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
->bi_rw
= WRITE_SYNC
;
2053 bio_add_page(bio
, page
, length
, pg_offset
);
2055 if (btrfsic_submit_bio_wait(bio
)) {
2056 /* try to remap that extent elsewhere? */
2057 btrfs_bio_counter_dec(fs_info
);
2059 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2063 btrfs_info_rl_in_rcu(fs_info
,
2064 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2065 btrfs_ino(inode
), start
,
2066 rcu_str_deref(dev
->name
), sector
);
2067 btrfs_bio_counter_dec(fs_info
);
2072 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2075 u64 start
= eb
->start
;
2076 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2079 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2082 for (i
= 0; i
< num_pages
; i
++) {
2083 struct page
*p
= eb
->pages
[i
];
2085 ret
= repair_io_failure(root
->fs_info
->btree_inode
, start
,
2086 PAGE_SIZE
, start
, p
,
2087 start
- page_offset(p
), mirror_num
);
2097 * each time an IO finishes, we do a fast check in the IO failure tree
2098 * to see if we need to process or clean up an io_failure_record
2100 int clean_io_failure(struct inode
*inode
, u64 start
, struct page
*page
,
2101 unsigned int pg_offset
)
2104 struct io_failure_record
*failrec
;
2105 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2106 struct extent_state
*state
;
2111 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2112 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2116 ret
= get_state_failrec(&BTRFS_I(inode
)->io_failure_tree
, start
,
2121 BUG_ON(!failrec
->this_mirror
);
2123 if (failrec
->in_validation
) {
2124 /* there was no real error, just free the record */
2125 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2129 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2132 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2133 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2136 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2138 if (state
&& state
->start
<= failrec
->start
&&
2139 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2140 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2142 if (num_copies
> 1) {
2143 repair_io_failure(inode
, start
, failrec
->len
,
2144 failrec
->logical
, page
,
2145 pg_offset
, failrec
->failed_mirror
);
2150 free_io_failure(inode
, failrec
);
2156 * Can be called when
2157 * - hold extent lock
2158 * - under ordered extent
2159 * - the inode is freeing
2161 void btrfs_free_io_failure_record(struct inode
*inode
, u64 start
, u64 end
)
2163 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2164 struct io_failure_record
*failrec
;
2165 struct extent_state
*state
, *next
;
2167 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2170 spin_lock(&failure_tree
->lock
);
2171 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2173 if (state
->start
> end
)
2176 ASSERT(state
->end
<= end
);
2178 next
= next_state(state
);
2180 failrec
= state
->failrec
;
2181 free_extent_state(state
);
2186 spin_unlock(&failure_tree
->lock
);
2189 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2190 struct io_failure_record
**failrec_ret
)
2192 struct io_failure_record
*failrec
;
2193 struct extent_map
*em
;
2194 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2195 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2196 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2200 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2202 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2206 failrec
->start
= start
;
2207 failrec
->len
= end
- start
+ 1;
2208 failrec
->this_mirror
= 0;
2209 failrec
->bio_flags
= 0;
2210 failrec
->in_validation
= 0;
2212 read_lock(&em_tree
->lock
);
2213 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2215 read_unlock(&em_tree
->lock
);
2220 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2221 free_extent_map(em
);
2224 read_unlock(&em_tree
->lock
);
2230 logical
= start
- em
->start
;
2231 logical
= em
->block_start
+ logical
;
2232 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2233 logical
= em
->block_start
;
2234 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2235 extent_set_compress_type(&failrec
->bio_flags
,
2239 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2240 logical
, start
, failrec
->len
);
2242 failrec
->logical
= logical
;
2243 free_extent_map(em
);
2245 /* set the bits in the private failure tree */
2246 ret
= set_extent_bits(failure_tree
, start
, end
,
2247 EXTENT_LOCKED
| EXTENT_DIRTY
);
2249 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2250 /* set the bits in the inode's tree */
2252 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2258 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2259 failrec
->logical
, failrec
->start
, failrec
->len
,
2260 failrec
->in_validation
);
2262 * when data can be on disk more than twice, add to failrec here
2263 * (e.g. with a list for failed_mirror) to make
2264 * clean_io_failure() clean all those errors at once.
2268 *failrec_ret
= failrec
;
2273 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2274 struct io_failure_record
*failrec
, int failed_mirror
)
2278 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2279 failrec
->logical
, failrec
->len
);
2280 if (num_copies
== 1) {
2282 * we only have a single copy of the data, so don't bother with
2283 * all the retry and error correction code that follows. no
2284 * matter what the error is, it is very likely to persist.
2286 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2287 num_copies
, failrec
->this_mirror
, failed_mirror
);
2292 * there are two premises:
2293 * a) deliver good data to the caller
2294 * b) correct the bad sectors on disk
2296 if (failed_bio
->bi_vcnt
> 1) {
2298 * to fulfill b), we need to know the exact failing sectors, as
2299 * we don't want to rewrite any more than the failed ones. thus,
2300 * we need separate read requests for the failed bio
2302 * if the following BUG_ON triggers, our validation request got
2303 * merged. we need separate requests for our algorithm to work.
2305 BUG_ON(failrec
->in_validation
);
2306 failrec
->in_validation
= 1;
2307 failrec
->this_mirror
= failed_mirror
;
2310 * we're ready to fulfill a) and b) alongside. get a good copy
2311 * of the failed sector and if we succeed, we have setup
2312 * everything for repair_io_failure to do the rest for us.
2314 if (failrec
->in_validation
) {
2315 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2316 failrec
->in_validation
= 0;
2317 failrec
->this_mirror
= 0;
2319 failrec
->failed_mirror
= failed_mirror
;
2320 failrec
->this_mirror
++;
2321 if (failrec
->this_mirror
== failed_mirror
)
2322 failrec
->this_mirror
++;
2325 if (failrec
->this_mirror
> num_copies
) {
2326 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2327 num_copies
, failrec
->this_mirror
, failed_mirror
);
2335 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2336 struct io_failure_record
*failrec
,
2337 struct page
*page
, int pg_offset
, int icsum
,
2338 bio_end_io_t
*endio_func
, void *data
)
2341 struct btrfs_io_bio
*btrfs_failed_bio
;
2342 struct btrfs_io_bio
*btrfs_bio
;
2344 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2348 bio
->bi_end_io
= endio_func
;
2349 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2350 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2351 bio
->bi_iter
.bi_size
= 0;
2352 bio
->bi_private
= data
;
2354 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2355 if (btrfs_failed_bio
->csum
) {
2356 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2357 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2359 btrfs_bio
= btrfs_io_bio(bio
);
2360 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2362 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2366 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2372 * this is a generic handler for readpage errors (default
2373 * readpage_io_failed_hook). if other copies exist, read those and write back
2374 * good data to the failed position. does not investigate in remapping the
2375 * failed extent elsewhere, hoping the device will be smart enough to do this as
2379 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2380 struct page
*page
, u64 start
, u64 end
,
2383 struct io_failure_record
*failrec
;
2384 struct inode
*inode
= page
->mapping
->host
;
2385 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2390 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
2392 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2396 ret
= btrfs_check_repairable(inode
, failed_bio
, failrec
, failed_mirror
);
2398 free_io_failure(inode
, failrec
);
2402 if (failed_bio
->bi_vcnt
> 1)
2403 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2405 read_mode
= READ_SYNC
;
2407 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2408 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2409 start
- page_offset(page
),
2410 (int)phy_offset
, failed_bio
->bi_end_io
,
2413 free_io_failure(inode
, failrec
);
2416 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
2418 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2419 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2421 ret
= tree
->ops
->submit_bio_hook(inode
, bio
, failrec
->this_mirror
,
2422 failrec
->bio_flags
, 0);
2424 free_io_failure(inode
, failrec
);
2431 /* lots and lots of room for performance fixes in the end_bio funcs */
2433 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2435 int uptodate
= (err
== 0);
2436 struct extent_io_tree
*tree
;
2439 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2441 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2442 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2443 end
, NULL
, uptodate
);
2449 ClearPageUptodate(page
);
2451 ret
= ret
< 0 ? ret
: -EIO
;
2452 mapping_set_error(page
->mapping
, ret
);
2457 * after a writepage IO is done, we need to:
2458 * clear the uptodate bits on error
2459 * clear the writeback bits in the extent tree for this IO
2460 * end_page_writeback if the page has no more pending IO
2462 * Scheduling is not allowed, so the extent state tree is expected
2463 * to have one and only one object corresponding to this IO.
