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 printk(KERN_ERR
"BTRFS: buffer leak start %llu len %lu "
80 eb
->start
, eb
->len
, atomic_read(&eb
->refs
));
81 list_del(&eb
->leak_list
);
82 kmem_cache_free(extent_buffer_cache
, eb
);
86 #define btrfs_debug_check_extent_io_range(tree, start, end) \
87 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
88 static inline void __btrfs_debug_check_extent_io_range(const char *caller
,
89 struct extent_io_tree
*tree
, u64 start
, u64 end
)
97 inode
= tree
->mapping
->host
;
98 isize
= i_size_read(inode
);
99 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
100 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
101 "%s: ino %llu isize %llu odd range [%llu,%llu]",
102 caller
, btrfs_ino(inode
), isize
, start
, end
);
106 #define btrfs_leak_debug_add(new, head) do {} while (0)
107 #define btrfs_leak_debug_del(entry) do {} while (0)
108 #define btrfs_leak_debug_check() do {} while (0)
109 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
112 #define BUFFER_LRU_MAX 64
117 struct rb_node rb_node
;
120 struct extent_page_data
{
122 struct extent_io_tree
*tree
;
123 get_extent_t
*get_extent
;
124 unsigned long bio_flags
;
126 /* tells writepage not to lock the state bits for this range
127 * it still does the unlocking
129 unsigned int extent_locked
:1;
131 /* tells the submit_bio code to use a WRITE_SYNC */
132 unsigned int sync_io
:1;
135 static void add_extent_changeset(struct extent_state
*state
, unsigned bits
,
136 struct extent_changeset
*changeset
,
143 if (set
&& (state
->state
& bits
) == bits
)
145 if (!set
&& (state
->state
& bits
) == 0)
147 changeset
->bytes_changed
+= state
->end
- state
->start
+ 1;
148 ret
= ulist_add(changeset
->range_changed
, state
->start
, state
->end
,
154 static noinline
void flush_write_bio(void *data
);
155 static inline struct btrfs_fs_info
*
156 tree_fs_info(struct extent_io_tree
*tree
)
160 return btrfs_sb(tree
->mapping
->host
->i_sb
);
163 int __init
extent_io_init(void)
165 extent_state_cache
= kmem_cache_create("btrfs_extent_state",
166 sizeof(struct extent_state
), 0,
167 SLAB_MEM_SPREAD
, NULL
);
168 if (!extent_state_cache
)
171 extent_buffer_cache
= kmem_cache_create("btrfs_extent_buffer",
172 sizeof(struct extent_buffer
), 0,
173 SLAB_MEM_SPREAD
, NULL
);
174 if (!extent_buffer_cache
)
175 goto free_state_cache
;
177 btrfs_bioset
= bioset_create(BIO_POOL_SIZE
,
178 offsetof(struct btrfs_io_bio
, bio
));
180 goto free_buffer_cache
;
182 if (bioset_integrity_create(btrfs_bioset
, BIO_POOL_SIZE
))
188 bioset_free(btrfs_bioset
);
192 kmem_cache_destroy(extent_buffer_cache
);
193 extent_buffer_cache
= NULL
;
196 kmem_cache_destroy(extent_state_cache
);
197 extent_state_cache
= NULL
;
201 void extent_io_exit(void)
203 btrfs_leak_debug_check();
206 * Make sure all delayed rcu free are flushed before we
210 kmem_cache_destroy(extent_state_cache
);
211 kmem_cache_destroy(extent_buffer_cache
);
213 bioset_free(btrfs_bioset
);
216 void extent_io_tree_init(struct extent_io_tree
*tree
,
217 struct address_space
*mapping
)
219 tree
->state
= RB_ROOT
;
221 tree
->dirty_bytes
= 0;
222 spin_lock_init(&tree
->lock
);
223 tree
->mapping
= mapping
;
226 static struct extent_state
*alloc_extent_state(gfp_t mask
)
228 struct extent_state
*state
;
230 state
= kmem_cache_alloc(extent_state_cache
, mask
);
234 state
->failrec
= NULL
;
235 RB_CLEAR_NODE(&state
->rb_node
);
236 btrfs_leak_debug_add(&state
->leak_list
, &states
);
237 atomic_set(&state
->refs
, 1);
238 init_waitqueue_head(&state
->wq
);
239 trace_alloc_extent_state(state
, mask
, _RET_IP_
);
243 void free_extent_state(struct extent_state
*state
)
247 if (atomic_dec_and_test(&state
->refs
)) {
248 WARN_ON(extent_state_in_tree(state
));
249 btrfs_leak_debug_del(&state
->leak_list
);
250 trace_free_extent_state(state
, _RET_IP_
);
251 kmem_cache_free(extent_state_cache
, state
);
255 static struct rb_node
*tree_insert(struct rb_root
*root
,
256 struct rb_node
*search_start
,
258 struct rb_node
*node
,
259 struct rb_node
***p_in
,
260 struct rb_node
**parent_in
)
263 struct rb_node
*parent
= NULL
;
264 struct tree_entry
*entry
;
266 if (p_in
&& parent_in
) {
272 p
= search_start
? &search_start
: &root
->rb_node
;
275 entry
= rb_entry(parent
, struct tree_entry
, rb_node
);
277 if (offset
< entry
->start
)
279 else if (offset
> entry
->end
)
286 rb_link_node(node
, parent
, p
);
287 rb_insert_color(node
, root
);
291 static struct rb_node
*__etree_search(struct extent_io_tree
*tree
, u64 offset
,
292 struct rb_node
**prev_ret
,
293 struct rb_node
**next_ret
,
294 struct rb_node
***p_ret
,
295 struct rb_node
**parent_ret
)
297 struct rb_root
*root
= &tree
->state
;
298 struct rb_node
**n
= &root
->rb_node
;
299 struct rb_node
*prev
= NULL
;
300 struct rb_node
*orig_prev
= NULL
;
301 struct tree_entry
*entry
;
302 struct tree_entry
*prev_entry
= NULL
;
306 entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
309 if (offset
< entry
->start
)
311 else if (offset
> entry
->end
)
324 while (prev
&& offset
> prev_entry
->end
) {
325 prev
= rb_next(prev
);
326 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
333 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
334 while (prev
&& offset
< prev_entry
->start
) {
335 prev
= rb_prev(prev
);
336 prev_entry
= rb_entry(prev
, struct tree_entry
, rb_node
);
343 static inline struct rb_node
*
344 tree_search_for_insert(struct extent_io_tree
*tree
,
346 struct rb_node
***p_ret
,
347 struct rb_node
**parent_ret
)
349 struct rb_node
*prev
= NULL
;
352 ret
= __etree_search(tree
, offset
, &prev
, NULL
, p_ret
, parent_ret
);
358 static inline struct rb_node
*tree_search(struct extent_io_tree
*tree
,
361 return tree_search_for_insert(tree
, offset
, NULL
, NULL
);
364 static void merge_cb(struct extent_io_tree
*tree
, struct extent_state
*new,
365 struct extent_state
*other
)
367 if (tree
->ops
&& tree
->ops
->merge_extent_hook
)
368 tree
->ops
->merge_extent_hook(tree
->mapping
->host
, new,
373 * utility function to look for merge candidates inside a given range.
374 * Any extents with matching state are merged together into a single
375 * extent in the tree. Extents with EXTENT_IO in their state field
376 * are not merged because the end_io handlers need to be able to do
377 * operations on them without sleeping (or doing allocations/splits).
379 * This should be called with the tree lock held.
381 static void merge_state(struct extent_io_tree
*tree
,
382 struct extent_state
*state
)
384 struct extent_state
*other
;
385 struct rb_node
*other_node
;
387 if (state
->state
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
390 other_node
= rb_prev(&state
->rb_node
);
392 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
393 if (other
->end
== state
->start
- 1 &&
394 other
->state
== state
->state
) {
395 merge_cb(tree
, state
, other
);
396 state
->start
= other
->start
;
397 rb_erase(&other
->rb_node
, &tree
->state
);
398 RB_CLEAR_NODE(&other
->rb_node
);
399 free_extent_state(other
);
402 other_node
= rb_next(&state
->rb_node
);
404 other
= rb_entry(other_node
, struct extent_state
, rb_node
);
405 if (other
->start
== state
->end
+ 1 &&
406 other
->state
== state
->state
) {
407 merge_cb(tree
, state
, other
);
408 state
->end
= other
->end
;
409 rb_erase(&other
->rb_node
, &tree
->state
);
410 RB_CLEAR_NODE(&other
->rb_node
);
411 free_extent_state(other
);
416 static void set_state_cb(struct extent_io_tree
*tree
,
417 struct extent_state
*state
, unsigned *bits
)
419 if (tree
->ops
&& tree
->ops
->set_bit_hook
)
420 tree
->ops
->set_bit_hook(tree
->mapping
->host
, state
, bits
);
423 static void clear_state_cb(struct extent_io_tree
*tree
,
424 struct extent_state
*state
, unsigned *bits
)
426 if (tree
->ops
&& tree
->ops
->clear_bit_hook
)
427 tree
->ops
->clear_bit_hook(tree
->mapping
->host
, state
, bits
);
430 static void set_state_bits(struct extent_io_tree
*tree
,
431 struct extent_state
*state
, unsigned *bits
,
432 struct extent_changeset
*changeset
);
435 * insert an extent_state struct into the tree. 'bits' are set on the
436 * struct before it is inserted.
438 * This may return -EEXIST if the extent is already there, in which case the
439 * state struct is freed.
441 * The tree lock is not taken internally. This is a utility function and
442 * probably isn't what you want to call (see set/clear_extent_bit).
444 static int insert_state(struct extent_io_tree
*tree
,
445 struct extent_state
*state
, u64 start
, u64 end
,
447 struct rb_node
**parent
,
448 unsigned *bits
, struct extent_changeset
*changeset
)
450 struct rb_node
*node
;
453 WARN(1, KERN_ERR
"BTRFS: end < start %llu %llu\n",
455 state
->start
= start
;
458 set_state_bits(tree
, state
, bits
, changeset
);
460 node
= tree_insert(&tree
->state
, NULL
, end
, &state
->rb_node
, p
, parent
);
462 struct extent_state
*found
;
463 found
= rb_entry(node
, struct extent_state
, rb_node
);
464 printk(KERN_ERR
"BTRFS: found node %llu %llu on insert of "
466 found
->start
, found
->end
, start
, end
);
469 merge_state(tree
, state
);
473 static void split_cb(struct extent_io_tree
*tree
, struct extent_state
*orig
,
476 if (tree
->ops
&& tree
->ops
->split_extent_hook
)
477 tree
->ops
->split_extent_hook(tree
->mapping
->host
, orig
, split
);
481 * split a given extent state struct in two, inserting the preallocated
482 * struct 'prealloc' as the newly created second half. 'split' indicates an
483 * offset inside 'orig' where it should be split.
486 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
487 * are two extent state structs in the tree:
488 * prealloc: [orig->start, split - 1]
489 * orig: [ split, orig->end ]
491 * The tree locks are not taken by this function. They need to be held
494 static int split_state(struct extent_io_tree
*tree
, struct extent_state
*orig
,
495 struct extent_state
*prealloc
, u64 split
)
497 struct rb_node
*node
;
499 split_cb(tree
, orig
, split
);
501 prealloc
->start
= orig
->start
;
502 prealloc
->end
= split
- 1;
503 prealloc
->state
= orig
->state
;
506 node
= tree_insert(&tree
->state
, &orig
->rb_node
, prealloc
->end
,
507 &prealloc
->rb_node
, NULL
, NULL
);
509 free_extent_state(prealloc
);
515 static struct extent_state
*next_state(struct extent_state
*state
)
517 struct rb_node
*next
= rb_next(&state
->rb_node
);
519 return rb_entry(next
, struct extent_state
, rb_node
);
525 * utility function to clear some bits in an extent state struct.
