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
;
2028 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2029 &map_length
, &bbio
, mirror_num
);
2034 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2035 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2036 bio
->bi_iter
.bi_sector
= sector
;
2037 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2038 btrfs_put_bbio(bbio
);
2039 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2043 bio
->bi_bdev
= dev
->bdev
;
2044 bio_add_page(bio
, page
, length
, pg_offset
);
2046 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
)) {
2047 /* try to remap that extent elsewhere? */
2049 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2053 btrfs_info_rl_in_rcu(fs_info
,
2054 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2055 btrfs_ino(inode
), start
,
2056 rcu_str_deref(dev
->name
), sector
);
2061 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2064 u64 start
= eb
->start
;
2065 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2068 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2071 for (i
= 0; i
< num_pages
; i
++) {
2072 struct page
*p
= eb
->pages
[i
];
2074 ret
= repair_io_failure(root
->fs_info
->btree_inode
, start
,
2075 PAGE_SIZE
, start
, p
,
2076 start
- page_offset(p
), mirror_num
);
2086 * each time an IO finishes, we do a fast check in the IO failure tree
2087 * to see if we need to process or clean up an io_failure_record
2089 int clean_io_failure(struct inode
*inode
, u64 start
, struct page
*page
,
2090 unsigned int pg_offset
)
2093 struct io_failure_record
*failrec
;
2094 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2095 struct extent_state
*state
;
2100 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2101 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2105 ret
= get_state_failrec(&BTRFS_I(inode
)->io_failure_tree
, start
,
2110 BUG_ON(!failrec
->this_mirror
);
2112 if (failrec
->in_validation
) {
2113 /* there was no real error, just free the record */
2114 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2118 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2121 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2122 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2125 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2127 if (state
&& state
->start
<= failrec
->start
&&
2128 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2129 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2131 if (num_copies
> 1) {
2132 repair_io_failure(inode
, start
, failrec
->len
,
2133 failrec
->logical
, page
,
2134 pg_offset
, failrec
->failed_mirror
);
2139 free_io_failure(inode
, failrec
);
2145 * Can be called when
2146 * - hold extent lock
2147 * - under ordered extent
2148 * - the inode is freeing
2150 void btrfs_free_io_failure_record(struct inode
*inode
, u64 start
, u64 end
)
2152 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2153 struct io_failure_record
*failrec
;
2154 struct extent_state
*state
, *next
;
2156 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2159 spin_lock(&failure_tree
->lock
);
2160 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2162 if (state
->start
> end
)
2165 ASSERT(state
->end
<= end
);
2167 next
= next_state(state
);
2169 failrec
= state
->failrec
;
2170 free_extent_state(state
);
2175 spin_unlock(&failure_tree
->lock
);
2178 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2179 struct io_failure_record
**failrec_ret
)
2181 struct io_failure_record
*failrec
;
2182 struct extent_map
*em
;
2183 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2184 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2185 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2189 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2191 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2195 failrec
->start
= start
;
2196 failrec
->len
= end
- start
+ 1;
2197 failrec
->this_mirror
= 0;
2198 failrec
->bio_flags
= 0;
2199 failrec
->in_validation
= 0;
2201 read_lock(&em_tree
->lock
);
2202 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2204 read_unlock(&em_tree
->lock
);
2209 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2210 free_extent_map(em
);
2213 read_unlock(&em_tree
->lock
);
2219 logical
= start
- em
->start
;
2220 logical
= em
->block_start
+ logical
;
2221 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2222 logical
= em
->block_start
;
2223 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2224 extent_set_compress_type(&failrec
->bio_flags
,
2228 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2229 logical
, start
, failrec
->len
);
2231 failrec
->logical
= logical
;
2232 free_extent_map(em
);
2234 /* set the bits in the private failure tree */
2235 ret
= set_extent_bits(failure_tree
, start
, end
,
2236 EXTENT_LOCKED
| EXTENT_DIRTY
);
2238 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2239 /* set the bits in the inode's tree */
2241 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2247 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2248 failrec
->logical
, failrec
->start
, failrec
->len
,
2249 failrec
->in_validation
);
2251 * when data can be on disk more than twice, add to failrec here
2252 * (e.g. with a list for failed_mirror) to make
2253 * clean_io_failure() clean all those errors at once.
2257 *failrec_ret
= failrec
;
2262 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2263 struct io_failure_record
*failrec
, int failed_mirror
)
2267 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2268 failrec
->logical
, failrec
->len
);
2269 if (num_copies
== 1) {
2271 * we only have a single copy of the data, so don't bother with
2272 * all the retry and error correction code that follows. no
2273 * matter what the error is, it is very likely to persist.
2275 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2276 num_copies
, failrec
->this_mirror
, failed_mirror
);
2281 * there are two premises:
2282 * a) deliver good data to the caller
2283 * b) correct the bad sectors on disk
2285 if (failed_bio
->bi_vcnt
> 1) {
2287 * to fulfill b), we need to know the exact failing sectors, as
2288 * we don't want to rewrite any more than the failed ones. thus,
2289 * we need separate read requests for the failed bio
2291 * if the following BUG_ON triggers, our validation request got
2292 * merged. we need separate requests for our algorithm to work.
2294 BUG_ON(failrec
->in_validation
);
2295 failrec
->in_validation
= 1;
2296 failrec
->this_mirror
= failed_mirror
;
2299 * we're ready to fulfill a) and b) alongside. get a good copy
2300 * of the failed sector and if we succeed, we have setup
2301 * everything for repair_io_failure to do the rest for us.
2303 if (failrec
->in_validation
) {
2304 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2305 failrec
->in_validation
= 0;
2306 failrec
->this_mirror
= 0;
2308 failrec
->failed_mirror
= failed_mirror
;
2309 failrec
->this_mirror
++;
2310 if (failrec
->this_mirror
== failed_mirror
)
2311 failrec
->this_mirror
++;
2314 if (failrec
->this_mirror
> num_copies
) {
2315 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2316 num_copies
, failrec
->this_mirror
, failed_mirror
);
2324 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2325 struct io_failure_record
*failrec
,
2326 struct page
*page
, int pg_offset
, int icsum
,
2327 bio_end_io_t
*endio_func
, void *data
)
2330 struct btrfs_io_bio
*btrfs_failed_bio
;
2331 struct btrfs_io_bio
*btrfs_bio
;
2333 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2337 bio
->bi_end_io
= endio_func
;
2338 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2339 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2340 bio
->bi_iter
.bi_size
= 0;
2341 bio
->bi_private
= data
;
2343 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2344 if (btrfs_failed_bio
->csum
) {
2345 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2346 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2348 btrfs_bio
= btrfs_io_bio(bio
);
2349 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2351 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2355 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2361 * this is a generic handler for readpage errors (default
2362 * readpage_io_failed_hook). if other copies exist, read those and write back
2363 * good data to the failed position. does not investigate in remapping the
2364 * failed extent elsewhere, hoping the device will be smart enough to do this as
2368 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2369 struct page
*page
, u64 start
, u64 end
,
2372 struct io_failure_record
*failrec
;
2373 struct inode
*inode
= page
->mapping
->host
;
2374 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2379 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2381 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2385 ret
= btrfs_check_repairable(inode
, failed_bio
, failrec
, failed_mirror
);
2387 free_io_failure(inode
, failrec
);
2391 if (failed_bio
->bi_vcnt
> 1)
2392 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2394 read_mode
= READ_SYNC
;
2396 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2397 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2398 start
- page_offset(page
),
2399 (int)phy_offset
, failed_bio
->bi_end_io
,
2402 free_io_failure(inode
, failrec
);
2406 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2407 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2409 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2410 failrec
->this_mirror
,
2411 failrec
->bio_flags
, 0);
2413 free_io_failure(inode
, failrec
);
2420 /* lots and lots of room for performance fixes in the end_bio funcs */
2422 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2424 int uptodate
= (err
== 0);
2425 struct extent_io_tree
*tree
;
2428 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2430 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2431 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2432 end
, NULL
, uptodate
);
2438 ClearPageUptodate(page
);
2440 ret
= ret
< 0 ? ret
: -EIO
;
2441 mapping_set_error(page
->mapping
, ret
);
2446 * after a writepage IO is done, we need to:
2447 * clear the uptodate bits on error
2448 * clear the writeback bits in the extent tree for this IO
2449 * end_page_writeback if the page has no more pending IO
2451 * Scheduling is not allowed, so the extent state tree is expected
2452 * to have one and only one object corresponding to this IO.
2454 static void end_bio_extent_writepage(struct bio
*bio
)
2456 struct bio_vec
*bvec
;
2461 bio_for_each_segment_all(bvec
, bio
, i
) {
2462 struct page
*page
= bvec
->bv_page
;
2464 /* We always issue full-page reads, but if some block
2465 * in a page fails to read, blk_update_request() will
2466 * advance bv_offset and adjust bv_len to compensate.
2467 * Print a warning for nonzero offsets, and an error
2468 * if they don't add up to a full page. */
2469 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2470 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2471 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2472 "partial page write in btrfs with offset %u and length %u",
2473 bvec
->bv_offset
, bvec
->bv_len
);
2475 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2476 "incomplete page write in btrfs with offset %u and "
2478 bvec
->bv_offset
, bvec
->bv_len
);
2481 start
= page_offset(page
);
2482 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2484 end_extent_writepage(page
, bio
->bi_error
, start
, end
);
2485 end_page_writeback(page
);
2492 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2495 struct extent_state
*cached
= NULL
;
2496 u64 end
= start
+ len
- 1;
2498 if (uptodate
&& tree
->track_uptodate
)
2499 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2500 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2504 * after a readpage IO is done, we need to:
2505 * clear the uptodate bits on error
2506 * set the uptodate bits if things worked
2507 * set the page up to date if all extents in the tree are uptodate
2508 * clear the lock bit in the extent tree
2509 * unlock the page if there are no other extents locked for it
2511 * Scheduling is not allowed, so the extent state tree is expected
2512 * to have one and only one object corresponding to this IO.
