1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
24 static struct kmem_cache
*extent_state_cache
;
25 static struct kmem_cache
*extent_buffer_cache
;
26 static struct bio_set
*btrfs_bioset
;
28 static inline bool extent_state_in_tree(const struct extent_state
*state
)
30 return !RB_EMPTY_NODE(&state
->rb_node
);
33 #ifdef CONFIG_BTRFS_DEBUG
34 static LIST_HEAD(buffers
);
35 static LIST_HEAD(states
);
37 static DEFINE_SPINLOCK(leak_lock
);
40 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
44 spin_lock_irqsave(&leak_lock
, flags
);
46 spin_unlock_irqrestore(&leak_lock
, flags
);
50 void btrfs_leak_debug_del(struct list_head
*entry
)
54 spin_lock_irqsave(&leak_lock
, flags
);
56 spin_unlock_irqrestore(&leak_lock
, flags
);
60 void btrfs_leak_debug_check(void)
62 struct extent_state
*state
;
63 struct extent_buffer
*eb
;
65 while (!list_empty(&states
)) {
66 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
67 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
68 state
->start
, state
->end
, state
->state
,
69 extent_state_in_tree(state
),
70 atomic_read(&state
->refs
));
71 list_del(&state
->leak_list
);
72 kmem_cache_free(extent_state_cache
, state
);
75 while (!list_empty(&buffers
)) {
76 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
77 printk(KERN_ERR
"BTRFS: buffer leak start %llu len %lu "
79 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
80 list_del(&eb
->leak_list
);
81 kmem_cache_free(extent_buffer_cache
, eb
);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
88 struct extent_io_tree
*tree
, u64 start
, u64 end
)
96 inode
= tree
->mapping
->host
;
97 isize
= i_size_read(inode
);
98 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
99 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
100 "%s: ino %llu isize %llu odd range [%llu,%llu]",
101 caller
, btrfs_ino(inode
), isize
, start
, end
);
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry) do {} while (0)
107 #define btrfs_leak_debug_check() do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
111 #define BUFFER_LRU_MAX 64
116 struct rb_node rb_node
;
119 struct extent_page_data
{
121 struct extent_io_tree
*tree
;
122 get_extent_t
*get_extent
;
123 unsigned long bio_flags
;
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
128 unsigned int extent_locked
:1;
130 /* tells the submit_bio code to use a WRITE_SYNC */
131 unsigned int sync_io
:1;
134 static void add_extent_changeset(struct extent_state
*state
, unsigned bits
,
135 struct extent_changeset
*changeset
,
142 if (set
&& (state
->state
& bits
) == bits
)
144 if (!set
&& (state
->state
& bits
) == 0)
146 changeset
->bytes_changed
+= state
->end
- state
->start
+ 1;
147 ret
= ulist_add(changeset
->range_changed
, state
->start
, state
->end
,
153 static noinline
void flush_write_bio(void *data
);
154 static inline struct btrfs_fs_info
*
155 tree_fs_info(struct extent_io_tree
*tree
)
159 return btrfs_sb(tree
->mapping
->host
->i_sb
);
162 int __init
extent_io_init(void)
164 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
165 sizeof(struct extent_state
), 0,
166 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
167 if (!extent_state_cache
)
170 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
171 sizeof(struct extent_buffer
), 0,
172 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
173 if (!extent_buffer_cache
)
174 goto free_state_cache
;
176 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
177 offsetof(struct btrfs_io_bio
, bio
));
179 goto free_buffer_cache
;
181 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
187 bioset_free(btrfs_bioset
);
191 kmem_cache_destroy(extent_buffer_cache
);
192 extent_buffer_cache
= NULL
;
195 kmem_cache_destroy(extent_state_cache
);
196 extent_state_cache
= NULL
;
200 void extent_io_exit(void)
202 btrfs_leak_debug_check();
205 * Make sure all delayed rcu free are flushed before we
209 if (extent_state_cache
)
210 kmem_cache_destroy(extent_state_cache
);
211 if (extent_buffer_cache
)
212 kmem_cache_destroy(extent_buffer_cache
);
214 bioset_free(btrfs_bioset
);
217 void extent_io_tree_init(struct extent_io_tree
*tree
,
218 struct address_space
*mapping
)
220 tree
->state
= RB_ROOT
;
222 tree
->dirty_bytes
= 0;
223 spin_lock_init(&tree
->lock
);
224 tree
->mapping
= mapping
;
227 static struct extent_state
*alloc_extent_state(gfp_t mask
)
229 struct extent_state
*state
;
231 state
= kmem_cache_alloc(extent_state_cache
, mask
);
236 RB_CLEAR_NODE(&state
->rb_node
);
237 btrfs_leak_debug_add(&state
->leak_list
, &states
);
238 atomic_set(&state
->refs
, 1);
239 init_waitqueue_head(&state
->wq
);
240 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
244 void free_extent_state(struct extent_state
*state
)
248 if (atomic_dec_and_test(&state
->refs
)) {
249 WARN_ON(extent_state_in_tree(state
));
250 btrfs_leak_debug_del(&state
->leak_list
);
251 trace_free_extent_state(state
, _RET_IP_
);
252 kmem_cache_free(extent_state_cache
, state
);
256 static struct rb_node
*tree_insert(struct rb_root
*root
,
257 struct rb_node
*search_start
,
259 struct rb_node
*node
,
260 struct rb_node
***p_in
,
261 struct rb_node
**parent_in
)
264 struct rb_node
*parent
= NULL
;
265 struct tree_entry
*entry
;
267 if (p_in
&& parent_in
) {
273 p
= search_start
? &search_start
: &root
->rb_node
;
276 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
278 if (offset
< entry
->start
)
280 else if (offset
> entry
->end
)
287 rb_link_node(node
, parent
, p
);
288 rb_insert_color(node
, root
);
292 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
293 struct rb_node
**prev_ret
,
294 struct rb_node
**next_ret
,
295 struct rb_node
***p_ret
,
296 struct rb_node
**parent_ret
)
298 struct rb_root
*root
= &tree
->state
;
299 struct rb_node
**n
= &root
->rb_node
;
300 struct rb_node
*prev
= NULL
;
301 struct rb_node
*orig_prev
= NULL
;
302 struct tree_entry
*entry
;
303 struct tree_entry
*prev_entry
= NULL
;
307 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
310 if (offset
< entry
->start
)
312 else if (offset
> entry
->end
)
325 while (prev
&& offset
> prev_entry
->end
) {
326 prev
= rb_next(prev
);
327 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
334 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
335 while (prev
&& offset
< prev_entry
->start
) {
336 prev
= rb_prev(prev
);
337 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
344 static inline struct rb_node
*
345 tree_search_for_insert(struct extent_io_tree
*tree
,
347 struct rb_node
***p_ret
,
348 struct rb_node
**parent_ret
)
350 struct rb_node
*prev
= NULL
;
353 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
359 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
362 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
365 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
366 struct extent_state
*other
)
368 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
369 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
374 * utility function to look for merge candidates inside a given range.
375 * Any extents with matching state are merged together into a single
376 * extent in the tree. Extents with EXTENT_IO in their state field
377 * are not merged because the end_io handlers need to be able to do
378 * operations on them without sleeping (or doing allocations/splits).
380 * This should be called with the tree lock held.
382 static void merge_state(struct extent_io_tree
*tree
,
383 struct extent_state
*state
)
385 struct extent_state
*other
;
386 struct rb_node
*other_node
;
388 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
391 other_node
= rb_prev(&state
->rb_node
);
393 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
394 if (other
->end
== state
->start
- 1 &&
395 other
->state
== state
->state
) {
396 merge_cb(tree
, state
, other
);
397 state
->start
= other
->start
;
398 rb_erase(&other
->rb_node
, &tree
->state
);
399 RB_CLEAR_NODE(&other
->rb_node
);
400 free_extent_state(other
);
403 other_node
= rb_next(&state
->rb_node
);
405 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
406 if (other
->start
== state
->end
+ 1 &&
407 other
->state
== state
->state
) {
408 merge_cb(tree
, state
, other
);
409 state
->end
= other
->end
;
410 rb_erase(&other
->rb_node
, &tree
->state
);
411 RB_CLEAR_NODE(&other
->rb_node
);
412 free_extent_state(other
);
417 static void set_state_cb(struct extent_io_tree
*tree
,
418 struct extent_state
*state
, unsigned *bits
)
420 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
421 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
424 static void clear_state_cb(struct extent_io_tree
*tree
,
425 struct extent_state
*state
, unsigned *bits
)
427 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
428 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
431 static void set_state_bits(struct extent_io_tree
*tree
,
432 struct extent_state
*state
, unsigned *bits
,
433 struct extent_changeset
*changeset
);
436 * insert an extent_state struct into the tree. 'bits' are set on the
437 * struct before it is inserted.
439 * This may return -EEXIST if the extent is already there, in which case the
440 * state struct is freed.
442 * The tree lock is not taken internally. This is a utility function and
443 * probably isn't what you want to call (see set/clear_extent_bit).
445 static int insert_state(struct extent_io_tree
*tree
,
446 struct extent_state
*state
, u64 start
, u64 end
,
448 struct rb_node
**parent
,
449 unsigned *bits
, struct extent_changeset
*changeset
)
451 struct rb_node
*node
;
454 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
456 state
->start
= start
;
459 set_state_bits(tree
, state
, bits
, changeset
);
461 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
463 struct extent_state
*found
;
464 found
= rb_entry(node
, struct extent_state
, rb_node
);
465 printk(KERN_ERR
"BTRFS: found node %llu %llu on insert of "
467 found
->start
, found
->end
, start
, end
);
470 merge_state(tree
, state
);
474 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
477 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
478 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
482 * split a given extent state struct in two, inserting the preallocated
483 * struct 'prealloc' as the newly created second half. 'split' indicates an
484 * offset inside 'orig' where it should be split.
487 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
488 * are two extent state structs in the tree:
489 * prealloc: [orig->start, split - 1]
490 * orig: [ split, orig->end ]
492 * The tree locks are not taken by this function. They need to be held
495 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
496 struct extent_state
*prealloc
, u64 split
)
498 struct rb_node
*node
;
500 split_cb(tree
, orig
, split
);
502 prealloc
->start
= orig
->start
;
503 prealloc
->end
= split
- 1;
504 prealloc
->state
= orig
->state
;
507 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
508 &prealloc
->rb_node
, NULL
, NULL
);
510 free_extent_state(prealloc
);
516 static struct extent_state
*next_state(struct extent_state
*state
)
518 struct rb_node
*next
= rb_next(&state
->rb_node
);
520 return rb_entry(next
, struct extent_state
, rb_node
);
526 * utility function to clear some bits in an extent state struct.
527 * it will optionally wake up any one waiting on this state (wake == 1).
529 * If no bits are set on the state struct after clearing things, the
530 * struct is freed and removed from the tree
532 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
533 struct extent_state
*state
,
534 unsigned *bits
, int wake
,
535 struct extent_changeset
*changeset
)
537 struct extent_state
*next
;
538 unsigned bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
540 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
541 u64 range
= state
->end
- state
->start
+ 1;
542 WARN_ON(range
> tree
->dirty_bytes
);
543 tree
->dirty_bytes
-= range
;
545 clear_state_cb(tree
, state
, bits
);
546 add_extent_changeset(state
, bits_to_clear
, changeset
, 0);
547 state
->state
&= ~bits_to_clear
;
550 if (state
->state
== 0) {
551 next
= next_state(state
);
552 if (extent_state_in_tree(state
)) {
553 rb_erase(&state
->rb_node
, &tree
->state
);
554 RB_CLEAR_NODE(&state
->rb_node
);
555 free_extent_state(state
);
560 merge_state(tree
, state
);
561 next
= next_state(state
);
566 static struct extent_state
*
567 alloc_extent_state_atomic(struct extent_state
*prealloc
)
570 prealloc
= alloc_extent_state(GFP_ATOMIC
);
575 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
577 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
578 "Extent tree was modified by another "
579 "thread while locked.");
583 * clear some bits on a range in the tree. This may require splitting
584 * or inserting elements in the tree, so the gfp mask is used to
585 * indicate which allocations or sleeping are allowed.
587 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
588 * the given range from the tree regardless of state (ie for truncate).
590 * the range [start, end] is inclusive.
592 * This takes the tree lock, and returns 0 on success and < 0 on error.
594 static int __clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
595 unsigned bits
, int wake
, int delete,
596 struct extent_state
**cached_state
,
597 gfp_t mask
, struct extent_changeset
*changeset
)
599 struct extent_state
*state
;
600 struct extent_state
*cached
;
601 struct extent_state
*prealloc
= NULL
;
602 struct rb_node
*node
;
607 btrfs_debug_check_extent_io_range(tree
, start
, end
);
609 if (bits
& EXTENT_DELALLOC
)
610 bits
|= EXTENT_NORESERVE
;
613 bits
|= ~EXTENT_CTLBITS
;
614 bits
|= EXTENT_FIRST_DELALLOC
;
616 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
619 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
621 * Don't care for allocation failure here because we might end
622 * up not needing the pre-allocated extent state at all, which
623 * is the case if we only have in the tree extent states that
624 * cover our input range and don't cover too any other range.
625 * If we end up needing a new extent state we allocate it later.
627 prealloc
= alloc_extent_state(mask
);
630 spin_lock(&tree
->lock
);
632 cached
= *cached_state
;
635 *cached_state
= NULL
;
639 if (cached
&& extent_state_in_tree(cached
) &&
640 cached
->start
<= start
&& cached
->end
> start
) {
642 atomic_dec(&cached
->refs
);
647 free_extent_state(cached
);
650 * this search will find the extents that end after
653 node
= tree_search(tree
, start
);
656 state
= rb_entry(node
, struct extent_state
, rb_node
);
658 if (state
->start
> end
)
660 WARN_ON(state
->end
< start
);
661 last_end
= state
->end
;
663 /* the state doesn't have the wanted bits, go ahead */
664 if (!(state
->state
& bits
)) {
665 state
= next_state(state
);
670 * | ---- desired range ---- |
672 * | ------------- state -------------- |
674 * We need to split the extent we found, and may flip
675 * bits on second half.
677 * If the extent we found extends past our range, we
678 * just split and search again. It'll get split again
679 * the next time though.
681 * If the extent we found is inside our range, we clear
682 * the desired bit on it.
685 if (state
->start
< start
) {
686 prealloc
= alloc_extent_state_atomic(prealloc
);
688 err
= split_state(tree
, state
, prealloc
, start
);
690 extent_io_tree_panic(tree
, err
);
695 if (state
->end
<= end
) {
696 state
= clear_state_bit(tree
, state
, &bits
, wake
,
703 * | ---- desired range ---- |
705 * We need to split the extent, and clear the bit
708 if (state
->start
<= end
&& state
->end
> end
) {
709 prealloc
= alloc_extent_state_atomic(prealloc
);
711 err
= split_state(tree
, state
, prealloc
, end
+ 1);
713 extent_io_tree_panic(tree
, err
);
718 clear_state_bit(tree
, prealloc
, &bits
, wake
, changeset
);
724 state
= clear_state_bit(tree
, state
, &bits
, wake
, changeset
);
726 if (last_end
== (u64
)-1)
728 start
= last_end
+ 1;
729 if (start
<= end
&& state
&& !need_resched())
734 spin_unlock(&tree
->lock
);
736 free_extent_state(prealloc
);
743 spin_unlock(&tree
->lock
);
744 if (gfpflags_allow_blocking(mask
))
749 static void wait_on_state(struct extent_io_tree
*tree
,
750 struct extent_state
*state
)
751 __releases(tree
->lock
)
752 __acquires(tree
->lock
)
755 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
756 spin_unlock(&tree
->lock
);
758 spin_lock(&tree
->lock
);
759 finish_wait(&state
->wq
, &wait
);
763 * waits for one or more bits to clear on a range in the state tree.