526 * it will optionally wake up any one waiting on this state (wake == 1).
528 * If no bits are set on the state struct after clearing things, the
529 * struct is freed and removed from the tree
531 static struct extent_state
*clear_state_bit(struct extent_io_tree
*tree
,
532 struct extent_state
*state
,
533 unsigned *bits
, int wake
,
534 struct extent_changeset
*changeset
)
536 struct extent_state
*next
;
537 unsigned bits_to_clear
= *bits
& ~EXTENT_CTLBITS
;
539 if ((bits_to_clear
& EXTENT_DIRTY
) && (state
->state
& EXTENT_DIRTY
)) {
540 u64 range
= state
->end
- state
->start
+ 1;
541 WARN_ON(range
> tree
->dirty_bytes
);
542 tree
->dirty_bytes
-= range
;
544 clear_state_cb(tree
, state
, bits
);
545 add_extent_changeset(state
, bits_to_clear
, changeset
, 0);
546 state
->state
&= ~bits_to_clear
;
549 if (state
->state
== 0) {
550 next
= next_state(state
);
551 if (extent_state_in_tree(state
)) {
552 rb_erase(&state
->rb_node
, &tree
->state
);
553 RB_CLEAR_NODE(&state
->rb_node
);
554 free_extent_state(state
);
559 merge_state(tree
, state
);
560 next
= next_state(state
);
565 static struct extent_state
*
566 alloc_extent_state_atomic(struct extent_state
*prealloc
)
569 prealloc
= alloc_extent_state(GFP_ATOMIC
);
574 static void extent_io_tree_panic(struct extent_io_tree
*tree
, int err
)
576 btrfs_panic(tree_fs_info(tree
), err
, "Locking error: "
577 "Extent tree was modified by another "
578 "thread while locked.");
582 * clear some bits on a range in the tree. This may require splitting
583 * or inserting elements in the tree, so the gfp mask is used to
584 * indicate which allocations or sleeping are allowed.
586 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
587 * the given range from the tree regardless of state (ie for truncate).
589 * the range [start, end] is inclusive.
591 * This takes the tree lock, and returns 0 on success and < 0 on error.
593 static int __clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
594 unsigned bits
, int wake
, int delete,
595 struct extent_state
**cached_state
,
596 gfp_t mask
, struct extent_changeset
*changeset
)
598 struct extent_state
*state
;
599 struct extent_state
*cached
;
600 struct extent_state
*prealloc
= NULL
;
601 struct rb_node
*node
;
606 btrfs_debug_check_extent_io_range(tree
, start
, end
);
608 if (bits
& EXTENT_DELALLOC
)
609 bits
|= EXTENT_NORESERVE
;
612 bits
|= ~EXTENT_CTLBITS
;
613 bits
|= EXTENT_FIRST_DELALLOC
;
615 if (bits
& (EXTENT_IOBITS
| EXTENT_BOUNDARY
))
618 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
620 * Don't care for allocation failure here because we might end
621 * up not needing the pre-allocated extent state at all, which
622 * is the case if we only have in the tree extent states that
623 * cover our input range and don't cover too any other range.
624 * If we end up needing a new extent state we allocate it later.
626 prealloc
= alloc_extent_state(mask
);
629 spin_lock(&tree
->lock
);
631 cached
= *cached_state
;
634 *cached_state
= NULL
;
638 if (cached
&& extent_state_in_tree(cached
) &&
639 cached
->start
<= start
&& cached
->end
> start
) {
641 atomic_dec(&cached
->refs
);
646 free_extent_state(cached
);
649 * this search will find the extents that end after
652 node
= tree_search(tree
, start
);
655 state
= rb_entry(node
, struct extent_state
, rb_node
);
657 if (state
->start
> end
)
659 WARN_ON(state
->end
< start
);
660 last_end
= state
->end
;
662 /* the state doesn't have the wanted bits, go ahead */
663 if (!(state
->state
& bits
)) {
664 state
= next_state(state
);
669 * | ---- desired range ---- |
671 * | ------------- state -------------- |
673 * We need to split the extent we found, and may flip
674 * bits on second half.
676 * If the extent we found extends past our range, we
677 * just split and search again. It'll get split again
678 * the next time though.
680 * If the extent we found is inside our range, we clear
681 * the desired bit on it.
684 if (state
->start
< start
) {
685 prealloc
= alloc_extent_state_atomic(prealloc
);
687 err
= split_state(tree
, state
, prealloc
, start
);
689 extent_io_tree_panic(tree
, err
);
694 if (state
->end
<= end
) {
695 state
= clear_state_bit(tree
, state
, &bits
, wake
,
702 * | ---- desired range ---- |
704 * We need to split the extent, and clear the bit
707 if (state
->start
<= end
&& state
->end
> end
) {
708 prealloc
= alloc_extent_state_atomic(prealloc
);
710 err
= split_state(tree
, state
, prealloc
, end
+ 1);
712 extent_io_tree_panic(tree
, err
);
717 clear_state_bit(tree
, prealloc
, &bits
, wake
, changeset
);
723 state
= clear_state_bit(tree
, state
, &bits
, wake
, changeset
);
725 if (last_end
== (u64
)-1)
727 start
= last_end
+ 1;
728 if (start
<= end
&& state
&& !need_resched())
734 spin_unlock(&tree
->lock
);
735 if (gfpflags_allow_blocking(mask
))
740 spin_unlock(&tree
->lock
);
742 free_extent_state(prealloc
);
748 static void wait_on_state(struct extent_io_tree
*tree
,
749 struct extent_state
*state
)
750 __releases(tree
->lock
)
751 __acquires(tree
->lock
)
754 prepare_to_wait(&state
->wq
, &wait
, TASK_UNINTERRUPTIBLE
);
755 spin_unlock(&tree
->lock
);
757 spin_lock(&tree
->lock
);
758 finish_wait(&state
->wq
, &wait
);
762 * waits for one or more bits to clear on a range in the state tree.
763 * The range [start, end] is inclusive.
764 * The tree lock is taken by this function
766 static void wait_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
769 struct extent_state
*state
;
770 struct rb_node
*node
;
772 btrfs_debug_check_extent_io_range(tree
, start
, end
);
774 spin_lock(&tree
->lock
);
778 * this search will find all the extents that end after
781 node
= tree_search(tree
, start
);
786 state
= rb_entry(node
, struct extent_state
, rb_node
);
788 if (state
->start
> end
)
791 if (state
->state
& bits
) {
792 start
= state
->start
;
793 atomic_inc(&state
->refs
);
794 wait_on_state(tree
, state
);
795 free_extent_state(state
);
798 start
= state
->end
+ 1;
803 if (!cond_resched_lock(&tree
->lock
)) {
804 node
= rb_next(node
);
809 spin_unlock(&tree
->lock
);
812 static void set_state_bits(struct extent_io_tree
*tree
,
813 struct extent_state
*state
,
814 unsigned *bits
, struct extent_changeset
*changeset
)
816 unsigned bits_to_set
= *bits
& ~EXTENT_CTLBITS
;
818 set_state_cb(tree
, state
, bits
);
819 if ((bits_to_set
& EXTENT_DIRTY
) && !(state
->state
& EXTENT_DIRTY
)) {
820 u64 range
= state
->end
- state
->start
+ 1;
821 tree
->dirty_bytes
+= range
;
823 add_extent_changeset(state
, bits_to_set
, changeset
, 1);
824 state
->state
|= bits_to_set
;
827 static void cache_state_if_flags(struct extent_state
*state
,
828 struct extent_state
**cached_ptr
,
831 if (cached_ptr
&& !(*cached_ptr
)) {
832 if (!flags
|| (state
->state
& flags
)) {
834 atomic_inc(&state
->refs
);
839 static void cache_state(struct extent_state
*state
,
840 struct extent_state
**cached_ptr
)
842 return cache_state_if_flags(state
, cached_ptr
,
843 EXTENT_IOBITS
| EXTENT_BOUNDARY
);
847 * set some bits on a range in the tree. This may require allocations or
848 * sleeping, so the gfp mask is used to indicate what is allowed.
850 * If any of the exclusive bits are set, this will fail with -EEXIST if some
851 * part of the range already has the desired bits set. The start of the
852 * existing range is returned in failed_start in this case.
854 * [start, end] is inclusive This takes the tree lock.
857 static int __must_check
858 __set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
859 unsigned bits
, unsigned exclusive_bits
,
860 u64
*failed_start
, struct extent_state
**cached_state
,
861 gfp_t mask
, struct extent_changeset
*changeset
)
863 struct extent_state
*state
;
864 struct extent_state
*prealloc
= NULL
;
865 struct rb_node
*node
;
867 struct rb_node
*parent
;
872 btrfs_debug_check_extent_io_range(tree
, start
, end
);
874 bits
|= EXTENT_FIRST_DELALLOC
;
876 if (!prealloc
&& gfpflags_allow_blocking(mask
)) {
878 * Don't care for allocation failure here because we might end
879 * up not needing the pre-allocated extent state at all, which
880 * is the case if we only have in the tree extent states that
881 * cover our input range and don't cover too any other range.
882 * If we end up needing a new extent state we allocate it later.
884 prealloc
= alloc_extent_state(mask
);
887 spin_lock(&tree
->lock
);
888 if (cached_state
&& *cached_state
) {
889 state
= *cached_state
;
890 if (state
->start
<= start
&& state
->end
> start
&&
891 extent_state_in_tree(state
)) {
892 node
= &state
->rb_node
;
897 * this search will find all the extents that end after
900 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
902 prealloc
= alloc_extent_state_atomic(prealloc
);
904 err
= insert_state(tree
, prealloc
, start
, end
,
905 &p
, &parent
, &bits
, changeset
);
907 extent_io_tree_panic(tree
, err
);
909 cache_state(prealloc
, cached_state
);
913 state
= rb_entry(node
, struct extent_state
, rb_node
);
915 last_start
= state
->start
;
916 last_end
= state
->end
;
919 * | ---- desired range ---- |
922 * Just lock what we found and keep going
924 if (state
->start
== start
&& state
->end
<= end
) {
925 if (state
->state
& exclusive_bits
) {
926 *failed_start
= state
->start
;
931 set_state_bits(tree
, state
, &bits
, changeset
);
932 cache_state(state
, cached_state
);
933 merge_state(tree
, state
);
934 if (last_end
== (u64
)-1)
936 start
= last_end
+ 1;
937 state
= next_state(state
);
938 if (start
< end
&& state
&& state
->start
== start
&&
945 * | ---- desired range ---- |
948 * | ------------- state -------------- |
950 * We need to split the extent we found, and may flip bits on
953 * If the extent we found extends past our
954 * range, we just split and search again. It'll get split
955 * again the next time though.