2514 static void end_bio_extent_readpage(struct bio
*bio
)
2516 struct bio_vec
*bvec
;
2517 int uptodate
= !bio
->bi_error
;
2518 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2519 struct extent_io_tree
*tree
;
2524 u64 extent_start
= 0;
2530 bio_for_each_segment_all(bvec
, bio
, i
) {
2531 struct page
*page
= bvec
->bv_page
;
2532 struct inode
*inode
= page
->mapping
->host
;
2534 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2535 "mirror=%u\n", (u64
)bio
->bi_iter
.bi_sector
,
2536 bio
->bi_error
, io_bio
->mirror_num
);
2537 tree
= &BTRFS_I(inode
)->io_tree
;
2539 /* We always issue full-page reads, but if some block
2540 * in a page fails to read, blk_update_request() will
2541 * advance bv_offset and adjust bv_len to compensate.
2542 * Print a warning for nonzero offsets, and an error
2543 * if they don't add up to a full page. */
2544 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2545 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2546 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2547 "partial page read in btrfs with offset %u and length %u",
2548 bvec
->bv_offset
, bvec
->bv_len
);
2550 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2551 "incomplete page read in btrfs with offset %u and "
2553 bvec
->bv_offset
, bvec
->bv_len
);
2556 start
= page_offset(page
);
2557 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2560 mirror
= io_bio
->mirror_num
;
2561 if (likely(uptodate
&& tree
->ops
&&
2562 tree
->ops
->readpage_end_io_hook
)) {
2563 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2569 clean_io_failure(inode
, start
, page
, 0);
2572 if (likely(uptodate
))
2575 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2576 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2577 if (!ret
&& !bio
->bi_error
)
2581 * The generic bio_readpage_error handles errors the
2582 * following way: If possible, new read requests are
2583 * created and submitted and will end up in
2584 * end_bio_extent_readpage as well (if we're lucky, not
2585 * in the !uptodate case). In that case it returns 0 and
2586 * we just go on with the next page in our bio. If it
2587 * can't handle the error it will return -EIO and we
2588 * remain responsible for that page.
2590 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2593 uptodate
= !bio
->bi_error
;
2599 if (likely(uptodate
)) {
2600 loff_t i_size
= i_size_read(inode
);
2601 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2604 /* Zero out the end if this page straddles i_size */
2605 off
= i_size
& (PAGE_SIZE
-1);
2606 if (page
->index
== end_index
&& off
)
2607 zero_user_segment(page
, off
, PAGE_SIZE
);
2608 SetPageUptodate(page
);
2610 ClearPageUptodate(page
);
2616 if (unlikely(!uptodate
)) {
2618 endio_readpage_release_extent(tree
,
2624 endio_readpage_release_extent(tree
, start
,
2625 end
- start
+ 1, 0);
2626 } else if (!extent_len
) {
2627 extent_start
= start
;
2628 extent_len
= end
+ 1 - start
;
2629 } else if (extent_start
+ extent_len
== start
) {
2630 extent_len
+= end
+ 1 - start
;
2632 endio_readpage_release_extent(tree
, extent_start
,
2633 extent_len
, uptodate
);
2634 extent_start
= start
;
2635 extent_len
= end
+ 1 - start
;
2640 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2643 io_bio
->end_io(io_bio
, bio
->bi_error
);
2648 * this allocates from the btrfs_bioset. We're returning a bio right now
2649 * but you can call btrfs_io_bio for the appropriate container_of magic
2652 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2655 struct btrfs_io_bio
*btrfs_bio
;
2658 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2660 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2661 while (!bio
&& (nr_vecs
/= 2)) {
2662 bio
= bio_alloc_bioset(gfp_flags
,
2663 nr_vecs
, btrfs_bioset
);
2668 bio
->bi_bdev
= bdev
;
2669 bio
->bi_iter
.bi_sector
= first_sector
;
2670 btrfs_bio
= btrfs_io_bio(bio
);
2671 btrfs_bio
->csum
= NULL
;
2672 btrfs_bio
->csum_allocated
= NULL
;
2673 btrfs_bio
->end_io
= NULL
;
2678 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2680 struct btrfs_io_bio
*btrfs_bio
;
2683 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2685 btrfs_bio
= btrfs_io_bio(new);
2686 btrfs_bio
->csum
= NULL
;
2687 btrfs_bio
->csum_allocated
= NULL
;
2688 btrfs_bio
->end_io
= NULL
;
2690 #ifdef CONFIG_BLK_CGROUP
2691 /* FIXME, put this into bio_clone_bioset */
2693 bio_associate_blkcg(new, bio
->bi_css
);
2699 /* this also allocates from the btrfs_bioset */
2700 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2702 struct btrfs_io_bio
*btrfs_bio
;
2705 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2707 btrfs_bio
= btrfs_io_bio(bio
);
2708 btrfs_bio
->csum
= NULL
;
2709 btrfs_bio
->csum_allocated
= NULL
;
2710 btrfs_bio
->end_io
= NULL
;
2716 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2717 int mirror_num
, unsigned long bio_flags
)
2720 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2721 struct page
*page
= bvec
->bv_page
;
2722 struct extent_io_tree
*tree
= bio
->bi_private
;
2725 start
= page_offset(page
) + bvec
->bv_offset
;
2727 bio
->bi_private
= NULL
;
2731 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2732 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2733 mirror_num
, bio_flags
, start
);
2735 btrfsic_submit_bio(rw
, bio
);
2741 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2742 unsigned long offset
, size_t size
, struct bio
*bio
,
2743 unsigned long bio_flags
)
2746 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2747 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2754 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2755 struct writeback_control
*wbc
,
2756 struct page
*page
, sector_t sector
,
2757 size_t size
, unsigned long offset
,
2758 struct block_device
*bdev
,
2759 struct bio
**bio_ret
,
2760 unsigned long max_pages
,
2761 bio_end_io_t end_io_func
,
2763 unsigned long prev_bio_flags
,
2764 unsigned long bio_flags
,
2765 bool force_bio_submit
)
2770 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2771 size_t page_size
= min_t(size_t, size
, PAGE_SIZE
);
2773 if (bio_ret
&& *bio_ret
) {
2776 contig
= bio
->bi_iter
.bi_sector
== sector
;
2778 contig
= bio_end_sector(bio
) == sector
;
2780 if (prev_bio_flags
!= bio_flags
|| !contig
||
2782 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2783 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2784 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2793 wbc_account_io(wbc
, page
, page_size
);
2798 bio
= btrfs_bio_alloc(bdev
, sector
, BIO_MAX_PAGES
,
2799 GFP_NOFS
| __GFP_HIGH
);
2803 bio_add_page(bio
, page
, page_size
, offset
);
2804 bio
->bi_end_io
= end_io_func
;
2805 bio
->bi_private
= tree
;
2807 wbc_init_bio(wbc
, bio
);
2808 wbc_account_io(wbc
, page
, page_size
);
2814 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2819 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2822 if (!PagePrivate(page
)) {
2823 SetPagePrivate(page
);
2825 set_page_private(page
, (unsigned long)eb
);
2827 WARN_ON(page
->private != (unsigned long)eb
);
2831 void set_page_extent_mapped(struct page
*page
)
2833 if (!PagePrivate(page
)) {
2834 SetPagePrivate(page
);
2836 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2840 static struct extent_map
*
2841 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2842 u64 start
, u64 len
, get_extent_t
*get_extent
,
2843 struct extent_map
**em_cached
)
2845 struct extent_map
*em
;
2847 if (em_cached
&& *em_cached
) {
2849 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2850 start
< extent_map_end(em
)) {
2851 atomic_inc(&em
->refs
);
2855 free_extent_map(em
);
2859 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2860 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2862 atomic_inc(&em
->refs
);
2868 * basic readpage implementation. Locked extent state structs are inserted
2869 * into the tree that are removed when the IO is done (by the end_io
2871 * XXX JDM: This needs looking at to ensure proper page locking
2873 static int __do_readpage(struct extent_io_tree
*tree
,
2875 get_extent_t
*get_extent
,
2876 struct extent_map
**em_cached
,
2877 struct bio
**bio
, int mirror_num
,
2878 unsigned long *bio_flags
, int rw
,
2881 struct inode
*inode
= page
->mapping
->host
;
2882 u64 start
= page_offset(page
);
2883 u64 page_end
= start
+ PAGE_SIZE
- 1;
2887 u64 last_byte
= i_size_read(inode
);
2891 struct extent_map
*em
;
2892 struct block_device
*bdev
;
2895 size_t pg_offset
= 0;
2897 size_t disk_io_size
;
2898 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2899 unsigned long this_bio_flag
= 0;
2901 set_page_extent_mapped(page
);
2904 if (!PageUptodate(page
)) {
2905 if (cleancache_get_page(page
) == 0) {
2906 BUG_ON(blocksize
!= PAGE_SIZE
);
2907 unlock_extent(tree
, start
, end
);
2912 if (page
->index
== last_byte
>> PAGE_SHIFT
) {
2914 size_t zero_offset
= last_byte
& (PAGE_SIZE
- 1);
2917 iosize
= PAGE_SIZE
- zero_offset
;
2918 userpage
= kmap_atomic(page
);
2919 memset(userpage
+ zero_offset
, 0, iosize
);
2920 flush_dcache_page(page
);
2921 kunmap_atomic(userpage
);
2924 while (cur
<= end
) {
2925 unsigned long pnr
= (last_byte
>> PAGE_SHIFT
) + 1;
2926 bool force_bio_submit
= false;
2928 if (cur
>= last_byte
) {
2930 struct extent_state
*cached
= NULL
;
2932 iosize
= PAGE_SIZE
- pg_offset
;
2933 userpage
= kmap_atomic(page
);
2934 memset(userpage
+ pg_offset
, 0, iosize
);
2935 flush_dcache_page(page
);
2936 kunmap_atomic(userpage
);
2937 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2939 unlock_extent_cached(tree
, cur
,
2944 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2945 end
- cur
+ 1, get_extent
, em_cached
);
2946 if (IS_ERR_OR_NULL(em
)) {
2948 unlock_extent(tree
, cur
, end
);
2951 extent_offset
= cur
- em
->start
;
2952 BUG_ON(extent_map_end(em
) <= cur
);
2955 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2956 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2957 extent_set_compress_type(&this_bio_flag
,
2961 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2962 cur_end
= min(extent_map_end(em
) - 1, end
);
2963 iosize
= ALIGN(iosize
, blocksize
);
2964 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2965 disk_io_size
= em
->block_len
;
2966 sector
= em
->block_start
>> 9;
2968 sector
= (em
->block_start
+ extent_offset
) >> 9;
2969 disk_io_size
= iosize
;
2972 block_start
= em
->block_start
;
2973 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2974 block_start
= EXTENT_MAP_HOLE
;
2977 * If we have a file range that points to a compressed extent
2978 * and it's followed by a consecutive file range that points to
2979 * to the same compressed extent (possibly with a different
2980 * offset and/or length, so it either points to the whole extent
2981 * or only part of it), we must make sure we do not submit a
2982 * single bio to populate the pages for the 2 ranges because
2983 * this makes the compressed extent read zero out the pages
2984 * belonging to the 2nd range. Imagine the following scenario:
2987 * [0 - 8K] [8K - 24K]
2990 * points to extent X, points to extent X,
2991 * offset 4K, length of 8K offset 0, length 16K
2993 * [extent X, compressed length = 4K uncompressed length = 16K]
2995 * If the bio to read the compressed extent covers both ranges,
2996 * it will decompress extent X into the pages belonging to the
2997 * first range and then it will stop, zeroing out the remaining
2998 * pages that belong to the other range that points to extent X.