764 * The range [start, end] is inclusive.
765 * The tree lock is taken by this function
767 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
770 struct extent_state
*state
;
771 struct rb_node
*node
;
773 btrfs_debug_check_extent_io_range(tree
, start
, end
);
775 spin_lock(&tree
->lock
);
779 * this search will find all the extents that end after
782 node
= tree_search(tree
, start
);
787 state
= rb_entry(node
, struct extent_state
, rb_node
);
789 if (state
->start
> end
)
792 if (state
->state
& bits
) {
793 start
= state
->start
;
794 atomic_inc(&state
->refs
);
795 wait_on_state(tree
, state
);
796 free_extent_state(state
);
799 start
= state
->end
+ 1;
804 if (!cond_resched_lock(&tree
->lock
)) {
805 node
= rb_next(node
);
810 spin_unlock(&tree
->lock
);
813 static void set_state_bits(struct extent_io_tree
*tree
,
814 struct extent_state
*state
,
815 unsigned *bits
, struct extent_changeset
*changeset
)
817 unsigned bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
819 set_state_cb(tree
, state
, bits
);
820 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
821 u64 range
= state
->end
- state
->start
+ 1;
822 tree
->dirty_bytes
+= range
;
824 add_extent_changeset(state
, bits_to_set
, changeset
, 1);
825 state
->state
|= bits_to_set
;
828 static void cache_state_if_flags(struct extent_state
*state
,
829 struct extent_state
**cached_ptr
,
832 if (cached_ptr
&& !(*cached_ptr
)) {
833 if (!flags
|| (state
->state
& flags
)) {
835 atomic_inc(&state
->refs
);
840 static void cache_state(struct extent_state
*state
,
841 struct extent_state
**cached_ptr
)
843 return cache_state_if_flags(state
, cached_ptr
,
844 EXTENT_IOBITS
| EXTENT_BOUNDARY
);
848 * set some bits on a range in the tree. This may require allocations or
849 * sleeping, so the gfp mask is used to indicate what is allowed.
851 * If any of the exclusive bits are set, this will fail with -EEXIST if some
852 * part of the range already has the desired bits set. The start of the
853 * existing range is returned in failed_start in this case.
855 * [start, end] is inclusive This takes the tree lock.
858 static int __must_check
859 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
860 unsigned bits
, unsigned exclusive_bits
,
861 u64
*failed_start
, struct extent_state
**cached_state
,
862 gfp_t mask
, struct extent_changeset
*changeset
)
864 struct extent_state
*state
;
865 struct extent_state
*prealloc
= NULL
;
866 struct rb_node
*node
;
868 struct rb_node
*parent
;
873 btrfs_debug_check_extent_io_range(tree
, start
, end
);
875 bits
|= EXTENT_FIRST_DELALLOC
;
877 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
878 prealloc
= alloc_extent_state(mask
);
882 spin_lock(&tree
->lock
);
883 if (cached_state
&& *cached_state
) {
884 state
= *cached_state
;
885 if (state
->start
<= start
&& state
->end
> start
&&
886 extent_state_in_tree(state
)) {
887 node
= &state
->rb_node
;
892 * this search will find all the extents that end after
895 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
897 prealloc
= alloc_extent_state_atomic(prealloc
);
899 err
= insert_state(tree
, prealloc
, start
, end
,
900 &p
, &parent
, &bits
, changeset
);
902 extent_io_tree_panic(tree
, err
);
904 cache_state(prealloc
, cached_state
);
908 state
= rb_entry(node
, struct extent_state
, rb_node
);
910 last_start
= state
->start
;
911 last_end
= state
->end
;
914 * | ---- desired range ---- |
917 * Just lock what we found and keep going
919 if (state
->start
== start
&& state
->end
<= end
) {
920 if (state
->state
& exclusive_bits
) {
921 *failed_start
= state
->start
;
926 set_state_bits(tree
, state
, &bits
, changeset
);
927 cache_state(state
, cached_state
);
928 merge_state(tree
, state
);
929 if (last_end
== (u64
)-1)
931 start
= last_end
+ 1;
932 state
= next_state(state
);
933 if (start
< end
&& state
&& state
->start
== start
&&
940 * | ---- desired range ---- |
943 * | ------------- state -------------- |
945 * We need to split the extent we found, and may flip bits on
948 * If the extent we found extends past our
949 * range, we just split and search again. It'll get split
950 * again the next time though.
952 * If the extent we found is inside our range, we set the
955 if (state
->start
< start
) {
956 if (state
->state
& exclusive_bits
) {
957 *failed_start
= start
;
962 prealloc
= alloc_extent_state_atomic(prealloc
);
964 err
= split_state(tree
, state
, prealloc
, start
);
966 extent_io_tree_panic(tree
, err
);
971 if (state
->end
<= end
) {
972 set_state_bits(tree
, state
, &bits
, changeset
);
973 cache_state(state
, cached_state
);
974 merge_state(tree
, state
);
975 if (last_end
== (u64
)-1)
977 start
= last_end
+ 1;
978 state
= next_state(state
);
979 if (start
< end
&& state
&& state
->start
== start
&&
986 * | ---- desired range ---- |
987 * | state | or | state |
989 * There's a hole, we need to insert something in it and
990 * ignore the extent we found.
992 if (state
->start
> start
) {
994 if (end
< last_start
)
997 this_end
= last_start
- 1;
999 prealloc
= alloc_extent_state_atomic(prealloc
);
1003 * Avoid to free 'prealloc' if it can be merged with
1006 err
= insert_state(tree
, prealloc
, start
, this_end
,
1007 NULL
, NULL
, &bits
, changeset
);
1009 extent_io_tree_panic(tree
, err
);
1011 cache_state(prealloc
, cached_state
);
1013 start
= this_end
+ 1;
1017 * | ---- desired range ---- |
1019 * We need to split the extent, and set the bit
1022 if (state
->start
<= end
&& state
->end
> end
) {
1023 if (state
->state
& exclusive_bits
) {
1024 *failed_start
= start
;
1029 prealloc
= alloc_extent_state_atomic(prealloc
);
1031 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1033 extent_io_tree_panic(tree
, err
);
1035 set_state_bits(tree
, prealloc
, &bits
, changeset
);
1036 cache_state(prealloc
, cached_state
);
1037 merge_state(tree
, prealloc
);
1045 spin_unlock(&tree
->lock
);
1047 free_extent_state(prealloc
);
1054 spin_unlock(&tree
->lock
);
1055 if (gfpflags_allow_blocking(mask
))
1060 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1061 unsigned bits
, u64
* failed_start
,
1062 struct extent_state
**cached_state
, gfp_t mask
)
1064 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1065 cached_state
, mask
, NULL
);
1070 * convert_extent_bit - convert all bits in a given range from one bit to
1072 * @tree: the io tree to search
1073 * @start: the start offset in bytes
1074 * @end: the end offset in bytes (inclusive)
1075 * @bits: the bits to set in this range
1076 * @clear_bits: the bits to clear in this range
1077 * @cached_state: state that we're going to cache
1078 * @mask: the allocation mask
1080 * This will go through and set bits for the given range. If any states exist
1081 * already in this range they are set with the given bit and cleared of the
1082 * clear_bits. This is only meant to be used by things that are mergeable, ie
1083 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1084 * boundary bits like LOCK.
1086 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1087 unsigned bits
, unsigned clear_bits
,
1088 struct extent_state
**cached_state
, gfp_t mask
)
1090 struct extent_state
*state
;
1091 struct extent_state
*prealloc
= NULL
;
1092 struct rb_node
*node
;
1094 struct rb_node
*parent
;
1098 bool first_iteration
= true;
1100 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1103 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
1105 * Best effort, don't worry if extent state allocation fails
1106 * here for the first iteration. We might have a cached state
1107 * that matches exactly the target range, in which case no
1108 * extent state allocations are needed. We'll only know this
1109 * after locking the tree.
1111 prealloc
= alloc_extent_state(mask
);
1112 if (!prealloc
&& !first_iteration
)
1116 spin_lock(&tree
->lock
);
1117 if (cached_state
&& *cached_state
) {
1118 state
= *cached_state
;
1119 if (state
->start
<= start
&& state
->end
> start
&&
1120 extent_state_in_tree(state
)) {
1121 node
= &state
->rb_node
;
1127 * this search will find all the extents that end after
1130 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1132 prealloc
= alloc_extent_state_atomic(prealloc
);
1137 err
= insert_state(tree
, prealloc
, start
, end
,
1138 &p
, &parent
, &bits
, NULL
);
1140 extent_io_tree_panic(tree
, err
);
1141 cache_state(prealloc
, cached_state
);
1145 state
= rb_entry(node
, struct extent_state
, rb_node
);
1147 last_start
= state
->start
;
1148 last_end
= state
->end
;
1151 * | ---- desired range ---- |
1154 * Just lock what we found and keep going
1156 if (state
->start
== start
&& state
->end
<= end
) {
1157 set_state_bits(tree
, state
, &bits
, NULL
);
1158 cache_state(state
, cached_state
);
1159 state
= clear_state_bit(tree
, state
, &clear_bits
, 0, NULL
);
1160 if (last_end
== (u64
)-1)
1162 start
= last_end
+ 1;
1163 if (start
< end
&& state
&& state
->start
== start
&&
1170 * | ---- desired range ---- |
1173 * | ------------- state -------------- |
1175 * We need to split the extent we found, and may flip bits on
1178 * If the extent we found extends past our
1179 * range, we just split and search again. It'll get split
1180 * again the next time though.
1182 * If the extent we found is inside our range, we set the
1183 * desired bit on it.
1185 if (state
->start
< start
) {
1186 prealloc
= alloc_extent_state_atomic(prealloc
);
1191 err
= split_state(tree
, state
, prealloc
, start
);
1193 extent_io_tree_panic(tree
, err
);
1197 if (state
->end
<= end
) {
1198 set_state_bits(tree
, state
, &bits
, NULL
);
1199 cache_state(state
, cached_state
);
1200 state
= clear_state_bit(tree
, state
, &clear_bits
, 0,
1202 if (last_end
== (u64
)-1)
1204 start
= last_end
+ 1;
1205 if (start
< end
&& state
&& state
->start
== start
&&
1212 * | ---- desired range ---- |
1213 * | state | or | state |
1215 * There's a hole, we need to insert something in it and
1216 * ignore the extent we found.
1218 if (state
->start
> start
) {
1220 if (end
< last_start
)
1223 this_end
= last_start
- 1;
1225 prealloc
= alloc_extent_state_atomic(prealloc
);
1232 * Avoid to free 'prealloc' if it can be merged with
1235 err
= insert_state(tree
, prealloc
, start
, this_end
,
1236 NULL
, NULL
, &bits
, NULL
);
1238 extent_io_tree_panic(tree
, err
);
1239 cache_state(prealloc
, cached_state
);
1241 start
= this_end
+ 1;
1245 * | ---- desired range ---- |
1247 * We need to split the extent, and set the bit
1250 if (state
->start
<= end
&& state
->end
> end
) {
1251 prealloc
= alloc_extent_state_atomic(prealloc
);
1257 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1259 extent_io_tree_panic(tree
, err
);
1261 set_state_bits(tree
, prealloc
, &bits
, NULL
);
1262 cache_state(prealloc
, cached_state
);
1263 clear_state_bit(tree
, prealloc
, &clear_bits
, 0, NULL
);
1271 spin_unlock(&tree
->lock
);
1273 free_extent_state(prealloc
);
1280 spin_unlock(&tree
->lock
);
1281 if (gfpflags_allow_blocking(mask
))
1283 first_iteration
= false;
1287 /* wrappers around set/clear extent bit */
1288 int set_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1291 return set_extent_bit(tree
, start
, end
, EXTENT_DIRTY
, NULL
,
1295 int set_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1296 unsigned bits
, gfp_t mask
)
1298 return set_extent_bit(tree
, start
, end
, bits
, NULL
,
1302 int set_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1303 unsigned bits
, gfp_t mask
,
1304 struct extent_changeset
*changeset
)
1307 * We don't support EXTENT_LOCKED yet, as current changeset will
1308 * record any bits changed, so for EXTENT_LOCKED case, it will
1309 * either fail with -EEXIST or changeset will record the whole
1312 BUG_ON(bits
& EXTENT_LOCKED
);
1314 return __set_extent_bit(tree
, start
, end
, bits
, 0, NULL
, NULL
, mask
,
1318 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1319 unsigned bits
, int wake
, int delete,
1320 struct extent_state
**cached
, gfp_t mask
)
1322 return __clear_extent_bit(tree
, start
, end
, bits
, wake
, delete,
1323 cached
, mask
, NULL
);
1326 int clear_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1327 unsigned bits
, gfp_t mask
)
1331 if (bits
& EXTENT_LOCKED
)
1334 return clear_extent_bit(tree
, start
, end
, bits
, wake
, 0, NULL
, mask
);
1337 int clear_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1338 unsigned bits
, gfp_t mask
,
1339 struct extent_changeset
*changeset
)
1342 * Don't support EXTENT_LOCKED case, same reason as
1343 * set_record_extent_bits().
1345 BUG_ON(bits
& EXTENT_LOCKED
);
1347 return __clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, mask
,
1351 int set_extent_delalloc(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1352 struct extent_state
**cached_state
, gfp_t mask
)
1354 return set_extent_bit(tree
, start
, end
,
1355 EXTENT_DELALLOC
| EXTENT_UPTODATE
,
1356 NULL
, cached_state
, mask
);
1359 int set_extent_defrag(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1360 struct extent_state
**cached_state
, gfp_t mask
)
1362 return set_extent_bit(tree
, start
, end
,
1363 EXTENT_DELALLOC
| EXTENT_UPTODATE
| EXTENT_DEFRAG
,
1364 NULL
, cached_state
, mask
);
1367 int clear_extent_dirty(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1370 return clear_extent_bit(tree
, start
, end
,
1371 EXTENT_DIRTY
| EXTENT_DELALLOC
|
1372 EXTENT_DO_ACCOUNTING
, 0, 0, NULL
, mask
);
1375 int set_extent_new(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1378 return set_extent_bit(tree
, start
, end
, EXTENT_NEW
, NULL
,
1382 int set_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1383 struct extent_state
**cached_state
, gfp_t mask
)
1385 return set_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, NULL
,
1386 cached_state
, mask
);
1389 int clear_extent_uptodate(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1390 struct extent_state
**cached_state
, gfp_t mask
)
1392 return clear_extent_bit(tree
, start
, end
, EXTENT_UPTODATE
, 0, 0,
1393 cached_state
, mask
);
1397 * either insert or lock state struct between start and end use mask to tell
1398 * us if waiting is desired.
