
*pages
, unsigned nr_pages
,
4166 get_extent_t get_extent
)
4168 struct bio
*bio
= NULL
;
4170 unsigned long bio_flags
= 0;
4171 struct page
*pagepool
[16];
4173 struct extent_map
*em_cached
= NULL
;
4175 u64 prev_em_start
= (u64
)-1;
4177 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4178 page
= list_entry(pages
->prev
, struct page
, lru
);
4180 prefetchw(&page
->flags
);
4181 list_del(&page
->lru
);
4182 if (add_to_page_cache_lru(page
, mapping
,
4183 page
->index
, GFP_NOFS
)) {
4188 pagepool
[nr
++] = page
;
4189 if (nr
< ARRAY_SIZE(pagepool
))
4191 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4192 &bio
, 0, &bio_flags
, READ
, &prev_em_start
);
4196 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4197 &bio
, 0, &bio_flags
, READ
, &prev_em_start
);
4200 free_extent_map(em_cached
);
4202 BUG_ON(!list_empty(pages
));
4204 return submit_one_bio(READ
, bio
, 0, bio_flags
);
4209 * basic invalidatepage code, this waits on any locked or writeback
4210 * ranges corresponding to the page, and then deletes any extent state
4211 * records from the tree
4213 int extent_invalidatepage(struct extent_io_tree
*tree
,
4214 struct page
*page
, unsigned long offset
)
4216 struct extent_state
*cached_state
= NULL
;
4217 u64 start
= page_offset(page
);
4218 u64 end
= start
+ PAGE_SIZE
- 1;
4219 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4221 start
+= ALIGN(offset
, blocksize
);
4225 lock_extent_bits(tree
, start
, end
, &cached_state
);
4226 wait_on_page_writeback(page
);
4227 clear_extent_bit(tree
, start
, end
,
4228 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4229 EXTENT_DO_ACCOUNTING
,
4230 1, 1, &cached_state
, GFP_NOFS
);
4235 * a helper for releasepage, this tests for areas of the page that
4236 * are locked or under IO and drops the related state bits if it is safe
4239 static int try_release_extent_state(struct extent_map_tree
*map
,
4240 struct extent_io_tree
*tree
,
4241 struct page
*page
, gfp_t mask
)
4243 u64 start
= page_offset(page
);
4244 u64 end
= start
+ PAGE_SIZE
- 1;
4247 if (test_range_bit(tree
, start
, end
,
4248 EXTENT_IOBITS
, 0, NULL
))
4251 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4254 * at this point we can safely clear everything except the
4255 * locked bit and the nodatasum bit
4257 ret
= clear_extent_bit(tree
, start
, end
,
4258 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4261 /* if clear_extent_bit failed for enomem reasons,
4262 * we can't allow the release to continue.
4273 * a helper for releasepage. As long as there are no locked extents
4274 * in the range corresponding to the page, both state records and extent
4275 * map records are removed
4277 int try_release_extent_mapping(struct extent_map_tree
*map
,
4278 struct extent_io_tree
*tree
, struct page
*page
,
4281 struct extent_map
*em
;
4282 u64 start
= page_offset(page
);
4283 u64 end
= start
+ PAGE_SIZE
- 1;
4285 if (gfpflags_allow_blocking(mask
) &&
4286 page
->mapping
->host
->i_size
> SZ_16M
) {
4288 while (start
<= end
) {
4289 len
= end
- start
+ 1;
4290 write_lock(&map
->lock
);
4291 em
= lookup_extent_mapping(map
, start
, len
);
4293 write_unlock(&map
->lock
);
4296 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4297 em
->start
!= start
) {
4298 write_unlock(&map
->lock
);
4299 free_extent_map(em
);
4302 if (!test_range_bit(tree
, em
->start
,
4303 extent_map_end(em
) - 1,
4304 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4306 remove_extent_mapping(map
, em
);
4307 /* once for the rb tree */
4308 free_extent_map(em
);
4310 start
= extent_map_end(em
);
4311 write_unlock(&map
->lock
);
4314 free_extent_map(em
);
4317 return try_release_extent_state(map
, tree
, page
, mask
);
4321 * helper function for fiemap, which doesn't want to see any holes.