2999 * So here we make sure we submit 2 bios, one for the first
3000 * range and another one for the third range. Both will target
3001 * the same physical extent from disk, but we can't currently
3002 * make the compressed bio endio callback populate the pages
3003 * for both ranges because each compressed bio is tightly
3004 * coupled with a single extent map, and each range can have
3005 * an extent map with a different offset value relative to the
3006 * uncompressed data of our extent and different lengths. This
3007 * is a corner case so we prioritize correctness over
3008 * non-optimal behavior (submitting 2 bios for the same extent).
3010 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3011 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3012 *prev_em_start
!= em
->orig_start
)
3013 force_bio_submit
= true;
3016 *prev_em_start
= em
->orig_start
;
3018 free_extent_map(em
);
3021 /* we've found a hole, just zero and go on */
3022 if (block_start
== EXTENT_MAP_HOLE
) {
3024 struct extent_state
*cached
= NULL
;
3026 userpage
= kmap_atomic(page
);
3027 memset(userpage
+ pg_offset
, 0, iosize
);
3028 flush_dcache_page(page
);
3029 kunmap_atomic(userpage
);
3031 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3033 unlock_extent_cached(tree
, cur
,
3037 pg_offset
+= iosize
;
3040 /* the get_extent function already copied into the page */
3041 if (test_range_bit(tree
, cur
, cur_end
,
3042 EXTENT_UPTODATE
, 1, NULL
)) {
3043 check_page_uptodate(tree
, page
);
3044 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3046 pg_offset
+= iosize
;
3049 /* we have an inline extent but it didn't get marked up
3050 * to date. Error out
3052 if (block_start
== EXTENT_MAP_INLINE
) {
3054 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3056 pg_offset
+= iosize
;
3061 ret
= submit_extent_page(rw
, tree
, NULL
, page
,
3062 sector
, disk_io_size
, pg_offset
,
3064 end_bio_extent_readpage
, mirror_num
,
3070 *bio_flags
= this_bio_flag
;
3073 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3076 pg_offset
+= iosize
;
3080 if (!PageError(page
))
3081 SetPageUptodate(page
);
3087 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3088 struct page
*pages
[], int nr_pages
,
3090 get_extent_t
*get_extent
,
3091 struct extent_map
**em_cached
,
3092 struct bio
**bio
, int mirror_num
,
3093 unsigned long *bio_flags
, int rw
,
3096 struct inode
*inode
;
3097 struct btrfs_ordered_extent
*ordered
;
3100 inode
= pages
[0]->mapping
->host
;
3102 lock_extent(tree
, start
, end
);
3103 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3107 unlock_extent(tree
, start
, end
);
3108 btrfs_start_ordered_extent(inode
, ordered
, 1);
3109 btrfs_put_ordered_extent(ordered
);
3112 for (index
= 0; index
< nr_pages
; index
++) {
3113 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3114 mirror_num
, bio_flags
, rw
, prev_em_start
);
3115 put_page(pages
[index
]);
3119 static void __extent_readpages(struct extent_io_tree
*tree
,
3120 struct page
*pages
[],
3121 int nr_pages
, get_extent_t
*get_extent
,
3122 struct extent_map
**em_cached
,
3123 struct bio
**bio
, int mirror_num
,
3124 unsigned long *bio_flags
, int rw
,
3131 int first_index
= 0;
3133 for (index
= 0; index
< nr_pages
; index
++) {
3134 page_start
= page_offset(pages
[index
]);
3137 end
= start
+ PAGE_SIZE
- 1;
3138 first_index
= index
;
3139 } else if (end
+ 1 == page_start
) {
3142 __do_contiguous_readpages(tree
, &pages
[first_index
],
3143 index
- first_index
, start
,
3144 end
, get_extent
, em_cached
,
3145 bio
, mirror_num
, bio_flags
,
3148 end
= start
+ PAGE_SIZE
- 1;
3149 first_index
= index
;
3154 __do_contiguous_readpages(tree
, &pages
[first_index
],
3155 index
- first_index
, start
,
3156 end
, get_extent
, em_cached
, bio
,
3157 mirror_num
, bio_flags
, rw
,
3161 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3163 get_extent_t
*get_extent
,
3164 struct bio
**bio
, int mirror_num
,
3165 unsigned long *bio_flags
, int rw
)
3167 struct inode
*inode
= page
->mapping
->host
;
3168 struct btrfs_ordered_extent
*ordered
;
3169 u64 start
= page_offset(page
);
3170 u64 end
= start
+ PAGE_SIZE
- 1;
3174 lock_extent(tree
, start
, end
);
3175 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3179 unlock_extent(tree
, start
, end
);
3180 btrfs_start_ordered_extent(inode
, ordered
, 1);
3181 btrfs_put_ordered_extent(ordered
);
3184 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3185 bio_flags
, rw
, NULL
);
3189 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3190 get_extent_t
*get_extent
, int mirror_num
)
3192 struct bio
*bio
= NULL
;
3193 unsigned long bio_flags
= 0;
3196 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3199 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3203 static void update_nr_written(struct page
*page
, struct writeback_control
*wbc
,
3204 unsigned long nr_written
)
3206 wbc
->nr_to_write
-= nr_written
;
3210 * helper for __extent_writepage, doing all of the delayed allocation setup.
3212 * This returns 1 if our fill_delalloc function did all the work required
3213 * to write the page (copy into inline extent). In this case the IO has
3214 * been started and the page is already unlocked.
3216 * This returns 0 if all went well (page still locked)
3217 * This returns < 0 if there were errors (page still locked)
3219 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3220 struct page
*page
, struct writeback_control
*wbc
,
3221 struct extent_page_data
*epd
,
3223 unsigned long *nr_written
)
3225 struct extent_io_tree
*tree
= epd
->tree
;
3226 u64 page_end
= delalloc_start
+ PAGE_SIZE
- 1;
3228 u64 delalloc_to_write
= 0;
3229 u64 delalloc_end
= 0;
3231 int page_started
= 0;
3233 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3236 while (delalloc_end
< page_end
) {
3237 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3241 BTRFS_MAX_EXTENT_SIZE
);
3242 if (nr_delalloc
== 0) {
3243 delalloc_start
= delalloc_end
+ 1;
3246 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3251 /* File system has been set read-only */
3254 /* fill_delalloc should be return < 0 for error
3255 * but just in case, we use > 0 here meaning the
3256 * IO is started, so we don't want to return > 0
3257 * unless things are going well.
3259 ret
= ret
< 0 ? ret
: -EIO
;
3263 * delalloc_end is already one less than the total length, so
3264 * we don't subtract one from PAGE_SIZE
3266 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3267 PAGE_SIZE
) >> PAGE_SHIFT
;
3268 delalloc_start
= delalloc_end
+ 1;
3270 if (wbc
->nr_to_write
< delalloc_to_write
) {
3273 if (delalloc_to_write
< thresh
* 2)
3274 thresh
= delalloc_to_write
;
3275 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3279 /* did the fill delalloc function already unlock and start
3284 * we've unlocked the page, so we can't update
3285 * the mapping's writeback index, just update
3288 wbc
->nr_to_write
-= *nr_written
;
3299 * helper for __extent_writepage. This calls the writepage start hooks,
3300 * and does the loop to map the page into extents and bios.