1400 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1401 unsigned bits
, struct extent_state
**cached_state
)
1407 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
| bits
,
1408 EXTENT_LOCKED
, &failed_start
,
1409 cached_state
, GFP_NOFS
, NULL
);
1410 if (err
== -EEXIST
) {
1411 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1412 start
= failed_start
;
1415 WARN_ON(start
> end
);
1420 int lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1422 return lock_extent_bits(tree
, start
, end
, 0, NULL
);
1425 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1430 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1431 &failed_start
, NULL
, GFP_NOFS
, NULL
);
1432 if (err
== -EEXIST
) {
1433 if (failed_start
> start
)
1434 clear_extent_bit(tree
, start
, failed_start
- 1,
1435 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1441 int unlock_extent_cached(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1442 struct extent_state
**cached
, gfp_t mask
)
1444 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, cached
,
1448 int unlock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1450 return clear_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, 1, 0, NULL
,
1454 void extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1456 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1457 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1460 while (index
<= end_index
) {
1461 page
= find_get_page(inode
->i_mapping
, index
);
1462 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1463 clear_page_dirty_for_io(page
);
1464 page_cache_release(page
);
1469 void extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1471 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1472 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1475 while (index
<= end_index
) {
1476 page
= find_get_page(inode
->i_mapping
, index
);
1477 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1478 __set_page_dirty_nobuffers(page
);
1479 account_page_redirty(page
);
1480 page_cache_release(page
);
1486 * helper function to set both pages and extents in the tree writeback
1488 static void set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1490 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1491 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1494 while (index
<= end_index
) {
1495 page
= find_get_page(tree
->mapping
, index
);
1496 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1497 set_page_writeback(page
);
1498 page_cache_release(page
);
1503 /* find the first state struct with 'bits' set after 'start', and
1504 * return it. tree->lock must be held. NULL will returned if
1505 * nothing was found after 'start'
1507 static struct extent_state
*
1508 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1509 u64 start
, unsigned bits
)
1511 struct rb_node
*node
;
1512 struct extent_state
*state
;
1515 * this search will find all the extents that end after
1518 node
= tree_search(tree
, start
);
1523 state
= rb_entry(node
, struct extent_state
, rb_node
);
1524 if (state
->end
>= start
&& (state
->state
& bits
))
1527 node
= rb_next(node
);
1536 * find the first offset in the io tree with 'bits' set. zero is
1537 * returned if we find something, and *start_ret and *end_ret are
1538 * set to reflect the state struct that was found.
1540 * If nothing was found, 1 is returned. If found something, return 0.
1542 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1543 u64
*start_ret
, u64
*end_ret
, unsigned bits
,
1544 struct extent_state
**cached_state
)
1546 struct extent_state
*state
;
1550 spin_lock(&tree
->lock
);
1551 if (cached_state
&& *cached_state
) {
1552 state
= *cached_state
;
1553 if (state
->end
== start
- 1 && extent_state_in_tree(state
)) {
1554 n
= rb_next(&state
->rb_node
);
1556 state
= rb_entry(n
, struct extent_state
,
1558 if (state
->state
& bits
)
1562 free_extent_state(*cached_state
);
1563 *cached_state
= NULL
;
1566 free_extent_state(*cached_state
);
1567 *cached_state
= NULL
;
1570 state
= find_first_extent_bit_state(tree
, start
, bits
);
1573 cache_state_if_flags(state
, cached_state
, 0);
1574 *start_ret
= state
->start
;
1575 *end_ret
= state
->end
;
1579 spin_unlock(&tree
->lock
);
1584 * find a contiguous range of bytes in the file marked as delalloc, not
1585 * more than 'max_bytes'. start and end are used to return the range,
1587 * 1 is returned if we find something, 0 if nothing was in the tree
1589 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1590 u64
*start
, u64
*end
, u64 max_bytes
,
1591 struct extent_state
**cached_state
)
1593 struct rb_node
*node
;
1594 struct extent_state
*state
;
1595 u64 cur_start
= *start
;
1597 u64 total_bytes
= 0;
1599 spin_lock(&tree
->lock
);
1602 * this search will find all the extents that end after
1605 node
= tree_search(tree
, cur_start
);
1613 state
= rb_entry(node
, struct extent_state
, rb_node
);
1614 if (found
&& (state
->start
!= cur_start
||
1615 (state
->state
& EXTENT_BOUNDARY
))) {
1618 if (!(state
->state
& EXTENT_DELALLOC
)) {
1624 *start
= state
->start
;
1625 *cached_state
= state
;
1626 atomic_inc(&state
->refs
);
1630 cur_start
= state
->end
+ 1;
1631 node
= rb_next(node
);
1632 total_bytes
+= state
->end
- state
->start
+ 1;
1633 if (total_bytes
>= max_bytes
)
1639 spin_unlock(&tree
->lock
);
1643 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1644 struct page
*locked_page
,
1648 struct page
*pages
[16];
1649 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1650 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1651 unsigned long nr_pages
= end_index
- index
+ 1;
1654 if (index
== locked_page
->index
&& end_index
== index
)
1657 while (nr_pages
> 0) {
1658 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1659 min_t(unsigned long, nr_pages
,
1660 ARRAY_SIZE(pages
)), pages
);
1661 for (i
= 0; i
< ret
; i
++) {
1662 if (pages
[i
] != locked_page
)
1663 unlock_page(pages
[i
]);
1664 page_cache_release(pages
[i
]);
1672 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1673 struct page
*locked_page
,
1677 unsigned long index
= delalloc_start
>> PAGE_CACHE_SHIFT
;
1678 unsigned long start_index
= index
;
1679 unsigned long end_index
= delalloc_end
>> PAGE_CACHE_SHIFT
;
1680 unsigned long pages_locked
= 0;
1681 struct page
*pages
[16];
1682 unsigned long nrpages
;
1686 /* the caller is responsible for locking the start index */
1687 if (index
== locked_page
->index
&& index
== end_index
)
1690 /* skip the page at the start index */
1691 nrpages
= end_index
- index
+ 1;
1692 while (nrpages
> 0) {
1693 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1694 min_t(unsigned long,
1695 nrpages
, ARRAY_SIZE(pages
)), pages
);
1700 /* now we have an array of pages, lock them all */
1701 for (i
= 0; i
< ret
; i
++) {
1703 * the caller is taking responsibility for
1706 if (pages
[i
] != locked_page
) {
1707 lock_page(pages
[i
]);
1708 if (!PageDirty(pages
[i
]) ||
1709 pages
[i
]->mapping
!= inode
->i_mapping
) {
1711 unlock_page(pages
[i
]);
1712 page_cache_release(pages
[i
]);
1716 page_cache_release(pages
[i
]);
1725 if (ret
&& pages_locked
) {
1726 __unlock_for_delalloc(inode
, locked_page
,
1728 ((u64
)(start_index
+ pages_locked
- 1)) <<
1735 * find a contiguous range of bytes in the file marked as delalloc, not
1736 * more than 'max_bytes'. start and end are used to return the range,
1738 * 1 is returned if we find something, 0 if nothing was in the tree
1740 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1741 struct extent_io_tree
*tree
,
1742 struct page
*locked_page
, u64
*start
,
1743 u64
*end
, u64 max_bytes
)
1748 struct extent_state
*cached_state
= NULL
;
1753 /* step one, find a bunch of delalloc bytes starting at start */
1754 delalloc_start
= *start
;
1756 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1757 max_bytes
, &cached_state
);
1758 if (!found
|| delalloc_end
<= *start
) {
1759 *start
= delalloc_start
;
1760 *end
= delalloc_end
;
1761 free_extent_state(cached_state
);
1766 * start comes from the offset of locked_page. We have to lock
1767 * pages in order, so we can't process delalloc bytes before
1770 if (delalloc_start
< *start
)
1771 delalloc_start
= *start
;
1774 * make sure to limit the number of pages we try to lock down
1776 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1777 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1779 /* step two, lock all the pages after the page that has start */
1780 ret
= lock_delalloc_pages(inode
, locked_page
,
1781 delalloc_start
, delalloc_end
);
1782 if (ret
== -EAGAIN
) {
1783 /* some of the pages are gone, lets avoid looping by
1784 * shortening the size of the delalloc range we're searching
1786 free_extent_state(cached_state
);
1787 cached_state
= NULL
;
1789 max_bytes
= PAGE_CACHE_SIZE
;
1797 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1799 /* step three, lock the state bits for the whole range */
1800 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, 0, &cached_state
);
1802 /* then test to make sure it is all still delalloc */
1803 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1804 EXTENT_DELALLOC
, 1, cached_state
);
1806 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1807 &cached_state
, GFP_NOFS
);
1808 __unlock_for_delalloc(inode
, locked_page
,
1809 delalloc_start
, delalloc_end
);
1813 free_extent_state(cached_state
);
1814 *start
= delalloc_start
;
1815 *end
= delalloc_end
;
1820 void extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1821 struct page
*locked_page
,
1822 unsigned clear_bits
,
1823 unsigned long page_ops
)
1825 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1827 struct page
*pages
[16];
1828 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
1829 unsigned long end_index
= end
>> PAGE_CACHE_SHIFT
;
1830 unsigned long nr_pages
= end_index
- index
+ 1;
1833 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1837 if ((page_ops
& PAGE_SET_ERROR
) && nr_pages
> 0)
1838 mapping_set_error(inode
->i_mapping
, -EIO
);
1840 while (nr_pages
> 0) {
1841 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1842 min_t(unsigned long,
1843 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1844 for (i
= 0; i
< ret
; i
++) {
1846 if (page_ops
& PAGE_SET_PRIVATE2
)
1847 SetPagePrivate2(pages
[i
]);
1849 if (pages
[i
] == locked_page
) {
1850 page_cache_release(pages
[i
]);
1853 if (page_ops
& PAGE_CLEAR_DIRTY
)
1854 clear_page_dirty_for_io(pages
[i
]);
1855 if (page_ops
& PAGE_SET_WRITEBACK
)
1856 set_page_writeback(pages
[i
]);
1857 if (page_ops
& PAGE_SET_ERROR
)
1858 SetPageError(pages
[i
]);
1859 if (page_ops
& PAGE_END_WRITEBACK
)
1860 end_page_writeback(pages
[i
]);
1861 if (page_ops
& PAGE_UNLOCK
)
1862 unlock_page(pages
[i
]);
1863 page_cache_release(pages
[i
]);
1872 * count the number of bytes in the tree that have a given bit(s)
1873 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1874 * cached. The total number found is returned.
1876 u64
count_range_bits(struct extent_io_tree
*tree
,
1877 u64
*start
, u64 search_end
, u64 max_bytes
,
1878 unsigned bits
, int contig
)
1880 struct rb_node
*node
;
1881 struct extent_state
*state
;
1882 u64 cur_start
= *start
;
1883 u64 total_bytes
= 0;
1887 if (WARN_ON(search_end
<= cur_start
))
1890 spin_lock(&tree
->lock
);
1891 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1892 total_bytes
= tree
->dirty_bytes
;
1896 * this search will find all the extents that end after
1899 node
= tree_search(tree
, cur_start
);
1904 state
= rb_entry(node
, struct extent_state
, rb_node
);
1905 if (state
->start
> search_end
)
1907 if (contig
&& found
&& state
->start
> last
+ 1)
1909 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1910 total_bytes
+= min(search_end
, state
->end
) + 1 -
1911 max(cur_start
, state
->start
);
1912 if (total_bytes
>= max_bytes
)
1915 *start
= max(cur_start
, state
->start
);
1919 } else if (contig
&& found
) {
1922 node
= rb_next(node
);
1927 spin_unlock(&tree
->lock
);
1932 * set the private field for a given byte offset in the tree. If there isn't
1933 * an extent_state there already, this does nothing.
1935 static int set_state_private(struct extent_io_tree
*tree
, u64 start
, u64
private)
1937 struct rb_node
*node
;
1938 struct extent_state
*state
;
1941 spin_lock(&tree
->lock
);
1943 * this search will find all the extents that end after
1946 node
= tree_search(tree
, start
);
1951 state
= rb_entry(node
, struct extent_state
, rb_node
);
1952 if (state
->start
!= start
) {
1956 state
->private = private;
1958 spin_unlock(&tree
->lock
);
1962 int get_state_private(struct extent_io_tree
*tree
, u64 start
, u64
*private)
1964 struct rb_node
*node
;
1965 struct extent_state
*state
;
1968 spin_lock(&tree
->lock
);
1970 * this search will find all the extents that end after
1973 node
= tree_search(tree
, start
);
1978 state
= rb_entry(node
, struct extent_state
, rb_node
);
1979 if (state
->start
!= start
) {
1983 *private = state
->private;
1985 spin_unlock(&tree
->lock
);
1990 * searches a range in the state tree for a given mask.
1991 * If 'filled' == 1, this returns 1 only if every extent in the tree
1992 * has the bits set. Otherwise, 1 is returned if any bit in the
1993 * range is found set.
1995 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1996 unsigned bits
, int filled
, struct extent_state
*cached
)
1998 struct extent_state
*state
= NULL
;
1999 struct rb_node
*node
;
2002 spin_lock(&tree
->lock
);
2003 if (cached
&& extent_state_in_tree(cached
) && cached
->start
<= start
&&
2004 cached
->end
> start
)
2005 node
= &cached
->rb_node
;
2007 node
= tree_search(tree
, start
);
2008 while (node
&& start
<= end
) {
2009 state
= rb_entry(node
, struct extent_state
, rb_node
);
2011 if (filled
&& state
->start
> start
) {
2016 if (state
->start
> end
)
2019 if (state
->state
& bits
) {
2023 } else if (filled
) {
2028 if (state
->end
== (u64
)-1)
2031 start
= state
->end
+ 1;
2034 node
= rb_next(node
);
2041 spin_unlock(&tree
->lock
);
2046 * helper function to set a given page up to date if all the
2047 * extents in the tree for that page are up to date
2049 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
2051 u64 start
= page_offset(page
);
2052 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
2053 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
2054 SetPageUptodate(page
);
2057 int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
)
2061 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2063 set_state_private(failure_tree
, rec
->start
, 0);
2064 ret
= clear_extent_bits(failure_tree
, rec
->start
,
2065 rec
->start
+ rec
->len
- 1,
2066 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2070 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
2071 rec
->start
+ rec
->len
- 1,
2072 EXTENT_DAMAGED
, GFP_NOFS
);
2081 * this bypasses the standard btrfs submit functions deliberately, as
2082 * the standard behavior is to write all copies in a raid setup. here we only
2083 * want to write the one bad copy. so we do the mapping for ourselves and issue
2084 * submit_bio directly.
2085 * to avoid any synchronization issues, wait for the data after writing, which
2086 * actually prevents the read that triggered the error from finishing.
2087 * currently, there can be no more than two copies of every data bit. thus,
2088 * exactly one rewrite is required.