4322 * This maps until we find something past 'last'
4324 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4327 get_extent_t
*get_extent
)
4329 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4330 struct extent_map
*em
;
4337 len
= last
- offset
;
4340 len
= ALIGN(len
, sectorsize
);
4341 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4342 if (IS_ERR_OR_NULL(em
))
4345 /* if this isn't a hole return it */
4346 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4347 em
->block_start
!= EXTENT_MAP_HOLE
) {
4351 /* this is a hole, advance to the next extent */
4352 offset
= extent_map_end(em
);
4353 free_extent_map(em
);
4360 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4361 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4365 u64 max
= start
+ len
;
4369 u64 last_for_get_extent
= 0;
4371 u64 isize
= i_size_read(inode
);
4372 struct btrfs_key found_key
;
4373 struct extent_map
*em
= NULL
;
4374 struct extent_state
*cached_state
= NULL
;
4375 struct btrfs_path
*path
;
4376 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4385 path
= btrfs_alloc_path();
4388 path
->leave_spinning
= 1;
4390 start
= round_down(start
, BTRFS_I(inode
)->root
->sectorsize
);
4391 len
= round_up(max
, BTRFS_I(inode
)->root
->sectorsize
) - start
;
4394 * lookup the last file extent. We're not using i_size here
4395 * because there might be preallocation past i_size
4397 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
), -1,
4400 btrfs_free_path(path
);
4409 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4410 found_type
= found_key
.type
;
4412 /* No extents, but there might be delalloc bits */
4413 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4414 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4415 /* have to trust i_size as the end */
4417 last_for_get_extent
= isize
;
4420 * remember the start of the last extent. There are a
4421 * bunch of different factors that go into the length of the
4422 * extent, so its much less complex to remember where it started
4424 last
= found_key
.offset
;
4425 last_for_get_extent
= last
+ 1;
4427 btrfs_release_path(path
);
4430 * we might have some extents allocated but more delalloc past those
4431 * extents. so, we trust isize unless the start of the last extent is
4436 last_for_get_extent
= isize
;
4439 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4442 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4452 u64 offset_in_extent
= 0;
4454 /* break if the extent we found is outside the range */
4455 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4459 * get_extent may return an extent that starts before our
4460 * requested range. We have to make sure the ranges
4461 * we return to fiemap always move forward and don't
4462 * overlap, so adjust the offsets here
4464 em_start
= max(em
->start
, off
);
4467 * record the offset from the start of the extent
4468 * for adjusting the disk offset below. Only do this if the
4469 * extent isn't compressed since our in ram offset may be past
4470 * what we have actually allocated on disk.
4472 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4473 offset_in_extent
= em_start
- em
->start
;
4474 em_end
= extent_map_end(em
);
4475 em_len
= em_end
- em_start
;
4480 * bump off for our next call to get_extent
4482 off
= extent_map_end(em
);
4486 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4488 flags
|= FIEMAP_EXTENT_LAST
;
4489 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4490 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4491 FIEMAP_EXTENT_NOT_ALIGNED
);
4492 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4493 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4494 FIEMAP_EXTENT_UNKNOWN
);
4495 } else if (fieinfo
->fi_extents_max
) {
4496 u64 bytenr
= em
->block_start
-
4497 (em
->start
- em
->orig_start
);
4499 disko
= em
->block_start
+ offset_in_extent
;
4502 * As btrfs supports shared space, this information
4503 * can be exported to userspace tools via
4504 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4505 * then we're just getting a count and we can skip the
4508 ret
= btrfs_check_shared(NULL
, root
->fs_info
,
4510 btrfs_ino(inode
), bytenr
);
4514 flags
|= FIEMAP_EXTENT_SHARED
;
4517 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4518 flags
|= FIEMAP_EXTENT_ENCODED
;
4519 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4520 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4522 free_extent_map(em
);
4524 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4525 (last
== (u64
)-1 && isize
<= em_end
)) {
4526 flags
|= FIEMAP_EXTENT_LAST
;
4530 /* now scan forward to see if this is really the last extent. */
4531 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4538 flags
|= FIEMAP_EXTENT_LAST
;
4541 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4550 free_extent_map(em
);
4552 btrfs_free_path(path
);
4553 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4554 &cached_state
, GFP_NOFS
);
4558 static void __free_extent_buffer(struct extent_buffer
*eb
)
4560 btrfs_leak_debug_del(&eb
->leak_list
);
4561 kmem_cache_free(extent_buffer_cache
, eb
);
4564 int extent_buffer_under_io(struct extent_buffer
*eb
)
4566 return (atomic_read(&eb
->io_pages
) ||
4567 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4568 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4572 * Helper for releasing extent buffer page.
