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"
23 #include "transaction.h"
25 static struct kmem_cache
*extent_state_cache
;
26 static struct kmem_cache
*extent_buffer_cache
;
27 static struct bio_set
*btrfs_bioset
;
29 static inline bool extent_state_in_tree(const struct extent_state
*state
)
31 return !RB_EMPTY_NODE(&state
->rb_node
);
34 #ifdef CONFIG_BTRFS_DEBUG
35 static LIST_HEAD(buffers
);
36 static LIST_HEAD(states
);
38 static DEFINE_SPINLOCK(leak_lock
);
41 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
45 spin_lock_irqsave(&leak_lock
, flags
);
47 spin_unlock_irqrestore(&leak_lock
, flags
);
51 void btrfs_leak_debug_del(struct list_head
*entry
)
55 spin_lock_irqsave(&leak_lock
, flags
);
57 spin_unlock_irqrestore(&leak_lock
, flags
);
61 void btrfs_leak_debug_check(void)
63 struct extent_state
*state
;
64 struct extent_buffer
*eb
;
66 while (!list_empty(&states
)) {
67 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
68 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
69 state
->start
, state
->end
, state
->state
,
70 extent_state_in_tree(state
),
71 atomic_read(&state
->refs
));
72 list_del(&state
->leak_list
);
73 kmem_cache_free(extent_state_cache
, state
);
76 while (!list_empty(&buffers
)) {
77 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
78 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
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(BTRFS_I(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 REQ_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_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_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
;
230 * The given mask might be not appropriate for the slab allocator,
231 * drop the unsupported bits
233 mask
&= ~(__GFP_DMA32
|__GFP_HIGHMEM
);
234 state
= kmem_cache_alloc(extent_state_cache
, mask
);
238 state
->failrec
= NULL
;
239 RB_CLEAR_NODE(&state
->rb_node
);
240 btrfs_leak_debug_add(&state
->leak_list
, &states
);
241 atomic_set(&state
->refs
, 1);
242 init_waitqueue_head(&state
->wq
);
243 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
247 void free_extent_state(struct extent_state
*state
)
251 if (atomic_dec_and_test(&state
->refs
)) {
252 WARN_ON(extent_state_in_tree(state
));
253 btrfs_leak_debug_del(&state
->leak_list
);
254 trace_free_extent_state(state
, _RET_IP_
);
255 kmem_cache_free(extent_state_cache
, state
);
259 static struct rb_node
*tree_insert(struct rb_root
*root
,
260 struct rb_node
*search_start
,
262 struct rb_node
*node
,
263 struct rb_node
***p_in
,
264 struct rb_node
**parent_in
)
267 struct rb_node
*parent
= NULL
;
268 struct tree_entry
*entry
;
270 if (p_in
&& parent_in
) {
276 p
= search_start
? &search_start
: &root
->rb_node
;
279 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
281 if (offset
< entry
->start
)
283 else if (offset
> entry
->end
)
290 rb_link_node(node
, parent
, p
);
291 rb_insert_color(node
, root
);
295 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
296 struct rb_node
**prev_ret
,
297 struct rb_node
**next_ret
,
298 struct rb_node
***p_ret
,
299 struct rb_node
**parent_ret
)
301 struct rb_root
*root
= &tree
->state
;
302 struct rb_node
**n
= &root
->rb_node
;
303 struct rb_node
*prev
= NULL
;
304 struct rb_node
*orig_prev
= NULL
;
305 struct tree_entry
*entry
;
306 struct tree_entry
*prev_entry
= NULL
;
310 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
313 if (offset
< entry
->start
)
315 else if (offset
> entry
->end
)
328 while (prev
&& offset
> prev_entry
->end
) {
329 prev
= rb_next(prev
);
330 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
337 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
338 while (prev
&& offset
< prev_entry
->start
) {
339 prev
= rb_prev(prev
);
340 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
347 static inline struct rb_node
*
348 tree_search_for_insert(struct extent_io_tree
*tree
,
350 struct rb_node
***p_ret
,
351 struct rb_node
**parent_ret
)
353 struct rb_node
*prev
= NULL
;
356 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
362 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
365 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
368 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
369 struct extent_state
*other
)
371 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
372 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
377 * utility function to look for merge candidates inside a given range.
378 * Any extents with matching state are merged together into a single
379 * extent in the tree. Extents with EXTENT_IO in their state field
380 * are not merged because the end_io handlers need to be able to do
381 * operations on them without sleeping (or doing allocations/splits).
383 * This should be called with the tree lock held.
385 static void merge_state(struct extent_io_tree
*tree
,
386 struct extent_state
*state
)
388 struct extent_state
*other
;
389 struct rb_node
*other_node
;
391 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
394 other_node
= rb_prev(&state
->rb_node
);
396 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
397 if (other
->end
== state
->start
- 1 &&
398 other
->state
== state
->state
) {
399 merge_cb(tree
, state
, other
);
400 state
->start
= other
->start
;
401 rb_erase(&other
->rb_node
, &tree
->state
);
402 RB_CLEAR_NODE(&other
->rb_node
);
403 free_extent_state(other
);
406 other_node
= rb_next(&state
->rb_node
);
408 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
409 if (other
->start
== state
->end
+ 1 &&
410 other
->state
== state
->state
) {
411 merge_cb(tree
, state
, other
);
412 state
->end
= other
->end
;
413 rb_erase(&other
->rb_node
, &tree
->state
);
414 RB_CLEAR_NODE(&other
->rb_node
);
415 free_extent_state(other
);
420 static void set_state_cb(struct extent_io_tree
*tree
,
421 struct extent_state
*state
, unsigned *bits
)
423 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
424 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
427 static void clear_state_cb(struct extent_io_tree
*tree
,
428 struct extent_state
*state
, unsigned *bits
)
430 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
431 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
434 static void set_state_bits(struct extent_io_tree
*tree
,
435 struct extent_state
*state
, unsigned *bits
,
436 struct extent_changeset
*changeset
);
439 * insert an extent_state struct into the tree. 'bits' are set on the
440 * struct before it is inserted.
442 * This may return -EEXIST if the extent is already there, in which case the
443 * state struct is freed.
445 * The tree lock is not taken internally. This is a utility function and
446 * probably isn't what you want to call (see set/clear_extent_bit).
448 static int insert_state(struct extent_io_tree
*tree
,
449 struct extent_state
*state
, u64 start
, u64 end
,
451 struct rb_node
**parent
,
452 unsigned *bits
, struct extent_changeset
*changeset
)
454 struct rb_node
*node
;
457 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
459 state
->start
= start
;
462 set_state_bits(tree
, state
, bits
, changeset
);
464 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
466 struct extent_state
*found
;
467 found
= rb_entry(node
, struct extent_state
, rb_node
);
468 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
469 found
->start
, found
->end
, start
, end
);
472 merge_state(tree
, state
);
476 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
479 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
480 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
484 * split a given extent state struct in two, inserting the preallocated
485 * struct 'prealloc' as the newly created second half. 'split' indicates an
486 * offset inside 'orig' where it should be split.
489 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
490 * are two extent state structs in the tree:
491 * prealloc: [orig->start, split - 1]
492 * orig: [ split, orig->end ]
494 * The tree locks are not taken by this function. They need to be held
497 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
498 struct extent_state
*prealloc
, u64 split
)
500 struct rb_node
*node
;
502 split_cb(tree
, orig
, split
);
504 prealloc
->start
= orig
->start
;
505 prealloc
->end
= split
- 1;
506 prealloc
->state
= orig
->state
;
509 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
510 &prealloc
->rb_node
, NULL
, NULL
);
512 free_extent_state(prealloc
);
518 static struct extent_state
*next_state(struct extent_state
*state
)
520 struct rb_node
*next
= rb_next(&state
->rb_node
);
522 return rb_entry(next
, struct extent_state
, rb_node
);
528 * utility function to clear some bits in an extent state struct.
529 * it will optionally wake up any one waiting on this state (wake == 1).
531 * If no bits are set on the state struct after clearing things, the
532 * struct is freed and removed from the tree
534 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
535 struct extent_state
*state
,
536 unsigned *bits
, int wake
,
537 struct extent_changeset
*changeset
)
539 struct extent_state
*next
;
540 unsigned bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
542 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
543 u64 range
= state
->end
- state
->start
+ 1;
544 WARN_ON(range
> tree
->dirty_bytes
);
545 tree
->dirty_bytes
-= range
;
547 clear_state_cb(tree
, state
, bits
);
548 add_extent_changeset(state
, bits_to_clear
, changeset
, 0);
549 state
->state
&= ~bits_to_clear
;
552 if (state
->state
== 0) {
553 next
= next_state(state
);
554 if (extent_state_in_tree(state
)) {
555 rb_erase(&state
->rb_node
, &tree
->state
);
556 RB_CLEAR_NODE(&state
->rb_node
);
557 free_extent_state(state
);
562 merge_state(tree
, state
);
563 next
= next_state(state
);
568 static struct extent_state
*
569 alloc_extent_state_atomic(struct extent_state
*prealloc
)
572 prealloc
= alloc_extent_state(GFP_ATOMIC
);
577 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
579 btrfs_panic(tree_fs_info(tree
), err
,
580 "Locking error: Extent tree was modified by another thread while locked.");
584 * clear some bits on a range in the tree. This may require splitting
585 * or inserting elements in the tree, so the gfp mask is used to
586 * indicate which allocations or sleeping are allowed.
588 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
589 * the given range from the tree regardless of state (ie for truncate).
591 * the range [start, end] is inclusive.
593 * This takes the tree lock, and returns 0 on success and < 0 on error.
595 static int __clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
596 unsigned bits
, int wake
, int delete,
597 struct extent_state
**cached_state
,
598 gfp_t mask
, struct extent_changeset
*changeset
)
600 struct extent_state
*state
;
601 struct extent_state
*cached
;
602 struct extent_state
*prealloc
= NULL
;
603 struct rb_node
*node
;
608 btrfs_debug_check_extent_io_range(tree
, start
, end
);
610 if (bits
& EXTENT_DELALLOC
)
611 bits
|= EXTENT_NORESERVE
;
614 bits
|= ~EXTENT_CTLBITS
;
615 bits
|= EXTENT_FIRST_DELALLOC
;
617 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
620 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
622 * Don't care for allocation failure here because we might end
623 * up not needing the pre-allocated extent state at all, which
624 * is the case if we only have in the tree extent states that
625 * cover our input range and don't cover too any other range.
626 * If we end up needing a new extent state we allocate it later.
628 prealloc
= alloc_extent_state(mask
);
631 spin_lock(&tree
->lock
);
633 cached
= *cached_state
;
636 *cached_state
= NULL
;
640 if (cached
&& extent_state_in_tree(cached
) &&
641 cached
->start
<= start
&& cached
->end
> start
) {
643 atomic_dec(&cached
->refs
);
648 free_extent_state(cached
);
651 * this search will find the extents that end after
654 node
= tree_search(tree
, start
);
657 state
= rb_entry(node
, struct extent_state
, rb_node
);
659 if (state
->start
> end
)
661 WARN_ON(state
->end
< start
);
662 last_end
= state
->end
;
664 /* the state doesn't have the wanted bits, go ahead */
665 if (!(state
->state
& bits
)) {
666 state
= next_state(state
);
671 * | ---- desired range ---- |
673 * | ------------- state -------------- |
675 * We need to split the extent we found, and may flip
676 * bits on second half.
678 * If the extent we found extends past our range, we
679 * just split and search again. It'll get split again
680 * the next time though.
682 * If the extent we found is inside our range, we clear
683 * the desired bit on it.
686 if (state
->start
< start
) {
687 prealloc
= alloc_extent_state_atomic(prealloc
);
689 err
= split_state(tree
, state
, prealloc
, start
);
691 extent_io_tree_panic(tree
, err
);
696 if (state
->end
<= end
) {
697 state
= clear_state_bit(tree
, state
, &bits
, wake
,
704 * | ---- desired range ---- |
706 * We need to split the extent, and clear the bit
709 if (state
->start
<= end
&& state
->end
> end
) {
710 prealloc
= alloc_extent_state_atomic(prealloc
);
712 err
= split_state(tree
, state
, prealloc
, end
+ 1);
714 extent_io_tree_panic(tree
, err
);
719 clear_state_bit(tree
, prealloc
, &bits
, wake
, changeset
);
725 state
= clear_state_bit(tree
, state
, &bits
, wake
, changeset
);
727 if (last_end
== (u64
)-1)
729 start
= last_end
+ 1;
730 if (start
<= end
&& state
&& !need_resched())
736 spin_unlock(&tree
->lock
);
737 if (gfpflags_allow_blocking(mask
))
742 spin_unlock(&tree
->lock
);
744 free_extent_state(prealloc
);
750 static void wait_on_state(struct extent_io_tree
*tree
,
751 struct extent_state
*state
)
752 __releases(tree
->lock
)
753 __acquires(tree
->lock
)
756 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
757 spin_unlock(&tree
->lock
);
759 spin_lock(&tree
->lock
);
760 finish_wait(&state
->wq
, &wait
);
764 * waits for one or more bits to clear on a range in the state tree.
765 * The range [start, end] is inclusive.
766 * The tree lock is taken by this function
768 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
771 struct extent_state
*state
;
772 struct rb_node
*node
;
774 btrfs_debug_check_extent_io_range(tree
, start
, end
);
776 spin_lock(&tree
->lock
);
780 * this search will find all the extents that end after
783 node
= tree_search(tree
, start
);
788 state
= rb_entry(node
, struct extent_state
, rb_node
);
790 if (state
->start
> end
)
793 if (state
->state
& bits
) {
794 start
= state
->start
;
795 atomic_inc(&state
->refs
);
796 wait_on_state(tree
, state
);
797 free_extent_state(state
);
800 start
= state
->end
+ 1;
805 if (!cond_resched_lock(&tree
->lock
)) {
806 node
= rb_next(node
);
811 spin_unlock(&tree
->lock
);
814 static void set_state_bits(struct extent_io_tree
*tree
,
815 struct extent_state
*state
,
816 unsigned *bits
, struct extent_changeset
*changeset
)
818 unsigned bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
820 set_state_cb(tree
, state
, bits
);
821 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
822 u64 range
= state
->end
- state
->start
+ 1;
823 tree
->dirty_bytes
+= range
;
825 add_extent_changeset(state
, bits_to_set
, changeset
, 1);
826 state
->state
|= bits_to_set
;
829 static void cache_state_if_flags(struct extent_state
*state
,
830 struct extent_state
**cached_ptr
,
833 if (cached_ptr
&& !(*cached_ptr
)) {
834 if (!flags
|| (state
->state
& flags
)) {
836 atomic_inc(&state
->refs
);
841 static void cache_state(struct extent_state
*state
,
842 struct extent_state
**cached_ptr
)
844 return cache_state_if_flags(state
, cached_ptr
,
845 EXTENT_IOBITS
| EXTENT_BOUNDARY
);
849 * set some bits on a range in the tree. This may require allocations or
850 * sleeping, so the gfp mask is used to indicate what is allowed.
852 * If any of the exclusive bits are set, this will fail with -EEXIST if some
853 * part of the range already has the desired bits set. The start of the
854 * existing range is returned in failed_start in this case.
856 * [start, end] is inclusive This takes the tree lock.