2090 int repair_io_failure(struct inode
*inode
, u64 start
, u64 length
, u64 logical
,
2091 struct page
*page
, unsigned int pg_offset
, int mirror_num
)
2093 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2095 struct btrfs_device
*dev
;
2098 struct btrfs_bio
*bbio
= NULL
;
2099 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2102 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2103 BUG_ON(!mirror_num
);
2105 /* we can't repair anything in raid56 yet */
2106 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2109 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2112 bio
->bi_iter
.bi_size
= 0;
2113 map_length
= length
;
2115 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2116 &map_length
, &bbio
, mirror_num
);
2121 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2122 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2123 bio
->bi_iter
.bi_sector
= sector
;
2124 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2125 btrfs_put_bbio(bbio
);
2126 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2130 bio
->bi_bdev
= dev
->bdev
;
2131 bio_add_page(bio
, page
, length
, pg_offset
);
2133 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
)) {
2134 /* try to remap that extent elsewhere? */
2136 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2140 btrfs_info_rl_in_rcu(fs_info
,
2141 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2142 btrfs_ino(inode
), start
,
2143 rcu_str_deref(dev
->name
), sector
);
2148 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2151 u64 start
= eb
->start
;
2152 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2155 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2158 for (i
= 0; i
< num_pages
; i
++) {
2159 struct page
*p
= eb
->pages
[i
];
2161 ret
= repair_io_failure(root
->fs_info
->btree_inode
, start
,
2162 PAGE_CACHE_SIZE
, start
, p
,
2163 start
- page_offset(p
), mirror_num
);
2166 start
+= PAGE_CACHE_SIZE
;
2173 * each time an IO finishes, we do a fast check in the IO failure tree
2174 * to see if we need to process or clean up an io_failure_record
2176 int clean_io_failure(struct inode
*inode
, u64 start
, struct page
*page
,
2177 unsigned int pg_offset
)
2180 u64 private_failure
;
2181 struct io_failure_record
*failrec
;
2182 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2183 struct extent_state
*state
;
2188 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2189 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2193 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
, start
,
2198 failrec
= (struct io_failure_record
*)(unsigned long) private_failure
;
2199 BUG_ON(!failrec
->this_mirror
);
2201 if (failrec
->in_validation
) {
2202 /* there was no real error, just free the record */
2203 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2207 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2210 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2211 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2214 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2216 if (state
&& state
->start
<= failrec
->start
&&
2217 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2218 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2220 if (num_copies
> 1) {
2221 repair_io_failure(inode
, start
, failrec
->len
,
2222 failrec
->logical
, page
,
2223 pg_offset
, failrec
->failed_mirror
);
2228 free_io_failure(inode
, failrec
);
2234 * Can be called when
2235 * - hold extent lock
2236 * - under ordered extent
2237 * - the inode is freeing
2239 void btrfs_free_io_failure_record(struct inode
*inode
, u64 start
, u64 end
)
2241 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2242 struct io_failure_record
*failrec
;
2243 struct extent_state
*state
, *next
;
2245 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2248 spin_lock(&failure_tree
->lock
);
2249 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2251 if (state
->start
> end
)
2254 ASSERT(state
->end
<= end
);
2256 next
= next_state(state
);
2258 failrec
= (struct io_failure_record
*)(unsigned long)state
->private;
2259 free_extent_state(state
);
2264 spin_unlock(&failure_tree
->lock
);
2267 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2268 struct io_failure_record
**failrec_ret
)
2270 struct io_failure_record
*failrec
;
2272 struct extent_map
*em
;
2273 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2274 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2275 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2279 ret
= get_state_private(failure_tree
, start
, &private);
2281 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2285 failrec
->start
= start
;
2286 failrec
->len
= end
- start
+ 1;
2287 failrec
->this_mirror
= 0;
2288 failrec
->bio_flags
= 0;
2289 failrec
->in_validation
= 0;
2291 read_lock(&em_tree
->lock
);
2292 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2294 read_unlock(&em_tree
->lock
);
2299 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2300 free_extent_map(em
);
2303 read_unlock(&em_tree
->lock
);
2309 logical
= start
- em
->start
;
2310 logical
= em
->block_start
+ logical
;
2311 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2312 logical
= em
->block_start
;
2313 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2314 extent_set_compress_type(&failrec
->bio_flags
,
2318 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2319 logical
, start
, failrec
->len
);
2321 failrec
->logical
= logical
;
2322 free_extent_map(em
);
2324 /* set the bits in the private failure tree */
2325 ret
= set_extent_bits(failure_tree
, start
, end
,
2326 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
2328 ret
= set_state_private(failure_tree
, start
,
2329 (u64
)(unsigned long)failrec
);
2330 /* set the bits in the inode's tree */
2332 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
,
2339 failrec
= (struct io_failure_record
*)(unsigned long)private;
2340 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2341 failrec
->logical
, failrec
->start
, failrec
->len
,
2342 failrec
->in_validation
);
2344 * when data can be on disk more than twice, add to failrec here
2345 * (e.g. with a list for failed_mirror) to make
2346 * clean_io_failure() clean all those errors at once.
2350 *failrec_ret
= failrec
;
2355 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2356 struct io_failure_record
*failrec
, int failed_mirror
)
2360 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2361 failrec
->logical
, failrec
->len
);
2362 if (num_copies
== 1) {
2364 * we only have a single copy of the data, so don't bother with
2365 * all the retry and error correction code that follows. no
2366 * matter what the error is, it is very likely to persist.
2368 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2369 num_copies
, failrec
->this_mirror
, failed_mirror
);
2374 * there are two premises:
2375 * a) deliver good data to the caller
2376 * b) correct the bad sectors on disk
2378 if (failed_bio
->bi_vcnt
> 1) {
2380 * to fulfill b), we need to know the exact failing sectors, as
2381 * we don't want to rewrite any more than the failed ones. thus,
2382 * we need separate read requests for the failed bio
2384 * if the following BUG_ON triggers, our validation request got
2385 * merged. we need separate requests for our algorithm to work.
2387 BUG_ON(failrec
->in_validation
);
2388 failrec
->in_validation
= 1;
2389 failrec
->this_mirror
= failed_mirror
;
2392 * we're ready to fulfill a) and b) alongside. get a good copy
2393 * of the failed sector and if we succeed, we have setup
2394 * everything for repair_io_failure to do the rest for us.
2396 if (failrec
->in_validation
) {
2397 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2398 failrec
->in_validation
= 0;
2399 failrec
->this_mirror
= 0;
2401 failrec
->failed_mirror
= failed_mirror
;
2402 failrec
->this_mirror
++;
2403 if (failrec
->this_mirror
== failed_mirror
)
2404 failrec
->this_mirror
++;
2407 if (failrec
->this_mirror
> num_copies
) {
2408 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2409 num_copies
, failrec
->this_mirror
, failed_mirror
);
2417 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2418 struct io_failure_record
*failrec
,
2419 struct page
*page
, int pg_offset
, int icsum
,
2420 bio_end_io_t
*endio_func
, void *data
)
2423 struct btrfs_io_bio
*btrfs_failed_bio
;
2424 struct btrfs_io_bio
*btrfs_bio
;
2426 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2430 bio
->bi_end_io
= endio_func
;
2431 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2432 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2433 bio
->bi_iter
.bi_size
= 0;
2434 bio
->bi_private
= data
;
2436 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2437 if (btrfs_failed_bio
->csum
) {
2438 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2439 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2441 btrfs_bio
= btrfs_io_bio(bio
);
2442 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2444 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2448 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2454 * this is a generic handler for readpage errors (default
2455 * readpage_io_failed_hook). if other copies exist, read those and write back
2456 * good data to the failed position. does not investigate in remapping the
2457 * failed extent elsewhere, hoping the device will be smart enough to do this as
2461 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2462 struct page
*page
, u64 start
, u64 end
,
2465 struct io_failure_record
*failrec
;
2466 struct inode
*inode
= page
->mapping
->host
;
2467 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2472 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
2474 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2478 ret
= btrfs_check_repairable(inode
, failed_bio
, failrec
, failed_mirror
);
2480 free_io_failure(inode
, failrec
);
2484 if (failed_bio
->bi_vcnt
> 1)
2485 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2487 read_mode
= READ_SYNC
;
2489 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2490 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2491 start
- page_offset(page
),
2492 (int)phy_offset
, failed_bio
->bi_end_io
,
2495 free_io_failure(inode
, failrec
);
2499 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2500 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2502 ret
= tree
->ops
->submit_bio_hook(inode
, read_mode
, bio
,
2503 failrec
->this_mirror
,
2504 failrec
->bio_flags
, 0);
2506 free_io_failure(inode
, failrec
);
2513 /* lots and lots of room for performance fixes in the end_bio funcs */
2515 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2517 int uptodate
= (err
== 0);
2518 struct extent_io_tree
*tree
;
2521 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2523 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2524 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2525 end
, NULL
, uptodate
);
2531 ClearPageUptodate(page
);
2533 ret
= ret
< 0 ? ret
: -EIO
;
2534 mapping_set_error(page
->mapping
, ret
);
2539 * after a writepage IO is done, we need to:
2540 * clear the uptodate bits on error
2541 * clear the writeback bits in the extent tree for this IO
2542 * end_page_writeback if the page has no more pending IO
2544 * Scheduling is not allowed, so the extent state tree is expected
2545 * to have one and only one object corresponding to this IO.
2547 static void end_bio_extent_writepage(struct bio
*bio
)
2549 struct bio_vec
*bvec
;
2554 bio_for_each_segment_all(bvec
, bio
, i
) {
2555 struct page
*page
= bvec
->bv_page
;
2557 /* We always issue full-page reads, but if some block
2558 * in a page fails to read, blk_update_request() will
2559 * advance bv_offset and adjust bv_len to compensate.
2560 * Print a warning for nonzero offsets, and an error
2561 * if they don't add up to a full page. */
2562 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2563 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2564 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2565 "partial page write in btrfs with offset %u and length %u",
2566 bvec
->bv_offset
, bvec
->bv_len
);
2568 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2569 "incomplete page write in btrfs with offset %u and "
2571 bvec
->bv_offset
, bvec
->bv_len
);
2574 start
= page_offset(page
);
2575 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2577 end_extent_writepage(page
, bio
->bi_error
, start
, end
);
2578 end_page_writeback(page
);
2585 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2588 struct extent_state
*cached
= NULL
;
2589 u64 end
= start
+ len
- 1;
2591 if (uptodate
&& tree
->track_uptodate
)
2592 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2593 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2597 * after a readpage IO is done, we need to:
2598 * clear the uptodate bits on error
2599 * set the uptodate bits if things worked
2600 * set the page up to date if all extents in the tree are uptodate
2601 * clear the lock bit in the extent tree
2602 * unlock the page if there are no other extents locked for it
2604 * Scheduling is not allowed, so the extent state tree is expected
2605 * to have one and only one object corresponding to this IO.
2607 static void end_bio_extent_readpage(struct bio
*bio
)
2609 struct bio_vec
*bvec
;
2610 int uptodate
= !bio
->bi_error
;
2611 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2612 struct extent_io_tree
*tree
;
2617 u64 extent_start
= 0;
2623 bio_for_each_segment_all(bvec
, bio
, i
) {
2624 struct page
*page
= bvec
->bv_page
;
2625 struct inode
*inode
= page
->mapping
->host
;
2627 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2628 "mirror=%u\n", (u64
)bio
->bi_iter
.bi_sector
,
2629 bio
->bi_error
, io_bio
->mirror_num
);
2630 tree
= &BTRFS_I(inode
)->io_tree
;
2632 /* We always issue full-page reads, but if some block
2633 * in a page fails to read, blk_update_request() will
2634 * advance bv_offset and adjust bv_len to compensate.
2635 * Print a warning for nonzero offsets, and an error
2636 * if they don't add up to a full page. */
2637 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_CACHE_SIZE
) {
2638 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_CACHE_SIZE
)
2639 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2640 "partial page read in btrfs with offset %u and length %u",
2641 bvec
->bv_offset
, bvec
->bv_len
);
2643 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2644 "incomplete page read in btrfs with offset %u and "
2646 bvec
->bv_offset
, bvec
->bv_len
);
2649 start
= page_offset(page
);
2650 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2653 mirror
= io_bio
->mirror_num
;
2654 if (likely(uptodate
&& tree
->ops
&&
2655 tree
->ops
->readpage_end_io_hook
)) {
2656 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2662 clean_io_failure(inode
, start
, page
, 0);
2665 if (likely(uptodate
))
2668 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2669 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2670 if (!ret
&& !bio
->bi_error
)
2674 * The generic bio_readpage_error handles errors the
2675 * following way: If possible, new read requests are
2676 * created and submitted and will end up in
2677 * end_bio_extent_readpage as well (if we're lucky, not
2678 * in the !uptodate case). In that case it returns 0 and
2679 * we just go on with the next page in our bio. If it
2680 * can't handle the error it will return -EIO and we
2681 * remain responsible for that page.
