1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
23 #include "transaction.h"
25 static struct kmem_cache
*extent_state_cache
;
26 static struct kmem_cache
*extent_buffer_cache
;
27 static struct bio_set
*btrfs_bioset
;
29 static inline bool extent_state_in_tree(const struct extent_state
*state
)
31 return !RB_EMPTY_NODE(&state
->rb_node
);
34 #ifdef CONFIG_BTRFS_DEBUG
35 static LIST_HEAD(buffers
);
36 static LIST_HEAD(states
);
38 static DEFINE_SPINLOCK(leak_lock
);
41 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
45 spin_lock_irqsave(&leak_lock
, flags
);
47 spin_unlock_irqrestore(&leak_lock
, flags
);
51 void btrfs_leak_debug_del(struct list_head
*entry
)
55 spin_lock_irqsave(&leak_lock
, flags
);
57 spin_unlock_irqrestore(&leak_lock
, flags
);
61 void btrfs_leak_debug_check(void)
63 struct extent_state
*state
;
64 struct extent_buffer
*eb
;
66 while (!list_empty(&states
)) {
67 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
68 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
69 state
->start
, state
->end
, state
->state
,
70 extent_state_in_tree(state
),
71 refcount_read(&state
->refs
));
72 list_del(&state
->leak_list
);
73 kmem_cache_free(extent_state_cache
, state
);
76 while (!list_empty(&buffers
)) {
77 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
78 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
79 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
80 list_del(&eb
->leak_list
);
81 kmem_cache_free(extent_buffer_cache
, eb
);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
88 struct extent_io_tree
*tree
, u64 start
, u64 end
)
96 inode
= tree
->mapping
->host
;
97 isize
= i_size_read(inode
);
98 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
99 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
100 "%s: ino %llu isize %llu odd range [%llu,%llu]",
101 caller
, btrfs_ino(BTRFS_I(inode
)), isize
, start
, end
);
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry) do {} while (0)
107 #define btrfs_leak_debug_check() do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
111 #define BUFFER_LRU_MAX 64
116 struct rb_node rb_node
;
119 struct extent_page_data
{
121 struct extent_io_tree
*tree
;
122 get_extent_t
*get_extent
;
123 unsigned long bio_flags
;
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
128 unsigned int extent_locked
:1;
130 /* tells the submit_bio code to use REQ_SYNC */
131 unsigned int sync_io
:1;
134 static void add_extent_changeset(struct extent_state
*state
, unsigned bits
,
135 struct extent_changeset
*changeset
,
142 if (set
&& (state
->state
& bits
) == bits
)
144 if (!set
&& (state
->state
& bits
) == 0)
146 changeset
->bytes_changed
+= state
->end
- state
->start
+ 1;
147 ret
= ulist_add(&changeset
->range_changed
, state
->start
, state
->end
,
153 static noinline
void flush_write_bio(void *data
);
154 static inline struct btrfs_fs_info
*
155 tree_fs_info(struct extent_io_tree
*tree
)
159 return btrfs_sb(tree
->mapping
->host
->i_sb
);
162 int __init
extent_io_init(void)
164 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
165 sizeof(struct extent_state
), 0,
166 SLAB_MEM_SPREAD
, NULL
);
167 if (!extent_state_cache
)
170 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
171 sizeof(struct extent_buffer
), 0,
172 SLAB_MEM_SPREAD
, NULL
);
173 if (!extent_buffer_cache
)
174 goto free_state_cache
;
176 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
177 offsetof(struct btrfs_io_bio
, bio
));
179 goto free_buffer_cache
;
181 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
187 bioset_free(btrfs_bioset
);
191 kmem_cache_destroy(extent_buffer_cache
);
192 extent_buffer_cache
= NULL
;
195 kmem_cache_destroy(extent_state_cache
);
196 extent_state_cache
= NULL
;
200 void extent_io_exit(void)
202 btrfs_leak_debug_check();
205 * Make sure all delayed rcu free are flushed before we
209 kmem_cache_destroy(extent_state_cache
);
210 kmem_cache_destroy(extent_buffer_cache
);
212 bioset_free(btrfs_bioset
);
215 void extent_io_tree_init(struct extent_io_tree
*tree
,
216 struct address_space
*mapping
)
218 tree
->state
= RB_ROOT
;
220 tree
->dirty_bytes
= 0;
221 spin_lock_init(&tree
->lock
);
222 tree
->mapping
= mapping
;
225 static struct extent_state
*alloc_extent_state(gfp_t mask
)
227 struct extent_state
*state
;
230 * The given mask might be not appropriate for the slab allocator,
231 * drop the unsupported bits
233 mask
&= ~(__GFP_DMA32
|__GFP_HIGHMEM
);
234 state
= kmem_cache_alloc(extent_state_cache
, mask
);
238 state
->failrec
= NULL
;
239 RB_CLEAR_NODE(&state
->rb_node
);
240 btrfs_leak_debug_add(&state
->leak_list
, &states
);
241 refcount_set(&state
->refs
, 1);
242 init_waitqueue_head(&state
->wq
);
243 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
247 void free_extent_state(struct extent_state
*state
)
251 if (refcount_dec_and_test(&state
->refs
)) {
252 WARN_ON(extent_state_in_tree(state
));
253 btrfs_leak_debug_del(&state
->leak_list
);
254 trace_free_extent_state(state
, _RET_IP_
);
255 kmem_cache_free(extent_state_cache
, state
);
259 static struct rb_node
*tree_insert(struct rb_root
*root
,
260 struct rb_node
*search_start
,
262 struct rb_node
*node
,
263 struct rb_node
***p_in
,
264 struct rb_node
**parent_in
)
267 struct rb_node
*parent
= NULL
;
268 struct tree_entry
*entry
;
270 if (p_in
&& parent_in
) {
276 p
= search_start
? &search_start
: &root
->rb_node
;
279 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
281 if (offset
< entry
->start
)
283 else if (offset
> entry
->end
)
290 rb_link_node(node
, parent
, p
);
291 rb_insert_color(node
, root
);
295 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
296 struct rb_node
**prev_ret
,
297 struct rb_node
**next_ret
,
298 struct rb_node
***p_ret
,
299 struct rb_node
**parent_ret
)
301 struct rb_root
*root
= &tree
->state
;
302 struct rb_node
**n
= &root
->rb_node
;
303 struct rb_node
*prev
= NULL
;
304 struct rb_node
*orig_prev
= NULL
;
305 struct tree_entry
*entry
;
306 struct tree_entry
*prev_entry
= NULL
;
310 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
313 if (offset
< entry
->start
)
315 else if (offset
> entry
->end
)
328 while (prev
&& offset
> prev_entry
->end
) {
329 prev
= rb_next(prev
);
330 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
337 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
338 while (prev
&& offset
< prev_entry
->start
) {
339 prev
= rb_prev(prev
);
340 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
347 static inline struct rb_node
*
348 tree_search_for_insert(struct extent_io_tree
*tree
,
350 struct rb_node
***p_ret
,
351 struct rb_node
**parent_ret
)
353 struct rb_node
*prev
= NULL
;
356 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
362 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
365 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
368 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
369 struct extent_state
*other
)
371 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
372 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
377 * utility function to look for merge candidates inside a given range.
378 * Any extents with matching state are merged together into a single
379 * extent in the tree. Extents with EXTENT_IO in their state field
380 * are not merged because the end_io handlers need to be able to do
381 * operations on them without sleeping (or doing allocations/splits).
383 * This should be called with the tree lock held.
385 static void merge_state(struct extent_io_tree
*tree
,
386 struct extent_state
*state
)
388 struct extent_state
*other
;
389 struct rb_node
*other_node
;
391 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
394 other_node
= rb_prev(&state
->rb_node
);
396 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
397 if (other
->end
== state
->start
- 1 &&
398 other
->state
== state
->state
) {
399 merge_cb(tree
, state
, other
);
400 state
->start
= other
->start
;
401 rb_erase(&other
->rb_node
, &tree
->state
);
402 RB_CLEAR_NODE(&other
->rb_node
);
403 free_extent_state(other
);
406 other_node
= rb_next(&state
->rb_node
);
408 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
409 if (other
->start
== state
->end
+ 1 &&
410 other
->state
== state
->state
) {
411 merge_cb(tree
, state
, other
);
412 state
->end
= other
->end
;
413 rb_erase(&other
->rb_node
, &tree
->state
);
414 RB_CLEAR_NODE(&other
->rb_node
);
415 free_extent_state(other
);
420 static void set_state_cb(struct extent_io_tree
*tree
,
421 struct extent_state
*state
, unsigned *bits
)
423 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
424 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
427 static void clear_state_cb(struct extent_io_tree
*tree
,
428 struct extent_state
*state
, unsigned *bits
)
430 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
431 tree
->ops
->clear_bit_hook(BTRFS_I(tree
->mapping
->host
),
435 static void set_state_bits(struct extent_io_tree
*tree
,
436 struct extent_state
*state
, unsigned *bits
,
437 struct extent_changeset
*changeset
);
440 * insert an extent_state struct into the tree. 'bits' are set on the
441 * struct before it is inserted.
443 * This may return -EEXIST if the extent is already there, in which case the
444 * state struct is freed.
446 * The tree lock is not taken internally. This is a utility function and
447 * probably isn't what you want to call (see set/clear_extent_bit).
449 static int insert_state(struct extent_io_tree
*tree
,
450 struct extent_state
*state
, u64 start
, u64 end
,
452 struct rb_node
**parent
,
453 unsigned *bits
, struct extent_changeset
*changeset
)
455 struct rb_node
*node
;
458 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
460 state
->start
= start
;
463 set_state_bits(tree
, state
, bits
, changeset
);
465 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
467 struct extent_state
*found
;
468 found
= rb_entry(node
, struct extent_state
, rb_node
);
469 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
470 found
->start
, found
->end
, start
, end
);
473 merge_state(tree
, state
);
477 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
480 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
481 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
485 * split a given extent state struct in two, inserting the preallocated
486 * struct 'prealloc' as the newly created second half. 'split' indicates an
487 * offset inside 'orig' where it should be split.
490 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
491 * are two extent state structs in the tree:
492 * prealloc: [orig->start, split - 1]
493 * orig: [ split, orig->end ]
495 * The tree locks are not taken by this function. They need to be held
498 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
499 struct extent_state
*prealloc
, u64 split
)
501 struct rb_node
*node
;
503 split_cb(tree
, orig
, split
);
505 prealloc
->start
= orig
->start
;
506 prealloc
->end
= split
- 1;
507 prealloc
->state
= orig
->state
;
510 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
511 &prealloc
->rb_node
, NULL
, NULL
);
513 free_extent_state(prealloc
);
519 static struct extent_state
*next_state(struct extent_state
*state
)
521 struct rb_node
*next
= rb_next(&state
->rb_node
);
523 return rb_entry(next
, struct extent_state
, rb_node
);
529 * utility function to clear some bits in an extent state struct.
530 * it will optionally wake up any one waiting on this state (wake == 1).
532 * If no bits are set on the state struct after clearing things, the
533 * struct is freed and removed from the tree
535 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
536 struct extent_state
*state
,
537 unsigned *bits
, int wake
,
538 struct extent_changeset
*changeset
)
540 struct extent_state
*next
;
541 unsigned bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
543 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
544 u64 range
= state
->end
- state
->start
+ 1;
545 WARN_ON(range
> tree
->dirty_bytes
);
546 tree
->dirty_bytes
-= range
;
548 clear_state_cb(tree
, state
, bits
);
549 add_extent_changeset(state
, bits_to_clear
, changeset
, 0);
550 state
->state
&= ~bits_to_clear
;
553 if (state
->state
== 0) {
554 next
= next_state(state
);
555 if (extent_state_in_tree(state
)) {
556 rb_erase(&state
->rb_node
, &tree
->state
);
557 RB_CLEAR_NODE(&state
->rb_node
);
558 free_extent_state(state
);
563 merge_state(tree
, state
);
564 next
= next_state(state
);
569 static struct extent_state
*
570 alloc_extent_state_atomic(struct extent_state
*prealloc
)
573 prealloc
= alloc_extent_state(GFP_ATOMIC
);
578 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
580 btrfs_panic(tree_fs_info(tree
), err
,
581 "Locking error: Extent tree was modified by another thread while locked.");
585 * clear some bits on a range in the tree. This may require splitting
586 * or inserting elements in the tree, so the gfp mask is used to
587 * indicate which allocations or sleeping are allowed.
589 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
590 * the given range from the tree regardless of state (ie for truncate).
592 * the range [start, end] is inclusive.
594 * This takes the tree lock, and returns 0 on success and < 0 on error.
596 static int __clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
597 unsigned bits
, int wake
, int delete,
598 struct extent_state
**cached_state
,
599 gfp_t mask
, struct extent_changeset
*changeset
)
601 struct extent_state
*state
;
602 struct extent_state
*cached
;
603 struct extent_state
*prealloc
= NULL
;
604 struct rb_node
*node
;
609 btrfs_debug_check_extent_io_range(tree
, start
, end
);
611 if (bits
& EXTENT_DELALLOC
)
612 bits
|= EXTENT_NORESERVE
;
615 bits
|= ~EXTENT_CTLBITS
;
616 bits
|= EXTENT_FIRST_DELALLOC
;
618 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
621 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
623 * Don't care for allocation failure here because we might end
624 * up not needing the pre-allocated extent state at all, which
625 * is the case if we only have in the tree extent states that
626 * cover our input range and don't cover too any other range.
627 * If we end up needing a new extent state we allocate it later.
629 prealloc
= alloc_extent_state(mask
);
632 spin_lock(&tree
->lock
);
634 cached
= *cached_state
;
637 *cached_state
= NULL
;
641 if (cached
&& extent_state_in_tree(cached
) &&
642 cached
->start
<= start
&& cached
->end
> start
) {
644 refcount_dec(&cached
->refs
);
649 free_extent_state(cached
);
652 * this search will find the extents that end after
655 node
= tree_search(tree
, start
);
658 state
= rb_entry(node
, struct extent_state
, rb_node
);
660 if (state
->start
> end
)
662 WARN_ON(state
->end
< start
);
663 last_end
= state
->end
;
665 /* the state doesn't have the wanted bits, go ahead */
666 if (!(state
->state
& bits
)) {
667 state
= next_state(state
);
672 * | ---- desired range ---- |
674 * | ------------- state -------------- |
676 * We need to split the extent we found, and may flip
677 * bits on second half.
679 * If the extent we found extends past our range, we
680 * just split and search again. It'll get split again
681 * the next time though.
683 * If the extent we found is inside our range, we clear
684 * the desired bit on it.
687 if (state
->start
< start
) {
688 prealloc
= alloc_extent_state_atomic(prealloc
);
690 err
= split_state(tree
, state
, prealloc
, start
);
692 extent_io_tree_panic(tree
, err
);
697 if (state
->end
<= end
) {
698 state
= clear_state_bit(tree
, state
, &bits
, wake
,
705 * | ---- desired range ---- |
707 * We need to split the extent, and clear the bit
710 if (state
->start
<= end
&& state
->end
> end
) {
711 prealloc
= alloc_extent_state_atomic(prealloc
);
713 err
= split_state(tree
, state
, prealloc
, end
+ 1);
715 extent_io_tree_panic(tree
, err
);
720 clear_state_bit(tree
, prealloc
, &bits
, wake
, changeset
);
726 state
= clear_state_bit(tree
, state
, &bits
, wake
, changeset
);
728 if (last_end
== (u64
)-1)
730 start
= last_end
+ 1;
731 if (start
<= end
&& state
&& !need_resched())
737 spin_unlock(&tree
->lock
);
738 if (gfpflags_allow_blocking(mask
))
743 spin_unlock(&tree
->lock
);
745 free_extent_state(prealloc
);
751 static void wait_on_state(struct extent_io_tree
*tree
,
752 struct extent_state
*state
)
753 __releases(tree
->lock
)
754 __acquires(tree
->lock
)
757 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
758 spin_unlock(&tree
->lock
);
760 spin_lock(&tree
->lock
);
761 finish_wait(&state
->wq
, &wait
);
765 * waits for one or more bits to clear on a range in the state tree.
766 * The range [start, end] is inclusive.
