1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
23 #include "transaction.h"
25 static struct kmem_cache
*extent_state_cache
;
26 static struct kmem_cache
*extent_buffer_cache
;
27 static struct bio_set
*btrfs_bioset
;
29 static inline bool extent_state_in_tree(const struct extent_state
*state
)
31 return !RB_EMPTY_NODE(&state
->rb_node
);
34 #ifdef CONFIG_BTRFS_DEBUG
35 static LIST_HEAD(buffers
);
36 static LIST_HEAD(states
);
38 static DEFINE_SPINLOCK(leak_lock
);
41 void btrfs_leak_debug_add(struct list_head
*new, struct list_head
*head
)
45 spin_lock_irqsave(&leak_lock
, flags
);
47 spin_unlock_irqrestore(&leak_lock
, flags
);
51 void btrfs_leak_debug_del(struct list_head
*entry
)
55 spin_lock_irqsave(&leak_lock
, flags
);
57 spin_unlock_irqrestore(&leak_lock
, flags
);
61 void btrfs_leak_debug_check(void)
63 struct extent_state
*state
;
64 struct extent_buffer
*eb
;
66 while (!list_empty(&states
)) {
67 state
= list_entry(states
.next
, struct extent_state
, leak_list
);
68 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
69 state
->start
, state
->end
, state
->state
,
70 extent_state_in_tree(state
),
71 atomic_read(&state
->refs
));
72 list_del(&state
->leak_list
);
73 kmem_cache_free(extent_state_cache
, state
);
76 while (!list_empty(&buffers
)) {
77 eb
= list_entry(buffers
.next
, struct extent_buffer
, leak_list
);
78 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
79 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
80 list_del(&eb
->leak_list
);
81 kmem_cache_free(extent_buffer_cache
, eb
);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
88 struct extent_io_tree
*tree
, u64 start
, u64 end
)
96 inode
= tree
->mapping
->host
;
97 isize
= i_size_read(inode
);
98 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
99 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
100 "%s: ino %llu isize %llu odd range [%llu,%llu]",
101 caller
, btrfs_ino(BTRFS_I(inode
)), isize
, start
, end
);
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry) do {} while (0)
107 #define btrfs_leak_debug_check() do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
111 #define BUFFER_LRU_MAX 64
116 struct rb_node rb_node
;
119 struct extent_page_data
{
121 struct extent_io_tree
*tree
;
122 get_extent_t
*get_extent
;
123 unsigned long bio_flags
;
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
128 unsigned int extent_locked
:1;
130 /* tells the submit_bio code to use REQ_SYNC */
131 unsigned int sync_io
:1;
134 static void add_extent_changeset(struct extent_state
*state
, unsigned bits
,
135 struct extent_changeset
*changeset
,
142 if (set
&& (state
->state
& bits
) == bits
)
144 if (!set
&& (state
->state
& bits
) == 0)
146 changeset
->bytes_changed
+= state
->end
- state
->start
+ 1;
147 ret
= ulist_add(&changeset
->range_changed
, state
->start
, state
->end
,
153 static noinline
void flush_write_bio(void *data
);
154 static inline struct btrfs_fs_info
*
155 tree_fs_info(struct extent_io_tree
*tree
)
159 return btrfs_sb(tree
->mapping
->host
->i_sb
);
162 int __init
extent_io_init(void)
164 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
165 sizeof(struct extent_state
), 0,
166 SLAB_MEM_SPREAD
, NULL
);
167 if (!extent_state_cache
)
170 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
171 sizeof(struct extent_buffer
), 0,
172 SLAB_MEM_SPREAD
, NULL
);
173 if (!extent_buffer_cache
)
174 goto free_state_cache
;
176 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
177 offsetof(struct btrfs_io_bio
, bio
));
179 goto free_buffer_cache
;
181 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
187 bioset_free(btrfs_bioset
);
191 kmem_cache_destroy(extent_buffer_cache
);
192 extent_buffer_cache
= NULL
;
195 kmem_cache_destroy(extent_state_cache
);
196 extent_state_cache
= NULL
;
200 void extent_io_exit(void)
202 btrfs_leak_debug_check();
205 * Make sure all delayed rcu free are flushed before we
209 kmem_cache_destroy(extent_state_cache
);
210 kmem_cache_destroy(extent_buffer_cache
);
212 bioset_free(btrfs_bioset
);
215 void extent_io_tree_init(struct extent_io_tree
*tree
,
216 struct address_space
*mapping
)
218 tree
->state
= RB_ROOT
;
220 tree
->dirty_bytes
= 0;
221 spin_lock_init(&tree
->lock
);
222 tree
->mapping
= mapping
;
225 static struct extent_state
*alloc_extent_state(gfp_t mask
)
227 struct extent_state
*state
;
230 * The given mask might be not appropriate for the slab allocator,
231 * drop the unsupported bits
233 mask
&= ~(__GFP_DMA32
|__GFP_HIGHMEM
);
234 state
= kmem_cache_alloc(extent_state_cache
, mask
);
238 state
->failrec
= NULL
;
239 RB_CLEAR_NODE(&state
->rb_node
);
240 btrfs_leak_debug_add(&state
->leak_list
, &states
);
241 atomic_set(&state
->refs
, 1);
242 init_waitqueue_head(&state
->wq
);
243 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
247 void free_extent_state(struct extent_state
*state
)
251 if (atomic_dec_and_test(&state
->refs
)) {
252 WARN_ON(extent_state_in_tree(state
));
253 btrfs_leak_debug_del(&state
->leak_list
);
254 trace_free_extent_state(state
, _RET_IP_
);
255 kmem_cache_free(extent_state_cache
, state
);
259 static struct rb_node
*tree_insert(struct rb_root
*root
,
260 struct rb_node
*search_start
,
262 struct rb_node
*node
,
263 struct rb_node
***p_in
,
264 struct rb_node
**parent_in
)
267 struct rb_node
*parent
= NULL
;
268 struct tree_entry
*entry
;
270 if (p_in
&& parent_in
) {
276 p
= search_start
? &search_start
: &root
->rb_node
;
279 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
281 if (offset
< entry
->start
)
283 else if (offset
> entry
->end
)
290 rb_link_node(node
, parent
, p
);
291 rb_insert_color(node
, root
);
295 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
296 struct rb_node
**prev_ret
,
297 struct rb_node
**next_ret
,
298 struct rb_node
***p_ret
,
299 struct rb_node
**parent_ret
)
301 struct rb_root
*root
= &tree
->state
;
302 struct rb_node
**n
= &root
->rb_node
;
303 struct rb_node
*prev
= NULL
;
304 struct rb_node
*orig_prev
= NULL
;
305 struct tree_entry
*entry
;
306 struct tree_entry
*prev_entry
= NULL
;
310 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
313 if (offset
< entry
->start
)
315 else if (offset
> entry
->end
)
328 while (prev
&& offset
> prev_entry
->end
) {
329 prev
= rb_next(prev
);
330 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
337 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
338 while (prev
&& offset
< prev_entry
->start
) {
339 prev
= rb_prev(prev
);
340 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
347 static inline struct rb_node
*
348 tree_search_for_insert(struct extent_io_tree
*tree
,
350 struct rb_node
***p_ret
,
351 struct rb_node
**parent_ret
)
353 struct rb_node
*prev
= NULL
;
356 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
362 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
365 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
368 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
369 struct extent_state
*other
)
371 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
372 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
377 * utility function to look for merge candidates inside a given range.
378 * Any extents with matching state are merged together into a single
379 * extent in the tree. Extents with EXTENT_IO in their state field
380 * are not merged because the end_io handlers need to be able to do
381 * operations on them without sleeping (or doing allocations/splits).
383 * This should be called with the tree lock held.
385 static void merge_state(struct extent_io_tree
*tree
,
386 struct extent_state
*state
)
388 struct extent_state
*other
;
389 struct rb_node
*other_node
;
391 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
394 other_node
= rb_prev(&state
->rb_node
);
396 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
397 if (other
->end
== state
->start
- 1 &&
398 other
->state
== state
->state
) {
399 merge_cb(tree
, state
, other
);
400 state
->start
= other
->start
;
401 rb_erase(&other
->rb_node
, &tree
->state
);
402 RB_CLEAR_NODE(&other
->rb_node
);
403 free_extent_state(other
);
406 other_node
= rb_next(&state
->rb_node
);
408 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
409 if (other
->start
== state
->end
+ 1 &&
410 other
->state
== state
->state
) {
411 merge_cb(tree
, state
, other
);
412 state
->end
= other
->end
;
413 rb_erase(&other
->rb_node
, &tree
->state
);
414 RB_CLEAR_NODE(&other
->rb_node
);
415 free_extent_state(other
);
420 static void set_state_cb(struct extent_io_tree
*tree
,
421 struct extent_state
*state
, unsigned *bits
)
423 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
424 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
427 static void clear_state_cb(struct extent_io_tree
*tree
,
428 struct extent_state
*state
, unsigned *bits
)
430 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
431 tree
->ops
->clear_bit_hook(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 atomic_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 atomic_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 atomic_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 atomic_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
;
2007 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2010 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2011 BUG_ON(!mirror_num
);
2013 /* we can't repair anything in raid56 yet */
2014 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2017 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2020 bio
->bi_iter
.bi_size
= 0;
2021 map_length
= length
;
2024 * Avoid races with device replace and make sure our bbio has devices
2025 * associated to its stripes that don't go away while we are doing the
2026 * read repair operation.
2028 btrfs_bio_counter_inc_blocked(fs_info
);
2029 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_WRITE
, logical
,
2030 &map_length
, &bbio
, mirror_num
);
2032 btrfs_bio_counter_dec(fs_info
);
2036 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2037 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2038 bio
->bi_iter
.bi_sector
= sector
;
2039 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2040 btrfs_put_bbio(bbio
);
2041 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2042 btrfs_bio_counter_dec(fs_info
);
2046 bio
->bi_bdev
= dev
->bdev
;
2047 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
;
2048 bio_add_page(bio
, page
, length
, pg_offset
);
2050 if (btrfsic_submit_bio_wait(bio
)) {
2051 /* try to remap that extent elsewhere? */
2052 btrfs_bio_counter_dec(fs_info
);
2054 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2058 btrfs_info_rl_in_rcu(fs_info
,
2059 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2060 btrfs_ino(inode
), start
,
2061 rcu_str_deref(dev
->name
), sector
);
2062 btrfs_bio_counter_dec(fs_info
);
2067 int repair_eb_io_failure(struct btrfs_fs_info
*fs_info
,
2068 struct extent_buffer
*eb
, int mirror_num
)
2070 u64 start
= eb
->start
;
2071 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2074 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2077 for (i
= 0; i
< num_pages
; i
++) {
2078 struct page
*p
= eb
->pages
[i
];
2080 ret
= repair_io_failure(BTRFS_I(fs_info
->btree_inode
), start
,
2081 PAGE_SIZE
, start
, p
,
2082 start
- page_offset(p
), mirror_num
);
2092 * each time an IO finishes, we do a fast check in the IO failure tree
2093 * to see if we need to process or clean up an io_failure_record
2095 int clean_io_failure(struct btrfs_inode
*inode
, u64 start
, struct page
*page
,
2096 unsigned int pg_offset
)
2099 struct io_failure_record
*failrec
;
2100 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2101 struct extent_state
*state
;
2106 ret
= count_range_bits(&inode
->io_failure_tree
, &private,
2107 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2111 ret
= get_state_failrec(&inode
->io_failure_tree
, start
,
2116 BUG_ON(!failrec
->this_mirror
);
2118 if (failrec
->in_validation
) {
2119 /* there was no real error, just free the record */
2120 btrfs_debug(fs_info
,
2121 "clean_io_failure: freeing dummy error at %llu",
2125 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2128 spin_lock(&inode
->io_tree
.lock
);
2129 state
= find_first_extent_bit_state(&inode
->io_tree
,
2132 spin_unlock(&inode
->io_tree
.lock
);
2134 if (state
&& state
->start
<= failrec
->start
&&
2135 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2136 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2138 if (num_copies
> 1) {
2139 repair_io_failure(inode
, start
, failrec
->len
,
2140 failrec
->logical
, page
,
2141 pg_offset
, failrec
->failed_mirror
);
2146 free_io_failure(inode
, failrec
);
2152 * Can be called when
2153 * - hold extent lock
2154 * - under ordered extent
2155 * - the inode is freeing
2157 void btrfs_free_io_failure_record(struct btrfs_inode
*inode
, u64 start
, u64 end
)
2159 struct extent_io_tree
*failure_tree
= &inode
->io_failure_tree
;
2160 struct io_failure_record
*failrec
;
2161 struct extent_state
*state
, *next
;
2163 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2166 spin_lock(&failure_tree
->lock
);
2167 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2169 if (state
->start
> end
)
2172 ASSERT(state
->end
<= end
);
2174 next
= next_state(state
);
2176 failrec
= state
->failrec
;
2177 free_extent_state(state
);
2182 spin_unlock(&failure_tree
->lock
);
2185 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2186 struct io_failure_record
**failrec_ret
)
2188 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2189 struct io_failure_record
*failrec
;
2190 struct extent_map
*em
;
2191 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2192 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2193 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2197 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2199 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2203 failrec
->start
= start
;
2204 failrec
->len
= end
- start
+ 1;
2205 failrec
->this_mirror
= 0;
2206 failrec
->bio_flags
= 0;
2207 failrec
->in_validation
= 0;
2209 read_lock(&em_tree
->lock
);
2210 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2212 read_unlock(&em_tree
->lock
);
2217 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2218 free_extent_map(em
);
2221 read_unlock(&em_tree
->lock
);
2227 logical
= start
- em
->start
;
2228 logical
= em
->block_start
+ logical
;
2229 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2230 logical
= em
->block_start
;
2231 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2232 extent_set_compress_type(&failrec
->bio_flags
,
2236 btrfs_debug(fs_info
,
2237 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2238 logical
, start
, failrec
->len
);
2240 failrec
->logical
= logical
;
2241 free_extent_map(em
);
2243 /* set the bits in the private failure tree */
2244 ret
= set_extent_bits(failure_tree
, start
, end
,
2245 EXTENT_LOCKED
| EXTENT_DIRTY
);
2247 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2248 /* set the bits in the inode's tree */
2250 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2256 btrfs_debug(fs_info
,
2257 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2258 failrec
->logical
, failrec
->start
, failrec
->len
,
2259 failrec
->in_validation
);
2261 * when data can be on disk more than twice, add to failrec here
2262 * (e.g. with a list for failed_mirror) to make
2263 * clean_io_failure() clean all those errors at once.