4574 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4576 unsigned long index
;
4578 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4580 BUG_ON(extent_buffer_under_io(eb
));
4582 index
= num_extent_pages(eb
->start
, eb
->len
);
4588 page
= eb
->pages
[index
];
4592 spin_lock(&page
->mapping
->private_lock
);
4594 * We do this since we'll remove the pages after we've
4595 * removed the eb from the radix tree, so we could race
4596 * and have this page now attached to the new eb. So
4597 * only clear page_private if it's still connected to
4600 if (PagePrivate(page
) &&
4601 page
->private == (unsigned long)eb
) {
4602 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4603 BUG_ON(PageDirty(page
));
4604 BUG_ON(PageWriteback(page
));
4606 * We need to make sure we haven't be attached
4609 ClearPagePrivate(page
);
4610 set_page_private(page
, 0);
4611 /* One for the page private */
4616 spin_unlock(&page
->mapping
->private_lock
);
4618 /* One for when we allocated the page */
4620 } while (index
!= 0);
4624 * Helper for releasing the extent buffer.
4626 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4628 btrfs_release_extent_buffer_page(eb
);
4629 __free_extent_buffer(eb
);
4632 static struct extent_buffer
*
4633 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4636 struct extent_buffer
*eb
= NULL
;
4638 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4641 eb
->fs_info
= fs_info
;
4643 rwlock_init(&eb
->lock
);
4644 atomic_set(&eb
->write_locks
, 0);
4645 atomic_set(&eb
->read_locks
, 0);
4646 atomic_set(&eb
->blocking_readers
, 0);
4647 atomic_set(&eb
->blocking_writers
, 0);
4648 atomic_set(&eb
->spinning_readers
, 0);
4649 atomic_set(&eb
->spinning_writers
, 0);
4650 eb
->lock_nested
= 0;
4651 init_waitqueue_head(&eb
->write_lock_wq
);
4652 init_waitqueue_head(&eb
->read_lock_wq
);
4654 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4656 spin_lock_init(&eb
->refs_lock
);
4657 atomic_set(&eb
->refs
, 1);
4658 atomic_set(&eb
->io_pages
, 0);
4661 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4663 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4664 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4665 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4670 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4674 struct extent_buffer
*new;
4675 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4677 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4681 for (i
= 0; i
< num_pages
; i
++) {
4682 p
= alloc_page(GFP_NOFS
);
4684 btrfs_release_extent_buffer(new);
4687 attach_extent_buffer_page(new, p
);
4688 WARN_ON(PageDirty(p
));
4693 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4694 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4695 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4700 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4701 u64 start
, unsigned long len
)
4703 struct extent_buffer
*eb
;
4704 unsigned long num_pages
;
4707 num_pages
= num_extent_pages(start
, len
);
4709 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4713 for (i
= 0; i
< num_pages
; i
++) {
4714 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4718 set_extent_buffer_uptodate(eb
);
4719 btrfs_set_header_nritems(eb
, 0);
4720 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4725 __free_page(eb
->pages
[i
- 1]);
4726 __free_extent_buffer(eb
);
4730 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4731 u64 start
, u32 nodesize
)
4737 * Called only from tests that don't always have a fs_info
4742 len
= fs_info
->tree_root
->nodesize
;
4745 return __alloc_dummy_extent_buffer(fs_info
, start
, len
);
4748 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4751 /* the ref bit is tricky. We have to make sure it is set
4752 * if we have the buffer dirty. Otherwise the
4753 * code to free a buffer can end up dropping a dirty
4756 * Once the ref bit is set, it won't go away while the
4757 * buffer is dirty or in writeback, and it also won't
4758 * go away while we have the reference count on the
4761 * We can't just set the ref bit without bumping the
4762 * ref on the eb because free_extent_buffer might
4763 * see the ref bit and try to clear it. If this happens
4764 * free_extent_buffer might end up dropping our original
4765 * ref by mistake and freeing the page before we are able
4766 * to add one more ref.
4768 * So bump the ref count first, then set the bit. If someone
4769 * beat us to it, drop the ref we added.