767 * The tree lock is taken by this function
769 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
772 struct extent_state
*state
;
773 struct rb_node
*node
;
775 btrfs_debug_check_extent_io_range(tree
, start
, end
);
777 spin_lock(&tree
->lock
);
781 * this search will find all the extents that end after
784 node
= tree_search(tree
, start
);
789 state
= rb_entry(node
, struct extent_state
, rb_node
);
791 if (state
->start
> end
)
794 if (state
->state
& bits
) {
795 start
= state
->start
;
796 refcount_inc(&state
->refs
);
797 wait_on_state(tree
, state
);
798 free_extent_state(state
);
801 start
= state
->end
+ 1;
806 if (!cond_resched_lock(&tree
->lock
)) {
807 node
= rb_next(node
);
812 spin_unlock(&tree
->lock
);
815 static void set_state_bits(struct extent_io_tree
*tree
,
816 struct extent_state
*state
,
817 unsigned *bits
, struct extent_changeset
*changeset
)
819 unsigned bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
821 set_state_cb(tree
, state
, bits
);
822 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
823 u64 range
= state
->end
- state
->start
+ 1;
824 tree
->dirty_bytes
+= range
;
826 add_extent_changeset(state
, bits_to_set
, changeset
, 1);
827 state
->state
|= bits_to_set
;
830 static void cache_state_if_flags(struct extent_state
*state
,
831 struct extent_state
**cached_ptr
,
834 if (cached_ptr
&& !(*cached_ptr
)) {
835 if (!flags
|| (state
->state
& flags
)) {
837 refcount_inc(&state
->refs
);
842 static void cache_state(struct extent_state
*state
,
843 struct extent_state
**cached_ptr
)
845 return cache_state_if_flags(state
, cached_ptr
,
846 EXTENT_IOBITS
| EXTENT_BOUNDARY
);
850 * set some bits on a range in the tree. This may require allocations or
851 * sleeping, so the gfp mask is used to indicate what is allowed.
853 * If any of the exclusive bits are set, this will fail with -EEXIST if some
854 * part of the range already has the desired bits set. The start of the
855 * existing range is returned in failed_start in this case.
857 * [start, end] is inclusive This takes the tree lock.
860 static int __must_check
861 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
862 unsigned bits
, unsigned exclusive_bits
,
863 u64
*failed_start
, struct extent_state
**cached_state
,
864 gfp_t mask
, struct extent_changeset
*changeset
)
866 struct extent_state
*state
;
867 struct extent_state
*prealloc
= NULL
;
868 struct rb_node
*node
;
870 struct rb_node
*parent
;
875 btrfs_debug_check_extent_io_range(tree
, start
, end
);
877 bits
|= EXTENT_FIRST_DELALLOC
;
879 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
881 * Don't care for allocation failure here because we might end
882 * up not needing the pre-allocated extent state at all, which
883 * is the case if we only have in the tree extent states that
884 * cover our input range and don't cover too any other range.
885 * If we end up needing a new extent state we allocate it later.
887 prealloc
= alloc_extent_state(mask
);
890 spin_lock(&tree
->lock
);
891 if (cached_state
&& *cached_state
) {
892 state
= *cached_state
;
893 if (state
->start
<= start
&& state
->end
> start
&&
894 extent_state_in_tree(state
)) {
895 node
= &state
->rb_node
;
900 * this search will find all the extents that end after
903 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
905 prealloc
= alloc_extent_state_atomic(prealloc
);
907 err
= insert_state(tree
, prealloc
, start
, end
,
908 &p
, &parent
, &bits
, changeset
);
910 extent_io_tree_panic(tree
, err
);
912 cache_state(prealloc
, cached_state
);
916 state
= rb_entry(node
, struct extent_state
, rb_node
);
918 last_start
= state
->start
;
919 last_end
= state
->end
;
922 * | ---- desired range ---- |
925 * Just lock what we found and keep going
927 if (state
->start
== start
&& state
->end
<= end
) {
928 if (state
->state
& exclusive_bits
) {
929 *failed_start
= state
->start
;
934 set_state_bits(tree
, state
, &bits
, changeset
);
935 cache_state(state
, cached_state
);
936 merge_state(tree
, state
);
937 if (last_end
== (u64
)-1)
939 start
= last_end
+ 1;
940 state
= next_state(state
);
941 if (start
< end
&& state
&& state
->start
== start
&&
948 * | ---- desired range ---- |
951 * | ------------- state -------------- |
953 * We need to split the extent we found, and may flip bits on
956 * If the extent we found extends past our
957 * range, we just split and search again. It'll get split
958 * again the next time though.
960 * If the extent we found is inside our range, we set the
963 if (state
->start
< start
) {
964 if (state
->state
& exclusive_bits
) {
965 *failed_start
= start
;
970 prealloc
= alloc_extent_state_atomic(prealloc
);
972 err
= split_state(tree
, state
, prealloc
, start
);
974 extent_io_tree_panic(tree
, err
);
979 if (state
->end
<= end
) {
980 set_state_bits(tree
, state
, &bits
, changeset
);
981 cache_state(state
, cached_state
);
982 merge_state(tree
, state
);
983 if (last_end
== (u64
)-1)
985 start
= last_end
+ 1;
986 state
= next_state(state
);
987 if (start
< end
&& state
&& state
->start
== start
&&
994 * | ---- desired range ---- |
995 * | state | or | state |
997 * There's a hole, we need to insert something in it and
998 * ignore the extent we found.
1000 if (state
->start
> start
) {
1002 if (end
< last_start
)
1005 this_end
= last_start
- 1;
1007 prealloc
= alloc_extent_state_atomic(prealloc
);
1011 * Avoid to free 'prealloc' if it can be merged with
1014 err
= insert_state(tree
, prealloc
, start
, this_end
,
1015 NULL
, NULL
, &bits
, changeset
);
1017 extent_io_tree_panic(tree
, err
);
1019 cache_state(prealloc
, cached_state
);
1021 start
= this_end
+ 1;
1025 * | ---- desired range ---- |
1027 * We need to split the extent, and set the bit
1030 if (state
->start
<= end
&& state
->end
> end
) {
1031 if (state
->state
& exclusive_bits
) {
1032 *failed_start
= start
;
1037 prealloc
= alloc_extent_state_atomic(prealloc
);
1039 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1041 extent_io_tree_panic(tree
, err
);
1043 set_state_bits(tree
, prealloc
, &bits
, changeset
);
1044 cache_state(prealloc
, cached_state
);
1045 merge_state(tree
, prealloc
);
1053 spin_unlock(&tree
->lock
);
1054 if (gfpflags_allow_blocking(mask
))
1059 spin_unlock(&tree
->lock
);
1061 free_extent_state(prealloc
);
1067 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1068 unsigned bits
, u64
* failed_start
,
1069 struct extent_state
**cached_state
, gfp_t mask
)
1071 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1072 cached_state
, mask
, NULL
);
1077 * convert_extent_bit - convert all bits in a given range from one bit to
1079 * @tree: the io tree to search
1080 * @start: the start offset in bytes
1081 * @end: the end offset in bytes (inclusive)
1082 * @bits: the bits to set in this range
1083 * @clear_bits: the bits to clear in this range
1084 * @cached_state: state that we're going to cache
1086 * This will go through and set bits for the given range. If any states exist
1087 * already in this range they are set with the given bit and cleared of the
1088 * clear_bits. This is only meant to be used by things that are mergeable, ie
1089 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1090 * boundary bits like LOCK.
1092 * All allocations are done with GFP_NOFS.
1094 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1095 unsigned bits
, unsigned clear_bits
,
1096 struct extent_state
**cached_state
)
1098 struct extent_state
*state
;
1099 struct extent_state
*prealloc
= NULL
;
1100 struct rb_node
*node
;
1102 struct rb_node
*parent
;
1106 bool first_iteration
= true;
1108 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1113 * Best effort, don't worry if extent state allocation fails
1114 * here for the first iteration. We might have a cached state
1115 * that matches exactly the target range, in which case no
1116 * extent state allocations are needed. We'll only know this
1117 * after locking the tree.
1119 prealloc
= alloc_extent_state(GFP_NOFS
);
1120 if (!prealloc
&& !first_iteration
)
1124 spin_lock(&tree
->lock
);
1125 if (cached_state
&& *cached_state
) {
1126 state
= *cached_state
;
1127 if (state
->start
<= start
&& state
->end
> start
&&
1128 extent_state_in_tree(state
)) {
1129 node
= &state
->rb_node
;
1135 * this search will find all the extents that end after
1138 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1140 prealloc
= alloc_extent_state_atomic(prealloc
);
1145 err
= insert_state(tree
, prealloc
, start
, end
,
1146 &p
, &parent
, &bits
, NULL
);
1148 extent_io_tree_panic(tree
, err
);
1149 cache_state(prealloc
, cached_state
);
1153 state
= rb_entry(node
, struct extent_state
, rb_node
);
1155 last_start
= state
->start
;
1156 last_end
= state
->end
;
1159 * | ---- desired range ---- |
1162 * Just lock what we found and keep going
1164 if (state
->start
== start
&& state
->end
<= end
) {
1165 set_state_bits(tree
, state
, &bits
, NULL
);
1166 cache_state(state
, cached_state
);
1167 state
= clear_state_bit(tree
, state
, &clear_bits
, 0, NULL
);
1168 if (last_end
== (u64
)-1)
1170 start
= last_end
+ 1;
1171 if (start
< end
&& state
&& state
->start
== start
&&
1178 * | ---- desired range ---- |
1181 * | ------------- state -------------- |
1183 * We need to split the extent we found, and may flip bits on
1186 * If the extent we found extends past our
1187 * range, we just split and search again. It'll get split
1188 * again the next time though.
1190 * If the extent we found is inside our range, we set the
1191 * desired bit on it.
1193 if (state
->start
< start
) {
1194 prealloc
= alloc_extent_state_atomic(prealloc
);
1199 err
= split_state(tree
, state
, prealloc
, start
);
1201 extent_io_tree_panic(tree
, err
);
1205 if (state
->end
<= end
) {
1206 set_state_bits(tree
, state
, &bits
, NULL
);
1207 cache_state(state
, cached_state
);
1208 state
= clear_state_bit(tree
, state
, &clear_bits
, 0,
1210 if (last_end
== (u64
)-1)
1212 start
= last_end
+ 1;
1213 if (start
< end
&& state
&& state
->start
== start
&&
1220 * | ---- desired range ---- |
1221 * | state | or | state |
1223 * There's a hole, we need to insert something in it and
1224 * ignore the extent we found.
1226 if (state
->start
> start
) {
1228 if (end
< last_start
)
1231 this_end
= last_start
- 1;
1233 prealloc
= alloc_extent_state_atomic(prealloc
);
1240 * Avoid to free 'prealloc' if it can be merged with
1243 err
= insert_state(tree
, prealloc
, start
, this_end
,
1244 NULL
, NULL
, &bits
, NULL
);
1246 extent_io_tree_panic(tree
, err
);
1247 cache_state(prealloc
, cached_state
);
1249 start
= this_end
+ 1;
1253 * | ---- desired range ---- |
1255 * We need to split the extent, and set the bit
1258 if (state
->start
<= end
&& state
->end
> end
) {
1259 prealloc
= alloc_extent_state_atomic(prealloc
);
1265 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1267 extent_io_tree_panic(tree
, err
);
1269 set_state_bits(tree
, prealloc
, &bits
, NULL
);
1270 cache_state(prealloc
, cached_state
);
1271 clear_state_bit(tree
, prealloc
, &clear_bits
, 0, NULL
);
1279 spin_unlock(&tree
->lock
);
1281 first_iteration
= false;
1285 spin_unlock(&tree
->lock
);
1287 free_extent_state(prealloc
);
1292 /* wrappers around set/clear extent bit */
1293 int set_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1294 unsigned bits
, struct extent_changeset
*changeset
)
1297 * We don't support EXTENT_LOCKED yet, as current changeset will
1298 * record any bits changed, so for EXTENT_LOCKED case, it will
1299 * either fail with -EEXIST or changeset will record the whole
1302 BUG_ON(bits
& EXTENT_LOCKED
);
1304 return __set_extent_bit(tree
, start
, end
, bits
, 0, NULL
, NULL
, GFP_NOFS
,
1308 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1309 unsigned bits
, int wake
, int delete,
1310 struct extent_state
**cached
, gfp_t mask
)
1312 return __clear_extent_bit(tree
, start
, end
, bits
, wake
, delete,
1313 cached
, mask
, NULL
);
1316 int clear_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1317 unsigned bits
, struct extent_changeset
*changeset
)
1320 * Don't support EXTENT_LOCKED case, same reason as
1321 * set_record_extent_bits().
1323 BUG_ON(bits
& EXTENT_LOCKED
);
1325 return __clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, GFP_NOFS
,
1330 * either insert or lock state struct between start and end use mask to tell
1331 * us if waiting is desired.
1333 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1334 struct extent_state
**cached_state
)
1340 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
,
1341 EXTENT_LOCKED
, &failed_start
,
1342 cached_state
, GFP_NOFS
, NULL
);
1343 if (err
== -EEXIST
) {
1344 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1345 start
= failed_start
;
1348 WARN_ON(start
> end
);
1353 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1358 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1359 &failed_start
, NULL
, GFP_NOFS
, NULL
);
1360 if (err
== -EEXIST
) {
1361 if (failed_start
> start
)
1362 clear_extent_bit(tree
, start
, failed_start
- 1,
1363 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1369 void extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1371 unsigned long index
= start
>> PAGE_SHIFT
;
1372 unsigned long end_index
= end
>> PAGE_SHIFT
;
1375 while (index
<= end_index
) {
1376 page
= find_get_page(inode
->i_mapping
, index
);
1377 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1378 clear_page_dirty_for_io(page
);
1384 void extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1386 unsigned long index
= start
>> PAGE_SHIFT
;
1387 unsigned long end_index
= end
>> PAGE_SHIFT
;
1390 while (index
<= end_index
) {
1391 page
= find_get_page(inode
->i_mapping
, index
);
1392 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1393 __set_page_dirty_nobuffers(page
);
1394 account_page_redirty(page
);
1401 * helper function to set both pages and extents in the tree writeback
1403 static void set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1405 unsigned long index
= start
>> PAGE_SHIFT
;
1406 unsigned long end_index
= end
>> PAGE_SHIFT
;
1409 while (index
<= end_index
) {
1410 page
= find_get_page(tree
->mapping
, index
);
1411 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1412 set_page_writeback(page
);
1418 /* find the first state struct with 'bits' set after 'start', and
1419 * return it. tree->lock must be held. NULL will returned if
1420 * nothing was found after 'start'
1422 static struct extent_state
*
1423 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1424 u64 start
, unsigned bits
)
1426 struct rb_node
*node
;
1427 struct extent_state
*state
;
1430 * this search will find all the extents that end after
1433 node
= tree_search(tree
, start
);
1438 state
= rb_entry(node
, struct extent_state
, rb_node
);
1439 if (state
->end
>= start
&& (state
->state
& bits
))
1442 node
= rb_next(node
);
1451 * find the first offset in the io tree with 'bits' set. zero is
1452 * returned if we find something, and *start_ret and *end_ret are
1453 * set to reflect the state struct that was found.
1455 * If nothing was found, 1 is returned. If found something, return 0.