2683 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2686 uptodate
= !bio
->bi_error
;
2692 if (likely(uptodate
)) {
2693 loff_t i_size
= i_size_read(inode
);
2694 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2697 /* Zero out the end if this page straddles i_size */
2698 off
= i_size
& (PAGE_CACHE_SIZE
-1);
2699 if (page
->index
== end_index
&& off
)
2700 zero_user_segment(page
, off
, PAGE_CACHE_SIZE
);
2701 SetPageUptodate(page
);
2703 ClearPageUptodate(page
);
2709 if (unlikely(!uptodate
)) {
2711 endio_readpage_release_extent(tree
,
2717 endio_readpage_release_extent(tree
, start
,
2718 end
- start
+ 1, 0);
2719 } else if (!extent_len
) {
2720 extent_start
= start
;
2721 extent_len
= end
+ 1 - start
;
2722 } else if (extent_start
+ extent_len
== start
) {
2723 extent_len
+= end
+ 1 - start
;
2725 endio_readpage_release_extent(tree
, extent_start
,
2726 extent_len
, uptodate
);
2727 extent_start
= start
;
2728 extent_len
= end
+ 1 - start
;
2733 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2736 io_bio
->end_io(io_bio
, bio
->bi_error
);
2741 * this allocates from the btrfs_bioset. We're returning a bio right now
2742 * but you can call btrfs_io_bio for the appropriate container_of magic
2745 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2748 struct btrfs_io_bio
*btrfs_bio
;
2751 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2753 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2754 while (!bio
&& (nr_vecs
/= 2)) {
2755 bio
= bio_alloc_bioset(gfp_flags
,
2756 nr_vecs
, btrfs_bioset
);
2761 bio
->bi_bdev
= bdev
;
2762 bio
->bi_iter
.bi_sector
= first_sector
;
2763 btrfs_bio
= btrfs_io_bio(bio
);
2764 btrfs_bio
->csum
= NULL
;
2765 btrfs_bio
->csum_allocated
= NULL
;
2766 btrfs_bio
->end_io
= NULL
;
2771 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2773 struct btrfs_io_bio
*btrfs_bio
;
2776 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2778 btrfs_bio
= btrfs_io_bio(new);
2779 btrfs_bio
->csum
= NULL
;
2780 btrfs_bio
->csum_allocated
= NULL
;
2781 btrfs_bio
->end_io
= NULL
;
2783 #ifdef CONFIG_BLK_CGROUP
2784 /* FIXME, put this into bio_clone_bioset */
2786 bio_associate_blkcg(new, bio
->bi_css
);
2792 /* this also allocates from the btrfs_bioset */
2793 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2795 struct btrfs_io_bio
*btrfs_bio
;
2798 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2800 btrfs_bio
= btrfs_io_bio(bio
);
2801 btrfs_bio
->csum
= NULL
;
2802 btrfs_bio
->csum_allocated
= NULL
;
2803 btrfs_bio
->end_io
= NULL
;
2809 static int __must_check
submit_one_bio(int rw
, struct bio
*bio
,
2810 int mirror_num
, unsigned long bio_flags
)
2813 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2814 struct page
*page
= bvec
->bv_page
;
2815 struct extent_io_tree
*tree
= bio
->bi_private
;
2818 start
= page_offset(page
) + bvec
->bv_offset
;
2820 bio
->bi_private
= NULL
;
2824 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2825 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, rw
, bio
,
2826 mirror_num
, bio_flags
, start
);
2828 btrfsic_submit_bio(rw
, bio
);
2834 static int merge_bio(int rw
, struct extent_io_tree
*tree
, struct page
*page
,
2835 unsigned long offset
, size_t size
, struct bio
*bio
,
2836 unsigned long bio_flags
)
2839 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2840 ret
= tree
->ops
->merge_bio_hook(rw
, page
, offset
, size
, bio
,
2847 static int submit_extent_page(int rw
, struct extent_io_tree
*tree
,
2848 struct writeback_control
*wbc
,
2849 struct page
*page
, sector_t sector
,
2850 size_t size
, unsigned long offset
,
2851 struct block_device
*bdev
,
2852 struct bio
**bio_ret
,
2853 unsigned long max_pages
,
2854 bio_end_io_t end_io_func
,
2856 unsigned long prev_bio_flags
,
2857 unsigned long bio_flags
,
2858 bool force_bio_submit
)
2863 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2864 size_t page_size
= min_t(size_t, size
, PAGE_CACHE_SIZE
);
2866 if (bio_ret
&& *bio_ret
) {
2869 contig
= bio
->bi_iter
.bi_sector
== sector
;
2871 contig
= bio_end_sector(bio
) == sector
;
2873 if (prev_bio_flags
!= bio_flags
|| !contig
||
2875 merge_bio(rw
, tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2876 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2877 ret
= submit_one_bio(rw
, bio
, mirror_num
,
2886 wbc_account_io(wbc
, page
, page_size
);
2891 bio
= btrfs_bio_alloc(bdev
, sector
, BIO_MAX_PAGES
,
2892 GFP_NOFS
| __GFP_HIGH
);
2896 bio_add_page(bio
, page
, page_size
, offset
);
2897 bio
->bi_end_io
= end_io_func
;
2898 bio
->bi_private
= tree
;
2900 wbc_init_bio(wbc
, bio
);
2901 wbc_account_io(wbc
, page
, page_size
);
2907 ret
= submit_one_bio(rw
, bio
, mirror_num
, bio_flags
);
2912 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2915 if (!PagePrivate(page
)) {
2916 SetPagePrivate(page
);
2917 page_cache_get(page
);
2918 set_page_private(page
, (unsigned long)eb
);
2920 WARN_ON(page
->private != (unsigned long)eb
);
2924 void set_page_extent_mapped(struct page
*page
)
2926 if (!PagePrivate(page
)) {
2927 SetPagePrivate(page
);
2928 page_cache_get(page
);
2929 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2933 static struct extent_map
*
2934 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2935 u64 start
, u64 len
, get_extent_t
*get_extent
,
2936 struct extent_map
**em_cached
)
2938 struct extent_map
*em
;
2940 if (em_cached
&& *em_cached
) {
2942 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2943 start
< extent_map_end(em
)) {
2944 atomic_inc(&em
->refs
);
2948 free_extent_map(em
);
2952 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2953 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2955 atomic_inc(&em
->refs
);
2961 * basic readpage implementation. Locked extent state structs are inserted
2962 * into the tree that are removed when the IO is done (by the end_io
2964 * XXX JDM: This needs looking at to ensure proper page locking
2966 static int __do_readpage(struct extent_io_tree
*tree
,
2968 get_extent_t
*get_extent
,
2969 struct extent_map
**em_cached
,
2970 struct bio
**bio
, int mirror_num
,
2971 unsigned long *bio_flags
, int rw
,
2974 struct inode
*inode
= page
->mapping
->host
;
2975 u64 start
= page_offset(page
);
2976 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
2980 u64 last_byte
= i_size_read(inode
);
2984 struct extent_map
*em
;
2985 struct block_device
*bdev
;
2988 int parent_locked
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2989 size_t pg_offset
= 0;
2991 size_t disk_io_size
;
2992 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2993 unsigned long this_bio_flag
= *bio_flags
& EXTENT_BIO_PARENT_LOCKED
;
2995 set_page_extent_mapped(page
);
2998 if (!PageUptodate(page
)) {
2999 if (cleancache_get_page(page
) == 0) {
3000 BUG_ON(blocksize
!= PAGE_SIZE
);
3001 unlock_extent(tree
, start
, end
);
3006 if (page
->index
== last_byte
>> PAGE_CACHE_SHIFT
) {
3008 size_t zero_offset
= last_byte
& (PAGE_CACHE_SIZE
- 1);
3011 iosize
= PAGE_CACHE_SIZE
- zero_offset
;
3012 userpage
= kmap_atomic(page
);
3013 memset(userpage
+ zero_offset
, 0, iosize
);
3014 flush_dcache_page(page
);
3015 kunmap_atomic(userpage
);
3018 while (cur
<= end
) {
3019 unsigned long pnr
= (last_byte
>> PAGE_CACHE_SHIFT
) + 1;
3020 bool force_bio_submit
= false;
3022 if (cur
>= last_byte
) {
3024 struct extent_state
*cached
= NULL
;
3026 iosize
= PAGE_CACHE_SIZE
- pg_offset
;
3027 userpage
= kmap_atomic(page
);
3028 memset(userpage
+ pg_offset
, 0, iosize
);
3029 flush_dcache_page(page
);
3030 kunmap_atomic(userpage
);
3031 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3034 unlock_extent_cached(tree
, cur
,
3039 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
3040 end
- cur
+ 1, get_extent
, em_cached
);
3041 if (IS_ERR_OR_NULL(em
)) {
3044 unlock_extent(tree
, cur
, end
);
3047 extent_offset
= cur
- em
->start
;
3048 BUG_ON(extent_map_end(em
) <= cur
);
3051 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3052 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
3053 extent_set_compress_type(&this_bio_flag
,
3057 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
3058 cur_end
= min(extent_map_end(em
) - 1, end
);
3059 iosize
= ALIGN(iosize
, blocksize
);
3060 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
3061 disk_io_size
= em
->block_len
;
3062 sector
= em
->block_start
>> 9;
3064 sector
= (em
->block_start
+ extent_offset
) >> 9;
3065 disk_io_size
= iosize
;
3068 block_start
= em
->block_start
;
3069 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
3070 block_start
= EXTENT_MAP_HOLE
;
3073 * If we have a file range that points to a compressed extent
3074 * and it's followed by a consecutive file range that points to
3075 * to the same compressed extent (possibly with a different
3076 * offset and/or length, so it either points to the whole extent
3077 * or only part of it), we must make sure we do not submit a
3078 * single bio to populate the pages for the 2 ranges because
3079 * this makes the compressed extent read zero out the pages
3080 * belonging to the 2nd range. Imagine the following scenario:
3083 * [0 - 8K] [8K - 24K]
3086 * points to extent X, points to extent X,
3087 * offset 4K, length of 8K offset 0, length 16K
3089 * [extent X, compressed length = 4K uncompressed length = 16K]
3091 * If the bio to read the compressed extent covers both ranges,
3092 * it will decompress extent X into the pages belonging to the
3093 * first range and then it will stop, zeroing out the remaining
3094 * pages that belong to the other range that points to extent X.
3095 * So here we make sure we submit 2 bios, one for the first
3096 * range and another one for the third range. Both will target
3097 * the same physical extent from disk, but we can't currently
3098 * make the compressed bio endio callback populate the pages
3099 * for both ranges because each compressed bio is tightly
3100 * coupled with a single extent map, and each range can have
3101 * an extent map with a different offset value relative to the
3102 * uncompressed data of our extent and different lengths. This
3103 * is a corner case so we prioritize correctness over
3104 * non-optimal behavior (submitting 2 bios for the same extent).
3106 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3107 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3108 *prev_em_start
!= em
->orig_start
)
3109 force_bio_submit
= true;
3112 *prev_em_start
= em
->orig_start
;
3114 free_extent_map(em
);
3117 /* we've found a hole, just zero and go on */
3118 if (block_start
== EXTENT_MAP_HOLE
) {
3120 struct extent_state
*cached
= NULL
;
3122 userpage
= kmap_atomic(page
);
3123 memset(userpage
+ pg_offset
, 0, iosize
);
3124 flush_dcache_page(page
);
3125 kunmap_atomic(userpage
);
3127 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3130 free_extent_state(cached
);
3132 unlock_extent_cached(tree
, cur
,
3136 pg_offset
+= iosize
;
3139 /* the get_extent function already copied into the page */
3140 if (test_range_bit(tree
, cur
, cur_end
,
3141 EXTENT_UPTODATE
, 1, NULL
)) {
3142 check_page_uptodate(tree
, page
);
3144 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3146 pg_offset
+= iosize
;
3149 /* we have an inline extent but it didn't get marked up
3150 * to date. Error out
3152 if (block_start
== EXTENT_MAP_INLINE
) {
3155 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3157 pg_offset
+= iosize
;
3162 ret
= submit_extent_page(rw
, tree
, NULL
, page
,
3163 sector
, disk_io_size
, pg_offset
,
3165 end_bio_extent_readpage
, mirror_num
,
3171 *bio_flags
= this_bio_flag
;
3175 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3178 pg_offset
+= iosize
;
3182 if (!PageError(page
))
3183 SetPageUptodate(page
);
3189 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3190 struct page
*pages
[], int nr_pages
,
3192 get_extent_t
*get_extent
,
3193 struct extent_map
**em_cached
,
3194 struct bio
**bio
, int mirror_num
,
3195 unsigned long *bio_flags
, int rw
,
3198 struct inode
*inode
;
3199 struct btrfs_ordered_extent
*ordered
;
3202 inode
= pages
[0]->mapping
->host
;
3204 lock_extent(tree
, start
, end
);
3205 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3209 unlock_extent(tree
, start
, end
);
3210 btrfs_start_ordered_extent(inode
, ordered
, 1);
3211 btrfs_put_ordered_extent(ordered
);
3214 for (index
= 0; index
< nr_pages
; index
++) {
3215 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3216 mirror_num
, bio_flags
, rw
, prev_em_start
);
3217 page_cache_release(pages
[index
]);
3221 static void __extent_readpages(struct extent_io_tree
*tree
,
3222 struct page
*pages
[],
3223 int nr_pages
, get_extent_t
*get_extent
,
3224 struct extent_map
**em_cached
,
3225 struct bio
**bio
, int mirror_num
,
3226 unsigned long *bio_flags
, int rw
,
3233 int first_index
= 0;
3235 for (index
= 0; index
< nr_pages
; index
++) {
3236 page_start
= page_offset(pages
[index
]);
3239 end
= start
+ PAGE_CACHE_SIZE
- 1;
3240 first_index
= index
;
3241 } else if (end
+ 1 == page_start
) {
3242 end
+= PAGE_CACHE_SIZE
;
3244 __do_contiguous_readpages(tree
, &pages
[first_index
],
3245 index
- first_index
, start
,
3246 end
, get_extent
, em_cached
,
3247 bio
, mirror_num
, bio_flags
,
3250 end
= start
+ PAGE_CACHE_SIZE
- 1;
3251 first_index
= index
;
3256 __do_contiguous_readpages(tree
, &pages
[first_index
],
3257 index
- first_index
, start
,
3258 end
, get_extent
, em_cached
, bio
,
3259 mirror_num
, bio_flags
, rw
,
3263 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3265 get_extent_t
*get_extent
,
3266 struct bio
**bio
, int mirror_num
,
3267 unsigned long *bio_flags
, int rw
)
3269 struct inode
*inode
= page
->mapping
->host
;
3270 struct btrfs_ordered_extent
*ordered
;
3271 u64 start
= page_offset(page
);
3272 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
3276 lock_extent(tree
, start
, end
);
3277 ordered
= btrfs_lookup_ordered_extent(inode
, start
);
3280 unlock_extent(tree
, start
, end
);
3281 btrfs_start_ordered_extent(inode
, ordered
, 1);
3282 btrfs_put_ordered_extent(ordered
);
3285 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3286 bio_flags
, rw
, NULL
);
3290 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3291 get_extent_t
*get_extent
, int mirror_num
)
3293 struct bio
*bio
= NULL
;
3294 unsigned long bio_flags
= 0;
3297 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3300 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3304 int extent_read_full_page_nolock(struct extent_io_tree
*tree
, struct page
*page
,
3305 get_extent_t
*get_extent
, int mirror_num
)
3307 struct bio
*bio
= NULL
;
3308 unsigned long bio_flags
= EXTENT_BIO_PARENT_LOCKED
;
3311 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, &bio
, mirror_num
,
3312 &bio_flags
, READ
, NULL
);
3314 ret
= submit_one_bio(READ
, bio
, mirror_num
, bio_flags
);
3318 static noinline
void update_nr_written(struct page
*page
,
3319 struct writeback_control
*wbc
,
3320 unsigned long nr_written
)
3322 wbc
->nr_to_write
-= nr_written
;
3323 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 &&
3324 wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
))
3325 page
->mapping
->writeback_index
= page
->index
+ nr_written
;
3329 * helper for __extent_writepage, doing all of the delayed allocation setup.
3331 * This returns 1 if our fill_delalloc function did all the work required
3332 * to write the page (copy into inline extent). In this case the IO has
3333 * been started and the page is already unlocked.
3335 * This returns 0 if all went well (page still locked)
3336 * This returns < 0 if there were errors (page still locked)
3338 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3339 struct page
*page
, struct writeback_control
*wbc
,
3340 struct extent_page_data
*epd
,
3342 unsigned long *nr_written
)
3344 struct extent_io_tree
*tree
= epd
->tree
;
3345 u64 page_end
= delalloc_start
+ PAGE_CACHE_SIZE
- 1;
3347 u64 delalloc_to_write
= 0;
3348 u64 delalloc_end
= 0;
3350 int page_started
= 0;
3352 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3355 while (delalloc_end
< page_end
) {
3356 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3360 BTRFS_MAX_EXTENT_SIZE
);
3361 if (nr_delalloc
== 0) {
3362 delalloc_start
= delalloc_end
+ 1;
3365 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3370 /* File system has been set read-only */
3373 /* fill_delalloc should be return < 0 for error
3374 * but just in case, we use > 0 here meaning the
3375 * IO is started, so we don't want to return > 0
3376 * unless things are going well.
3378 ret
= ret
< 0 ? ret
: -EIO
;
3382 * delalloc_end is already one less than the total
3383 * length, so we don't subtract one from
3386 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3389 delalloc_start
= delalloc_end
+ 1;
3391 if (wbc
->nr_to_write
< delalloc_to_write
) {
3394 if (delalloc_to_write
< thresh
* 2)
3395 thresh
= delalloc_to_write
;
3396 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3400 /* did the fill delalloc function already unlock and start
3405 * we've unlocked the page, so we can't update
3406 * the mapping's writeback index, just update
3409 wbc
->nr_to_write
-= *nr_written
;
3420 * helper for __extent_writepage. This calls the writepage start hooks,
3421 * and does the loop to map the page into extents and bios.