2267 *failrec_ret
= failrec
;
2272 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2273 struct io_failure_record
*failrec
, int failed_mirror
)
2275 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2278 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
2279 if (num_copies
== 1) {
2281 * we only have a single copy of the data, so don't bother with
2282 * all the retry and error correction code that follows. no
2283 * matter what the error is, it is very likely to persist.
2285 btrfs_debug(fs_info
,
2286 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2287 num_copies
, failrec
->this_mirror
, failed_mirror
);
2292 * there are two premises:
2293 * a) deliver good data to the caller
2294 * b) correct the bad sectors on disk
2296 if (failed_bio
->bi_vcnt
> 1) {
2298 * to fulfill b), we need to know the exact failing sectors, as
2299 * we don't want to rewrite any more than the failed ones. thus,
2300 * we need separate read requests for the failed bio
2302 * if the following BUG_ON triggers, our validation request got
2303 * merged. we need separate requests for our algorithm to work.
2305 BUG_ON(failrec
->in_validation
);
2306 failrec
->in_validation
= 1;
2307 failrec
->this_mirror
= failed_mirror
;
2310 * we're ready to fulfill a) and b) alongside. get a good copy
2311 * of the failed sector and if we succeed, we have setup
2312 * everything for repair_io_failure to do the rest for us.
2314 if (failrec
->in_validation
) {
2315 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2316 failrec
->in_validation
= 0;
2317 failrec
->this_mirror
= 0;
2319 failrec
->failed_mirror
= failed_mirror
;
2320 failrec
->this_mirror
++;
2321 if (failrec
->this_mirror
== failed_mirror
)
2322 failrec
->this_mirror
++;
2325 if (failrec
->this_mirror
> num_copies
) {
2326 btrfs_debug(fs_info
,
2327 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2328 num_copies
, failrec
->this_mirror
, failed_mirror
);
2336 struct bio
*btrfs_create_repair_bio(struct inode
*inode
, struct bio
*failed_bio
,
2337 struct io_failure_record
*failrec
,
2338 struct page
*page
, int pg_offset
, int icsum
,
2339 bio_end_io_t
*endio_func
, void *data
)
2341 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2343 struct btrfs_io_bio
*btrfs_failed_bio
;
2344 struct btrfs_io_bio
*btrfs_bio
;
2346 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2350 bio
->bi_end_io
= endio_func
;
2351 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2352 bio
->bi_bdev
= fs_info
->fs_devices
->latest_bdev
;
2353 bio
->bi_iter
.bi_size
= 0;
2354 bio
->bi_private
= data
;
2356 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2357 if (btrfs_failed_bio
->csum
) {
2358 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
2360 btrfs_bio
= btrfs_io_bio(bio
);
2361 btrfs_bio
->csum
= btrfs_bio
->csum_inline
;
2363 memcpy(btrfs_bio
->csum
, btrfs_failed_bio
->csum
+ icsum
,
2367 bio_add_page(bio
, page
, failrec
->len
, pg_offset
);
2373 * this is a generic handler for readpage errors (default
2374 * readpage_io_failed_hook). if other copies exist, read those and write back
2375 * good data to the failed position. does not investigate in remapping the
2376 * failed extent elsewhere, hoping the device will be smart enough to do this as
2380 static int bio_readpage_error(struct bio
*failed_bio
, u64 phy_offset
,
2381 struct page
*page
, u64 start
, u64 end
,
2384 struct io_failure_record
*failrec
;
2385 struct inode
*inode
= page
->mapping
->host
;
2386 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2391 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
2393 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
2397 ret
= btrfs_check_repairable(inode
, failed_bio
, failrec
, failed_mirror
);
2399 free_io_failure(BTRFS_I(inode
), failrec
);
2403 if (failed_bio
->bi_vcnt
> 1)
2404 read_mode
|= REQ_FAILFAST_DEV
;
2406 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2407 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2408 start
- page_offset(page
),
2409 (int)phy_offset
, failed_bio
->bi_end_io
,
2412 free_io_failure(BTRFS_I(inode
), failrec
);
2415 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
2417 btrfs_debug(btrfs_sb(inode
->i_sb
),
2418 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2419 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2421 ret
= tree
->ops
->submit_bio_hook(inode
, bio
, failrec
->this_mirror
,
2422 failrec
->bio_flags
, 0);
2424 free_io_failure(BTRFS_I(inode
), failrec
);
2431 /* lots and lots of room for performance fixes in the end_bio funcs */
2433 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2435 int uptodate
= (err
== 0);
2436 struct extent_io_tree
*tree
;
2439 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2441 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
2442 tree
->ops
->writepage_end_io_hook(page
, start
, end
, NULL
,
2446 ClearPageUptodate(page
);
2448 ret
= ret
< 0 ? ret
: -EIO
;
2449 mapping_set_error(page
->mapping
, ret
);
2454 * after a writepage IO is done, we need to:
2455 * clear the uptodate bits on error
2456 * clear the writeback bits in the extent tree for this IO
2457 * end_page_writeback if the page has no more pending IO
2459 * Scheduling is not allowed, so the extent state tree is expected
2460 * to have one and only one object corresponding to this IO.
2462 static void end_bio_extent_writepage(struct bio
*bio
)
2464 struct bio_vec
*bvec
;
2469 bio_for_each_segment_all(bvec
, bio
, i
) {
2470 struct page
*page
= bvec
->bv_page
;
2471 struct inode
*inode
= page
->mapping
->host
;
2472 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2474 /* We always issue full-page reads, but if some block
2475 * in a page fails to read, blk_update_request() will
2476 * advance bv_offset and adjust bv_len to compensate.
2477 * Print a warning for nonzero offsets, and an error
2478 * if they don't add up to a full page. */
2479 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2480 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2482 "partial page write in btrfs with offset %u and length %u",
2483 bvec
->bv_offset
, bvec
->bv_len
);
2486 "incomplete page write in btrfs with offset %u and length %u",
2487 bvec
->bv_offset
, bvec
->bv_len
);
2490 start
= page_offset(page
);
2491 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2493 end_extent_writepage(page
, bio
->bi_error
, start
, end
);
2494 end_page_writeback(page
);
2501 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2504 struct extent_state
*cached
= NULL
;
2505 u64 end
= start
+ len
- 1;
2507 if (uptodate
&& tree
->track_uptodate
)
2508 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2509 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2513 * after a readpage IO is done, we need to:
2514 * clear the uptodate bits on error
2515 * set the uptodate bits if things worked
2516 * set the page up to date if all extents in the tree are uptodate
2517 * clear the lock bit in the extent tree
2518 * unlock the page if there are no other extents locked for it
2520 * Scheduling is not allowed, so the extent state tree is expected
2521 * to have one and only one object corresponding to this IO.
2523 static void end_bio_extent_readpage(struct bio
*bio
)
2525 struct bio_vec
*bvec
;
2526 int uptodate
= !bio
->bi_error
;
2527 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2528 struct extent_io_tree
*tree
;
2533 u64 extent_start
= 0;
2539 bio_for_each_segment_all(bvec
, bio
, i
) {
2540 struct page
*page
= bvec
->bv_page
;
2541 struct inode
*inode
= page
->mapping
->host
;
2542 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2544 btrfs_debug(fs_info
,
2545 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2546 (u64
)bio
->bi_iter
.bi_sector
, bio
->bi_error
,
2547 io_bio
->mirror_num
);
2548 tree
= &BTRFS_I(inode
)->io_tree
;
2550 /* We always issue full-page reads, but if some block
2551 * in a page fails to read, blk_update_request() will
2552 * advance bv_offset and adjust bv_len to compensate.
2553 * Print a warning for nonzero offsets, and an error
2554 * if they don't add up to a full page. */
2555 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2556 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2558 "partial page read in btrfs with offset %u and length %u",
2559 bvec
->bv_offset
, bvec
->bv_len
);
2562 "incomplete page read in btrfs with offset %u and length %u",
2563 bvec
->bv_offset
, bvec
->bv_len
);
2566 start
= page_offset(page
);
2567 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2570 mirror
= io_bio
->mirror_num
;
2571 if (likely(uptodate
&& tree
->ops
)) {
2572 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2578 clean_io_failure(BTRFS_I(inode
), start
,
2582 if (likely(uptodate
))
2586 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2587 if (ret
== -EAGAIN
) {
2589 * Data inode's readpage_io_failed_hook() always
2592 * The generic bio_readpage_error handles errors
2593 * the following way: If possible, new read
2594 * requests are created and submitted and will
2595 * end up in end_bio_extent_readpage as well (if
2596 * we're lucky, not in the !uptodate case). In
2597 * that case it returns 0 and we just go on with
2598 * the next page in our bio. If it can't handle
2599 * the error it will return -EIO and we remain
2600 * responsible for that page.
2602 ret
= bio_readpage_error(bio
, offset
, page
,
2603 start
, end
, mirror
);
2605 uptodate
= !bio
->bi_error
;
2612 * metadata's readpage_io_failed_hook() always returns
2613 * -EIO and fixes nothing. -EIO is also returned if
2614 * data inode error could not be fixed.