4771 refs
= atomic_read(&eb
->refs
);
4772 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4775 spin_lock(&eb
->refs_lock
);
4776 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4777 atomic_inc(&eb
->refs
);
4778 spin_unlock(&eb
->refs_lock
);
4781 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4782 struct page
*accessed
)
4784 unsigned long num_pages
, i
;
4786 check_buffer_tree_ref(eb
);
4788 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4789 for (i
= 0; i
< num_pages
; i
++) {
4790 struct page
*p
= eb
->pages
[i
];
4793 mark_page_accessed(p
);
4797 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4800 struct extent_buffer
*eb
;
4803 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4804 start
>> PAGE_SHIFT
);
4805 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4808 * Lock our eb's refs_lock to avoid races with
4809 * free_extent_buffer. When we get our eb it might be flagged
4810 * with EXTENT_BUFFER_STALE and another task running
4811 * free_extent_buffer might have seen that flag set,
4812 * eb->refs == 2, that the buffer isn't under IO (dirty and
4813 * writeback flags not set) and it's still in the tree (flag
4814 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4815 * of decrementing the extent buffer's reference count twice.
4816 * So here we could race and increment the eb's reference count,
4817 * clear its stale flag, mark it as dirty and drop our reference
4818 * before the other task finishes executing free_extent_buffer,
4819 * which would later result in an attempt to free an extent
4820 * buffer that is dirty.
4822 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4823 spin_lock(&eb
->refs_lock
);
4824 spin_unlock(&eb
->refs_lock
);
4826 mark_extent_buffer_accessed(eb
, NULL
);
4834 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4835 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4836 u64 start
, u32 nodesize
)
4838 struct extent_buffer
*eb
, *exists
= NULL
;
4841 eb
= find_extent_buffer(fs_info
, start
);
4844 eb
= alloc_dummy_extent_buffer(fs_info
, start
, nodesize
);
4847 eb
->fs_info
= fs_info
;
4849 ret
= radix_tree_preload(GFP_NOFS
);
4852 spin_lock(&fs_info
->buffer_lock
);
4853 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4854 start
>> PAGE_SHIFT
, eb
);
4855 spin_unlock(&fs_info
->buffer_lock
);
4856 radix_tree_preload_end();
4857 if (ret
== -EEXIST
) {
4858 exists
= find_extent_buffer(fs_info
, start
);
4864 check_buffer_tree_ref(eb
);
4865 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4868 * We will free dummy extent buffer's if they come into
4869 * free_extent_buffer with a ref count of 2, but if we are using this we
4870 * want the buffers to stay in memory until we're done with them, so
4871 * bump the ref count again.
4873 atomic_inc(&eb
->refs
);
4876 btrfs_release_extent_buffer(eb
);
4881 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4884 unsigned long len
= fs_info
->tree_root
->nodesize
;
4885 unsigned long num_pages
= num_extent_pages(start
, len
);
4887 unsigned long index
= start
>> PAGE_SHIFT
;
4888 struct extent_buffer
*eb
;
4889 struct extent_buffer
*exists
= NULL
;
4891 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4895 if (!IS_ALIGNED(start
, fs_info
->tree_root
->sectorsize
)) {
4896 btrfs_err(fs_info
, "bad tree block start %llu", start
);
4897 return ERR_PTR(-EINVAL
);
4900 eb
= find_extent_buffer(fs_info
, start
);
4904 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4906 return ERR_PTR(-ENOMEM
);
4908 for (i
= 0; i
< num_pages
; i
++, index
++) {
4909 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4911 exists
= ERR_PTR(-ENOMEM
);
4915 spin_lock(&mapping
->private_lock
);
4916 if (PagePrivate(p
)) {
4918 * We could have already allocated an eb for this page
4919 * and attached one so lets see if we can get a ref on
4920 * the existing eb, and if we can we know it's good and
4921 * we can just return that one, else we know we can just
4922 * overwrite page->private.
4924 exists
= (struct extent_buffer
*)p
->private;
4925 if (atomic_inc_not_zero(&exists
->refs
)) {
4926 spin_unlock(&mapping
->private_lock
);
4929 mark_extent_buffer_accessed(exists
, p
);
4935 * Do this so attach doesn't complain and we need to
4936 * drop the ref the old guy had.