3423 * We return 1 if the IO is started and the page is unlocked,
3424 * 0 if all went well (page still locked)
3425 * < 0 if there were errors (page still locked)
3427 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3429 struct writeback_control
*wbc
,
3430 struct extent_page_data
*epd
,
3432 unsigned long nr_written
,
3433 int write_flags
, int *nr_ret
)
3435 struct extent_io_tree
*tree
= epd
->tree
;
3436 u64 start
= page_offset(page
);
3437 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3444 struct extent_state
*cached_state
= NULL
;
3445 struct extent_map
*em
;
3446 struct block_device
*bdev
;
3447 size_t pg_offset
= 0;
3453 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3454 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3457 /* Fixup worker will requeue */
3459 wbc
->pages_skipped
++;
3461 redirty_page_for_writepage(wbc
, page
);
3463 update_nr_written(page
, wbc
, nr_written
);
3471 * we don't want to touch the inode after unlocking the page,
3472 * so we update the mapping writeback index now
3474 update_nr_written(page
, wbc
, nr_written
+ 1);
3477 if (i_size
<= start
) {
3478 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3479 tree
->ops
->writepage_end_io_hook(page
, start
,
3484 blocksize
= inode
->i_sb
->s_blocksize
;
3486 while (cur
<= end
) {
3488 if (cur
>= i_size
) {
3489 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3490 tree
->ops
->writepage_end_io_hook(page
, cur
,
3494 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3496 if (IS_ERR_OR_NULL(em
)) {
3498 ret
= PTR_ERR_OR_ZERO(em
);
3502 extent_offset
= cur
- em
->start
;
3503 em_end
= extent_map_end(em
);
3504 BUG_ON(em_end
<= cur
);
3506 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3507 iosize
= ALIGN(iosize
, blocksize
);
3508 sector
= (em
->block_start
+ extent_offset
) >> 9;
3510 block_start
= em
->block_start
;
3511 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3512 free_extent_map(em
);
3516 * compressed and inline extents are written through other
3519 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3520 block_start
== EXTENT_MAP_INLINE
) {
3522 * end_io notification does not happen here for
3523 * compressed extents
3525 if (!compressed
&& tree
->ops
&&
3526 tree
->ops
->writepage_end_io_hook
)
3527 tree
->ops
->writepage_end_io_hook(page
, cur
,
3530 else if (compressed
) {
3531 /* we don't want to end_page_writeback on
3532 * a compressed extent. this happens
3539 pg_offset
+= iosize
;
3543 if (tree
->ops
&& tree
->ops
->writepage_io_hook
) {
3544 ret
= tree
->ops
->writepage_io_hook(page
, cur
,
3552 unsigned long max_nr
= (i_size
>> PAGE_CACHE_SHIFT
) + 1;
3554 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3555 if (!PageWriteback(page
)) {
3556 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3557 "page %lu not writeback, cur %llu end %llu",
3558 page
->index
, cur
, end
);
3561 ret
= submit_extent_page(write_flags
, tree
, wbc
, page
,
3562 sector
, iosize
, pg_offset
,
3563 bdev
, &epd
->bio
, max_nr
,
3564 end_bio_extent_writepage
,
3570 pg_offset
+= iosize
;
3578 /* drop our reference on any cached states */
3579 free_extent_state(cached_state
);
3584 * the writepage semantics are similar to regular writepage. extent
3585 * records are inserted to lock ranges in the tree, and as dirty areas
3586 * are found, they are marked writeback. Then the lock bits are removed
3587 * and the end_io handler clears the writeback ranges
3589 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3592 struct inode
*inode
= page
->mapping
->host
;
3593 struct extent_page_data
*epd
= data
;
3594 u64 start
= page_offset(page
);
3595 u64 page_end
= start
+ PAGE_CACHE_SIZE
- 1;
3598 size_t pg_offset
= 0;
3599 loff_t i_size
= i_size_read(inode
);
3600 unsigned long end_index
= i_size
>> PAGE_CACHE_SHIFT
;
3602 unsigned long nr_written
= 0;
3604 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3605 write_flags
= WRITE_SYNC
;
3607 write_flags
= WRITE
;
3609 trace___extent_writepage(page
, inode
, wbc
);
3611 WARN_ON(!PageLocked(page
));
3613 ClearPageError(page
);
3615 pg_offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
3616 if (page
->index
> end_index
||
3617 (page
->index
== end_index
&& !pg_offset
)) {
3618 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
3623 if (page
->index
== end_index
) {
3626 userpage
= kmap_atomic(page
);
3627 memset(userpage
+ pg_offset
, 0,
3628 PAGE_CACHE_SIZE
- pg_offset
);
3629 kunmap_atomic(userpage
);
3630 flush_dcache_page(page
);
3635 set_page_extent_mapped(page
);
3637 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3643 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3644 i_size
, nr_written
, write_flags
, &nr
);
3650 /* make sure the mapping tag for page dirty gets cleared */
3651 set_page_writeback(page
);
3652 end_page_writeback(page
);
3654 if (PageError(page
)) {
3655 ret
= ret
< 0 ? ret
: -EIO
;
3656 end_extent_writepage(page
, ret
, start
, page_end
);
3665 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3667 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3668 TASK_UNINTERRUPTIBLE
);
3671 static noinline_for_stack
int
3672 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3673 struct btrfs_fs_info
*fs_info
,
3674 struct extent_page_data
*epd
)
3676 unsigned long i
, num_pages
;
3680 if (!btrfs_try_tree_write_lock(eb
)) {
3682 flush_write_bio(epd
);
3683 btrfs_tree_lock(eb
);
3686 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3687 btrfs_tree_unlock(eb
);
3691 flush_write_bio(epd
);
3695 wait_on_extent_buffer_writeback(eb
);
3696 btrfs_tree_lock(eb
);
3697 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3699 btrfs_tree_unlock(eb
);
3704 * We need to do this to prevent races in people who check if the eb is
3705 * under IO since we can end up having no IO bits set for a short period
3708 spin_lock(&eb
->refs_lock
);
3709 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3710 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3711 spin_unlock(&eb
->refs_lock
);
3712 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3713 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3715 fs_info
->dirty_metadata_batch
);
3718 spin_unlock(&eb
->refs_lock
);
3721 btrfs_tree_unlock(eb
);
3726 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3727 for (i
= 0; i
< num_pages
; i
++) {
3728 struct page
*p
= eb
->pages
[i
];
3730 if (!trylock_page(p
)) {
3732 flush_write_bio(epd
);
3742 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3744 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3745 smp_mb__after_atomic();
3746 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3749 static void set_btree_ioerr(struct page
*page
)
3751 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3752 struct btrfs_inode
*btree_ino
= BTRFS_I(eb
->fs_info
->btree_inode
);
3755 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3759 * If writeback for a btree extent that doesn't belong to a log tree
3760 * failed, increment the counter transaction->eb_write_errors.
3761 * We do this because while the transaction is running and before it's
3762 * committing (when we call filemap_fdata[write|wait]_range against
3763 * the btree inode), we might have
3764 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3765 * returns an error or an error happens during writeback, when we're
3766 * committing the transaction we wouldn't know about it, since the pages
3767 * can be no longer dirty nor marked anymore for writeback (if a
3768 * subsequent modification to the extent buffer didn't happen before the
3769 * transaction commit), which makes filemap_fdata[write|wait]_range not
3770 * able to find the pages tagged with SetPageError at transaction
3771 * commit time. So if this happens we must abort the transaction,
3772 * otherwise we commit a super block with btree roots that point to
3773 * btree nodes/leafs whose content on disk is invalid - either garbage
3774 * or the content of some node/leaf from a past generation that got
3775 * cowed or deleted and is no longer valid.
3777 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3778 * not be enough - we need to distinguish between log tree extents vs
3779 * non-log tree extents, and the next filemap_fdatawait_range() call
3780 * will catch and clear such errors in the mapping - and that call might
3781 * be from a log sync and not from a transaction commit. Also, checking
3782 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3783 * not done and would not be reliable - the eb might have been released
3784 * from memory and reading it back again means that flag would not be
3785 * set (since it's a runtime flag, not persisted on disk).
3787 * Using the flags below in the btree inode also makes us achieve the
3788 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3789 * writeback for all dirty pages and before filemap_fdatawait_range()
3790 * is called, the writeback for all dirty pages had already finished
3791 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3792 * filemap_fdatawait_range() would return success, as it could not know
3793 * that writeback errors happened (the pages were no longer tagged for
3796 switch (eb
->log_index
) {
3798 set_bit(BTRFS_INODE_BTREE_ERR
, &btree_ino
->runtime_flags
);
3801 set_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
3804 set_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
3807 BUG(); /* unexpected, logic error */
3811 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3813 struct bio_vec
*bvec
;
3814 struct extent_buffer
*eb
;
3817 bio_for_each_segment_all(bvec
, bio
, i
) {
3818 struct page
*page
= bvec
->bv_page
;
3820 eb
= (struct extent_buffer
*)page
->private;
3822 done
= atomic_dec_and_test(&eb
->io_pages
);
3824 if (bio
->bi_error
||
3825 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3826 ClearPageUptodate(page
);
3827 set_btree_ioerr(page
);
3830 end_page_writeback(page
);
3835 end_extent_buffer_writeback(eb
);
3841 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3842 struct btrfs_fs_info
*fs_info
,
3843 struct writeback_control
*wbc
,
3844 struct extent_page_data
*epd
)
3846 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3847 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3848 u64 offset
= eb
->start
;
3849 unsigned long i
, num_pages
;
3850 unsigned long bio_flags
= 0;
3851 int rw
= (epd
->sync_io
? WRITE_SYNC
: WRITE
) | REQ_META
;
3854 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3855 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3856 atomic_set(&eb
->io_pages
, num_pages
);
3857 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3858 bio_flags
= EXTENT_BIO_TREE_LOG
;
3860 for (i
= 0; i
< num_pages
; i
++) {
3861 struct page
*p
= eb
->pages
[i
];
3863 clear_page_dirty_for_io(p
);
3864 set_page_writeback(p
);
3865 ret
= submit_extent_page(rw
, tree
, wbc
, p
, offset
>> 9,
3866 PAGE_CACHE_SIZE
, 0, bdev
, &epd
->bio
,
3867 -1, end_bio_extent_buffer_writepage
,
3868 0, epd
->bio_flags
, bio_flags
, false);
3869 epd
->bio_flags
= bio_flags
;
3872 end_page_writeback(p
);
3873 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3874 end_extent_buffer_writeback(eb
);
3878 offset
+= PAGE_CACHE_SIZE
;
3879 update_nr_written(p
, wbc
, 1);
3883 if (unlikely(ret
)) {
3884 for (; i
< num_pages
; i
++) {
3885 struct page
*p
= eb
->pages
[i
];
3886 clear_page_dirty_for_io(p
);
3894 int btree_write_cache_pages(struct address_space
*mapping
,
3895 struct writeback_control
*wbc
)
3897 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3898 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3899 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3900 struct extent_page_data epd
= {
3904 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3909 int nr_to_write_done
= 0;
3910 struct pagevec pvec
;
3913 pgoff_t end
; /* Inclusive */
3917 pagevec_init(&pvec
, 0);
3918 if (wbc
->range_cyclic
) {
3919 index
= mapping
->writeback_index
; /* Start from prev offset */
3922 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
3923 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
3926 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3927 tag
= PAGECACHE_TAG_TOWRITE
;
3929 tag
= PAGECACHE_TAG_DIRTY
;
3931 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3932 tag_pages_for_writeback(mapping
, index
, end
);
3933 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3934 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3935 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3939 for (i
= 0; i
< nr_pages
; i
++) {
3940 struct page
*page
= pvec
.pages
[i
];
3942 if (!PagePrivate(page
))
3945 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3950 spin_lock(&mapping
->private_lock
);
3951 if (!PagePrivate(page
)) {
3952 spin_unlock(&mapping
->private_lock
);
3956 eb
= (struct extent_buffer
*)page
->private;
3959 * Shouldn't happen and normally this would be a BUG_ON
3960 * but no sense in crashing the users box for something
3961 * we can survive anyway.
3964 spin_unlock(&mapping
->private_lock
);
3968 if (eb
== prev_eb
) {
3969 spin_unlock(&mapping
->private_lock
);
3973 ret
= atomic_inc_not_zero(&eb
->refs
);
3974 spin_unlock(&mapping
->private_lock
);
3979 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3981 free_extent_buffer(eb
);
3985 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3988 free_extent_buffer(eb
);
3991 free_extent_buffer(eb
);
3994 * the filesystem may choose to bump up nr_to_write.
3995 * We have to make sure to honor the new nr_to_write
3998 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4000 pagevec_release(&pvec
);
4003 if (!scanned
&& !done
) {
4005 * We hit the last page and there is more work to be done: wrap
4006 * back to the start of the file
4012 flush_write_bio(&epd
);
4017 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
4018 * @mapping: address space structure to write
4019 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
4020 * @writepage: function called for each page
4021 * @data: data passed to writepage function
4023 * If a page is already under I/O, write_cache_pages() skips it, even
4024 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
4025 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
4026 * and msync() need to guarantee that all the data which was dirty at the time
4027 * the call was made get new I/O started against them. If wbc->sync_mode is
4028 * WB_SYNC_ALL then we were called for data integrity and we must wait for
4029 * existing IO to complete.
4031 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
4032 struct address_space
*mapping
,
4033 struct writeback_control
*wbc
,
4034 writepage_t writepage
, void *data
,
4035 void (*flush_fn
)(void *))
4037 struct inode
*inode
= mapping
->host
;
4041 int nr_to_write_done
= 0;
4042 struct pagevec pvec
;
4045 pgoff_t end
; /* Inclusive */
4050 * We have to hold onto the inode so that ordered extents can do their
4051 * work when the IO finishes. The alternative to this is failing to add
4052 * an ordered extent if the igrab() fails there and that is a huge pain
4053 * to deal with, so instead just hold onto the inode throughout the
4054 * writepages operation. If it fails here we are freeing up the inode
4055 * anyway and we'd rather not waste our time writing out stuff that is
4056 * going to be truncated anyway.
4061 pagevec_init(&pvec
, 0);
4062 if (wbc
->range_cyclic
) {
4063 index
= mapping
->writeback_index
; /* Start from prev offset */
4066 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
4067 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
4070 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4071 tag
= PAGECACHE_TAG_TOWRITE
;
4073 tag
= PAGECACHE_TAG_DIRTY
;
4075 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4076 tag_pages_for_writeback(mapping
, index
, end
);
4077 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
4078 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
4079 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
4083 for (i
= 0; i
< nr_pages
; i
++) {
4084 struct page
*page
= pvec
.pages
[i
];
4087 * At this point we hold neither mapping->tree_lock nor
4088 * lock on the page itself: the page may be truncated or
4089 * invalidated (changing page->mapping to NULL), or even
4090 * swizzled back from swapper_space to tmpfs file
4093 if (!trylock_page(page
)) {
4098 if (unlikely(page
->mapping
!= mapping
)) {
4103 if (!wbc
->range_cyclic
&& page
->index
> end
) {
4109 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4110 if (PageWriteback(page
))
4112 wait_on_page_writeback(page
);
4115 if (PageWriteback(page
) ||
4116 !clear_page_dirty_for_io(page
)) {
4121 ret
= (*writepage
)(page
, wbc
, data
);
4123 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4127 if (!err
&& ret
< 0)
4131 * the filesystem may choose to bump up nr_to_write.