2616 ASSERT(ret
== -EIO
);
2619 if (likely(uptodate
)) {
2620 loff_t i_size
= i_size_read(inode
);
2621 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2624 /* Zero out the end if this page straddles i_size */
2625 off
= i_size
& (PAGE_SIZE
-1);
2626 if (page
->index
== end_index
&& off
)
2627 zero_user_segment(page
, off
, PAGE_SIZE
);
2628 SetPageUptodate(page
);
2630 ClearPageUptodate(page
);
2636 if (unlikely(!uptodate
)) {
2638 endio_readpage_release_extent(tree
,
2644 endio_readpage_release_extent(tree
, start
,
2645 end
- start
+ 1, 0);
2646 } else if (!extent_len
) {
2647 extent_start
= start
;
2648 extent_len
= end
+ 1 - start
;
2649 } else if (extent_start
+ extent_len
== start
) {
2650 extent_len
+= end
+ 1 - start
;
2652 endio_readpage_release_extent(tree
, extent_start
,
2653 extent_len
, uptodate
);
2654 extent_start
= start
;
2655 extent_len
= end
+ 1 - start
;
2660 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2663 io_bio
->end_io(io_bio
, bio
->bi_error
);
2668 * this allocates from the btrfs_bioset. We're returning a bio right now
2669 * but you can call btrfs_io_bio for the appropriate container_of magic
2672 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2675 struct btrfs_io_bio
*btrfs_bio
;
2678 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2680 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2681 while (!bio
&& (nr_vecs
/= 2)) {
2682 bio
= bio_alloc_bioset(gfp_flags
,
2683 nr_vecs
, btrfs_bioset
);
2688 bio
->bi_bdev
= bdev
;
2689 bio
->bi_iter
.bi_sector
= first_sector
;
2690 btrfs_bio
= btrfs_io_bio(bio
);
2691 btrfs_bio
->csum
= NULL
;
2692 btrfs_bio
->csum_allocated
= NULL
;
2693 btrfs_bio
->end_io
= NULL
;
2698 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2700 struct btrfs_io_bio
*btrfs_bio
;
2703 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2705 btrfs_bio
= btrfs_io_bio(new);
2706 btrfs_bio
->csum
= NULL
;
2707 btrfs_bio
->csum_allocated
= NULL
;
2708 btrfs_bio
->end_io
= NULL
;
2713 /* this also allocates from the btrfs_bioset */
2714 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2716 struct btrfs_io_bio
*btrfs_bio
;
2719 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2721 btrfs_bio
= btrfs_io_bio(bio
);
2722 btrfs_bio
->csum
= NULL
;
2723 btrfs_bio
->csum_allocated
= NULL
;
2724 btrfs_bio
->end_io
= NULL
;
2730 static int __must_check
submit_one_bio(struct bio
*bio
, int mirror_num
,
2731 unsigned long bio_flags
)
2734 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2735 struct page
*page
= bvec
->bv_page
;
2736 struct extent_io_tree
*tree
= bio
->bi_private
;
2739 start
= page_offset(page
) + bvec
->bv_offset
;
2741 bio
->bi_private
= NULL
;
2745 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, bio
,
2746 mirror_num
, bio_flags
, start
);
2748 btrfsic_submit_bio(bio
);
2754 static int merge_bio(struct extent_io_tree
*tree
, struct page
*page
,
2755 unsigned long offset
, size_t size
, struct bio
*bio
,
2756 unsigned long bio_flags
)
2760 ret
= tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
,
2766 static int submit_extent_page(int op
, int op_flags
, struct extent_io_tree
*tree
,
2767 struct writeback_control
*wbc
,
2768 struct page
*page
, sector_t sector
,
2769 size_t size
, unsigned long offset
,
2770 struct block_device
*bdev
,
2771 struct bio
**bio_ret
,
2772 bio_end_io_t end_io_func
,
2774 unsigned long prev_bio_flags
,
2775 unsigned long bio_flags
,
2776 bool force_bio_submit
)
2781 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2782 size_t page_size
= min_t(size_t, size
, PAGE_SIZE
);
2784 if (bio_ret
&& *bio_ret
) {
2787 contig
= bio
->bi_iter
.bi_sector
== sector
;
2789 contig
= bio_end_sector(bio
) == sector
;
2791 if (prev_bio_flags
!= bio_flags
|| !contig
||
2793 merge_bio(tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2794 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2795 ret
= submit_one_bio(bio
, mirror_num
, prev_bio_flags
);
2803 wbc_account_io(wbc
, page
, page_size
);
2808 bio
= btrfs_bio_alloc(bdev
, sector
, BIO_MAX_PAGES
,
2809 GFP_NOFS
| __GFP_HIGH
);
2813 bio_add_page(bio
, page
, page_size
, offset
);
2814 bio
->bi_end_io
= end_io_func
;
2815 bio
->bi_private
= tree
;
2816 bio_set_op_attrs(bio
, op
, op_flags
);
2818 wbc_init_bio(wbc
, bio
);
2819 wbc_account_io(wbc
, page
, page_size
);
2825 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
2830 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2833 if (!PagePrivate(page
)) {
2834 SetPagePrivate(page
);
2836 set_page_private(page
, (unsigned long)eb
);
2838 WARN_ON(page
->private != (unsigned long)eb
);
2842 void set_page_extent_mapped(struct page
*page
)
2844 if (!PagePrivate(page
)) {
2845 SetPagePrivate(page
);
2847 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2851 static struct extent_map
*
2852 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2853 u64 start
, u64 len
, get_extent_t
*get_extent
,
2854 struct extent_map
**em_cached
)
2856 struct extent_map
*em
;
2858 if (em_cached
&& *em_cached
) {
2860 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2861 start
< extent_map_end(em
)) {
2862 atomic_inc(&em
->refs
);
2866 free_extent_map(em
);
2870 em
= get_extent(BTRFS_I(inode
), page
, pg_offset
, start
, len
, 0);
2871 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2873 atomic_inc(&em
->refs
);
2879 * basic readpage implementation. Locked extent state structs are inserted
2880 * into the tree that are removed when the IO is done (by the end_io
2882 * XXX JDM: This needs looking at to ensure proper page locking
2883 * return 0 on success, otherwise return error
2885 static int __do_readpage(struct extent_io_tree
*tree
,
2887 get_extent_t
*get_extent
,
2888 struct extent_map
**em_cached
,
2889 struct bio
**bio
, int mirror_num
,
2890 unsigned long *bio_flags
, int read_flags
,
2893 struct inode
*inode
= page
->mapping
->host
;
2894 u64 start
= page_offset(page
);
2895 u64 page_end
= start
+ PAGE_SIZE
- 1;
2899 u64 last_byte
= i_size_read(inode
);
2903 struct extent_map
*em
;
2904 struct block_device
*bdev
;
2907 size_t pg_offset
= 0;
2909 size_t disk_io_size
;
2910 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2911 unsigned long this_bio_flag
= 0;
2913 set_page_extent_mapped(page
);
2916 if (!PageUptodate(page
)) {
2917 if (cleancache_get_page(page
) == 0) {
2918 BUG_ON(blocksize
!= PAGE_SIZE
);
2919 unlock_extent(tree
, start
, end
);
2924 if (page
->index
== last_byte
>> PAGE_SHIFT
) {
2926 size_t zero_offset
= last_byte
& (PAGE_SIZE
- 1);
2929 iosize
= PAGE_SIZE
- zero_offset
;
2930 userpage
= kmap_atomic(page
);
2931 memset(userpage
+ zero_offset
, 0, iosize
);
2932 flush_dcache_page(page
);
2933 kunmap_atomic(userpage
);
2936 while (cur
<= end
) {
2937 bool force_bio_submit
= false;
2939 if (cur
>= last_byte
) {
2941 struct extent_state
*cached
= NULL
;
2943 iosize
= PAGE_SIZE
- pg_offset
;
2944 userpage
= kmap_atomic(page
);
2945 memset(userpage
+ pg_offset
, 0, iosize
);
2946 flush_dcache_page(page
);
2947 kunmap_atomic(userpage
);
2948 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2950 unlock_extent_cached(tree
, cur
,
2955 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2956 end
- cur
+ 1, get_extent
, em_cached
);
2957 if (IS_ERR_OR_NULL(em
)) {
2959 unlock_extent(tree
, cur
, end
);
2962 extent_offset
= cur
- em
->start
;
2963 BUG_ON(extent_map_end(em
) <= cur
);
2966 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2967 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2968 extent_set_compress_type(&this_bio_flag
,
2972 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2973 cur_end
= min(extent_map_end(em
) - 1, end
);
2974 iosize
= ALIGN(iosize
, blocksize
);
2975 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2976 disk_io_size
= em
->block_len
;
2977 sector
= em
->block_start
>> 9;
2979 sector
= (em
->block_start
+ extent_offset
) >> 9;
2980 disk_io_size
= iosize
;
2983 block_start
= em
->block_start
;
2984 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2985 block_start
= EXTENT_MAP_HOLE
;
2988 * If we have a file range that points to a compressed extent
2989 * and it's followed by a consecutive file range that points to
2990 * to the same compressed extent (possibly with a different
2991 * offset and/or length, so it either points to the whole extent
2992 * or only part of it), we must make sure we do not submit a
2993 * single bio to populate the pages for the 2 ranges because
2994 * this makes the compressed extent read zero out the pages
2995 * belonging to the 2nd range. Imagine the following scenario:
2998 * [0 - 8K] [8K - 24K]
3001 * points to extent X, points to extent X,
3002 * offset 4K, length of 8K offset 0, length 16K
3004 * [extent X, compressed length = 4K uncompressed length = 16K]
3006 * If the bio to read the compressed extent covers both ranges,
3007 * it will decompress extent X into the pages belonging to the
3008 * first range and then it will stop, zeroing out the remaining
3009 * pages that belong to the other range that points to extent X.
3010 * So here we make sure we submit 2 bios, one for the first
3011 * range and another one for the third range. Both will target
3012 * the same physical extent from disk, but we can't currently
3013 * make the compressed bio endio callback populate the pages
3014 * for both ranges because each compressed bio is tightly
3015 * coupled with a single extent map, and each range can have
3016 * an extent map with a different offset value relative to the
3017 * uncompressed data of our extent and different lengths. This
3018 * is a corner case so we prioritize correctness over
3019 * non-optimal behavior (submitting 2 bios for the same extent).
3021 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3022 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3023 *prev_em_start
!= em
->orig_start
)
3024 force_bio_submit
= true;
3027 *prev_em_start
= em
->orig_start
;
3029 free_extent_map(em
);
3032 /* we've found a hole, just zero and go on */
3033 if (block_start
== EXTENT_MAP_HOLE
) {
3035 struct extent_state
*cached
= NULL
;
3037 userpage
= kmap_atomic(page
);
3038 memset(userpage
+ pg_offset
, 0, iosize
);
3039 flush_dcache_page(page
);
3040 kunmap_atomic(userpage
);
3042 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3044 unlock_extent_cached(tree
, cur
,
3048 pg_offset
+= iosize
;
3051 /* the get_extent function already copied into the page */
3052 if (test_range_bit(tree
, cur
, cur_end
,
3053 EXTENT_UPTODATE
, 1, NULL
)) {
3054 check_page_uptodate(tree
, page
);
3055 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3057 pg_offset
+= iosize
;
3060 /* we have an inline extent but it didn't get marked up
3061 * to date. Error out
3063 if (block_start
== EXTENT_MAP_INLINE
) {
3065 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3067 pg_offset
+= iosize
;
3071 ret
= submit_extent_page(REQ_OP_READ
, read_flags
, tree
, NULL
,
3072 page
, sector
, disk_io_size
, pg_offset
,
3074 end_bio_extent_readpage
, mirror_num
,
3080 *bio_flags
= this_bio_flag
;
3083 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3087 pg_offset
+= iosize
;
3091 if (!PageError(page
))
3092 SetPageUptodate(page
);
3098 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3099 struct page
*pages
[], int nr_pages
,
3101 get_extent_t
*get_extent
,
3102 struct extent_map
**em_cached
,
3103 struct bio
**bio
, int mirror_num
,
3104 unsigned long *bio_flags
,
3107 struct inode
*inode
;
3108 struct btrfs_ordered_extent
*ordered
;
3111 inode
= pages
[0]->mapping
->host
;
3113 lock_extent(tree
, start
, end
);
3114 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
3118 unlock_extent(tree
, start
, end
);
3119 btrfs_start_ordered_extent(inode
, ordered
, 1);
3120 btrfs_put_ordered_extent(ordered
);
3123 for (index
= 0; index
< nr_pages
; index
++) {
3124 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3125 mirror_num
, bio_flags
, 0, prev_em_start
);
3126 put_page(pages
[index
]);
3130 static void __extent_readpages(struct extent_io_tree
*tree
,
3131 struct page
*pages
[],
3132 int nr_pages
, get_extent_t
*get_extent
,
3133 struct extent_map
**em_cached
,
3134 struct bio
**bio
, int mirror_num
,
3135 unsigned long *bio_flags
,
3142 int first_index
= 0;
3144 for (index
= 0; index
< nr_pages
; index
++) {
3145 page_start
= page_offset(pages
[index
]);
3148 end
= start
+ PAGE_SIZE
- 1;
3149 first_index
= index
;
3150 } else if (end
+ 1 == page_start
) {
3153 __do_contiguous_readpages(tree
, &pages
[first_index
],
3154 index
- first_index
, start
,
3155 end
, get_extent
, em_cached
,
3156 bio
, mirror_num
, bio_flags
,
3159 end
= start
+ PAGE_SIZE
- 1;
3160 first_index
= index
;
3165 __do_contiguous_readpages(tree
, &pages
[first_index
],
3166 index
- first_index
, start
,
3167 end
, get_extent
, em_cached
, bio
,
3168 mirror_num
, bio_flags
,
3172 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3174 get_extent_t
*get_extent
,
3175 struct bio
**bio
, int mirror_num
,
3176 unsigned long *bio_flags
, int read_flags
)
3178 struct inode
*inode
= page
->mapping
->host
;
3179 struct btrfs_ordered_extent
*ordered
;
3180 u64 start
= page_offset(page
);
3181 u64 end
= start
+ PAGE_SIZE
- 1;
3185 lock_extent(tree
, start
, end
);
3186 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
3190 unlock_extent(tree
, start
, end
);
3191 btrfs_start_ordered_extent(inode
, ordered
, 1);
3192 btrfs_put_ordered_extent(ordered
);
3195 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3196 bio_flags
, read_flags
, NULL
);
3200 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3201 get_extent_t
*get_extent
, int mirror_num
)
3203 struct bio
*bio
= NULL
;
3204 unsigned long bio_flags
= 0;
3207 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3210 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
3214 static void update_nr_written(struct writeback_control
*wbc
,
3215 unsigned long nr_written
)
3217 wbc
->nr_to_write
-= nr_written
;
3221 * helper for __extent_writepage, doing all of the delayed allocation setup.