3302 * We return 1 if the IO is started and the page is unlocked,
3303 * 0 if all went well (page still locked)
3304 * < 0 if there were errors (page still locked)
3306 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3308 struct writeback_control
*wbc
,
3309 struct extent_page_data
*epd
,
3311 unsigned long nr_written
,
3312 int write_flags
, int *nr_ret
)
3314 struct extent_io_tree
*tree
= epd
->tree
;
3315 u64 start
= page_offset(page
);
3316 u64 page_end
= start
+ PAGE_SIZE
- 1;
3323 struct extent_state
*cached_state
= NULL
;
3324 struct extent_map
*em
;
3325 struct block_device
*bdev
;
3326 size_t pg_offset
= 0;
3332 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3333 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3336 /* Fixup worker will requeue */
3338 wbc
->pages_skipped
++;
3340 redirty_page_for_writepage(wbc
, page
);
3342 update_nr_written(page
, wbc
, nr_written
);
3350 * we don't want to touch the inode after unlocking the page,
3351 * so we update the mapping writeback index now
3353 update_nr_written(page
, wbc
, nr_written
+ 1);
3356 if (i_size
<= start
) {
3357 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3358 tree
->ops
->writepage_end_io_hook(page
, start
,
3363 blocksize
= inode
->i_sb
->s_blocksize
;
3365 while (cur
<= end
) {
3367 unsigned long max_nr
;
3369 if (cur
>= i_size
) {
3370 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3371 tree
->ops
->writepage_end_io_hook(page
, cur
,
3375 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3377 if (IS_ERR_OR_NULL(em
)) {
3379 ret
= PTR_ERR_OR_ZERO(em
);
3383 extent_offset
= cur
- em
->start
;
3384 em_end
= extent_map_end(em
);
3385 BUG_ON(em_end
<= cur
);
3387 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3388 iosize
= ALIGN(iosize
, blocksize
);
3389 sector
= (em
->block_start
+ extent_offset
) >> 9;
3391 block_start
= em
->block_start
;
3392 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3393 free_extent_map(em
);
3397 * compressed and inline extents are written through other
3400 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3401 block_start
== EXTENT_MAP_INLINE
) {
3403 * end_io notification does not happen here for
3404 * compressed extents
3406 if (!compressed
&& tree
->ops
&&
3407 tree
->ops
->writepage_end_io_hook
)
3408 tree
->ops
->writepage_end_io_hook(page
, cur
,
3411 else if (compressed
) {
3412 /* we don't want to end_page_writeback on
3413 * a compressed extent. this happens
3420 pg_offset
+= iosize
;
3424 max_nr
= (i_size
>> PAGE_SHIFT
) + 1;
3426 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3427 if (!PageWriteback(page
)) {
3428 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3429 "page %lu not writeback, cur %llu end %llu",
3430 page
->index
, cur
, end
);
3433 ret
= submit_extent_page(write_flags
, tree
, wbc
, page
,
3434 sector
, iosize
, pg_offset
,
3435 bdev
, &epd
->bio
, max_nr
,
3436 end_bio_extent_writepage
,
3442 pg_offset
+= iosize
;
3450 /* drop our reference on any cached states */
3451 free_extent_state(cached_state
);
3456 * the writepage semantics are similar to regular writepage. extent
3457 * records are inserted to lock ranges in the tree, and as dirty areas
3458 * are found, they are marked writeback. Then the lock bits are removed
3459 * and the end_io handler clears the writeback ranges
3461 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3464 struct inode
*inode
= page
->mapping
->host
;
3465 struct extent_page_data
*epd
= data
;
3466 u64 start
= page_offset(page
);
3467 u64 page_end
= start
+ PAGE_SIZE
- 1;
3470 size_t pg_offset
= 0;
3471 loff_t i_size
= i_size_read(inode
);
3472 unsigned long end_index
= i_size
>> PAGE_SHIFT
;
3474 unsigned long nr_written
= 0;
3476 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3477 write_flags
= WRITE_SYNC
;
3479 write_flags
= WRITE
;
3481 trace___extent_writepage(page
, inode
, wbc
);
3483 WARN_ON(!PageLocked(page
));
3485 ClearPageError(page
);
3487 pg_offset
= i_size
& (PAGE_SIZE
- 1);
3488 if (page
->index
> end_index
||
3489 (page
->index
== end_index
&& !pg_offset
)) {
3490 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3495 if (page
->index
== end_index
) {
3498 userpage
= kmap_atomic(page
);
3499 memset(userpage
+ pg_offset
, 0,
3500 PAGE_SIZE
- pg_offset
);
3501 kunmap_atomic(userpage
);
3502 flush_dcache_page(page
);
3507 set_page_extent_mapped(page
);
3509 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3515 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3516 i_size
, nr_written
, write_flags
, &nr
);
3522 /* make sure the mapping tag for page dirty gets cleared */
3523 set_page_writeback(page
);
3524 end_page_writeback(page
);
3526 if (PageError(page
)) {
3527 ret
= ret
< 0 ? ret
: -EIO
;
3528 end_extent_writepage(page
, ret
, start
, page_end
);
3537 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3539 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3540 TASK_UNINTERRUPTIBLE
);
3543 static noinline_for_stack
int
3544 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3545 struct btrfs_fs_info
*fs_info
,
3546 struct extent_page_data
*epd
)
3548 unsigned long i
, num_pages
;
3552 if (!btrfs_try_tree_write_lock(eb
)) {
3554 flush_write_bio(epd
);
3555 btrfs_tree_lock(eb
);
3558 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3559 btrfs_tree_unlock(eb
);
3563 flush_write_bio(epd
);
3567 wait_on_extent_buffer_writeback(eb
);
3568 btrfs_tree_lock(eb
);
3569 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3571 btrfs_tree_unlock(eb
);
3576 * We need to do this to prevent races in people who check if the eb is
3577 * under IO since we can end up having no IO bits set for a short period
3580 spin_lock(&eb
->refs_lock
);
3581 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3582 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3583 spin_unlock(&eb
->refs_lock
);
3584 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3585 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3587 fs_info
->dirty_metadata_batch
);
3590 spin_unlock(&eb
->refs_lock
);
3593 btrfs_tree_unlock(eb
);
3598 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3599 for (i
= 0; i
< num_pages
; i
++) {
3600 struct page
*p
= eb
->pages
[i
];
3602 if (!trylock_page(p
)) {
3604 flush_write_bio(epd
);
3614 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3616 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3617 smp_mb__after_atomic();
3618 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3621 static void set_btree_ioerr(struct page
*page
)
3623 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3624 struct btrfs_inode
*btree_ino
= BTRFS_I(eb
->fs_info
->btree_inode
);
3627 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3631 * If writeback for a btree extent that doesn't belong to a log tree
3632 * failed, increment the counter transaction->eb_write_errors.
3633 * We do this because while the transaction is running and before it's
3634 * committing (when we call filemap_fdata[write|wait]_range against
3635 * the btree inode), we might have
3636 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3637 * returns an error or an error happens during writeback, when we're
3638 * committing the transaction we wouldn't know about it, since the pages
3639 * can be no longer dirty nor marked anymore for writeback (if a
3640 * subsequent modification to the extent buffer didn't happen before the
3641 * transaction commit), which makes filemap_fdata[write|wait]_range not
3642 * able to find the pages tagged with SetPageError at transaction
3643 * commit time. So if this happens we must abort the transaction,
3644 * otherwise we commit a super block with btree roots that point to
3645 * btree nodes/leafs whose content on disk is invalid - either garbage
3646 * or the content of some node/leaf from a past generation that got
3647 * cowed or deleted and is no longer valid.
3649 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3650 * not be enough - we need to distinguish between log tree extents vs
3651 * non-log tree extents, and the next filemap_fdatawait_range() call
3652 * will catch and clear such errors in the mapping - and that call might
3653 * be from a log sync and not from a transaction commit. Also, checking
3654 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3655 * not done and would not be reliable - the eb might have been released
3656 * from memory and reading it back again means that flag would not be
3657 * set (since it's a runtime flag, not persisted on disk).
3659 * Using the flags below in the btree inode also makes us achieve the
3660 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3661 * writeback for all dirty pages and before filemap_fdatawait_range()
3662 * is called, the writeback for all dirty pages had already finished
3663 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3664 * filemap_fdatawait_range() would return success, as it could not know
3665 * that writeback errors happened (the pages were no longer tagged for
3668 switch (eb
->log_index
) {
3670 set_bit(BTRFS_INODE_BTREE_ERR
, &btree_ino
->runtime_flags
);
3673 set_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
3676 set_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
3679 BUG(); /* unexpected, logic error */
3683 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3685 struct bio_vec
*bvec
;
3686 struct extent_buffer
*eb
;
3689 bio_for_each_segment_all(bvec
, bio
, i
) {
3690 struct page
*page
= bvec
->bv_page
;
3692 eb
= (struct extent_buffer
*)page
->private;
3694 done
= atomic_dec_and_test(&eb
->io_pages
);
3696 if (bio
->bi_error
||
3697 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3698 ClearPageUptodate(page
);
3699 set_btree_ioerr(page
);
3702 end_page_writeback(page
);
3707 end_extent_buffer_writeback(eb
);
3713 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3714 struct btrfs_fs_info
*fs_info
,
3715 struct writeback_control
*wbc
,
3716 struct extent_page_data
*epd
)
3718 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3719 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3720 u64 offset
= eb
->start
;
3721 unsigned long i
, num_pages
;
3722 unsigned long bio_flags
= 0;
3723 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3726 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3727 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3728 atomic_set(&eb
->io_pages
, num_pages
);
3729 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3730 bio_flags
= EXTENT_BIO_TREE_LOG
;
3732 for (i
= 0; i
< num_pages
; i
++) {
3733 struct page
*p
= eb
->pages
[i
];
3735 clear_page_dirty_for_io(p
);
3736 set_page_writeback(p
);
3737 ret
= submit_extent_page(rw
, tree
, wbc
, p
, offset
>> 9,
3738 PAGE_SIZE
, 0, bdev
, &epd
->bio
,
3739 -1, end_bio_extent_buffer_writepage
,
3740 0, epd
->bio_flags
, bio_flags
, false);
3741 epd
->bio_flags
= bio_flags
;
3744 end_page_writeback(p
);
3745 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3746 end_extent_buffer_writeback(eb
);
3750 offset
+= PAGE_SIZE
;
3751 update_nr_written(p
, wbc
, 1);
3755 if (unlikely(ret
)) {
3756 for (; i
< num_pages
; i
++) {
3757 struct page
*p
= eb
->pages
[i
];
3758 clear_page_dirty_for_io(p
);
3766 int btree_write_cache_pages(struct address_space
*mapping
,
3767 struct writeback_control
*wbc
)
3769 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3770 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3771 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3772 struct extent_page_data epd
= {
3776 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3781 int nr_to_write_done
= 0;
3782 struct pagevec pvec
;
3785 pgoff_t end
; /* Inclusive */
3789 pagevec_init(&pvec
, 0);
3790 if (wbc
->range_cyclic
) {
3791 index
= mapping
->writeback_index
; /* Start from prev offset */
3794 index
= wbc
->range_start
>> PAGE_SHIFT
;
3795 end
= wbc
->range_end
>> PAGE_SHIFT
;
3798 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3799 tag
= PAGECACHE_TAG_TOWRITE
;
3801 tag
= PAGECACHE_TAG_DIRTY
;
3803 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3804 tag_pages_for_writeback(mapping
, index
, end
);
3805 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3806 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3807 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3811 for (i
= 0; i
< nr_pages
; i
++) {
3812 struct page
*page
= pvec
.pages
[i
];
3814 if (!PagePrivate(page
))
3817 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3822 spin_lock(&mapping
->private_lock
);
3823 if (!PagePrivate(page
)) {
3824 spin_unlock(&mapping
->private_lock
);
3828 eb
= (struct extent_buffer
*)page
->private;
3831 * Shouldn't happen and normally this would be a BUG_ON
3832 * but no sense in crashing the users box for something
3833 * we can survive anyway.