4938 ClearPagePrivate(p
);
4939 WARN_ON(PageDirty(p
));
4942 attach_extent_buffer_page(eb
, p
);
4943 spin_unlock(&mapping
->private_lock
);
4944 WARN_ON(PageDirty(p
));
4946 if (!PageUptodate(p
))
4950 * see below about how we avoid a nasty race with release page
4951 * and why we unlock later
4955 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4957 ret
= radix_tree_preload(GFP_NOFS
);
4959 exists
= ERR_PTR(ret
);
4963 spin_lock(&fs_info
->buffer_lock
);
4964 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4965 start
>> PAGE_SHIFT
, eb
);
4966 spin_unlock(&fs_info
->buffer_lock
);
4967 radix_tree_preload_end();
4968 if (ret
== -EEXIST
) {
4969 exists
= find_extent_buffer(fs_info
, start
);
4975 /* add one reference for the tree */
4976 check_buffer_tree_ref(eb
);
4977 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4980 * there is a race where release page may have
4981 * tried to find this extent buffer in the radix
4982 * but failed. It will tell the VM it is safe to
4983 * reclaim the, and it will clear the page private bit.
4984 * We must make sure to set the page private bit properly
4985 * after the extent buffer is in the radix tree so
4986 * it doesn't get lost
4988 SetPageChecked(eb
->pages
[0]);
4989 for (i
= 1; i
< num_pages
; i
++) {
4991 ClearPageChecked(p
);
4994 unlock_page(eb
->pages
[0]);
4998 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4999 for (i
= 0; i
< num_pages
; i
++) {
5001 unlock_page(eb
->pages
[i
]);
5004 btrfs_release_extent_buffer(eb
);
5008 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
5010 struct extent_buffer
*eb
=
5011 container_of(head
, struct extent_buffer
, rcu_head
);
5013 __free_extent_buffer(eb
);
5016 /* Expects to have eb->eb_lock already held */
5017 static int release_extent_buffer(struct extent_buffer
*eb
)
5019 WARN_ON(atomic_read(&eb
->refs
) == 0);
5020 if (atomic_dec_and_test(&eb
->refs
)) {
5021 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5022 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5024 spin_unlock(&eb
->refs_lock
);
5026 spin_lock(&fs_info
->buffer_lock
);
5027 radix_tree_delete(&fs_info
->buffer_radix
,
5028 eb
->start
>> PAGE_SHIFT
);
5029 spin_unlock(&fs_info
->buffer_lock
);
5031 spin_unlock(&eb
->refs_lock
);
5034 /* Should be safe to release our pages at this point */
5035 btrfs_release_extent_buffer_page(eb
);
5036 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5037 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5038 __free_extent_buffer(eb
);
5042 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5045 spin_unlock(&eb
->refs_lock
);
5050 void free_extent_buffer(struct extent_buffer
*eb
)
5058 refs
= atomic_read(&eb
->refs
);
5061 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5066 spin_lock(&eb
->refs_lock
);
5067 if (atomic_read(&eb
->refs
) == 2 &&
5068 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5069 atomic_dec(&eb
->refs
);
5071 if (atomic_read(&eb
->refs
) == 2 &&
5072 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5073 !extent_buffer_under_io(eb
) &&
5074 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5075 atomic_dec(&eb
->refs
);
5078 * I know this is terrible, but it's temporary until we stop tracking
5079 * the uptodate bits and such for the extent buffers.