859 static int __must_check
860 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
861 unsigned bits
, unsigned exclusive_bits
,
862 u64
*failed_start
, struct extent_state
**cached_state
,
863 gfp_t mask
, struct extent_changeset
*changeset
)
865 struct extent_state
*state
;
866 struct extent_state
*prealloc
= NULL
;
867 struct rb_node
*node
;
869 struct rb_node
*parent
;
874 btrfs_debug_check_extent_io_range(tree
, start
, end
);
876 bits
|= EXTENT_FIRST_DELALLOC
;
878 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
880 * Don't care for allocation failure here because we might end
881 * up not needing the pre-allocated extent state at all, which
882 * is the case if we only have in the tree extent states that
883 * cover our input range and don't cover too any other range.
884 * If we end up needing a new extent state we allocate it later.
886 prealloc
= alloc_extent_state(mask
);
889 spin_lock(&tree
->lock
);
890 if (cached_state
&& *cached_state
) {
891 state
= *cached_state
;
892 if (state
->start
<= start
&& state
->end
> start
&&
893 extent_state_in_tree(state
)) {
894 node
= &state
->rb_node
;
899 * this search will find all the extents that end after
902 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
904 prealloc
= alloc_extent_state_atomic(prealloc
);
906 err
= insert_state(tree
, prealloc
, start
, end
,
907 &p
, &parent
, &bits
, changeset
);
909 extent_io_tree_panic(tree
, err
);
911 cache_state(prealloc
, cached_state
);
915 state
= rb_entry(node
, struct extent_state
, rb_node
);
917 last_start
= state
->start
;
918 last_end
= state
->end
;
921 * | ---- desired range ---- |
924 * Just lock what we found and keep going
926 if (state
->start
== start
&& state
->end
<= end
) {
927 if (state
->state
& exclusive_bits
) {
928 *failed_start
= state
->start
;
933 set_state_bits(tree
, state
, &bits
, changeset
);
934 cache_state(state
, cached_state
);
935 merge_state(tree
, state
);
936 if (last_end
== (u64
)-1)
938 start
= last_end
+ 1;
939 state
= next_state(state
);
940 if (start
< end
&& state
&& state
->start
== start
&&
947 * | ---- desired range ---- |
950 * | ------------- state -------------- |
952 * We need to split the extent we found, and may flip bits on
955 * If the extent we found extends past our
956 * range, we just split and search again. It'll get split
957 * again the next time though.
959 * If the extent we found is inside our range, we set the
962 if (state
->start
< start
) {
963 if (state
->state
& exclusive_bits
) {
964 *failed_start
= start
;
969 prealloc
= alloc_extent_state_atomic(prealloc
);
971 err
= split_state(tree
, state
, prealloc
, start
);
973 extent_io_tree_panic(tree
, err
);
978 if (state
->end
<= end
) {
979 set_state_bits(tree
, state
, &bits
, changeset
);
980 cache_state(state
, cached_state
);
981 merge_state(tree
, state
);
982 if (last_end
== (u64
)-1)
984 start
= last_end
+ 1;
985 state
= next_state(state
);
986 if (start
< end
&& state
&& state
->start
== start
&&
993 * | ---- desired range ---- |
994 * | state | or | state |
996 * There's a hole, we need to insert something in it and
997 * ignore the extent we found.
999 if (state
->start
> start
) {
1001 if (end
< last_start
)
1004 this_end
= last_start
- 1;
1006 prealloc
= alloc_extent_state_atomic(prealloc
);
1010 * Avoid to free 'prealloc' if it can be merged with
1013 err
= insert_state(tree
, prealloc
, start
, this_end
,
1014 NULL
, NULL
, &bits
, changeset
);
1016 extent_io_tree_panic(tree
, err
);
1018 cache_state(prealloc
, cached_state
);
1020 start
= this_end
+ 1;
1024 * | ---- desired range ---- |
1026 * We need to split the extent, and set the bit
1029 if (state
->start
<= end
&& state
->end
> end
) {
1030 if (state
->state
& exclusive_bits
) {
1031 *failed_start
= start
;
1036 prealloc
= alloc_extent_state_atomic(prealloc
);
1038 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1040 extent_io_tree_panic(tree
, err
);
1042 set_state_bits(tree
, prealloc
, &bits
, changeset
);
1043 cache_state(prealloc
, cached_state
);
1044 merge_state(tree
, prealloc
);
1052 spin_unlock(&tree
->lock
);
1053 if (gfpflags_allow_blocking(mask
))
1058 spin_unlock(&tree
->lock
);
1060 free_extent_state(prealloc
);
1066 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1067 unsigned bits
, u64
* failed_start
,
1068 struct extent_state
**cached_state
, gfp_t mask
)
1070 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1071 cached_state
, mask
, NULL
);
1076 * convert_extent_bit - convert all bits in a given range from one bit to
1078 * @tree: the io tree to search
1079 * @start: the start offset in bytes
1080 * @end: the end offset in bytes (inclusive)
1081 * @bits: the bits to set in this range
1082 * @clear_bits: the bits to clear in this range
1083 * @cached_state: state that we're going to cache
1085 * This will go through and set bits for the given range. If any states exist
1086 * already in this range they are set with the given bit and cleared of the
1087 * clear_bits. This is only meant to be used by things that are mergeable, ie
1088 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1089 * boundary bits like LOCK.
1091 * All allocations are done with GFP_NOFS.
1093 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1094 unsigned bits
, unsigned clear_bits
,
1095 struct extent_state
**cached_state
)
1097 struct extent_state
*state
;
1098 struct extent_state
*prealloc
= NULL
;
1099 struct rb_node
*node
;
1101 struct rb_node
*parent
;
1105 bool first_iteration
= true;
1107 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1112 * Best effort, don't worry if extent state allocation fails
1113 * here for the first iteration. We might have a cached state
1114 * that matches exactly the target range, in which case no
1115 * extent state allocations are needed. We'll only know this
1116 * after locking the tree.
1118 prealloc
= alloc_extent_state(GFP_NOFS
);
1119 if (!prealloc
&& !first_iteration
)
1123 spin_lock(&tree
->lock
);
1124 if (cached_state
&& *cached_state
) {
1125 state
= *cached_state
;
1126 if (state
->start
<= start
&& state
->end
> start
&&
1127 extent_state_in_tree(state
)) {
1128 node
= &state
->rb_node
;
1134 * this search will find all the extents that end after
1137 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1139 prealloc
= alloc_extent_state_atomic(prealloc
);
1144 err
= insert_state(tree
, prealloc
, start
, end
,
1145 &p
, &parent
, &bits
, NULL
);
1147 extent_io_tree_panic(tree
, err
);
1148 cache_state(prealloc
, cached_state
);
1152 state
= rb_entry(node
, struct extent_state
, rb_node
);
1154 last_start
= state
->start
;
1155 last_end
= state
->end
;
1158 * | ---- desired range ---- |
1161 * Just lock what we found and keep going
1163 if (state
->start
== start
&& state
->end
<= end
) {
1164 set_state_bits(tree
, state
, &bits
, NULL
);
1165 cache_state(state
, cached_state
);
1166 state
= clear_state_bit(tree
, state
, &clear_bits
, 0, NULL
);
1167 if (last_end
== (u64
)-1)
1169 start
= last_end
+ 1;
1170 if (start
< end
&& state
&& state
->start
== start
&&
1177 * | ---- desired range ---- |
1180 * | ------------- state -------------- |
1182 * We need to split the extent we found, and may flip bits on
1185 * If the extent we found extends past our
1186 * range, we just split and search again. It'll get split
1187 * again the next time though.
1189 * If the extent we found is inside our range, we set the
1190 * desired bit on it.
1192 if (state
->start
< start
) {
1193 prealloc
= alloc_extent_state_atomic(prealloc
);
1198 err
= split_state(tree
, state
, prealloc
, start
);
1200 extent_io_tree_panic(tree
, err
);
1204 if (state
->end
<= end
) {
1205 set_state_bits(tree
, state
, &bits
, NULL
);
1206 cache_state(state
, cached_state
);
1207 state
= clear_state_bit(tree
, state
, &clear_bits
, 0,
1209 if (last_end
== (u64
)-1)
1211 start
= last_end
+ 1;
1212 if (start
< end
&& state
&& state
->start
== start
&&
1219 * | ---- desired range ---- |
1220 * | state | or | state |
1222 * There's a hole, we need to insert something in it and
1223 * ignore the extent we found.
1225 if (state
->start
> start
) {
1227 if (end
< last_start
)
1230 this_end
= last_start
- 1;
1232 prealloc
= alloc_extent_state_atomic(prealloc
);
1239 * Avoid to free 'prealloc' if it can be merged with
1242 err
= insert_state(tree
, prealloc
, start
, this_end
,
1243 NULL
, NULL
, &bits
, NULL
);
1245 extent_io_tree_panic(tree
, err
);
1246 cache_state(prealloc
, cached_state
);
1248 start
= this_end
+ 1;
1252 * | ---- desired range ---- |
1254 * We need to split the extent, and set the bit
1257 if (state
->start
<= end
&& state
->end
> end
) {
1258 prealloc
= alloc_extent_state_atomic(prealloc
);
1264 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1266 extent_io_tree_panic(tree
, err
);
1268 set_state_bits(tree
, prealloc
, &bits
, NULL
);
1269 cache_state(prealloc
, cached_state
);
1270 clear_state_bit(tree
, prealloc
, &clear_bits
, 0, NULL
);
1278 spin_unlock(&tree
->lock
);
1280 first_iteration
= false;
1284 spin_unlock(&tree
->lock
);
1286 free_extent_state(prealloc
);
1291 /* wrappers around set/clear extent bit */
1292 int set_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1293 unsigned bits
, struct extent_changeset
*changeset
)
1296 * We don't support EXTENT_LOCKED yet, as current changeset will
1297 * record any bits changed, so for EXTENT_LOCKED case, it will
1298 * either fail with -EEXIST or changeset will record the whole
1301 BUG_ON(bits
& EXTENT_LOCKED
);
1303 return __set_extent_bit(tree
, start
, end
, bits
, 0, NULL
, NULL
, GFP_NOFS
,
1307 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1308 unsigned bits
, int wake
, int delete,
1309 struct extent_state
**cached
, gfp_t mask
)
1311 return __clear_extent_bit(tree
, start
, end
, bits
, wake
, delete,
1312 cached
, mask
, NULL
);
1315 int clear_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1316 unsigned bits
, struct extent_changeset
*changeset
)
1319 * Don't support EXTENT_LOCKED case, same reason as
1320 * set_record_extent_bits().
1322 BUG_ON(bits
& EXTENT_LOCKED
);
1324 return __clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, GFP_NOFS
,
1329 * either insert or lock state struct between start and end use mask to tell
1330 * us if waiting is desired.
1332 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1333 struct extent_state
**cached_state
)
1339 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
,
1340 EXTENT_LOCKED
, &failed_start
,
1341 cached_state
, GFP_NOFS
, NULL
);
1342 if (err
== -EEXIST
) {
1343 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1344 start
= failed_start
;
1347 WARN_ON(start
> end
);
1352 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1357 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1358 &failed_start
, NULL
, GFP_NOFS
, NULL
);
1359 if (err
== -EEXIST
) {
1360 if (failed_start
> start
)
1361 clear_extent_bit(tree
, start
, failed_start
- 1,
1362 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1368 void extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1370 unsigned long index
= start
>> PAGE_SHIFT
;
1371 unsigned long end_index
= end
>> PAGE_SHIFT
;
1374 while (index
<= end_index
) {
1375 page
= find_get_page(inode
->i_mapping
, index
);
1376 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1377 clear_page_dirty_for_io(page
);
1383 void extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1385 unsigned long index
= start
>> PAGE_SHIFT
;
1386 unsigned long end_index
= end
>> PAGE_SHIFT
;
1389 while (index
<= end_index
) {
1390 page
= find_get_page(inode
->i_mapping
, index
);
1391 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1392 __set_page_dirty_nobuffers(page
);
1393 account_page_redirty(page
);
1400 * helper function to set both pages and extents in the tree writeback
1402 static void set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1404 unsigned long index
= start
>> PAGE_SHIFT
;
1405 unsigned long end_index
= end
>> PAGE_SHIFT
;
1408 while (index
<= end_index
) {
1409 page
= find_get_page(tree
->mapping
, index
);
1410 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1411 set_page_writeback(page
);
1417 /* find the first state struct with 'bits' set after 'start', and
1418 * return it. tree->lock must be held. NULL will returned if
1419 * nothing was found after 'start'
1421 static struct extent_state
*
1422 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1423 u64 start
, unsigned bits
)
1425 struct rb_node
*node
;
1426 struct extent_state
*state
;
1429 * this search will find all the extents that end after
1432 node
= tree_search(tree
, start
);
1437 state
= rb_entry(node
, struct extent_state
, rb_node
);
1438 if (state
->end
>= start
&& (state
->state
& bits
))
1441 node
= rb_next(node
);
1450 * find the first offset in the io tree with 'bits' set. zero is
1451 * returned if we find something, and *start_ret and *end_ret are
1452 * set to reflect the state struct that was found.
1454 * If nothing was found, 1 is returned. If found something, return 0.