3223 * This returns 1 if our fill_delalloc function did all the work required
3224 * to write the page (copy into inline extent). In this case the IO has
3225 * been started and the page is already unlocked.
3227 * This returns 0 if all went well (page still locked)
3228 * This returns < 0 if there were errors (page still locked)
3230 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3231 struct page
*page
, struct writeback_control
*wbc
,
3232 struct extent_page_data
*epd
,
3234 unsigned long *nr_written
)
3236 struct extent_io_tree
*tree
= epd
->tree
;
3237 u64 page_end
= delalloc_start
+ PAGE_SIZE
- 1;
3239 u64 delalloc_to_write
= 0;
3240 u64 delalloc_end
= 0;
3242 int page_started
= 0;
3244 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3247 while (delalloc_end
< page_end
) {
3248 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3252 BTRFS_MAX_EXTENT_SIZE
);
3253 if (nr_delalloc
== 0) {
3254 delalloc_start
= delalloc_end
+ 1;
3257 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3262 /* File system has been set read-only */
3265 /* fill_delalloc should be return < 0 for error
3266 * but just in case, we use > 0 here meaning the
3267 * IO is started, so we don't want to return > 0
3268 * unless things are going well.
3270 ret
= ret
< 0 ? ret
: -EIO
;
3274 * delalloc_end is already one less than the total length, so
3275 * we don't subtract one from PAGE_SIZE
3277 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3278 PAGE_SIZE
) >> PAGE_SHIFT
;
3279 delalloc_start
= delalloc_end
+ 1;
3281 if (wbc
->nr_to_write
< delalloc_to_write
) {
3284 if (delalloc_to_write
< thresh
* 2)
3285 thresh
= delalloc_to_write
;
3286 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3290 /* did the fill delalloc function already unlock and start
3295 * we've unlocked the page, so we can't update
3296 * the mapping's writeback index, just update
3299 wbc
->nr_to_write
-= *nr_written
;
3310 * helper for __extent_writepage. This calls the writepage start hooks,
3311 * and does the loop to map the page into extents and bios.
3313 * We return 1 if the IO is started and the page is unlocked,
3314 * 0 if all went well (page still locked)
3315 * < 0 if there were errors (page still locked)
3317 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3319 struct writeback_control
*wbc
,
3320 struct extent_page_data
*epd
,
3322 unsigned long nr_written
,
3323 int write_flags
, int *nr_ret
)
3325 struct extent_io_tree
*tree
= epd
->tree
;
3326 u64 start
= page_offset(page
);
3327 u64 page_end
= start
+ PAGE_SIZE
- 1;
3334 struct extent_map
*em
;
3335 struct block_device
*bdev
;
3336 size_t pg_offset
= 0;
3342 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3343 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3346 /* Fixup worker will requeue */
3348 wbc
->pages_skipped
++;
3350 redirty_page_for_writepage(wbc
, page
);
3352 update_nr_written(wbc
, nr_written
);
3359 * we don't want to touch the inode after unlocking the page,
3360 * so we update the mapping writeback index now
3362 update_nr_written(wbc
, nr_written
+ 1);
3365 if (i_size
<= start
) {
3366 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3367 tree
->ops
->writepage_end_io_hook(page
, start
,
3372 blocksize
= inode
->i_sb
->s_blocksize
;
3374 while (cur
<= end
) {
3377 if (cur
>= i_size
) {
3378 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3379 tree
->ops
->writepage_end_io_hook(page
, cur
,
3383 em
= epd
->get_extent(BTRFS_I(inode
), page
, pg_offset
, cur
,
3385 if (IS_ERR_OR_NULL(em
)) {
3387 ret
= PTR_ERR_OR_ZERO(em
);
3391 extent_offset
= cur
- em
->start
;
3392 em_end
= extent_map_end(em
);
3393 BUG_ON(em_end
<= cur
);
3395 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3396 iosize
= ALIGN(iosize
, blocksize
);
3397 sector
= (em
->block_start
+ extent_offset
) >> 9;
3399 block_start
= em
->block_start
;
3400 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3401 free_extent_map(em
);
3405 * compressed and inline extents are written through other
3408 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3409 block_start
== EXTENT_MAP_INLINE
) {
3411 * end_io notification does not happen here for
3412 * compressed extents
3414 if (!compressed
&& tree
->ops
&&
3415 tree
->ops
->writepage_end_io_hook
)
3416 tree
->ops
->writepage_end_io_hook(page
, cur
,
3419 else if (compressed
) {
3420 /* we don't want to end_page_writeback on
3421 * a compressed extent. this happens
3428 pg_offset
+= iosize
;
3432 set_range_writeback(tree
, cur
, cur
+ iosize
- 1);
3433 if (!PageWriteback(page
)) {
3434 btrfs_err(BTRFS_I(inode
)->root
->fs_info
,
3435 "page %lu not writeback, cur %llu end %llu",
3436 page
->index
, cur
, end
);
3439 ret
= submit_extent_page(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3440 page
, sector
, iosize
, pg_offset
,
3442 end_bio_extent_writepage
,
3446 if (PageWriteback(page
))
3447 end_page_writeback(page
);
3451 pg_offset
+= iosize
;
3460 * the writepage semantics are similar to regular writepage. extent
3461 * records are inserted to lock ranges in the tree, and as dirty areas
3462 * are found, they are marked writeback. Then the lock bits are removed
3463 * and the end_io handler clears the writeback ranges
3465 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3468 struct inode
*inode
= page
->mapping
->host
;
3469 struct extent_page_data
*epd
= data
;
3470 u64 start
= page_offset(page
);
3471 u64 page_end
= start
+ PAGE_SIZE
- 1;
3474 size_t pg_offset
= 0;
3475 loff_t i_size
= i_size_read(inode
);
3476 unsigned long end_index
= i_size
>> PAGE_SHIFT
;
3477 int write_flags
= 0;
3478 unsigned long nr_written
= 0;
3480 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3481 write_flags
= REQ_SYNC
;
3483 trace___extent_writepage(page
, inode
, wbc
);
3485 WARN_ON(!PageLocked(page
));
3487 ClearPageError(page
);
3489 pg_offset
= i_size
& (PAGE_SIZE
- 1);
3490 if (page
->index
> end_index
||
3491 (page
->index
== end_index
&& !pg_offset
)) {
3492 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3497 if (page
->index
== end_index
) {
3500 userpage
= kmap_atomic(page
);
3501 memset(userpage
+ pg_offset
, 0,
3502 PAGE_SIZE
- pg_offset
);
3503 kunmap_atomic(userpage
);
3504 flush_dcache_page(page
);
3509 set_page_extent_mapped(page
);
3511 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3517 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3518 i_size
, nr_written
, write_flags
, &nr
);
3524 /* make sure the mapping tag for page dirty gets cleared */
3525 set_page_writeback(page
);
3526 end_page_writeback(page
);
3528 if (PageError(page
)) {
3529 ret
= ret
< 0 ? ret
: -EIO
;
3530 end_extent_writepage(page
, ret
, start
, page_end
);
3539 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3541 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3542 TASK_UNINTERRUPTIBLE
);
3545 static noinline_for_stack
int
3546 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3547 struct btrfs_fs_info
*fs_info
,
3548 struct extent_page_data
*epd
)
3550 unsigned long i
, num_pages
;
3554 if (!btrfs_try_tree_write_lock(eb
)) {
3556 flush_write_bio(epd
);
3557 btrfs_tree_lock(eb
);
3560 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3561 btrfs_tree_unlock(eb
);
3565 flush_write_bio(epd
);
3569 wait_on_extent_buffer_writeback(eb
);
3570 btrfs_tree_lock(eb
);
3571 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3573 btrfs_tree_unlock(eb
);
3578 * We need to do this to prevent races in people who check if the eb is
3579 * under IO since we can end up having no IO bits set for a short period
3582 spin_lock(&eb
->refs_lock
);
3583 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3584 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3585 spin_unlock(&eb
->refs_lock
);
3586 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3587 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3589 fs_info
->dirty_metadata_batch
);
3592 spin_unlock(&eb
->refs_lock
);
3595 btrfs_tree_unlock(eb
);
3600 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3601 for (i
= 0; i
< num_pages
; i
++) {
3602 struct page
*p
= eb
->pages
[i
];
3604 if (!trylock_page(p
)) {
3606 flush_write_bio(epd
);
3616 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3618 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3619 smp_mb__after_atomic();
3620 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3623 static void set_btree_ioerr(struct page
*page
)
3625 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3628 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3632 * If writeback for a btree extent that doesn't belong to a log tree
3633 * failed, increment the counter transaction->eb_write_errors.
3634 * We do this because while the transaction is running and before it's
3635 * committing (when we call filemap_fdata[write|wait]_range against
3636 * the btree inode), we might have
3637 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3638 * returns an error or an error happens during writeback, when we're
3639 * committing the transaction we wouldn't know about it, since the pages
3640 * can be no longer dirty nor marked anymore for writeback (if a
3641 * subsequent modification to the extent buffer didn't happen before the
3642 * transaction commit), which makes filemap_fdata[write|wait]_range not
3643 * able to find the pages tagged with SetPageError at transaction
3644 * commit time. So if this happens we must abort the transaction,
3645 * otherwise we commit a super block with btree roots that point to
3646 * btree nodes/leafs whose content on disk is invalid - either garbage
3647 * or the content of some node/leaf from a past generation that got
3648 * cowed or deleted and is no longer valid.
3650 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3651 * not be enough - we need to distinguish between log tree extents vs
3652 * non-log tree extents, and the next filemap_fdatawait_range() call
3653 * will catch and clear such errors in the mapping - and that call might
3654 * be from a log sync and not from a transaction commit. Also, checking
3655 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3656 * not done and would not be reliable - the eb might have been released
3657 * from memory and reading it back again means that flag would not be
3658 * set (since it's a runtime flag, not persisted on disk).