3836 spin_unlock(&mapping
->private_lock
);
3840 if (eb
== prev_eb
) {
3841 spin_unlock(&mapping
->private_lock
);
3845 ret
= atomic_inc_not_zero(&eb
->refs
);
3846 spin_unlock(&mapping
->private_lock
);
3851 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3853 free_extent_buffer(eb
);
3857 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3860 free_extent_buffer(eb
);
3863 free_extent_buffer(eb
);
3866 * the filesystem may choose to bump up nr_to_write.
3867 * We have to make sure to honor the new nr_to_write
3870 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3872 pagevec_release(&pvec
);
3875 if (!scanned
&& !done
) {
3877 * We hit the last page and there is more work to be done: wrap
3878 * back to the start of the file
3884 flush_write_bio(&epd
);
3889 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3890 * @mapping: address space structure to write
3891 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3892 * @writepage: function called for each page
3893 * @data: data passed to writepage function
3895 * If a page is already under I/O, write_cache_pages() skips it, even
3896 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3897 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3898 * and msync() need to guarantee that all the data which was dirty at the time
3899 * the call was made get new I/O started against them. If wbc->sync_mode is
3900 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3901 * existing IO to complete.
3903 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3904 struct address_space
*mapping
,
3905 struct writeback_control
*wbc
,
3906 writepage_t writepage
, void *data
,
3907 void (*flush_fn
)(void *))
3909 struct inode
*inode
= mapping
->host
;
3912 int nr_to_write_done
= 0;
3913 struct pagevec pvec
;
3916 pgoff_t end
; /* Inclusive */
3918 int range_whole
= 0;
3923 * We have to hold onto the inode so that ordered extents can do their
3924 * work when the IO finishes. The alternative to this is failing to add
3925 * an ordered extent if the igrab() fails there and that is a huge pain
3926 * to deal with, so instead just hold onto the inode throughout the
3927 * writepages operation. If it fails here we are freeing up the inode
3928 * anyway and we'd rather not waste our time writing out stuff that is
3929 * going to be truncated anyway.
3934 pagevec_init(&pvec
, 0);
3935 if (wbc
->range_cyclic
) {
3936 index
= mapping
->writeback_index
; /* Start from prev offset */
3939 index
= wbc
->range_start
>> PAGE_SHIFT
;
3940 end
= wbc
->range_end
>> PAGE_SHIFT
;
3941 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3945 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3946 tag
= PAGECACHE_TAG_TOWRITE
;
3948 tag
= PAGECACHE_TAG_DIRTY
;
3950 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3951 tag_pages_for_writeback(mapping
, index
, end
);
3953 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3954 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3955 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3959 for (i
= 0; i
< nr_pages
; i
++) {
3960 struct page
*page
= pvec
.pages
[i
];
3962 done_index
= page
->index
;
3964 * At this point we hold neither mapping->tree_lock nor
3965 * lock on the page itself: the page may be truncated or
3966 * invalidated (changing page->mapping to NULL), or even
3967 * swizzled back from swapper_space to tmpfs file
3970 if (!trylock_page(page
)) {
3975 if (unlikely(page
->mapping
!= mapping
)) {
3980 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3986 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3987 if (PageWriteback(page
))
3989 wait_on_page_writeback(page
);
3992 if (PageWriteback(page
) ||
3993 !clear_page_dirty_for_io(page
)) {
3998 ret
= (*writepage
)(page
, wbc
, data
);
4000 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4006 * done_index is set past this page,
4007 * so media errors will not choke
4008 * background writeout for the entire
4009 * file. This has consequences for
4010 * range_cyclic semantics (ie. it may
4011 * not be suitable for data integrity
4014 done_index
= page
->index
+ 1;
4020 * the filesystem may choose to bump up nr_to_write.
4021 * We have to make sure to honor the new nr_to_write
4024 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4026 pagevec_release(&pvec
);
4029 if (!scanned
&& !done
) {
4031 * We hit the last page and there is more work to be done: wrap
4032 * back to the start of the file
4039 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4040 mapping
->writeback_index
= done_index
;
4042 btrfs_add_delayed_iput(inode
);
4046 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4055 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
4056 BUG_ON(ret
< 0); /* -ENOMEM */
4061 static noinline
void flush_write_bio(void *data
)
4063 struct extent_page_data
*epd
= data
;
4064 flush_epd_write_bio(epd
);
4067 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4068 get_extent_t
*get_extent
,
4069 struct writeback_control
*wbc
)
4072 struct extent_page_data epd
= {
4075 .get_extent
= get_extent
,
4077 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4081 ret
= __extent_writepage(page
, wbc
, &epd
);
4083 flush_epd_write_bio(&epd
);
4087 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4088 u64 start
, u64 end
, get_extent_t
*get_extent
,
4092 struct address_space
*mapping
= inode
->i_mapping
;
4094 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4097 struct extent_page_data epd
= {
4100 .get_extent
= get_extent
,
4102 .sync_io
= mode
== WB_SYNC_ALL
,
4105 struct writeback_control wbc_writepages
= {
4107 .nr_to_write
= nr_pages
* 2,
4108 .range_start
= start
,
4109 .range_end
= end
+ 1,
4112 while (start
<= end
) {
4113 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4114 if (clear_page_dirty_for_io(page
))
4115 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4117 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4118 tree
->ops
->writepage_end_io_hook(page
, start
,
4119 start
+ PAGE_SIZE
- 1,
4127 flush_epd_write_bio(&epd
);
4131 int extent_writepages(struct extent_io_tree
*tree
,
4132 struct address_space
*mapping
,
4133 get_extent_t
*get_extent
,
4134 struct writeback_control
*wbc
)
4137 struct extent_page_data epd
= {
4140 .get_extent
= get_extent
,
4142 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4146 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
4147 __extent_writepage
, &epd
,
4149 flush_epd_write_bio(&epd
);
4153 int extent_readpages(struct extent_io_tree
*tree
,
4154 struct address_space
*mapping
,
4155 struct list_head
*pages
, unsigned nr_pages
,
4156 get_extent_t get_extent
)
4158 struct bio
*bio
= NULL
;
4160 unsigned long bio_flags
= 0;
4161 struct page
*pagepool
[16];
4163 struct extent_map
*em_cached
= NULL
;
4165 u64 prev_em_start
= (u64
)-1;
4167 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4168 page
= list_entry(pages
->prev
, struct page
, lru
);
4170 prefetchw(&page
->flags
);
4171 list_del(&page
->lru
);
4172 if (add_to_page_cache_lru(page
, mapping
,
4173 page
->index
, GFP_NOFS
)) {
4178 pagepool
[nr
++] = page
;
4179 if (nr
< ARRAY_SIZE(pagepool
))
4181 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4182 &bio
, 0, &bio_flags
, READ
, &prev_em_start
);
4186 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4187 &bio
, 0, &bio_flags
, READ
, &prev_em_start
);
4190 free_extent_map(em_cached
);
4192 BUG_ON(!list_empty(pages
));
4194 return submit_one_bio(READ
, bio
, 0, bio_flags
);
4199 * basic invalidatepage code, this waits on any locked or writeback
4200 * ranges corresponding to the page, and then deletes any extent state
4201 * records from the tree
4203 int extent_invalidatepage(struct extent_io_tree
*tree
,
4204 struct page
*page
, unsigned long offset
)
4206 struct extent_state
*cached_state
= NULL
;
4207 u64 start
= page_offset(page
);
4208 u64 end
= start
+ PAGE_SIZE
- 1;
4209 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4211 start
+= ALIGN(offset
, blocksize
);
4215 lock_extent_bits(tree
, start
, end
, &cached_state
);
4216 wait_on_page_writeback(page
);
4217 clear_extent_bit(tree
, start
, end
,
4218 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4219 EXTENT_DO_ACCOUNTING
,
4220 1, 1, &cached_state
, GFP_NOFS
);
4225 * a helper for releasepage, this tests for areas of the page that
4226 * are locked or under IO and drops the related state bits if it is safe
4229 static int try_release_extent_state(struct extent_map_tree
*map
,
4230 struct extent_io_tree
*tree
,
4231 struct page
*page
, gfp_t mask
)
4233 u64 start
= page_offset(page
);
4234 u64 end
= start
+ PAGE_SIZE
- 1;
4237 if (test_range_bit(tree
, start
, end
,
4238 EXTENT_IOBITS
, 0, NULL
))
4241 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4244 * at this point we can safely clear everything except the
4245 * locked bit and the nodatasum bit
4247 ret
= clear_extent_bit(tree
, start
, end
,
4248 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4251 /* if clear_extent_bit failed for enomem reasons,
4252 * we can't allow the release to continue.