1456 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1457 u64
*start_ret
, u64
*end_ret
, unsigned bits
,
1458 struct extent_state
**cached_state
)
1460 struct extent_state
*state
;
1464 spin_lock(&tree
->lock
);
1465 if (cached_state
&& *cached_state
) {
1466 state
= *cached_state
;
1467 if (state
->end
== start
- 1 && extent_state_in_tree(state
)) {
1468 n
= rb_next(&state
->rb_node
);
1470 state
= rb_entry(n
, struct extent_state
,
1472 if (state
->state
& bits
)
1476 free_extent_state(*cached_state
);
1477 *cached_state
= NULL
;
1480 free_extent_state(*cached_state
);
1481 *cached_state
= NULL
;
1484 state
= find_first_extent_bit_state(tree
, start
, bits
);
1487 cache_state_if_flags(state
, cached_state
, 0);
1488 *start_ret
= state
->start
;
1489 *end_ret
= state
->end
;
1493 spin_unlock(&tree
->lock
);
1498 * find a contiguous range of bytes in the file marked as delalloc, not
1499 * more than 'max_bytes'. start and end are used to return the range,
1501 * 1 is returned if we find something, 0 if nothing was in the tree
1503 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1504 u64
*start
, u64
*end
, u64 max_bytes
,
1505 struct extent_state
**cached_state
)
1507 struct rb_node
*node
;
1508 struct extent_state
*state
;
1509 u64 cur_start
= *start
;
1511 u64 total_bytes
= 0;
1513 spin_lock(&tree
->lock
);
1516 * this search will find all the extents that end after
1519 node
= tree_search(tree
, cur_start
);
1527 state
= rb_entry(node
, struct extent_state
, rb_node
);
1528 if (found
&& (state
->start
!= cur_start
||
1529 (state
->state
& EXTENT_BOUNDARY
))) {
1532 if (!(state
->state
& EXTENT_DELALLOC
)) {
1538 *start
= state
->start
;
1539 *cached_state
= state
;
1540 atomic_inc(&state
->refs
);
1544 cur_start
= state
->end
+ 1;
1545 node
= rb_next(node
);
1546 total_bytes
+= state
->end
- state
->start
+ 1;
1547 if (total_bytes
>= max_bytes
)
1553 spin_unlock(&tree
->lock
);
1557 static int __process_pages_contig(struct address_space
*mapping
,
1558 struct page
*locked_page
,
1559 pgoff_t start_index
, pgoff_t end_index
,
1560 unsigned long page_ops
, pgoff_t
*index_ret
);
1562 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1563 struct page
*locked_page
,
1566 unsigned long index
= start
>> PAGE_SHIFT
;
1567 unsigned long end_index
= end
>> PAGE_SHIFT
;
1569 ASSERT(locked_page
);
1570 if (index
== locked_page
->index
&& end_index
== index
)
1573 __process_pages_contig(inode
->i_mapping
, locked_page
, index
, end_index
,
1577 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1578 struct page
*locked_page
,
1582 unsigned long index
= delalloc_start
>> PAGE_SHIFT
;
1583 unsigned long index_ret
= index
;
1584 unsigned long end_index
= delalloc_end
>> PAGE_SHIFT
;
1587 ASSERT(locked_page
);
1588 if (index
== locked_page
->index
&& index
== end_index
)
1591 ret
= __process_pages_contig(inode
->i_mapping
, locked_page
, index
,
1592 end_index
, PAGE_LOCK
, &index_ret
);
1594 __unlock_for_delalloc(inode
, locked_page
, delalloc_start
,
1595 (u64
)index_ret
<< PAGE_SHIFT
);
1600 * find a contiguous range of bytes in the file marked as delalloc, not
1601 * more than 'max_bytes'. start and end are used to return the range,
1603 * 1 is returned if we find something, 0 if nothing was in the tree
1605 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1606 struct extent_io_tree
*tree
,
1607 struct page
*locked_page
, u64
*start
,
1608 u64
*end
, u64 max_bytes
)
1613 struct extent_state
*cached_state
= NULL
;
1618 /* step one, find a bunch of delalloc bytes starting at start */
1619 delalloc_start
= *start
;
1621 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1622 max_bytes
, &cached_state
);
1623 if (!found
|| delalloc_end
<= *start
) {
1624 *start
= delalloc_start
;
1625 *end
= delalloc_end
;
1626 free_extent_state(cached_state
);
1631 * start comes from the offset of locked_page. We have to lock
1632 * pages in order, so we can't process delalloc bytes before
1635 if (delalloc_start
< *start
)
1636 delalloc_start
= *start
;
1639 * make sure to limit the number of pages we try to lock down
1641 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1642 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1644 /* step two, lock all the pages after the page that has start */
1645 ret
= lock_delalloc_pages(inode
, locked_page
,
1646 delalloc_start
, delalloc_end
);
1647 if (ret
== -EAGAIN
) {
1648 /* some of the pages are gone, lets avoid looping by
1649 * shortening the size of the delalloc range we're searching
1651 free_extent_state(cached_state
);
1652 cached_state
= NULL
;
1654 max_bytes
= PAGE_SIZE
;
1662 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1664 /* step three, lock the state bits for the whole range */
1665 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, &cached_state
);
1667 /* then test to make sure it is all still delalloc */
1668 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1669 EXTENT_DELALLOC
, 1, cached_state
);
1671 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1672 &cached_state
, GFP_NOFS
);
1673 __unlock_for_delalloc(inode
, locked_page
,
1674 delalloc_start
, delalloc_end
);
1678 free_extent_state(cached_state
);
1679 *start
= delalloc_start
;
1680 *end
= delalloc_end
;
1685 static int __process_pages_contig(struct address_space
*mapping
,
1686 struct page
*locked_page
,
1687 pgoff_t start_index
, pgoff_t end_index
,
1688 unsigned long page_ops
, pgoff_t
*index_ret
)
1690 unsigned long nr_pages
= end_index
- start_index
+ 1;
1691 unsigned long pages_locked
= 0;
1692 pgoff_t index
= start_index
;
1693 struct page
*pages
[16];
1698 if (page_ops
& PAGE_LOCK
) {
1699 ASSERT(page_ops
== PAGE_LOCK
);
1700 ASSERT(index_ret
&& *index_ret
== start_index
);
1703 if ((page_ops
& PAGE_SET_ERROR
) && nr_pages
> 0)
1704 mapping_set_error(mapping
, -EIO
);
1706 while (nr_pages
> 0) {
1707 ret
= find_get_pages_contig(mapping
, index
,
1708 min_t(unsigned long,
1709 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1712 * Only if we're going to lock these pages,
1713 * can we find nothing at @index.
1715 ASSERT(page_ops
& PAGE_LOCK
);
1719 for (i
= 0; i
< ret
; i
++) {
1720 if (page_ops
& PAGE_SET_PRIVATE2
)
1721 SetPagePrivate2(pages
[i
]);
1723 if (pages
[i
] == locked_page
) {
1728 if (page_ops
& PAGE_CLEAR_DIRTY
)
1729 clear_page_dirty_for_io(pages
[i
]);
1730 if (page_ops
& PAGE_SET_WRITEBACK
)
1731 set_page_writeback(pages
[i
]);
1732 if (page_ops
& PAGE_SET_ERROR
)
1733 SetPageError(pages
[i
]);
1734 if (page_ops
& PAGE_END_WRITEBACK
)
1735 end_page_writeback(pages
[i
]);
1736 if (page_ops
& PAGE_UNLOCK
)
1737 unlock_page(pages
[i
]);
1738 if (page_ops
& PAGE_LOCK
) {
1739 lock_page(pages
[i
]);
1740 if (!PageDirty(pages
[i
]) ||
1741 pages
[i
]->mapping
!= mapping
) {
1742 unlock_page(pages
[i
]);
1756 if (err
&& index_ret
)
1757 *index_ret
= start_index
+ pages_locked
- 1;
1761 void extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1762 u64 delalloc_end
, struct page
*locked_page
,
1763 unsigned clear_bits
,
1764 unsigned long page_ops
)
1766 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, clear_bits
, 1, 0,
1769 __process_pages_contig(inode
->i_mapping
, locked_page
,
1770 start
>> PAGE_SHIFT
, end
>> PAGE_SHIFT
,
1775 * count the number of bytes in the tree that have a given bit(s)
1776 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1777 * cached. The total number found is returned.
1779 u64
count_range_bits(struct extent_io_tree
*tree
,
1780 u64
*start
, u64 search_end
, u64 max_bytes
,
1781 unsigned bits
, int contig
)
1783 struct rb_node
*node
;
1784 struct extent_state
*state
;
1785 u64 cur_start
= *start
;
1786 u64 total_bytes
= 0;
1790 if (WARN_ON(search_end
<= cur_start
))
1793 spin_lock(&tree
->lock
);
1794 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1795 total_bytes
= tree
->dirty_bytes
;
1799 * this search will find all the extents that end after
1802 node
= tree_search(tree
, cur_start
);
1807 state
= rb_entry(node
, struct extent_state
, rb_node
);
1808 if (state
->start
> search_end
)
1810 if (contig
&& found
&& state
->start
> last
+ 1)
1812 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1813 total_bytes
+= min(search_end
, state
->end
) + 1 -
1814 max(cur_start
, state
->start
);
1815 if (total_bytes
>= max_bytes
)
1818 *start
= max(cur_start
, state
->start
);
1822 } else if (contig
&& found
) {
1825 node
= rb_next(node
);
1830 spin_unlock(&tree
->lock
);
1835 * set the private field for a given byte offset in the tree. If there isn't
1836 * an extent_state there already, this does nothing.
1838 static noinline
int set_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1839 struct io_failure_record
*failrec
)
1841 struct rb_node
*node
;
1842 struct extent_state
*state
;
1845 spin_lock(&tree
->lock
);
1847 * this search will find all the extents that end after
1850 node
= tree_search(tree
, start
);
1855 state
= rb_entry(node
, struct extent_state
, rb_node
);
1856 if (state
->start
!= start
) {
1860 state
->failrec
= failrec
;
1862 spin_unlock(&tree
->lock
);
1866 static noinline
int get_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1867 struct io_failure_record
**failrec
)
1869 struct rb_node
*node
;
1870 struct extent_state
*state
;
1873 spin_lock(&tree
->lock
);
1875 * this search will find all the extents that end after
1878 node
= tree_search(tree
, start
);
1883 state
= rb_entry(node
, struct extent_state
, rb_node
);
1884 if (state
->start
!= start
) {
1888 *failrec
= state
->failrec
;
1890 spin_unlock(&tree
->lock
);
1895 * searches a range in the state tree for a given mask.
1896 * If 'filled' == 1, this returns 1 only if every extent in the tree
1897 * has the bits set. Otherwise, 1 is returned if any bit in the
1898 * range is found set.
1900 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1901 unsigned bits
, int filled
, struct extent_state
*cached
)
1903 struct extent_state
*state
= NULL
;
1904 struct rb_node
*node
;
1907 spin_lock(&tree
->lock
);
1908 if (cached
&& extent_state_in_tree(cached
) && cached
->start
<= start
&&
1909 cached
->end
> start
)
1910 node
= &cached
->rb_node
;
1912 node
= tree_search(tree
, start
);
1913 while (node
&& start
<= end
) {
1914 state
= rb_entry(node
, struct extent_state
, rb_node
);
1916 if (filled
&& state
->start
> start
) {
1921 if (state
->start
> end
)
1924 if (state
->state
& bits
) {
1928 } else if (filled
) {
1933 if (state
->end
== (u64
)-1)
1936 start
= state
->end
+ 1;
1939 node
= rb_next(node
);
1946 spin_unlock(&tree
->lock
);
1951 * helper function to set a given page up to date if all the
1952 * extents in the tree for that page are up to date
1954 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1956 u64 start
= page_offset(page
);
1957 u64 end
= start
+ PAGE_SIZE
- 1;
1958 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1959 SetPageUptodate(page
);
1962 int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
)
1966 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1968 set_state_failrec(failure_tree
, rec
->start
, NULL
);
1969 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1970 rec
->start
+ rec
->len
- 1,
1971 EXTENT_LOCKED
| EXTENT_DIRTY
);
1975 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1976 rec
->start
+ rec
->len
- 1,
1986 * this bypasses the standard btrfs submit functions deliberately, as
1987 * the standard behavior is to write all copies in a raid setup. here we only
1988 * want to write the one bad copy. so we do the mapping for ourselves and issue
1989 * submit_bio directly.
1990 * to avoid any synchronization issues, wait for the data after writing, which
1991 * actually prevents the read that triggered the error from finishing.
1992 * currently, there can be no more than two copies of every data bit. thus,
1993 * exactly one rewrite is required.
1995 int repair_io_failure(struct inode
*inode
, u64 start
, u64 length
, u64 logical
,
1996 struct page
*page
, unsigned int pg_offset
, int mirror_num
)
1998 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2000 struct btrfs_device
*dev
;
2003 struct btrfs_bio
*bbio
= NULL
;
2004 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2007 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2008 BUG_ON(!mirror_num
);
2010 /* we can't repair anything in raid56 yet */
2011 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2014 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2017 bio
->bi_iter
.bi_size
= 0;
2018 map_length
= length
;
2021 * Avoid races with device replace and make sure our bbio has devices
2022 * associated to its stripes that don't go away while we are doing the
2023 * read repair operation.
2025 btrfs_bio_counter_inc_blocked(fs_info
);
2026 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_WRITE
, logical
,
2027 &map_length
, &bbio
, mirror_num
);
2029 btrfs_bio_counter_dec(fs_info
);
2033 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2034 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2035 bio
->bi_iter
.bi_sector
= sector
;
2036 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2037 btrfs_put_bbio(bbio
);
2038 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2039 btrfs_bio_counter_dec(fs_info
);
2043 bio
->bi_bdev
= dev
->bdev
;
2044 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
;
2045 bio_add_page(bio
, page
, length
, pg_offset
);
2047 if (btrfsic_submit_bio_wait(bio
)) {
2048 /* try to remap that extent elsewhere? */
2049 btrfs_bio_counter_dec(fs_info
);
2051 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2055 btrfs_info_rl_in_rcu(fs_info
,
2056 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2057 btrfs_ino(BTRFS_I(inode
)), start
,
2058 rcu_str_deref(dev
->name
), sector
);
2059 btrfs_bio_counter_dec(fs_info
);
2064 int repair_eb_io_failure(struct btrfs_fs_info
*fs_info
,
2065 struct extent_buffer
*eb
, int mirror_num
)
2067 u64 start
= eb
->start
;
2068 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2071 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2074 for (i
= 0; i
< num_pages
; i
++) {
2075 struct page
*p
= eb
->pages
[i
];
2077 ret
= repair_io_failure(fs_info
->btree_inode
, start
,
2078 PAGE_SIZE
, start
, p
,
2079 start
- page_offset(p
), mirror_num
);
2089 * each time an IO finishes, we do a fast check in the IO failure tree
2090 * to see if we need to process or clean up an io_failure_record
2092 int clean_io_failure(struct inode
*inode
, u64 start
, struct page
*page
,
2093 unsigned int pg_offset
)
2096 struct io_failure_record
*failrec
;
2097 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2098 struct extent_state
*state
;
2103 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2104 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2108 ret
= get_state_failrec(&BTRFS_I(inode
)->io_failure_tree
, start
,
2113 BUG_ON(!failrec
->this_mirror
);
2115 if (failrec
->in_validation
) {
2116 /* there was no real error, just free the record */
2117 btrfs_debug(fs_info
,
2118 "clean_io_failure: freeing dummy error at %llu",
2122 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2125 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2126 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2129 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2131 if (state
&& state
->start
<= failrec
->start
&&
2132 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2133 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2135 if (num_copies
> 1) {
2136 repair_io_failure(inode
, start
, failrec
->len
,
2137 failrec
->logical
, page
,
2138 pg_offset
, failrec
->failed_mirror
);
2143 free_io_failure(inode
, failrec
);
2149 * Can be called when
2150 * - hold extent lock
2151 * - under ordered extent
2152 * - the inode is freeing
2154 void btrfs_free_io_failure_record(struct inode
*inode
, u64 start
, u64 end
)
2156 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2157 struct io_failure_record
*failrec
;
2158 struct extent_state
*state
, *next
;
2160 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2163 spin_lock(&failure_tree
->lock
);
2164 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2166 if (state
->start
> end
)
2169 ASSERT(state
->end
<= end
);
2171 next
= next_state(state
);
2173 failrec
= state
->failrec
;
2174 free_extent_state(state
);
2179 spin_unlock(&failure_tree
->lock
);
2182 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2183 struct io_failure_record
**failrec_ret
)
2185 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2186 struct io_failure_record
*failrec
;
2187 struct extent_map
*em
;
2188 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2189 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2190 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2194 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2196 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2200 failrec
->start
= start
;
2201 failrec
->len
= end
- start
+ 1;
2202 failrec
->this_mirror
= 0;
2203 failrec
->bio_flags
= 0;
2204 failrec
->in_validation
= 0;
2206 read_lock(&em_tree
->lock
);
2207 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2209 read_unlock(&em_tree
->lock
);
2214 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2215 free_extent_map(em
);
2218 read_unlock(&em_tree
->lock
);
2224 logical
= start
- em
->start
;
2225 logical
= em
->block_start
+ logical
;
2226 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2227 logical
= em
->block_start
;
2228 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2229 extent_set_compress_type(&failrec
->bio_flags
,
2233 btrfs_debug(fs_info
,
2234 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2235 logical
, start
, failrec
->len
);
2237 failrec
->logical
= logical
;
2238 free_extent_map(em
);
2240 /* set the bits in the private failure tree */
2241 ret
= set_extent_bits(failure_tree
, start
, end
,
2242 EXTENT_LOCKED
| EXTENT_DIRTY
);
2244 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2245 /* set the bits in the inode's tree */
2247 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2253 btrfs_debug(fs_info
,
2254 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2255 failrec
->logical
, failrec
->start
, failrec
->len
,
2256 failrec
->in_validation
);
2258 * when data can be on disk more than twice, add to failrec here
2259 * (e.g. with a list for failed_mirror) to make
2260 * clean_io_failure() clean all those errors at once.