3660 * Using the flags below in the btree inode also makes us achieve the
3661 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3662 * writeback for all dirty pages and before filemap_fdatawait_range()
3663 * is called, the writeback for all dirty pages had already finished
3664 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3665 * filemap_fdatawait_range() would return success, as it could not know
3666 * that writeback errors happened (the pages were no longer tagged for
3669 switch (eb
->log_index
) {
3671 set_bit(BTRFS_FS_BTREE_ERR
, &eb
->fs_info
->flags
);
3674 set_bit(BTRFS_FS_LOG1_ERR
, &eb
->fs_info
->flags
);
3677 set_bit(BTRFS_FS_LOG2_ERR
, &eb
->fs_info
->flags
);
3680 BUG(); /* unexpected, logic error */
3684 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3686 struct bio_vec
*bvec
;
3687 struct extent_buffer
*eb
;
3690 bio_for_each_segment_all(bvec
, bio
, i
) {
3691 struct page
*page
= bvec
->bv_page
;
3693 eb
= (struct extent_buffer
*)page
->private;
3695 done
= atomic_dec_and_test(&eb
->io_pages
);
3697 if (bio
->bi_error
||
3698 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3699 ClearPageUptodate(page
);
3700 set_btree_ioerr(page
);
3703 end_page_writeback(page
);
3708 end_extent_buffer_writeback(eb
);
3714 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3715 struct btrfs_fs_info
*fs_info
,
3716 struct writeback_control
*wbc
,
3717 struct extent_page_data
*epd
)
3719 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3720 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3721 u64 offset
= eb
->start
;
3723 unsigned long i
, num_pages
;
3724 unsigned long bio_flags
= 0;
3725 unsigned long start
, end
;
3726 int write_flags
= (epd
->sync_io
? REQ_SYNC
: 0) | REQ_META
;
3729 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3730 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3731 atomic_set(&eb
->io_pages
, num_pages
);
3732 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3733 bio_flags
= EXTENT_BIO_TREE_LOG
;
3735 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3736 nritems
= btrfs_header_nritems(eb
);
3737 if (btrfs_header_level(eb
) > 0) {
3738 end
= btrfs_node_key_ptr_offset(nritems
);
3740 memzero_extent_buffer(eb
, end
, eb
->len
- end
);
3744 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3746 start
= btrfs_item_nr_offset(nritems
);
3747 end
= btrfs_leaf_data(eb
) + leaf_data_end(fs_info
, eb
);
3748 memzero_extent_buffer(eb
, start
, end
- start
);
3751 for (i
= 0; i
< num_pages
; i
++) {
3752 struct page
*p
= eb
->pages
[i
];
3754 clear_page_dirty_for_io(p
);
3755 set_page_writeback(p
);
3756 ret
= submit_extent_page(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3757 p
, offset
>> 9, PAGE_SIZE
, 0, bdev
,
3759 end_bio_extent_buffer_writepage
,
3760 0, epd
->bio_flags
, bio_flags
, false);
3761 epd
->bio_flags
= bio_flags
;
3764 if (PageWriteback(p
))
3765 end_page_writeback(p
);
3766 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3767 end_extent_buffer_writeback(eb
);
3771 offset
+= PAGE_SIZE
;
3772 update_nr_written(wbc
, 1);
3776 if (unlikely(ret
)) {
3777 for (; i
< num_pages
; i
++) {
3778 struct page
*p
= eb
->pages
[i
];
3779 clear_page_dirty_for_io(p
);
3787 int btree_write_cache_pages(struct address_space
*mapping
,
3788 struct writeback_control
*wbc
)
3790 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3791 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3792 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3793 struct extent_page_data epd
= {
3797 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3802 int nr_to_write_done
= 0;
3803 struct pagevec pvec
;
3806 pgoff_t end
; /* Inclusive */
3810 pagevec_init(&pvec
, 0);
3811 if (wbc
->range_cyclic
) {
3812 index
= mapping
->writeback_index
; /* Start from prev offset */
3815 index
= wbc
->range_start
>> PAGE_SHIFT
;
3816 end
= wbc
->range_end
>> PAGE_SHIFT
;
3819 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3820 tag
= PAGECACHE_TAG_TOWRITE
;
3822 tag
= PAGECACHE_TAG_DIRTY
;
3824 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3825 tag_pages_for_writeback(mapping
, index
, end
);
3826 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3827 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3828 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3832 for (i
= 0; i
< nr_pages
; i
++) {
3833 struct page
*page
= pvec
.pages
[i
];
3835 if (!PagePrivate(page
))
3838 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3843 spin_lock(&mapping
->private_lock
);
3844 if (!PagePrivate(page
)) {
3845 spin_unlock(&mapping
->private_lock
);
3849 eb
= (struct extent_buffer
*)page
->private;
3852 * Shouldn't happen and normally this would be a BUG_ON
3853 * but no sense in crashing the users box for something
3854 * we can survive anyway.
3857 spin_unlock(&mapping
->private_lock
);
3861 if (eb
== prev_eb
) {
3862 spin_unlock(&mapping
->private_lock
);
3866 ret
= atomic_inc_not_zero(&eb
->refs
);
3867 spin_unlock(&mapping
->private_lock
);
3872 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3874 free_extent_buffer(eb
);
3878 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3881 free_extent_buffer(eb
);
3884 free_extent_buffer(eb
);
3887 * the filesystem may choose to bump up nr_to_write.
3888 * We have to make sure to honor the new nr_to_write
3891 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3893 pagevec_release(&pvec
);
3896 if (!scanned
&& !done
) {
3898 * We hit the last page and there is more work to be done: wrap
3899 * back to the start of the file
3905 flush_write_bio(&epd
);
3910 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3911 * @mapping: address space structure to write
3912 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3913 * @writepage: function called for each page
3914 * @data: data passed to writepage function
3916 * If a page is already under I/O, write_cache_pages() skips it, even
3917 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3918 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3919 * and msync() need to guarantee that all the data which was dirty at the time
3920 * the call was made get new I/O started against them. If wbc->sync_mode is
3921 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3922 * existing IO to complete.
3924 static int extent_write_cache_pages(struct address_space
*mapping
,
3925 struct writeback_control
*wbc
,
3926 writepage_t writepage
, void *data
,
3927 void (*flush_fn
)(void *))
3929 struct inode
*inode
= mapping
->host
;
3932 int nr_to_write_done
= 0;
3933 struct pagevec pvec
;
3936 pgoff_t end
; /* Inclusive */
3938 int range_whole
= 0;
3943 * We have to hold onto the inode so that ordered extents can do their
3944 * work when the IO finishes. The alternative to this is failing to add
3945 * an ordered extent if the igrab() fails there and that is a huge pain
3946 * to deal with, so instead just hold onto the inode throughout the
3947 * writepages operation. If it fails here we are freeing up the inode
3948 * anyway and we'd rather not waste our time writing out stuff that is
3949 * going to be truncated anyway.
3954 pagevec_init(&pvec
, 0);
3955 if (wbc
->range_cyclic
) {
3956 index
= mapping
->writeback_index
; /* Start from prev offset */
3959 index
= wbc
->range_start
>> PAGE_SHIFT
;
3960 end
= wbc
->range_end
>> PAGE_SHIFT
;
3961 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3965 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3966 tag
= PAGECACHE_TAG_TOWRITE
;
3968 tag
= PAGECACHE_TAG_DIRTY
;
3970 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3971 tag_pages_for_writeback(mapping
, index
, end
);
3973 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3974 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3975 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3979 for (i
= 0; i
< nr_pages
; i
++) {
3980 struct page
*page
= pvec
.pages
[i
];
3982 done_index
= page
->index
;
3984 * At this point we hold neither mapping->tree_lock nor
3985 * lock on the page itself: the page may be truncated or
3986 * invalidated (changing page->mapping to NULL), or even
3987 * swizzled back from swapper_space to tmpfs file
3990 if (!trylock_page(page
)) {
3995 if (unlikely(page
->mapping
!= mapping
)) {
4000 if (!wbc
->range_cyclic
&& page
->index
> end
) {
4006 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4007 if (PageWriteback(page
))
4009 wait_on_page_writeback(page
);
4012 if (PageWriteback(page
) ||
4013 !clear_page_dirty_for_io(page
)) {
4018 ret
= (*writepage
)(page
, wbc
, data
);
4020 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4026 * done_index is set past this page,
4027 * so media errors will not choke
4028 * background writeout for the entire
4029 * file. This has consequences for
4030 * range_cyclic semantics (ie. it may
4031 * not be suitable for data integrity
4034 done_index
= page
->index
+ 1;
4040 * the filesystem may choose to bump up nr_to_write.
4041 * We have to make sure to honor the new nr_to_write
4044 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4046 pagevec_release(&pvec
);
4049 if (!scanned
&& !done
) {
4051 * We hit the last page and there is more work to be done: wrap
4052 * back to the start of the file
4059 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4060 mapping
->writeback_index
= done_index
;
4062 btrfs_add_delayed_iput(inode
);
4066 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4071 bio_set_op_attrs(epd
->bio
, REQ_OP_WRITE
,
4072 epd
->sync_io
? REQ_SYNC
: 0);
4074 ret
= submit_one_bio(epd
->bio
, 0, epd
->bio_flags
);
4075 BUG_ON(ret
< 0); /* -ENOMEM */
4080 static noinline
void flush_write_bio(void *data
)
4082 struct extent_page_data
*epd
= data
;
4083 flush_epd_write_bio(epd
);
4086 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4087 get_extent_t
*get_extent
,
4088 struct writeback_control
*wbc
)
4091 struct extent_page_data epd
= {
4094 .get_extent
= get_extent
,
4096 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4100 ret
= __extent_writepage(page
, wbc
, &epd
);
4102 flush_epd_write_bio(&epd
);
4106 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4107 u64 start
, u64 end
, get_extent_t
*get_extent
,
4111 struct address_space
*mapping
= inode
->i_mapping
;
4113 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4116 struct extent_page_data epd
= {
4119 .get_extent
= get_extent
,
4121 .sync_io
= mode
== WB_SYNC_ALL
,
4124 struct writeback_control wbc_writepages
= {
4126 .nr_to_write
= nr_pages
* 2,
4127 .range_start
= start
,
4128 .range_end
= end
+ 1,
4131 while (start
<= end
) {
4132 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4133 if (clear_page_dirty_for_io(page
))
4134 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4136 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4137 tree
->ops
->writepage_end_io_hook(page
, start
,
4138 start
+ PAGE_SIZE
- 1,
4146 flush_epd_write_bio(&epd
);
4150 int extent_writepages(struct extent_io_tree
*tree
,
4151 struct address_space
*mapping
,
4152 get_extent_t
*get_extent
,
4153 struct writeback_control
*wbc
)
4156 struct extent_page_data epd
= {
4159 .get_extent
= get_extent
,
4161 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4165 ret
= extent_write_cache_pages(mapping
, wbc
, __extent_writepage
, &epd
,
4167 flush_epd_write_bio(&epd
);
4171 int extent_readpages(struct extent_io_tree
*tree
,
4172 struct address_space
*mapping
,
4173 struct list_head
*pages
, unsigned nr_pages
,
4174 get_extent_t get_extent
)
4176 struct bio
*bio
= NULL
;
4178 unsigned long bio_flags
= 0;
4179 struct page
*pagepool
[16];
4181 struct extent_map
*em_cached
= NULL
;
4183 u64 prev_em_start
= (u64
)-1;
4185 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4186 page
= list_entry(pages
->prev
, struct page
, lru
);
4188 prefetchw(&page
->flags
);
4189 list_del(&page
->lru
);
4190 if (add_to_page_cache_lru(page
, mapping
,
4192 readahead_gfp_mask(mapping
))) {
4197 pagepool
[nr
++] = page
;
4198 if (nr
< ARRAY_SIZE(pagepool
))
4200 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4201 &bio
, 0, &bio_flags
, &prev_em_start
);
4205 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4206 &bio
, 0, &bio_flags
, &prev_em_start
);
4209 free_extent_map(em_cached
);
4211 BUG_ON(!list_empty(pages
));
4213 return submit_one_bio(bio
, 0, bio_flags
);
4218 * basic invalidatepage code, this waits on any locked or writeback
4219 * ranges corresponding to the page, and then deletes any extent state
4220 * records from the tree
4222 int extent_invalidatepage(struct extent_io_tree
*tree
,
4223 struct page
*page
, unsigned long offset
)
4225 struct extent_state
*cached_state
= NULL
;
4226 u64 start
= page_offset(page
);
4227 u64 end
= start
+ PAGE_SIZE
- 1;
4228 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4230 start
+= ALIGN(offset
, blocksize
);
4234 lock_extent_bits(tree
, start
, end
, &cached_state
);
4235 wait_on_page_writeback(page
);
4236 clear_extent_bit(tree
, start
, end
,
4237 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4238 EXTENT_DO_ACCOUNTING
,
4239 1, 1, &cached_state
, GFP_NOFS
);
4244 * a helper for releasepage, this tests for areas of the page that
4245 * are locked or under IO and drops the related state bits if it is safe
4248 static int try_release_extent_state(struct extent_map_tree
*map
,
4249 struct extent_io_tree
*tree
,
4250 struct page
*page
, gfp_t mask
)
4252 u64 start
= page_offset(page
);
4253 u64 end
= start
+ PAGE_SIZE
- 1;
4256 if (test_range_bit(tree
, start
, end
,
4257 EXTENT_IOBITS
, 0, NULL
))
4261 * at this point we can safely clear everything except the
4262 * locked bit and the nodatasum bit
4264 ret
= clear_extent_bit(tree
, start
, end
,
4265 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4268 /* if clear_extent_bit failed for enomem reasons,
4269 * we can't allow the release to continue.