4263 * a helper for releasepage. As long as there are no locked extents
4264 * in the range corresponding to the page, both state records and extent
4265 * map records are removed
4267 int try_release_extent_mapping(struct extent_map_tree
*map
,
4268 struct extent_io_tree
*tree
, struct page
*page
,
4271 struct extent_map
*em
;
4272 u64 start
= page_offset(page
);
4273 u64 end
= start
+ PAGE_SIZE
- 1;
4275 if (gfpflags_allow_blocking(mask
) &&
4276 page
->mapping
->host
->i_size
> SZ_16M
) {
4278 while (start
<= end
) {
4279 len
= end
- start
+ 1;
4280 write_lock(&map
->lock
);
4281 em
= lookup_extent_mapping(map
, start
, len
);
4283 write_unlock(&map
->lock
);
4286 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4287 em
->start
!= start
) {
4288 write_unlock(&map
->lock
);
4289 free_extent_map(em
);
4292 if (!test_range_bit(tree
, em
->start
,
4293 extent_map_end(em
) - 1,
4294 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4296 remove_extent_mapping(map
, em
);
4297 /* once for the rb tree */
4298 free_extent_map(em
);
4300 start
= extent_map_end(em
);
4301 write_unlock(&map
->lock
);
4304 free_extent_map(em
);
4307 return try_release_extent_state(map
, tree
, page
, mask
);
4311 * helper function for fiemap, which doesn't want to see any holes.
4312 * This maps until we find something past 'last'
4314 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4317 get_extent_t
*get_extent
)
4319 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4320 struct extent_map
*em
;
4327 len
= last
- offset
;
4330 len
= ALIGN(len
, sectorsize
);
4331 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4332 if (IS_ERR_OR_NULL(em
))
4335 /* if this isn't a hole return it */
4336 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4337 em
->block_start
!= EXTENT_MAP_HOLE
) {
4341 /* this is a hole, advance to the next extent */
4342 offset
= extent_map_end(em
);
4343 free_extent_map(em
);
4350 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4351 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4355 u64 max
= start
+ len
;
4359 u64 last_for_get_extent
= 0;
4361 u64 isize
= i_size_read(inode
);
4362 struct btrfs_key found_key
;
4363 struct extent_map
*em
= NULL
;
4364 struct extent_state
*cached_state
= NULL
;
4365 struct btrfs_path
*path
;
4366 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4375 path
= btrfs_alloc_path();
4378 path
->leave_spinning
= 1;
4380 start
= round_down(start
, BTRFS_I(inode
)->root
->sectorsize
);
4381 len
= round_up(max
, BTRFS_I(inode
)->root
->sectorsize
) - start
;
4384 * lookup the last file extent. We're not using i_size here
4385 * because there might be preallocation past i_size
4387 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
), -1,
4390 btrfs_free_path(path
);
4399 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4400 found_type
= found_key
.type
;
4402 /* No extents, but there might be delalloc bits */
4403 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4404 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4405 /* have to trust i_size as the end */
4407 last_for_get_extent
= isize
;
4410 * remember the start of the last extent. There are a
4411 * bunch of different factors that go into the length of the
4412 * extent, so its much less complex to remember where it started
4414 last
= found_key
.offset
;
4415 last_for_get_extent
= last
+ 1;
4417 btrfs_release_path(path
);
4420 * we might have some extents allocated but more delalloc past those
4421 * extents. so, we trust isize unless the start of the last extent is
4426 last_for_get_extent
= isize
;
4429 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4432 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4442 u64 offset_in_extent
= 0;
4444 /* break if the extent we found is outside the range */
4445 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4449 * get_extent may return an extent that starts before our
4450 * requested range. We have to make sure the ranges
4451 * we return to fiemap always move forward and don't
4452 * overlap, so adjust the offsets here
4454 em_start
= max(em
->start
, off
);
4457 * record the offset from the start of the extent
4458 * for adjusting the disk offset below. Only do this if the
4459 * extent isn't compressed since our in ram offset may be past
4460 * what we have actually allocated on disk.
4462 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4463 offset_in_extent
= em_start
- em
->start
;
4464 em_end
= extent_map_end(em
);
4465 em_len
= em_end
- em_start
;
4470 * bump off for our next call to get_extent
4472 off
= extent_map_end(em
);
4476 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4478 flags
|= FIEMAP_EXTENT_LAST
;
4479 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4480 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4481 FIEMAP_EXTENT_NOT_ALIGNED
);
4482 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4483 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4484 FIEMAP_EXTENT_UNKNOWN
);
4485 } else if (fieinfo
->fi_extents_max
) {
4486 u64 bytenr
= em
->block_start
-
4487 (em
->start
- em
->orig_start
);
4489 disko
= em
->block_start
+ offset_in_extent
;
4492 * As btrfs supports shared space, this information
4493 * can be exported to userspace tools via
4494 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4495 * then we're just getting a count and we can skip the
4498 ret
= btrfs_check_shared(NULL
, root
->fs_info
,
4500 btrfs_ino(inode
), bytenr
);
4504 flags
|= FIEMAP_EXTENT_SHARED
;
4507 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4508 flags
|= FIEMAP_EXTENT_ENCODED
;
4509 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4510 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4512 free_extent_map(em
);
4514 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4515 (last
== (u64
)-1 && isize
<= em_end
)) {
4516 flags
|= FIEMAP_EXTENT_LAST
;
4520 /* now scan forward to see if this is really the last extent. */
4521 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4528 flags
|= FIEMAP_EXTENT_LAST
;
4531 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4540 free_extent_map(em
);
4542 btrfs_free_path(path
);
4543 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4544 &cached_state
, GFP_NOFS
);
4548 static void __free_extent_buffer(struct extent_buffer
*eb
)
4550 btrfs_leak_debug_del(&eb
->leak_list
);
4551 kmem_cache_free(extent_buffer_cache
, eb
);
4554 int extent_buffer_under_io(struct extent_buffer
*eb
)
4556 return (atomic_read(&eb
->io_pages
) ||
4557 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4558 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4562 * Helper for releasing extent buffer page.
4564 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4566 unsigned long index
;
4568 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4570 BUG_ON(extent_buffer_under_io(eb
));
4572 index
= num_extent_pages(eb
->start
, eb
->len
);
4578 page
= eb
->pages
[index
];
4582 spin_lock(&page
->mapping
->private_lock
);
4584 * We do this since we'll remove the pages after we've
4585 * removed the eb from the radix tree, so we could race
4586 * and have this page now attached to the new eb. So
4587 * only clear page_private if it's still connected to
4590 if (PagePrivate(page
) &&
4591 page
->private == (unsigned long)eb
) {
4592 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4593 BUG_ON(PageDirty(page
));
4594 BUG_ON(PageWriteback(page
));
4596 * We need to make sure we haven't be attached
4599 ClearPagePrivate(page
);
4600 set_page_private(page
, 0);
4601 /* One for the page private */
4606 spin_unlock(&page
->mapping
->private_lock
);
4608 /* One for when we allocated the page */
4610 } while (index
!= 0);
4614 * Helper for releasing the extent buffer.
4616 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4618 btrfs_release_extent_buffer_page(eb
);
4619 __free_extent_buffer(eb
);
4622 static struct extent_buffer
*
4623 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4626 struct extent_buffer
*eb
= NULL
;
4628 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4631 eb
->fs_info
= fs_info
;
4633 rwlock_init(&eb
->lock
);
4634 atomic_set(&eb
->write_locks
, 0);
4635 atomic_set(&eb
->read_locks
, 0);
4636 atomic_set(&eb
->blocking_readers
, 0);
4637 atomic_set(&eb
->blocking_writers
, 0);
4638 atomic_set(&eb
->spinning_readers
, 0);
4639 atomic_set(&eb
->spinning_writers
, 0);
4640 eb
->lock_nested
= 0;
4641 init_waitqueue_head(&eb
->write_lock_wq
);
4642 init_waitqueue_head(&eb
->read_lock_wq
);
4644 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4646 spin_lock_init(&eb
->refs_lock
);
4647 atomic_set(&eb
->refs
, 1);
4648 atomic_set(&eb
->io_pages
, 0);
4651 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4653 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4654 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4655 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4660 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4664 struct extent_buffer
*new;
4665 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4667 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4671 for (i
= 0; i
< num_pages
; i
++) {
4672 p
= alloc_page(GFP_NOFS
);
4674 btrfs_release_extent_buffer(new);
4677 attach_extent_buffer_page(new, p
);
4678 WARN_ON(PageDirty(p
));
4683 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4684 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4685 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4690 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4691 u64 start
, unsigned long len
)
4693 struct extent_buffer
*eb
;
4694 unsigned long num_pages
;
4697 num_pages
= num_extent_pages(start
, len
);
4699 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4703 for (i
= 0; i
< num_pages
; i
++) {
4704 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4708 set_extent_buffer_uptodate(eb
);
4709 btrfs_set_header_nritems(eb
, 0);
4710 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4715 __free_page(eb
->pages
[i
- 1]);
4716 __free_extent_buffer(eb
);
4720 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4727 * Called only from tests that don't always have a fs_info
4728 * available, but we know that nodesize is 4096
4732 len
= fs_info
->tree_root
->nodesize
;
4735 return __alloc_dummy_extent_buffer(fs_info
, start
, len
);
4738 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4741 /* the ref bit is tricky. We have to make sure it is set
4742 * if we have the buffer dirty. Otherwise the
4743 * code to free a buffer can end up dropping a dirty
4746 * Once the ref bit is set, it won't go away while the
4747 * buffer is dirty or in writeback, and it also won't
4748 * go away while we have the reference count on the
4751 * We can't just set the ref bit without bumping the
4752 * ref on the eb because free_extent_buffer might
4753 * see the ref bit and try to clear it. If this happens
4754 * free_extent_buffer might end up dropping our original
4755 * ref by mistake and freeing the page before we are able
4756 * to add one more ref.
4758 * So bump the ref count first, then set the bit. If someone
4759 * beat us to it, drop the ref we added.
4761 refs
= atomic_read(&eb
->refs
);
4762 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4765 spin_lock(&eb
->refs_lock
);
4766 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4767 atomic_inc(&eb
->refs
);
4768 spin_unlock(&eb
->refs_lock
);
4771 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4772 struct page
*accessed
)
4774 unsigned long num_pages
, i
;
4776 check_buffer_tree_ref(eb
);
4778 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4779 for (i
= 0; i
< num_pages
; i
++) {
4780 struct page
*p
= eb
->pages
[i
];
4783 mark_page_accessed(p
);
4787 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4790 struct extent_buffer
*eb
;
4793 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4794 start
>> PAGE_SHIFT
);
4795 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4798 * Lock our eb's refs_lock to avoid races with
4799 * free_extent_buffer. When we get our eb it might be flagged
4800 * with EXTENT_BUFFER_STALE and another task running
4801 * free_extent_buffer might have seen that flag set,
4802 * eb->refs == 2, that the buffer isn't under IO (dirty and
4803 * writeback flags not set) and it's still in the tree (flag
4804 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4805 * of decrementing the extent buffer's reference count twice.