2465 static void end_bio_extent_writepage(struct bio
*bio
)
2467 struct bio_vec
*bvec
;
2472 bio_for_each_segment_all(bvec
, bio
, i
) {
2473 struct page
*page
= bvec
->bv_page
;
2475 /* We always issue full-page reads, but if some block
2476 * in a page fails to read, blk_update_request() will
2477 * advance bv_offset and adjust bv_len to compensate.
2478 * Print a warning for nonzero offsets, and an error
2479 * if they don't add up to a full page. */
2480 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2481 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2482 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2483 "partial page write in btrfs with offset %u and length %u",
2484 bvec
->bv_offset
, bvec
->bv_len
);
2486 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2487 "incomplete page write in btrfs with offset %u and "
2489 bvec
->bv_offset
, bvec
->bv_len
);
2492 start
= page_offset(page
);
2493 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2495 end_extent_writepage(page
, bio
->bi_error
, start
, end
);
2496 end_page_writeback(page
);
2503 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2506 struct extent_state
*cached
= NULL
;
2507 u64 end
= start
+ len
- 1;
2509 if (uptodate
&& tree
->track_uptodate
)
2510 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2511 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2515 * after a readpage IO is done, we need to:
2516 * clear the uptodate bits on error
2517 * set the uptodate bits if things worked
2518 * set the page up to date if all extents in the tree are uptodate
2519 * clear the lock bit in the extent tree
2520 * unlock the page if there are no other extents locked for it
2522 * Scheduling is not allowed, so the extent state tree is expected
2523 * to have one and only one object corresponding to this IO.
2525 static void end_bio_extent_readpage(struct bio
*bio
)
2527 struct bio_vec
*bvec
;
2528 int uptodate
= !bio
->bi_error
;
2529 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2530 struct extent_io_tree
*tree
;
2535 u64 extent_start
= 0;
2541 bio_for_each_segment_all(bvec
, bio
, i
) {
2542 struct page
*page
= bvec
->bv_page
;
2543 struct inode
*inode
= page
->mapping
->host
;
2545 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2546 "mirror=%u\n", (u64
)bio
->bi_iter
.bi_sector
,
2547 bio
->bi_error
, io_bio
->mirror_num
);
2548 tree
= &BTRFS_I(inode
)->io_tree
;
2550 /* We always issue full-page reads, but if some block
2551 * in a page fails to read, blk_update_request() will
2552 * advance bv_offset and adjust bv_len to compensate.
2553 * Print a warning for nonzero offsets, and an error
2554 * if they don't add up to a full page. */
2555 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2556 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2557 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2558 "partial page read in btrfs with offset %u and length %u",
2559 bvec
->bv_offset
, bvec
->bv_len
);
2561 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2562 "incomplete page read in btrfs with offset %u and "
2564 bvec
->bv_offset
, bvec
->bv_len
);
2567 start
= page_offset(page
);
2568 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2571 mirror
= io_bio
->mirror_num
;
2572 if (likely(uptodate
&& tree
->ops
&&
2573 tree
->ops
->readpage_end_io_hook
)) {
2574 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2580 clean_io_failure(inode
, start
, page
, 0);
2583 if (likely(uptodate
))
2586 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2587 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2588 if (!ret
&& !bio
->bi_error
)
2592 * The generic bio_readpage_error handles errors the
2593 * following way: If possible, new read requests are
2594 * created and submitted and will end up in
2595 * end_bio_extent_readpage as well (if we're lucky, not
2596 * in the !uptodate case). In that case it returns 0 and
2597 * we just go on with the next page in our bio. If it
2598 * can't handle the error it will return -EIO and we
2599 * remain responsible for that page.
2601 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2604 uptodate
= !bio
->bi_error
;
2610 if (likely(uptodate
)) {
2611 loff_t i_size
= i_size_read(inode
);
2612 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2615 /* Zero out the end if this page straddles i_size */
2616 off
= i_size
& (PAGE_SIZE
-1);
2617 if (page
->index
== end_index
&& off
)
2618 zero_user_segment(page
, off
, PAGE_SIZE
);
2619 SetPageUptodate(page
);
2621 ClearPageUptodate(page
);
2627 if (unlikely(!uptodate
)) {
2629 endio_readpage_release_extent(tree
,
2635 endio_readpage_release_extent(tree
, start
,
2636 end
- start
+ 1, 0);
2637 } else if (!extent_len
) {
2638 extent_start
= start
;
2639 extent_len
= end
+ 1 - start
;
2640 } else if (extent_start
+ extent_len
== start
) {
2641 extent_len
+= end
+ 1 - start
;
2643 endio_readpage_release_extent(tree
, extent_start
,
2644 extent_len
, uptodate
);
2645 extent_start
= start
;
2646 extent_len
= end
+ 1 - start
;
2651 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2654 io_bio
->end_io(io_bio
, bio
->bi_error
);
2659 * this allocates from the btrfs_bioset. We're returning a bio right now
2660 * but you can call btrfs_io_bio for the appropriate container_of magic
2663 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2666 struct btrfs_io_bio
*btrfs_bio
;
2669 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2671 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2672 while (!bio
&& (nr_vecs
/= 2)) {
2673 bio
= bio_alloc_bioset(gfp_flags
,
2674 nr_vecs
, btrfs_bioset
);
2679 bio
->bi_bdev
= bdev
;
2680 bio
->bi_iter
.bi_sector
= first_sector
;
2681 btrfs_bio
= btrfs_io_bio(bio
);
2682 btrfs_bio
->csum
= NULL
;
2683 btrfs_bio
->csum_allocated
= NULL
;
2684 btrfs_bio
->end_io
= NULL
;
2689 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2691 struct btrfs_io_bio
*btrfs_bio
;
2694 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2696 btrfs_bio
= btrfs_io_bio(new);
2697 btrfs_bio
->csum
= NULL
;
2698 btrfs_bio
->csum_allocated
= NULL
;
2699 btrfs_bio
->end_io
= NULL
;
2701 #ifdef CONFIG_BLK_CGROUP
2702 /* FIXME, put this into bio_clone_bioset */
2704 bio_associate_blkcg(new, bio
->bi_css
);
2710 /* this also allocates from the btrfs_bioset */
2711 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2713 struct btrfs_io_bio
*btrfs_bio
;
2716 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2718 btrfs_bio
= btrfs_io_bio(bio
);
2719 btrfs_bio
->csum
= NULL
;
2720 btrfs_bio
->csum_allocated
= NULL
;
2721 btrfs_bio
->end_io
= NULL
;
2727 static int __must_check
submit_one_bio(struct bio
*bio
, int mirror_num
,
2728 unsigned long bio_flags
)
2731 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2732 struct page
*page
= bvec
->bv_page
;
2733 struct extent_io_tree
*tree
= bio
->bi_private
;
2736 start
= page_offset(page
) + bvec
->bv_offset
;
2738 bio
->bi_private
= NULL
;
2741 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2742 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, bio
,
2743 mirror_num
, bio_flags
, start
);
2745 btrfsic_submit_bio(bio
);
2751 static int merge_bio(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(page
, offset
, size
, bio
,
2764 static int submit_extent_page(int op
, int op_flags
, 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(tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2793 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2794 ret
= submit_one_bio(bio
, mirror_num
, prev_bio_flags
);
2802 wbc_account_io(wbc
, page
, page_size
);
2807 bio
= btrfs_bio_alloc(bdev
, sector
, BIO_MAX_PAGES
,
2808 GFP_NOFS
| __GFP_HIGH
);
2812 bio_add_page(bio
, page
, page_size
, offset
);
2813 bio
->bi_end_io
= end_io_func
;
2814 bio
->bi_private
= tree
;
2815 bio_set_op_attrs(bio
, op
, op_flags
);
2817 wbc_init_bio(wbc
, bio
);
2818 wbc_account_io(wbc
, page
, page_size
);
2824 ret
= submit_one_bio(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 read_flags
,
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(REQ_OP_READ
, read_flags
, tree
, NULL
,
3072 page
, 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
,
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
, 0, 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
,
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
,
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 read_flags
)
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
, read_flags
, 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(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(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3444 page
, 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
;
3483 int write_flags
= 0;
3484 unsigned long nr_written
= 0;
3486 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3487 write_flags
= WRITE_SYNC
;
3489 trace___extent_writepage(page
, inode
, wbc
);