4280 * a helper for releasepage. As long as there are no locked extents
4281 * in the range corresponding to the page, both state records and extent
4282 * map records are removed
4284 int try_release_extent_mapping(struct extent_map_tree
*map
,
4285 struct extent_io_tree
*tree
, struct page
*page
,
4288 struct extent_map
*em
;
4289 u64 start
= page_offset(page
);
4290 u64 end
= start
+ PAGE_SIZE
- 1;
4292 if (gfpflags_allow_blocking(mask
) &&
4293 page
->mapping
->host
->i_size
> SZ_16M
) {
4295 while (start
<= end
) {
4296 len
= end
- start
+ 1;
4297 write_lock(&map
->lock
);
4298 em
= lookup_extent_mapping(map
, start
, len
);
4300 write_unlock(&map
->lock
);
4303 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4304 em
->start
!= start
) {
4305 write_unlock(&map
->lock
);
4306 free_extent_map(em
);
4309 if (!test_range_bit(tree
, em
->start
,
4310 extent_map_end(em
) - 1,
4311 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4313 remove_extent_mapping(map
, em
);
4314 /* once for the rb tree */
4315 free_extent_map(em
);
4317 start
= extent_map_end(em
);
4318 write_unlock(&map
->lock
);
4321 free_extent_map(em
);
4324 return try_release_extent_state(map
, tree
, page
, mask
);
4328 * helper function for fiemap, which doesn't want to see any holes.
4329 * This maps until we find something past 'last'
4331 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4334 get_extent_t
*get_extent
)
4336 u64 sectorsize
= btrfs_inode_sectorsize(inode
);
4337 struct extent_map
*em
;
4344 len
= last
- offset
;
4347 len
= ALIGN(len
, sectorsize
);
4348 em
= get_extent(BTRFS_I(inode
), NULL
, 0, offset
, len
, 0);
4349 if (IS_ERR_OR_NULL(em
))
4352 /* if this isn't a hole return it */
4353 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4354 em
->block_start
!= EXTENT_MAP_HOLE
) {
4358 /* this is a hole, advance to the next extent */
4359 offset
= extent_map_end(em
);
4360 free_extent_map(em
);
4367 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4368 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4372 u64 max
= start
+ len
;
4376 u64 last_for_get_extent
= 0;
4378 u64 isize
= i_size_read(inode
);
4379 struct btrfs_key found_key
;
4380 struct extent_map
*em
= NULL
;
4381 struct extent_state
*cached_state
= NULL
;
4382 struct btrfs_path
*path
;
4383 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4392 path
= btrfs_alloc_path();
4395 path
->leave_spinning
= 1;
4397 start
= round_down(start
, btrfs_inode_sectorsize(inode
));
4398 len
= round_up(max
, btrfs_inode_sectorsize(inode
)) - start
;
4401 * lookup the last file extent. We're not using i_size here
4402 * because there might be preallocation past i_size
4404 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
4405 btrfs_ino(BTRFS_I(inode
)), -1, 0);
4407 btrfs_free_path(path
);
4416 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4417 found_type
= found_key
.type
;
4419 /* No extents, but there might be delalloc bits */
4420 if (found_key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
4421 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4422 /* have to trust i_size as the end */
4424 last_for_get_extent
= isize
;
4427 * remember the start of the last extent. There are a
4428 * bunch of different factors that go into the length of the
4429 * extent, so its much less complex to remember where it started
4431 last
= found_key
.offset
;
4432 last_for_get_extent
= last
+ 1;
4434 btrfs_release_path(path
);
4437 * we might have some extents allocated but more delalloc past those
4438 * extents. so, we trust isize unless the start of the last extent is
4443 last_for_get_extent
= isize
;
4446 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4449 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4459 u64 offset_in_extent
= 0;
4461 /* break if the extent we found is outside the range */
4462 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4466 * get_extent may return an extent that starts before our
4467 * requested range. We have to make sure the ranges
4468 * we return to fiemap always move forward and don't
4469 * overlap, so adjust the offsets here
4471 em_start
= max(em
->start
, off
);
4474 * record the offset from the start of the extent
4475 * for adjusting the disk offset below. Only do this if the
4476 * extent isn't compressed since our in ram offset may be past
4477 * what we have actually allocated on disk.
4479 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4480 offset_in_extent
= em_start
- em
->start
;
4481 em_end
= extent_map_end(em
);
4482 em_len
= em_end
- em_start
;
4487 * bump off for our next call to get_extent
4489 off
= extent_map_end(em
);
4493 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4495 flags
|= FIEMAP_EXTENT_LAST
;
4496 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4497 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4498 FIEMAP_EXTENT_NOT_ALIGNED
);
4499 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4500 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4501 FIEMAP_EXTENT_UNKNOWN
);
4502 } else if (fieinfo
->fi_extents_max
) {
4503 struct btrfs_trans_handle
*trans
;
4505 u64 bytenr
= em
->block_start
-
4506 (em
->start
- em
->orig_start
);
4508 disko
= em
->block_start
+ offset_in_extent
;
4511 * We need a trans handle to get delayed refs
4513 trans
= btrfs_join_transaction(root
);
4515 * It's OK if we can't start a trans we can still check
4522 * As btrfs supports shared space, this information
4523 * can be exported to userspace tools via
4524 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4525 * then we're just getting a count and we can skip the
4528 ret
= btrfs_check_shared(trans
, root
->fs_info
,
4530 btrfs_ino(BTRFS_I(inode
)), bytenr
);
4532 btrfs_end_transaction(trans
);
4536 flags
|= FIEMAP_EXTENT_SHARED
;
4539 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4540 flags
|= FIEMAP_EXTENT_ENCODED
;
4541 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4542 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4544 free_extent_map(em
);
4546 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4547 (last
== (u64
)-1 && isize
<= em_end
)) {
4548 flags
|= FIEMAP_EXTENT_LAST
;
4552 /* now scan forward to see if this is really the last extent. */
4553 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4560 flags
|= FIEMAP_EXTENT_LAST
;
4563 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4572 free_extent_map(em
);
4574 btrfs_free_path(path
);
4575 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4576 &cached_state
, GFP_NOFS
);
4580 static void __free_extent_buffer(struct extent_buffer
*eb
)
4582 btrfs_leak_debug_del(&eb
->leak_list
);
4583 kmem_cache_free(extent_buffer_cache
, eb
);
4586 int extent_buffer_under_io(struct extent_buffer
*eb
)
4588 return (atomic_read(&eb
->io_pages
) ||
4589 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4590 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4594 * Helper for releasing extent buffer page.
4596 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4598 unsigned long index
;
4600 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4602 BUG_ON(extent_buffer_under_io(eb
));
4604 index
= num_extent_pages(eb
->start
, eb
->len
);
4610 page
= eb
->pages
[index
];
4614 spin_lock(&page
->mapping
->private_lock
);
4616 * We do this since we'll remove the pages after we've
4617 * removed the eb from the radix tree, so we could race
4618 * and have this page now attached to the new eb. So
4619 * only clear page_private if it's still connected to
4622 if (PagePrivate(page
) &&
4623 page
->private == (unsigned long)eb
) {
4624 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4625 BUG_ON(PageDirty(page
));
4626 BUG_ON(PageWriteback(page
));
4628 * We need to make sure we haven't be attached
4631 ClearPagePrivate(page
);
4632 set_page_private(page
, 0);
4633 /* One for the page private */
4638 spin_unlock(&page
->mapping
->private_lock
);
4640 /* One for when we allocated the page */
4642 } while (index
!= 0);
4646 * Helper for releasing the extent buffer.
4648 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4650 btrfs_release_extent_buffer_page(eb
);
4651 __free_extent_buffer(eb
);
4654 static struct extent_buffer
*
4655 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4658 struct extent_buffer
*eb
= NULL
;
4660 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4663 eb
->fs_info
= fs_info
;
4665 rwlock_init(&eb
->lock
);
4666 atomic_set(&eb
->write_locks
, 0);
4667 atomic_set(&eb
->read_locks
, 0);
4668 atomic_set(&eb
->blocking_readers
, 0);
4669 atomic_set(&eb
->blocking_writers
, 0);
4670 atomic_set(&eb
->spinning_readers
, 0);
4671 atomic_set(&eb
->spinning_writers
, 0);
4672 eb
->lock_nested
= 0;
4673 init_waitqueue_head(&eb
->write_lock_wq
);
4674 init_waitqueue_head(&eb
->read_lock_wq
);
4676 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4678 spin_lock_init(&eb
->refs_lock
);
4679 atomic_set(&eb
->refs
, 1);
4680 atomic_set(&eb
->io_pages
, 0);
4683 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4685 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4686 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4687 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4692 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4696 struct extent_buffer
*new;
4697 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4699 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4703 for (i
= 0; i
< num_pages
; i
++) {
4704 p
= alloc_page(GFP_NOFS
);
4706 btrfs_release_extent_buffer(new);
4709 attach_extent_buffer_page(new, p
);
4710 WARN_ON(PageDirty(p
));
4713 copy_page(page_address(p
), page_address(src
->pages
[i
]));
4716 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4717 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4722 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4723 u64 start
, unsigned long len
)
4725 struct extent_buffer
*eb
;
4726 unsigned long num_pages
;
4729 num_pages
= num_extent_pages(start
, len
);
4731 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4735 for (i
= 0; i
< num_pages
; i
++) {
4736 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4740 set_extent_buffer_uptodate(eb
);
4741 btrfs_set_header_nritems(eb
, 0);
4742 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4747 __free_page(eb
->pages
[i
- 1]);
4748 __free_extent_buffer(eb
);
4752 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4755 return __alloc_dummy_extent_buffer(fs_info
, start
, fs_info
->nodesize
);
4758 static void check_buffer_tree_ref(struct extent_buffer
*eb
)
4761 /* the ref bit is tricky. We have to make sure it is set
4762 * if we have the buffer dirty. Otherwise the
4763 * code to free a buffer can end up dropping a dirty
4766 * Once the ref bit is set, it won't go away while the
4767 * buffer is dirty or in writeback, and it also won't
4768 * go away while we have the reference count on the
4771 * We can't just set the ref bit without bumping the
4772 * ref on the eb because free_extent_buffer might
4773 * see the ref bit and try to clear it. If this happens
4774 * free_extent_buffer might end up dropping our original
4775 * ref by mistake and freeing the page before we are able
4776 * to add one more ref.
4778 * So bump the ref count first, then set the bit. If someone
4779 * beat us to it, drop the ref we added.