2264 *failrec_ret
= failrec
;
2269 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2270 struct io_failure_record
*failrec
, int failed_mirror
)
2272 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2275 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
2276 if (num_copies
== 1) {
2278 * we only have a single copy of the data, so don't bother with
2279 * all the retry and error correction code that follows. no
2280 * matter what the error is, it is very likely to persist.
2282 btrfs_debug(fs_info
,
2283 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2284 num_copies
, failrec
->this_mirror
, failed_mirror
);
2289 * there are two premises:
2290 * a) deliver good data to the caller
2291 * b) correct the bad sectors on disk
2293 if (failed_bio
->bi_vcnt
> 1) {
2295 * to fulfill b), we need to know the exact failing sectors, as
2296 * we don't want to rewrite any more than the failed ones. thus,
2297 * we need separate read requests for the failed bio
2299 * if the following BUG_ON triggers, our validation request got
2300 * merged. we need separate requests for our algorithm to work.
2302 BUG_ON(failrec
->in_validation
);
2303 failrec
->in_validation
= 1;
2304 failrec
->this_mirror
= failed_mirror
;
2307 * we're ready to fulfill a) and b) alongside. get a good copy
2308 * of the failed sector and if we succeed, we have setup
2309 * everything for repair_io_failure to do the rest for us.
2311 if (failrec
->in_validation
) {
2312 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2313 failrec
->in_validation
= 0;
2314 failrec
->this_mirror
= 0;
2316 failrec
->failed_mirror
= failed_mirror
;
2317 failrec
->this_mirror
++;
2318 if (failrec
->this_mirror
== failed_mirror
)
2319 failrec
->this_mirror
++;
2322 if (failrec
->this_mirror
> num_copies
) {
2323 btrfs_debug(fs_info
,
2324 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2325 num_copies
, failrec
->this_mirror
, failed_mirror
);
2333 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2334 struct io_failure_record
*failrec
,
2335 struct page
*page
, int pg_offset
, int icsum
,
2336 bio_end_io_t
*endio_func
, void *data
)
2338 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2340 struct btrfs_io_bio
*btrfs_failed_bio
;
2341 struct btrfs_io_bio
*btrfs_bio
;
2343 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2347 bio
->bi_end_io
= endio_func
;
2348 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2349 bio
->bi_bdev
= fs_info
->fs_devices
->latest_bdev
;
2350 bio
->bi_iter
.bi_size
= 0;
2351 bio
->bi_private
= data
;
2353 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2354 if (btrfs_failed_bio
->csum
) {
2355 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2357 btrfs_bio
= btrfs_io_bio(bio
);
2358 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2360 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2364 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2370 * this is a generic handler for readpage errors (default
2371 * readpage_io_failed_hook). if other copies exist, read those and write back
2372 * good data to the failed position. does not investigate in remapping the
2373 * failed extent elsewhere, hoping the device will be smart enough to do this as
2377 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2378 struct page
*page
, u64 start
, u64 end
,
2381 struct io_failure_record
*failrec
;
2382 struct inode
*inode
= page
->mapping
->host
;
2383 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2388 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
2390 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2394 ret
= btrfs_check_repairable(inode
, failed_bio
, failrec
, failed_mirror
);
2396 free_io_failure(inode
, failrec
);
2400 if (failed_bio
->bi_vcnt
> 1)
2401 read_mode
|= REQ_FAILFAST_DEV
;
2403 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2404 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2405 start
- page_offset(page
),
2406 (int)phy_offset
, failed_bio
->bi_end_io
,
2409 free_io_failure(inode
, failrec
);
2412 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
2414 btrfs_debug(btrfs_sb(inode
->i_sb
),
2415 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2416 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2418 ret
= tree
->ops
->submit_bio_hook(inode
, bio
, failrec
->this_mirror
,
2419 failrec
->bio_flags
, 0);
2421 free_io_failure(inode
, failrec
);
2428 /* lots and lots of room for performance fixes in the end_bio funcs */
2430 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2432 int uptodate
= (err
== 0);
2433 struct extent_io_tree
*tree
;
2436 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2438 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2439 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2440 end
, NULL
, uptodate
);
2446 ClearPageUptodate(page
);
2448 ret
= ret
< 0 ? ret
: -EIO
;
2449 mapping_set_error(page
->mapping
, ret
);
2454 * after a writepage IO is done, we need to:
2455 * clear the uptodate bits on error
2456 * clear the writeback bits in the extent tree for this IO
2457 * end_page_writeback if the page has no more pending IO
2459 * Scheduling is not allowed, so the extent state tree is expected
2460 * to have one and only one object corresponding to this IO.
2462 static void end_bio_extent_writepage(struct bio
*bio
)
2464 struct bio_vec
*bvec
;
2469 bio_for_each_segment_all(bvec
, bio
, i
) {
2470 struct page
*page
= bvec
->bv_page
;
2471 struct inode
*inode
= page
->mapping
->host
;
2472 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2474 /* We always issue full-page reads, but if some block
2475 * in a page fails to read, blk_update_request() will
2476 * advance bv_offset and adjust bv_len to compensate.
2477 * Print a warning for nonzero offsets, and an error
2478 * if they don't add up to a full page. */
2479 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2480 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2482 "partial page write in btrfs with offset %u and length %u",
2483 bvec
->bv_offset
, bvec
->bv_len
);
2486 "incomplete page write in btrfs with offset %u and length %u",
2487 bvec
->bv_offset
, bvec
->bv_len
);
2490 start
= page_offset(page
);
2491 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2493 end_extent_writepage(page
, bio
->bi_error
, start
, end
);
2494 end_page_writeback(page
);
2501 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2504 struct extent_state
*cached
= NULL
;
2505 u64 end
= start
+ len
- 1;
2507 if (uptodate
&& tree
->track_uptodate
)
2508 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2509 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2513 * after a readpage IO is done, we need to:
2514 * clear the uptodate bits on error
2515 * set the uptodate bits if things worked
2516 * set the page up to date if all extents in the tree are uptodate
2517 * clear the lock bit in the extent tree
2518 * unlock the page if there are no other extents locked for it
2520 * Scheduling is not allowed, so the extent state tree is expected
2521 * to have one and only one object corresponding to this IO.
2523 static void end_bio_extent_readpage(struct bio
*bio
)
2525 struct bio_vec
*bvec
;
2526 int uptodate
= !bio
->bi_error
;
2527 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2528 struct extent_io_tree
*tree
;
2533 u64 extent_start
= 0;
2539 bio_for_each_segment_all(bvec
, bio
, i
) {
2540 struct page
*page
= bvec
->bv_page
;
2541 struct inode
*inode
= page
->mapping
->host
;
2542 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2544 btrfs_debug(fs_info
,
2545 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2546 (u64
)bio
->bi_iter
.bi_sector
, bio
->bi_error
,
2547 io_bio
->mirror_num
);
2548 tree
= &BTRFS_I(inode
)->io_tree
;
2550 /* We always issue full-page reads, but if some block
2551 * in a page fails to read, blk_update_request() will
2552 * advance bv_offset and adjust bv_len to compensate.
2553 * Print a warning for nonzero offsets, and an error
2554 * if they don't add up to a full page. */
2555 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2556 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2558 "partial page read in btrfs with offset %u and length %u",
2559 bvec
->bv_offset
, bvec
->bv_len
);
2562 "incomplete page read in btrfs with offset %u and length %u",
2563 bvec
->bv_offset
, bvec
->bv_len
);
2566 start
= page_offset(page
);
2567 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2570 mirror
= io_bio
->mirror_num
;
2571 if (likely(uptodate
&& tree
->ops
&&
2572 tree
->ops
->readpage_end_io_hook
)) {
2573 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2579 clean_io_failure(inode
, start
, page
, 0);
2582 if (likely(uptodate
))
2585 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2586 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2587 if (!ret
&& !bio
->bi_error
)
2591 * The generic bio_readpage_error handles errors the
2592 * following way: If possible, new read requests are
2593 * created and submitted and will end up in
2594 * end_bio_extent_readpage as well (if we're lucky, not
2595 * in the !uptodate case). In that case it returns 0 and
2596 * we just go on with the next page in our bio. If it
2597 * can't handle the error it will return -EIO and we
2598 * remain responsible for that page.
2600 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2603 uptodate
= !bio
->bi_error
;
2609 if (likely(uptodate
)) {
2610 loff_t i_size
= i_size_read(inode
);
2611 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2614 /* Zero out the end if this page straddles i_size */
2615 off
= i_size
& (PAGE_SIZE
-1);
2616 if (page
->index
== end_index
&& off
)
2617 zero_user_segment(page
, off
, PAGE_SIZE
);
2618 SetPageUptodate(page
);
2620 ClearPageUptodate(page
);
2626 if (unlikely(!uptodate
)) {
2628 endio_readpage_release_extent(tree
,
2634 endio_readpage_release_extent(tree
, start
,
2635 end
- start
+ 1, 0);
2636 } else if (!extent_len
) {
2637 extent_start
= start
;
2638 extent_len
= end
+ 1 - start
;
2639 } else if (extent_start
+ extent_len
== start
) {
2640 extent_len
+= end
+ 1 - start
;
2642 endio_readpage_release_extent(tree
, extent_start
,
2643 extent_len
, uptodate
);
2644 extent_start
= start
;
2645 extent_len
= end
+ 1 - start
;
2650 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2653 io_bio
->end_io(io_bio
, bio
->bi_error
);
2658 * this allocates from the btrfs_bioset. We're returning a bio right now
2659 * but you can call btrfs_io_bio for the appropriate container_of magic
2662 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2665 struct btrfs_io_bio
*btrfs_bio
;
2668 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2670 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2671 while (!bio
&& (nr_vecs
/= 2)) {
2672 bio
= bio_alloc_bioset(gfp_flags
,
2673 nr_vecs
, btrfs_bioset
);
2678 bio
->bi_bdev
= bdev
;
2679 bio
->bi_iter
.bi_sector
= first_sector
;
2680 btrfs_bio
= btrfs_io_bio(bio
);
2681 btrfs_bio
->csum
= NULL
;
2682 btrfs_bio
->csum_allocated
= NULL
;
2683 btrfs_bio
->end_io
= NULL
;
2688 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2690 struct btrfs_io_bio
*btrfs_bio
;
2693 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2695 btrfs_bio
= btrfs_io_bio(new);
2696 btrfs_bio
->csum
= NULL
;
2697 btrfs_bio
->csum_allocated
= NULL
;
2698 btrfs_bio
->end_io
= NULL
;
2703 /* this also allocates from the btrfs_bioset */
2704 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2706 struct btrfs_io_bio
*btrfs_bio
;
2709 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2711 btrfs_bio
= btrfs_io_bio(bio
);
2712 btrfs_bio
->csum
= NULL
;
2713 btrfs_bio
->csum_allocated
= NULL
;
2714 btrfs_bio
->end_io
= NULL
;
2720 static int __must_check
submit_one_bio(struct bio
*bio
, int mirror_num
,
2721 unsigned long bio_flags
)
2724 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2725 struct page
*page
= bvec
->bv_page
;
2726 struct extent_io_tree
*tree
= bio
->bi_private
;
2729 start
= page_offset(page
) + bvec
->bv_offset
;
2731 bio
->bi_private
= NULL
;
2734 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2735 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, bio
,
2736 mirror_num
, bio_flags
, start
);
2738 btrfsic_submit_bio(bio
);
2744 static int merge_bio(struct extent_io_tree
*tree
, struct page
*page
,
2745 unsigned long offset
, size_t size
, struct bio
*bio
,
2746 unsigned long bio_flags
)
2749 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2750 ret
= tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
,
2756 static int submit_extent_page(int op
, int op_flags
, struct extent_io_tree
*tree
,
2757 struct writeback_control
*wbc
,
2758 struct page
*page
, sector_t sector
,
2759 size_t size
, unsigned long offset
,
2760 struct block_device
*bdev
,
2761 struct bio
**bio_ret
,
2762 bio_end_io_t end_io_func
,
2764 unsigned long prev_bio_flags
,
2765 unsigned long bio_flags
,
2766 bool force_bio_submit
)
2771 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2772 size_t page_size
= min_t(size_t, size
, PAGE_SIZE
);
2774 if (bio_ret
&& *bio_ret
) {
2777 contig
= bio
->bi_iter
.bi_sector
== sector
;
2779 contig
= bio_end_sector(bio
) == sector
;
2781 if (prev_bio_flags
!= bio_flags
|| !contig
||
2783 merge_bio(tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2784 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2785 ret
= submit_one_bio(bio
, mirror_num
, prev_bio_flags
);
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
;
2806 bio_set_op_attrs(bio
, op
, op_flags
);
2808 wbc_init_bio(wbc
, bio
);
2809 wbc_account_io(wbc
, page
, page_size
);
2815 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
2820 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2823 if (!PagePrivate(page
)) {
2824 SetPagePrivate(page
);
2826 set_page_private(page
, (unsigned long)eb
);
2828 WARN_ON(page
->private != (unsigned long)eb
);
2832 void set_page_extent_mapped(struct page
*page
)
2834 if (!PagePrivate(page
)) {
2835 SetPagePrivate(page
);
2837 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2841 static struct extent_map
*
2842 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2843 u64 start
, u64 len
, get_extent_t
*get_extent
,
2844 struct extent_map
**em_cached
)
2846 struct extent_map
*em
;
2848 if (em_cached
&& *em_cached
) {
2850 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2851 start
< extent_map_end(em
)) {
2852 atomic_inc(&em
->refs
);
2856 free_extent_map(em
);
2860 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2861 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2863 atomic_inc(&em
->refs
);
2869 * basic readpage implementation. Locked extent state structs are inserted
2870 * into the tree that are removed when the IO is done (by the end_io
2872 * XXX JDM: This needs looking at to ensure proper page locking
2873 * return 0 on success, otherwise return error
2875 static int __do_readpage(struct extent_io_tree
*tree
,
2877 get_extent_t
*get_extent
,
2878 struct extent_map
**em_cached
,
2879 struct bio
**bio
, int mirror_num
,
2880 unsigned long *bio_flags
, int read_flags
,
2883 struct inode
*inode
= page
->mapping
->host
;
2884 u64 start
= page_offset(page
);
2885 u64 page_end
= start
+ PAGE_SIZE
- 1;
2889 u64 last_byte
= i_size_read(inode
);
2893 struct extent_map
*em
;
2894 struct block_device
*bdev
;
2897 size_t pg_offset
= 0;
2899 size_t disk_io_size
;
2900 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2901 unsigned long this_bio_flag
= 0;
2903 set_page_extent_mapped(page
);
2906 if (!PageUptodate(page
)) {
2907 if (cleancache_get_page(page
) == 0) {
2908 BUG_ON(blocksize
!= PAGE_SIZE
);
2909 unlock_extent(tree
, start
, end
);
2914 if (page
->index
== last_byte
>> PAGE_SHIFT
) {
2916 size_t zero_offset
= last_byte
& (PAGE_SIZE
- 1);
2919 iosize
= PAGE_SIZE
- zero_offset
;
2920 userpage
= kmap_atomic(page
);
2921 memset(userpage
+ zero_offset
, 0, iosize
);
2922 flush_dcache_page(page
);
2923 kunmap_atomic(userpage
);
2926 while (cur
<= end
) {
2927 bool force_bio_submit
= false;
2929 if (cur
>= last_byte
) {
2931 struct extent_state
*cached
= NULL
;
2933 iosize
= PAGE_SIZE
- pg_offset
;
2934 userpage
= kmap_atomic(page
);
2935 memset(userpage
+ pg_offset
, 0, iosize
);
2936 flush_dcache_page(page
);
2937 kunmap_atomic(userpage
);
2938 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2940 unlock_extent_cached(tree
, cur
,
2945 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2946 end
- cur
+ 1, get_extent
, em_cached
);
2947 if (IS_ERR_OR_NULL(em
)) {
2949 unlock_extent(tree
, cur
, end
);
2952 extent_offset
= cur
- em
->start
;
2953 BUG_ON(extent_map_end(em
) <= cur
);
2956 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2957 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2958 extent_set_compress_type(&this_bio_flag
,
2962 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2963 cur_end
= min(extent_map_end(em
) - 1, end
);
2964 iosize
= ALIGN(iosize
, blocksize
);
2965 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2966 disk_io_size
= em
->block_len
;
2967 sector
= em
->block_start
>> 9;
2969 sector
= (em
->block_start
+ extent_offset
) >> 9;
2970 disk_io_size
= iosize
;
2973 block_start
= em
->block_start
;
2974 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2975 block_start
= EXTENT_MAP_HOLE
;
2978 * If we have a file range that points to a compressed extent
2979 * and it's followed by a consecutive file range that points to
2980 * to the same compressed extent (possibly with a different
2981 * offset and/or length, so it either points to the whole extent
2982 * or only part of it), we must make sure we do not submit a
2983 * single bio to populate the pages for the 2 ranges because
2984 * this makes the compressed extent read zero out the pages
2985 * belonging to the 2nd range. Imagine the following scenario:
2988 * [0 - 8K] [8K - 24K]
2991 * points to extent X, points to extent X,
2992 * offset 4K, length of 8K offset 0, length 16K
2994 * [extent X, compressed length = 4K uncompressed length = 16K]
2996 * If the bio to read the compressed extent covers both ranges,
2997 * it will decompress extent X into the pages belonging to the
2998 * first range and then it will stop, zeroing out the remaining
2999 * pages that belong to the other range that points to extent X.