3491 WARN_ON(!PageLocked(page
));
3493 ClearPageError(page
);
3495 pg_offset
= i_size
& (PAGE_SIZE
- 1);
3496 if (page
->index
> end_index
||
3497 (page
->index
== end_index
&& !pg_offset
)) {
3498 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3503 if (page
->index
== end_index
) {
3506 userpage
= kmap_atomic(page
);
3507 memset(userpage
+ pg_offset
, 0,
3508 PAGE_SIZE
- pg_offset
);
3509 kunmap_atomic(userpage
);
3510 flush_dcache_page(page
);
3515 set_page_extent_mapped(page
);
3517 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3523 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3524 i_size
, nr_written
, write_flags
, &nr
);
3530 /* make sure the mapping tag for page dirty gets cleared */
3531 set_page_writeback(page
);
3532 end_page_writeback(page
);
3534 if (PageError(page
)) {
3535 ret
= ret
< 0 ? ret
: -EIO
;
3536 end_extent_writepage(page
, ret
, start
, page_end
);
3545 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3547 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3548 TASK_UNINTERRUPTIBLE
);
3551 static noinline_for_stack
int
3552 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3553 struct btrfs_fs_info
*fs_info
,
3554 struct extent_page_data
*epd
)
3556 unsigned long i
, num_pages
;
3560 if (!btrfs_try_tree_write_lock(eb
)) {
3562 flush_write_bio(epd
);
3563 btrfs_tree_lock(eb
);
3566 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3567 btrfs_tree_unlock(eb
);
3571 flush_write_bio(epd
);
3575 wait_on_extent_buffer_writeback(eb
);
3576 btrfs_tree_lock(eb
);
3577 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3579 btrfs_tree_unlock(eb
);
3584 * We need to do this to prevent races in people who check if the eb is
3585 * under IO since we can end up having no IO bits set for a short period
3588 spin_lock(&eb
->refs_lock
);
3589 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3590 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3591 spin_unlock(&eb
->refs_lock
);
3592 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3593 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3595 fs_info
->dirty_metadata_batch
);
3598 spin_unlock(&eb
->refs_lock
);
3601 btrfs_tree_unlock(eb
);
3606 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3607 for (i
= 0; i
< num_pages
; i
++) {
3608 struct page
*p
= eb
->pages
[i
];
3610 if (!trylock_page(p
)) {
3612 flush_write_bio(epd
);
3622 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3624 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3625 smp_mb__after_atomic();
3626 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3629 static void set_btree_ioerr(struct page
*page
)
3631 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3632 struct btrfs_inode
*btree_ino
= BTRFS_I(eb
->fs_info
->btree_inode
);
3635 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3639 * If writeback for a btree extent that doesn't belong to a log tree
3640 * failed, increment the counter transaction->eb_write_errors.
3641 * We do this because while the transaction is running and before it's
3642 * committing (when we call filemap_fdata[write|wait]_range against
3643 * the btree inode), we might have
3644 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3645 * returns an error or an error happens during writeback, when we're
3646 * committing the transaction we wouldn't know about it, since the pages
3647 * can be no longer dirty nor marked anymore for writeback (if a
3648 * subsequent modification to the extent buffer didn't happen before the
3649 * transaction commit), which makes filemap_fdata[write|wait]_range not
3650 * able to find the pages tagged with SetPageError at transaction
3651 * commit time. So if this happens we must abort the transaction,
3652 * otherwise we commit a super block with btree roots that point to
3653 * btree nodes/leafs whose content on disk is invalid - either garbage
3654 * or the content of some node/leaf from a past generation that got
3655 * cowed or deleted and is no longer valid.
3657 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3658 * not be enough - we need to distinguish between log tree extents vs
3659 * non-log tree extents, and the next filemap_fdatawait_range() call
3660 * will catch and clear such errors in the mapping - and that call might
3661 * be from a log sync and not from a transaction commit. Also, checking
3662 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3663 * not done and would not be reliable - the eb might have been released
3664 * from memory and reading it back again means that flag would not be
3665 * set (since it's a runtime flag, not persisted on disk).
3667 * Using the flags below in the btree inode also makes us achieve the
3668 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3669 * writeback for all dirty pages and before filemap_fdatawait_range()
3670 * is called, the writeback for all dirty pages had already finished
3671 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3672 * filemap_fdatawait_range() would return success, as it could not know
3673 * that writeback errors happened (the pages were no longer tagged for
3676 switch (eb
->log_index
) {
3678 set_bit(BTRFS_INODE_BTREE_ERR
, &btree_ino
->runtime_flags
);
3681 set_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
3684 set_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
3687 BUG(); /* unexpected, logic error */
3691 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3693 struct bio_vec
*bvec
;
3694 struct extent_buffer
*eb
;
3697 bio_for_each_segment_all(bvec
, bio
, i
) {
3698 struct page
*page
= bvec
->bv_page
;
3700 eb
= (struct extent_buffer
*)page
->private;
3702 done
= atomic_dec_and_test(&eb
->io_pages
);
3704 if (bio
->bi_error
||
3705 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3706 ClearPageUptodate(page
);
3707 set_btree_ioerr(page
);
3710 end_page_writeback(page
);
3715 end_extent_buffer_writeback(eb
);
3721 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3722 struct btrfs_fs_info
*fs_info
,
3723 struct writeback_control
*wbc
,
3724 struct extent_page_data
*epd
)
3726 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3727 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3728 u64 offset
= eb
->start
;
3729 unsigned long i
, num_pages
;
3730 unsigned long bio_flags
= 0;
3731 int write_flags
= (epd
->sync_io
? WRITE_SYNC
: 0) | REQ_META
;
3734 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3735 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3736 atomic_set(&eb
->io_pages
, num_pages
);
3737 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3738 bio_flags
= EXTENT_BIO_TREE_LOG
;
3740 for (i
= 0; i
< num_pages
; i
++) {
3741 struct page
*p
= eb
->pages
[i
];
3743 clear_page_dirty_for_io(p
);
3744 set_page_writeback(p
);
3745 ret
= submit_extent_page(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3746 p
, offset
>> 9, PAGE_SIZE
, 0, bdev
,
3748 end_bio_extent_buffer_writepage
,
3749 0, epd
->bio_flags
, bio_flags
, false);
3750 epd
->bio_flags
= bio_flags
;
3753 end_page_writeback(p
);
3754 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3755 end_extent_buffer_writeback(eb
);
3759 offset
+= PAGE_SIZE
;
3760 update_nr_written(p
, wbc
, 1);
3764 if (unlikely(ret
)) {
3765 for (; i
< num_pages
; i
++) {
3766 struct page
*p
= eb
->pages
[i
];
3767 clear_page_dirty_for_io(p
);
3775 int btree_write_cache_pages(struct address_space
*mapping
,
3776 struct writeback_control
*wbc
)
3778 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3779 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3780 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3781 struct extent_page_data epd
= {
3785 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3790 int nr_to_write_done
= 0;
3791 struct pagevec pvec
;
3794 pgoff_t end
; /* Inclusive */
3798 pagevec_init(&pvec
, 0);
3799 if (wbc
->range_cyclic
) {
3800 index
= mapping
->writeback_index
; /* Start from prev offset */
3803 index
= wbc
->range_start
>> PAGE_SHIFT
;
3804 end
= wbc
->range_end
>> PAGE_SHIFT
;
3807 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3808 tag
= PAGECACHE_TAG_TOWRITE
;
3810 tag
= PAGECACHE_TAG_DIRTY
;
3812 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3813 tag_pages_for_writeback(mapping
, index
, end
);
3814 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3815 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3816 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3820 for (i
= 0; i
< nr_pages
; i
++) {
3821 struct page
*page
= pvec
.pages
[i
];
3823 if (!PagePrivate(page
))
3826 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3831 spin_lock(&mapping
->private_lock
);
3832 if (!PagePrivate(page
)) {
3833 spin_unlock(&mapping
->private_lock
);
3837 eb
= (struct extent_buffer
*)page
->private;
3840 * Shouldn't happen and normally this would be a BUG_ON
3841 * but no sense in crashing the users box for something
3842 * we can survive anyway.