957 * If the extent we found is inside our range, we set the
960 if (state
->start
< start
) {
961 if (state
->state
& exclusive_bits
) {
962 *failed_start
= start
;
967 prealloc
= alloc_extent_state_atomic(prealloc
);
969 err
= split_state(tree
, state
, prealloc
, start
);
971 extent_io_tree_panic(tree
, err
);
976 if (state
->end
<= end
) {
977 set_state_bits(tree
, state
, &bits
, changeset
);
978 cache_state(state
, cached_state
);
979 merge_state(tree
, state
);
980 if (last_end
== (u64
)-1)
982 start
= last_end
+ 1;
983 state
= next_state(state
);
984 if (start
< end
&& state
&& state
->start
== start
&&
991 * | ---- desired range ---- |
992 * | state | or | state |
994 * There's a hole, we need to insert something in it and
995 * ignore the extent we found.
997 if (state
->start
> start
) {
999 if (end
< last_start
)
1002 this_end
= last_start
- 1;
1004 prealloc
= alloc_extent_state_atomic(prealloc
);
1008 * Avoid to free 'prealloc' if it can be merged with
1011 err
= insert_state(tree
, prealloc
, start
, this_end
,
1012 NULL
, NULL
, &bits
, changeset
);
1014 extent_io_tree_panic(tree
, err
);
1016 cache_state(prealloc
, cached_state
);
1018 start
= this_end
+ 1;
1022 * | ---- desired range ---- |
1024 * We need to split the extent, and set the bit
1027 if (state
->start
<= end
&& state
->end
> end
) {
1028 if (state
->state
& exclusive_bits
) {
1029 *failed_start
= start
;
1034 prealloc
= alloc_extent_state_atomic(prealloc
);
1036 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1038 extent_io_tree_panic(tree
, err
);
1040 set_state_bits(tree
, prealloc
, &bits
, changeset
);
1041 cache_state(prealloc
, cached_state
);
1042 merge_state(tree
, prealloc
);
1050 spin_unlock(&tree
->lock
);
1051 if (gfpflags_allow_blocking(mask
))
1056 spin_unlock(&tree
->lock
);
1058 free_extent_state(prealloc
);
1064 int set_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1065 unsigned bits
, u64
* failed_start
,
1066 struct extent_state
**cached_state
, gfp_t mask
)
1068 return __set_extent_bit(tree
, start
, end
, bits
, 0, failed_start
,
1069 cached_state
, mask
, NULL
);
1074 * convert_extent_bit - convert all bits in a given range from one bit to
1076 * @tree: the io tree to search
1077 * @start: the start offset in bytes
1078 * @end: the end offset in bytes (inclusive)
1079 * @bits: the bits to set in this range
1080 * @clear_bits: the bits to clear in this range
1081 * @cached_state: state that we're going to cache
1083 * This will go through and set bits for the given range. If any states exist
1084 * already in this range they are set with the given bit and cleared of the
1085 * clear_bits. This is only meant to be used by things that are mergeable, ie
1086 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1087 * boundary bits like LOCK.
1089 * All allocations are done with GFP_NOFS.
1091 int convert_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1092 unsigned bits
, unsigned clear_bits
,
1093 struct extent_state
**cached_state
)
1095 struct extent_state
*state
;
1096 struct extent_state
*prealloc
= NULL
;
1097 struct rb_node
*node
;
1099 struct rb_node
*parent
;
1103 bool first_iteration
= true;
1105 btrfs_debug_check_extent_io_range(tree
, start
, end
);
1110 * Best effort, don't worry if extent state allocation fails
1111 * here for the first iteration. We might have a cached state
1112 * that matches exactly the target range, in which case no
1113 * extent state allocations are needed. We'll only know this
1114 * after locking the tree.
1116 prealloc
= alloc_extent_state(GFP_NOFS
);
1117 if (!prealloc
&& !first_iteration
)
1121 spin_lock(&tree
->lock
);
1122 if (cached_state
&& *cached_state
) {
1123 state
= *cached_state
;
1124 if (state
->start
<= start
&& state
->end
> start
&&
1125 extent_state_in_tree(state
)) {
1126 node
= &state
->rb_node
;
1132 * this search will find all the extents that end after
1135 node
= tree_search_for_insert(tree
, start
, &p
, &parent
);
1137 prealloc
= alloc_extent_state_atomic(prealloc
);
1142 err
= insert_state(tree
, prealloc
, start
, end
,
1143 &p
, &parent
, &bits
, NULL
);
1145 extent_io_tree_panic(tree
, err
);
1146 cache_state(prealloc
, cached_state
);
1150 state
= rb_entry(node
, struct extent_state
, rb_node
);
1152 last_start
= state
->start
;
1153 last_end
= state
->end
;
1156 * | ---- desired range ---- |
1159 * Just lock what we found and keep going
1161 if (state
->start
== start
&& state
->end
<= end
) {
1162 set_state_bits(tree
, state
, &bits
, NULL
);
1163 cache_state(state
, cached_state
);
1164 state
= clear_state_bit(tree
, state
, &clear_bits
, 0, NULL
);
1165 if (last_end
== (u64
)-1)
1167 start
= last_end
+ 1;
1168 if (start
< end
&& state
&& state
->start
== start
&&
1175 * | ---- desired range ---- |
1178 * | ------------- state -------------- |
1180 * We need to split the extent we found, and may flip bits on
1183 * If the extent we found extends past our
1184 * range, we just split and search again. It'll get split
1185 * again the next time though.
1187 * If the extent we found is inside our range, we set the
1188 * desired bit on it.
1190 if (state
->start
< start
) {
1191 prealloc
= alloc_extent_state_atomic(prealloc
);
1196 err
= split_state(tree
, state
, prealloc
, start
);
1198 extent_io_tree_panic(tree
, err
);
1202 if (state
->end
<= end
) {
1203 set_state_bits(tree
, state
, &bits
, NULL
);
1204 cache_state(state
, cached_state
);
1205 state
= clear_state_bit(tree
, state
, &clear_bits
, 0,
1207 if (last_end
== (u64
)-1)
1209 start
= last_end
+ 1;
1210 if (start
< end
&& state
&& state
->start
== start
&&
1217 * | ---- desired range ---- |
1218 * | state | or | state |
1220 * There's a hole, we need to insert something in it and
1221 * ignore the extent we found.
1223 if (state
->start
> start
) {
1225 if (end
< last_start
)
1228 this_end
= last_start
- 1;
1230 prealloc
= alloc_extent_state_atomic(prealloc
);
1237 * Avoid to free 'prealloc' if it can be merged with
1240 err
= insert_state(tree
, prealloc
, start
, this_end
,
1241 NULL
, NULL
, &bits
, NULL
);
1243 extent_io_tree_panic(tree
, err
);
1244 cache_state(prealloc
, cached_state
);
1246 start
= this_end
+ 1;
1250 * | ---- desired range ---- |
1252 * We need to split the extent, and set the bit
1255 if (state
->start
<= end
&& state
->end
> end
) {
1256 prealloc
= alloc_extent_state_atomic(prealloc
);
1262 err
= split_state(tree
, state
, prealloc
, end
+ 1);
1264 extent_io_tree_panic(tree
, err
);
1266 set_state_bits(tree
, prealloc
, &bits
, NULL
);
1267 cache_state(prealloc
, cached_state
);
1268 clear_state_bit(tree
, prealloc
, &clear_bits
, 0, NULL
);
1276 spin_unlock(&tree
->lock
);
1278 first_iteration
= false;
1282 spin_unlock(&tree
->lock
);
1284 free_extent_state(prealloc
);
1289 /* wrappers around set/clear extent bit */
1290 int set_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1291 unsigned bits
, struct extent_changeset
*changeset
)
1294 * We don't support EXTENT_LOCKED yet, as current changeset will
1295 * record any bits changed, so for EXTENT_LOCKED case, it will
1296 * either fail with -EEXIST or changeset will record the whole
1299 BUG_ON(bits
& EXTENT_LOCKED
);
1301 return __set_extent_bit(tree
, start
, end
, bits
, 0, NULL
, NULL
, GFP_NOFS
,
1305 int clear_extent_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1306 unsigned bits
, int wake
, int delete,
1307 struct extent_state
**cached
, gfp_t mask
)
1309 return __clear_extent_bit(tree
, start
, end
, bits
, wake
, delete,
1310 cached
, mask
, NULL
);
1313 int clear_record_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1314 unsigned bits
, struct extent_changeset
*changeset
)
1317 * Don't support EXTENT_LOCKED case, same reason as
1318 * set_record_extent_bits().
1320 BUG_ON(bits
& EXTENT_LOCKED
);
1322 return __clear_extent_bit(tree
, start
, end
, bits
, 0, 0, NULL
, GFP_NOFS
,
1327 * either insert or lock state struct between start and end use mask to tell
1328 * us if waiting is desired.
1330 int lock_extent_bits(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1331 struct extent_state
**cached_state
)
1337 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
,
1338 EXTENT_LOCKED
, &failed_start
,
1339 cached_state
, GFP_NOFS
, NULL
);
1340 if (err
== -EEXIST
) {
1341 wait_extent_bit(tree
, failed_start
, end
, EXTENT_LOCKED
);
1342 start
= failed_start
;
1345 WARN_ON(start
> end
);
1350 int try_lock_extent(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1355 err
= __set_extent_bit(tree
, start
, end
, EXTENT_LOCKED
, EXTENT_LOCKED
,
1356 &failed_start
, NULL
, GFP_NOFS
, NULL
);
1357 if (err
== -EEXIST
) {
1358 if (failed_start
> start
)
1359 clear_extent_bit(tree
, start
, failed_start
- 1,
1360 EXTENT_LOCKED
, 1, 0, NULL
, GFP_NOFS
);
1366 void extent_range_clear_dirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1368 unsigned long index
= start
>> PAGE_SHIFT
;
1369 unsigned long end_index
= end
>> PAGE_SHIFT
;
1372 while (index
<= end_index
) {
1373 page
= find_get_page(inode
->i_mapping
, index
);
1374 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1375 clear_page_dirty_for_io(page
);
1381 void extent_range_redirty_for_io(struct inode
*inode
, u64 start
, u64 end
)
1383 unsigned long index
= start
>> PAGE_SHIFT
;
1384 unsigned long end_index
= end
>> PAGE_SHIFT
;
1387 while (index
<= end_index
) {
1388 page
= find_get_page(inode
->i_mapping
, index
);
1389 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1390 __set_page_dirty_nobuffers(page
);
1391 account_page_redirty(page
);
1398 * helper function to set both pages and extents in the tree writeback
1400 static void set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
1402 unsigned long index
= start
>> PAGE_SHIFT
;
1403 unsigned long end_index
= end
>> PAGE_SHIFT
;
1406 while (index
<= end_index
) {
1407 page
= find_get_page(tree
->mapping
, index
);
1408 BUG_ON(!page
); /* Pages should be in the extent_io_tree */
1409 set_page_writeback(page
);
1415 /* find the first state struct with 'bits' set after 'start', and
1416 * return it. tree->lock must be held. NULL will returned if
1417 * nothing was found after 'start'
1419 static struct extent_state
*
1420 find_first_extent_bit_state(struct extent_io_tree
*tree
,
1421 u64 start
, unsigned bits
)
1423 struct rb_node
*node
;
1424 struct extent_state
*state
;
1427 * this search will find all the extents that end after
1430 node
= tree_search(tree
, start
);
1435 state
= rb_entry(node
, struct extent_state
, rb_node
);
1436 if (state
->end
>= start
&& (state
->state
& bits
))
1439 node
= rb_next(node
);
1448 * find the first offset in the io tree with 'bits' set. zero is
1449 * returned if we find something, and *start_ret and *end_ret are
1450 * set to reflect the state struct that was found.