1457 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1458 u64
*start_ret
, u64
*end_ret
, unsigned bits
,
1459 struct extent_state
**cached_state
)
1461 struct extent_state
*state
;
1465 spin_lock(&tree
->lock
);
1466 if (cached_state
&& *cached_state
) {
1467 state
= *cached_state
;
1468 if (state
->end
== start
- 1 && extent_state_in_tree(state
)) {
1469 n
= rb_next(&state
->rb_node
);
1471 state
= rb_entry(n
, struct extent_state
,
1473 if (state
->state
& bits
)
1477 free_extent_state(*cached_state
);
1478 *cached_state
= NULL
;
1481 free_extent_state(*cached_state
);
1482 *cached_state
= NULL
;
1485 state
= find_first_extent_bit_state(tree
, start
, bits
);
1488 cache_state_if_flags(state
, cached_state
, 0);
1489 *start_ret
= state
->start
;
1490 *end_ret
= state
->end
;
1494 spin_unlock(&tree
->lock
);
1499 * find a contiguous range of bytes in the file marked as delalloc, not
1500 * more than 'max_bytes'. start and end are used to return the range,
1502 * 1 is returned if we find something, 0 if nothing was in the tree
1504 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1505 u64
*start
, u64
*end
, u64 max_bytes
,
1506 struct extent_state
**cached_state
)
1508 struct rb_node
*node
;
1509 struct extent_state
*state
;
1510 u64 cur_start
= *start
;
1512 u64 total_bytes
= 0;
1514 spin_lock(&tree
->lock
);
1517 * this search will find all the extents that end after
1520 node
= tree_search(tree
, cur_start
);
1528 state
= rb_entry(node
, struct extent_state
, rb_node
);
1529 if (found
&& (state
->start
!= cur_start
||
1530 (state
->state
& EXTENT_BOUNDARY
))) {
1533 if (!(state
->state
& EXTENT_DELALLOC
)) {
1539 *start
= state
->start
;
1540 *cached_state
= state
;
1541 refcount_inc(&state
->refs
);
1545 cur_start
= state
->end
+ 1;
1546 node
= rb_next(node
);
1547 total_bytes
+= state
->end
- state
->start
+ 1;
1548 if (total_bytes
>= max_bytes
)
1554 spin_unlock(&tree
->lock
);
1558 static int __process_pages_contig(struct address_space
*mapping
,
1559 struct page
*locked_page
,
1560 pgoff_t start_index
, pgoff_t end_index
,
1561 unsigned long page_ops
, pgoff_t
*index_ret
);
1563 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1564 struct page
*locked_page
,
1567 unsigned long index
= start
>> PAGE_SHIFT
;
1568 unsigned long end_index
= end
>> PAGE_SHIFT
;
1570 ASSERT(locked_page
);
1571 if (index
== locked_page
->index
&& end_index
== index
)
1574 __process_pages_contig(inode
->i_mapping
, locked_page
, index
, end_index
,
1578 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1579 struct page
*locked_page
,
1583 unsigned long index
= delalloc_start
>> PAGE_SHIFT
;
1584 unsigned long index_ret
= index
;
1585 unsigned long end_index
= delalloc_end
>> PAGE_SHIFT
;
1588 ASSERT(locked_page
);
1589 if (index
== locked_page
->index
&& index
== end_index
)
1592 ret
= __process_pages_contig(inode
->i_mapping
, locked_page
, index
,
1593 end_index
, PAGE_LOCK
, &index_ret
);
1595 __unlock_for_delalloc(inode
, locked_page
, delalloc_start
,
1596 (u64
)index_ret
<< PAGE_SHIFT
);
1601 * find a contiguous range of bytes in the file marked as delalloc, not
1602 * more than 'max_bytes'. start and end are used to return the range,
1604 * 1 is returned if we find something, 0 if nothing was in the tree
1606 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1607 struct extent_io_tree
*tree
,
1608 struct page
*locked_page
, u64
*start
,
1609 u64
*end
, u64 max_bytes
)
1614 struct extent_state
*cached_state
= NULL
;
1619 /* step one, find a bunch of delalloc bytes starting at start */
1620 delalloc_start
= *start
;
1622 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1623 max_bytes
, &cached_state
);
1624 if (!found
|| delalloc_end
<= *start
) {
1625 *start
= delalloc_start
;
1626 *end
= delalloc_end
;
1627 free_extent_state(cached_state
);
1632 * start comes from the offset of locked_page. We have to lock
1633 * pages in order, so we can't process delalloc bytes before
1636 if (delalloc_start
< *start
)
1637 delalloc_start
= *start
;
1640 * make sure to limit the number of pages we try to lock down
1642 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1643 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1645 /* step two, lock all the pages after the page that has start */
1646 ret
= lock_delalloc_pages(inode
, locked_page
,
1647 delalloc_start
, delalloc_end
);
1648 if (ret
== -EAGAIN
) {
1649 /* some of the pages are gone, lets avoid looping by
1650 * shortening the size of the delalloc range we're searching
1652 free_extent_state(cached_state
);
1653 cached_state
= NULL
;
1655 max_bytes
= PAGE_SIZE
;
1663 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1665 /* step three, lock the state bits for the whole range */
1666 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, &cached_state
);
1668 /* then test to make sure it is all still delalloc */
1669 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1670 EXTENT_DELALLOC
, 1, cached_state
);
1672 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1673 &cached_state
, GFP_NOFS
);
1674 __unlock_for_delalloc(inode
, locked_page
,
1675 delalloc_start
, delalloc_end
);
1679 free_extent_state(cached_state
);
1680 *start
= delalloc_start
;
1681 *end
= delalloc_end
;
1686 static int __process_pages_contig(struct address_space
*mapping
,
1687 struct page
*locked_page
,
1688 pgoff_t start_index
, pgoff_t end_index
,
1689 unsigned long page_ops
, pgoff_t
*index_ret
)
1691 unsigned long nr_pages
= end_index
- start_index
+ 1;
1692 unsigned long pages_locked
= 0;
1693 pgoff_t index
= start_index
;
1694 struct page
*pages
[16];
1699 if (page_ops
& PAGE_LOCK
) {
1700 ASSERT(page_ops
== PAGE_LOCK
);
1701 ASSERT(index_ret
&& *index_ret
== start_index
);
1704 if ((page_ops
& PAGE_SET_ERROR
) && nr_pages
> 0)
1705 mapping_set_error(mapping
, -EIO
);
1707 while (nr_pages
> 0) {
1708 ret
= find_get_pages_contig(mapping
, index
,
1709 min_t(unsigned long,
1710 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1713 * Only if we're going to lock these pages,
1714 * can we find nothing at @index.
1716 ASSERT(page_ops
& PAGE_LOCK
);
1721 for (i
= 0; i
< ret
; i
++) {
1722 if (page_ops
& PAGE_SET_PRIVATE2
)
1723 SetPagePrivate2(pages
[i
]);
1725 if (pages
[i
] == locked_page
) {
1730 if (page_ops
& PAGE_CLEAR_DIRTY
)
1731 clear_page_dirty_for_io(pages
[i
]);
1732 if (page_ops
& PAGE_SET_WRITEBACK
)
1733 set_page_writeback(pages
[i
]);
1734 if (page_ops
& PAGE_SET_ERROR
)
1735 SetPageError(pages
[i
]);
1736 if (page_ops
& PAGE_END_WRITEBACK
)
1737 end_page_writeback(pages
[i
]);
1738 if (page_ops
& PAGE_UNLOCK
)
1739 unlock_page(pages
[i
]);
1740 if (page_ops
& PAGE_LOCK
) {
1741 lock_page(pages
[i
]);
1742 if (!PageDirty(pages
[i
]) ||
1743 pages
[i
]->mapping
!= mapping
) {
1744 unlock_page(pages
[i
]);
1758 if (err
&& index_ret
)
1759 *index_ret
= start_index
+ pages_locked
- 1;
1763 void extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1764 u64 delalloc_end
, struct page
*locked_page
,
1765 unsigned clear_bits
,
1766 unsigned long page_ops
)
1768 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, clear_bits
, 1, 0,
1771 __process_pages_contig(inode
->i_mapping
, locked_page
,
1772 start
>> PAGE_SHIFT
, end
>> PAGE_SHIFT
,
1777 * count the number of bytes in the tree that have a given bit(s)
1778 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1779 * cached. The total number found is returned.
1781 u64
count_range_bits(struct extent_io_tree
*tree
,
1782 u64
*start
, u64 search_end
, u64 max_bytes
,
1783 unsigned bits
, int contig
)
1785 struct rb_node
*node
;
1786 struct extent_state
*state
;
1787 u64 cur_start
= *start
;
1788 u64 total_bytes
= 0;
1792 if (WARN_ON(search_end
<= cur_start
))
1795 spin_lock(&tree
->lock
);
1796 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1797 total_bytes
= tree
->dirty_bytes
;
1801 * this search will find all the extents that end after
1804 node
= tree_search(tree
, cur_start
);
1809 state
= rb_entry(node
, struct extent_state
, rb_node
);
1810 if (state
->start
> search_end
)
1812 if (contig
&& found
&& state
->start
> last
+ 1)
1814 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1815 total_bytes
+= min(search_end
, state
->end
) + 1 -
1816 max(cur_start
, state
->start
);
1817 if (total_bytes
>= max_bytes
)
1820 *start
= max(cur_start
, state
->start
);
1824 } else if (contig
&& found
) {
1827 node
= rb_next(node
);
1832 spin_unlock(&tree
->lock
);
1837 * set the private field for a given byte offset in the tree. If there isn't
1838 * an extent_state there already, this does nothing.
1840 static noinline
int set_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1841 struct io_failure_record
*failrec
)
1843 struct rb_node
*node
;
1844 struct extent_state
*state
;
1847 spin_lock(&tree
->lock
);
1849 * this search will find all the extents that end after
1852 node
= tree_search(tree
, start
);
1857 state
= rb_entry(node
, struct extent_state
, rb_node
);
1858 if (state
->start
!= start
) {
1862 state
->failrec
= failrec
;
1864 spin_unlock(&tree
->lock
);
1868 static noinline
int get_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1869 struct io_failure_record
**failrec
)
1871 struct rb_node
*node
;
1872 struct extent_state
*state
;
1875 spin_lock(&tree
->lock
);
1877 * this search will find all the extents that end after
1880 node
= tree_search(tree
, start
);
1885 state
= rb_entry(node
, struct extent_state
, rb_node
);
1886 if (state
->start
!= start
) {
1890 *failrec
= state
->failrec
;
1892 spin_unlock(&tree
->lock
);
1897 * searches a range in the state tree for a given mask.
1898 * If 'filled' == 1, this returns 1 only if every extent in the tree
1899 * has the bits set. Otherwise, 1 is returned if any bit in the
1900 * range is found set.
1902 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1903 unsigned bits
, int filled
, struct extent_state
*cached
)
1905 struct extent_state
*state
= NULL
;
1906 struct rb_node
*node
;
1909 spin_lock(&tree
->lock
);
1910 if (cached
&& extent_state_in_tree(cached
) && cached
->start
<= start
&&
1911 cached
->end
> start
)
1912 node
= &cached
->rb_node
;
1914 node
= tree_search(tree
, start
);
1915 while (node
&& start
<= end
) {
1916 state
= rb_entry(node
, struct extent_state
, rb_node
);
1918 if (filled
&& state
->start
> start
) {
1923 if (state
->start
> end
)
1926 if (state
->state
& bits
) {
1930 } else if (filled
) {
1935 if (state
->end
== (u64
)-1)
1938 start
= state
->end
+ 1;
1941 node
= rb_next(node
);
1948 spin_unlock(&tree
->lock
);
1953 * helper function to set a given page up to date if all the
1954 * extents in the tree for that page are up to date
1956 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1958 u64 start
= page_offset(page
);
1959 u64 end
= start
+ PAGE_SIZE
- 1;
1960 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1961 SetPageUptodate(page
);
1964 int free_io_failure(struct btrfs_inode
*inode
, struct io_failure_record
*rec
)
1968 struct extent_io_tree
*failure_tree
= &inode
->io_failure_tree
;
1970 set_state_failrec(failure_tree
, rec
->start
, NULL
);
1971 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1972 rec
->start
+ rec
->len
- 1,
1973 EXTENT_LOCKED
| EXTENT_DIRTY
);
1977 ret
= clear_extent_bits(&inode
->io_tree
, rec
->start
,
1978 rec
->start
+ rec
->len
- 1,
1988 * this bypasses the standard btrfs submit functions deliberately, as
1989 * the standard behavior is to write all copies in a raid setup. here we only
1990 * want to write the one bad copy. so we do the mapping for ourselves and issue
1991 * submit_bio directly.
1992 * to avoid any synchronization issues, wait for the data after writing, which
1993 * actually prevents the read that triggered the error from finishing.
1994 * currently, there can be no more than two copies of every data bit. thus,
1995 * exactly one rewrite is required.
1997 int repair_io_failure(struct btrfs_inode
*inode
, u64 start
, u64 length
,
1998 u64 logical
, struct page
*page
,
1999 unsigned int pg_offset
, int mirror_num
)
2001 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2003 struct btrfs_device
*dev
;
2006 struct btrfs_bio
*bbio
= NULL
;
2009 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2010 BUG_ON(!mirror_num
);
2012 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2015 bio
->bi_iter
.bi_size
= 0;
2016 map_length
= length
;
2019 * Avoid races with device replace and make sure our bbio has devices
2020 * associated to its stripes that don't go away while we are doing the
2021 * read repair operation.
2023 btrfs_bio_counter_inc_blocked(fs_info
);
2024 if (btrfs_is_parity_mirror(fs_info
, logical
, length
, mirror_num
)) {
2026 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2027 * to update all raid stripes, but here we just want to correct
2028 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2029 * stripe's dev and sector.
2031 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_READ
, logical
,
2032 &map_length
, &bbio
, 0);
2034 btrfs_bio_counter_dec(fs_info
);
2038 ASSERT(bbio
->mirror_num
== 1);
2040 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_WRITE
, logical
,
2041 &map_length
, &bbio
, mirror_num
);
2043 btrfs_bio_counter_dec(fs_info
);
2047 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2050 sector
= bbio
->stripes
[bbio
->mirror_num
- 1].physical
>> 9;
2051 bio
->bi_iter
.bi_sector
= sector
;
2052 dev
= bbio
->stripes
[bbio
->mirror_num
- 1].dev
;
2053 btrfs_put_bbio(bbio
);
2054 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2055 btrfs_bio_counter_dec(fs_info
);
2059 bio
->bi_bdev
= dev
->bdev
;
2060 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
;
2061 bio_add_page(bio
, page
, length
, pg_offset
);
2063 if (btrfsic_submit_bio_wait(bio
)) {
2064 /* try to remap that extent elsewhere? */
2065 btrfs_bio_counter_dec(fs_info
);
2067 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2071 btrfs_info_rl_in_rcu(fs_info
,
2072 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2073 btrfs_ino(inode
), start
,
2074 rcu_str_deref(dev
->name
), sector
);
2075 btrfs_bio_counter_dec(fs_info
);
2080 int repair_eb_io_failure(struct btrfs_fs_info
*fs_info
,
2081 struct extent_buffer
*eb
, int mirror_num
)
2083 u64 start
= eb
->start
;
2084 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2087 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2090 for (i
= 0; i
< num_pages
; i
++) {
2091 struct page
*p
= eb
->pages
[i
];
2093 ret
= repair_io_failure(BTRFS_I(fs_info
->btree_inode
), start
,
2094 PAGE_SIZE
, start
, p
,
2095 start
- page_offset(p
), mirror_num
);
2105 * each time an IO finishes, we do a fast check in the IO failure tree
2106 * to see if we need to process or clean up an io_failure_record
2108 int clean_io_failure(struct btrfs_inode
*inode
, u64 start
, struct page
*page
,
2109 unsigned int pg_offset
)
2112 struct io_failure_record
*failrec
;
2113 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2114 struct extent_state
*state
;
2119 ret
= count_range_bits(&inode
->io_failure_tree
, &private,
2120 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2124 ret
= get_state_failrec(&inode
->io_failure_tree
, start
,
2129 BUG_ON(!failrec
->this_mirror
);
2131 if (failrec
->in_validation
) {
2132 /* there was no real error, just free the record */
2133 btrfs_debug(fs_info
,
2134 "clean_io_failure: freeing dummy error at %llu",
2138 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2141 spin_lock(&inode
->io_tree
.lock
);
2142 state
= find_first_extent_bit_state(&inode
->io_tree
,
2145 spin_unlock(&inode
->io_tree
.lock
);
2147 if (state
&& state
->start
<= failrec
->start
&&
2148 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2149 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2151 if (num_copies
> 1) {
2152 repair_io_failure(inode
, start
, failrec
->len
,
2153 failrec
->logical
, page
,
2154 pg_offset
, failrec
->failed_mirror
);
2159 free_io_failure(inode
, failrec
);
2165 * Can be called when
2166 * - hold extent lock
2167 * - under ordered extent
2168 * - the inode is freeing
2170 void btrfs_free_io_failure_record(struct btrfs_inode
*inode
, u64 start
, u64 end
)
2172 struct extent_io_tree
*failure_tree
= &inode
->io_failure_tree
;
2173 struct io_failure_record
*failrec
;
2174 struct extent_state
*state
, *next
;
2176 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2179 spin_lock(&failure_tree
->lock
);
2180 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2182 if (state
->start
> end
)
2185 ASSERT(state
->end
<= end
);
2187 next
= next_state(state
);
2189 failrec
= state
->failrec
;
2190 free_extent_state(state
);
2195 spin_unlock(&failure_tree
->lock
);
2198 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2199 struct io_failure_record
**failrec_ret
)
2201 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2202 struct io_failure_record
*failrec
;
2203 struct extent_map
*em
;
2204 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2205 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2206 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2210 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2212 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2216 failrec
->start
= start
;
2217 failrec
->len
= end
- start
+ 1;
2218 failrec
->this_mirror
= 0;
2219 failrec
->bio_flags
= 0;
2220 failrec
->in_validation
= 0;
2222 read_lock(&em_tree
->lock
);
2223 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2225 read_unlock(&em_tree
->lock
);
2230 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2231 free_extent_map(em
);
2234 read_unlock(&em_tree
->lock
);
2240 logical
= start
- em
->start
;
2241 logical
= em
->block_start
+ logical
;
2242 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2243 logical
= em
->block_start
;
2244 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2245 extent_set_compress_type(&failrec
->bio_flags
,
2249 btrfs_debug(fs_info
,
2250 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2251 logical
, start
, failrec
->len
);
2253 failrec
->logical
= logical
;
2254 free_extent_map(em
);
2256 /* set the bits in the private failure tree */
2257 ret
= set_extent_bits(failure_tree
, start
, end
,
2258 EXTENT_LOCKED
| EXTENT_DIRTY
);
2260 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2261 /* set the bits in the inode's tree */
2263 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2269 btrfs_debug(fs_info
,
2270 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2271 failrec
->logical
, failrec
->start
, failrec
->len
,
2272 failrec
->in_validation
);
2274 * when data can be on disk more than twice, add to failrec here
2275 * (e.g. with a list for failed_mirror) to make
2276 * clean_io_failure() clean all those errors at once.