4132 * We have to make sure to honor the new nr_to_write
4135 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4137 pagevec_release(&pvec
);
4140 if (!scanned
&& !done
&& !err
) {
4142 * We hit the last page and there is more work to be done: wrap
4143 * back to the start of the file
4149 btrfs_add_delayed_iput(inode
);
4153 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4162 ret
= submit_one_bio(rw
, epd
->bio
, 0, epd
->bio_flags
);
4163 BUG_ON(ret
< 0); /* -ENOMEM */
4168 static noinline
void flush_write_bio(void *data
)
4170 struct extent_page_data
*epd
= data
;
4171 flush_epd_write_bio(epd
);
4174 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4175 get_extent_t
*get_extent
,
4176 struct writeback_control
*wbc
)
4179 struct extent_page_data epd
= {
4182 .get_extent
= get_extent
,
4184 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4188 ret
= __extent_writepage(page
, wbc
, &epd
);
4190 flush_epd_write_bio(&epd
);
4194 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4195 u64 start
, u64 end
, get_extent_t
*get_extent
,
4199 struct address_space
*mapping
= inode
->i_mapping
;
4201 unsigned long nr_pages
= (end
- start
+ PAGE_CACHE_SIZE
) >>
4204 struct extent_page_data epd
= {
4207 .get_extent
= get_extent
,
4209 .sync_io
= mode
== WB_SYNC_ALL
,
4212 struct writeback_control wbc_writepages
= {
4214 .nr_to_write
= nr_pages
* 2,
4215 .range_start
= start
,
4216 .range_end
= end
+ 1,
4219 while (start
<= end
) {
4220 page
= find_get_page(mapping
, start
>> PAGE_CACHE_SHIFT
);
4221 if (clear_page_dirty_for_io(page
))
4222 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4224 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4225 tree
->ops
->writepage_end_io_hook(page
, start
,
4226 start
+ PAGE_CACHE_SIZE
- 1,
4230 page_cache_release(page
);
4231 start
+= PAGE_CACHE_SIZE
;
4234 flush_epd_write_bio(&epd
);
4238 int extent_writepages(struct extent_io_tree
*tree
,
4239 struct address_space
*mapping
,
4240 get_extent_t
*get_extent
,
4241 struct writeback_control
*wbc
)
4244 struct extent_page_data epd
= {
4247 .get_extent
= get_extent
,
4249 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4253 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
4254 __extent_writepage
, &epd
,
4256 flush_epd_write_bio(&epd
);
4260 int extent_readpages(struct extent_io_tree
*tree
,
4261 struct address_space
*mapping
,
4262 struct list_head
*pages
, unsigned nr_pages
,
4263 get_extent_t get_extent
)
4265 struct bio
*bio
= NULL
;
4267 unsigned long bio_flags
= 0;
4268 struct page
*pagepool
[16];
4270 struct extent_map
*em_cached
= NULL
;
4272 u64 prev_em_start
= (u64
)-1;
4274 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4275 page
= list_entry(pages
->prev
, struct page
, lru
);
4277 prefetchw(&page
->flags
);
4278 list_del(&page
->lru
);
4279 if (add_to_page_cache_lru(page
, mapping
,
4280 page
->index
, GFP_NOFS
)) {
4281 page_cache_release(page
);
4285 pagepool
[nr
++] = page
;
4286 if (nr
< ARRAY_SIZE(pagepool
))
4288 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4289 &bio
, 0, &bio_flags
, READ
, &prev_em_start
);
4293 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4294 &bio
, 0, &bio_flags
, READ
, &prev_em_start
);
4297 free_extent_map(em_cached
);
4299 BUG_ON(!list_empty(pages
));
4301 return submit_one_bio(READ
, bio
, 0, bio_flags
);
4306 * basic invalidatepage code, this waits on any locked or writeback
4307 * ranges corresponding to the page, and then deletes any extent state
4308 * records from the tree
4310 int extent_invalidatepage(struct extent_io_tree
*tree
,
4311 struct page
*page
, unsigned long offset
)
4313 struct extent_state
*cached_state
= NULL
;
4314 u64 start
= page_offset(page
);
4315 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4316 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4318 start
+= ALIGN(offset
, blocksize
);
4322 lock_extent_bits(tree
, start
, end
, 0, &cached_state
);
4323 wait_on_page_writeback(page
);
4324 clear_extent_bit(tree
, start
, end
,
4325 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4326 EXTENT_DO_ACCOUNTING
,
4327 1, 1, &cached_state
, GFP_NOFS
);
4332 * a helper for releasepage, this tests for areas of the page that
4333 * are locked or under IO and drops the related state bits if it is safe
4336 static int try_release_extent_state(struct extent_map_tree
*map
,
4337 struct extent_io_tree
*tree
,
4338 struct page
*page
, gfp_t mask
)
4340 u64 start
= page_offset(page
);
4341 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4344 if (test_range_bit(tree
, start
, end
,
4345 EXTENT_IOBITS
, 0, NULL
))
4348 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4351 * at this point we can safely clear everything except the
4352 * locked bit and the nodatasum bit
4354 ret
= clear_extent_bit(tree
, start
, end
,
4355 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4358 /* if clear_extent_bit failed for enomem reasons,
4359 * we can't allow the release to continue.
4370 * a helper for releasepage. As long as there are no locked extents
4371 * in the range corresponding to the page, both state records and extent
4372 * map records are removed
4374 int try_release_extent_mapping(struct extent_map_tree
*map
,
4375 struct extent_io_tree
*tree
, struct page
*page
,
4378 struct extent_map
*em
;
4379 u64 start
= page_offset(page
);
4380 u64 end
= start
+ PAGE_CACHE_SIZE
- 1;
4382 if (gfpflags_allow_blocking(mask
) &&
4383 page
->mapping
->host
->i_size
> 16 * 1024 * 1024) {
4385 while (start
<= end
) {
4386 len
= end
- start
+ 1;
4387 write_lock(&map
->lock
);
4388 em
= lookup_extent_mapping(map
, start
, len
);
4390 write_unlock(&map
->lock
);
4393 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4394 em
->start
!= start
) {
4395 write_unlock(&map
->lock
);
4396 free_extent_map(em
);
4399 if (!test_range_bit(tree
, em
->start
,
4400 extent_map_end(em
) - 1,
4401 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4403 remove_extent_mapping(map
, em
);
4404 /* once for the rb tree */
4405 free_extent_map(em
);
4407 start
= extent_map_end(em
);
4408 write_unlock(&map
->lock
);
4411 free_extent_map(em
);
4414 return try_release_extent_state(map
, tree
, page
, mask
);
4418 * helper function for fiemap, which doesn't want to see any holes.
4419 * This maps until we find something past 'last'
4421 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4424 get_extent_t
*get_extent
)
4426 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4427 struct extent_map
*em
;
4434 len
= last
- offset
;
4437 len
= ALIGN(len
, sectorsize
);
4438 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4439 if (IS_ERR_OR_NULL(em
))
4442 /* if this isn't a hole return it */
4443 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4444 em
->block_start
!= EXTENT_MAP_HOLE
) {
4448 /* this is a hole, advance to the next extent */
4449 offset
= extent_map_end(em
);
4450 free_extent_map(em
);
4457 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4458 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4462 u64 max
= start
+ len
;
4466 u64 last_for_get_extent
= 0;
4468 u64 isize
= i_size_read(inode
);
4469 struct btrfs_key found_key
;
4470 struct extent_map
*em
= NULL
;
4471 struct extent_state
*cached_state
= NULL
;
4472 struct btrfs_path
*path
;
4473 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4482 path
= btrfs_alloc_path();
4485 path
->leave_spinning
= 1;
4487 start
= round_down(start
, BTRFS_I(inode
)->root
->sectorsize
);
4488 len
= round_up(max
, BTRFS_I(inode
)->root
->sectorsize
) - start
;
4491 * lookup the last file extent. We're not using i_size here
4492 * because there might be preallocation past i_size
4494 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
), -1,
4497 btrfs_free_path(path
);
4502 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4503 found_type
= found_key
.type
;
4505 /* No extents, but there might be delalloc bits */
4506 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4507 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4508 /* have to trust i_size as the end */
4510 last_for_get_extent
= isize
;
4513 * remember the start of the last extent. There are a
4514 * bunch of different factors that go into the length of the
4515 * extent, so its much less complex to remember where it started
4517 last
= found_key
.offset
;
4518 last_for_get_extent
= last
+ 1;
4520 btrfs_release_path(path
);
4523 * we might have some extents allocated but more delalloc past those
4524 * extents. so, we trust isize unless the start of the last extent is
4529 last_for_get_extent
= isize
;
4532 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1, 0,
4535 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4545 u64 offset_in_extent
= 0;
4547 /* break if the extent we found is outside the range */
4548 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4552 * get_extent may return an extent that starts before our
4553 * requested range. We have to make sure the ranges
4554 * we return to fiemap always move forward and don't
4555 * overlap, so adjust the offsets here
4557 em_start
= max(em
->start
, off
);
4560 * record the offset from the start of the extent
4561 * for adjusting the disk offset below. Only do this if the
4562 * extent isn't compressed since our in ram offset may be past
4563 * what we have actually allocated on disk.
4565 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4566 offset_in_extent
= em_start
- em
->start
;
4567 em_end
= extent_map_end(em
);
4568 em_len
= em_end
- em_start
;
4573 * bump off for our next call to get_extent
4575 off
= extent_map_end(em
);
4579 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4581 flags
|= FIEMAP_EXTENT_LAST
;
4582 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4583 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4584 FIEMAP_EXTENT_NOT_ALIGNED
);
4585 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4586 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4587 FIEMAP_EXTENT_UNKNOWN
);
4588 } else if (fieinfo
->fi_extents_max
) {
4589 u64 bytenr
= em
->block_start
-
4590 (em
->start
- em
->orig_start
);
4592 disko
= em
->block_start
+ offset_in_extent
;
4595 * As btrfs supports shared space, this information
4596 * can be exported to userspace tools via
4597 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4598 * then we're just getting a count and we can skip the
4601 ret
= btrfs_check_shared(NULL
, root
->fs_info
,
4603 btrfs_ino(inode
), bytenr
);
4607 flags
|= FIEMAP_EXTENT_SHARED
;
4610 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4611 flags
|= FIEMAP_EXTENT_ENCODED
;
4612 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4613 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4615 free_extent_map(em
);
4617 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4618 (last
== (u64
)-1 && isize
<= em_end
)) {
4619 flags
|= FIEMAP_EXTENT_LAST
;
4623 /* now scan forward to see if this is really the last extent. */
4624 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4631 flags
|= FIEMAP_EXTENT_LAST
;
4634 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4643 free_extent_map(em
);
4645 btrfs_free_path(path
);
4646 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4647 &cached_state
, GFP_NOFS
);
4651 static void __free_extent_buffer(struct extent_buffer
*eb
)
4653 btrfs_leak_debug_del(&eb
->leak_list
);
4654 kmem_cache_free(extent_buffer_cache
, eb
);
4657 int extent_buffer_under_io(struct extent_buffer
*eb
)
4659 return (atomic_read(&eb
->io_pages
) ||
4660 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4661 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4665 * Helper for releasing extent buffer page.
4667 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4669 unsigned long index
;
4671 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4673 BUG_ON(extent_buffer_under_io(eb
));
4675 index
= num_extent_pages(eb
->start
, eb
->len
);
4681 page
= eb
->pages
[index
];
4685 spin_lock(&page
->mapping
->private_lock
);
4687 * We do this since we'll remove the pages after we've
4688 * removed the eb from the radix tree, so we could race
4689 * and have this page now attached to the new eb. So
4690 * only clear page_private if it's still connected to
4693 if (PagePrivate(page
) &&
4694 page
->private == (unsigned long)eb
) {
4695 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4696 BUG_ON(PageDirty(page
));
4697 BUG_ON(PageWriteback(page
));
4699 * We need to make sure we haven't be attached
4702 ClearPagePrivate(page
);
4703 set_page_private(page
, 0);
4704 /* One for the page private */
4705 page_cache_release(page
);
4709 spin_unlock(&page
->mapping
->private_lock
);
4711 /* One for when we alloced the page */
4712 page_cache_release(page
);
4713 } while (index
!= 0);
4717 * Helper for releasing the extent buffer.
4719 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4721 btrfs_release_extent_buffer_page(eb
);
4722 __free_extent_buffer(eb
);
4725 static struct extent_buffer
*
4726 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4729 struct extent_buffer
*eb
= NULL
;
4731 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4734 eb
->fs_info
= fs_info
;
4736 rwlock_init(&eb
->lock
);
4737 atomic_set(&eb
->write_locks
, 0);
4738 atomic_set(&eb
->read_locks
, 0);
4739 atomic_set(&eb
->blocking_readers
, 0);
4740 atomic_set(&eb
->blocking_writers
, 0);
4741 atomic_set(&eb
->spinning_readers
, 0);
4742 atomic_set(&eb
->spinning_writers
, 0);
4743 eb
->lock_nested
= 0;
4744 init_waitqueue_head(&eb
->write_lock_wq
);
4745 init_waitqueue_head(&eb
->read_lock_wq
);
4747 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4749 spin_lock_init(&eb
->refs_lock
);
4750 atomic_set(&eb
->refs
, 1);
4751 atomic_set(&eb
->io_pages
, 0);
4754 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4756 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4757 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4758 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4763 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4767 struct extent_buffer
*new;
4768 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4770 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4774 for (i
= 0; i
< num_pages
; i
++) {
4775 p
= alloc_page(GFP_NOFS
);
4777 btrfs_release_extent_buffer(new);
4780 attach_extent_buffer_page(new, p
);
4781 WARN_ON(PageDirty(p
));
4786 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4787 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4788 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4793 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4796 struct extent_buffer
*eb
;
4798 unsigned long num_pages
;
4803 * Called only from tests that don't always have a fs_info
4804 * available, but we know that nodesize is 4096
4808 len
= fs_info
->tree_root
->nodesize
;
4810 num_pages
= num_extent_pages(0, len
);
4812 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4816 for (i
= 0; i
< num_pages
; i
++) {
4817 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4821 set_extent_buffer_uptodate(eb
);
4822 btrfs_set_header_nritems(eb
, 0);
4823 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4828 __free_page(eb
->pages
[i
- 1]);
4829 __free_extent_buffer(eb
);
4833 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4836 /* the ref bit is tricky. We have to make sure it is set
4837 * if we have the buffer dirty. Otherwise the
4838 * code to free a buffer can end up dropping a dirty
4841 * Once the ref bit is set, it won't go away while the
4842 * buffer is dirty or in writeback, and it also won't
4843 * go away while we have the reference count on the
4846 * We can't just set the ref bit without bumping the
4847 * ref on the eb because free_extent_buffer might
4848 * see the ref bit and try to clear it. If this happens
4849 * free_extent_buffer might end up dropping our original
4850 * ref by mistake and freeing the page before we are able
4851 * to add one more ref.
4853 * So bump the ref count first, then set the bit. If someone
4854 * beat us to it, drop the ref we added.
4856 refs
= atomic_read(&eb
->refs
);
4857 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4860 spin_lock(&eb
->refs_lock
);
4861 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4862 atomic_inc(&eb
->refs
);
4863 spin_unlock(&eb
->refs_lock
);
4866 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4867 struct page
*accessed
)
4869 unsigned long num_pages
, i
;
4871 check_buffer_tree_ref(eb
);
4873 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4874 for (i
= 0; i
< num_pages
; i
++) {
4875 struct page
*p
= eb
->pages
[i
];
4878 mark_page_accessed(p
);
4882 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4885 struct extent_buffer
*eb
;
4888 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4889 start
>> PAGE_CACHE_SHIFT
);
4890 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4893 * Lock our eb's refs_lock to avoid races with
4894 * free_extent_buffer. When we get our eb it might be flagged
4895 * with EXTENT_BUFFER_STALE and another task running
4896 * free_extent_buffer might have seen that flag set,
4897 * eb->refs == 2, that the buffer isn't under IO (dirty and
4898 * writeback flags not set) and it's still in the tree (flag
4899 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4900 * of decrementing the extent buffer's reference count twice.
4901 * So here we could race and increment the eb's reference count,
4902 * clear its stale flag, mark it as dirty and drop our reference
4903 * before the other task finishes executing free_extent_buffer,
4904 * which would later result in an attempt to free an extent
4905 * buffer that is dirty.
4907 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4908 spin_lock(&eb
->refs_lock
);
4909 spin_unlock(&eb
->refs_lock
);
4911 mark_extent_buffer_accessed(eb
, NULL
);
4919 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4920 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4923 struct extent_buffer
*eb
, *exists
= NULL
;
4926 eb
= find_extent_buffer(fs_info
, start
);
4929 eb
= alloc_dummy_extent_buffer(fs_info
, start
);
4932 eb
->fs_info
= fs_info
;
4934 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
4937 spin_lock(&fs_info
->buffer_lock
);
4938 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4939 start
>> PAGE_CACHE_SHIFT
, eb
);
4940 spin_unlock(&fs_info
->buffer_lock
);
4941 radix_tree_preload_end();
4942 if (ret
== -EEXIST
) {
4943 exists
= find_extent_buffer(fs_info
, start
);
4949 check_buffer_tree_ref(eb
);
4950 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4953 * We will free dummy extent buffer's if they come into
4954 * free_extent_buffer with a ref count of 2, but if we are using this we
4955 * want the buffers to stay in memory until we're done with them, so
4956 * bump the ref count again.