5081 release_extent_buffer(eb
);
5084 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5089 spin_lock(&eb
->refs_lock
);
5090 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5092 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5093 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5094 atomic_dec(&eb
->refs
);
5095 release_extent_buffer(eb
);
5098 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5101 unsigned long num_pages
;
5104 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5106 for (i
= 0; i
< num_pages
; i
++) {
5107 page
= eb
->pages
[i
];
5108 if (!PageDirty(page
))
5112 WARN_ON(!PagePrivate(page
));
5114 clear_page_dirty_for_io(page
);
5115 spin_lock_irq(&page
->mapping
->tree_lock
);
5116 if (!PageDirty(page
)) {
5117 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5119 PAGECACHE_TAG_DIRTY
);
5121 spin_unlock_irq(&page
->mapping
->tree_lock
);
5122 ClearPageError(page
);
5125 WARN_ON(atomic_read(&eb
->refs
) == 0);
5128 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5131 unsigned long num_pages
;
5134 check_buffer_tree_ref(eb
);
5136 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5138 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5139 WARN_ON(atomic_read(&eb
->refs
) == 0);
5140 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5142 for (i
= 0; i
< num_pages
; i
++)
5143 set_page_dirty(eb
->pages
[i
]);
5147 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5151 unsigned long num_pages
;
5153 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5154 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5155 for (i
= 0; i
< num_pages
; i
++) {
5156 page
= eb
->pages
[i
];
5158 ClearPageUptodate(page
);
5162 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5166 unsigned long num_pages
;
5168 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5169 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5170 for (i
= 0; i
< num_pages
; i
++) {
5171 page
= eb
->pages
[i
];
5172 SetPageUptodate(page
);
5176 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5178 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5181 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5182 struct extent_buffer
*eb
, u64 start
, int wait
,
5183 get_extent_t
*get_extent
, int mirror_num
)
5186 unsigned long start_i
;
5190 int locked_pages
= 0;
5191 int all_uptodate
= 1;
5192 unsigned long num_pages
;
5193 unsigned long num_reads
= 0;
5194 struct bio
*bio
= NULL
;
5195 unsigned long bio_flags
= 0;
5197 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5201 WARN_ON(start
< eb
->start
);
5202 start_i
= (start
>> PAGE_SHIFT
) -
5203 (eb
->start
>> PAGE_SHIFT
);
5208 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5209 for (i
= start_i
; i
< num_pages
; i
++) {
5210 page
= eb
->pages
[i
];
5211 if (wait
== WAIT_NONE
) {
5212 if (!trylock_page(page
))
5218 if (!PageUptodate(page
)) {
5225 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5229 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5230 eb
->read_mirror
= 0;
5231 atomic_set(&eb
->io_pages
, num_reads
);
5232 for (i
= start_i
; i
< num_pages
; i
++) {
5233 page
= eb
->pages
[i
];
5234 if (!PageUptodate(page
)) {
5235 ClearPageError(page
);
5236 err
= __extent_read_full_page(tree
, page
,
5238 mirror_num
, &bio_flags
,
5248 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
5254 if (ret
|| wait
!= WAIT_COMPLETE
)
5257 for (i
= start_i
; i
< num_pages
; i
++) {
5258 page
= eb
->pages
[i
];
5259 wait_on_page_locked(page
);
5260 if (!PageUptodate(page
))
5268 while (locked_pages
> 0) {
5269 page
= eb
->pages
[i
];
5277 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5278 unsigned long start
,
5285 char *dst
= (char *)dstv
;
5286 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5287 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5289 WARN_ON(start
> eb
->len
);
5290 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5292 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5295 page
= eb
->pages
[i
];
5297 cur
= min(len
, (PAGE_SIZE
- offset
));
5298 kaddr
= page_address(page
);
5299 memcpy(dst
, kaddr
+ offset
, cur
);
5308 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5309 unsigned long start
,
5316 char __user
*dst
= (char __user
*)dstv
;
5317 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5318 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5321 WARN_ON(start
> eb
->len
);
5322 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5324 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5327 page
= eb
->pages
[i
];
5329 cur
= min(len
, (PAGE_SIZE
- offset
));
5330 kaddr
= page_address(page
);
5331 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5346 * return 0 if the item is found within a page.
5347 * return 1 if the item spans two pages.
5348 * return -EINVAL otherwise.