4781 refs
= atomic_read(&eb
->refs
);
4782 if (refs
>= 2 && test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4785 spin_lock(&eb
->refs_lock
);
4786 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
4787 atomic_inc(&eb
->refs
);
4788 spin_unlock(&eb
->refs_lock
);
4791 static void mark_extent_buffer_accessed(struct extent_buffer
*eb
,
4792 struct page
*accessed
)
4794 unsigned long num_pages
, i
;
4796 check_buffer_tree_ref(eb
);
4798 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
4799 for (i
= 0; i
< num_pages
; i
++) {
4800 struct page
*p
= eb
->pages
[i
];
4803 mark_page_accessed(p
);
4807 struct extent_buffer
*find_extent_buffer(struct btrfs_fs_info
*fs_info
,
4810 struct extent_buffer
*eb
;
4813 eb
= radix_tree_lookup(&fs_info
->buffer_radix
,
4814 start
>> PAGE_SHIFT
);
4815 if (eb
&& atomic_inc_not_zero(&eb
->refs
)) {
4818 * Lock our eb's refs_lock to avoid races with
4819 * free_extent_buffer. When we get our eb it might be flagged
4820 * with EXTENT_BUFFER_STALE and another task running
4821 * free_extent_buffer might have seen that flag set,
4822 * eb->refs == 2, that the buffer isn't under IO (dirty and
4823 * writeback flags not set) and it's still in the tree (flag
4824 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4825 * of decrementing the extent buffer's reference count twice.
4826 * So here we could race and increment the eb's reference count,
4827 * clear its stale flag, mark it as dirty and drop our reference
4828 * before the other task finishes executing free_extent_buffer,
4829 * which would later result in an attempt to free an extent
4830 * buffer that is dirty.
4832 if (test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
)) {
4833 spin_lock(&eb
->refs_lock
);
4834 spin_unlock(&eb
->refs_lock
);
4836 mark_extent_buffer_accessed(eb
, NULL
);
4844 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4845 struct extent_buffer
*alloc_test_extent_buffer(struct btrfs_fs_info
*fs_info
,
4848 struct extent_buffer
*eb
, *exists
= NULL
;
4851 eb
= find_extent_buffer(fs_info
, start
);
4854 eb
= alloc_dummy_extent_buffer(fs_info
, start
);
4857 eb
->fs_info
= fs_info
;
4859 ret
= radix_tree_preload(GFP_NOFS
);
4862 spin_lock(&fs_info
->buffer_lock
);
4863 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4864 start
>> PAGE_SHIFT
, eb
);
4865 spin_unlock(&fs_info
->buffer_lock
);
4866 radix_tree_preload_end();
4867 if (ret
== -EEXIST
) {
4868 exists
= find_extent_buffer(fs_info
, start
);
4874 check_buffer_tree_ref(eb
);
4875 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4878 * We will free dummy extent buffer's if they come into
4879 * free_extent_buffer with a ref count of 2, but if we are using this we
4880 * want the buffers to stay in memory until we're done with them, so
4881 * bump the ref count again.
4883 atomic_inc(&eb
->refs
);
4886 btrfs_release_extent_buffer(eb
);
4891 struct extent_buffer
*alloc_extent_buffer(struct btrfs_fs_info
*fs_info
,
4894 unsigned long len
= fs_info
->nodesize
;
4895 unsigned long num_pages
= num_extent_pages(start
, len
);
4897 unsigned long index
= start
>> PAGE_SHIFT
;
4898 struct extent_buffer
*eb
;
4899 struct extent_buffer
*exists
= NULL
;
4901 struct address_space
*mapping
= fs_info
->btree_inode
->i_mapping
;
4905 if (!IS_ALIGNED(start
, fs_info
->sectorsize
)) {
4906 btrfs_err(fs_info
, "bad tree block start %llu", start
);
4907 return ERR_PTR(-EINVAL
);
4910 eb
= find_extent_buffer(fs_info
, start
);
4914 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4916 return ERR_PTR(-ENOMEM
);
4918 for (i
= 0; i
< num_pages
; i
++, index
++) {
4919 p
= find_or_create_page(mapping
, index
, GFP_NOFS
|__GFP_NOFAIL
);
4921 exists
= ERR_PTR(-ENOMEM
);
4925 spin_lock(&mapping
->private_lock
);
4926 if (PagePrivate(p
)) {
4928 * We could have already allocated an eb for this page
4929 * and attached one so lets see if we can get a ref on
4930 * the existing eb, and if we can we know it's good and
4931 * we can just return that one, else we know we can just
4932 * overwrite page->private.
4934 exists
= (struct extent_buffer
*)p
->private;
4935 if (atomic_inc_not_zero(&exists
->refs
)) {
4936 spin_unlock(&mapping
->private_lock
);
4939 mark_extent_buffer_accessed(exists
, p
);
4945 * Do this so attach doesn't complain and we need to
4946 * drop the ref the old guy had.
4948 ClearPagePrivate(p
);
4949 WARN_ON(PageDirty(p
));
4952 attach_extent_buffer_page(eb
, p
);
4953 spin_unlock(&mapping
->private_lock
);
4954 WARN_ON(PageDirty(p
));
4956 if (!PageUptodate(p
))
4960 * see below about how we avoid a nasty race with release page
4961 * and why we unlock later
4965 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
4967 ret
= radix_tree_preload(GFP_NOFS
);
4969 exists
= ERR_PTR(ret
);
4973 spin_lock(&fs_info
->buffer_lock
);
4974 ret
= radix_tree_insert(&fs_info
->buffer_radix
,
4975 start
>> PAGE_SHIFT
, eb
);
4976 spin_unlock(&fs_info
->buffer_lock
);
4977 radix_tree_preload_end();
4978 if (ret
== -EEXIST
) {
4979 exists
= find_extent_buffer(fs_info
, start
);
4985 /* add one reference for the tree */
4986 check_buffer_tree_ref(eb
);
4987 set_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
);
4990 * there is a race where release page may have
4991 * tried to find this extent buffer in the radix
4992 * but failed. It will tell the VM it is safe to
4993 * reclaim the, and it will clear the page private bit.
4994 * We must make sure to set the page private bit properly
4995 * after the extent buffer is in the radix tree so
4996 * it doesn't get lost
4998 SetPageChecked(eb
->pages
[0]);
4999 for (i
= 1; i
< num_pages
; i
++) {
5001 ClearPageChecked(p
);
5004 unlock_page(eb
->pages
[0]);
5008 WARN_ON(!atomic_dec_and_test(&eb
->refs
));
5009 for (i
= 0; i
< num_pages
; i
++) {
5011 unlock_page(eb
->pages
[i
]);
5014 btrfs_release_extent_buffer(eb
);
5018 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head
*head
)
5020 struct extent_buffer
*eb
=
5021 container_of(head
, struct extent_buffer
, rcu_head
);
5023 __free_extent_buffer(eb
);
5026 /* Expects to have eb->eb_lock already held */
5027 static int release_extent_buffer(struct extent_buffer
*eb
)
5029 WARN_ON(atomic_read(&eb
->refs
) == 0);
5030 if (atomic_dec_and_test(&eb
->refs
)) {
5031 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE
, &eb
->bflags
)) {
5032 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
5034 spin_unlock(&eb
->refs_lock
);
5036 spin_lock(&fs_info
->buffer_lock
);
5037 radix_tree_delete(&fs_info
->buffer_radix
,
5038 eb
->start
>> PAGE_SHIFT
);
5039 spin_unlock(&fs_info
->buffer_lock
);
5041 spin_unlock(&eb
->refs_lock
);
5044 /* Should be safe to release our pages at this point */
5045 btrfs_release_extent_buffer_page(eb
);
5046 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5047 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))) {
5048 __free_extent_buffer(eb
);
5052 call_rcu(&eb
->rcu_head
, btrfs_release_extent_buffer_rcu
);
5055 spin_unlock(&eb
->refs_lock
);
5060 void free_extent_buffer(struct extent_buffer
*eb
)
5068 refs
= atomic_read(&eb
->refs
);
5071 old
= atomic_cmpxchg(&eb
->refs
, refs
, refs
- 1);
5076 spin_lock(&eb
->refs_lock
);
5077 if (atomic_read(&eb
->refs
) == 2 &&
5078 test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
))
5079 atomic_dec(&eb
->refs
);
5081 if (atomic_read(&eb
->refs
) == 2 &&
5082 test_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
) &&
5083 !extent_buffer_under_io(eb
) &&
5084 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5085 atomic_dec(&eb
->refs
);
5088 * I know this is terrible, but it's temporary until we stop tracking
5089 * the uptodate bits and such for the extent buffers.
5091 release_extent_buffer(eb
);
5094 void free_extent_buffer_stale(struct extent_buffer
*eb
)
5099 spin_lock(&eb
->refs_lock
);
5100 set_bit(EXTENT_BUFFER_STALE
, &eb
->bflags
);
5102 if (atomic_read(&eb
->refs
) == 2 && !extent_buffer_under_io(eb
) &&
5103 test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
))
5104 atomic_dec(&eb
->refs
);
5105 release_extent_buffer(eb
);
5108 void clear_extent_buffer_dirty(struct extent_buffer
*eb
)
5111 unsigned long num_pages
;
5114 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5116 for (i
= 0; i
< num_pages
; i
++) {
5117 page
= eb
->pages
[i
];
5118 if (!PageDirty(page
))
5122 WARN_ON(!PagePrivate(page
));
5124 clear_page_dirty_for_io(page
);
5125 spin_lock_irq(&page
->mapping
->tree_lock
);
5126 if (!PageDirty(page
)) {
5127 radix_tree_tag_clear(&page
->mapping
->page_tree
,
5129 PAGECACHE_TAG_DIRTY
);
5131 spin_unlock_irq(&page
->mapping
->tree_lock
);
5132 ClearPageError(page
);
5135 WARN_ON(atomic_read(&eb
->refs
) == 0);
5138 int set_extent_buffer_dirty(struct extent_buffer
*eb
)
5141 unsigned long num_pages
;
5144 check_buffer_tree_ref(eb
);
5146 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
5148 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5149 WARN_ON(atomic_read(&eb
->refs
) == 0);
5150 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
));
5152 for (i
= 0; i
< num_pages
; i
++)
5153 set_page_dirty(eb
->pages
[i
]);
5157 void clear_extent_buffer_uptodate(struct extent_buffer
*eb
)
5161 unsigned long num_pages
;
5163 clear_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5164 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5165 for (i
= 0; i
< num_pages
; i
++) {
5166 page
= eb
->pages
[i
];
5168 ClearPageUptodate(page
);
5172 void set_extent_buffer_uptodate(struct extent_buffer
*eb
)
5176 unsigned long num_pages
;
5178 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5179 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5180 for (i
= 0; i
< num_pages
; i
++) {
5181 page
= eb
->pages
[i
];
5182 SetPageUptodate(page
);
5186 int extent_buffer_uptodate(struct extent_buffer
*eb
)
5188 return test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5191 int read_extent_buffer_pages(struct extent_io_tree
*tree
,
5192 struct extent_buffer
*eb
, int wait
,
5193 get_extent_t
*get_extent
, int mirror_num
)
5199 int locked_pages
= 0;
5200 int all_uptodate
= 1;
5201 unsigned long num_pages
;
5202 unsigned long num_reads
= 0;
5203 struct bio
*bio
= NULL
;
5204 unsigned long bio_flags
= 0;
5206 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5209 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5210 for (i
= 0; i
< num_pages
; i
++) {
5211 page
= eb
->pages
[i
];
5212 if (wait
== WAIT_NONE
) {
5213 if (!trylock_page(page
))
5221 * We need to firstly lock all pages to make sure that
5222 * the uptodate bit of our pages won't be affected by
5223 * clear_extent_buffer_uptodate().
5225 for (i
= 0; i
< num_pages
; i
++) {
5226 page
= eb
->pages
[i
];
5227 if (!PageUptodate(page
)) {
5234 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5238 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5239 eb
->read_mirror
= 0;
5240 atomic_set(&eb
->io_pages
, num_reads
);
5241 for (i
= 0; i
< num_pages
; i
++) {
5242 page
= eb
->pages
[i
];
5244 if (!PageUptodate(page
)) {
5246 atomic_dec(&eb
->io_pages
);
5251 ClearPageError(page
);
5252 err
= __extent_read_full_page(tree
, page
,
5254 mirror_num
, &bio_flags
,
5259 * We use &bio in above __extent_read_full_page,
5260 * so we ensure that if it returns error, the
5261 * current page fails to add itself to bio and
5262 * it's been unlocked.
5264 * We must dec io_pages by ourselves.