3000 * So here we make sure we submit 2 bios, one for the first
3001 * range and another one for the third range. Both will target
3002 * the same physical extent from disk, but we can't currently
3003 * make the compressed bio endio callback populate the pages
3004 * for both ranges because each compressed bio is tightly
3005 * coupled with a single extent map, and each range can have
3006 * an extent map with a different offset value relative to the
3007 * uncompressed data of our extent and different lengths. This
3008 * is a corner case so we prioritize correctness over
3009 * non-optimal behavior (submitting 2 bios for the same extent).
3011 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3012 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3013 *prev_em_start
!= em
->orig_start
)
3014 force_bio_submit
= true;
3017 *prev_em_start
= em
->orig_start
;
3019 free_extent_map(em
);
3022 /* we've found a hole, just zero and go on */
3023 if (block_start
== EXTENT_MAP_HOLE
) {
3025 struct extent_state
*cached
= NULL
;
3027 userpage
= kmap_atomic(page
);
3028 memset(userpage
+ pg_offset
, 0, iosize
);
3029 flush_dcache_page(page
);
3030 kunmap_atomic(userpage
);
3032 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3034 unlock_extent_cached(tree
, cur
,
3038 pg_offset
+= iosize
;
3041 /* the get_extent function already copied into the page */
3042 if (test_range_bit(tree
, cur
, cur_end
,
3043 EXTENT_UPTODATE
, 1, NULL
)) {
3044 check_page_uptodate(tree
, page
);
3045 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3047 pg_offset
+= iosize
;
3050 /* we have an inline extent but it didn't get marked up
3051 * to date. Error out
3053 if (block_start
== EXTENT_MAP_INLINE
) {
3055 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3057 pg_offset
+= iosize
;
3061 ret
= submit_extent_page(REQ_OP_READ
, read_flags
, tree
, NULL
,
3062 page
, 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);
3077 pg_offset
+= iosize
;
3081 if (!PageError(page
))
3082 SetPageUptodate(page
);
3088 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3089 struct page
*pages
[], int nr_pages
,
3091 get_extent_t
*get_extent
,
3092 struct extent_map
**em_cached
,
3093 struct bio
**bio
, int mirror_num
,
3094 unsigned long *bio_flags
,
3097 struct inode
*inode
;
3098 struct btrfs_ordered_extent
*ordered
;
3101 inode
= pages
[0]->mapping
->host
;
3103 lock_extent(tree
, start
, end
);
3104 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3108 unlock_extent(tree
, start
, end
);
3109 btrfs_start_ordered_extent(inode
, ordered
, 1);
3110 btrfs_put_ordered_extent(ordered
);
3113 for (index
= 0; index
< nr_pages
; index
++) {
3114 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3115 mirror_num
, bio_flags
, 0, prev_em_start
);
3116 put_page(pages
[index
]);
3120 static void __extent_readpages(struct extent_io_tree
*tree
,
3121 struct page
*pages
[],
3122 int nr_pages
, get_extent_t
*get_extent
,
3123 struct extent_map
**em_cached
,
3124 struct bio
**bio
, int mirror_num
,
3125 unsigned long *bio_flags
,
3132 int first_index
= 0;
3134 for (index
= 0; index
< nr_pages
; index
++) {
3135 page_start
= page_offset(pages
[index
]);
3138 end
= start
+ PAGE_SIZE
- 1;
3139 first_index
= index
;
3140 } else if (end
+ 1 == page_start
) {
3143 __do_contiguous_readpages(tree
, &pages
[first_index
],
3144 index
- first_index
, start
,
3145 end
, get_extent
, em_cached
,
3146 bio
, mirror_num
, bio_flags
,
3149 end
= start
+ PAGE_SIZE
- 1;
3150 first_index
= index
;
3155 __do_contiguous_readpages(tree
, &pages
[first_index
],
3156 index
- first_index
, start
,
3157 end
, get_extent
, em_cached
, bio
,
3158 mirror_num
, bio_flags
,
3162 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3164 get_extent_t
*get_extent
,
3165 struct bio
**bio
, int mirror_num
,
3166 unsigned long *bio_flags
, int read_flags
)
3168 struct inode
*inode
= page
->mapping
->host
;
3169 struct btrfs_ordered_extent
*ordered
;
3170 u64 start
= page_offset(page
);
3171 u64 end
= start
+ PAGE_SIZE
- 1;
3175 lock_extent(tree
, start
, end
);
3176 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3180 unlock_extent(tree
, start
, end
);
3181 btrfs_start_ordered_extent(inode
, ordered
, 1);
3182 btrfs_put_ordered_extent(ordered
);
3185 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3186 bio_flags
, read_flags
, NULL
);
3190 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3191 get_extent_t
*get_extent
, int mirror_num
)
3193 struct bio
*bio
= NULL
;
3194 unsigned long bio_flags
= 0;
3197 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3200 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
3204 static void update_nr_written(struct writeback_control
*wbc
,
3205 unsigned long nr_written
)
3207 wbc
->nr_to_write
-= nr_written
;
3211 * helper for __extent_writepage, doing all of the delayed allocation setup.
3213 * This returns 1 if our fill_delalloc function did all the work required
3214 * to write the page (copy into inline extent). In this case the IO has
3215 * been started and the page is already unlocked.
3217 * This returns 0 if all went well (page still locked)
3218 * This returns < 0 if there were errors (page still locked)
3220 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3221 struct page
*page
, struct writeback_control
*wbc
,
3222 struct extent_page_data
*epd
,
3224 unsigned long *nr_written
)
3226 struct extent_io_tree
*tree
= epd
->tree
;
3227 u64 page_end
= delalloc_start
+ PAGE_SIZE
- 1;
3229 u64 delalloc_to_write
= 0;
3230 u64 delalloc_end
= 0;
3232 int page_started
= 0;
3234 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3237 while (delalloc_end
< page_end
) {
3238 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3242 BTRFS_MAX_EXTENT_SIZE
);
3243 if (nr_delalloc
== 0) {
3244 delalloc_start
= delalloc_end
+ 1;
3247 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3252 /* File system has been set read-only */
3255 /* fill_delalloc should be return < 0 for error
3256 * but just in case, we use > 0 here meaning the
3257 * IO is started, so we don't want to return > 0
3258 * unless things are going well.
3260 ret
= ret
< 0 ? ret
: -EIO
;
3264 * delalloc_end is already one less than the total length, so
3265 * we don't subtract one from PAGE_SIZE
3267 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3268 PAGE_SIZE
) >> PAGE_SHIFT
;
3269 delalloc_start
= delalloc_end
+ 1;
3271 if (wbc
->nr_to_write
< delalloc_to_write
) {
3274 if (delalloc_to_write
< thresh
* 2)
3275 thresh
= delalloc_to_write
;
3276 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3280 /* did the fill delalloc function already unlock and start
3285 * we've unlocked the page, so we can't update
3286 * the mapping's writeback index, just update
3289 wbc
->nr_to_write
-= *nr_written
;
3300 * helper for __extent_writepage. This calls the writepage start hooks,
3301 * and does the loop to map the page into extents and bios.
3303 * We return 1 if the IO is started and the page is unlocked,
3304 * 0 if all went well (page still locked)
3305 * < 0 if there were errors (page still locked)
3307 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3309 struct writeback_control
*wbc
,
3310 struct extent_page_data
*epd
,
3312 unsigned long nr_written
,
3313 int write_flags
, int *nr_ret
)
3315 struct extent_io_tree
*tree
= epd
->tree
;
3316 u64 start
= page_offset(page
);
3317 u64 page_end
= start
+ PAGE_SIZE
- 1;
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(wbc
, nr_written
);
3349 * we don't want to touch the inode after unlocking the page,
3350 * so we update the mapping writeback index now
3352 update_nr_written(wbc
, nr_written
+ 1);
3355 if (i_size
<= start
) {
3356 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3357 tree
->ops
->writepage_end_io_hook(page
, start
,
3362 blocksize
= inode
->i_sb
->s_blocksize
;
3364 while (cur
<= end
) {
3367 if (cur
>= i_size
) {
3368 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3369 tree
->ops
->writepage_end_io_hook(page
, cur
,
3373 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3375 if (IS_ERR_OR_NULL(em
)) {
3377 ret
= PTR_ERR_OR_ZERO(em
);
3381 extent_offset
= cur
- em
->start
;
3382 em_end
= extent_map_end(em
);
3383 BUG_ON(em_end
<= cur
);
3385 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3386 iosize
= ALIGN(iosize
, blocksize
);
3387 sector
= (em
->block_start
+ extent_offset
) >> 9;
3389 block_start
= em
->block_start
;
3390 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3391 free_extent_map(em
);
3395 * compressed and inline extents are written through other
3398 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3399 block_start
== EXTENT_MAP_INLINE
) {
3401 * end_io notification does not happen here for
3402 * compressed extents
3404 if (!compressed
&& tree
->ops
&&
3405 tree
->ops
->writepage_end_io_hook
)
3406 tree
->ops
->writepage_end_io_hook(page
, cur
,
3409 else if (compressed
) {
3410 /* we don't want to end_page_writeback on
3411 * a compressed extent. this happens
3418 pg_offset
+= iosize
;
3422 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3423 if (!PageWriteback(page
)) {
3424 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3425 "page %lu not writeback, cur %llu end %llu",
3426 page
->index
, cur
, end
);
3429 ret
= submit_extent_page(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3430 page
, sector
, iosize
, pg_offset
,
3432 end_bio_extent_writepage
,
3436 if (PageWriteback(page
))
3437 end_page_writeback(page
);
3441 pg_offset
+= iosize
;
3450 * the writepage semantics are similar to regular writepage. extent
3451 * records are inserted to lock ranges in the tree, and as dirty areas
3452 * are found, they are marked writeback. Then the lock bits are removed
3453 * and the end_io handler clears the writeback ranges
3455 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3458 struct inode
*inode
= page
->mapping
->host
;
3459 struct extent_page_data
*epd
= data
;
3460 u64 start
= page_offset(page
);
3461 u64 page_end
= start
+ PAGE_SIZE
- 1;
3464 size_t pg_offset
= 0;
3465 loff_t i_size
= i_size_read(inode
);
3466 unsigned long end_index
= i_size
>> PAGE_SHIFT
;
3467 int write_flags
= 0;
3468 unsigned long nr_written
= 0;
3470 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3471 write_flags
= REQ_SYNC
;
3473 trace___extent_writepage(page
, inode
, wbc
);
3475 WARN_ON(!PageLocked(page
));
3477 ClearPageError(page
);
3479 pg_offset
= i_size
& (PAGE_SIZE
- 1);
3480 if (page
->index
> end_index
||
3481 (page
->index
== end_index
&& !pg_offset
)) {
3482 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3487 if (page
->index
== end_index
) {
3490 userpage
= kmap_atomic(page
);
3491 memset(userpage
+ pg_offset
, 0,
3492 PAGE_SIZE
- pg_offset
);
3493 kunmap_atomic(userpage
);
3494 flush_dcache_page(page
);
3499 set_page_extent_mapped(page
);
3501 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3507 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3508 i_size
, nr_written
, write_flags
, &nr
);
3514 /* make sure the mapping tag for page dirty gets cleared */
3515 set_page_writeback(page
);
3516 end_page_writeback(page
);
3518 if (PageError(page
)) {
3519 ret
= ret
< 0 ? ret
: -EIO
;
3520 end_extent_writepage(page
, ret
, start
, page_end
);
3529 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3531 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3532 TASK_UNINTERRUPTIBLE
);
3535 static noinline_for_stack
int
3536 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3537 struct btrfs_fs_info
*fs_info
,
3538 struct extent_page_data
*epd
)
3540 unsigned long i
, num_pages
;
3544 if (!btrfs_try_tree_write_lock(eb
)) {
3546 flush_write_bio(epd
);
3547 btrfs_tree_lock(eb
);
3550 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3551 btrfs_tree_unlock(eb
);
3555 flush_write_bio(epd
);
3559 wait_on_extent_buffer_writeback(eb
);
3560 btrfs_tree_lock(eb
);
3561 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3563 btrfs_tree_unlock(eb
);
3568 * We need to do this to prevent races in people who check if the eb is
3569 * under IO since we can end up having no IO bits set for a short period
3572 spin_lock(&eb
->refs_lock
);
3573 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3574 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3575 spin_unlock(&eb
->refs_lock
);
3576 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3577 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3579 fs_info
->dirty_metadata_batch
);
3582 spin_unlock(&eb
->refs_lock
);
3585 btrfs_tree_unlock(eb
);
3590 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3591 for (i
= 0; i
< num_pages
; i
++) {
3592 struct page
*p
= eb
->pages
[i
];
3594 if (!trylock_page(p
)) {
3596 flush_write_bio(epd
);
3606 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3608 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3609 smp_mb__after_atomic();
3610 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3613 static void set_btree_ioerr(struct page
*page
)
3615 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3618 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3622 * If writeback for a btree extent that doesn't belong to a log tree
3623 * failed, increment the counter transaction->eb_write_errors.
3624 * We do this because while the transaction is running and before it's
3625 * committing (when we call filemap_fdata[write|wait]_range against
3626 * the btree inode), we might have
3627 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3628 * returns an error or an error happens during writeback, when we're
3629 * committing the transaction we wouldn't know about it, since the pages
3630 * can be no longer dirty nor marked anymore for writeback (if a
3631 * subsequent modification to the extent buffer didn't happen before the
3632 * transaction commit), which makes filemap_fdata[write|wait]_range not
3633 * able to find the pages tagged with SetPageError at transaction
3634 * commit time. So if this happens we must abort the transaction,
3635 * otherwise we commit a super block with btree roots that point to
3636 * btree nodes/leafs whose content on disk is invalid - either garbage
3637 * or the content of some node/leaf from a past generation that got
3638 * cowed or deleted and is no longer valid.