5350 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5351 unsigned long min_len
, char **map
,
5352 unsigned long *map_start
,
5353 unsigned long *map_len
)
5355 size_t offset
= start
& (PAGE_SIZE
- 1);
5358 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5359 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5360 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5367 offset
= start_offset
;
5371 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5374 if (start
+ min_len
> eb
->len
) {
5375 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
5377 eb
->start
, eb
->len
, start
, min_len
);
5382 kaddr
= page_address(p
);
5383 *map
= kaddr
+ offset
;
5384 *map_len
= PAGE_SIZE
- offset
;
5388 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5389 unsigned long start
,
5396 char *ptr
= (char *)ptrv
;
5397 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5398 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5401 WARN_ON(start
> eb
->len
);
5402 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5404 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5407 page
= eb
->pages
[i
];
5409 cur
= min(len
, (PAGE_SIZE
- offset
));
5411 kaddr
= page_address(page
);
5412 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5424 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5425 unsigned long start
, unsigned long len
)
5431 char *src
= (char *)srcv
;
5432 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5433 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5435 WARN_ON(start
> eb
->len
);
5436 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5438 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5441 page
= eb
->pages
[i
];
5442 WARN_ON(!PageUptodate(page
));
5444 cur
= min(len
, PAGE_SIZE
- offset
);
5445 kaddr
= page_address(page
);
5446 memcpy(kaddr
+ offset
, src
, cur
);
5455 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5456 unsigned long start
, unsigned long len
)
5462 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5463 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5465 WARN_ON(start
> eb
->len
);
5466 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5468 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5471 page
= eb
->pages
[i
];
5472 WARN_ON(!PageUptodate(page
));
5474 cur
= min(len
, PAGE_SIZE
- offset
);
5475 kaddr
= page_address(page
);
5476 memset(kaddr
+ offset
, c
, cur
);
5484 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5485 unsigned long dst_offset
, unsigned long src_offset
,
5488 u64 dst_len
= dst
->len
;
5493 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5494 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5496 WARN_ON(src
->len
!= dst_len
);
5498 offset
= (start_offset
+ dst_offset
) &
5502 page
= dst
->pages
[i
];
5503 WARN_ON(!PageUptodate(page
));
5505 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5507 kaddr
= page_address(page
);
5508 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5518 * The extent buffer bitmap operations are done with byte granularity because
5519 * bitmap items are not guaranteed to be aligned to a word and therefore a
5520 * single word in a bitmap may straddle two pages in the extent buffer.
5522 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
5523 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
5524 #define BITMAP_FIRST_BYTE_MASK(start) \
5525 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
5526 #define BITMAP_LAST_BYTE_MASK(nbits) \
5527 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
5530 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5532 * @eb: the extent buffer
5533 * @start: offset of the bitmap item in the extent buffer
5535 * @page_index: return index of the page in the extent buffer that contains the
5537 * @page_offset: return offset into the page given by page_index
5539 * This helper hides the ugliness of finding the byte in an extent buffer which
5540 * contains a given bit.
5542 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5543 unsigned long start
, unsigned long nr
,
5544 unsigned long *page_index
,
5545 size_t *page_offset
)
5547 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5548 size_t byte_offset
= BIT_BYTE(nr
);
5552 * The byte we want is the offset of the extent buffer + the offset of
5553 * the bitmap item in the extent buffer + the offset of the byte in the
5556 offset
= start_offset
+ start
+ byte_offset
;
5558 *page_index
= offset
>> PAGE_SHIFT
;
5559 *page_offset
= offset
& (PAGE_SIZE
- 1);
5563 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5564 * @eb: the extent buffer
5565 * @start: offset of the bitmap item in the extent buffer
5566 * @nr: bit number to test
5568 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5576 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5577 page
= eb
->pages
[i
];
5578 WARN_ON(!PageUptodate(page
));
5579 kaddr
= page_address(page
);
5580 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5584 * extent_buffer_bitmap_set - set an area of a bitmap
5585 * @eb: the extent buffer
5586 * @start: offset of the bitmap item in the extent buffer
5587 * @pos: bit number of the first bit
5588 * @len: number of bits to set
5590 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5591 unsigned long pos
, unsigned long len
)
5597 const unsigned int size
= pos
+ len
;
5598 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5599 unsigned int mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5601 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5602 page