2280 *failrec_ret
= failrec
;
2285 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2286 struct io_failure_record
*failrec
, int failed_mirror
)
2288 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2291 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
2292 if (num_copies
== 1) {
2294 * we only have a single copy of the data, so don't bother with
2295 * all the retry and error correction code that follows. no
2296 * matter what the error is, it is very likely to persist.
2298 btrfs_debug(fs_info
,
2299 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2300 num_copies
, failrec
->this_mirror
, failed_mirror
);
2305 * there are two premises:
2306 * a) deliver good data to the caller
2307 * b) correct the bad sectors on disk
2309 if (failed_bio
->bi_vcnt
> 1) {
2311 * to fulfill b), we need to know the exact failing sectors, as
2312 * we don't want to rewrite any more than the failed ones. thus,
2313 * we need separate read requests for the failed bio
2315 * if the following BUG_ON triggers, our validation request got
2316 * merged. we need separate requests for our algorithm to work.
2318 BUG_ON(failrec
->in_validation
);
2319 failrec
->in_validation
= 1;
2320 failrec
->this_mirror
= failed_mirror
;
2323 * we're ready to fulfill a) and b) alongside. get a good copy
2324 * of the failed sector and if we succeed, we have setup
2325 * everything for repair_io_failure to do the rest for us.
2327 if (failrec
->in_validation
) {
2328 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2329 failrec
->in_validation
= 0;
2330 failrec
->this_mirror
= 0;
2332 failrec
->failed_mirror
= failed_mirror
;
2333 failrec
->this_mirror
++;
2334 if (failrec
->this_mirror
== failed_mirror
)
2335 failrec
->this_mirror
++;
2338 if (failrec
->this_mirror
> num_copies
) {
2339 btrfs_debug(fs_info
,
2340 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2341 num_copies
, failrec
->this_mirror
, failed_mirror
);
2349 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2350 struct io_failure_record
*failrec
,
2351 struct page
*page
, int pg_offset
, int icsum
,
2352 bio_end_io_t
*endio_func
, void *data
)
2354 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2356 struct btrfs_io_bio
*btrfs_failed_bio
;
2357 struct btrfs_io_bio
*btrfs_bio
;
2359 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2363 bio
->bi_end_io
= endio_func
;
2364 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2365 bio
->bi_bdev
= fs_info
->fs_devices
->latest_bdev
;
2366 bio
->bi_iter
.bi_size
= 0;
2367 bio
->bi_private
= data
;
2369 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2370 if (btrfs_failed_bio
->csum
) {
2371 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2373 btrfs_bio
= btrfs_io_bio(bio
);
2374 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2376 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2380 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2386 * this is a generic handler for readpage errors (default
2387 * readpage_io_failed_hook). if other copies exist, read those and write back
2388 * good data to the failed position. does not investigate in remapping the
2389 * failed extent elsewhere, hoping the device will be smart enough to do this as
2393 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2394 struct page
*page
, u64 start
, u64 end
,
2397 struct io_failure_record
*failrec
;
2398 struct inode
*inode
= page
->mapping
->host
;
2399 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2402 blk_status_t status
;
2405 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
2407 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2411 ret
= btrfs_check_repairable(inode
, failed_bio
, failrec
, failed_mirror
);
2413 free_io_failure(BTRFS_I(inode
), failrec
);
2417 if (failed_bio
->bi_vcnt
> 1)
2418 read_mode
|= REQ_FAILFAST_DEV
;
2420 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2421 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2422 start
- page_offset(page
),
2423 (int)phy_offset
, failed_bio
->bi_end_io
,
2426 free_io_failure(BTRFS_I(inode
), failrec
);
2429 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
2431 btrfs_debug(btrfs_sb(inode
->i_sb
),
2432 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2433 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2435 status
= tree
->ops
->submit_bio_hook(inode
, bio
, failrec
->this_mirror
,
2436 failrec
->bio_flags
, 0);
2438 free_io_failure(BTRFS_I(inode
), failrec
);
2440 ret
= blk_status_to_errno(status
);
2446 /* lots and lots of room for performance fixes in the end_bio funcs */
2448 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2450 int uptodate
= (err
== 0);
2451 struct extent_io_tree
*tree
;
2454 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2456 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2457 tree
->ops
->writepage_end_io_hook(page
, start
, end
, NULL
,
2461 ClearPageUptodate(page
);
2463 ret
= ret
< 0 ? ret
: -EIO
;
2464 mapping_set_error(page
->mapping
, ret
);
2469 * after a writepage IO is done, we need to:
2470 * clear the uptodate bits on error
2471 * clear the writeback bits in the extent tree for this IO
2472 * end_page_writeback if the page has no more pending IO
2474 * Scheduling is not allowed, so the extent state tree is expected
2475 * to have one and only one object corresponding to this IO.
2477 static void end_bio_extent_writepage(struct bio
*bio
)
2479 int error
= blk_status_to_errno(bio
->bi_status
);
2480 struct bio_vec
*bvec
;
2485 bio_for_each_segment_all(bvec
, bio
, i
) {
2486 struct page
*page
= bvec
->bv_page
;
2487 struct inode
*inode
= page
->mapping
->host
;
2488 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2490 /* We always issue full-page reads, but if some block
2491 * in a page fails to read, blk_update_request() will
2492 * advance bv_offset and adjust bv_len to compensate.
2493 * Print a warning for nonzero offsets, and an error
2494 * if they don't add up to a full page. */
2495 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2496 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2498 "partial page write in btrfs with offset %u and length %u",
2499 bvec
->bv_offset
, bvec
->bv_len
);
2502 "incomplete page write in btrfs with offset %u and length %u",
2503 bvec
->bv_offset
, bvec
->bv_len
);
2506 start
= page_offset(page
);
2507 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2509 end_extent_writepage(page
, error
, start
, end
);
2510 end_page_writeback(page
);
2517 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2520 struct extent_state
*cached
= NULL
;
2521 u64 end
= start
+ len
- 1;
2523 if (uptodate
&& tree
->track_uptodate
)
2524 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2525 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2529 * after a readpage IO is done, we need to:
2530 * clear the uptodate bits on error
2531 * set the uptodate bits if things worked
2532 * set the page up to date if all extents in the tree are uptodate
2533 * clear the lock bit in the extent tree
2534 * unlock the page if there are no other extents locked for it
2536 * Scheduling is not allowed, so the extent state tree is expected
2537 * to have one and only one object corresponding to this IO.
2539 static void end_bio_extent_readpage(struct bio
*bio
)
2541 struct bio_vec
*bvec
;
2542 int uptodate
= !bio
->bi_status
;
2543 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2544 struct extent_io_tree
*tree
;
2549 u64 extent_start
= 0;
2555 bio_for_each_segment_all(bvec
, bio
, i
) {
2556 struct page
*page
= bvec
->bv_page
;
2557 struct inode
*inode
= page
->mapping
->host
;
2558 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2560 btrfs_debug(fs_info
,
2561 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2562 (u64
)bio
->bi_iter
.bi_sector
, bio
->bi_status
,
2563 io_bio
->mirror_num
);
2564 tree
= &BTRFS_I(inode
)->io_tree
;
2566 /* We always issue full-page reads, but if some block
2567 * in a page fails to read, blk_update_request() will
2568 * advance bv_offset and adjust bv_len to compensate.
2569 * Print a warning for nonzero offsets, and an error
2570 * if they don't add up to a full page. */
2571 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2572 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2574 "partial page read in btrfs with offset %u and length %u",
2575 bvec
->bv_offset
, bvec
->bv_len
);
2578 "incomplete page read in btrfs with offset %u and length %u",
2579 bvec
->bv_offset
, bvec
->bv_len
);
2582 start
= page_offset(page
);
2583 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2586 mirror
= io_bio
->mirror_num
;
2587 if (likely(uptodate
&& tree
->ops
)) {
2588 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2594 clean_io_failure(BTRFS_I(inode
), start
,
2598 if (likely(uptodate
))
2602 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2603 if (ret
== -EAGAIN
) {
2605 * Data inode's readpage_io_failed_hook() always
2608 * The generic bio_readpage_error handles errors
2609 * the following way: If possible, new read
2610 * requests are created and submitted and will
2611 * end up in end_bio_extent_readpage as well (if
2612 * we're lucky, not in the !uptodate case). In
2613 * that case it returns 0 and we just go on with
2614 * the next page in our bio. If it can't handle
2615 * the error it will return -EIO and we remain
2616 * responsible for that page.
2618 ret
= bio_readpage_error(bio
, offset
, page
,
2619 start
, end
, mirror
);
2621 uptodate
= !bio
->bi_status
;
2628 * metadata's readpage_io_failed_hook() always returns
2629 * -EIO and fixes nothing. -EIO is also returned if
2630 * data inode error could not be fixed.
2632 ASSERT(ret
== -EIO
);
2635 if (likely(uptodate
)) {
2636 loff_t i_size
= i_size_read(inode
);
2637 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2640 /* Zero out the end if this page straddles i_size */
2641 off
= i_size
& (PAGE_SIZE
-1);
2642 if (page
->index
== end_index
&& off
)
2643 zero_user_segment(page
, off
, PAGE_SIZE
);
2644 SetPageUptodate(page
);
2646 ClearPageUptodate(page
);
2652 if (unlikely(!uptodate
)) {
2654 endio_readpage_release_extent(tree
,
2660 endio_readpage_release_extent(tree
, start
,
2661 end
- start
+ 1, 0);
2662 } else if (!extent_len
) {
2663 extent_start
= start
;
2664 extent_len
= end
+ 1 - start
;
2665 } else if (extent_start
+ extent_len
== start
) {
2666 extent_len
+= end
+ 1 - start
;
2668 endio_readpage_release_extent(tree
, extent_start
,
2669 extent_len
, uptodate
);
2670 extent_start
= start
;
2671 extent_len
= end
+ 1 - start
;
2676 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2679 io_bio
->end_io(io_bio
, blk_status_to_errno(bio
->bi_status
));
2684 * this allocates from the btrfs_bioset. We're returning a bio right now
2685 * but you can call btrfs_io_bio for the appropriate container_of magic
2688 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2691 struct btrfs_io_bio
*btrfs_bio
;
2694 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2696 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2697 while (!bio
&& (nr_vecs
/= 2)) {
2698 bio
= bio_alloc_bioset(gfp_flags
,
2699 nr_vecs
, btrfs_bioset
);
2704 bio
->bi_bdev
= bdev
;
2705 bio
->bi_iter
.bi_sector
= first_sector
;
2706 btrfs_bio
= btrfs_io_bio(bio
);
2707 btrfs_bio
->csum
= NULL
;
2708 btrfs_bio
->csum_allocated
= NULL
;
2709 btrfs_bio
->end_io
= NULL
;
2714 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2716 struct btrfs_io_bio
*btrfs_bio
;
2719 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2721 btrfs_bio
= btrfs_io_bio(new);
2722 btrfs_bio
->csum
= NULL
;
2723 btrfs_bio
->csum_allocated
= NULL
;
2724 btrfs_bio
->end_io
= NULL
;
2729 /* this also allocates from the btrfs_bioset */
2730 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2732 struct btrfs_io_bio
*btrfs_bio
;
2735 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2737 btrfs_bio
= btrfs_io_bio(bio
);
2738 btrfs_bio
->csum
= NULL
;
2739 btrfs_bio
->csum_allocated
= NULL
;
2740 btrfs_bio
->end_io
= NULL
;
2746 static int __must_check
submit_one_bio(struct bio
*bio
, int mirror_num
,
2747 unsigned long bio_flags
)
2749 blk_status_t ret
= 0;
2750 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2751 struct page
*page
= bvec
->bv_page
;
2752 struct extent_io_tree
*tree
= bio
->bi_private
;
2755 start
= page_offset(page
) + bvec
->bv_offset
;
2757 bio
->bi_private
= NULL
;
2761 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, bio
,
2762 mirror_num
, bio_flags
, start
);
2764 btrfsic_submit_bio(bio
);
2767 return blk_status_to_errno(ret
);
2770 static int merge_bio(struct extent_io_tree
*tree
, struct page
*page
,
2771 unsigned long offset
, size_t size
, struct bio
*bio
,
2772 unsigned long bio_flags
)
2776 ret
= tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
,
2782 static int submit_extent_page(int op
, int op_flags
, struct extent_io_tree
*tree
,
2783 struct writeback_control
*wbc
,
2784 struct page
*page
, sector_t sector
,
2785 size_t size
, unsigned long offset
,
2786 struct block_device
*bdev
,
2787 struct bio
**bio_ret
,
2788 bio_end_io_t end_io_func
,
2790 unsigned long prev_bio_flags
,
2791 unsigned long bio_flags
,
2792 bool force_bio_submit
)
2797 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2798 size_t page_size
= min_t(size_t, size
, PAGE_SIZE
);
2800 if (bio_ret
&& *bio_ret
) {
2803 contig
= bio
->bi_iter
.bi_sector
== sector
;
2805 contig
= bio_end_sector(bio
) == sector
;
2807 if (prev_bio_flags
!= bio_flags
|| !contig
||
2809 merge_bio(tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2810 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2811 ret
= submit_one_bio(bio
, mirror_num
, prev_bio_flags
);
2819 wbc_account_io(wbc
, page
, page_size
);
2824 bio
= btrfs_bio_alloc(bdev
, sector
, BIO_MAX_PAGES
,
2825 GFP_NOFS
| __GFP_HIGH
);
2829 bio_add_page(bio
, page
, page_size
, offset
);
2830 bio
->bi_end_io
= end_io_func
;
2831 bio
->bi_private
= tree
;
2832 bio_set_op_attrs(bio
, op
, op_flags
);
2834 wbc_init_bio(wbc
, bio
);
2835 wbc_account_io(wbc
, page
, page_size
);
2841 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
2846 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2849 if (!PagePrivate(page
)) {
2850 SetPagePrivate(page
);
2852 set_page_private(page
, (unsigned long)eb
);
2854 WARN_ON(page
->private != (unsigned long)eb
);
2858 void set_page_extent_mapped(struct page
*page
)
2860 if (!PagePrivate(page
)) {
2861 SetPagePrivate(page
);
2863 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2867 static struct extent_map
*
2868 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2869 u64 start
, u64 len
, get_extent_t
*get_extent
,
2870 struct extent_map
**em_cached
)
2872 struct extent_map
*em
;
2874 if (em_cached
&& *em_cached
) {
2876 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2877 start
< extent_map_end(em
)) {
2878 refcount_inc(&em
->refs
);
2882 free_extent_map(em
);
2886 em
= get_extent(BTRFS_I(inode
), page
, pg_offset
, start
, len
, 0);
2887 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2889 refcount_inc(&em
->refs
);
2895 * basic readpage implementation. Locked extent state structs are inserted
2896 * into the tree that are removed when the IO is done (by the end_io
2898 * XXX JDM: This needs looking at to ensure proper page locking
2899 * return 0 on success, otherwise return error
2901 static int __do_readpage(struct extent_io_tree
*tree
,
2903 get_extent_t
*get_extent
,
2904 struct extent_map
**em_cached
,
2905 struct bio
**bio
, int mirror_num
,
2906 unsigned long *bio_flags
, int read_flags
,
2909 struct inode
*inode
= page
->mapping
->host
;
2910 u64 start
= page_offset(page
);
2911 u64 page_end
= start
+ PAGE_SIZE
- 1;
2915 u64 last_byte
= i_size_read(inode
);
2919 struct extent_map
*em
;
2920 struct block_device
*bdev
;
2923 size_t pg_offset
= 0;
2925 size_t disk_io_size
;
2926 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2927 unsigned long this_bio_flag
= 0;
2929 set_page_extent_mapped(page
);
2932 if (!PageUptodate(page
)) {
2933 if (cleancache_get_page(page
) == 0) {
2934 BUG_ON(blocksize
!= PAGE_SIZE
);
2935 unlock_extent(tree
, start
, end
);
2940 if (page
->index
== last_byte
>> PAGE_SHIFT
) {
2942 size_t zero_offset
= last_byte
& (PAGE_SIZE
- 1);
2945 iosize
= PAGE_SIZE
- zero_offset
;
2946 userpage
= kmap_atomic(page
);
2947 memset(userpage
+ zero_offset
, 0, iosize
);
2948 flush_dcache_page(page
);
2949 kunmap_atomic(userpage
);
2952 while (cur
<= end
) {
2953 bool force_bio_submit
= false;
2955 if (cur
>= last_byte
) {
2957 struct extent_state
*cached
= NULL
;
2959 iosize
= PAGE_SIZE
- pg_offset
;
2960 userpage
= kmap_atomic(page
);
2961 memset(userpage
+ pg_offset
, 0, iosize
);
2962 flush_dcache_page(page
);
2963 kunmap_atomic(userpage
);
2964 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2966 unlock_extent_cached(tree
, cur
,
2971 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2972 end
- cur
+ 1, get_extent
, em_cached
);
2973 if (IS_ERR_OR_NULL(em
)) {
2975 unlock_extent(tree
, cur
, end
);
2978 extent_offset
= cur
- em
->start
;
2979 BUG_ON(extent_map_end(em
) <= cur
);
2982 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2983 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2984 extent_set_compress_type(&this_bio_flag
,
2988 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2989 cur_end
= min(extent_map_end(em
) - 1, end
);
2990 iosize
= ALIGN(iosize
, blocksize
);
2991 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2992 disk_io_size
= em
->block_len
;
2993 sector
= em
->block_start
>> 9;
2995 sector
= (em
->block_start
+ extent_offset
) >> 9;
2996 disk_io_size
= iosize
;
2999 block_start
= em
->block_start
;
3000 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
3001 block_start
= EXTENT_MAP_HOLE
;
3004 * If we have a file range that points to a compressed extent
3005 * and it's followed by a consecutive file range that points to
3006 * to the same compressed extent (possibly with a different
3007 * offset and/or length, so it either points to the whole extent
3008 * or only part of it), we must make sure we do not submit a
3009 * single bio to populate the pages for the 2 ranges because
3010 * this makes the compressed extent read zero out the pages
3011 * belonging to the 2nd range. Imagine the following scenario:
3014 * [0 - 8K] [8K - 24K]
3017 * points to extent X, points to extent X,
3018 * offset 4K, length of 8K offset 0, length 16K
3020 * [extent X, compressed length = 4K uncompressed length = 16K]
3022 * If the bio to read the compressed extent covers both ranges,
3023 * it will decompress extent X into the pages belonging to the
3024 * first range and then it will stop, zeroing out the remaining
3025 * pages that belong to the other range that points to extent X.