3640 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3641 * not be enough - we need to distinguish between log tree extents vs
3642 * non-log tree extents, and the next filemap_fdatawait_range() call
3643 * will catch and clear such errors in the mapping - and that call might
3644 * be from a log sync and not from a transaction commit. Also, checking
3645 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3646 * not done and would not be reliable - the eb might have been released
3647 * from memory and reading it back again means that flag would not be
3648 * set (since it's a runtime flag, not persisted on disk).
3650 * Using the flags below in the btree inode also makes us achieve the
3651 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3652 * writeback for all dirty pages and before filemap_fdatawait_range()
3653 * is called, the writeback for all dirty pages had already finished
3654 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3655 * filemap_fdatawait_range() would return success, as it could not know
3656 * that writeback errors happened (the pages were no longer tagged for
3659 switch (eb
->log_index
) {
3661 set_bit(BTRFS_FS_BTREE_ERR
, &eb
->fs_info
->flags
);
3664 set_bit(BTRFS_FS_LOG1_ERR
, &eb
->fs_info
->flags
);
3667 set_bit(BTRFS_FS_LOG2_ERR
, &eb
->fs_info
->flags
);
3670 BUG(); /* unexpected, logic error */
3674 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3676 struct bio_vec
*bvec
;
3677 struct extent_buffer
*eb
;
3680 bio_for_each_segment_all(bvec
, bio
, i
) {
3681 struct page
*page
= bvec
->bv_page
;
3683 eb
= (struct extent_buffer
*)page
->private;
3685 done
= atomic_dec_and_test(&eb
->io_pages
);
3687 if (bio
->bi_error
||
3688 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3689 ClearPageUptodate(page
);
3690 set_btree_ioerr(page
);
3693 end_page_writeback(page
);
3698 end_extent_buffer_writeback(eb
);
3704 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3705 struct btrfs_fs_info
*fs_info
,
3706 struct writeback_control
*wbc
,
3707 struct extent_page_data
*epd
)
3709 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3710 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3711 u64 offset
= eb
->start
;
3713 unsigned long i
, num_pages
;
3714 unsigned long bio_flags
= 0;
3715 unsigned long start
, end
;
3716 int write_flags
= (epd
->sync_io
? REQ_SYNC
: 0) | REQ_META
;
3719 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3720 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3721 atomic_set(&eb
->io_pages
, num_pages
);
3722 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3723 bio_flags
= EXTENT_BIO_TREE_LOG
;
3725 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3726 nritems
= btrfs_header_nritems(eb
);
3727 if (btrfs_header_level(eb
) > 0) {
3728 end
= btrfs_node_key_ptr_offset(nritems
);
3730 memzero_extent_buffer(eb
, end
, eb
->len
- end
);
3734 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3736 start
= btrfs_item_nr_offset(nritems
);
3737 end
= btrfs_leaf_data(eb
) + leaf_data_end(fs_info
, eb
);
3738 memzero_extent_buffer(eb
, start
, end
- start
);
3741 for (i
= 0; i
< num_pages
; i
++) {
3742 struct page
*p
= eb
->pages
[i
];
3744 clear_page_dirty_for_io(p
);
3745 set_page_writeback(p
);
3746 ret
= submit_extent_page(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3747 p
, offset
>> 9, PAGE_SIZE
, 0, bdev
,
3749 end_bio_extent_buffer_writepage
,
3750 0, epd
->bio_flags
, bio_flags
, false);
3751 epd
->bio_flags
= bio_flags
;
3754 if (PageWriteback(p
))
3755 end_page_writeback(p
);
3756 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3757 end_extent_buffer_writeback(eb
);
3761 offset
+= PAGE_SIZE
;
3762 update_nr_written(wbc
, 1);
3766 if (unlikely(ret
)) {
3767 for (; i
< num_pages
; i
++) {
3768 struct page
*p
= eb
->pages
[i
];
3769 clear_page_dirty_for_io(p
);
3777 int btree_write_cache_pages(struct address_space
*mapping
,
3778 struct writeback_control
*wbc
)
3780 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3781 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3782 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3783 struct extent_page_data epd
= {
3787 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3792 int nr_to_write_done
= 0;
3793 struct pagevec pvec
;
3796 pgoff_t end
; /* Inclusive */
3800 pagevec_init(&pvec
, 0);
3801 if (wbc
->range_cyclic
) {
3802 index
= mapping
->writeback_index
; /* Start from prev offset */
3805 index
= wbc
->range_start
>> PAGE_SHIFT
;
3806 end
= wbc
->range_end
>> PAGE_SHIFT
;
3809 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3810 tag
= PAGECACHE_TAG_TOWRITE
;
3812 tag
= PAGECACHE_TAG_DIRTY
;
3814 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3815 tag_pages_for_writeback(mapping
, index
, end
);
3816 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3817 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3818 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3822 for (i
= 0; i
< nr_pages
; i
++) {
3823 struct page
*page
= pvec
.pages
[i
];
3825 if (!PagePrivate(page
))
3828 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3833 spin_lock(&mapping
->private_lock
);
3834 if (!PagePrivate(page
)) {
3835 spin_unlock(&mapping
->private_lock
);
3839 eb
= (struct extent_buffer
*)page
->private;
3842 * Shouldn't happen and normally this would be a BUG_ON
3843 * but no sense in crashing the users box for something
3844 * we can survive anyway.
3847 spin_unlock(&mapping
->private_lock
);
3851 if (eb
== prev_eb
) {
3852 spin_unlock(&mapping
->private_lock
);
3856 ret
= atomic_inc_not_zero(&eb
->refs
);
3857 spin_unlock(&mapping
->private_lock
);
3862 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3864 free_extent_buffer(eb
);
3868 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3871 free_extent_buffer(eb
);
3874 free_extent_buffer(eb
);
3877 * the filesystem may choose to bump up nr_to_write.
3878 * We have to make sure to honor the new nr_to_write
3881 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3883 pagevec_release(&pvec
);
3886 if (!scanned
&& !done
) {
3888 * We hit the last page and there is more work to be done: wrap
3889 * back to the start of the file
3895 flush_write_bio(&epd
);
3900 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3901 * @mapping: address space structure to write
3902 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3903 * @writepage: function called for each page
3904 * @data: data passed to writepage function
3906 * If a page is already under I/O, write_cache_pages() skips it, even
3907 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3908 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3909 * and msync() need to guarantee that all the data which was dirty at the time
3910 * the call was made get new I/O started against them. If wbc->sync_mode is
3911 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3912 * existing IO to complete.
3914 static int extent_write_cache_pages(struct address_space
*mapping
,
3915 struct writeback_control
*wbc
,
3916 writepage_t writepage
, void *data
,
3917 void (*flush_fn
)(void *))
3919 struct inode
*inode
= mapping
->host
;
3922 int nr_to_write_done
= 0;
3923 struct pagevec pvec
;
3926 pgoff_t end
; /* Inclusive */
3928 int range_whole
= 0;
3933 * We have to hold onto the inode so that ordered extents can do their
3934 * work when the IO finishes. The alternative to this is failing to add
3935 * an ordered extent if the igrab() fails there and that is a huge pain
3936 * to deal with, so instead just hold onto the inode throughout the
3937 * writepages operation. If it fails here we are freeing up the inode
3938 * anyway and we'd rather not waste our time writing out stuff that is
3939 * going to be truncated anyway.
3944 pagevec_init(&pvec
, 0);
3945 if (wbc
->range_cyclic
) {
3946 index
= mapping
->writeback_index
; /* Start from prev offset */
3949 index
= wbc
->range_start
>> PAGE_SHIFT
;
3950 end
= wbc
->range_end
>> PAGE_SHIFT
;
3951 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3955 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3956 tag
= PAGECACHE_TAG_TOWRITE
;
3958 tag
= PAGECACHE_TAG_DIRTY
;
3960 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3961 tag_pages_for_writeback(mapping
, index
, end
);
3963 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3964 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3965 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3969 for (i
= 0; i
< nr_pages
; i
++) {
3970 struct page
*page
= pvec
.pages
[i
];
3972 done_index
= page
->index
;
3974 * At this point we hold neither mapping->tree_lock nor
3975 * lock on the page itself: the page may be truncated or
3976 * invalidated (changing page->mapping to NULL), or even
3977 * swizzled back from swapper_space to tmpfs file
3980 if (!trylock_page(page
)) {
3985 if (unlikely(page
->mapping
!= mapping
)) {
3990 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3996 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3997 if (PageWriteback(page
))
3999 wait_on_page_writeback(page
);
4002 if (PageWriteback(page
) ||
4003 !clear_page_dirty_for_io(page
)) {
4008 ret
= (*writepage
)(page
, wbc
, data
);
4010 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4016 * done_index is set past this page,
4017 * so media errors will not choke
4018 * background writeout for the entire
4019 * file. This has consequences for
4020 * range_cyclic semantics (ie. it may
4021 * not be suitable for data integrity
4024 done_index
= page
->index
+ 1;
4030 * the filesystem may choose to bump up nr_to_write.
4031 * We have to make sure to honor the new nr_to_write
4034 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4036 pagevec_release(&pvec
);
4039 if (!scanned
&& !done
) {
4041 * We hit the last page and there is more work to be done: wrap
4042 * back to the start of the file
4049 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4050 mapping
->writeback_index
= done_index
;
4052 btrfs_add_delayed_iput(inode
);
4056 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4061 bio_set_op_attrs(epd
->bio
, REQ_OP_WRITE
,
4062 epd
->sync_io
? REQ_SYNC
: 0);
4064 ret
= submit_one_bio(epd
->bio
, 0, epd
->bio_flags
);
4065 BUG_ON(ret
< 0); /* -ENOMEM */
4070 static noinline
void flush_write_bio(void *data
)
4072 struct extent_page_data
*epd
= data
;
4073 flush_epd_write_bio(epd
);
4076 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4077 get_extent_t
*get_extent
,
4078 struct writeback_control
*wbc
)
4081 struct extent_page_data epd
= {
4084 .get_extent
= get_extent
,
4086 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4090 ret
= __extent_writepage(page
, wbc
, &epd
);
4092 flush_epd_write_bio(&epd
);
4096 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4097 u64 start
, u64 end
, get_extent_t
*get_extent
,
4101 struct address_space
*mapping
= inode
->i_mapping
;
4103 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4106 struct extent_page_data epd
= {
4109 .get_extent
= get_extent
,
4111 .sync_io
= mode
== WB_SYNC_ALL
,
4114 struct writeback_control wbc_writepages
= {
4116 .nr_to_write
= nr_pages
* 2,
4117 .range_start
= start
,
4118 .range_end
= end
+ 1,
4121 while (start
<= end
) {
4122 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4123 if (clear_page_dirty_for_io(page
))
4124 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4126 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4127 tree
->ops
->writepage_end_io_hook(page
, start
,
4128 start
+ PAGE_SIZE
- 1,
4136 flush_epd_write_bio(&epd
);
4140 int extent_writepages(struct extent_io_tree
*tree
,
4141 struct address_space
*mapping
,
4142 get_extent_t
*get_extent
,
4143 struct writeback_control
*wbc
)
4146 struct extent_page_data epd
= {
4149 .get_extent
= get_extent
,
4151 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4155 ret
= extent_write_cache_pages(mapping
, wbc
, __extent_writepage
, &epd
,
4157 flush_epd_write_bio(&epd
);
4161 int extent_readpages(struct extent_io_tree
*tree
,
4162 struct address_space
*mapping
,
4163 struct list_head
*pages
, unsigned nr_pages
,
4164 get_extent_t get_extent
)
4166 struct bio
*bio
= NULL
;
4168 unsigned long bio_flags
= 0;
4169 struct page
*pagepool
[16];
4171 struct extent_map
*em_cached
= NULL
;
4173 u64 prev_em_start
= (u64
)-1;
4175 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4176 page
= list_entry(pages
->prev
, struct page
, lru
);
4178 prefetchw(&page
->flags
);
4179 list_del(&page
->lru
);
4180 if (add_to_page_cache_lru(page
, mapping
,
4182 readahead_gfp_mask(mapping
))) {
4187 pagepool
[nr
++] = page
;
4188 if (nr
< ARRAY_SIZE(pagepool
))
4190 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4191 &bio
, 0, &bio_flags
, &prev_em_start
);
4195 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4196 &bio
, 0, &bio_flags
, &prev_em_start
);
4199 free_extent_map(em_cached
);
4201 BUG_ON(!list_empty(pages
));
4203 return submit_one_bio(bio
, 0, bio_flags
);
4208 * basic invalidatepage code, this waits on any locked or writeback
4209 * ranges corresponding to the page, and then deletes any extent state
4210 * records from the tree
4212 int extent_invalidatepage(struct extent_io_tree
*tree
,
4213 struct page
*page
, unsigned long offset
)
4215 struct extent_state
*cached_state
= NULL
;
4216 u64 start
= page_offset(page
);
4217 u64 end
= start
+ PAGE_SIZE
- 1;
4218 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4220 start
+= ALIGN(offset
, blocksize
);
4224 lock_extent_bits(tree
, start
, end
, &cached_state
);
4225 wait_on_page_writeback(page
);
4226 clear_extent_bit(tree
, start
, end
,
4227 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4228 EXTENT_DO_ACCOUNTING
,
4229 1, 1, &cached_state
, GFP_NOFS
);
4234 * a helper for releasepage, this tests for areas of the page that
4235 * are locked or under IO and drops the related state bits if it is safe
4238 static int try_release_extent_state(struct extent_map_tree
*map
,
4239 struct extent_io_tree
*tree
,
4240 struct page
*page
, gfp_t mask
)
4242 u64 start
= page_offset(page
);
4243 u64 end
= start
+ PAGE_SIZE
- 1;
4246 if (test_range_bit(tree
, start
, end
,
4247 EXTENT_IOBITS
, 0, NULL
))
4251 * at this point we can safely clear everything except the
4252 * locked bit and the nodatasum bit
4254 ret
= clear_extent_bit(tree
, start
, end
,
4255 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4258 /* if clear_extent_bit failed for enomem reasons,
4259 * we can't allow the release to continue.
4270 * a helper for releasepage. As long as there are no locked extents
4271 * in the range corresponding to the page, both state records and extent
4272 * map records are removed
4274 int try_release_extent_mapping(struct extent_map_tree
*map
,
4275 struct extent_io_tree
*tree
, struct page
*page
,
4278 struct extent_map
*em
;
4279 u64 start
= page_offset(page
);
4280 u64 end
= start
+ PAGE_SIZE
- 1;
4282 if (gfpflags_allow_blocking(mask
) &&
4283 page
->mapping
->host
->i_size
> SZ_16M
) {
4285 while (start
<= end
) {
4286 len
= end
- start
+ 1;
4287 write_lock(&map
->lock
);
4288 em
= lookup_extent_mapping(map
, start
, len
);
4290 write_unlock(&map
->lock
);
4293 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4294 em
->start
!= start
) {
4295 write_unlock(&map
->lock
);
4296 free_extent_map(em
);
4299 if (!test_range_bit(tree
, em
->start
,
4300 extent_map_end(em
) - 1,
4301 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4303 remove_extent_mapping(map
, em
);
4304 /* once for the rb tree */
4305 free_extent_map(em
);
4307 start
= extent_map_end(em
);
4308 write_unlock(&map
->lock
);
4311 free_extent_map(em
);
4314 return try_release_extent_state(map
, tree
, page
, mask
);
4318 * helper function for fiemap, which doesn't want to see any holes.