3845 spin_unlock(&mapping
->private_lock
);
3849 if (eb
== prev_eb
) {
3850 spin_unlock(&mapping
->private_lock
);
3854 ret
= atomic_inc_not_zero(&eb
->refs
);
3855 spin_unlock(&mapping
->private_lock
);
3860 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3862 free_extent_buffer(eb
);
3866 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3869 free_extent_buffer(eb
);
3872 free_extent_buffer(eb
);
3875 * the filesystem may choose to bump up nr_to_write.
3876 * We have to make sure to honor the new nr_to_write
3879 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3881 pagevec_release(&pvec
);
3884 if (!scanned
&& !done
) {
3886 * We hit the last page and there is more work to be done: wrap
3887 * back to the start of the file
3893 flush_write_bio(&epd
);
3898 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3899 * @mapping: address space structure to write
3900 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3901 * @writepage: function called for each page
3902 * @data: data passed to writepage function
3904 * If a page is already under I/O, write_cache_pages() skips it, even
3905 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3906 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3907 * and msync() need to guarantee that all the data which was dirty at the time
3908 * the call was made get new I/O started against them. If wbc->sync_mode is
3909 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3910 * existing IO to complete.
3912 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3913 struct address_space
*mapping
,
3914 struct writeback_control
*wbc
,
3915 writepage_t writepage
, void *data
,
3916 void (*flush_fn
)(void *))
3918 struct inode
*inode
= mapping
->host
;
3921 int nr_to_write_done
= 0;
3922 struct pagevec pvec
;
3925 pgoff_t end
; /* Inclusive */
3927 int range_whole
= 0;
3932 * We have to hold onto the inode so that ordered extents can do their
3933 * work when the IO finishes. The alternative to this is failing to add
3934 * an ordered extent if the igrab() fails there and that is a huge pain
3935 * to deal with, so instead just hold onto the inode throughout the
3936 * writepages operation. If it fails here we are freeing up the inode
3937 * anyway and we'd rather not waste our time writing out stuff that is
3938 * going to be truncated anyway.
3943 pagevec_init(&pvec
, 0);
3944 if (wbc
->range_cyclic
) {
3945 index
= mapping
->writeback_index
; /* Start from prev offset */
3948 index
= wbc
->range_start
>> PAGE_SHIFT
;
3949 end
= wbc
->range_end
>> PAGE_SHIFT
;
3950 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3954 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3955 tag
= PAGECACHE_TAG_TOWRITE
;
3957 tag
= PAGECACHE_TAG_DIRTY
;
3959 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3960 tag_pages_for_writeback(mapping
, index
, end
);
3962 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3963 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3964 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3968 for (i
= 0; i
< nr_pages
; i
++) {
3969 struct page
*page
= pvec
.pages
[i
];
3971 done_index
= page
->index
;
3973 * At this point we hold neither mapping->tree_lock nor
3974 * lock on the page itself: the page may be truncated or
3975 * invalidated (changing page->mapping to NULL), or even
3976 * swizzled back from swapper_space to tmpfs file
3979 if (!trylock_page(page
)) {
3984 if (unlikely(page
->mapping
!= mapping
)) {
3989 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3995 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3996 if (PageWriteback(page
))
3998 wait_on_page_writeback(page
);
4001 if (PageWriteback(page
) ||
4002 !clear_page_dirty_for_io(page
)) {
4007 ret
= (*writepage
)(page
, wbc
, data
);
4009 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4015 * done_index is set past this page,
4016 * so media errors will not choke
4017 * background writeout for the entire
4018 * file. This has consequences for
4019 * range_cyclic semantics (ie. it may
4020 * not be suitable for data integrity
4023 done_index
= page
->index
+ 1;
4029 * the filesystem may choose to bump up nr_to_write.
4030 * We have to make sure to honor the new nr_to_write
4033 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4035 pagevec_release(&pvec
);
4038 if (!scanned
&& !done
) {
4040 * We hit the last page and there is more work to be done: wrap
4041 * back to the start of the file
4048 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4049 mapping
->writeback_index
= done_index
;
4051 btrfs_add_delayed_iput(inode
);
4055 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4060 bio_set_op_attrs(epd
->bio
, REQ_OP_WRITE
,
4061 epd
->sync_io
? WRITE_SYNC
: 0);
4063 ret
= submit_one_bio(epd
->bio
, 0, epd
->bio_flags
);
4064 BUG_ON(ret
< 0); /* -ENOMEM */
4069 static noinline
void flush_write_bio(void *data
)
4071 struct extent_page_data
*epd
= data
;
4072 flush_epd_write_bio(epd
);
4075 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4076 get_extent_t
*get_extent
,
4077 struct writeback_control
*wbc
)
4080 struct extent_page_data epd
= {
4083 .get_extent
= get_extent
,
4085 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4089 ret
= __extent_writepage(page
, wbc
, &epd
);
4091 flush_epd_write_bio(&epd
);
4095 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4096 u64 start
, u64 end
, get_extent_t
*get_extent
,
4100 struct address_space
*mapping
= inode
->i_mapping
;
4102 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4105 struct extent_page_data epd
= {
4108 .get_extent
= get_extent
,
4110 .sync_io
= mode
== WB_SYNC_ALL
,
4113 struct writeback_control wbc_writepages
= {
4115 .nr_to_write
= nr_pages
* 2,
4116 .range_start
= start
,
4117 .range_end
= end
+ 1,
4120 while (start
<= end
) {
4121 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4122 if (clear_page_dirty_for_io(page
))
4123 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4125 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4126 tree
->ops
->writepage_end_io_hook(page
, start
,
4127 start
+ PAGE_SIZE
- 1,
4135 flush_epd_write_bio(&epd
);
4139 int extent_writepages(struct extent_io_tree
*tree
,
4140 struct address_space
*mapping
,
4141 get_extent_t
*get_extent
,
4142 struct writeback_control
*wbc
)
4145 struct extent_page_data epd
= {
4148 .get_extent
= get_extent
,
4150 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4154 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
4155 __extent_writepage
, &epd
,
4157 flush_epd_write_bio(&epd
);
4161 int extent_readpages(struct extent_io_tree
*tree
,
4162 struct address_space
*mapping
,
4163 struct list_head
*pages
, unsigned nr_pages
,
4164 get_extent_t get_extent
)
4166 struct bio
*bio
= NULL
;
4168 unsigned long bio_flags
= 0;
4169 struct page
*pagepool
[16];
4171 struct extent_map
*em_cached
= NULL
;
4173 u64 prev_em_start
= (u64
)-1;
4175 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4176 page
= list_entry(pages
->prev
, struct page
, lru
);
4178 prefetchw(&page
->flags
);
4179 list_del(&page
->lru
);
4180 if (add_to_page_cache_lru(page
, mapping
,
4182 readahead_gfp_mask(mapping
))) {
4187 pagepool
[nr
++] = page
;
4188 if (nr
< ARRAY_SIZE(pagepool
))
4190 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4191 &bio
, 0, &bio_flags
, &prev_em_start
);
4195 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4196 &bio
, 0, &bio_flags
, &prev_em_start
);
4199 free_extent_map(em_cached
);
4201 BUG_ON(!list_empty(pages
));
4203 return submit_one_bio(bio
, 0, bio_flags
);
4208 * basic invalidatepage code, this waits on any locked or writeback
4209 * ranges corresponding to the page, and then deletes any extent state
4210 * records from the tree
4212 int extent_invalidatepage(struct extent_io_tree
*tree
,
4213 struct page
*page
, unsigned long offset
)
4215 struct extent_state
*cached_state
= NULL
;
4216 u64 start
= page_offset(page
);
4217 u64 end
= start
+ PAGE_SIZE
- 1;
4218 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4220 start
+= ALIGN(offset
, blocksize
);
4224 lock_extent_bits(tree
, start
, end
, &cached_state
);
4225 wait_on_page_writeback(page
);
4226 clear_extent_bit(tree
, start
, end
,
4227 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4228 EXTENT_DO_ACCOUNTING
,
4229 1, 1, &cached_state
, GFP_NOFS
);
4234 * a helper for releasepage, this tests for areas of the page that
4235 * are locked or under IO and drops the related state bits if it is safe
4238 static int try_release_extent_state(struct extent_map_tree
*map
,
4239 struct extent_io_tree
*tree
,
4240 struct page
*page
, gfp_t mask
)
4242 u64 start
= page_offset(page
);
4243 u64 end
= start
+ PAGE_SIZE
- 1;
4246 if (test_range_bit(tree
, start
, end
,
4247 EXTENT_IOBITS
, 0, NULL
))
4250 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4253 * at this point we can safely clear everything except the
4254 * locked bit and the nodatasum bit
4256 ret
= clear_extent_bit(tree
, start
, end
,
4257 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4260 /* if clear_extent_bit failed for enomem reasons,
4261 * we can't allow the release to continue.