3026 * So here we make sure we submit 2 bios, one for the first
3027 * range and another one for the third range. Both will target
3028 * the same physical extent from disk, but we can't currently
3029 * make the compressed bio endio callback populate the pages
3030 * for both ranges because each compressed bio is tightly
3031 * coupled with a single extent map, and each range can have
3032 * an extent map with a different offset value relative to the
3033 * uncompressed data of our extent and different lengths. This
3034 * is a corner case so we prioritize correctness over
3035 * non-optimal behavior (submitting 2 bios for the same extent).
3037 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3038 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3039 *prev_em_start
!= em
->orig_start
)
3040 force_bio_submit
= true;
3043 *prev_em_start
= em
->orig_start
;
3045 free_extent_map(em
);
3048 /* we've found a hole, just zero and go on */
3049 if (block_start
== EXTENT_MAP_HOLE
) {
3051 struct extent_state
*cached
= NULL
;
3053 userpage
= kmap_atomic(page
);
3054 memset(userpage
+ pg_offset
, 0, iosize
);
3055 flush_dcache_page(page
);
3056 kunmap_atomic(userpage
);
3058 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3060 unlock_extent_cached(tree
, cur
,
3064 pg_offset
+= iosize
;
3067 /* the get_extent function already copied into the page */
3068 if (test_range_bit(tree
, cur
, cur_end
,
3069 EXTENT_UPTODATE
, 1, NULL
)) {
3070 check_page_uptodate(tree
, page
);
3071 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3073 pg_offset
+= iosize
;
3076 /* we have an inline extent but it didn't get marked up
3077 * to date. Error out
3079 if (block_start
== EXTENT_MAP_INLINE
) {
3081 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3083 pg_offset
+= iosize
;
3087 ret
= submit_extent_page(REQ_OP_READ
, read_flags
, tree
, NULL
,
3088 page
, sector
, disk_io_size
, pg_offset
,
3090 end_bio_extent_readpage
, mirror_num
,
3096 *bio_flags
= this_bio_flag
;
3099 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3103 pg_offset
+= iosize
;
3107 if (!PageError(page
))
3108 SetPageUptodate(page
);
3114 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3115 struct page
*pages
[], int nr_pages
,
3117 get_extent_t
*get_extent
,
3118 struct extent_map
**em_cached
,
3119 struct bio
**bio
, int mirror_num
,
3120 unsigned long *bio_flags
,
3123 struct inode
*inode
;
3124 struct btrfs_ordered_extent
*ordered
;
3127 inode
= pages
[0]->mapping
->host
;
3129 lock_extent(tree
, start
, end
);
3130 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
3134 unlock_extent(tree
, start
, end
);
3135 btrfs_start_ordered_extent(inode
, ordered
, 1);
3136 btrfs_put_ordered_extent(ordered
);
3139 for (index
= 0; index
< nr_pages
; index
++) {
3140 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3141 mirror_num
, bio_flags
, 0, prev_em_start
);
3142 put_page(pages
[index
]);
3146 static void __extent_readpages(struct extent_io_tree
*tree
,
3147 struct page
*pages
[],
3148 int nr_pages
, get_extent_t
*get_extent
,
3149 struct extent_map
**em_cached
,
3150 struct bio
**bio
, int mirror_num
,
3151 unsigned long *bio_flags
,
3158 int first_index
= 0;
3160 for (index
= 0; index
< nr_pages
; index
++) {
3161 page_start
= page_offset(pages
[index
]);
3164 end
= start
+ PAGE_SIZE
- 1;
3165 first_index
= index
;
3166 } else if (end
+ 1 == page_start
) {
3169 __do_contiguous_readpages(tree
, &pages
[first_index
],
3170 index
- first_index
, start
,
3171 end
, get_extent
, em_cached
,
3172 bio
, mirror_num
, bio_flags
,
3175 end
= start
+ PAGE_SIZE
- 1;
3176 first_index
= index
;
3181 __do_contiguous_readpages(tree
, &pages
[first_index
],
3182 index
- first_index
, start
,
3183 end
, get_extent
, em_cached
, bio
,
3184 mirror_num
, bio_flags
,
3188 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3190 get_extent_t
*get_extent
,
3191 struct bio
**bio
, int mirror_num
,
3192 unsigned long *bio_flags
, int read_flags
)
3194 struct inode
*inode
= page
->mapping
->host
;
3195 struct btrfs_ordered_extent
*ordered
;
3196 u64 start
= page_offset(page
);
3197 u64 end
= start
+ PAGE_SIZE
- 1;
3201 lock_extent(tree
, start
, end
);
3202 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
3206 unlock_extent(tree
, start
, end
);
3207 btrfs_start_ordered_extent(inode
, ordered
, 1);
3208 btrfs_put_ordered_extent(ordered
);
3211 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3212 bio_flags
, read_flags
, NULL
);
3216 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3217 get_extent_t
*get_extent
, int mirror_num
)
3219 struct bio
*bio
= NULL
;
3220 unsigned long bio_flags
= 0;
3223 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3226 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
3230 static void update_nr_written(struct writeback_control
*wbc
,
3231 unsigned long nr_written
)
3233 wbc
->nr_to_write
-= nr_written
;
3237 * helper for __extent_writepage, doing all of the delayed allocation setup.
3239 * This returns 1 if our fill_delalloc function did all the work required
3240 * to write the page (copy into inline extent). In this case the IO has
3241 * been started and the page is already unlocked.
3243 * This returns 0 if all went well (page still locked)
3244 * This returns < 0 if there were errors (page still locked)
3246 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3247 struct page
*page
, struct writeback_control
*wbc
,
3248 struct extent_page_data
*epd
,
3250 unsigned long *nr_written
)
3252 struct extent_io_tree
*tree
= epd
->tree
;
3253 u64 page_end
= delalloc_start
+ PAGE_SIZE
- 1;
3255 u64 delalloc_to_write
= 0;
3256 u64 delalloc_end
= 0;
3258 int page_started
= 0;
3260 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3263 while (delalloc_end
< page_end
) {
3264 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3268 BTRFS_MAX_EXTENT_SIZE
);
3269 if (nr_delalloc
== 0) {
3270 delalloc_start
= delalloc_end
+ 1;
3273 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3278 /* File system has been set read-only */
3281 /* fill_delalloc should be return < 0 for error
3282 * but just in case, we use > 0 here meaning the
3283 * IO is started, so we don't want to return > 0
3284 * unless things are going well.
3286 ret
= ret
< 0 ? ret
: -EIO
;
3290 * delalloc_end is already one less than the total length, so
3291 * we don't subtract one from PAGE_SIZE
3293 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3294 PAGE_SIZE
) >> PAGE_SHIFT
;
3295 delalloc_start
= delalloc_end
+ 1;
3297 if (wbc
->nr_to_write
< delalloc_to_write
) {
3300 if (delalloc_to_write
< thresh
* 2)
3301 thresh
= delalloc_to_write
;
3302 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3306 /* did the fill delalloc function already unlock and start
3311 * we've unlocked the page, so we can't update
3312 * the mapping's writeback index, just update
3315 wbc
->nr_to_write
-= *nr_written
;
3326 * helper for __extent_writepage. This calls the writepage start hooks,
3327 * and does the loop to map the page into extents and bios.
3329 * We return 1 if the IO is started and the page is unlocked,
3330 * 0 if all went well (page still locked)
3331 * < 0 if there were errors (page still locked)
3333 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3335 struct writeback_control
*wbc
,
3336 struct extent_page_data
*epd
,
3338 unsigned long nr_written
,
3339 int write_flags
, int *nr_ret
)
3341 struct extent_io_tree
*tree
= epd
->tree
;
3342 u64 start
= page_offset(page
);
3343 u64 page_end
= start
+ PAGE_SIZE
- 1;
3350 struct extent_map
*em
;
3351 struct block_device
*bdev
;
3352 size_t pg_offset
= 0;
3358 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3359 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3362 /* Fixup worker will requeue */
3364 wbc
->pages_skipped
++;
3366 redirty_page_for_writepage(wbc
, page
);
3368 update_nr_written(wbc
, nr_written
);
3375 * we don't want to touch the inode after unlocking the page,
3376 * so we update the mapping writeback index now
3378 update_nr_written(wbc
, nr_written
+ 1);
3381 if (i_size
<= start
) {
3382 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3383 tree
->ops
->writepage_end_io_hook(page
, start
,
3388 blocksize
= inode
->i_sb
->s_blocksize
;
3390 while (cur
<= end
) {
3393 if (cur
>= i_size
) {
3394 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3395 tree
->ops
->writepage_end_io_hook(page
, cur
,
3399 em
= epd
->get_extent(BTRFS_I(inode
), page
, pg_offset
, cur
,
3401 if (IS_ERR_OR_NULL(em
)) {
3403 ret
= PTR_ERR_OR_ZERO(em
);
3407 extent_offset
= cur
- em
->start
;
3408 em_end
= extent_map_end(em
);
3409 BUG_ON(em_end
<= cur
);
3411 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3412 iosize
= ALIGN(iosize
, blocksize
);
3413 sector
= (em
->block_start
+ extent_offset
) >> 9;
3415 block_start
= em
->block_start
;
3416 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3417 free_extent_map(em
);
3421 * compressed and inline extents are written through other
3424 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3425 block_start
== EXTENT_MAP_INLINE
) {
3427 * end_io notification does not happen here for
3428 * compressed extents
3430 if (!compressed
&& tree
->ops
&&
3431 tree
->ops
->writepage_end_io_hook
)
3432 tree
->ops
->writepage_end_io_hook(page
, cur
,
3435 else if (compressed
) {
3436 /* we don't want to end_page_writeback on
3437 * a compressed extent. this happens
3444 pg_offset
+= iosize
;
3448 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3449 if (!PageWriteback(page
)) {
3450 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3451 "page %lu not writeback, cur %llu end %llu",
3452 page
->index
, cur
, end
);
3455 ret
= submit_extent_page(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3456 page
, sector
, iosize
, pg_offset
,
3458 end_bio_extent_writepage
,
3462 if (PageWriteback(page
))
3463 end_page_writeback(page
);
3467 pg_offset
+= iosize
;
3476 * the writepage semantics are similar to regular writepage. extent
3477 * records are inserted to lock ranges in the tree, and as dirty areas
3478 * are found, they are marked writeback. Then the lock bits are removed
3479 * and the end_io handler clears the writeback ranges
3481 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3484 struct inode
*inode
= page
->mapping
->host
;
3485 struct extent_page_data
*epd
= data
;
3486 u64 start
= page_offset(page
);
3487 u64 page_end
= start
+ PAGE_SIZE
- 1;
3490 size_t pg_offset
= 0;
3491 loff_t i_size
= i_size_read(inode
);
3492 unsigned long end_index
= i_size
>> PAGE_SHIFT
;
3493 int write_flags
= 0;
3494 unsigned long nr_written
= 0;
3496 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3497 write_flags
= REQ_SYNC
;
3499 trace___extent_writepage(page
, inode
, wbc
);
3501 WARN_ON(!PageLocked(page
));
3503 ClearPageError(page
);
3505 pg_offset
= i_size
& (PAGE_SIZE
- 1);
3506 if (page
->index
> end_index
||
3507 (page
->index
== end_index
&& !pg_offset
)) {
3508 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3513 if (page
->index
== end_index
) {
3516 userpage
= kmap_atomic(page
);
3517 memset(userpage
+ pg_offset
, 0,
3518 PAGE_SIZE
- pg_offset
);
3519 kunmap_atomic(userpage
);
3520 flush_dcache_page(page
);
3525 set_page_extent_mapped(page
);
3527 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3533 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3534 i_size
, nr_written
, write_flags
, &nr
);
3540 /* make sure the mapping tag for page dirty gets cleared */
3541 set_page_writeback(page
);
3542 end_page_writeback(page
);
3544 if (PageError(page
)) {
3545 ret
= ret
< 0 ? ret
: -EIO
;
3546 end_extent_writepage(page
, ret
, start
, page_end
);
3555 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3557 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3558 TASK_UNINTERRUPTIBLE
);
3561 static noinline_for_stack
int
3562 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3563 struct btrfs_fs_info
*fs_info
,
3564 struct extent_page_data
*epd
)
3566 unsigned long i
, num_pages
;
3570 if (!btrfs_try_tree_write_lock(eb
)) {
3572 flush_write_bio(epd
);
3573 btrfs_tree_lock(eb
);
3576 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3577 btrfs_tree_unlock(eb
);
3581 flush_write_bio(epd
);
3585 wait_on_extent_buffer_writeback(eb
);
3586 btrfs_tree_lock(eb
);
3587 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3589 btrfs_tree_unlock(eb
);
3594 * We need to do this to prevent races in people who check if the eb is
3595 * under IO since we can end up having no IO bits set for a short period
3598 spin_lock(&eb
->refs_lock
);
3599 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3600 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3601 spin_unlock(&eb
->refs_lock
);
3602 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3603 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3605 fs_info
->dirty_metadata_batch
);
3608 spin_unlock(&eb
->refs_lock
);
3611 btrfs_tree_unlock(eb
);
3616 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3617 for (i
= 0; i
< num_pages
; i
++) {
3618 struct page
*p
= eb
->pages
[i
];
3620 if (!trylock_page(p
)) {
3622 flush_write_bio(epd
);
3632 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3634 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3635 smp_mb__after_atomic();
3636 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3639 static void set_btree_ioerr(struct page
*page
)
3641 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3644 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3648 * If writeback for a btree extent that doesn't belong to a log tree
3649 * failed, increment the counter transaction->eb_write_errors.