4319 * This maps until we find something past 'last'
4321 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4324 get_extent_t
*get_extent
)
4326 u64 sectorsize
= btrfs_inode_sectorsize(inode
);
4327 struct extent_map
*em
;
4334 len
= last
- offset
;
4337 len
= ALIGN(len
, sectorsize
);
4338 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4339 if (IS_ERR_OR_NULL(em
))
4342 /* if this isn't a hole return it */
4343 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4344 em
->block_start
!= EXTENT_MAP_HOLE
) {
4348 /* this is a hole, advance to the next extent */
4349 offset
= extent_map_end(em
);
4350 free_extent_map(em
);
4357 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4358 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4362 u64 max
= start
+ len
;
4366 u64 last_for_get_extent
= 0;
4368 u64 isize
= i_size_read(inode
);
4369 struct btrfs_key found_key
;
4370 struct extent_map
*em
= NULL
;
4371 struct extent_state
*cached_state
= NULL
;
4372 struct btrfs_path
*path
;
4373 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4382 path
= btrfs_alloc_path();
4385 path
->leave_spinning
= 1;
4387 start
= round_down(start
, btrfs_inode_sectorsize(inode
));
4388 len
= round_up(max
, btrfs_inode_sectorsize(inode
)) - start
;
4391 * lookup the last file extent. We're not using i_size here
4392 * because there might be preallocation past i_size
4394 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
4395 btrfs_ino(BTRFS_I(inode
)), -1, 0);
4397 btrfs_free_path(path
);
4406 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4407 found_type
= found_key
.type
;
4409 /* No extents, but there might be delalloc bits */
4410 if (found_key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
4411 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4412 /* have to trust i_size as the end */
4414 last_for_get_extent
= isize
;
4417 * remember the start of the last extent. There are a
4418 * bunch of different factors that go into the length of the
4419 * extent, so its much less complex to remember where it started
4421 last
= found_key
.offset
;
4422 last_for_get_extent
= last
+ 1;
4424 btrfs_release_path(path
);
4427 * we might have some extents allocated but more delalloc past those
4428 * extents. so, we trust isize unless the start of the last extent is
4433 last_for_get_extent
= isize
;
4436 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4439 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4449 u64 offset_in_extent
= 0;
4451 /* break if the extent we found is outside the range */
4452 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4456 * get_extent may return an extent that starts before our
4457 * requested range. We have to make sure the ranges
4458 * we return to fiemap always move forward and don't
4459 * overlap, so adjust the offsets here
4461 em_start
= max(em
->start
, off
);
4464 * record the offset from the start of the extent
4465 * for adjusting the disk offset below. Only do this if the
4466 * extent isn't compressed since our in ram offset may be past
4467 * what we have actually allocated on disk.
4469 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4470 offset_in_extent
= em_start
- em
->start
;
4471 em_end
= extent_map_end(em
);
4472 em_len
= em_end
- em_start
;
4477 * bump off for our next call to get_extent
4479 off
= extent_map_end(em
);
4483 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4485 flags
|= FIEMAP_EXTENT_LAST
;
4486 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4487 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4488 FIEMAP_EXTENT_NOT_ALIGNED
);
4489 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4490 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4491 FIEMAP_EXTENT_UNKNOWN
);
4492 } else if (fieinfo
->fi_extents_max
) {
4493 struct btrfs_trans_handle
*trans
;
4495 u64 bytenr
= em
->block_start
-
4496 (em
->start
- em
->orig_start
);
4498 disko
= em
->block_start
+ offset_in_extent
;
4501 * We need a trans handle to get delayed refs
4503 trans
= btrfs_join_transaction(root
);
4505 * It's OK if we can't start a trans we can still check
4512 * As btrfs supports shared space, this information
4513 * can be exported to userspace tools via
4514 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4515 * then we're just getting a count and we can skip the
4518 ret
= btrfs_check_shared(trans
, root
->fs_info
,
4520 btrfs_ino(BTRFS_I(inode
)), bytenr
);
4522 btrfs_end_transaction(trans
);
4526 flags
|= FIEMAP_EXTENT_SHARED
;
4529 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4530 flags
|= FIEMAP_EXTENT_ENCODED
;
4531 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4532 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4534 free_extent_map(em
);
4536 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4537 (last
== (u64
)-1 && isize
<= em_end
)) {
4538 flags
|= FIEMAP_EXTENT_LAST
;
4542 /* now scan forward to see if this is really the last extent. */
4543 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4550 flags
|= FIEMAP_EXTENT_LAST
;
4553 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4562 free_extent_map(em
);
4564 btrfs_free_path(path
);
4565 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4566 &cached_state
, GFP_NOFS
);
4570 static void __free_extent_buffer(struct extent_buffer
*eb
)
4572 btrfs_leak_debug_del(&eb
->leak_list
);
4573 kmem_cache_free(extent_buffer_cache
, eb
);
4576 int extent_buffer_under_io(struct extent_buffer
*eb
)
4578 return (atomic_read(&eb
->io_pages
) ||
4579 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4580 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4584 * Helper for releasing extent buffer page.
4586 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4588 unsigned long index
;
4590 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4592 BUG_ON(extent_buffer_under_io(eb
));
4594 index
= num_extent_pages(eb
->start
, eb
->len
);
4600 page
= eb
->pages
[index
];
4604 spin_lock(&page
->mapping
->private_lock
);
4606 * We do this since we'll remove the pages after we've
4607 * removed the eb from the radix tree, so we could race
4608 * and have this page now attached to the new eb. So
4609 * only clear page_private if it's still connected to
4612 if (PagePrivate(page
) &&
4613 page
->private == (unsigned long)eb
) {
4614 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4615 BUG_ON(PageDirty(page
));
4616 BUG_ON(PageWriteback(page
));
4618 * We need to make sure we haven't be attached
4621 ClearPagePrivate(page
);
4622 set_page_private(page
, 0);
4623 /* One for the page private */
4628 spin_unlock(&page
->mapping
->private_lock
);
4630 /* One for when we allocated the page */
4632 } while (index
!= 0);
4636 * Helper for releasing the extent buffer.
4638 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4640 btrfs_release_extent_buffer_page(eb
);
4641 __free_extent_buffer(eb
);
4644 static struct extent_buffer
*
4645 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4648 struct extent_buffer
*eb
= NULL
;
4650 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4653 eb
->fs_info
= fs_info
;
4655 rwlock_init(&eb
->lock
);
4656 atomic_set(&eb
->write_locks
, 0);
4657 atomic_set(&eb
->read_locks
, 0);
4658 atomic_set(&eb
->blocking_readers
, 0);
4659 atomic_set(&eb
->blocking_writers
, 0);
4660 atomic_set(&eb
->spinning_readers
, 0);
4661 atomic_set(&eb
->spinning_writers
, 0);
4662 eb
->lock_nested
= 0;
4663 init_waitqueue_head(&eb
->write_lock_wq
);
4664 init_waitqueue_head(&eb
->read_lock_wq
);
4666 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4668 spin_lock_init(&eb
->refs_lock
);
4669 atomic_set(&eb
->refs
, 1);
4670 atomic_set(&eb
->io_pages
, 0);
4673 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4675 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4676 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4677 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4682 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4686 struct extent_buffer
*new;
4687 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4689 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4693 for (i
= 0; i
< num_pages
; i
++) {
4694 p
= alloc_page(GFP_NOFS
);
4696 btrfs_release_extent_buffer(new);
4699 attach_extent_buffer_page(new, p
);
4700 WARN_ON(PageDirty(p
));
4703 copy_page(page_address(p
), page_address(src
->pages
[i
]));
4706 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4707 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4712 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4713 u64 start
, unsigned long len
)
4715 struct extent_buffer
*eb
;
4716 unsigned long num_pages
;
4719 num_pages
= num_extent_pages(start
, len
);
4721 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4725 for (i
= 0; i
< num_pages
; i
++) {
4726 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4730 set_extent_buffer_uptodate(eb
);
4731 btrfs_set_header_nritems(eb
, 0);
4732 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4737 __free_page(eb
->pages
[i
- 1]);
4738 __free_extent_buffer(eb
);
4742 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4745 return __alloc_dummy_extent_buffer(fs_info
, start
, fs_info
->nodesize
);
4748 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4751 /* the ref bit is tricky. We have to make sure it is set
4752 * if we have the buffer dirty. Otherwise the
4753 * code to free a buffer can end up dropping a dirty
4756 * Once the ref bit is set, it won't go away while the
4757 * buffer is dirty or in writeback, and it also won't
4758 * go away while we have the reference count on the
4761 * We can't just set the ref bit without bumping the
4762 * ref on the eb because free_extent_buffer might
4763 * see the ref bit and try to clear it. If this happens
4764 * free_extent_buffer might end up dropping our original
4765 * ref by mistake and freeing the page before we are able
4766 * to add one more ref.
4768 * So bump the ref count first, then set the bit. If someone
4769 * beat us to it, drop the ref we added.
4771 refs
= atomic_read(&eb
->refs
);
4772 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4775 spin_lock(&eb
->refs_lock
);
4776 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4777 atomic_inc(&eb
->refs
);
4778 spin_unlock(&eb
->refs_lock
);
4781 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4782 struct page
*accessed
)
4784 unsigned long num_pages
, i
;
4786 check_buffer_tree_ref(eb
);
4788 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4789 for (i
= 0; i
< num_pages
; i
++) {
4790 struct page
*p
= eb
->pages
[i
];
4793 mark_page_accessed(p
);
4797 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4800 struct extent_buffer
*eb
;
4803 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4804 start
>> PAGE_SHIFT
);
4805 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4808 * Lock our eb's refs_lock to avoid races with
4809 * free_extent_buffer. When we get our eb it might be flagged
4810 * with EXTENT_BUFFER_STALE and another task running
4811 * free_extent_buffer might have seen that flag set,
4812 * eb->refs == 2, that the buffer isn't under IO (dirty and
4813 * writeback flags not set) and it's still in the tree (flag
4814 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4815 * of decrementing the extent buffer's reference count twice.
4816 * So here we could race and increment the eb's reference count,
4817 * clear its stale flag, mark it as dirty and drop our reference
4818 * before the other task finishes executing free_extent_buffer,
4819 * which would later result in an attempt to free an extent
4820 * buffer that is dirty.
4822 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4823 spin_lock(&eb
->refs_lock
);
4824 spin_unlock(&eb
->refs_lock
);
4826 mark_extent_buffer_accessed(eb
, NULL
);
4834 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4835 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4838 struct extent_buffer
*eb
, *exists
= NULL
;
4841 eb
= find_extent_buffer(fs_info
, start
);
4844 eb
= alloc_dummy_extent_buffer(fs_info
, start
);
4847 eb
->fs_info
= fs_info
;
4849 ret
= radix_tree_preload(GFP_NOFS
);
4852 spin_lock(&fs_info
->buffer_lock
);
4853 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4854 start
>> PAGE_SHIFT
, eb
);
4855 spin_unlock(&fs_info
->buffer_lock
);
4856 radix_tree_preload_end();
4857 if (ret
== -EEXIST
) {
4858 exists
= find_extent_buffer(fs_info
, start
);
4864 check_buffer_tree_ref(eb
);
4865 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4868 * We will free dummy extent buffer's if they come into
4869 * free_extent_buffer with a ref count of 2, but if we are using this we
4870 * want the buffers to stay in memory until we're done with them, so
4871 * bump the ref count again.
4873 atomic_inc(&eb
->refs
);
4876 btrfs_release_extent_buffer(eb
);
4881 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4884 unsigned long len
= fs_info
->nodesize
;
4885 unsigned long num_pages
= num_extent_pages(start
, len
);
4887 unsigned long index
= start
>> PAGE_SHIFT
;
4888 struct extent_buffer
*eb
;
4889 struct extent_buffer
*exists
= NULL
;
4891 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4895 if (!IS_ALIGNED(start
, fs_info
->sectorsize
)) {
4896 btrfs_err(fs_info
, "bad tree block start %llu", start
);
4897 return ERR_PTR(-EINVAL
);
4900 eb
= find_extent_buffer(fs_info
, start
);
4904 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4906 return ERR_PTR(-ENOMEM
);
4908 for (i
= 0; i
< num_pages
; i
++, index
++) {
4909 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4911 exists
= ERR_PTR(-ENOMEM
);
4915 spin_lock(&mapping
->private_lock
);
4916 if (PagePrivate(p
)) {
4918 * We could have already allocated an eb for this page
4919 * and attached one so lets see if we can get a ref on
4920 * the existing eb, and if we can we know it's good and
4921 * we can just return that one, else we know we can just
4922 * overwrite page->private.
4924 exists
= (struct extent_buffer
*)p
->private;
4925 if (atomic_inc_not_zero(&exists
->refs
)) {
4926 spin_unlock(&mapping
->private_lock
);
4929 mark_extent_buffer_accessed(exists
, p
);
4935 * Do this so attach doesn't complain and we need to
4936 * drop the ref the old guy had.
4938 ClearPagePrivate(p
);
4939 WARN_ON(PageDirty(p
));
4942 attach_extent_buffer_page(eb
, p
);
4943 spin_unlock(&mapping
->private_lock
);
4944 WARN_ON(PageDirty(p
));
4946 if (!PageUptodate(p
))
4950 * see below about how we avoid a nasty race with release page
4951 * and why we unlock later
4955 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4957 ret
= radix_tree_preload(GFP_NOFS
);
4959 exists
= ERR_PTR(ret
);
4963 spin_lock(&fs_info
->buffer_lock
);
4964 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4965 start
>> PAGE_SHIFT
, eb
);
4966 spin_unlock(&fs_info
->buffer_lock
);
4967 radix_tree_preload_end();
4968 if (ret
== -EEXIST
) {
4969 exists
= find_extent_buffer(fs_info
, start
);
4975 /* add one reference for the tree */
4976 check_buffer_tree_ref(eb
);
4977 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4980 * there is a race where release page may have
4981 * tried to find this extent buffer in the radix
4982 * but failed. It will tell the VM it is safe to
4983 * reclaim the, and it will clear the page private bit.
4984 * We must make sure to set the page private bit properly
4985 * after the extent buffer is in the radix tree so
4986 * it doesn't get lost
4988 SetPageChecked(eb
->pages
[0]);
4989 for (i
= 1; i
< num_pages
; i
++) {
4991 ClearPageChecked(p
);
4994 unlock_page(eb
->pages
[0]);
4998 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
4999 for (i
= 0; i
< num_pages
; i
++) {
5001 unlock_page(eb
->pages
[i
]);
5004 btrfs_release_extent_buffer(eb
);
5008 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
5010 struct extent_buffer
*eb
=
5011 container_of(head
, struct extent_buffer
, rcu_head
);
5013 __free_extent_buffer(eb
);
5016 /* Expects to have eb->eb_lock already held */
5017 static int release_extent_buffer(struct extent_buffer
*eb
)
5019 WARN_ON(atomic_read(&eb
->refs
) == 0);
5020 if (atomic_dec_and_test(&eb
->refs
)) {
5021 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5022 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5024 spin_unlock(&eb
->refs_lock
);
5026 spin_lock(&fs_info
->buffer_lock
);
5027 radix_tree_delete(&fs_info
->buffer_radix
,
5028 eb
->start
>> PAGE_SHIFT
);
5029 spin_unlock(&fs_info
->buffer_lock
);
5031 spin_unlock(&eb
->refs_lock
);
5034 /* Should be safe to release our pages at this point */
5035 btrfs_release_extent_buffer_page(eb
);
5036 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5037 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5038 __free_extent_buffer(eb
);
5042 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5045 spin_unlock(&eb
->refs_lock
);
5050 void free_extent_buffer(struct extent_buffer
*eb
)
5058 refs
= atomic_read(&eb
->refs
);
5061 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5066 spin_lock(&eb
->refs_lock
);
5067 if (atomic_read(&eb
->refs
) == 2 &&
5068 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5069 atomic_dec(&eb
->refs
);
5071 if (atomic_read(&eb
->refs
) == 2 &&
5072 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5073 !extent_buffer_under_io(eb
) &&
5074 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5075 atomic_dec(&eb
->refs
);
5078 * I know this is terrible, but it's temporary until we stop tracking
5079 * the uptodate bits and such for the extent buffers.