4272 * a helper for releasepage. As long as there are no locked extents
4273 * in the range corresponding to the page, both state records and extent
4274 * map records are removed
4276 int try_release_extent_mapping(struct extent_map_tree
*map
,
4277 struct extent_io_tree
*tree
, struct page
*page
,
4280 struct extent_map
*em
;
4281 u64 start
= page_offset(page
);
4282 u64 end
= start
+ PAGE_SIZE
- 1;
4284 if (gfpflags_allow_blocking(mask
) &&
4285 page
->mapping
->host
->i_size
> SZ_16M
) {
4287 while (start
<= end
) {
4288 len
= end
- start
+ 1;
4289 write_lock(&map
->lock
);
4290 em
= lookup_extent_mapping(map
, start
, len
);
4292 write_unlock(&map
->lock
);
4295 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4296 em
->start
!= start
) {
4297 write_unlock(&map
->lock
);
4298 free_extent_map(em
);
4301 if (!test_range_bit(tree
, em
->start
,
4302 extent_map_end(em
) - 1,
4303 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4305 remove_extent_mapping(map
, em
);
4306 /* once for the rb tree */
4307 free_extent_map(em
);
4309 start
= extent_map_end(em
);
4310 write_unlock(&map
->lock
);
4313 free_extent_map(em
);
4316 return try_release_extent_state(map
, tree
, page
, mask
);
4320 * helper function for fiemap, which doesn't want to see any holes.
4321 * This maps until we find something past 'last'
4323 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4326 get_extent_t
*get_extent
)
4328 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4329 struct extent_map
*em
;
4336 len
= last
- offset
;
4339 len
= ALIGN(len
, sectorsize
);
4340 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4341 if (IS_ERR_OR_NULL(em
))
4344 /* if this isn't a hole return it */
4345 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4346 em
->block_start
!= EXTENT_MAP_HOLE
) {
4350 /* this is a hole, advance to the next extent */
4351 offset
= extent_map_end(em
);
4352 free_extent_map(em
);
4359 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4360 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4364 u64 max
= start
+ len
;
4368 u64 last_for_get_extent
= 0;
4370 u64 isize
= i_size_read(inode
);
4371 struct btrfs_key found_key
;
4372 struct extent_map
*em
= NULL
;
4373 struct extent_state
*cached_state
= NULL
;
4374 struct btrfs_path
*path
;
4375 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4384 path
= btrfs_alloc_path();
4387 path
->leave_spinning
= 1;
4389 start
= round_down(start
, BTRFS_I(inode
)->root
->sectorsize
);
4390 len
= round_up(max
, BTRFS_I(inode
)->root
->sectorsize
) - start
;
4393 * lookup the last file extent. We're not using i_size here
4394 * because there might be preallocation past i_size
4396 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
), -1,
4399 btrfs_free_path(path
);
4408 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4409 found_type
= found_key
.type
;
4411 /* No extents, but there might be delalloc bits */
4412 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4413 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4414 /* have to trust i_size as the end */
4416 last_for_get_extent
= isize
;
4419 * remember the start of the last extent. There are a
4420 * bunch of different factors that go into the length of the
4421 * extent, so its much less complex to remember where it started
4423 last
= found_key
.offset
;
4424 last_for_get_extent
= last
+ 1;
4426 btrfs_release_path(path
);
4429 * we might have some extents allocated but more delalloc past those
4430 * extents. so, we trust isize unless the start of the last extent is
4435 last_for_get_extent
= isize
;
4438 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4441 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4451 u64 offset_in_extent
= 0;
4453 /* break if the extent we found is outside the range */
4454 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4458 * get_extent may return an extent that starts before our
4459 * requested range. We have to make sure the ranges
4460 * we return to fiemap always move forward and don't
4461 * overlap, so adjust the offsets here
4463 em_start
= max(em
->start
, off
);
4466 * record the offset from the start of the extent
4467 * for adjusting the disk offset below. Only do this if the
4468 * extent isn't compressed since our in ram offset may be past
4469 * what we have actually allocated on disk.
4471 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4472 offset_in_extent
= em_start
- em
->start
;
4473 em_end
= extent_map_end(em
);
4474 em_len
= em_end
- em_start
;
4479 * bump off for our next call to get_extent
4481 off
= extent_map_end(em
);
4485 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4487 flags
|= FIEMAP_EXTENT_LAST
;
4488 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4489 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4490 FIEMAP_EXTENT_NOT_ALIGNED
);
4491 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4492 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4493 FIEMAP_EXTENT_UNKNOWN
);
4494 } else if (fieinfo
->fi_extents_max
) {
4495 u64 bytenr
= em
->block_start
-
4496 (em
->start
- em
->orig_start
);
4498 disko
= em
->block_start
+ offset_in_extent
;
4501 * As btrfs supports shared space, this information
4502 * can be exported to userspace tools via
4503 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4504 * then we're just getting a count and we can skip the
4507 ret
= btrfs_check_shared(NULL
, root
->fs_info
,
4509 btrfs_ino(inode
), bytenr
);
4513 flags
|= FIEMAP_EXTENT_SHARED
;
4516 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4517 flags
|= FIEMAP_EXTENT_ENCODED
;
4518 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4519 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4521 free_extent_map(em
);
4523 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4524 (last
== (u64
)-1 && isize
<= em_end
)) {
4525 flags
|= FIEMAP_EXTENT_LAST
;
4529 /* now scan forward to see if this is really the last extent. */
4530 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4537 flags
|= FIEMAP_EXTENT_LAST
;
4540 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4549 free_extent_map(em
);
4551 btrfs_free_path(path
);
4552 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4553 &cached_state
, GFP_NOFS
);
4557 static void __free_extent_buffer(struct extent_buffer
*eb
)
4559 btrfs_leak_debug_del(&eb
->leak_list
);
4560 kmem_cache_free(extent_buffer_cache
, eb
);
4563 int extent_buffer_under_io(struct extent_buffer
*eb
)
4565 return (atomic_read(&eb
->io_pages
) ||
4566 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4567 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4571 * Helper for releasing extent buffer page.
4573 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4575 unsigned long index
;
4577 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4579 BUG_ON(extent_buffer_under_io(eb
));
4581 index
= num_extent_pages(eb
->start
, eb
->len
);
4587 page
= eb
->pages
[index
];
4591 spin_lock(&page
->mapping
->private_lock
);
4593 * We do this since we'll remove the pages after we've
4594 * removed the eb from the radix tree, so we could race
4595 * and have this page now attached to the new eb. So
4596 * only clear page_private if it's still connected to
4599 if (PagePrivate(page
) &&
4600 page
->private == (unsigned long)eb
) {
4601 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4602 BUG_ON(PageDirty(page
));
4603 BUG_ON(PageWriteback(page
));
4605 * We need to make sure we haven't be attached
4608 ClearPagePrivate(page
);
4609 set_page_private(page
, 0);
4610 /* One for the page private */
4615 spin_unlock(&page
->mapping
->private_lock
);
4617 /* One for when we allocated the page */
4619 } while (index
!= 0);
4623 * Helper for releasing the extent buffer.