1452 * If nothing was found, 1 is returned. If found something, return 0.
1454 int find_first_extent_bit(struct extent_io_tree
*tree
, u64 start
,
1455 u64
*start_ret
, u64
*end_ret
, unsigned bits
,
1456 struct extent_state
**cached_state
)
1458 struct extent_state
*state
;
1462 spin_lock(&tree
->lock
);
1463 if (cached_state
&& *cached_state
) {
1464 state
= *cached_state
;
1465 if (state
->end
== start
- 1 && extent_state_in_tree(state
)) {
1466 n
= rb_next(&state
->rb_node
);
1468 state
= rb_entry(n
, struct extent_state
,
1470 if (state
->state
& bits
)
1474 free_extent_state(*cached_state
);
1475 *cached_state
= NULL
;
1478 free_extent_state(*cached_state
);
1479 *cached_state
= NULL
;
1482 state
= find_first_extent_bit_state(tree
, start
, bits
);
1485 cache_state_if_flags(state
, cached_state
, 0);
1486 *start_ret
= state
->start
;
1487 *end_ret
= state
->end
;
1491 spin_unlock(&tree
->lock
);
1496 * find a contiguous range of bytes in the file marked as delalloc, not
1497 * more than 'max_bytes'. start and end are used to return the range,
1499 * 1 is returned if we find something, 0 if nothing was in the tree
1501 static noinline u64
find_delalloc_range(struct extent_io_tree
*tree
,
1502 u64
*start
, u64
*end
, u64 max_bytes
,
1503 struct extent_state
**cached_state
)
1505 struct rb_node
*node
;
1506 struct extent_state
*state
;
1507 u64 cur_start
= *start
;
1509 u64 total_bytes
= 0;
1511 spin_lock(&tree
->lock
);
1514 * this search will find all the extents that end after
1517 node
= tree_search(tree
, cur_start
);
1525 state
= rb_entry(node
, struct extent_state
, rb_node
);
1526 if (found
&& (state
->start
!= cur_start
||
1527 (state
->state
& EXTENT_BOUNDARY
))) {
1530 if (!(state
->state
& EXTENT_DELALLOC
)) {
1536 *start
= state
->start
;
1537 *cached_state
= state
;
1538 atomic_inc(&state
->refs
);
1542 cur_start
= state
->end
+ 1;
1543 node
= rb_next(node
);
1544 total_bytes
+= state
->end
- state
->start
+ 1;
1545 if (total_bytes
>= max_bytes
)
1551 spin_unlock(&tree
->lock
);
1555 static noinline
void __unlock_for_delalloc(struct inode
*inode
,
1556 struct page
*locked_page
,
1560 struct page
*pages
[16];
1561 unsigned long index
= start
>> PAGE_SHIFT
;
1562 unsigned long end_index
= end
>> PAGE_SHIFT
;
1563 unsigned long nr_pages
= end_index
- index
+ 1;
1566 if (index
== locked_page
->index
&& end_index
== index
)
1569 while (nr_pages
> 0) {
1570 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1571 min_t(unsigned long, nr_pages
,
1572 ARRAY_SIZE(pages
)), pages
);
1573 for (i
= 0; i
< ret
; i
++) {
1574 if (pages
[i
] != locked_page
)
1575 unlock_page(pages
[i
]);
1584 static noinline
int lock_delalloc_pages(struct inode
*inode
,
1585 struct page
*locked_page
,
1589 unsigned long index
= delalloc_start
>> PAGE_SHIFT
;
1590 unsigned long start_index
= index
;
1591 unsigned long end_index
= delalloc_end
>> PAGE_SHIFT
;
1592 unsigned long pages_locked
= 0;
1593 struct page
*pages
[16];
1594 unsigned long nrpages
;
1598 /* the caller is responsible for locking the start index */
1599 if (index
== locked_page
->index
&& index
== end_index
)
1602 /* skip the page at the start index */
1603 nrpages
= end_index
- index
+ 1;
1604 while (nrpages
> 0) {
1605 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1606 min_t(unsigned long,
1607 nrpages
, ARRAY_SIZE(pages
)), pages
);
1612 /* now we have an array of pages, lock them all */
1613 for (i
= 0; i
< ret
; i
++) {
1615 * the caller is taking responsibility for
1618 if (pages
[i
] != locked_page
) {
1619 lock_page(pages
[i
]);
1620 if (!PageDirty(pages
[i
]) ||
1621 pages
[i
]->mapping
!= inode
->i_mapping
) {
1623 unlock_page(pages
[i
]);
1637 if (ret
&& pages_locked
) {
1638 __unlock_for_delalloc(inode
, locked_page
,
1640 ((u64
)(start_index
+ pages_locked
- 1)) <<
1647 * find a contiguous range of bytes in the file marked as delalloc, not
1648 * more than 'max_bytes'. start and end are used to return the range,
1650 * 1 is returned if we find something, 0 if nothing was in the tree
1652 STATIC u64
find_lock_delalloc_range(struct inode
*inode
,
1653 struct extent_io_tree
*tree
,
1654 struct page
*locked_page
, u64
*start
,
1655 u64
*end
, u64 max_bytes
)
1660 struct extent_state
*cached_state
= NULL
;
1665 /* step one, find a bunch of delalloc bytes starting at start */
1666 delalloc_start
= *start
;
1668 found
= find_delalloc_range(tree
, &delalloc_start
, &delalloc_end
,
1669 max_bytes
, &cached_state
);
1670 if (!found
|| delalloc_end
<= *start
) {
1671 *start
= delalloc_start
;
1672 *end
= delalloc_end
;
1673 free_extent_state(cached_state
);
1678 * start comes from the offset of locked_page. We have to lock
1679 * pages in order, so we can't process delalloc bytes before
1682 if (delalloc_start
< *start
)
1683 delalloc_start
= *start
;
1686 * make sure to limit the number of pages we try to lock down
1688 if (delalloc_end
+ 1 - delalloc_start
> max_bytes
)
1689 delalloc_end
= delalloc_start
+ max_bytes
- 1;
1691 /* step two, lock all the pages after the page that has start */
1692 ret
= lock_delalloc_pages(inode
, locked_page
,
1693 delalloc_start
, delalloc_end
);
1694 if (ret
== -EAGAIN
) {
1695 /* some of the pages are gone, lets avoid looping by
1696 * shortening the size of the delalloc range we're searching
1698 free_extent_state(cached_state
);
1699 cached_state
= NULL
;
1701 max_bytes
= PAGE_SIZE
;
1709 BUG_ON(ret
); /* Only valid values are 0 and -EAGAIN */
1711 /* step three, lock the state bits for the whole range */
1712 lock_extent_bits(tree
, delalloc_start
, delalloc_end
, &cached_state
);
1714 /* then test to make sure it is all still delalloc */
1715 ret
= test_range_bit(tree
, delalloc_start
, delalloc_end
,
1716 EXTENT_DELALLOC
, 1, cached_state
);
1718 unlock_extent_cached(tree
, delalloc_start
, delalloc_end
,
1719 &cached_state
, GFP_NOFS
);
1720 __unlock_for_delalloc(inode
, locked_page
,
1721 delalloc_start
, delalloc_end
);
1725 free_extent_state(cached_state
);
1726 *start
= delalloc_start
;
1727 *end
= delalloc_end
;
1732 void extent_clear_unlock_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1733 struct page
*locked_page
,
1734 unsigned clear_bits
,
1735 unsigned long page_ops
)
1737 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
1739 struct page
*pages
[16];
1740 unsigned long index
= start
>> PAGE_SHIFT
;
1741 unsigned long end_index
= end
>> PAGE_SHIFT
;
1742 unsigned long nr_pages
= end_index
- index
+ 1;
1745 clear_extent_bit(tree
, start
, end
, clear_bits
, 1, 0, NULL
, GFP_NOFS
);
1749 if ((page_ops
& PAGE_SET_ERROR
) && nr_pages
> 0)
1750 mapping_set_error(inode
->i_mapping
, -EIO
);
1752 while (nr_pages
> 0) {
1753 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
1754 min_t(unsigned long,
1755 nr_pages
, ARRAY_SIZE(pages
)), pages
);
1756 for (i
= 0; i
< ret
; i
++) {
1758 if (page_ops
& PAGE_SET_PRIVATE2
)
1759 SetPagePrivate2(pages
[i
]);
1761 if (pages
[i
] == locked_page
) {
1765 if (page_ops
& PAGE_CLEAR_DIRTY
)
1766 clear_page_dirty_for_io(pages
[i
]);
1767 if (page_ops
& PAGE_SET_WRITEBACK
)
1768 set_page_writeback(pages
[i
]);
1769 if (page_ops
& PAGE_SET_ERROR
)
1770 SetPageError(pages
[i
]);
1771 if (page_ops
& PAGE_END_WRITEBACK
)
1772 end_page_writeback(pages
[i
]);
1773 if (page_ops
& PAGE_UNLOCK
)
1774 unlock_page(pages
[i
]);
1784 * count the number of bytes in the tree that have a given bit(s)
1785 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1786 * cached. The total number found is returned.
1788 u64
count_range_bits(struct extent_io_tree
*tree
,
1789 u64
*start
, u64 search_end
, u64 max_bytes
,
1790 unsigned bits
, int contig
)
1792 struct rb_node
*node
;
1793 struct extent_state
*state
;
1794 u64 cur_start
= *start
;
1795 u64 total_bytes
= 0;
1799 if (WARN_ON(search_end
<= cur_start
))
1802 spin_lock(&tree
->lock
);
1803 if (cur_start
== 0 && bits
== EXTENT_DIRTY
) {
1804 total_bytes
= tree
->dirty_bytes
;
1808 * this search will find all the extents that end after
1811 node
= tree_search(tree
, cur_start
);
1816 state
= rb_entry(node
, struct extent_state
, rb_node
);
1817 if (state
->start
> search_end
)
1819 if (contig
&& found
&& state
->start
> last
+ 1)
1821 if (state
->end
>= cur_start
&& (state
->state
& bits
) == bits
) {
1822 total_bytes
+= min(search_end
, state
->end
) + 1 -
1823 max(cur_start
, state
->start
);
1824 if (total_bytes
>= max_bytes
)
1827 *start
= max(cur_start
, state
->start
);
1831 } else if (contig
&& found
) {
1834 node
= rb_next(node
);
1839 spin_unlock(&tree
->lock
);
1844 * set the private field for a given byte offset in the tree. If there isn't
1845 * an extent_state there already, this does nothing.
1847 static noinline
int set_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1848 struct io_failure_record
*failrec
)
1850 struct rb_node
*node
;
1851 struct extent_state
*state
;
1854 spin_lock(&tree
->lock
);
1856 * this search will find all the extents that end after
1859 node
= tree_search(tree
, start
);
1864 state
= rb_entry(node
, struct extent_state
, rb_node
);
1865 if (state
->start
!= start
) {
1869 state
->failrec
= failrec
;
1871 spin_unlock(&tree
->lock
);
1875 static noinline
int get_state_failrec(struct extent_io_tree
*tree
, u64 start
,
1876 struct io_failure_record
**failrec
)
1878 struct rb_node
*node
;
1879 struct extent_state
*state
;
1882 spin_lock(&tree
->lock
);
1884 * this search will find all the extents that end after
1887 node
= tree_search(tree
, start
);
1892 state
= rb_entry(node
, struct extent_state
, rb_node
);
1893 if (state
->start
!= start
) {
1897 *failrec
= state
->failrec
;
1899 spin_unlock(&tree
->lock
);
1904 * searches a range in the state tree for a given mask.