5081 release_extent_buffer(eb
);
5084 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5089 spin_lock(&eb
->refs_lock
);
5090 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5092 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5093 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5094 atomic_dec(&eb
->refs
);
5095 release_extent_buffer(eb
);
5098 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5101 unsigned long num_pages
;
5104 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5106 for (i
= 0; i
< num_pages
; i
++) {
5107 page
= eb
->pages
[i
];
5108 if (!PageDirty(page
))
5112 WARN_ON(!PagePrivate(page
));
5114 clear_page_dirty_for_io(page
);
5115 spin_lock_irq(&page
->mapping
->tree_lock
);
5116 if (!PageDirty(page
)) {
5117 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5119 PAGECACHE_TAG_DIRTY
);
5121 spin_unlock_irq(&page
->mapping
->tree_lock
);
5122 ClearPageError(page
);
5125 WARN_ON(atomic_read(&eb
->refs
) == 0);
5128 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5131 unsigned long num_pages
;
5134 check_buffer_tree_ref(eb
);
5136 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5138 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5139 WARN_ON(atomic_read(&eb
->refs
) == 0);
5140 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5142 for (i
= 0; i
< num_pages
; i
++)
5143 set_page_dirty(eb
->pages
[i
]);
5147 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5151 unsigned long num_pages
;
5153 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5154 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5155 for (i
= 0; i
< num_pages
; i
++) {
5156 page
= eb
->pages
[i
];
5158 ClearPageUptodate(page
);
5162 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5166 unsigned long num_pages
;
5168 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5169 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5170 for (i
= 0; i
< num_pages
; i
++) {
5171 page
= eb
->pages
[i
];
5172 SetPageUptodate(page
);
5176 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5178 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5181 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5182 struct extent_buffer
*eb
, int wait
,
5183 get_extent_t
*get_extent
, int mirror_num
)
5189 int locked_pages
= 0;
5190 int all_uptodate
= 1;
5191 unsigned long num_pages
;
5192 unsigned long num_reads
= 0;
5193 struct bio
*bio
= NULL
;
5194 unsigned long bio_flags
= 0;
5196 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5199 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5200 for (i
= 0; i
< num_pages
; i
++) {
5201 page
= eb
->pages
[i
];
5202 if (wait
== WAIT_NONE
) {
5203 if (!trylock_page(page
))
5211 * We need to firstly lock all pages to make sure that
5212 * the uptodate bit of our pages won't be affected by
5213 * clear_extent_buffer_uptodate().
5215 for (i
= 0; i
< num_pages
; i
++) {
5216 page
= eb
->pages
[i
];
5217 if (!PageUptodate(page
)) {
5224 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5228 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5229 eb
->read_mirror
= 0;
5230 atomic_set(&eb
->io_pages
, num_reads
);
5231 for (i
= 0; i
< num_pages
; i
++) {
5232 page
= eb
->pages
[i
];
5234 if (!PageUptodate(page
)) {
5236 atomic_dec(&eb
->io_pages
);
5241 ClearPageError(page
);
5242 err
= __extent_read_full_page(tree
, page
,
5244 mirror_num
, &bio_flags
,
5249 * We use &bio in above __extent_read_full_page,
5250 * so we ensure that if it returns error, the
5251 * current page fails to add itself to bio and
5252 * it's been unlocked.
5254 * We must dec io_pages by ourselves.
5256 atomic_dec(&eb
->io_pages
);
5264 err
= submit_one_bio(bio
, mirror_num
, bio_flags
);
5269 if (ret
|| wait
!= WAIT_COMPLETE
)
5272 for (i
= 0; i
< num_pages
; i
++) {
5273 page
= eb
->pages
[i
];
5274 wait_on_page_locked(page
);
5275 if (!PageUptodate(page
))
5282 while (locked_pages
> 0) {
5284 page
= eb
->pages
[locked_pages
];
5290 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5291 unsigned long start
,
5298 char *dst
= (char *)dstv
;
5299 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5300 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 memcpy(dst
, kaddr
+ offset
, cur
);
5321 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5322 unsigned long start
,
5329 char __user
*dst
= (char __user
*)dstv
;
5330 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5331 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5334 WARN_ON(start
> eb
->len
);
5335 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5337 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5340 page
= eb
->pages
[i
];
5342 cur
= min(len
, (PAGE_SIZE
- offset
));
5343 kaddr
= page_address(page
);
5344 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5359 * return 0 if the item is found within a page.
5360 * return 1 if the item spans two pages.
5361 * return -EINVAL otherwise.
5363 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5364 unsigned long min_len
, char **map
,
5365 unsigned long *map_start
,
5366 unsigned long *map_len
)
5368 size_t offset
= start
& (PAGE_SIZE
- 1);
5371 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5372 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5373 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5380 offset
= start_offset
;
5384 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5387 if (start
+ min_len
> eb
->len
) {
5388 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5389 eb
->start
, eb
->len
, start
, min_len
);
5394 kaddr
= page_address(p
);
5395 *map
= kaddr
+ offset
;
5396 *map_len
= PAGE_SIZE
- offset
;
5400 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5401 unsigned long start
,
5408 char *ptr
= (char *)ptrv
;
5409 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5410 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5413 WARN_ON(start
> eb
->len
);
5414 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5416 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5419 page
= eb
->pages
[i
];
5421 cur
= min(len
, (PAGE_SIZE
- offset
));
5423 kaddr
= page_address(page
);
5424 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5436 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer
*eb
,
5441 WARN_ON(!PageUptodate(eb
->pages
[0]));
5442 kaddr
= page_address(eb
->pages
[0]);
5443 memcpy(kaddr
+ offsetof(struct btrfs_header
, chunk_tree_uuid
), srcv
,
5447 void write_extent_buffer_fsid(struct extent_buffer
*eb
, const void *srcv
)
5451 WARN_ON(!PageUptodate(eb
->pages
[0]));
5452 kaddr
= page_address(eb
->pages
[0]);
5453 memcpy(kaddr
+ offsetof(struct btrfs_header
, fsid
), srcv
,
5457 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5458 unsigned long start
, unsigned long len
)
5464 char *src
= (char *)srcv
;
5465 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5466 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5468 WARN_ON(start
> eb
->len
);
5469 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5471 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5474 page
= eb
->pages
[i
];
5475 WARN_ON(!PageUptodate(page
));
5477 cur
= min(len
, PAGE_SIZE
- offset
);
5478 kaddr
= page_address(page
);
5479 memcpy(kaddr
+ offset
, src
, cur
);
5488 void memzero_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5495 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5496 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5498 WARN_ON(start
> eb
->len
);
5499 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5501 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5504 page
= eb
->pages
[i
];
5505 WARN_ON(!PageUptodate(page
));
5507 cur
= min(len
, PAGE_SIZE
- offset
);
5508 kaddr
= page_address(page
);
5509 memset(kaddr
+ offset
, 0, cur
);
5517 void copy_extent_buffer_full(struct extent_buffer
*dst
,
5518 struct extent_buffer
*src
)
5523 ASSERT(dst
->len
== src
->len
);
5525 num_pages
= num_extent_pages(dst
->start
, dst
->len
);
5526 for (i
= 0; i
< num_pages
; i
++)
5527 copy_page(page_address(dst
->pages
[i
]),
5528 page_address(src
->pages
[i
]));
5531 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5532 unsigned long dst_offset
, unsigned long src_offset
,
5535 u64 dst_len
= dst
->len
;
5540 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5541 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5543 WARN_ON(src
->len
!= dst_len
);
5545 offset
= (start_offset
+ dst_offset
) &
5549 page
= dst
->pages
[i
];
5550 WARN_ON(!PageUptodate(page
));
5552 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5554 kaddr
= page_address(page
);
5555 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5564 void le_bitmap_set(u8
*map
, unsigned int start
, int len
)
5566 u8
*p
= map
+ BIT_BYTE(start
);
5567 const unsigned int size
= start
+ len
;
5568 int bits_to_set
= BITS_PER_BYTE
- (start
% BITS_PER_BYTE
);
5569 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(start
);
5571 while (len
- bits_to_set
>= 0) {
5574 bits_to_set
= BITS_PER_BYTE
;
5579 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5584 void le_bitmap_clear(u8
*map
, unsigned int start
, int len
)
5586 u8
*p
= map
+ BIT_BYTE(start
);
5587 const unsigned int size
= start
+ len
;
5588 int bits_to_clear
= BITS_PER_BYTE
- (start
% BITS_PER_BYTE
);
5589 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(start
);
5591 while (len
- bits_to_clear
>= 0) {
5592 *p
&= ~mask_to_clear
;
5593 len
-= bits_to_clear
;
5594 bits_to_clear
= BITS_PER_BYTE
;
5599 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5600 *p
&= ~mask_to_clear
;
5605 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5607 * @eb: the extent buffer
5608 * @start: offset of the bitmap item in the extent buffer
5610 * @page_index: return index of the page in the extent buffer that contains the
5612 * @page_offset: return offset into the page given by page_index
5614 * This helper hides the ugliness of finding the byte in an extent buffer which
5615 * contains a given bit.
5617 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5618 unsigned long start
, unsigned long nr
,
5619 unsigned long *page_index
,
5620 size_t *page_offset
)
5622 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5623 size_t byte_offset
= BIT_BYTE(nr
);
5627 * The byte we want is the offset of the extent buffer + the offset of
5628 * the bitmap item in the extent buffer + the offset of the byte in the
5631 offset
= start_offset
+ start
+ byte_offset
;
5633 *page_index
= offset
>> PAGE_SHIFT
;
5634 *page_offset
= offset
& (PAGE_SIZE
- 1);
5638 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5639 * @eb: the extent buffer
5640 * @start: offset of the bitmap item in the extent buffer
5641 * @nr: bit number to test
5643 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5651 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5652 page
= eb
->pages
[i
];
5653 WARN_ON(!PageUptodate(page
));
5654 kaddr
= page_address(page
);
5655 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5659 * extent_buffer_bitmap_set - set an area of a bitmap
5660 * @eb: the extent buffer
5661 * @start: offset of the bitmap item in the extent buffer
5662 * @pos: bit number of the first bit
5663 * @len: number of bits to set
5665 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5666 unsigned long pos
, unsigned long len
)
5672 const unsigned int size
= pos
+ len
;
5673 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5674 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5676 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5677 page
= eb
->pages
[i
];
5678 WARN_ON(!PageUptodate(page
));
5679 kaddr
= page_address(page
);
5681 while (len
>= bits_to_set
) {
5682 kaddr
[offset
] |= mask_to_set
;
5684 bits_to_set
= BITS_PER_BYTE
;
5686 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5688 page
= eb
->pages
[++i
];
5689 WARN_ON(!PageUptodate(page
));
5690 kaddr
= page_address(page
);
5694 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5695 kaddr
[offset
] |= mask_to_set
;
5701 * extent_buffer_bitmap_clear - clear an area of a bitmap
5702 * @eb: the extent buffer
5703 * @start: offset of the bitmap item in the extent buffer
5704 * @pos: bit number of the first bit
5705 * @len: number of bits to clear
5707 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5708 unsigned long pos
, unsigned long len
)
5714 const unsigned int size
= pos
+ len
;
5715 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5716 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5718 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5719 page
= eb
->pages
[i
];
5720 WARN_ON(!PageUptodate(page
));
5721 kaddr
= page_address(page
);
5723 while (len
>= bits_to_clear
) {
5724 kaddr
[offset
] &= ~mask_to_clear
;
5725 len
-= bits_to_clear
;
5726 bits_to_clear
= BITS_PER_BYTE
;
5728 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5730 page
= eb
->pages
[++i
];
5731 WARN_ON(!PageUptodate(page
));
5732 kaddr
= page_address(page
);
5736 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5737 kaddr
[offset
] &= ~mask_to_clear
;
5741 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5743 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5744 return distance
< len
;
5747 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5748 unsigned long dst_off
, unsigned long src_off
,
5751 char *dst_kaddr
= page_address(dst_page
);
5753 int must_memmove
= 0;
5755 if (dst_page
!= src_page
) {
5756 src_kaddr
= page_address(src_page
);
5758 src_kaddr
= dst_kaddr
;
5759 if (areas_overlap(src_off
, dst_off
, len
))
5764 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5766 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5769 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5770 unsigned long src_offset
, unsigned long len
)
5772 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5774 size_t dst_off_in_page
;
5775 size_t src_off_in_page
;
5776 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5777 unsigned long dst_i
;
5778 unsigned long src_i
;
5780 if (src_offset
+ len
> dst
->len
) {
5782 "memmove bogus src_offset %lu move len %lu dst len %lu",
5783 src_offset
, len
, dst
->len
);
5786 if (dst_offset
+ len
> dst
->len
) {
5788 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5789 dst_offset
, len
, dst
->len
);
5794 dst_off_in_page
= (start_offset
+ dst_offset
) &
5796 src_off_in_page
= (start_offset
+ src_offset
) &
5799 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5800 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5802 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5804 cur
= min_t(unsigned long, cur
,
5805 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5807 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5808 dst_off_in_page
, src_off_in_page
, cur
);
5816 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5817 unsigned long src_offset
, unsigned long len
)
5819 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5821 size_t dst_off_in_page
;
5822 size_t src_off_in_page
;
5823 unsigned long dst_end
= dst_offset
+ len
- 1;
5824 unsigned long src_end
= src_offset
+ len
- 1;
5825 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5826 unsigned long dst_i
;
5827 unsigned long src_i
;
5829 if (src_offset
+ len
> dst
->len
) {
5831 "memmove bogus src_offset %lu move len %lu len %lu",
5832 src_offset
, len
, dst
->len
);
5835 if (dst_offset
+ len
> dst
->len
) {
5837 "memmove bogus dst_offset %lu move len %lu len %lu",
5838 dst_offset
, len
, dst
->len
);
5841 if (dst_offset
< src_offset
) {
5842 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5846 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5847 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5849 dst_off_in_page
= (start_offset
+ dst_end
) &
5851 src_off_in_page
= (start_offset
+ src_end
) &
5854 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5855 cur
= min(cur
, dst_off_in_page
+ 1);
5856 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5857 dst_off_in_page
- cur
+ 1,
5858 src_off_in_page
- cur
+ 1, cur
);
5866 int try_release_extent_buffer(struct page
*page
)
5868 struct extent_buffer
*eb
;
5871 * We need to make sure nobody is attaching this page to an eb right
5874 spin_lock(&page
->mapping
->private_lock
);
5875 if (!PagePrivate(page
)) {
5876 spin_unlock(&page
->mapping
->private_lock
);
5880 eb
= (struct extent_buffer
*)page
->private;
5884 * This is a little awful but should be ok, we need to make sure that
5885 * the eb doesn't disappear out from under us while we're looking at
5888 spin_lock(&eb
->refs_lock
);
5889 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5890 spin_unlock(&eb
->refs_lock
);
5891 spin_unlock(&page
->mapping
->private_lock
);
5894 spin_unlock(&page
->mapping
->private_lock
);
5897 * If tree ref isn't set then we know the ref on this eb is a real ref,
5898 * so just return, this page will likely be freed soon anyway.
5900 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5901 spin_unlock(&eb
->refs_lock
);
5905 return release_extent_buffer(eb
);