4625 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4627 btrfs_release_extent_buffer_page(eb
);
4628 __free_extent_buffer(eb
);
4631 static struct extent_buffer
*
4632 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4635 struct extent_buffer
*eb
= NULL
;
4637 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4640 eb
->fs_info
= fs_info
;
4642 rwlock_init(&eb
->lock
);
4643 atomic_set(&eb
->write_locks
, 0);
4644 atomic_set(&eb
->read_locks
, 0);
4645 atomic_set(&eb
->blocking_readers
, 0);
4646 atomic_set(&eb
->blocking_writers
, 0);
4647 atomic_set(&eb
->spinning_readers
, 0);
4648 atomic_set(&eb
->spinning_writers
, 0);
4649 eb
->lock_nested
= 0;
4650 init_waitqueue_head(&eb
->write_lock_wq
);
4651 init_waitqueue_head(&eb
->read_lock_wq
);
4653 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4655 spin_lock_init(&eb
->refs_lock
);
4656 atomic_set(&eb
->refs
, 1);
4657 atomic_set(&eb
->io_pages
, 0);
4660 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4662 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4663 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4664 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4669 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4673 struct extent_buffer
*new;
4674 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4676 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4680 for (i
= 0; i
< num_pages
; i
++) {
4681 p
= alloc_page(GFP_NOFS
);
4683 btrfs_release_extent_buffer(new);
4686 attach_extent_buffer_page(new, p
);
4687 WARN_ON(PageDirty(p
));
4692 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4693 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4694 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4699 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4700 u64 start
, unsigned long len
)
4702 struct extent_buffer
*eb
;
4703 unsigned long num_pages
;
4706 num_pages
= num_extent_pages(start
, len
);
4708 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4712 for (i
= 0; i
< num_pages
; i
++) {
4713 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4717 set_extent_buffer_uptodate(eb
);
4718 btrfs_set_header_nritems(eb
, 0);
4719 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4724 __free_page(eb
->pages
[i
- 1]);
4725 __free_extent_buffer(eb
);
4729 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4730 u64 start
, u32 nodesize
)
4736 * Called only from tests that don't always have a fs_info
4741 len
= fs_info
->tree_root
->nodesize
;
4744 return __alloc_dummy_extent_buffer(fs_info
, start
, len
);
4747 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4750 /* the ref bit is tricky. We have to make sure it is set
4751 * if we have the buffer dirty. Otherwise the
4752 * code to free a buffer can end up dropping a dirty
4755 * Once the ref bit is set, it won't go away while the
4756 * buffer is dirty or in writeback, and it also won't
4757 * go away while we have the reference count on the
4760 * We can't just set the ref bit without bumping the
4761 * ref on the eb because free_extent_buffer might
4762 * see the ref bit and try to clear it. If this happens
4763 * free_extent_buffer might end up dropping our original
4764 * ref by mistake and freeing the page before we are able
4765 * to add one more ref.
4767 * So bump the ref count first, then set the bit. If someone
4768 * beat us to it, drop the ref we added.
4770 refs
= atomic_read(&eb
->refs
);
4771 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4774 spin_lock(&eb
->refs_lock
);
4775 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4776 atomic_inc(&eb
->refs
);
4777 spin_unlock(&eb
->refs_lock
);
4780 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4781 struct page
*accessed
)
4783 unsigned long num_pages
, i
;
4785 check_buffer_tree_ref(eb
);
4787 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4788 for (i
= 0; i
< num_pages
; i
++) {
4789 struct page
*p
= eb
->pages
[i
];
4792 mark_page_accessed(p
);
4796 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4799 struct extent_buffer
*eb
;
4802 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4803 start
>> PAGE_SHIFT
);
4804 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4807 * Lock our eb's refs_lock to avoid races with
4808 * free_extent_buffer. When we get our eb it might be flagged
4809 * with EXTENT_BUFFER_STALE and another task running
4810 * free_extent_buffer might have seen that flag set,
4811 * eb->refs == 2, that the buffer isn't under IO (dirty and
4812 * writeback flags not set) and it's still in the tree (flag
4813 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4814 * of decrementing the extent buffer's reference count twice.
4815 * So here we could race and increment the eb's reference count,
4816 * clear its stale flag, mark it as dirty and drop our reference
4817 * before the other task finishes executing free_extent_buffer,
4818 * which would later result in an attempt to free an extent
4819 * buffer that is dirty.
4821 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4822 spin_lock(&eb
->refs_lock
);
4823 spin_unlock(&eb
->refs_lock
);
4825 mark_extent_buffer_accessed(eb
, NULL
);
4833 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4834 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4835 u64 start
, u32 nodesize
)
4837 struct extent_buffer
*eb
, *exists
= NULL
;
4840 eb
= find_extent_buffer(fs_info
, start
);
4843 eb
= alloc_dummy_extent_buffer(fs_info
, start
, nodesize
);
4846 eb
->fs_info
= fs_info
;
4848 ret
= radix_tree_preload(GFP_NOFS
);
4851 spin_lock(&fs_info
->buffer_lock
);
4852 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4853 start
>> PAGE_SHIFT
, eb
);
4854 spin_unlock(&fs_info
->buffer_lock
);
4855 radix_tree_preload_end();
4856 if (ret
== -EEXIST
) {
4857 exists
= find_extent_buffer(fs_info
, start
);
4863 check_buffer_tree_ref(eb
);
4864 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4867 * We will free dummy extent buffer's if they come into
4868 * free_extent_buffer with a ref count of 2, but if we are using this we
4869 * want the buffers to stay in memory until we're done with them, so
4870 * bump the ref count again.
4872 atomic_inc(&eb
->refs
);
4875 btrfs_release_extent_buffer(eb
);
4880 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4883 unsigned long len
= fs_info
->tree_root
->nodesize
;
4884 unsigned long num_pages
= num_extent_pages(start
, len
);
4886 unsigned long index
= start
>> PAGE_SHIFT
;
4887 struct extent_buffer
*eb
;
4888 struct extent_buffer
*exists
= NULL
;
4890 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4894 if (!IS_ALIGNED(start
, fs_info
->tree_root
->sectorsize
)) {
4895 btrfs_err(fs_info
, "bad tree block start %llu", start
);
4896 return ERR_PTR(-EINVAL
);
4899 eb
= find_extent_buffer(fs_info
, start
);
4903 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4905 return ERR_PTR(-ENOMEM
);
4907 for (i
= 0; i
< num_pages
; i
++, index
++) {
4908 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4910 exists
= ERR_PTR(-ENOMEM
);
4914 spin_lock(&mapping
->private_lock
);
4915 if (PagePrivate(p
)) {
4917 * We could have already allocated an eb for this page
4918 * and attached one so lets see if we can get a ref on
4919 * the existing eb, and if we can we know it's good and
4920 * we can just return that one, else we know we can just
4921 * overwrite page->private.
4923 exists
= (struct extent_buffer
*)p
->private;
4924 if (atomic_inc_not_zero(&exists
->refs
)) {
4925 spin_unlock(&mapping
->private_lock
);
4928 mark_extent_buffer_accessed(exists
, p
);
4934 * Do this so attach doesn't complain and we need to
4935 * drop the ref the old guy had.
4937 ClearPagePrivate(p
);
4938 WARN_ON(PageDirty(p
));
4941 attach_extent_buffer_page(eb
, p
);
4942 spin_unlock(&mapping
->private_lock
);
4943 WARN_ON(PageDirty(p
));
4945 if (!PageUptodate(p
))
4949 * see below about how we avoid a nasty race with release page
4950 * and why we unlock later
4954 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4956 ret
= radix_tree_preload(GFP_NOFS
);
4958 exists
= ERR_PTR(ret
);
4962 spin_lock(&fs_info
->buffer_lock
);
4963 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4964 start
>> PAGE_SHIFT
, eb
);
4965 spin_unlock(&fs_info
->buffer_lock
);
4966 radix_tree_preload_end();
4967 if (ret
== -EEXIST
) {
4968 exists
= find_extent_buffer(fs_info
, start
);
4974 /* add one reference for the tree */
4975 check_buffer_tree_ref(eb
);
4976 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4979 * there is a race where release page may have
4980 * tried to find this extent buffer in the radix
4981 * but failed. It will tell the VM it is safe to
4982 * reclaim the, and it will clear the page private bit.