1905 * If 'filled' == 1, this returns 1 only if every extent in the tree
1906 * has the bits set. Otherwise, 1 is returned if any bit in the
1907 * range is found set.
1909 int test_range_bit(struct extent_io_tree
*tree
, u64 start
, u64 end
,
1910 unsigned bits
, int filled
, struct extent_state
*cached
)
1912 struct extent_state
*state
= NULL
;
1913 struct rb_node
*node
;
1916 spin_lock(&tree
->lock
);
1917 if (cached
&& extent_state_in_tree(cached
) && cached
->start
<= start
&&
1918 cached
->end
> start
)
1919 node
= &cached
->rb_node
;
1921 node
= tree_search(tree
, start
);
1922 while (node
&& start
<= end
) {
1923 state
= rb_entry(node
, struct extent_state
, rb_node
);
1925 if (filled
&& state
->start
> start
) {
1930 if (state
->start
> end
)
1933 if (state
->state
& bits
) {
1937 } else if (filled
) {
1942 if (state
->end
== (u64
)-1)
1945 start
= state
->end
+ 1;
1948 node
= rb_next(node
);
1955 spin_unlock(&tree
->lock
);
1960 * helper function to set a given page up to date if all the
1961 * extents in the tree for that page are up to date
1963 static void check_page_uptodate(struct extent_io_tree
*tree
, struct page
*page
)
1965 u64 start
= page_offset(page
);
1966 u64 end
= start
+ PAGE_SIZE
- 1;
1967 if (test_range_bit(tree
, start
, end
, EXTENT_UPTODATE
, 1, NULL
))
1968 SetPageUptodate(page
);
1971 int free_io_failure(struct inode
*inode
, struct io_failure_record
*rec
)
1975 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1977 set_state_failrec(failure_tree
, rec
->start
, NULL
);
1978 ret
= clear_extent_bits(failure_tree
, rec
->start
,
1979 rec
->start
+ rec
->len
- 1,
1980 EXTENT_LOCKED
| EXTENT_DIRTY
);
1984 ret
= clear_extent_bits(&BTRFS_I(inode
)->io_tree
, rec
->start
,
1985 rec
->start
+ rec
->len
- 1,
1995 * this bypasses the standard btrfs submit functions deliberately, as
1996 * the standard behavior is to write all copies in a raid setup. here we only
1997 * want to write the one bad copy. so we do the mapping for ourselves and issue
1998 * submit_bio directly.
1999 * to avoid any synchronization issues, wait for the data after writing, which
2000 * actually prevents the read that triggered the error from finishing.
2001 * currently, there can be no more than two copies of every data bit. thus,
2002 * exactly one rewrite is required.
2004 int repair_io_failure(struct inode
*inode
, u64 start
, u64 length
, u64 logical
,
2005 struct page
*page
, unsigned int pg_offset
, int mirror_num
)
2007 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2009 struct btrfs_device
*dev
;
2012 struct btrfs_bio
*bbio
= NULL
;
2013 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
2016 ASSERT(!(fs_info
->sb
->s_flags
& MS_RDONLY
));
2017 BUG_ON(!mirror_num
);
2019 /* we can't repair anything in raid56 yet */
2020 if (btrfs_is_parity_mirror(map_tree
, logical
, length
, mirror_num
))
2023 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2026 bio
->bi_iter
.bi_size
= 0;
2027 map_length
= length
;
2030 * Avoid races with device replace and make sure our bbio has devices
2031 * associated to its stripes that don't go away while we are doing the
2032 * read repair operation.
2034 btrfs_bio_counter_inc_blocked(fs_info
);
2035 ret
= btrfs_map_block(fs_info
, WRITE
, logical
,
2036 &map_length
, &bbio
, mirror_num
);
2038 btrfs_bio_counter_dec(fs_info
);
2042 BUG_ON(mirror_num
!= bbio
->mirror_num
);
2043 sector
= bbio
->stripes
[mirror_num
-1].physical
>> 9;
2044 bio
->bi_iter
.bi_sector
= sector
;
2045 dev
= bbio
->stripes
[mirror_num
-1].dev
;
2046 btrfs_put_bbio(bbio
);
2047 if (!dev
|| !dev
->bdev
|| !dev
->writeable
) {
2048 btrfs_bio_counter_dec(fs_info
);
2052 bio
->bi_bdev
= dev
->bdev
;
2053 bio_set_op_attrs(bio
, REQ_OP_WRITE
, WRITE_SYNC
);
2054 bio_add_page(bio
, page
, length
, pg_offset
);
2056 if (btrfsic_submit_bio_wait(bio
)) {
2057 /* try to remap that extent elsewhere? */
2058 btrfs_bio_counter_dec(fs_info
);
2060 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
2064 btrfs_info_rl_in_rcu(fs_info
,
2065 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2066 btrfs_ino(inode
), start
,
2067 rcu_str_deref(dev
->name
), sector
);
2068 btrfs_bio_counter_dec(fs_info
);
2073 int repair_eb_io_failure(struct btrfs_root
*root
, struct extent_buffer
*eb
,
2076 u64 start
= eb
->start
;
2077 unsigned long i
, num_pages
= num_extent_pages(eb
->start
, eb
->len
);
2080 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2083 for (i
= 0; i
< num_pages
; i
++) {
2084 struct page
*p
= eb
->pages
[i
];
2086 ret
= repair_io_failure(root
->fs_info
->btree_inode
, start
,
2087 PAGE_SIZE
, start
, p
,
2088 start
- page_offset(p
), mirror_num
);
2098 * each time an IO finishes, we do a fast check in the IO failure tree
2099 * to see if we need to process or clean up an io_failure_record
2101 int clean_io_failure(struct inode
*inode
, u64 start
, struct page
*page
,
2102 unsigned int pg_offset
)
2105 struct io_failure_record
*failrec
;
2106 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2107 struct extent_state
*state
;
2112 ret
= count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
2113 (u64
)-1, 1, EXTENT_DIRTY
, 0);
2117 ret
= get_state_failrec(&BTRFS_I(inode
)->io_failure_tree
, start
,
2122 BUG_ON(!failrec
->this_mirror
);
2124 if (failrec
->in_validation
) {
2125 /* there was no real error, just free the record */
2126 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2130 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
2133 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
2134 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
2137 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
2139 if (state
&& state
->start
<= failrec
->start
&&
2140 state
->end
>= failrec
->start
+ failrec
->len
- 1) {
2141 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
,
2143 if (num_copies
> 1) {
2144 repair_io_failure(inode
, start
, failrec
->len
,
2145 failrec
->logical
, page
,
2146 pg_offset
, failrec
->failed_mirror
);
2151 free_io_failure(inode
, failrec
);
2157 * Can be called when
2158 * - hold extent lock
2159 * - under ordered extent
2160 * - the inode is freeing
2162 void btrfs_free_io_failure_record(struct inode
*inode
, u64 start
, u64 end
)
2164 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2165 struct io_failure_record
*failrec
;
2166 struct extent_state
*state
, *next
;
2168 if (RB_EMPTY_ROOT(&failure_tree
->state
))
2171 spin_lock(&failure_tree
->lock
);
2172 state
= find_first_extent_bit_state(failure_tree
, start
, EXTENT_DIRTY
);
2174 if (state
->start
> end
)
2177 ASSERT(state
->end
<= end
);
2179 next
= next_state(state
);
2181 failrec
= state
->failrec
;
2182 free_extent_state(state
);
2187 spin_unlock(&failure_tree
->lock
);
2190 int btrfs_get_io_failure_record(struct inode
*inode
, u64 start
, u64 end
,
2191 struct io_failure_record
**failrec_ret
)
2193 struct io_failure_record
*failrec
;
2194 struct extent_map
*em
;
2195 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
2196 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
2197 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2201 ret
= get_state_failrec(failure_tree
, start
, &failrec
);
2203 failrec
= kzalloc(sizeof(*failrec
), GFP_NOFS
);
2207 failrec
->start
= start
;
2208 failrec
->len
= end
- start
+ 1;
2209 failrec
->this_mirror
= 0;
2210 failrec
->bio_flags
= 0;
2211 failrec
->in_validation
= 0;
2213 read_lock(&em_tree
->lock
);
2214 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
2216 read_unlock(&em_tree
->lock
);
2221 if (em
->start
> start
|| em
->start
+ em
->len
<= start
) {
2222 free_extent_map(em
);
2225 read_unlock(&em_tree
->lock
);
2231 logical
= start
- em
->start
;
2232 logical
= em
->block_start
+ logical
;
2233 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2234 logical
= em
->block_start
;
2235 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
2236 extent_set_compress_type(&failrec
->bio_flags
,
2240 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2241 logical
, start
, failrec
->len
);
2243 failrec
->logical
= logical
;
2244 free_extent_map(em
);
2246 /* set the bits in the private failure tree */
2247 ret
= set_extent_bits(failure_tree
, start
, end
,
2248 EXTENT_LOCKED
| EXTENT_DIRTY
);
2250 ret
= set_state_failrec(failure_tree
, start
, failrec
);
2251 /* set the bits in the inode's tree */
2253 ret
= set_extent_bits(tree
, start
, end
, EXTENT_DAMAGED
);
2259 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2260 failrec
->logical
, failrec
->start
, failrec
->len
,
2261 failrec
->in_validation
);
2263 * when data can be on disk more than twice, add to failrec here
2264 * (e.g. with a list for failed_mirror) to make
2265 * clean_io_failure() clean all those errors at once.
2269 *failrec_ret
= failrec
;
2274 int btrfs_check_repairable(struct inode
*inode
, struct bio
*failed_bio
,
2275 struct io_failure_record
*failrec
, int failed_mirror
)
2279 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
2280 failrec
->logical
, failrec
->len
);
2281 if (num_copies
== 1) {
2283 * we only have a single copy of the data, so don't bother with
2284 * all the retry and error correction code that follows. no
2285 * matter what the error is, it is very likely to persist.
2287 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2288 num_copies
, failrec
->this_mirror
, failed_mirror
);
2293 * there are two premises:
2294 * a) deliver good data to the caller
2295 * b) correct the bad sectors on disk
2297 if (failed_bio
->bi_vcnt
> 1) {
2299 * to fulfill b), we need to know the exact failing sectors, as
2300 * we don't want to rewrite any more than the failed ones. thus,
2301 * we need separate read requests for the failed bio
2303 * if the following BUG_ON triggers, our validation request got
2304 * merged. we need separate requests for our algorithm to work.
2306 BUG_ON(failrec
->in_validation
);
2307 failrec
->in_validation
= 1;
2308 failrec
->this_mirror
= failed_mirror
;
2311 * we're ready to fulfill a) and b) alongside. get a good copy
2312 * of the failed sector and if we succeed, we have setup
2313 * everything for repair_io_failure to do the rest for us.