3650 * We do this because while the transaction is running and before it's
3651 * committing (when we call filemap_fdata[write|wait]_range against
3652 * the btree inode), we might have
3653 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3654 * returns an error or an error happens during writeback, when we're
3655 * committing the transaction we wouldn't know about it, since the pages
3656 * can be no longer dirty nor marked anymore for writeback (if a
3657 * subsequent modification to the extent buffer didn't happen before the
3658 * transaction commit), which makes filemap_fdata[write|wait]_range not
3659 * able to find the pages tagged with SetPageError at transaction
3660 * commit time. So if this happens we must abort the transaction,
3661 * otherwise we commit a super block with btree roots that point to
3662 * btree nodes/leafs whose content on disk is invalid - either garbage
3663 * or the content of some node/leaf from a past generation that got
3664 * cowed or deleted and is no longer valid.
3666 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3667 * not be enough - we need to distinguish between log tree extents vs
3668 * non-log tree extents, and the next filemap_fdatawait_range() call
3669 * will catch and clear such errors in the mapping - and that call might
3670 * be from a log sync and not from a transaction commit. Also, checking
3671 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3672 * not done and would not be reliable - the eb might have been released
3673 * from memory and reading it back again means that flag would not be
3674 * set (since it's a runtime flag, not persisted on disk).
3676 * Using the flags below in the btree inode also makes us achieve the
3677 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3678 * writeback for all dirty pages and before filemap_fdatawait_range()
3679 * is called, the writeback for all dirty pages had already finished
3680 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3681 * filemap_fdatawait_range() would return success, as it could not know
3682 * that writeback errors happened (the pages were no longer tagged for
3685 switch (eb
->log_index
) {
3687 set_bit(BTRFS_FS_BTREE_ERR
, &eb
->fs_info
->flags
);
3690 set_bit(BTRFS_FS_LOG1_ERR
, &eb
->fs_info
->flags
);
3693 set_bit(BTRFS_FS_LOG2_ERR
, &eb
->fs_info
->flags
);
3696 BUG(); /* unexpected, logic error */
3700 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3702 struct bio_vec
*bvec
;
3703 struct extent_buffer
*eb
;
3706 bio_for_each_segment_all(bvec
, bio
, i
) {
3707 struct page
*page
= bvec
->bv_page
;
3709 eb
= (struct extent_buffer
*)page
->private;
3711 done
= atomic_dec_and_test(&eb
->io_pages
);
3713 if (bio
->bi_status
||
3714 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3715 ClearPageUptodate(page
);
3716 set_btree_ioerr(page
);
3719 end_page_writeback(page
);
3724 end_extent_buffer_writeback(eb
);
3730 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3731 struct btrfs_fs_info
*fs_info
,
3732 struct writeback_control
*wbc
,
3733 struct extent_page_data
*epd
)
3735 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3736 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3737 u64 offset
= eb
->start
;
3739 unsigned long i
, num_pages
;
3740 unsigned long bio_flags
= 0;
3741 unsigned long start
, end
;
3742 int write_flags
= (epd
->sync_io
? REQ_SYNC
: 0) | REQ_META
;
3745 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3746 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3747 atomic_set(&eb
->io_pages
, num_pages
);
3748 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3749 bio_flags
= EXTENT_BIO_TREE_LOG
;
3751 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3752 nritems
= btrfs_header_nritems(eb
);
3753 if (btrfs_header_level(eb
) > 0) {
3754 end
= btrfs_node_key_ptr_offset(nritems
);
3756 memzero_extent_buffer(eb
, end
, eb
->len
- end
);
3760 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3762 start
= btrfs_item_nr_offset(nritems
);
3763 end
= btrfs_leaf_data(eb
) + leaf_data_end(fs_info
, eb
);
3764 memzero_extent_buffer(eb
, start
, end
- start
);
3767 for (i
= 0; i
< num_pages
; i
++) {
3768 struct page
*p
= eb
->pages
[i
];
3770 clear_page_dirty_for_io(p
);
3771 set_page_writeback(p
);
3772 ret
= submit_extent_page(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3773 p
, offset
>> 9, PAGE_SIZE
, 0, bdev
,
3775 end_bio_extent_buffer_writepage
,
3776 0, epd
->bio_flags
, bio_flags
, false);
3777 epd
->bio_flags
= bio_flags
;
3780 if (PageWriteback(p
))
3781 end_page_writeback(p
);
3782 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3783 end_extent_buffer_writeback(eb
);
3787 offset
+= PAGE_SIZE
;
3788 update_nr_written(wbc
, 1);
3792 if (unlikely(ret
)) {
3793 for (; i
< num_pages
; i
++) {
3794 struct page
*p
= eb
->pages
[i
];
3795 clear_page_dirty_for_io(p
);
3803 int btree_write_cache_pages(struct address_space
*mapping
,
3804 struct writeback_control
*wbc
)
3806 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3807 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3808 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3809 struct extent_page_data epd
= {
3813 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3818 int nr_to_write_done
= 0;
3819 struct pagevec pvec
;
3822 pgoff_t end
; /* Inclusive */
3826 pagevec_init(&pvec
, 0);
3827 if (wbc
->range_cyclic
) {
3828 index
= mapping
->writeback_index
; /* Start from prev offset */
3831 index
= wbc
->range_start
>> PAGE_SHIFT
;
3832 end
= wbc
->range_end
>> PAGE_SHIFT
;
3835 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3836 tag
= PAGECACHE_TAG_TOWRITE
;
3838 tag
= PAGECACHE_TAG_DIRTY
;
3840 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3841 tag_pages_for_writeback(mapping
, index
, end
);
3842 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3843 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3844 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3848 for (i
= 0; i
< nr_pages
; i
++) {
3849 struct page
*page
= pvec
.pages
[i
];
3851 if (!PagePrivate(page
))
3854 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3859 spin_lock(&mapping
->private_lock
);
3860 if (!PagePrivate(page
)) {
3861 spin_unlock(&mapping
->private_lock
);
3865 eb
= (struct extent_buffer
*)page
->private;
3868 * Shouldn't happen and normally this would be a BUG_ON
3869 * but no sense in crashing the users box for something
3870 * we can survive anyway.
3873 spin_unlock(&mapping
->private_lock
);
3877 if (eb
== prev_eb
) {
3878 spin_unlock(&mapping
->private_lock
);
3882 ret
= atomic_inc_not_zero(&eb
->refs
);
3883 spin_unlock(&mapping
->private_lock
);
3888 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3890 free_extent_buffer(eb
);
3894 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3897 free_extent_buffer(eb
);
3900 free_extent_buffer(eb
);
3903 * the filesystem may choose to bump up nr_to_write.
3904 * We have to make sure to honor the new nr_to_write
3907 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3909 pagevec_release(&pvec
);
3912 if (!scanned
&& !done
) {
3914 * We hit the last page and there is more work to be done: wrap
3915 * back to the start of the file
3921 flush_write_bio(&epd
);
3926 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3927 * @mapping: address space structure to write
3928 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3929 * @writepage: function called for each page
3930 * @data: data passed to writepage function
3932 * If a page is already under I/O, write_cache_pages() skips it, even
3933 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3934 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3935 * and msync() need to guarantee that all the data which was dirty at the time
3936 * the call was made get new I/O started against them. If wbc->sync_mode is
3937 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3938 * existing IO to complete.
3940 static int extent_write_cache_pages(struct address_space
*mapping
,
3941 struct writeback_control
*wbc
,
3942 writepage_t writepage
, void *data
,
3943 void (*flush_fn
)(void *))
3945 struct inode
*inode
= mapping
->host
;
3948 int nr_to_write_done
= 0;
3949 struct pagevec pvec
;
3952 pgoff_t end
; /* Inclusive */
3954 int range_whole
= 0;
3959 * We have to hold onto the inode so that ordered extents can do their
3960 * work when the IO finishes. The alternative to this is failing to add
3961 * an ordered extent if the igrab() fails there and that is a huge pain
3962 * to deal with, so instead just hold onto the inode throughout the
3963 * writepages operation. If it fails here we are freeing up the inode
3964 * anyway and we'd rather not waste our time writing out stuff that is
3965 * going to be truncated anyway.
3970 pagevec_init(&pvec
, 0);
3971 if (wbc
->range_cyclic
) {
3972 index
= mapping
->writeback_index
; /* Start from prev offset */
3975 index
= wbc
->range_start
>> PAGE_SHIFT
;
3976 end
= wbc
->range_end
>> PAGE_SHIFT
;
3977 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3981 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3982 tag
= PAGECACHE_TAG_TOWRITE
;
3984 tag
= PAGECACHE_TAG_DIRTY
;
3986 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3987 tag_pages_for_writeback(mapping
, index
, end
);
3989 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3990 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3991 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3995 for (i
= 0; i
< nr_pages
; i
++) {
3996 struct page
*page
= pvec
.pages
[i
];
3998 done_index
= page
->index
;
4000 * At this point we hold neither mapping->tree_lock nor
4001 * lock on the page itself: the page may be truncated or
4002 * invalidated (changing page->mapping to NULL), or even
4003 * swizzled back from swapper_space to tmpfs file
4006 if (!trylock_page(page
)) {
4011 if (unlikely(page
->mapping
!= mapping
)) {
4016 if (!wbc
->range_cyclic
&& page
->index
> end
) {
4022 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4023 if (PageWriteback(page
))
4025 wait_on_page_writeback(page
);
4028 if (PageWriteback(page
) ||
4029 !clear_page_dirty_for_io(page
)) {
4034 ret
= (*writepage
)(page
, wbc
, data
);
4036 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4042 * done_index is set past this page,
4043 * so media errors will not choke
4044 * background writeout for the entire
4045 * file. This has consequences for
4046 * range_cyclic semantics (ie. it may
4047 * not be suitable for data integrity
4050 done_index
= page
->index
+ 1;
4056 * the filesystem may choose to bump up nr_to_write.
4057 * We have to make sure to honor the new nr_to_write
4060 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4062 pagevec_release(&pvec
);
4065 if (!scanned
&& !done
) {
4067 * We hit the last page and there is more work to be done: wrap
4068 * back to the start of the file
4075 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4076 mapping
->writeback_index
= done_index
;
4078 btrfs_add_delayed_iput(inode
);
4082 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4087 bio_set_op_attrs(epd
->bio
, REQ_OP_WRITE
,
4088 epd
->sync_io
? REQ_SYNC
: 0);
4090 ret
= submit_one_bio(epd
->bio
, 0, epd
->bio_flags
);
4091 BUG_ON(ret
< 0); /* -ENOMEM */
4096 static noinline
void flush_write_bio(void *data
)
4098 struct extent_page_data
*epd
= data
;
4099 flush_epd_write_bio(epd
);
4102 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4103 get_extent_t
*get_extent
,
4104 struct writeback_control
*wbc
)
4107 struct extent_page_data epd
= {
4110 .get_extent
= get_extent
,
4112 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4116 ret
= __extent_writepage(page
, wbc
, &epd
);
4118 flush_epd_write_bio(&epd
);
4122 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4123 u64 start
, u64 end
, get_extent_t
*get_extent
,
4127 struct address_space
*mapping
= inode
->i_mapping
;
4129 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4132 struct extent_page_data epd
= {
4135 .get_extent
= get_extent
,
4137 .sync_io
= mode
== WB_SYNC_ALL
,
4140 struct writeback_control wbc_writepages
= {
4142 .nr_to_write
= nr_pages
* 2,
4143 .range_start
= start
,
4144 .range_end
= end
+ 1,
4147 while (start
<= end
) {
4148 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4149 if (clear_page_dirty_for_io(page
))
4150 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4152 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4153 tree
->ops
->writepage_end_io_hook(page
, start
,
4154 start
+ PAGE_SIZE
- 1,
4162 flush_epd_write_bio(&epd
);
4166 int extent_writepages(struct extent_io_tree
*tree
,
4167 struct address_space
*mapping
,
4168 get_extent_t
*get_extent
,
4169 struct writeback_control
*wbc
)
4172 struct extent_page_data epd
= {
4175 .get_extent
= get_extent
,
4177 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4181 ret
= extent_write_cache_pages(mapping
, wbc
, __extent_writepage
, &epd
,
4183 flush_epd_write_bio(&epd
);
4187 int extent_readpages(struct extent_io_tree
*tree
,
4188 struct address_space
*mapping
,
4189 struct list_head
*pages
, unsigned nr_pages
,
4190 get_extent_t get_extent
)
4192 struct bio
*bio
= NULL
;
4194 unsigned long bio_flags
= 0;
4195 struct page
*pagepool
[16];
4197 struct extent_map
*em_cached
= NULL
;
4199 u64 prev_em_start
= (u64
)-1;
4201 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4202 page
= list_entry(pages
->prev
, struct page
, lru
);
4204 prefetchw(&page
->flags
);
4205 list_del(&page
->lru
);
4206 if (add_to_page_cache_lru(page
, mapping
,
4208 readahead_gfp_mask(mapping
))) {
4213 pagepool
[nr
++] = page
;
4214 if (nr
< ARRAY_SIZE(pagepool
))
4216 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4217 &bio
, 0, &bio_flags
, &prev_em_start
);
4221 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4222 &bio
, 0, &bio_flags
, &prev_em_start
);
4225 free_extent_map(em_cached
);
4227 BUG_ON(!list_empty(pages
));
4229 return submit_one_bio(bio
, 0, bio_flags
);
4234 * basic invalidatepage code, this waits on any locked or writeback
4235 * ranges corresponding to the page, and then deletes any extent state
4236 * records from the tree
4238 int extent_invalidatepage(struct extent_io_tree
*tree
,
4239 struct page
*page
, unsigned long offset
)
4241 struct extent_state
*cached_state
= NULL
;
4242 u64 start
= page_offset(page
);
4243 u64 end
= start
+ PAGE_SIZE
- 1;
4244 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4246 start
+= ALIGN(offset
, blocksize
);
4250 lock_extent_bits(tree
, start
, end
, &cached_state
);
4251 wait_on_page_writeback(page
);
4252 clear_extent_bit(tree
, start
, end
,
4253 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4254 EXTENT_DO_ACCOUNTING
,
4255 1, 1, &cached_state
, GFP_NOFS
);
4260 * a helper for releasepage, this tests for areas of the page that
4261 * are locked or under IO and drops the related state bits if it is safe
4264 static int try_release_extent_state(struct extent_map_tree
*map
,
4265 struct extent_io_tree
*tree
,
4266 struct page
*page
, gfp_t mask
)
4268 u64 start
= page_offset(page
);
4269 u64 end
= start
+ PAGE_SIZE
- 1;
4272 if (test_range_bit(tree
, start
, end
,
4273 EXTENT_IOBITS
, 0, NULL
))
4277 * at this point we can safely clear everything except the
4278 * locked bit and the nodatasum bit
4280 ret
= clear_extent_bit(tree
, start
, end
,
4281 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4284 /* if clear_extent_bit failed for enomem reasons,
4285 * we can't allow the release to continue.