4958 atomic_inc(&eb
->refs
);
4961 btrfs_release_extent_buffer(eb
);
4966 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4969 unsigned long len
= fs_info
->tree_root
->nodesize
;
4970 unsigned long num_pages
= num_extent_pages(start
, len
);
4972 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
4973 struct extent_buffer
*eb
;
4974 struct extent_buffer
*exists
= NULL
;
4976 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4980 eb
= find_extent_buffer(fs_info
, start
);
4984 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4988 for (i
= 0; i
< num_pages
; i
++, index
++) {
4989 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4993 spin_lock(&mapping
->private_lock
);
4994 if (PagePrivate(p
)) {
4996 * We could have already allocated an eb for this page
4997 * and attached one so lets see if we can get a ref on
4998 * the existing eb, and if we can we know it's good and
4999 * we can just return that one, else we know we can just
5000 * overwrite page->private.
5002 exists
= (struct extent_buffer
*)p
->private;
5003 if (atomic_inc_not_zero(&exists
->refs
)) {
5004 spin_unlock(&mapping
->private_lock
);
5006 page_cache_release(p
);
5007 mark_extent_buffer_accessed(exists
, p
);
5013 * Do this so attach doesn't complain and we need to
5014 * drop the ref the old guy had.
5016 ClearPagePrivate(p
);
5017 WARN_ON(PageDirty(p
));
5018 page_cache_release(p
);
5020 attach_extent_buffer_page(eb
, p
);
5021 spin_unlock(&mapping
->private_lock
);
5022 WARN_ON(PageDirty(p
));
5024 if (!PageUptodate(p
))
5028 * see below about how we avoid a nasty race with release page
5029 * and why we unlock later
5033 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5035 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
5039 spin_lock(&fs_info
->buffer_lock
);
5040 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
5041 start
>> PAGE_CACHE_SHIFT
, eb
);
5042 spin_unlock(&fs_info
->buffer_lock
);
5043 radix_tree_preload_end();
5044 if (ret
== -EEXIST
) {
5045 exists
= find_extent_buffer(fs_info
, start
);
5051 /* add one reference for the tree */
5052 check_buffer_tree_ref(eb
);
5053 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
5056 * there is a race where release page may have
5057 * tried to find this extent buffer in the radix
5058 * but failed. It will tell the VM it is safe to
5059 * reclaim the, and it will clear the page private bit.
5060 * We must make sure to set the page private bit properly
5061 * after the extent buffer is in the radix tree so
5062 * it doesn't get lost
5064 SetPageChecked(eb
->pages
[0]);
5065 for (i
= 1; i
< num_pages
; i
++) {
5067 ClearPageChecked(p
);
5070 unlock_page(eb
->pages
[0]);
5074 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
5075 for (i
= 0; i
< num_pages
; i
++) {
5077 unlock_page(eb
->pages
[i
]);
5080 btrfs_release_extent_buffer(eb
);
5084 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
5086 struct extent_buffer
*eb
=
5087 container_of(head
, struct extent_buffer
, rcu_head
);
5089 __free_extent_buffer(eb
);
5092 /* Expects to have eb->eb_lock already held */
5093 static int release_extent_buffer(struct extent_buffer
*eb
)
5095 WARN_ON(atomic_read(&eb
->refs
) == 0);
5096 if (atomic_dec_and_test(&eb
->refs
)) {
5097 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5098 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5100 spin_unlock(&eb
->refs_lock
);
5102 spin_lock(&fs_info
->buffer_lock
);
5103 radix_tree_delete(&fs_info
->buffer_radix
,
5104 eb
->start
>> PAGE_CACHE_SHIFT
);
5105 spin_unlock(&fs_info
->buffer_lock
);
5107 spin_unlock(&eb
->refs_lock
);
5110 /* Should be safe to release our pages at this point */
5111 btrfs_release_extent_buffer_page(eb
);
5112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5113 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5114 __free_extent_buffer(eb
);
5118 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5121 spin_unlock(&eb
->refs_lock
);
5126 void free_extent_buffer(struct extent_buffer
*eb
)
5134 refs
= atomic_read(&eb
->refs
);
5137 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5142 spin_lock(&eb
->refs_lock
);
5143 if (atomic_read(&eb
->refs
) == 2 &&
5144 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5145 atomic_dec(&eb
->refs
);
5147 if (atomic_read(&eb
->refs
) == 2 &&
5148 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5149 !extent_buffer_under_io(eb
) &&
5150 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5151 atomic_dec(&eb
->refs
);
5154 * I know this is terrible, but it's temporary until we stop tracking
5155 * the uptodate bits and such for the extent buffers.
5157 release_extent_buffer(eb
);
5160 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5165 spin_lock(&eb
->refs_lock
);
5166 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5168 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5169 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5170 atomic_dec(&eb
->refs
);
5171 release_extent_buffer(eb
);
5174 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5177 unsigned long num_pages
;
5180 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5182 for (i
= 0; i
< num_pages
; i
++) {
5183 page
= eb
->pages
[i
];
5184 if (!PageDirty(page
))
5188 WARN_ON(!PagePrivate(page
));
5190 clear_page_dirty_for_io(page
);
5191 spin_lock_irq(&page
->mapping
->tree_lock
);
5192 if (!PageDirty(page
)) {
5193 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5195 PAGECACHE_TAG_DIRTY
);
5197 spin_unlock_irq(&page
->mapping
->tree_lock
);
5198 ClearPageError(page
);
5201 WARN_ON(atomic_read(&eb
->refs
) == 0);
5204 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5207 unsigned long num_pages
;
5210 check_buffer_tree_ref(eb
);
5212 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5214 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5215 WARN_ON(atomic_read(&eb
->refs
) == 0);
5216 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5218 for (i
= 0; i
< num_pages
; i
++)
5219 set_page_dirty(eb
->pages
[i
]);
5223 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5227 unsigned long num_pages
;
5229 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5230 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5231 for (i
= 0; i
< num_pages
; i
++) {
5232 page
= eb
->pages
[i
];
5234 ClearPageUptodate(page
);
5238 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5242 unsigned long num_pages
;
5244 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5245 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5246 for (i
= 0; i
< num_pages
; i
++) {
5247 page
= eb
->pages
[i
];
5248 SetPageUptodate(page
);
5252 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5254 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5257 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5258 struct extent_buffer
*eb
, u64 start
, int wait
,
5259 get_extent_t
*get_extent
, int mirror_num
)
5262 unsigned long start_i
;
5266 int locked_pages
= 0;
5267 int all_uptodate
= 1;
5268 unsigned long num_pages
;
5269 unsigned long num_reads
= 0;
5270 struct bio
*bio
= NULL
;
5271 unsigned long bio_flags
= 0;
5273 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5277 WARN_ON(start
< eb
->start
);
5278 start_i
= (start
>> PAGE_CACHE_SHIFT
) -
5279 (eb
->start
>> PAGE_CACHE_SHIFT
);
5284 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5285 for (i
= start_i
; i
< num_pages
; i
++) {
5286 page
= eb
->pages
[i
];
5287 if (wait
== WAIT_NONE
) {
5288 if (!trylock_page(page
))
5294 if (!PageUptodate(page
)) {
5301 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5305 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5306 eb
->read_mirror
= 0;
5307 atomic_set(&eb
->io_pages
, num_reads
);
5308 for (i
= start_i
; i
< num_pages
; i
++) {
5309 page
= eb
->pages
[i
];
5310 if (!PageUptodate(page
)) {
5311 ClearPageError(page
);
5312 err
= __extent_read_full_page(tree
, page
,
5314 mirror_num
, &bio_flags
,
5324 err
= submit_one_bio(READ
| REQ_META
, bio
, mirror_num
,
5330 if (ret
|| wait
!= WAIT_COMPLETE
)
5333 for (i
= start_i
; i
< num_pages
; i
++) {
5334 page
= eb
->pages
[i
];
5335 wait_on_page_locked(page
);
5336 if (!PageUptodate(page
))
5344 while (locked_pages
> 0) {
5345 page
= eb
->pages
[i
];
5353 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5354 unsigned long start
,
5361 char *dst
= (char *)dstv
;
5362 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5363 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5365 WARN_ON(start
> eb
->len
);
5366 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5368 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5371 page
= eb
->pages
[i
];
5373 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5374 kaddr
= page_address(page
);
5375 memcpy(dst
, kaddr
+ offset
, cur
);
5384 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5385 unsigned long start
,
5392 char __user
*dst
= (char __user
*)dstv
;
5393 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5394 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5397 WARN_ON(start
> eb
->len
);
5398 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5400 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5403 page
= eb
->pages
[i
];
5405 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5406 kaddr
= page_address(page
);
5407 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5421 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5422 unsigned long min_len
, char **map
,
5423 unsigned long *map_start
,
5424 unsigned long *map_len
)
5426 size_t offset
= start
& (PAGE_CACHE_SIZE
- 1);
5429 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5430 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5431 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5438 offset
= start_offset
;
5442 *map_start
= ((u64
)i
<< PAGE_CACHE_SHIFT
) - start_offset
;
5445 if (start
+ min_len
> eb
->len
) {
5446 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
5448 eb
->start
, eb
->len
, start
, min_len
);
5453 kaddr
= page_address(p
);
5454 *map
= kaddr
+ offset
;
5455 *map_len
= PAGE_CACHE_SIZE
- offset
;
5459 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5460 unsigned long start
,
5467 char *ptr
= (char *)ptrv
;
5468 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5469 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5472 WARN_ON(start
> eb
->len
);
5473 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5475 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5478 page
= eb
->pages
[i
];
5480 cur
= min(len
, (PAGE_CACHE_SIZE
- offset
));
5482 kaddr
= page_address(page
);
5483 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5495 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5496 unsigned long start
, unsigned long len
)
5502 char *src
= (char *)srcv
;
5503 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5504 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5506 WARN_ON(start
> eb
->len
);
5507 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5509 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5512 page
= eb
->pages
[i
];
5513 WARN_ON(!PageUptodate(page
));
5515 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5516 kaddr
= page_address(page
);
5517 memcpy(kaddr
+ offset
, src
, cur
);
5526 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5527 unsigned long start
, unsigned long len
)
5533 size_t start_offset
= eb
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5534 unsigned long i
= (start_offset
+ start
) >> PAGE_CACHE_SHIFT
;
5536 WARN_ON(start
> eb
->len
);
5537 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5539 offset
= (start_offset
+ start
) & (PAGE_CACHE_SIZE
- 1);
5542 page
= eb
->pages
[i
];
5543 WARN_ON(!PageUptodate(page
));
5545 cur
= min(len
, PAGE_CACHE_SIZE
- offset
);
5546 kaddr
= page_address(page
);
5547 memset(kaddr
+ offset
, c
, cur
);
5555 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5556 unsigned long dst_offset
, unsigned long src_offset
,
5559 u64 dst_len
= dst
->len
;
5564 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5565 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5567 WARN_ON(src
->len
!= dst_len
);
5569 offset
= (start_offset
+ dst_offset
) &
5570 (PAGE_CACHE_SIZE
- 1);
5573 page
= dst
->pages
[i
];
5574 WARN_ON(!PageUptodate(page
));
5576 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
- offset
));
5578 kaddr
= page_address(page
);
5579 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5588 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5590 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5591 return distance
< len
;
5594 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5595 unsigned long dst_off
, unsigned long src_off
,
5598 char *dst_kaddr
= page_address(dst_page
);
5600 int must_memmove
= 0;
5602 if (dst_page
!= src_page
) {
5603 src_kaddr
= page_address(src_page
);
5605 src_kaddr
= dst_kaddr
;
5606 if (areas_overlap(src_off
, dst_off
, len
))
5611 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5613 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5616 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5617 unsigned long src_offset
, unsigned long len
)
5620 size_t dst_off_in_page
;
5621 size_t src_off_in_page
;
5622 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5623 unsigned long dst_i
;
5624 unsigned long src_i
;
5626 if (src_offset
+ len
> dst
->len
) {
5627 btrfs_err(dst
->fs_info
,
5628 "memmove bogus src_offset %lu move "
5629 "len %lu dst len %lu", src_offset
, len
, dst
->len
);
5632 if (dst_offset
+ len
> dst
->len
) {
5633 btrfs_err(dst
->fs_info
,
5634 "memmove bogus dst_offset %lu move "
5635 "len %lu dst len %lu", dst_offset
, len
, dst
->len
);
5640 dst_off_in_page
= (start_offset
+ dst_offset
) &
5641 (PAGE_CACHE_SIZE
- 1);
5642 src_off_in_page
= (start_offset
+ src_offset
) &
5643 (PAGE_CACHE_SIZE
- 1);
5645 dst_i
= (start_offset
+ dst_offset
) >> PAGE_CACHE_SHIFT
;
5646 src_i
= (start_offset
+ src_offset
) >> PAGE_CACHE_SHIFT
;
5648 cur
= min(len
, (unsigned long)(PAGE_CACHE_SIZE
-
5650 cur
= min_t(unsigned long, cur
,
5651 (unsigned long)(PAGE_CACHE_SIZE
- dst_off_in_page
));
5653 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5654 dst_off_in_page
, src_off_in_page
, cur
);
5662 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5663 unsigned long src_offset
, unsigned long len
)
5666 size_t dst_off_in_page
;
5667 size_t src_off_in_page
;
5668 unsigned long dst_end
= dst_offset
+ len
- 1;
5669 unsigned long src_end
= src_offset
+ len
- 1;
5670 size_t start_offset
= dst
->start
& ((u64
)PAGE_CACHE_SIZE
- 1);
5671 unsigned long dst_i
;
5672 unsigned long src_i
;
5674 if (src_offset
+ len
> dst
->len
) {
5675 btrfs_err(dst
->fs_info
, "memmove bogus src_offset %lu move "
5676 "len %lu len %lu", src_offset
, len
, dst
->len
);
5679 if (dst_offset
+ len
> dst
->len
) {
5680 btrfs_err(dst
->fs_info
, "memmove bogus dst_offset %lu move "
5681 "len %lu len %lu", dst_offset
, len
, dst
->len
);
5684 if (dst_offset
< src_offset
) {
5685 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5689 dst_i
= (start_offset
+ dst_end
) >> PAGE_CACHE_SHIFT
;
5690 src_i
= (start_offset
+ src_end
) >> PAGE_CACHE_SHIFT
;
5692 dst_off_in_page
= (start_offset
+ dst_end
) &
5693 (PAGE_CACHE_SIZE
- 1);
5694 src_off_in_page
= (start_offset
+ src_end
) &
5695 (PAGE_CACHE_SIZE
- 1);
5697 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5698 cur
= min(cur
, dst_off_in_page
+ 1);
5699 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5700 dst_off_in_page
- cur
+ 1,
5701 src_off_in_page
- cur
+ 1, cur
);
5709 int try_release_extent_buffer(struct page
*page
)
5711 struct extent_buffer
*eb
;
5714 * We need to make sure noboody is attaching this page to an eb right
5717 spin_lock(&page
->mapping
->private_lock
);
5718 if (!PagePrivate(page
)) {
5719 spin_unlock(&page
->mapping
->private_lock
);
5723 eb
= (struct extent_buffer
*)page
->private;
5727 * This is a little awful but should be ok, we need to make sure that
5728 * the eb doesn't disappear out from under us while we're looking at
5731 spin_lock(&eb
->refs_lock
);
5732 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5733 spin_unlock(&eb
->refs_lock
);
5734 spin_unlock(&page
->mapping
->private_lock
);
5737 spin_unlock(&page
->mapping
->private_lock
);
5740 * If tree ref isn't set then we know the ref on this eb is a real ref,
5741 * so just return, this page will likely be freed soon anyway.
5743 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5744 spin_unlock(&eb
->refs_lock
);
5748 return release_extent_buffer(eb
);