2315 if (failrec
->in_validation
) {
2316 BUG_ON(failrec
->this_mirror
!= failed_mirror
);
2317 failrec
->in_validation
= 0;
2318 failrec
->this_mirror
= 0;
2320 failrec
->failed_mirror
= failed_mirror
;
2321 failrec
->this_mirror
++;
2322 if (failrec
->this_mirror
== failed_mirror
)
2323 failrec
->this_mirror
++;
2326 if (failrec
->this_mirror
> num_copies
) {
2327 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
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
)
2342 struct btrfs_io_bio
*btrfs_failed_bio
;
2343 struct btrfs_io_bio
*btrfs_bio
;
2345 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
2349 bio
->bi_end_io
= endio_func
;
2350 bio
->bi_iter
.bi_sector
= failrec
->logical
>> 9;
2351 bio
->bi_bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
2352 bio
->bi_iter
.bi_size
= 0;
2353 bio
->bi_private
= data
;
2355 btrfs_failed_bio
= btrfs_io_bio(failed_bio
);
2356 if (btrfs_failed_bio
->csum
) {
2357 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
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(inode
, failrec
);
2403 if (failed_bio
->bi_vcnt
> 1)
2404 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
2406 read_mode
= READ_SYNC
;
2408 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2409 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
2410 start
- page_offset(page
),
2411 (int)phy_offset
, failed_bio
->bi_end_io
,
2414 free_io_failure(inode
, failrec
);
2417 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
2419 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2420 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
2422 ret
= tree
->ops
->submit_bio_hook(inode
, bio
, failrec
->this_mirror
,
2423 failrec
->bio_flags
, 0);
2425 free_io_failure(inode
, failrec
);
2432 /* lots and lots of room for performance fixes in the end_bio funcs */
2434 void end_extent_writepage(struct page
*page
, int err
, u64 start
, u64 end
)
2436 int uptodate
= (err
== 0);
2437 struct extent_io_tree
*tree
;
2440 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
2442 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
) {
2443 ret
= tree
->ops
->writepage_end_io_hook(page
, start
,
2444 end
, NULL
, uptodate
);
2450 ClearPageUptodate(page
);
2452 ret
= ret
< 0 ? ret
: -EIO
;
2453 mapping_set_error(page
->mapping
, ret
);
2458 * after a writepage IO is done, we need to:
2459 * clear the uptodate bits on error
2460 * clear the writeback bits in the extent tree for this IO
2461 * end_page_writeback if the page has no more pending IO
2463 * Scheduling is not allowed, so the extent state tree is expected
2464 * to have one and only one object corresponding to this IO.
2466 static void end_bio_extent_writepage(struct bio
*bio
)
2468 struct bio_vec
*bvec
;
2473 bio_for_each_segment_all(bvec
, bio
, i
) {
2474 struct page
*page
= bvec
->bv_page
;
2476 /* We always issue full-page reads, but if some block
2477 * in a page fails to read, blk_update_request() will
2478 * advance bv_offset and adjust bv_len to compensate.
2479 * Print a warning for nonzero offsets, and an error
2480 * if they don't add up to a full page. */
2481 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2482 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2483 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2484 "partial page write in btrfs with offset %u and length %u",
2485 bvec
->bv_offset
, bvec
->bv_len
);
2487 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2488 "incomplete page write in btrfs with offset %u and "
2490 bvec
->bv_offset
, bvec
->bv_len
);
2493 start
= page_offset(page
);
2494 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2496 end_extent_writepage(page
, bio
->bi_error
, start
, end
);
2497 end_page_writeback(page
);
2504 endio_readpage_release_extent(struct extent_io_tree
*tree
, u64 start
, u64 len
,
2507 struct extent_state
*cached
= NULL
;
2508 u64 end
= start
+ len
- 1;
2510 if (uptodate
&& tree
->track_uptodate
)
2511 set_extent_uptodate(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2512 unlock_extent_cached(tree
, start
, end
, &cached
, GFP_ATOMIC
);
2516 * after a readpage IO is done, we need to:
2517 * clear the uptodate bits on error
2518 * set the uptodate bits if things worked
2519 * set the page up to date if all extents in the tree are uptodate
2520 * clear the lock bit in the extent tree
2521 * unlock the page if there are no other extents locked for it
2523 * Scheduling is not allowed, so the extent state tree is expected
2524 * to have one and only one object corresponding to this IO.
2526 static void end_bio_extent_readpage(struct bio
*bio
)
2528 struct bio_vec
*bvec
;
2529 int uptodate
= !bio
->bi_error
;
2530 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
2531 struct extent_io_tree
*tree
;
2536 u64 extent_start
= 0;
2542 bio_for_each_segment_all(bvec
, bio
, i
) {
2543 struct page
*page
= bvec
->bv_page
;
2544 struct inode
*inode
= page
->mapping
->host
;
2546 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2547 "mirror=%u\n", (u64
)bio
->bi_iter
.bi_sector
,
2548 bio
->bi_error
, io_bio
->mirror_num
);
2549 tree
= &BTRFS_I(inode
)->io_tree
;
2551 /* We always issue full-page reads, but if some block
2552 * in a page fails to read, blk_update_request() will
2553 * advance bv_offset and adjust bv_len to compensate.
2554 * Print a warning for nonzero offsets, and an error
2555 * if they don't add up to a full page. */
2556 if (bvec
->bv_offset
|| bvec
->bv_len
!= PAGE_SIZE
) {
2557 if (bvec
->bv_offset
+ bvec
->bv_len
!= PAGE_SIZE
)
2558 btrfs_err(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2559 "partial page read in btrfs with offset %u and length %u",
2560 bvec
->bv_offset
, bvec
->bv_len
);
2562 btrfs_info(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
2563 "incomplete page read in btrfs with offset %u and "
2565 bvec
->bv_offset
, bvec
->bv_len
);
2568 start
= page_offset(page
);
2569 end
= start
+ bvec
->bv_offset
+ bvec
->bv_len
- 1;
2572 mirror
= io_bio
->mirror_num
;
2573 if (likely(uptodate
&& tree
->ops
&&
2574 tree
->ops
->readpage_end_io_hook
)) {
2575 ret
= tree
->ops
->readpage_end_io_hook(io_bio
, offset
,
2581 clean_io_failure(inode
, start
, page
, 0);
2584 if (likely(uptodate
))
2587 if (tree
->ops
&& tree
->ops
->readpage_io_failed_hook
) {
2588 ret
= tree
->ops
->readpage_io_failed_hook(page
, mirror
);
2589 if (!ret
&& !bio
->bi_error
)
2593 * The generic bio_readpage_error handles errors the
2594 * following way: If possible, new read requests are
2595 * created and submitted and will end up in
2596 * end_bio_extent_readpage as well (if we're lucky, not
2597 * in the !uptodate case). In that case it returns 0 and
2598 * we just go on with the next page in our bio. If it
2599 * can't handle the error it will return -EIO and we
2600 * remain responsible for that page.
2602 ret
= bio_readpage_error(bio
, offset
, page
, start
, end
,
2605 uptodate
= !bio
->bi_error
;
2611 if (likely(uptodate
)) {
2612 loff_t i_size
= i_size_read(inode
);
2613 pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2616 /* Zero out the end if this page straddles i_size */
2617 off
= i_size
& (PAGE_SIZE
-1);
2618 if (page
->index
== end_index
&& off
)
2619 zero_user_segment(page
, off
, PAGE_SIZE
);
2620 SetPageUptodate(page
);
2622 ClearPageUptodate(page
);
2628 if (unlikely(!uptodate
)) {
2630 endio_readpage_release_extent(tree
,
2636 endio_readpage_release_extent(tree
, start
,
2637 end
- start
+ 1, 0);
2638 } else if (!extent_len
) {
2639 extent_start
= start
;
2640 extent_len
= end
+ 1 - start
;
2641 } else if (extent_start
+ extent_len
== start
) {
2642 extent_len
+= end
+ 1 - start
;
2644 endio_readpage_release_extent(tree
, extent_start
,
2645 extent_len
, uptodate
);
2646 extent_start
= start
;
2647 extent_len
= end
+ 1 - start
;
2652 endio_readpage_release_extent(tree
, extent_start
, extent_len
,
2655 io_bio
->end_io(io_bio
, bio
->bi_error
);
2660 * this allocates from the btrfs_bioset. We're returning a bio right now
2661 * but you can call btrfs_io_bio for the appropriate container_of magic
2664 btrfs_bio_alloc(struct block_device
*bdev
, u64 first_sector
, int nr_vecs
,
2667 struct btrfs_io_bio
*btrfs_bio
;
2670 bio
= bio_alloc_bioset(gfp_flags
, nr_vecs
, btrfs_bioset
);
2672 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
2673 while (!bio
&& (nr_vecs
/= 2)) {
2674 bio
= bio_alloc_bioset(gfp_flags
,
2675 nr_vecs
, btrfs_bioset
);
2680 bio
->bi_bdev
= bdev
;
2681 bio
->bi_iter
.bi_sector
= first_sector
;
2682 btrfs_bio
= btrfs_io_bio(bio
);
2683 btrfs_bio
->csum
= NULL
;
2684 btrfs_bio
->csum_allocated
= NULL
;
2685 btrfs_bio
->end_io
= NULL
;
2690 struct bio
*btrfs_bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
2692 struct btrfs_io_bio
*btrfs_bio
;
2695 new = bio_clone_bioset(bio
, gfp_mask
, btrfs_bioset
);
2697 btrfs_bio
= btrfs_io_bio(new);
2698 btrfs_bio
->csum
= NULL
;
2699 btrfs_bio
->csum_allocated
= NULL
;
2700 btrfs_bio
->end_io
= NULL
;
2705 /* this also allocates from the btrfs_bioset */
2706 struct bio
*btrfs_io_bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
2708 struct btrfs_io_bio
*btrfs_bio
;
2711 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, btrfs_bioset
);
2713 btrfs_bio
= btrfs_io_bio(bio
);
2714 btrfs_bio
->csum
= NULL
;
2715 btrfs_bio
->csum_allocated
= NULL
;
2716 btrfs_bio
->end_io
= NULL
;
2722 static int __must_check
submit_one_bio(struct bio
*bio
, int mirror_num
,
2723 unsigned long bio_flags
)
2726 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
2727 struct page
*page
= bvec
->bv_page
;
2728 struct extent_io_tree
*tree
= bio
->bi_private
;
2731 start
= page_offset(page
) + bvec
->bv_offset
;
2733 bio
->bi_private
= NULL
;
2736 if (tree
->ops
&& tree
->ops
->submit_bio_hook
)
2737 ret
= tree
->ops
->submit_bio_hook(page
->mapping
->host
, bio
,
2738 mirror_num
, bio_flags
, start
);
2740 btrfsic_submit_bio(bio
);
2746 static int merge_bio(struct extent_io_tree
*tree
, struct page
*page
,
2747 unsigned long offset
, size_t size
, struct bio
*bio
,
2748 unsigned long bio_flags
)
2751 if (tree
->ops
&& tree
->ops
->merge_bio_hook
)
2752 ret
= tree
->ops
->merge_bio_hook(page
, offset
, size
, bio
,
2758 static int submit_extent_page(int op
, int op_flags
, struct extent_io_tree
*tree
,
2759 struct writeback_control
*wbc
,
2760 struct page
*page
, sector_t sector
,
2761 size_t size
, unsigned long offset
,
2762 struct block_device
*bdev
,
2763 struct bio
**bio_ret
,
2764 unsigned long max_pages
,
2765 bio_end_io_t end_io_func
,
2767 unsigned long prev_bio_flags
,
2768 unsigned long bio_flags
,
2769 bool force_bio_submit
)
2774 int old_compressed
= prev_bio_flags
& EXTENT_BIO_COMPRESSED
;