4296 * a helper for releasepage. As long as there are no locked extents
4297 * in the range corresponding to the page, both state records and extent
4298 * map records are removed
4300 int try_release_extent_mapping(struct extent_map_tree
*map
,
4301 struct extent_io_tree
*tree
, struct page
*page
,
4304 struct extent_map
*em
;
4305 u64 start
= page_offset(page
);
4306 u64 end
= start
+ PAGE_SIZE
- 1;
4308 if (gfpflags_allow_blocking(mask
) &&
4309 page
->mapping
->host
->i_size
> SZ_16M
) {
4311 while (start
<= end
) {
4312 len
= end
- start
+ 1;
4313 write_lock(&map
->lock
);
4314 em
= lookup_extent_mapping(map
, start
, len
);
4316 write_unlock(&map
->lock
);
4319 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4320 em
->start
!= start
) {
4321 write_unlock(&map
->lock
);
4322 free_extent_map(em
);
4325 if (!test_range_bit(tree
, em
->start
,
4326 extent_map_end(em
) - 1,
4327 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4329 remove_extent_mapping(map
, em
);
4330 /* once for the rb tree */
4331 free_extent_map(em
);
4333 start
= extent_map_end(em
);
4334 write_unlock(&map
->lock
);
4337 free_extent_map(em
);
4340 return try_release_extent_state(map
, tree
, page
, mask
);
4344 * helper function for fiemap, which doesn't want to see any holes.
4345 * This maps until we find something past 'last'
4347 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4350 get_extent_t
*get_extent
)
4352 u64 sectorsize
= btrfs_inode_sectorsize(inode
);
4353 struct extent_map
*em
;
4360 len
= last
- offset
;
4363 len
= ALIGN(len
, sectorsize
);
4364 em
= get_extent(BTRFS_I(inode
), NULL
, 0, offset
, len
, 0);
4365 if (IS_ERR_OR_NULL(em
))
4368 /* if this isn't a hole return it */
4369 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4370 em
->block_start
!= EXTENT_MAP_HOLE
) {
4374 /* this is a hole, advance to the next extent */
4375 offset
= extent_map_end(em
);
4376 free_extent_map(em
);
4383 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4384 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4388 u64 max
= start
+ len
;
4392 u64 last_for_get_extent
= 0;
4394 u64 isize
= i_size_read(inode
);
4395 struct btrfs_key found_key
;
4396 struct extent_map
*em
= NULL
;
4397 struct extent_state
*cached_state
= NULL
;
4398 struct btrfs_path
*path
;
4399 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4408 path
= btrfs_alloc_path();
4411 path
->leave_spinning
= 1;
4413 start
= round_down(start
, btrfs_inode_sectorsize(inode
));
4414 len
= round_up(max
, btrfs_inode_sectorsize(inode
)) - start
;
4417 * lookup the last file extent. We're not using i_size here
4418 * because there might be preallocation past i_size
4420 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
4421 btrfs_ino(BTRFS_I(inode
)), -1, 0);
4423 btrfs_free_path(path
);
4432 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4433 found_type
= found_key
.type
;
4435 /* No extents, but there might be delalloc bits */
4436 if (found_key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
4437 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4438 /* have to trust i_size as the end */
4440 last_for_get_extent
= isize
;
4443 * remember the start of the last extent. There are a
4444 * bunch of different factors that go into the length of the
4445 * extent, so its much less complex to remember where it started
4447 last
= found_key
.offset
;
4448 last_for_get_extent
= last
+ 1;
4450 btrfs_release_path(path
);
4453 * we might have some extents allocated but more delalloc past those
4454 * extents. so, we trust isize unless the start of the last extent is
4459 last_for_get_extent
= isize
;
4462 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4465 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4475 u64 offset_in_extent
= 0;
4477 /* break if the extent we found is outside the range */
4478 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4482 * get_extent may return an extent that starts before our
4483 * requested range. We have to make sure the ranges
4484 * we return to fiemap always move forward and don't
4485 * overlap, so adjust the offsets here
4487 em_start
= max(em
->start
, off
);
4490 * record the offset from the start of the extent
4491 * for adjusting the disk offset below. Only do this if the
4492 * extent isn't compressed since our in ram offset may be past
4493 * what we have actually allocated on disk.
4495 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4496 offset_in_extent
= em_start
- em
->start
;
4497 em_end
= extent_map_end(em
);
4498 em_len
= em_end
- em_start
;
4503 * bump off for our next call to get_extent
4505 off
= extent_map_end(em
);
4509 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4511 flags
|= FIEMAP_EXTENT_LAST
;
4512 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4513 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4514 FIEMAP_EXTENT_NOT_ALIGNED
);
4515 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4516 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4517 FIEMAP_EXTENT_UNKNOWN
);
4518 } else if (fieinfo
->fi_extents_max
) {
4519 struct btrfs_trans_handle
*trans
;
4521 u64 bytenr
= em
->block_start
-
4522 (em
->start
- em
->orig_start
);
4524 disko
= em
->block_start
+ offset_in_extent
;
4527 * We need a trans handle to get delayed refs
4529 trans
= btrfs_join_transaction(root
);
4531 * It's OK if we can't start a trans we can still check
4538 * As btrfs supports shared space, this information
4539 * can be exported to userspace tools via
4540 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4541 * then we're just getting a count and we can skip the
4544 ret
= btrfs_check_shared(trans
, root
->fs_info
,
4546 btrfs_ino(BTRFS_I(inode
)), bytenr
);
4548 btrfs_end_transaction(trans
);
4552 flags
|= FIEMAP_EXTENT_SHARED
;
4555 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4556 flags
|= FIEMAP_EXTENT_ENCODED
;
4557 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4558 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4560 free_extent_map(em
);
4562 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4563 (last
== (u64
)-1 && isize
<= em_end
)) {
4564 flags
|= FIEMAP_EXTENT_LAST
;
4568 /* now scan forward to see if this is really the last extent. */
4569 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4576 flags
|= FIEMAP_EXTENT_LAST
;
4579 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4588 free_extent_map(em
);
4590 btrfs_free_path(path
);
4591 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4592 &cached_state
, GFP_NOFS
);
4596 static void __free_extent_buffer(struct extent_buffer
*eb
)
4598 btrfs_leak_debug_del(&eb
->leak_list
);
4599 kmem_cache_free(extent_buffer_cache
, eb
);
4602 int extent_buffer_under_io(struct extent_buffer
*eb
)
4604 return (atomic_read(&eb
->io_pages
) ||
4605 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4606 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4610 * Helper for releasing extent buffer page.
4612 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4614 unsigned long index
;
4616 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4618 BUG_ON(extent_buffer_under_io(eb
));
4620 index
= num_extent_pages(eb
->start
, eb
->len
);
4626 page
= eb
->pages
[index
];
4630 spin_lock(&page
->mapping
->private_lock
);
4632 * We do this since we'll remove the pages after we've
4633 * removed the eb from the radix tree, so we could race
4634 * and have this page now attached to the new eb. So
4635 * only clear page_private if it's still connected to
4638 if (PagePrivate(page
) &&
4639 page
->private == (unsigned long)eb
) {
4640 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4641 BUG_ON(PageDirty(page
));
4642 BUG_ON(PageWriteback(page
));
4644 * We need to make sure we haven't be attached
4647 ClearPagePrivate(page
);
4648 set_page_private(page
, 0);
4649 /* One for the page private */
4654 spin_unlock(&page
->mapping
->private_lock
);
4656 /* One for when we allocated the page */
4658 } while (index
!= 0);
4662 * Helper for releasing the extent buffer.
4664 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4666 btrfs_release_extent_buffer_page(eb
);
4667 __free_extent_buffer(eb
);
4670 static struct extent_buffer
*
4671 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4674 struct extent_buffer
*eb
= NULL
;
4676 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4679 eb
->fs_info
= fs_info
;
4681 rwlock_init(&eb
->lock
);
4682 atomic_set(&eb
->write_locks
, 0);
4683 atomic_set(&eb
->read_locks
, 0);
4684 atomic_set(&eb
->blocking_readers
, 0);
4685 atomic_set(&eb
->blocking_writers
, 0);
4686 atomic_set(&eb
->spinning_readers
, 0);
4687 atomic_set(&eb
->spinning_writers
, 0);
4688 eb
->lock_nested
= 0;
4689 init_waitqueue_head(&eb
->write_lock_wq
);
4690 init_waitqueue_head(&eb
->read_lock_wq
);
4692 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4694 spin_lock_init(&eb
->refs_lock
);
4695 atomic_set(&eb
->refs
, 1);
4696 atomic_set(&eb
->io_pages
, 0);
4699 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4701 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4702 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4703 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4708 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4712 struct extent_buffer
*new;
4713 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4715 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4719 for (i
= 0; i
< num_pages
; i
++) {
4720 p
= alloc_page(GFP_NOFS
);
4722 btrfs_release_extent_buffer(new);
4725 attach_extent_buffer_page(new, p
);
4726 WARN_ON(PageDirty(p
));
4729 copy_page(page_address(p
), page_address(src
->pages
[i
]));
4732 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4733 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4738 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4739 u64 start
, unsigned long len
)
4741 struct extent_buffer
*eb
;
4742 unsigned long num_pages
;
4745 num_pages
= num_extent_pages(start
, len
);
4747 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4751 for (i
= 0; i
< num_pages
; i
++) {
4752 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4756 set_extent_buffer_uptodate(eb
);
4757 btrfs_set_header_nritems(eb
, 0);
4758 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4763 __free_page(eb
->pages
[i
- 1]);
4764 __free_extent_buffer(eb
);
4768 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4771 return __alloc_dummy_extent_buffer(fs_info
, start
, fs_info
->nodesize
);
4774 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4777 /* the ref bit is tricky. We have to make sure it is set
4778 * if we have the buffer dirty. Otherwise the
4779 * code to free a buffer can end up dropping a dirty
4782 * Once the ref bit is set, it won't go away while the
4783 * buffer is dirty or in writeback, and it also won't
4784 * go away while we have the reference count on the
4787 * We can't just set the ref bit without bumping the
4788 * ref on the eb because free_extent_buffer might
4789 * see the ref bit and try to clear it. If this happens
4790 * free_extent_buffer might end up dropping our original
4791 * ref by mistake and freeing the page before we are able
4792 * to add one more ref.
4794 * So bump the ref count first, then set the bit. If someone
4795 * beat us to it, drop the ref we added.
4797 refs
= atomic_read(&eb
->refs
);
4798 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4801 spin_lock(&eb
->refs_lock
);
4802 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4803 atomic_inc(&eb
->refs
);
4804 spin_unlock(&eb
->refs_lock
);
4807 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4808 struct page
*accessed
)
4810 unsigned long num_pages
, i
;
4812 check_buffer_tree_ref(eb
);
4814 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4815 for (i
= 0; i
< num_pages
; i
++) {
4816 struct page
*p
= eb
->pages
[i
];
4819 mark_page_accessed(p
);
4823 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4826 struct extent_buffer
*eb
;
4829 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4830 start
>> PAGE_SHIFT
);
4831 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4834 * Lock our eb's refs_lock to avoid races with
4835 * free_extent_buffer. When we get our eb it might be flagged
4836 * with EXTENT_BUFFER_STALE and another task running
4837 * free_extent_buffer might have seen that flag set,
4838 * eb->refs == 2, that the buffer isn't under IO (dirty and
4839 * writeback flags not set) and it's still in the tree (flag
4840 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4841 * of decrementing the extent buffer's reference count twice.
4842 * So here we could race and increment the eb's reference count,
4843 * clear its stale flag, mark it as dirty and drop our reference
4844 * before the other task finishes executing free_extent_buffer,
4845 * which would later result in an attempt to free an extent
4846 * buffer that is dirty.
4848 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4849 spin_lock(&eb
->refs_lock
);
4850 spin_unlock(&eb
->refs_lock
);
4852 mark_extent_buffer_accessed(eb
, NULL
);
4860 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4861 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4864 struct extent_buffer
*eb
, *exists
= NULL
;
4867 eb
= find_extent_buffer(fs_info
, start
);
4870 eb
= alloc_dummy_extent_buffer(fs_info
, start
);
4873 eb
->fs_info
= fs_info
;
4875 ret
= radix_tree_preload(GFP_NOFS
);
4878 spin_lock(&fs_info
->buffer_lock
);
4879 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4880 start
>> PAGE_SHIFT
, eb
);
4881 spin_unlock(&fs_info
->buffer_lock
);
4882 radix_tree_preload_end();
4883 if (ret
== -EEXIST
) {
4884 exists
= find_extent_buffer(fs_info
, start
);
4890 check_buffer_tree_ref(eb
);
4891 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4894 * We will free dummy extent buffer's if they come into
4895 * free_extent_buffer with a ref count of 2, but if we are using this we
4896 * want the buffers to stay in memory until we're done with them, so
4897 * bump the ref count again.
4899 atomic_inc(&eb
->refs
);
4902 btrfs_release_extent_buffer(eb
);
4907 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4910 unsigned long len
= fs_info
->nodesize
;
4911 unsigned long num_pages
= num_extent_pages(start
, len
);
4913 unsigned long index
= start
>> PAGE_SHIFT
;
4914 struct extent_buffer
*eb
;
4915 struct extent_buffer
*exists
= NULL
;
4917 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4921 if (!IS_ALIGNED(start
, fs_info
->sectorsize
)) {
4922 btrfs_err(fs_info
, "bad tree block start %llu", start
);
4923 return ERR_PTR(-EINVAL
);
4926 eb
= find_extent_buffer(fs_info
, start
);
4930 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4932 return ERR_PTR(-ENOMEM
);
4934 for (i
= 0; i
< num_pages
; i
++, index
++) {
4935 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4937 exists
= ERR_PTR(-ENOMEM
);
4941 spin_lock(&mapping
->private_lock
);
4942 if (PagePrivate(p
)) {
4944 * We could have already allocated an eb for this page
4945 * and attached one so lets see if we can get a ref on
4946 * the existing eb, and if we can we know it's good and
4947 * we can just return that one, else we know we can just
4948 * overwrite page->private.
4950 exists
= (struct extent_buffer
*)p
->private;
4951 if (atomic_inc_not_zero(&exists
->refs
)) {
4952 spin_unlock(&mapping
->private_lock
);
4955 mark_extent_buffer_accessed(exists
, p
);
4961 * Do this so attach doesn't complain and we need to
4962 * drop the ref the old guy had.
4964 ClearPagePrivate(p
);
4965 WARN_ON(PageDirty(p
));
4968 attach_extent_buffer_page(eb
, p
);
4969 spin_unlock(&mapping
->private_lock
);
4970 WARN_ON(PageDirty(p
));
4972 if (!PageUptodate(p
))
4976 * see below about how we avoid a nasty race with release page
4977 * and why we unlock later
4981 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4983 ret
= radix_tree_preload(GFP_NOFS
);
4985 exists
= ERR_PTR(ret
);
4989 spin_lock(&fs_info
->buffer_lock
);
4990 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4991 start
>> PAGE_SHIFT
, eb
);
4992 spin_unlock(&fs_info
->buffer_lock
);
4993 radix_tree_preload_end();
4994 if (ret
== -EEXIST
) {
4995 exists
= find_extent_buffer(fs_info
, start
);
5001 /* add one reference for the tree */
5002 check_buffer_tree_ref(eb
);
5003 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
5006 * there is a race where release page may have
5007 * tried to find this extent buffer in the radix
5008 * but failed. It will tell the VM it is safe to
5009 * reclaim the, and it will clear the page private bit.
5010 * We must make sure to set the page private bit properly
5011 * after the extent buffer is in the radix tree so
5012 * it doesn't get lost
5014 SetPageChecked(eb
->pages
[0]);
5015 for (i
= 1; i
< num_pages
; i
++) {
5017 ClearPageChecked(p
);
5020 unlock_page(eb
->pages
[0]);
5024 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
5025 for (i
= 0; i
< num_pages
; i
++) {
5027 unlock_page(eb
->pages
[i
]);
5030 btrfs_release_extent_buffer(eb
);
5034 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
5036 struct extent_buffer
*eb
=
5037 container_of(head
, struct extent_buffer
, rcu_head
);
5039 __free_extent_buffer(eb
);
5042 /* Expects to have eb->eb_lock already held */
5043 static int release_extent_buffer(struct extent_buffer
*eb
)
5045 WARN_ON(atomic_read(&eb
->refs
) == 0);
5046 if (atomic_dec_and_test(&eb
->refs
)) {
5047 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5048 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5050 spin_unlock(&eb
->refs_lock
);
5052 spin_lock(&fs_info
->buffer_lock
);
5053 radix_tree_delete(&fs_info
->buffer_radix
,
5054 eb
->start
>> PAGE_SHIFT
);
5055 spin_unlock(&fs_info
->buffer_lock
);
5057 spin_unlock(&eb
->refs_lock
);
5060 /* Should be safe to release our pages at this point */
5061 btrfs_release_extent_buffer_page(eb
);
5062 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5063 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5064 __free_extent_buffer(eb
);
5068 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5071 spin_unlock(&eb
->refs_lock
);
5076 void free_extent_buffer(struct extent_buffer
*eb
)
5084 refs
= atomic_read(&eb
->refs
);
5087 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5092 spin_lock(&eb
->refs_lock
);
5093 if (atomic_read(&eb
->refs
) == 2 &&
5094 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5095 atomic_dec(&eb
->refs
);
5097 if (atomic_read(&eb
->refs
) == 2 &&
5098 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5099 !extent_buffer_under_io(eb
) &&
5100 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5101 atomic_dec(&eb
->refs
);
5104 * I know this is terrible, but it's temporary until we stop tracking
5105 * the uptodate bits and such for the extent buffers.