5266 atomic_dec(&eb
->io_pages
);
5274 err
= submit_one_bio(bio
, mirror_num
, bio_flags
);
5279 if (ret
|| wait
!= WAIT_COMPLETE
)
5282 for (i
= 0; i
< num_pages
; i
++) {
5283 page
= eb
->pages
[i
];
5284 wait_on_page_locked(page
);
5285 if (!PageUptodate(page
))
5292 while (locked_pages
> 0) {
5294 page
= eb
->pages
[locked_pages
];
5300 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5301 unsigned long start
,
5308 char *dst
= (char *)dstv
;
5309 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5310 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5312 WARN_ON(start
> eb
->len
);
5313 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5315 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5318 page
= eb
->pages
[i
];
5320 cur
= min(len
, (PAGE_SIZE
- offset
));
5321 kaddr
= page_address(page
);
5322 memcpy(dst
, kaddr
+ offset
, cur
);
5331 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5332 unsigned long start
,
5339 char __user
*dst
= (char __user
*)dstv
;
5340 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5341 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5344 WARN_ON(start
> eb
->len
);
5345 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5347 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5350 page
= eb
->pages
[i
];
5352 cur
= min(len
, (PAGE_SIZE
- offset
));
5353 kaddr
= page_address(page
);
5354 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5369 * return 0 if the item is found within a page.
5370 * return 1 if the item spans two pages.
5371 * return -EINVAL otherwise.
5373 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5374 unsigned long min_len
, char **map
,
5375 unsigned long *map_start
,
5376 unsigned long *map_len
)
5378 size_t offset
= start
& (PAGE_SIZE
- 1);
5381 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5382 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5383 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5390 offset
= start_offset
;
5394 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5397 if (start
+ min_len
> eb
->len
) {
5398 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5399 eb
->start
, eb
->len
, start
, min_len
);
5404 kaddr
= page_address(p
);
5405 *map
= kaddr
+ offset
;
5406 *map_len
= PAGE_SIZE
- offset
;
5410 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5411 unsigned long start
,
5418 char *ptr
= (char *)ptrv
;
5419 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5420 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5423 WARN_ON(start
> eb
->len
);
5424 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5426 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5429 page
= eb
->pages
[i
];
5431 cur
= min(len
, (PAGE_SIZE
- offset
));
5433 kaddr
= page_address(page
);
5434 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5446 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer
*eb
,
5451 WARN_ON(!PageUptodate(eb
->pages
[0]));
5452 kaddr
= page_address(eb
->pages
[0]);
5453 memcpy(kaddr
+ offsetof(struct btrfs_header
, chunk_tree_uuid
), srcv
,
5457 void write_extent_buffer_fsid(struct extent_buffer
*eb
, const void *srcv
)
5461 WARN_ON(!PageUptodate(eb
->pages
[0]));
5462 kaddr
= page_address(eb
->pages
[0]);
5463 memcpy(kaddr
+ offsetof(struct btrfs_header
, fsid
), srcv
,
5467 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5468 unsigned long start
, unsigned long len
)
5474 char *src
= (char *)srcv
;
5475 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5476 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5478 WARN_ON(start
> eb
->len
);
5479 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5481 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5484 page
= eb
->pages
[i
];
5485 WARN_ON(!PageUptodate(page
));
5487 cur
= min(len
, PAGE_SIZE
- offset
);
5488 kaddr
= page_address(page
);
5489 memcpy(kaddr
+ offset
, src
, cur
);
5498 void memzero_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5505 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5506 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5508 WARN_ON(start
> eb
->len
);
5509 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5511 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5514 page
= eb
->pages
[i
];
5515 WARN_ON(!PageUptodate(page
));
5517 cur
= min(len
, PAGE_SIZE
- offset
);
5518 kaddr
= page_address(page
);
5519 memset(kaddr
+ offset
, 0, cur
);
5527 void copy_extent_buffer_full(struct extent_buffer
*dst
,
5528 struct extent_buffer
*src
)
5533 ASSERT(dst
->len
== src
->len
);
5535 num_pages
= num_extent_pages(dst
->start
, dst
->len
);
5536 for (i
= 0; i
< num_pages
; i
++)
5537 copy_page(page_address(dst
->pages
[i
]),
5538 page_address(src
->pages
[i
]));
5541 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5542 unsigned long dst_offset
, unsigned long src_offset
,
5545 u64 dst_len
= dst
->len
;
5550 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5551 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5553 WARN_ON(src
->len
!= dst_len
);
5555 offset
= (start_offset
+ dst_offset
) &
5559 page
= dst
->pages
[i
];
5560 WARN_ON(!PageUptodate(page
));
5562 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5564 kaddr
= page_address(page
);
5565 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5574 void le_bitmap_set(u8
*map
, unsigned int start
, int len
)
5576 u8
*p
= map
+ BIT_BYTE(start
);
5577 const unsigned int size
= start
+ len
;
5578 int bits_to_set
= BITS_PER_BYTE
- (start
% BITS_PER_BYTE
);
5579 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(start
);
5581 while (len
- bits_to_set
>= 0) {
5584 bits_to_set
= BITS_PER_BYTE
;
5589 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5594 void le_bitmap_clear(u8
*map
, unsigned int start
, int len
)
5596 u8
*p
= map
+ BIT_BYTE(start
);
5597 const unsigned int size
= start
+ len
;
5598 int bits_to_clear
= BITS_PER_BYTE
- (start
% BITS_PER_BYTE
);
5599 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(start
);
5601 while (len
- bits_to_clear
>= 0) {
5602 *p
&= ~mask_to_clear
;
5603 len
-= bits_to_clear
;
5604 bits_to_clear
= BITS_PER_BYTE
;
5609 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5610 *p
&= ~mask_to_clear
;
5615 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5617 * @eb: the extent buffer
5618 * @start: offset of the bitmap item in the extent buffer
5620 * @page_index: return index of the page in the extent buffer that contains the
5622 * @page_offset: return offset into the page given by page_index
5624 * This helper hides the ugliness of finding the byte in an extent buffer which
5625 * contains a given bit.
5627 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5628 unsigned long start
, unsigned long nr
,
5629 unsigned long *page_index
,
5630 size_t *page_offset
)
5632 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5633 size_t byte_offset
= BIT_BYTE(nr
);
5637 * The byte we want is the offset of the extent buffer + the offset of
5638 * the bitmap item in the extent buffer + the offset of the byte in the
5641 offset
= start_offset
+ start
+ byte_offset
;
5643 *page_index
= offset
>> PAGE_SHIFT
;
5644 *page_offset
= offset
& (PAGE_SIZE
- 1);
5648 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5649 * @eb: the extent buffer
5650 * @start: offset of the bitmap item in the extent buffer
5651 * @nr: bit number to test
5653 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5661 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5662 page
= eb
->pages
[i
];
5663 WARN_ON(!PageUptodate(page
));
5664 kaddr
= page_address(page
);
5665 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5669 * extent_buffer_bitmap_set - set an area of a bitmap
5670 * @eb: the extent buffer
5671 * @start: offset of the bitmap item in the extent buffer
5672 * @pos: bit number of the first bit
5673 * @len: number of bits to set
5675 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5676 unsigned long pos
, unsigned long len
)
5682 const unsigned int size
= pos
+ len
;
5683 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5684 u8 mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5686 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5687 page
= eb
->pages
[i
];
5688 WARN_ON(!PageUptodate(page
));
5689 kaddr
= page_address(page
);
5691 while (len
>= bits_to_set
) {
5692 kaddr
[offset
] |= mask_to_set
;
5694 bits_to_set
= BITS_PER_BYTE
;
5696 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5698 page
= eb
->pages
[++i
];
5699 WARN_ON(!PageUptodate(page
));
5700 kaddr
= page_address(page
);
5704 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5705 kaddr
[offset
] |= mask_to_set
;
5711 * extent_buffer_bitmap_clear - clear an area of a bitmap
5712 * @eb: the extent buffer
5713 * @start: offset of the bitmap item in the extent buffer
5714 * @pos: bit number of the first bit
5715 * @len: number of bits to clear
5717 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5718 unsigned long pos
, unsigned long len
)
5724 const unsigned int size
= pos
+ len
;
5725 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5726 u8 mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5728 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5729 page
= eb
->pages
[i
];
5730 WARN_ON(!PageUptodate(page
));
5731 kaddr
= page_address(page
);
5733 while (len
>= bits_to_clear
) {
5734 kaddr
[offset
] &= ~mask_to_clear
;
5735 len
-= bits_to_clear
;
5736 bits_to_clear
= BITS_PER_BYTE
;
5738 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5740 page
= eb
->pages
[++i
];
5741 WARN_ON(!PageUptodate(page
));
5742 kaddr
= page_address(page
);
5746 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5747 kaddr
[offset
] &= ~mask_to_clear
;
5751 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5753 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5754 return distance
< len
;
5757 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5758 unsigned long dst_off
, unsigned long src_off
,
5761 char *dst_kaddr
= page_address(dst_page
);
5763 int must_memmove
= 0;
5765 if (dst_page
!= src_page
) {
5766 src_kaddr
= page_address(src_page
);
5768 src_kaddr
= dst_kaddr
;
5769 if (areas_overlap(src_off
, dst_off
, len
))
5774 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5776 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5779 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5780 unsigned long src_offset
, unsigned long len
)
5782 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5784 size_t dst_off_in_page
;
5785 size_t src_off_in_page
;
5786 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5787 unsigned long dst_i
;
5788 unsigned long src_i
;
5790 if (src_offset
+ len
> dst
->len
) {
5792 "memmove bogus src_offset %lu move len %lu dst len %lu",
5793 src_offset
, len
, dst
->len
);
5796 if (dst_offset
+ len
> dst
->len
) {
5798 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5799 dst_offset
, len
, dst
->len
);
5804 dst_off_in_page
= (start_offset
+ dst_offset
) &
5806 src_off_in_page
= (start_offset
+ src_offset
) &
5809 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5810 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5812 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5814 cur
= min_t(unsigned long, cur
,
5815 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5817 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5818 dst_off_in_page
, src_off_in_page
, cur
);
5826 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5827 unsigned long src_offset
, unsigned long len
)
5829 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
5831 size_t dst_off_in_page
;
5832 size_t src_off_in_page
;
5833 unsigned long dst_end
= dst_offset
+ len
- 1;
5834 unsigned long src_end
= src_offset
+ len
- 1;
5835 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5836 unsigned long dst_i
;
5837 unsigned long src_i
;
5839 if (src_offset
+ len
> dst
->len
) {
5841 "memmove bogus src_offset %lu move len %lu len %lu",
5842 src_offset
, len
, dst
->len
);
5845 if (dst_offset
+ len
> dst
->len
) {
5847 "memmove bogus dst_offset %lu move len %lu len %lu",
5848 dst_offset
, len
, dst
->len
);
5851 if (dst_offset
< src_offset
) {
5852 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5856 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5857 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5859 dst_off_in_page
= (start_offset
+ dst_end
) &
5861 src_off_in_page
= (start_offset
+ src_end
) &
5864 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5865 cur
= min(cur
, dst_off_in_page
+ 1);
5866 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5867 dst_off_in_page
- cur
+ 1,
5868 src_off_in_page
- cur
+ 1, cur
);
5876 int try_release_extent_buffer(struct page
*page
)
5878 struct extent_buffer
*eb
;
5881 * We need to make sure nobody is attaching this page to an eb right
5884 spin_lock(&page
->mapping
->private_lock
);
5885 if (!PagePrivate(page
)) {
5886 spin_unlock(&page
->mapping
->private_lock
);
5890 eb
= (struct extent_buffer
*)page
->private;
5894 * This is a little awful but should be ok, we need to make sure that
5895 * the eb doesn't disappear out from under us while we're looking at
5898 spin_lock(&eb
->refs_lock
);
5899 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5900 spin_unlock(&eb
->refs_lock
);
5901 spin_unlock(&page
->mapping
->private_lock
);
5904 spin_unlock(&page
->mapping
->private_lock
);
5907 * If tree ref isn't set then we know the ref on this eb is a real ref,
5908 * so just return, this page will likely be freed soon anyway.
5910 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5911 spin_unlock(&eb
->refs_lock
);
5915 return release_extent_buffer(eb
);