2775 size_t page_size
= min_t(size_t, size
, PAGE_SIZE
);
2777 if (bio_ret
&& *bio_ret
) {
2780 contig
= bio
->bi_iter
.bi_sector
== sector
;
2782 contig
= bio_end_sector(bio
) == sector
;
2784 if (prev_bio_flags
!= bio_flags
|| !contig
||
2786 merge_bio(tree
, page
, offset
, page_size
, bio
, bio_flags
) ||
2787 bio_add_page(bio
, page
, page_size
, offset
) < page_size
) {
2788 ret
= submit_one_bio(bio
, mirror_num
, prev_bio_flags
);
2796 wbc_account_io(wbc
, page
, page_size
);
2801 bio
= btrfs_bio_alloc(bdev
, sector
, BIO_MAX_PAGES
,
2802 GFP_NOFS
| __GFP_HIGH
);
2806 bio_add_page(bio
, page
, page_size
, offset
);
2807 bio
->bi_end_io
= end_io_func
;
2808 bio
->bi_private
= tree
;
2809 bio_set_op_attrs(bio
, op
, op_flags
);
2811 wbc_init_bio(wbc
, bio
);
2812 wbc_account_io(wbc
, page
, page_size
);
2818 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
2823 static void attach_extent_buffer_page(struct extent_buffer
*eb
,
2826 if (!PagePrivate(page
)) {
2827 SetPagePrivate(page
);
2829 set_page_private(page
, (unsigned long)eb
);
2831 WARN_ON(page
->private != (unsigned long)eb
);
2835 void set_page_extent_mapped(struct page
*page
)
2837 if (!PagePrivate(page
)) {
2838 SetPagePrivate(page
);
2840 set_page_private(page
, EXTENT_PAGE_PRIVATE
);
2844 static struct extent_map
*
2845 __get_extent_map(struct inode
*inode
, struct page
*page
, size_t pg_offset
,
2846 u64 start
, u64 len
, get_extent_t
*get_extent
,
2847 struct extent_map
**em_cached
)
2849 struct extent_map
*em
;
2851 if (em_cached
&& *em_cached
) {
2853 if (extent_map_in_tree(em
) && start
>= em
->start
&&
2854 start
< extent_map_end(em
)) {
2855 atomic_inc(&em
->refs
);
2859 free_extent_map(em
);
2863 em
= get_extent(inode
, page
, pg_offset
, start
, len
, 0);
2864 if (em_cached
&& !IS_ERR_OR_NULL(em
)) {
2866 atomic_inc(&em
->refs
);
2872 * basic readpage implementation. Locked extent state structs are inserted
2873 * into the tree that are removed when the IO is done (by the end_io
2875 * XXX JDM: This needs looking at to ensure proper page locking
2876 * return 0 on success, otherwise return error
2878 static int __do_readpage(struct extent_io_tree
*tree
,
2880 get_extent_t
*get_extent
,
2881 struct extent_map
**em_cached
,
2882 struct bio
**bio
, int mirror_num
,
2883 unsigned long *bio_flags
, int read_flags
,
2886 struct inode
*inode
= page
->mapping
->host
;
2887 u64 start
= page_offset(page
);
2888 u64 page_end
= start
+ PAGE_SIZE
- 1;
2892 u64 last_byte
= i_size_read(inode
);
2896 struct extent_map
*em
;
2897 struct block_device
*bdev
;
2900 size_t pg_offset
= 0;
2902 size_t disk_io_size
;
2903 size_t blocksize
= inode
->i_sb
->s_blocksize
;
2904 unsigned long this_bio_flag
= 0;
2906 set_page_extent_mapped(page
);
2909 if (!PageUptodate(page
)) {
2910 if (cleancache_get_page(page
) == 0) {
2911 BUG_ON(blocksize
!= PAGE_SIZE
);
2912 unlock_extent(tree
, start
, end
);
2917 if (page
->index
== last_byte
>> PAGE_SHIFT
) {
2919 size_t zero_offset
= last_byte
& (PAGE_SIZE
- 1);
2922 iosize
= PAGE_SIZE
- zero_offset
;
2923 userpage
= kmap_atomic(page
);
2924 memset(userpage
+ zero_offset
, 0, iosize
);
2925 flush_dcache_page(page
);
2926 kunmap_atomic(userpage
);
2929 while (cur
<= end
) {
2930 unsigned long pnr
= (last_byte
>> PAGE_SHIFT
) + 1;
2931 bool force_bio_submit
= false;
2933 if (cur
>= last_byte
) {
2935 struct extent_state
*cached
= NULL
;
2937 iosize
= PAGE_SIZE
- pg_offset
;
2938 userpage
= kmap_atomic(page
);
2939 memset(userpage
+ pg_offset
, 0, iosize
);
2940 flush_dcache_page(page
);
2941 kunmap_atomic(userpage
);
2942 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
2944 unlock_extent_cached(tree
, cur
,
2949 em
= __get_extent_map(inode
, page
, pg_offset
, cur
,
2950 end
- cur
+ 1, get_extent
, em_cached
);
2951 if (IS_ERR_OR_NULL(em
)) {
2953 unlock_extent(tree
, cur
, end
);
2956 extent_offset
= cur
- em
->start
;
2957 BUG_ON(extent_map_end(em
) <= cur
);
2960 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
2961 this_bio_flag
|= EXTENT_BIO_COMPRESSED
;
2962 extent_set_compress_type(&this_bio_flag
,
2966 iosize
= min(extent_map_end(em
) - cur
, end
- cur
+ 1);
2967 cur_end
= min(extent_map_end(em
) - 1, end
);
2968 iosize
= ALIGN(iosize
, blocksize
);
2969 if (this_bio_flag
& EXTENT_BIO_COMPRESSED
) {
2970 disk_io_size
= em
->block_len
;
2971 sector
= em
->block_start
>> 9;
2973 sector
= (em
->block_start
+ extent_offset
) >> 9;
2974 disk_io_size
= iosize
;
2977 block_start
= em
->block_start
;
2978 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
2979 block_start
= EXTENT_MAP_HOLE
;
2982 * If we have a file range that points to a compressed extent
2983 * and it's followed by a consecutive file range that points to
2984 * to the same compressed extent (possibly with a different
2985 * offset and/or length, so it either points to the whole extent
2986 * or only part of it), we must make sure we do not submit a
2987 * single bio to populate the pages for the 2 ranges because
2988 * this makes the compressed extent read zero out the pages
2989 * belonging to the 2nd range. Imagine the following scenario:
2992 * [0 - 8K] [8K - 24K]
2995 * points to extent X, points to extent X,
2996 * offset 4K, length of 8K offset 0, length 16K
2998 * [extent X, compressed length = 4K uncompressed length = 16K]
3000 * If the bio to read the compressed extent covers both ranges,
3001 * it will decompress extent X into the pages belonging to the
3002 * first range and then it will stop, zeroing out the remaining
3003 * pages that belong to the other range that points to extent X.
3004 * So here we make sure we submit 2 bios, one for the first
3005 * range and another one for the third range. Both will target
3006 * the same physical extent from disk, but we can't currently
3007 * make the compressed bio endio callback populate the pages
3008 * for both ranges because each compressed bio is tightly
3009 * coupled with a single extent map, and each range can have
3010 * an extent map with a different offset value relative to the
3011 * uncompressed data of our extent and different lengths. This
3012 * is a corner case so we prioritize correctness over
3013 * non-optimal behavior (submitting 2 bios for the same extent).
3015 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) &&
3016 prev_em_start
&& *prev_em_start
!= (u64
)-1 &&
3017 *prev_em_start
!= em
->orig_start
)
3018 force_bio_submit
= true;
3021 *prev_em_start
= em
->orig_start
;
3023 free_extent_map(em
);
3026 /* we've found a hole, just zero and go on */
3027 if (block_start
== EXTENT_MAP_HOLE
) {
3029 struct extent_state
*cached
= NULL
;
3031 userpage
= kmap_atomic(page
);
3032 memset(userpage
+ pg_offset
, 0, iosize
);
3033 flush_dcache_page(page
);
3034 kunmap_atomic(userpage
);
3036 set_extent_uptodate(tree
, cur
, cur
+ iosize
- 1,
3038 unlock_extent_cached(tree
, cur
,
3042 pg_offset
+= iosize
;
3045 /* the get_extent function already copied into the page */
3046 if (test_range_bit(tree
, cur
, cur_end
,
3047 EXTENT_UPTODATE
, 1, NULL
)) {
3048 check_page_uptodate(tree
, page
);
3049 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3051 pg_offset
+= iosize
;
3054 /* we have an inline extent but it didn't get marked up
3055 * to date. Error out
3057 if (block_start
== EXTENT_MAP_INLINE
) {
3059 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3061 pg_offset
+= iosize
;
3066 ret
= submit_extent_page(REQ_OP_READ
, read_flags
, tree
, NULL
,
3067 page
, sector
, disk_io_size
, pg_offset
,
3069 end_bio_extent_readpage
, mirror_num
,
3075 *bio_flags
= this_bio_flag
;
3078 unlock_extent(tree
, cur
, cur
+ iosize
- 1);
3082 pg_offset
+= iosize
;
3086 if (!PageError(page
))
3087 SetPageUptodate(page
);
3093 static inline void __do_contiguous_readpages(struct extent_io_tree
*tree
,
3094 struct page
*pages
[], int nr_pages
,
3096 get_extent_t
*get_extent
,
3097 struct extent_map
**em_cached
,
3098 struct bio
**bio
, int mirror_num
,
3099 unsigned long *bio_flags
,
3102 struct inode
*inode
;
3103 struct btrfs_ordered_extent
*ordered
;
3106 inode
= pages
[0]->mapping
->host
;
3108 lock_extent(tree
, start
, end
);
3109 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3113 unlock_extent(tree
, start
, end
);
3114 btrfs_start_ordered_extent(inode
, ordered
, 1);
3115 btrfs_put_ordered_extent(ordered
);
3118 for (index
= 0; index
< nr_pages
; index
++) {
3119 __do_readpage(tree
, pages
[index
], get_extent
, em_cached
, bio
,
3120 mirror_num
, bio_flags
, 0, prev_em_start
);
3121 put_page(pages
[index
]);
3125 static void __extent_readpages(struct extent_io_tree
*tree
,
3126 struct page
*pages
[],
3127 int nr_pages
, get_extent_t
*get_extent
,
3128 struct extent_map
**em_cached
,
3129 struct bio
**bio
, int mirror_num
,
3130 unsigned long *bio_flags
,
3137 int first_index
= 0;
3139 for (index
= 0; index
< nr_pages
; index
++) {
3140 page_start
= page_offset(pages
[index
]);
3143 end
= start
+ PAGE_SIZE
- 1;
3144 first_index
= index
;
3145 } else if (end
+ 1 == page_start
) {
3148 __do_contiguous_readpages(tree
, &pages
[first_index
],
3149 index
- first_index
, start
,
3150 end
, get_extent
, em_cached
,
3151 bio
, mirror_num
, bio_flags
,
3154 end
= start
+ PAGE_SIZE
- 1;
3155 first_index
= index
;
3160 __do_contiguous_readpages(tree
, &pages
[first_index
],
3161 index
- first_index
, start
,
3162 end
, get_extent
, em_cached
, bio
,
3163 mirror_num
, bio_flags
,
3167 static int __extent_read_full_page(struct extent_io_tree
*tree
,
3169 get_extent_t
*get_extent
,
3170 struct bio
**bio
, int mirror_num
,
3171 unsigned long *bio_flags
, int read_flags
)
3173 struct inode
*inode
= page
->mapping
->host
;
3174 struct btrfs_ordered_extent
*ordered
;
3175 u64 start
= page_offset(page
);
3176 u64 end
= start
+ PAGE_SIZE
- 1;
3180 lock_extent(tree
, start
, end
);
3181 ordered
= btrfs_lookup_ordered_range(inode
, start
,
3185 unlock_extent(tree
, start
, end
);
3186 btrfs_start_ordered_extent(inode
, ordered
, 1);
3187 btrfs_put_ordered_extent(ordered
);
3190 ret
= __do_readpage(tree
, page
, get_extent
, NULL
, bio
, mirror_num
,
3191 bio_flags
, read_flags
, NULL
);
3195 int extent_read_full_page(struct extent_io_tree
*tree
, struct page
*page
,
3196 get_extent_t
*get_extent
, int mirror_num
)
3198 struct bio
*bio
= NULL
;
3199 unsigned long bio_flags
= 0;
3202 ret
= __extent_read_full_page(tree
, page
, get_extent
, &bio
, mirror_num
,
3205 ret
= submit_one_bio(bio
, mirror_num
, bio_flags
);
3209 static void update_nr_written(struct page
*page
, struct writeback_control
*wbc
,
3210 unsigned long nr_written
)
3212 wbc
->nr_to_write
-= nr_written
;
3216 * helper for __extent_writepage, doing all of the delayed allocation setup.