5107 release_extent_buffer(eb
);
5110 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5115 spin_lock(&eb
->refs_lock
);
5116 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5118 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5119 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5120 atomic_dec(&eb
->refs
);
5121 release_extent_buffer(eb
);
5124 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5127 unsigned long num_pages
;
5130 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5132 for (i
= 0; i
< num_pages
; i
++) {
5133 page
= eb
->pages
[i
];
5134 if (!PageDirty(page
))
5138 WARN_ON(!PagePrivate(page
));
5140 clear_page_dirty_for_io(page
);
5141 spin_lock_irq(&page
->mapping
->tree_lock
);
5142 if (!PageDirty(page
)) {
5143 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5145 PAGECACHE_TAG_DIRTY
);
5147 spin_unlock_irq(&page
->mapping
->tree_lock
);
5148 ClearPageError(page
);
5151 WARN_ON(atomic_read(&eb
->refs
) == 0);
5154 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5157 unsigned long num_pages
;
5160 check_buffer_tree_ref(eb
);
5162 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5164 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5165 WARN_ON(atomic_read(&eb
->refs
) == 0);
5166 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5168 for (i
= 0; i
< num_pages
; i
++)
5169 set_page_dirty(eb
->pages
[i
]);
5173 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5177 unsigned long num_pages
;
5179 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5180 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5181 for (i
= 0; i
< num_pages
; i
++) {
5182 page
= eb
->pages
[i
];
5184 ClearPageUptodate(page
);
5188 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5192 unsigned long num_pages
;
5194 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5195 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5196 for (i
= 0; i
< num_pages
; i
++) {
5197 page
= eb
->pages
[i
];
5198 SetPageUptodate(page
);
5202 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5204 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5207 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5208 struct extent_buffer
*eb
, int wait
,
5209 get_extent_t
*get_extent
, int mirror_num
)
5215 int locked_pages
= 0;
5216 int all_uptodate
= 1;
5217 unsigned long num_pages
;
5218 unsigned long num_reads
= 0;
5219 struct bio
*bio
= NULL
;
5220 unsigned long bio_flags
= 0;
5222 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5225 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5226 for (i
= 0; i
< num_pages
; i
++) {
5227 page
= eb
->pages
[i
];
5228 if (wait
== WAIT_NONE
) {
5229 if (!trylock_page(page
))
5237 * We need to firstly lock all pages to make sure that
5238 * the uptodate bit of our pages won't be affected by
5239 * clear_extent_buffer_uptodate().
5241 for (i
= 0; i
< num_pages
; i
++) {
5242 page
= eb
->pages
[i
];
5243 if (!PageUptodate(page
)) {
5250 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5254 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5255 eb
->read_mirror
= 0;
5256 atomic_set(&eb
->io_pages
, num_reads
);
5257 for (i
= 0; i
< num_pages
; i
++) {
5258 page
= eb
->pages
[i
];
5260 if (!PageUptodate(page
)) {
5262 atomic_dec(&eb
->io_pages
);
5267 ClearPageError(page
);
5268 err
= __extent_read_full_page(tree
, page
,
5270 mirror_num
, &bio_flags
,
5275 * We use &bio in above __extent_read_full_page,
5276 * so we ensure that if it returns error, the
5277 * current page fails to add itself to bio and
5278 * it's been unlocked.
5280 * We must dec io_pages by ourselves.
5282 atomic_dec(&eb
->io_pages
);
5290 err
= submit_one_bio(bio
, mirror_num
, bio_flags
);
5295 if (ret
|| wait
!= WAIT_COMPLETE
)
5298 for (i
= 0; i
< num_pages
; i
++) {
5299 page
= eb
->pages
[i
];
5300 wait_on_page_locked(page
);
5301 if (!PageUptodate(page
))
5308 while (locked_pages
> 0) {
5310 page
= eb
->pages
[locked_pages
];
5316 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5317 unsigned long start
,
5324 char *dst
= (char *)dstv
;
5325 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5326 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5328 WARN_ON(start
> eb
->len
);
5329 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5331 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5334 page
= eb
->pages
[i
];
5336 cur
= min(len
, (PAGE_SIZE
- offset
));
5337 kaddr
= page_address(page
);
5338 memcpy(dst
, kaddr
+ offset
, cur
);
5347 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5348 unsigned long start
,
5355 char __user
*dst
= (char __user
*)dstv
;
5356 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5357 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5360 WARN_ON(start
> eb
->len
);
5361 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5363 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5366 page
= eb
->pages
[i
];
5368 cur
= min(len
, (PAGE_SIZE
- offset
));
5369 kaddr
= page_address(page
);
5370 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5385 * return 0 if the item is found within a page.
5386 * return 1 if the item spans two pages.
5387 * return -EINVAL otherwise.
5389 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5390 unsigned long min_len
, char **map
,
5391 unsigned long *map_start
,
5392 unsigned long *map_len
)
5394 size_t offset
= start
& (PAGE_SIZE
- 1);
5397 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5398 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5399 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5406 offset
= start_offset
;
5410 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5413 if (start
+ min_len
> eb
->len
) {
5414 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5415 eb
->start
, eb
->len
, start
, min_len
);
5420 kaddr
= page_address(p
);
5421 *map
= kaddr
+ offset
;
5422 *map_len
= PAGE_SIZE
- offset
;
5426 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5427 unsigned long start
,
5434 char *ptr
= (char *)ptrv
;
5435 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5436 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5439 WARN_ON(start
> eb
->len
);
5440 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5442 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5445 page
= eb
->pages
[i
];
5447 cur
= min(len
, (PAGE_SIZE
- offset
));
5449 kaddr
= page_address(page
);
5450 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5462 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer
*eb
,
5467 WARN_ON(!PageUptodate(eb
->pages
[0]));
5468 kaddr
= page_address(eb
->pages
[0]);
5469 memcpy(kaddr
+ offsetof(struct btrfs_header
, chunk_tree_uuid
), srcv
,
5473 void write_extent_buffer_fsid(struct extent_buffer
*eb
, const void *srcv
)
5477 WARN_ON(!PageUptodate(eb
->pages
[0]));
5478 kaddr
= page_address(eb
->pages
[0]);
5479 memcpy(kaddr
+ offsetof(struct btrfs_header
, fsid
), srcv
,
5483 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5484 unsigned long start
, unsigned long len
)
5490 char *src
= (char *)srcv
;
5491 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5492 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5494 WARN_ON(start
> eb
->len
);
5495 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5497 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5500 page
= eb
->pages
[i
];
5501 WARN_ON(!PageUptodate(page
));
5503 cur
= min(len
, PAGE_SIZE
- offset
);
5504 kaddr
= page_address(page
);
5505 memcpy(kaddr
+ offset
, src
, cur
);
5514 void memzero_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5521 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5522 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5524 WARN_ON(start
> eb
->len
);
5525 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5527 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5530 page
= eb
->pages
[i
];
5531 WARN_ON(!PageUptodate(page
));
5533 cur
= min(len
, PAGE_SIZE
- offset
);
5534 kaddr
= page_address(page
);
5535 memset(kaddr
+ offset
, 0, cur
);
5543 void copy_extent_buffer_full(struct extent_buffer
*dst
,
5544 struct extent_buffer
*src
)
5549 ASSERT(dst
->len
== src
->len
);
5551 num_pages
= num_extent_pages(dst
->start
, dst
->len
);
5552 for (i
= 0; i
< num_pages
; i
++)
5553 copy_page(page_address(dst
->pages
[i
]),
5554 page_address(src
->pages
[i
]));
5557 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5558 unsigned long dst_offset
, unsigned long src_offset
,
5561 u64 dst_len
= dst
->len
;
5566 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5567 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5569 WARN_ON(src
->len
!= dst_len
);
5571 offset
= (start_offset
+ dst_offset
) &
5575 page
= dst
->pages
[i
];
5576 WARN_ON(!PageUptodate(page
));
5578 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5580 kaddr
= page_address(page
);
5581 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5590 void le_bitmap_set(u8
*map
, unsigned int start
, int len
)
5592 u8
*p
= map
+ BIT_BYTE(start
);
5593 const unsigned int size
= start
+ len
;
5594 int bits_to_set
= BITS_PER_BYTE
- (start
% BITS_PER_BYTE
);
5595 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(start
);
5597 while (len
- bits_to_set
>= 0) {
5600 bits_to_set
= BITS_PER_BYTE
;
5605 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5610 void le_bitmap_clear(u8
*map
, unsigned int start
, int len
)
5612 u8
*p
= map
+ BIT_BYTE(start
);
5613 const unsigned int size
= start
+ len
;
5614 int bits_to_clear
= BITS_PER_BYTE
- (start
% BITS_PER_BYTE
);
5615 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(start
);
5617 while (len
- bits_to_clear
>= 0) {
5618 *p
&= ~mask_to_clear
;
5619 len
-= bits_to_clear
;
5620 bits_to_clear
= BITS_PER_BYTE
;
5625 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5626 *p
&= ~mask_to_clear
;
5631 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5633 * @eb: the extent buffer
5634 * @start: offset of the bitmap item in the extent buffer
5636 * @page_index: return index of the page in the extent buffer that contains the
5638 * @page_offset: return offset into the page given by page_index
5640 * This helper hides the ugliness of finding the byte in an extent buffer which
5641 * contains a given bit.
5643 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5644 unsigned long start
, unsigned long nr
,
5645 unsigned long *page_index
,
5646 size_t *page_offset
)
5648 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5649 size_t byte_offset
= BIT_BYTE(nr
);
5653 * The byte we want is the offset of the extent buffer + the offset of
5654 * the bitmap item in the extent buffer + the offset of the byte in the
5657 offset
= start_offset
+ start
+ byte_offset
;
5659 *page_index
= offset
>> PAGE_SHIFT
;
5660 *page_offset
= offset
& (PAGE_SIZE
- 1);
5664 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5665 * @eb: the extent buffer
5666 * @start: offset of the bitmap item in the extent buffer
5667 * @nr: bit number to test
5669 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5677 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5678 page
= eb
->pages
[i
];
5679 WARN_ON(!PageUptodate(page
));
5680 kaddr
= page_address(page
);
5681 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5685 * extent_buffer_bitmap_set - set an area of a bitmap
5686 * @eb: the extent buffer
5687 * @start: offset of the bitmap item in the extent buffer
5688 * @pos: bit number of the first bit
5689 * @len: number of bits to set
5691 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5692 unsigned long pos
, unsigned long len
)
5698 const unsigned int size
= pos
+ len
;
5699 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5700 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5702 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5703 page
= eb
->pages
[i
];
5704 WARN_ON(!PageUptodate(page
));
5705 kaddr
= page_address(page
);
5707 while (len
>= bits_to_set
) {
5708 kaddr
[offset
] |= mask_to_set
;
5710 bits_to_set
= BITS_PER_BYTE
;
5712 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5714 page
= eb
->pages
[++i
];
5715 WARN_ON(!PageUptodate(page
));
5716 kaddr
= page_address(page
);
5720 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5721 kaddr
[offset
] |= mask_to_set
;
5727 * extent_buffer_bitmap_clear - clear an area of a bitmap
5728 * @eb: the extent buffer
5729 * @start: offset of the bitmap item in the extent buffer
5730 * @pos: bit number of the first bit
5731 * @len: number of bits to clear
5733 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5734 unsigned long pos
, unsigned long len
)
5740 const unsigned int size
= pos
+ len
;
5741 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5742 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5744 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5745 page
= eb
->pages
[i
];
5746 WARN_ON(!PageUptodate(page
));
5747 kaddr
= page_address(page
);
5749 while (len
>= bits_to_clear
) {
5750 kaddr
[offset
] &= ~mask_to_clear
;
5751 len
-= bits_to_clear
;
5752 bits_to_clear
= BITS_PER_BYTE
;
5754 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5756 page
= eb
->pages
[++i
];
5757 WARN_ON(!PageUptodate(page
));
5758 kaddr
= page_address(page
);
5762 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5763 kaddr
[offset
] &= ~mask_to_clear
;
5767 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5769 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5770 return distance
< len
;
5773 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5774 unsigned long dst_off
, unsigned long src_off
,
5777 char *dst_kaddr
= page_address(dst_page
);
5779 int must_memmove
= 0;
5781 if (dst_page
!= src_page
) {
5782 src_kaddr
= page_address(src_page
);
5784 src_kaddr
= dst_kaddr
;
5785 if (areas_overlap(src_off
, dst_off
, len
))
5790 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5792 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5795 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5796 unsigned long src_offset
, unsigned long len
)
5798 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5800 size_t dst_off_in_page
;
5801 size_t src_off_in_page
;
5802 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5803 unsigned long dst_i
;
5804 unsigned long src_i
;
5806 if (src_offset
+ len
> dst
->len
) {
5808 "memmove bogus src_offset %lu move len %lu dst len %lu",
5809 src_offset
, len
, dst
->len
);
5812 if (dst_offset
+ len
> dst
->len
) {
5814 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5815 dst_offset
, len
, dst
->len
);
5820 dst_off_in_page
= (start_offset
+ dst_offset
) &
5822 src_off_in_page
= (start_offset
+ src_offset
) &
5825 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5826 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5828 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5830 cur
= min_t(unsigned long, cur
,
5831 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5833 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5834 dst_off_in_page
, src_off_in_page
, cur
);
5842 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5843 unsigned long src_offset
, unsigned long len
)
5845 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5847 size_t dst_off_in_page
;
5848 size_t src_off_in_page
;
5849 unsigned long dst_end
= dst_offset
+ len
- 1;
5850 unsigned long src_end
= src_offset
+ len
- 1;
5851 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5852 unsigned long dst_i
;
5853 unsigned long src_i
;
5855 if (src_offset
+ len
> dst
->len
) {
5857 "memmove bogus src_offset %lu move len %lu len %lu",
5858 src_offset
, len
, dst
->len
);
5861 if (dst_offset
+ len
> dst
->len
) {
5863 "memmove bogus dst_offset %lu move len %lu len %lu",
5864 dst_offset
, len
, dst
->len
);
5867 if (dst_offset
< src_offset
) {
5868 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5872 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5873 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5875 dst_off_in_page
= (start_offset
+ dst_end
) &
5877 src_off_in_page
= (start_offset
+ src_end
) &
5880 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5881 cur
= min(cur
, dst_off_in_page
+ 1);
5882 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5883 dst_off_in_page
- cur
+ 1,
5884 src_off_in_page
- cur
+ 1, cur
);
5892 int try_release_extent_buffer(struct page
*page
)
5894 struct extent_buffer
*eb
;
5897 * We need to make sure nobody is attaching this page to an eb right
5900 spin_lock(&page
->mapping
->private_lock
);
5901 if (!PagePrivate(page
)) {
5902 spin_unlock(&page
->mapping
->private_lock
);
5906 eb
= (struct extent_buffer
*)page
->private;
5910 * This is a little awful but should be ok, we need to make sure that
5911 * the eb doesn't disappear out from under us while we're looking at
5914 spin_lock(&eb
->refs_lock
);
5915 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5916 spin_unlock(&eb
->refs_lock
);
5917 spin_unlock(&page
->mapping
->private_lock
);
5920 spin_unlock(&page
->mapping
->private_lock
);
5923 * If tree ref isn't set then we know the ref on this eb is a real ref,
5924 * so just return, this page will likely be freed soon anyway.
5926 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5927 spin_unlock(&eb
->refs_lock
);
5931 return release_extent_buffer(eb
);