4983 * We must make sure to set the page private bit properly
4984 * after the extent buffer is in the radix tree so
4985 * it doesn't get lost
4987 SetPageChecked(eb
->pages
[0]);
4988 for (i
= 1; i
< num_pages
; i
++) {
4990 ClearPageChecked(p
);
4993 unlock_page(eb
->pages
[0]);
4997 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4998 for (i
= 0; i
< num_pages
; i
++) {
5000 unlock_page(eb
->pages
[i
]);
5003 btrfs_release_extent_buffer(eb
);
5007 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
5009 struct extent_buffer
*eb
=
5010 container_of(head
, struct extent_buffer
, rcu_head
);
5012 __free_extent_buffer(eb
);
5015 /* Expects to have eb->eb_lock already held */
5016 static int release_extent_buffer(struct extent_buffer
*eb
)
5018 WARN_ON(atomic_read(&eb
->refs
) == 0);
5019 if (atomic_dec_and_test(&eb
->refs
)) {
5020 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5021 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5023 spin_unlock(&eb
->refs_lock
);
5025 spin_lock(&fs_info
->buffer_lock
);
5026 radix_tree_delete(&fs_info
->buffer_radix
,
5027 eb
->start
>> PAGE_SHIFT
);
5028 spin_unlock(&fs_info
->buffer_lock
);
5030 spin_unlock(&eb
->refs_lock
);
5033 /* Should be safe to release our pages at this point */
5034 btrfs_release_extent_buffer_page(eb
);
5035 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5036 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5037 __free_extent_buffer(eb
);
5041 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5044 spin_unlock(&eb
->refs_lock
);
5049 void free_extent_buffer(struct extent_buffer
*eb
)
5057 refs
= atomic_read(&eb
->refs
);
5060 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5065 spin_lock(&eb
->refs_lock
);
5066 if (atomic_read(&eb
->refs
) == 2 &&
5067 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5068 atomic_dec(&eb
->refs
);
5070 if (atomic_read(&eb
->refs
) == 2 &&
5071 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5072 !extent_buffer_under_io(eb
) &&
5073 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5074 atomic_dec(&eb
->refs
);
5077 * I know this is terrible, but it's temporary until we stop tracking
5078 * the uptodate bits and such for the extent buffers.
5080 release_extent_buffer(eb
);
5083 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5088 spin_lock(&eb
->refs_lock
);
5089 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5091 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5092 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5093 atomic_dec(&eb
->refs
);
5094 release_extent_buffer(eb
);
5097 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5100 unsigned long num_pages
;
5103 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5105 for (i
= 0; i
< num_pages
; i
++) {
5106 page
= eb
->pages
[i
];
5107 if (!PageDirty(page
))
5111 WARN_ON(!PagePrivate(page
));
5113 clear_page_dirty_for_io(page
);
5114 spin_lock_irq(&page
->mapping
->tree_lock
);
5115 if (!PageDirty(page
)) {
5116 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5118 PAGECACHE_TAG_DIRTY
);
5120 spin_unlock_irq(&page
->mapping
->tree_lock
);
5121 ClearPageError(page
);
5124 WARN_ON(atomic_read(&eb
->refs
) == 0);
5127 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5130 unsigned long num_pages
;
5133 check_buffer_tree_ref(eb
);
5135 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5137 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5138 WARN_ON(atomic_read(&eb
->refs
) == 0);
5139 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5141 for (i
= 0; i
< num_pages
; i
++)
5142 set_page_dirty(eb
->pages
[i
]);
5146 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5150 unsigned long num_pages
;
5152 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5153 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5154 for (i
= 0; i
< num_pages
; i
++) {
5155 page
= eb
->pages
[i
];
5157 ClearPageUptodate(page
);
5161 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5165 unsigned long num_pages
;
5167 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5168 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5169 for (i
= 0; i
< num_pages
; i
++) {
5170 page
= eb
->pages
[i
];
5171 SetPageUptodate(page
);
5175 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5177 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5180 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5181 struct extent_buffer
*eb
, u64 start
, int wait
,
5182 get_extent_t
*get_extent
, int mirror_num
)
5185 unsigned long start_i
;
5189 int locked_pages
= 0;
5190 int all_uptodate
= 1;
5191 unsigned long num_pages
;
5192 unsigned long num_reads
= 0;
5193 struct bio
*bio
= NULL
;
5194 unsigned long bio_flags
= 0;
5196 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5200 WARN_ON(start
< eb
->start
);
5201 start_i
= (start
>> PAGE_SHIFT
) -
5202 (eb
->start
>> PAGE_SHIFT
);
5207 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5208 for (i
= start_i
; i
< num_pages
; i
++) {
5209 page
= eb
->pages
[i
];
5210 if (wait
== WAIT_NONE
) {
5211 if (!trylock_page(page
))
5217 if (!PageUptodate(page
)) {
5224 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5228 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5229 eb
->read_mirror
= 0;
5230 atomic_set(&eb
->io_pages
, num_reads
);
5231 for (i
= start_i
; i
< num_pages
; i
++) {
5232 page
= eb
->pages
[i
];
5233 if (!PageUptodate(page
)) {
5234 ClearPageError(page
);
5235 err
= __extent_read_full_page(tree
, page
,
5237 mirror_num
, &bio_flags
,
5247 err
= submit_one_bio(bio
, mirror_num
, bio_flags
);
5252 if (ret
|| wait
!= WAIT_COMPLETE
)
5255 for (i
= start_i
; i
< num_pages
; i
++) {
5256 page
= eb
->pages
[i
];
5257 wait_on_page_locked(page
);
5258 if (!PageUptodate(page
))
5266 while (locked_pages
> 0) {
5267 page
= eb
->pages
[i
];
5275 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5276 unsigned long start
,
5283 char *dst
= (char *)dstv
;
5284 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5285 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5287 WARN_ON(start
> eb
->len
);
5288 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5290 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5293 page
= eb
->pages
[i
];
5295 cur
= min(len
, (PAGE_SIZE
- offset
));
5296 kaddr
= page_address(page
);
5297 memcpy(dst
, kaddr
+ offset
, cur
);
5306 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5307 unsigned long start
,
5314 char __user
*dst
= (char __user
*)dstv
;
5315 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5316 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5319 WARN_ON(start
> eb
->len
);
5320 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5322 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5325 page
= eb
->pages
[i
];
5327 cur
= min(len
, (PAGE_SIZE
- offset
));
5328 kaddr
= page_address(page
);
5329 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5344 * return 0 if the item is found within a page.
5345 * return 1 if the item spans two pages.
5346 * return -EINVAL otherwise.
5348 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5349 unsigned long min_len
, char **map
,
5350 unsigned long *map_start
,
5351 unsigned long *map_len
)
5353 size_t offset
= start
& (PAGE_SIZE
- 1);
5356 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5357 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5358 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5365 offset
= start_offset
;
5369 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5372 if (start
+ min_len
> eb
->len
) {
5373 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
5375 eb
->start
, eb
->len
, start
, min_len
);
5380 kaddr
= page_address(p
);
5381 *map
= kaddr
+ offset
;
5382 *map_len
= PAGE_SIZE
- offset
;
5386 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5387 unsigned long start
,
5394 char *ptr
= (char *)ptrv
;
5395 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5396 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5399 WARN_ON(start
> eb
->len
);
5400 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5402 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5405 page
= eb
->pages
[i
];
5407 cur
= min(len
, (PAGE_SIZE
- offset
));
5409 kaddr
= page_address(page
);
5410 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5422 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5423 unsigned long start
, unsigned long len
)
5429 char *src
= (char *)srcv
;
5430 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5431 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5433 WARN_ON(start
> eb
->len
);
5434 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5436 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5439 page
= eb
->pages
[i
];
5440 WARN_ON(!PageUptodate(page
));
5442 cur
= min(len
, PAGE_SIZE
- offset
);
5443 kaddr
= page_address(page
);
5444 memcpy(kaddr
+ offset
, src
, cur
);
5453 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5454 unsigned long start
, unsigned long len
)
5460 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5461 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5463 WARN_ON(start
> eb
->len
);
5464 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5466 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5469 page
= eb
->pages
[i
];
5470 WARN_ON(!PageUptodate(page
));
5472 cur
= min(len
, PAGE_SIZE
- offset
);
5473 kaddr
= page_address(page
);
5474 memset(kaddr
+ offset
, c
, cur
);
5482 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5483 unsigned long dst_offset
, unsigned long src_offset
,
5486 u64 dst_len
= dst
->len
;
5491 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5492 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5494 WARN_ON(src
->len
!= dst_len
);
5496 offset
= (start_offset
+ dst_offset
) &
5500 page
= dst
->pages
[i
];
5501 WARN_ON(!PageUptodate(page
));
5503 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5505 kaddr
= page_address(page
);
5506 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5516 * The extent buffer bitmap operations are done with byte granularity because
5517 * bitmap items are not guaranteed to be aligned to a word and therefore a
5518 * single word in a bitmap may straddle two pages in the extent buffer.