= eb
->pages
[i
];
5603 WARN_ON(!PageUptodate(page
));
5604 kaddr
= page_address(page
);
5606 while (len
>= bits_to_set
) {
5607 kaddr
[offset
] |= mask_to_set
;
5609 bits_to_set
= BITS_PER_BYTE
;
5611 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5613 page
= eb
->pages
[++i
];
5614 WARN_ON(!PageUptodate(page
));
5615 kaddr
= page_address(page
);
5619 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5620 kaddr
[offset
] |= mask_to_set
;
5626 * extent_buffer_bitmap_clear - clear an area of a bitmap
5627 * @eb: the extent buffer
5628 * @start: offset of the bitmap item in the extent buffer
5629 * @pos: bit number of the first bit
5630 * @len: number of bits to clear
5632 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5633 unsigned long pos
, unsigned long len
)
5639 const unsigned int size
= pos
+ len
;
5640 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5641 unsigned int mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5643 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5644 page
= eb
->pages
[i
];
5645 WARN_ON(!PageUptodate(page
));
5646 kaddr
= page_address(page
);
5648 while (len
>= bits_to_clear
) {
5649 kaddr
[offset
] &= ~mask_to_clear
;
5650 len
-= bits_to_clear
;
5651 bits_to_clear
= BITS_PER_BYTE
;
5652 mask_to_clear
= ~0U;
5653 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5655 page
= eb
->pages
[++i
];
5656 WARN_ON(!PageUptodate(page
));
5657 kaddr
= page_address(page
);
5661 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5662 kaddr
[offset
] &= ~mask_to_clear
;
5666 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5668 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5669 return distance
< len
;
5672 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5673 unsigned long dst_off
, unsigned long src_off
,
5676 char *dst_kaddr
= page_address(dst_page
);
5678 int must_memmove
= 0;
5680 if (dst_page
!= src_page
) {
5681 src_kaddr
= page_address(src_page
);
5683 src_kaddr
= dst_kaddr
;
5684 if (areas_overlap(src_off
, dst_off
, len
))
5689 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5691 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5694 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5695 unsigned long src_offset
, unsigned long len
)
5698 size_t dst_off_in_page
;
5699 size_t src_off_in_page
;
5700 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5701 unsigned long dst_i
;
5702 unsigned long src_i
;
5704 if (src_offset
+ len
> dst
->len
) {
5705 btrfs_err(dst
->fs_info
,
5706 "memmove bogus src_offset %lu move "
5707 "len %lu dst len %lu", src_offset
, len
, dst
->len
);
5710 if (dst_offset
+ len
> dst
->len
) {
5711 btrfs_err(dst
->fs_info
,
5712 "memmove bogus dst_offset %lu move "
5713 "len %lu dst len %lu", dst_offset
, len
, dst
->len
);
5718 dst_off_in_page
= (start_offset
+ dst_offset
) &
5720 src_off_in_page
= (start_offset
+ src_offset
) &
5723 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5724 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5726 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5728 cur
= min_t(unsigned long, cur
,
5729 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5731 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5732 dst_off_in_page
, src_off_in_page
, cur
);
5740 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5741 unsigned long src_offset
, unsigned long len
)
5744 size_t dst_off_in_page
;
5745 size_t src_off_in_page
;
5746 unsigned long dst_end
= dst_offset
+ len
- 1;
5747 unsigned long src_end
= src_offset
+ len
- 1;
5748 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5749 unsigned long dst_i
;
5750 unsigned long src_i
;
5752 if (src_offset
+ len
> dst
->len
) {
5753 btrfs_err(dst
->fs_info
, "memmove bogus src_offset %lu move "
5754 "len %lu len %lu", src_offset
, len
, dst
->len
);
5757 if (dst_offset
+ len
> dst
->len
) {
5758 btrfs_err(dst
->fs_info
, "memmove bogus dst_offset %lu move "
5759 "len %lu len %lu", dst_offset
, len
, dst
->len
);
5762 if (dst_offset
< src_offset
) {
5763 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5767 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5768 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5770 dst_off_in_page
= (start_offset
+ dst_end
) &
5772 src_off_in_page
= (start_offset
+ src_end
) &
5775 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5776 cur
= min(cur
, dst_off_in_page
+ 1);
5777 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5778 dst_off_in_page
- cur
+ 1,
5779 src_off_in_page
- cur
+ 1, cur
);
5787 int try_release_extent_buffer(struct page
*page
)
5789 struct extent_buffer
*eb
;
5792 * We need to make sure nobody is attaching this page to an eb right
5795 spin_lock(&page
->mapping
->private_lock
);
5796 if (!PagePrivate(page
)) {
5797 spin_unlock(&page
->mapping
->private_lock
);
5801 eb
= (struct extent_buffer
*)page
->private;
5805 * This is a little awful but should be ok, we need to make sure that
5806 * the eb doesn't disappear out from under us while we're looking at
5809 spin_lock(&eb
->refs_lock
);
5810 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5811 spin_unlock(&eb
->refs_lock
);
5812 spin_unlock(&page
->mapping
->private_lock
);
5815 spin_unlock(&page
->mapping
->private_lock
);
5818 * If tree ref isn't set then we know the ref on this eb is a real ref,
5819 * so just return, this page will likely be freed soon anyway.
5821 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5822 spin_unlock(&eb
->refs_lock
);
5826 return release_extent_buffer(eb
);