4806 * So here we could race and increment the eb's reference count,
4807 * clear its stale flag, mark it as dirty and drop our reference
4808 * before the other task finishes executing free_extent_buffer,
4809 * which would later result in an attempt to free an extent
4810 * buffer that is dirty.
4812 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4813 spin_lock(&eb
->refs_lock
);
4814 spin_unlock(&eb
->refs_lock
);
4816 mark_extent_buffer_accessed(eb
, NULL
);
4824 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4825 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4828 struct extent_buffer
*eb
, *exists
= NULL
;
4831 eb
= find_extent_buffer(fs_info
, start
);
4834 eb
= alloc_dummy_extent_buffer(fs_info
, start
);
4837 eb
->fs_info
= fs_info
;
4839 ret
= radix_tree_preload(GFP_NOFS
);
4842 spin_lock(&fs_info
->buffer_lock
);
4843 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4844 start
>> PAGE_SHIFT
, eb
);
4845 spin_unlock(&fs_info
->buffer_lock
);
4846 radix_tree_preload_end();
4847 if (ret
== -EEXIST
) {
4848 exists
= find_extent_buffer(fs_info
, start
);
4854 check_buffer_tree_ref(eb
);
4855 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4858 * We will free dummy extent buffer's if they come into
4859 * free_extent_buffer with a ref count of 2, but if we are using this we
4860 * want the buffers to stay in memory until we're done with them, so
4861 * bump the ref count again.
4863 atomic_inc(&eb
->refs
);
4866 btrfs_release_extent_buffer(eb
);
4871 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4874 unsigned long len
= fs_info
->tree_root
->nodesize
;
4875 unsigned long num_pages
= num_extent_pages(start
, len
);
4877 unsigned long index
= start
>> PAGE_SHIFT
;
4878 struct extent_buffer
*eb
;
4879 struct extent_buffer
*exists
= NULL
;
4881 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4885 eb
= find_extent_buffer(fs_info
, start
);
4889 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4893 for (i
= 0; i
< num_pages
; i
++, index
++) {
4894 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4898 spin_lock(&mapping
->private_lock
);
4899 if (PagePrivate(p
)) {
4901 * We could have already allocated an eb for this page
4902 * and attached one so lets see if we can get a ref on
4903 * the existing eb, and if we can we know it's good and
4904 * we can just return that one, else we know we can just
4905 * overwrite page->private.
4907 exists
= (struct extent_buffer
*)p
->private;
4908 if (atomic_inc_not_zero(&exists
->refs
)) {
4909 spin_unlock(&mapping
->private_lock
);
4912 mark_extent_buffer_accessed(exists
, p
);
4918 * Do this so attach doesn't complain and we need to
4919 * drop the ref the old guy had.
4921 ClearPagePrivate(p
);
4922 WARN_ON(PageDirty(p
));
4925 attach_extent_buffer_page(eb
, p
);
4926 spin_unlock(&mapping
->private_lock
);
4927 WARN_ON(PageDirty(p
));
4929 if (!PageUptodate(p
))
4933 * see below about how we avoid a nasty race with release page
4934 * and why we unlock later
4938 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4940 ret
= radix_tree_preload(GFP_NOFS
);
4944 spin_lock(&fs_info
->buffer_lock
);
4945 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4946 start
>> PAGE_SHIFT
, eb
);
4947 spin_unlock(&fs_info
->buffer_lock
);
4948 radix_tree_preload_end();
4949 if (ret
== -EEXIST
) {
4950 exists
= find_extent_buffer(fs_info
, start
);
4956 /* add one reference for the tree */
4957 check_buffer_tree_ref(eb
);
4958 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4961 * there is a race where release page may have
4962 * tried to find this extent buffer in the radix
4963 * but failed. It will tell the VM it is safe to
4964 * reclaim the, and it will clear the page private bit.
4965 * We must make sure to set the page private bit properly
4966 * after the extent buffer is in the radix tree so
4967 * it doesn't get lost
4969 SetPageChecked(eb
->pages
[0]);
4970 for (i
= 1; i
< num_pages
; i
++) {
4972 ClearPageChecked(p
);
4975 unlock_page(eb
->pages
[0]);
4979 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4980 for (i
= 0; i
< num_pages
; i
++) {
4982 unlock_page(eb
->pages
[i
]);
4985 btrfs_release_extent_buffer(eb
);
4989 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
4991 struct extent_buffer
*eb
=
4992 container_of(head
, struct extent_buffer
, rcu_head
);
4994 __free_extent_buffer(eb
);
4997 /* Expects to have eb->eb_lock already held */
4998 static int release_extent_buffer(struct extent_buffer
*eb
)
5000 WARN_ON(atomic_read(&eb
->refs
) == 0);
5001 if (atomic_dec_and_test(&eb
->refs
)) {
5002 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5003 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5005 spin_unlock(&eb
->refs_lock
);
5007 spin_lock(&fs_info
->buffer_lock
);
5008 radix_tree_delete(&fs_info
->buffer_radix
,
5009 eb
->start
>> PAGE_SHIFT
);
5010 spin_unlock(&fs_info
->buffer_lock
);
5012 spin_unlock(&eb
->refs_lock
);
5015 /* Should be safe to release our pages at this point */
5016 btrfs_release_extent_buffer_page(eb
);
5017 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5018 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5019 __free_extent_buffer(eb
);
5023 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5026 spin_unlock(&eb
->refs_lock
);
5031 void free_extent_buffer(struct extent_buffer
*eb
)
5039 refs
= atomic_read(&eb
->refs
);
5042 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5047 spin_lock(&eb
->refs_lock
);
5048 if (atomic_read(&eb
->refs
) == 2 &&
5049 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5050 atomic_dec(&eb
->refs
);
5052 if (atomic_read(&eb
->refs
) == 2 &&
5053 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5054 !extent_buffer_under_io(eb
) &&
5055 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5056 atomic_dec(&eb
->refs
);
5059 * I know this is terrible, but it's temporary until we stop tracking
5060 * the uptodate bits and such for the extent buffers.
5062 release_extent_buffer(eb
);
5065 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5070 spin_lock(&eb
->refs_lock
);
5071 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5073 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5074 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5075 atomic_dec(&eb
->refs
);
5076 release_extent_buffer(eb
);
5079 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5082 unsigned long num_pages
;
5085 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5087 for (i
= 0; i
< num_pages
; i
++) {
5088 page
= eb
->pages
[i
];
5089 if (!PageDirty(page
))
5093 WARN_ON(!PagePrivate(page
));
5095 clear_page_dirty_for_io(page
);
5096 spin_lock_irq(&page
->mapping
->tree_lock
);
5097 if (!PageDirty(page
)) {
5098 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5100 PAGECACHE_TAG_DIRTY
);
5102 spin_unlock_irq(&page
->mapping
->tree_lock
);
5103 ClearPageError(page
);
5106 WARN_ON(atomic_read(&eb
->refs
) == 0);
5109 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5112 unsigned long num_pages
;
5115 check_buffer_tree_ref(eb
);
5117 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5119 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5120 WARN_ON(atomic_read(&eb
->refs
) == 0);
5121 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5123 for (i
= 0; i
< num_pages
; i
++)
5124 set_page_dirty(eb
->pages
[i
]);
5128 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5132 unsigned long num_pages
;
5134 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5135 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5136 for (i
= 0; i
< num_pages
; i
++) {
5137 page
= eb
->pages
[i
];
5139 ClearPageUptodate(page
);
5143 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5147 unsigned long num_pages
;
5149 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5150 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5151 for (i
= 0; i
< num_pages
; i
++) {
5152 page
= eb
->pages
[i
];
5153 SetPageUptodate(page
);
5157 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5159 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5162 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5163 struct extent_buffer
*eb
, u64 start
, int wait
,
5164 get_extent_t
*get_extent
, int mirror_num
)
5167 unsigned long start_i
;
5171 int locked_pages
= 0;
5172 int all_uptodate
= 1;
5173 unsigned long num_pages
;
5174 unsigned long num_reads
= 0;
5175 struct bio
*bio
= NULL
;
5176 unsigned long bio_flags
= 0;
5178 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5182 WARN_ON(start
< eb
->start
);
5183 start_i
= (start
>> PAGE_SHIFT
) -
5184 (eb
->start
>> PAGE_SHIFT
);
5189 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5190 for (i
= start_i
; i
< num_pages
; i
++) {
5191 page
= eb
->pages
[i
];
5192 if (wait
== WAIT_NONE
) {
5193 if (!trylock_page(page
))
5199 if (!PageUptodate(page
)) {
5206 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5210 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5211 eb
->read_mirror
= 0;
5212 atomic_set(&eb
->io_pages
, num_reads
);
5213 for (i
= start_i
; i
< num_pages
; i
++) {
5214 page
= eb
->pages
[i
];
5215 if (!PageUptodate(page
)) {
5216 ClearPageError(page
);
5217 err
= __extent_read_full_page(tree
, page
,
5219 mirror_num
, &bio_flags
,
5229 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
5235 if (ret
|| wait
!= WAIT_COMPLETE
)
5238 for (i
= start_i
; i
< num_pages
; i
++) {
5239 page
= eb
->pages
[i
];
5240 wait_on_page_locked(page
);
5241 if (!PageUptodate(page
))
5249 while (locked_pages
> 0) {
5250 page
= eb
->pages
[i
];
5258 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5259 unsigned long start
,
5266 char *dst
= (char *)dstv
;
5267 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5268 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5270 WARN_ON(start
> eb
->len
);
5271 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5273 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5276 page
= eb
->pages
[i
];
5278 cur
= min(len
, (PAGE_SIZE
- offset
));
5279 kaddr
= page_address(page
);
5280 memcpy(dst
, kaddr
+ offset
, cur
);
5289 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5290 unsigned long start
,
5297 char __user
*dst
= (char __user
*)dstv
;
5298 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5299 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5302 WARN_ON(start
> eb
->len
);
5303 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5305 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5308 page
= eb
->pages
[i
];
5310 cur
= min(len
, (PAGE_SIZE
- offset
));
5311 kaddr
= page_address(page
);
5312 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5326 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5327 unsigned long min_len
, char **map
,
5328 unsigned long *map_start
,
5329 unsigned long *map_len
)
5331 size_t offset
= start
& (PAGE_SIZE
- 1);
5334 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5335 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5336 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5343 offset
= start_offset
;
5347 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5350 if (start
+ min_len
> eb
->len
) {
5351 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
5353 eb
->start
, eb
->len
, start
, min_len
);
5358 kaddr
= page_address(p
);
5359 *map
= kaddr
+ offset
;
5360 *map_len
= PAGE_SIZE
- offset
;
5364 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5365 unsigned long start
,
5372 char *ptr
= (char *)ptrv
;
5373 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5374 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5377 WARN_ON(start
> eb
->len
);
5378 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5380 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5383 page
= eb
->pages
[i
];
5385 cur
= min(len
, (PAGE_SIZE
- offset
));
5387 kaddr
= page_address(page
);
5388 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5400 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5401 unsigned long start
, unsigned long len
)
5407 char *src
= (char *)srcv
;
5408 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5409 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5411 WARN_ON(start
> eb
->len
);
5412 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5414 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5417 page
= eb
->pages
[i
];
5418 WARN_ON(!PageUptodate(page
));
5420 cur
= min(len
, PAGE_SIZE
- offset
);
5421 kaddr
= page_address(page
);
5422 memcpy(kaddr
+ offset
, src
, cur
);
5431 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5432 unsigned long start
, unsigned long len
)
5438 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5439 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5441 WARN_ON(start
> eb
->len
);
5442 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5444 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5447 page
= eb
->pages
[i
];
5448 WARN_ON(!PageUptodate(page
));
5450 cur
= min(len
, PAGE_SIZE
- offset
);
5451 kaddr
= page_address(page
);
5452 memset(kaddr
+ offset
, c
, cur
);
5460 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5461 unsigned long dst_offset
, unsigned long src_offset
,
5464 u64 dst_len
= dst
->len
;
5469 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5470 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5472 WARN_ON(src
->len
!= dst_len
);
5474 offset
= (start_offset
+ dst_offset
) &
5478 page
= dst
->pages
[i
];
5479 WARN_ON(!PageUptodate(page
));
5481 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5483 kaddr
= page_address(page
);
5484 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5494 * The extent buffer bitmap operations are done with byte granularity because
5495 * bitmap items are not guaranteed to be aligned to a word and therefore a
5496 * single word in a bitmap may straddle two pages in the extent buffer.