5520 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
5521 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
5522 #define BITMAP_FIRST_BYTE_MASK(start) \
5523 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
5524 #define BITMAP_LAST_BYTE_MASK(nbits) \
5525 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
5528 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5530 * @eb: the extent buffer
5531 * @start: offset of the bitmap item in the extent buffer
5533 * @page_index: return index of the page in the extent buffer that contains the
5535 * @page_offset: return offset into the page given by page_index
5537 * This helper hides the ugliness of finding the byte in an extent buffer which
5538 * contains a given bit.
5540 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5541 unsigned long start
, unsigned long nr
,
5542 unsigned long *page_index
,
5543 size_t *page_offset
)
5545 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5546 size_t byte_offset
= BIT_BYTE(nr
);
5550 * The byte we want is the offset of the extent buffer + the offset of
5551 * the bitmap item in the extent buffer + the offset of the byte in the
5554 offset
= start_offset
+ start
+ byte_offset
;
5556 *page_index
= offset
>> PAGE_SHIFT
;
5557 *page_offset
= offset
& (PAGE_SIZE
- 1);
5561 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5562 * @eb: the extent buffer
5563 * @start: offset of the bitmap item in the extent buffer
5564 * @nr: bit number to test
5566 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5574 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5575 page
= eb
->pages
[i
];
5576 WARN_ON(!PageUptodate(page
));
5577 kaddr
= page_address(page
);
5578 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5582 * extent_buffer_bitmap_set - set an area of a bitmap
5583 * @eb: the extent buffer
5584 * @start: offset of the bitmap item in the extent buffer
5585 * @pos: bit number of the first bit
5586 * @len: number of bits to set
5588 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5589 unsigned long pos
, unsigned long len
)
5595 const unsigned int size
= pos
+ len
;
5596 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5597 unsigned int mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5599 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5600 page
= eb
->pages
[i
];
5601 WARN_ON(!PageUptodate(page
));
5602 kaddr
= page_address(page
);
5604 while (len
>= bits_to_set
) {
5605 kaddr
[offset
] |= mask_to_set
;
5607 bits_to_set
= BITS_PER_BYTE
;
5609 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5611 page
= eb
->pages
[++i
];
5612 WARN_ON(!PageUptodate(page
));
5613 kaddr
= page_address(page
);
5617 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5618 kaddr
[offset
] |= mask_to_set
;
5624 * extent_buffer_bitmap_clear - clear an area of a bitmap
5625 * @eb: the extent buffer
5626 * @start: offset of the bitmap item in the extent buffer
5627 * @pos: bit number of the first bit
5628 * @len: number of bits to clear
5630 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5631 unsigned long pos
, unsigned long len
)
5637 const unsigned int size
= pos
+ len
;
5638 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5639 unsigned int mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5641 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5642 page
= eb
->pages
[i
];
5643 WARN_ON(!PageUptodate(page
));
5644 kaddr
= page_address(page
);
5646 while (len
>= bits_to_clear
) {
5647 kaddr
[offset
] &= ~mask_to_clear
;
5648 len
-= bits_to_clear
;
5649 bits_to_clear
= BITS_PER_BYTE
;
5650 mask_to_clear
= ~0U;
5651 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5653 page
= eb
->pages
[++i
];
5654 WARN_ON(!PageUptodate(page
));
5655 kaddr
= page_address(page
);
5659 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5660 kaddr
[offset
] &= ~mask_to_clear
;
5664 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5666 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5667 return distance
< len
;
5670 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5671 unsigned long dst_off
, unsigned long src_off
,
5674 char *dst_kaddr
= page_address(dst_page
);
5676 int must_memmove
= 0;
5678 if (dst_page
!= src_page
) {
5679 src_kaddr
= page_address(src_page
);
5681 src_kaddr
= dst_kaddr
;
5682 if (areas_overlap(src_off
, dst_off
, len
))
5687 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5689 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5692 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5693 unsigned long src_offset
, unsigned long len
)
5696 size_t dst_off_in_page
;
5697 size_t src_off_in_page
;
5698 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5699 unsigned long dst_i
;
5700 unsigned long src_i
;
5702 if (src_offset
+ len
> dst
->len
) {
5703 btrfs_err(dst
->fs_info
,
5704 "memmove bogus src_offset %lu move "
5705 "len %lu dst len %lu", src_offset
, len
, dst
->len
);
5708 if (dst_offset
+ len
> dst
->len
) {
5709 btrfs_err(dst
->fs_info
,
5710 "memmove bogus dst_offset %lu move "
5711 "len %lu dst len %lu", dst_offset
, len
, dst
->len
);
5716 dst_off_in_page
= (start_offset
+ dst_offset
) &
5718 src_off_in_page
= (start_offset
+ src_offset
) &
5721 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5722 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5724 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5726 cur
= min_t(unsigned long, cur
,
5727 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5729 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5730 dst_off_in_page
, src_off_in_page
, cur
);
5738 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5739 unsigned long src_offset
, unsigned long len
)
5742 size_t dst_off_in_page
;
5743 size_t src_off_in_page
;
5744 unsigned long dst_end
= dst_offset
+ len
- 1;
5745 unsigned long src_end
= src_offset
+ len
- 1;
5746 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5747 unsigned long dst_i
;
5748 unsigned long src_i
;
5750 if (src_offset
+ len
> dst
->len
) {
5751 btrfs_err(dst
->fs_info
, "memmove bogus src_offset %lu move "
5752 "len %lu len %lu", src_offset
, len
, dst
->len
);
5755 if (dst_offset
+ len
> dst
->len
) {
5756 btrfs_err(dst
->fs_info
, "memmove bogus dst_offset %lu move "
5757 "len %lu len %lu", dst_offset
, len
, dst
->len
);
5760 if (dst_offset
< src_offset
) {
5761 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5765 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5766 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5768 dst_off_in_page
= (start_offset
+ dst_end
) &
5770 src_off_in_page
= (start_offset
+ src_end
) &
5773 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5774 cur
= min(cur
, dst_off_in_page
+ 1);
5775 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5776 dst_off_in_page
- cur
+ 1,
5777 src_off_in_page
- cur
+ 1, cur
);
5785 int try_release_extent_buffer(struct page
*page
)
5787 struct extent_buffer
*eb
;
5790 * We need to make sure nobody is attaching this page to an eb right
5793 spin_lock(&page
->mapping
->private_lock
);
5794 if (!PagePrivate(page
)) {
5795 spin_unlock(&page
->mapping
->private_lock
);
5799 eb
= (struct extent_buffer
*)page
->private;
5803 * This is a little awful but should be ok, we need to make sure that
5804 * the eb doesn't disappear out from under us while we're looking at
5807 spin_lock(&eb
->refs_lock
);
5808 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5809 spin_unlock(&eb
->refs_lock
);
5810 spin_unlock(&page
->mapping
->private_lock
);
5813 spin_unlock(&page
->mapping
->private_lock
);
5816 * If tree ref isn't set then we know the ref on this eb is a real ref,
5817 * so just return, this page will likely be freed soon anyway.
5819 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5820 spin_unlock(&eb
->refs_lock
);
5824 return release_extent_buffer(eb
);