5498 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
5499 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
5500 #define BITMAP_FIRST_BYTE_MASK(start) \
5501 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
5502 #define BITMAP_LAST_BYTE_MASK(nbits) \
5503 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
5506 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5508 * @eb: the extent buffer
5509 * @start: offset of the bitmap item in the extent buffer
5511 * @page_index: return index of the page in the extent buffer that contains the
5513 * @page_offset: return offset into the page given by page_index
5515 * This helper hides the ugliness of finding the byte in an extent buffer which
5516 * contains a given bit.
5518 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5519 unsigned long start
, unsigned long nr
,
5520 unsigned long *page_index
,
5521 size_t *page_offset
)
5523 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5524 size_t byte_offset
= BIT_BYTE(nr
);
5528 * The byte we want is the offset of the extent buffer + the offset of
5529 * the bitmap item in the extent buffer + the offset of the byte in the
5532 offset
= start_offset
+ start
+ byte_offset
;
5534 *page_index
= offset
>> PAGE_SHIFT
;
5535 *page_offset
= offset
& (PAGE_SIZE
- 1);
5539 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5540 * @eb: the extent buffer
5541 * @start: offset of the bitmap item in the extent buffer
5542 * @nr: bit number to test
5544 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5552 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5553 page
= eb
->pages
[i
];
5554 WARN_ON(!PageUptodate(page
));
5555 kaddr
= page_address(page
);
5556 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5560 * extent_buffer_bitmap_set - set an area of a bitmap
5561 * @eb: the extent buffer
5562 * @start: offset of the bitmap item in the extent buffer
5563 * @pos: bit number of the first bit
5564 * @len: number of bits to set
5566 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5567 unsigned long pos
, unsigned long len
)
5573 const unsigned int size
= pos
+ len
;
5574 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5575 unsigned int mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5577 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5578 page
= eb
->pages
[i
];
5579 WARN_ON(!PageUptodate(page
));
5580 kaddr
= page_address(page
);
5582 while (len
>= bits_to_set
) {
5583 kaddr
[offset
] |= mask_to_set
;
5585 bits_to_set
= BITS_PER_BYTE
;
5587 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5589 page
= eb
->pages
[++i
];
5590 WARN_ON(!PageUptodate(page
));
5591 kaddr
= page_address(page
);
5595 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5596 kaddr
[offset
] |= mask_to_set
;
5602 * extent_buffer_bitmap_clear - clear an area of a bitmap
5603 * @eb: the extent buffer
5604 * @start: offset of the bitmap item in the extent buffer
5605 * @pos: bit number of the first bit
5606 * @len: number of bits to clear
5608 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5609 unsigned long pos
, unsigned long len
)
5615 const unsigned int size
= pos
+ len
;
5616 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5617 unsigned int mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5619 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5620 page
= eb
->pages
[i
];
5621 WARN_ON(!PageUptodate(page
));
5622 kaddr
= page_address(page
);
5624 while (len
>= bits_to_clear
) {
5625 kaddr
[offset
] &= ~mask_to_clear
;
5626 len
-= bits_to_clear
;
5627 bits_to_clear
= BITS_PER_BYTE
;
5628 mask_to_clear
= ~0U;
5629 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5631 page
= eb
->pages
[++i
];
5632 WARN_ON(!PageUptodate(page
));
5633 kaddr
= page_address(page
);
5637 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5638 kaddr
[offset
] &= ~mask_to_clear
;
5642 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5644 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5645 return distance
< len
;
5648 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5649 unsigned long dst_off
, unsigned long src_off
,
5652 char *dst_kaddr
= page_address(dst_page
);
5654 int must_memmove
= 0;
5656 if (dst_page
!= src_page
) {
5657 src_kaddr
= page_address(src_page
);
5659 src_kaddr
= dst_kaddr
;
5660 if (areas_overlap(src_off
, dst_off
, len
))
5665 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5667 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5670 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5671 unsigned long src_offset
, unsigned long len
)
5674 size_t dst_off_in_page
;
5675 size_t src_off_in_page
;
5676 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5677 unsigned long dst_i
;
5678 unsigned long src_i
;
5680 if (src_offset
+ len
> dst
->len
) {
5681 btrfs_err(dst
->fs_info
,
5682 "memmove bogus src_offset %lu move "
5683 "len %lu dst len %lu", src_offset
, len
, dst
->len
);
5686 if (dst_offset
+ len
> dst
->len
) {
5687 btrfs_err(dst
->fs_info
,
5688 "memmove bogus dst_offset %lu move "
5689 "len %lu dst len %lu", dst_offset
, len
, dst
->len
);
5694 dst_off_in_page
= (start_offset
+ dst_offset
) &
5696 src_off_in_page
= (start_offset
+ src_offset
) &
5699 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5700 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5702 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5704 cur
= min_t(unsigned long, cur
,
5705 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5707 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5708 dst_off_in_page
, src_off_in_page
, cur
);
5716 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5717 unsigned long src_offset
, unsigned long len
)
5720 size_t dst_off_in_page
;
5721 size_t src_off_in_page
;
5722 unsigned long dst_end
= dst_offset
+ len
- 1;
5723 unsigned long src_end
= src_offset
+ len
- 1;
5724 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5725 unsigned long dst_i
;
5726 unsigned long src_i
;
5728 if (src_offset
+ len
> dst
->len
) {
5729 btrfs_err(dst
->fs_info
, "memmove bogus src_offset %lu move "
5730 "len %lu len %lu", src_offset
, len
, dst
->len
);
5733 if (dst_offset
+ len
> dst
->len
) {
5734 btrfs_err(dst
->fs_info
, "memmove bogus dst_offset %lu move "
5735 "len %lu len %lu", dst_offset
, len
, dst
->len
);
5738 if (dst_offset
< src_offset
) {
5739 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5743 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5744 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5746 dst_off_in_page
= (start_offset
+ dst_end
) &
5748 src_off_in_page
= (start_offset
+ src_end
) &
5751 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5752 cur
= min(cur
, dst_off_in_page
+ 1);
5753 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5754 dst_off_in_page
- cur
+ 1,
5755 src_off_in_page
- cur
+ 1, cur
);
5763 int try_release_extent_buffer(struct page
*page
)
5765 struct extent_buffer
*eb
;
5768 * We need to make sure nobody is attaching this page to an eb right
5771 spin_lock(&page
->mapping
->private_lock
);
5772 if (!PagePrivate(page
)) {
5773 spin_unlock(&page
->mapping
->private_lock
);
5777 eb
= (struct extent_buffer
*)page
->private;
5781 * This is a little awful but should be ok, we need to make sure that
5782 * the eb doesn't disappear out from under us while we're looking at
5785 spin_lock(&eb
->refs_lock
);
5786 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5787 spin_unlock(&eb
->refs_lock
);
5788 spin_unlock(&page
->mapping
->private_lock
);
5791 spin_unlock(&page
->mapping
->private_lock
);
5794 * If tree ref isn't set then we know the ref on this eb is a real ref,
5795 * so just return, this page will likely be freed soon anyway.
5797 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5798 spin_unlock(&eb
->refs_lock
);
5802 return release_extent_buffer(eb
);