3218 * This returns 1 if our fill_delalloc function did all the work required
3219 * to write the page (copy into inline extent). In this case the IO has
3220 * been started and the page is already unlocked.
3222 * This returns 0 if all went well (page still locked)
3223 * This returns < 0 if there were errors (page still locked)
3225 static noinline_for_stack
int writepage_delalloc(struct inode
*inode
,
3226 struct page
*page
, struct writeback_control
*wbc
,
3227 struct extent_page_data
*epd
,
3229 unsigned long *nr_written
)
3231 struct extent_io_tree
*tree
= epd
->tree
;
3232 u64 page_end
= delalloc_start
+ PAGE_SIZE
- 1;
3234 u64 delalloc_to_write
= 0;
3235 u64 delalloc_end
= 0;
3237 int page_started
= 0;
3239 if (epd
->extent_locked
|| !tree
->ops
|| !tree
->ops
->fill_delalloc
)
3242 while (delalloc_end
< page_end
) {
3243 nr_delalloc
= find_lock_delalloc_range(inode
, tree
,
3247 BTRFS_MAX_EXTENT_SIZE
);
3248 if (nr_delalloc
== 0) {
3249 delalloc_start
= delalloc_end
+ 1;
3252 ret
= tree
->ops
->fill_delalloc(inode
, page
,
3257 /* File system has been set read-only */
3260 /* fill_delalloc should be return < 0 for error
3261 * but just in case, we use > 0 here meaning the
3262 * IO is started, so we don't want to return > 0
3263 * unless things are going well.
3265 ret
= ret
< 0 ? ret
: -EIO
;
3269 * delalloc_end is already one less than the total length, so
3270 * we don't subtract one from PAGE_SIZE
3272 delalloc_to_write
+= (delalloc_end
- delalloc_start
+
3273 PAGE_SIZE
) >> PAGE_SHIFT
;
3274 delalloc_start
= delalloc_end
+ 1;
3276 if (wbc
->nr_to_write
< delalloc_to_write
) {
3279 if (delalloc_to_write
< thresh
* 2)
3280 thresh
= delalloc_to_write
;
3281 wbc
->nr_to_write
= min_t(u64
, delalloc_to_write
,
3285 /* did the fill delalloc function already unlock and start
3290 * we've unlocked the page, so we can't update
3291 * the mapping's writeback index, just update
3294 wbc
->nr_to_write
-= *nr_written
;
3305 * helper for __extent_writepage. This calls the writepage start hooks,
3306 * and does the loop to map the page into extents and bios.
3308 * We return 1 if the IO is started and the page is unlocked,
3309 * 0 if all went well (page still locked)
3310 * < 0 if there were errors (page still locked)
3312 static noinline_for_stack
int __extent_writepage_io(struct inode
*inode
,
3314 struct writeback_control
*wbc
,
3315 struct extent_page_data
*epd
,
3317 unsigned long nr_written
,
3318 int write_flags
, int *nr_ret
)
3320 struct extent_io_tree
*tree
= epd
->tree
;
3321 u64 start
= page_offset(page
);
3322 u64 page_end
= start
+ PAGE_SIZE
- 1;
3329 struct extent_state
*cached_state
= NULL
;
3330 struct extent_map
*em
;
3331 struct block_device
*bdev
;
3332 size_t pg_offset
= 0;
3338 if (tree
->ops
&& tree
->ops
->writepage_start_hook
) {
3339 ret
= tree
->ops
->writepage_start_hook(page
, start
,
3342 /* Fixup worker will requeue */
3344 wbc
->pages_skipped
++;
3346 redirty_page_for_writepage(wbc
, page
);
3348 update_nr_written(page
, wbc
, nr_written
);
3356 * we don't want to touch the inode after unlocking the page,
3357 * so we update the mapping writeback index now
3359 update_nr_written(page
, wbc
, nr_written
+ 1);
3362 if (i_size
<= start
) {
3363 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3364 tree
->ops
->writepage_end_io_hook(page
, start
,
3369 blocksize
= inode
->i_sb
->s_blocksize
;
3371 while (cur
<= end
) {
3373 unsigned long max_nr
;
3375 if (cur
>= i_size
) {
3376 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
3377 tree
->ops
->writepage_end_io_hook(page
, cur
,
3381 em
= epd
->get_extent(inode
, page
, pg_offset
, cur
,
3383 if (IS_ERR_OR_NULL(em
)) {
3385 ret
= PTR_ERR_OR_ZERO(em
);
3389 extent_offset
= cur
- em
->start
;
3390 em_end
= extent_map_end(em
);
3391 BUG_ON(em_end
<= cur
);
3393 iosize
= min(em_end
- cur
, end
- cur
+ 1);
3394 iosize
= ALIGN(iosize
, blocksize
);
3395 sector
= (em
->block_start
+ extent_offset
) >> 9;
3397 block_start
= em
->block_start
;
3398 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3399 free_extent_map(em
);
3403 * compressed and inline extents are written through other
3406 if (compressed
|| block_start
== EXTENT_MAP_HOLE
||
3407 block_start
== EXTENT_MAP_INLINE
) {
3409 * end_io notification does not happen here for
3410 * compressed extents
3412 if (!compressed
&& tree
->ops
&&
3413 tree
->ops
->writepage_end_io_hook
)
3414 tree
->ops
->writepage_end_io_hook(page
, cur
,
3417 else if (compressed
) {
3418 /* we don't want to end_page_writeback on
3419 * a compressed extent. this happens
3426 pg_offset
+= iosize
;
3430 max_nr
= (i_size
>> PAGE_SHIFT
) + 1;
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
,
3441 bdev
, &epd
->bio
, max_nr
,
3442 end_bio_extent_writepage
,
3448 pg_offset
+= iosize
;
3456 /* drop our reference on any cached states */
3457 free_extent_state(cached_state
);
3462 * the writepage semantics are similar to regular writepage. extent
3463 * records are inserted to lock ranges in the tree, and as dirty areas
3464 * are found, they are marked writeback. Then the lock bits are removed
3465 * and the end_io handler clears the writeback ranges
3467 static int __extent_writepage(struct page
*page
, struct writeback_control
*wbc
,
3470 struct inode
*inode
= page
->mapping
->host
;
3471 struct extent_page_data
*epd
= data
;
3472 u64 start
= page_offset(page
);
3473 u64 page_end
= start
+ PAGE_SIZE
- 1;
3476 size_t pg_offset
= 0;
3477 loff_t i_size
= i_size_read(inode
);
3478 unsigned long end_index
= i_size
>> PAGE_SHIFT
;
3479 int write_flags
= 0;
3480 unsigned long nr_written
= 0;
3482 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3483 write_flags
= WRITE_SYNC
;
3485 trace___extent_writepage(page
, inode
, wbc
);
3487 WARN_ON(!PageLocked(page
));
3489 ClearPageError(page
);
3491 pg_offset
= i_size
& (PAGE_SIZE
- 1);
3492 if (page
->index
> end_index
||
3493 (page
->index
== end_index
&& !pg_offset
)) {
3494 page
->mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
3499 if (page
->index
== end_index
) {
3502 userpage
= kmap_atomic(page
);
3503 memset(userpage
+ pg_offset
, 0,
3504 PAGE_SIZE
- pg_offset
);
3505 kunmap_atomic(userpage
);
3506 flush_dcache_page(page
);
3511 set_page_extent_mapped(page
);
3513 ret
= writepage_delalloc(inode
, page
, wbc
, epd
, start
, &nr_written
);
3519 ret
= __extent_writepage_io(inode
, page
, wbc
, epd
,
3520 i_size
, nr_written
, write_flags
, &nr
);
3526 /* make sure the mapping tag for page dirty gets cleared */
3527 set_page_writeback(page
);
3528 end_page_writeback(page
);
3530 if (PageError(page
)) {
3531 ret
= ret
< 0 ? ret
: -EIO
;
3532 end_extent_writepage(page
, ret
, start
, page_end
);
3541 void wait_on_extent_buffer_writeback(struct extent_buffer
*eb
)
3543 wait_on_bit_io(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
,
3544 TASK_UNINTERRUPTIBLE
);
3547 static noinline_for_stack
int
3548 lock_extent_buffer_for_io(struct extent_buffer
*eb
,
3549 struct btrfs_fs_info
*fs_info
,
3550 struct extent_page_data
*epd
)
3552 unsigned long i
, num_pages
;
3556 if (!btrfs_try_tree_write_lock(eb
)) {
3558 flush_write_bio(epd
);
3559 btrfs_tree_lock(eb
);
3562 if (test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
)) {
3563 btrfs_tree_unlock(eb
);
3567 flush_write_bio(epd
);
3571 wait_on_extent_buffer_writeback(eb
);
3572 btrfs_tree_lock(eb
);
3573 if (!test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
))
3575 btrfs_tree_unlock(eb
);
3580 * We need to do this to prevent races in people who check if the eb is
3581 * under IO since we can end up having no IO bits set for a short period
3584 spin_lock(&eb
->refs_lock
);
3585 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3586 set_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3587 spin_unlock(&eb
->refs_lock
);
3588 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3589 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
3591 fs_info
->dirty_metadata_batch
);
3594 spin_unlock(&eb
->refs_lock
);
3597 btrfs_tree_unlock(eb
);
3602 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3603 for (i
= 0; i
< num_pages
; i
++) {
3604 struct page
*p
= eb
->pages
[i
];
3606 if (!trylock_page(p
)) {
3608 flush_write_bio(epd
);
3618 static void end_extent_buffer_writeback(struct extent_buffer
*eb
)
3620 clear_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
);
3621 smp_mb__after_atomic();
3622 wake_up_bit(&eb
->bflags
, EXTENT_BUFFER_WRITEBACK
);
3625 static void set_btree_ioerr(struct page
*page
)
3627 struct extent_buffer
*eb
= (struct extent_buffer
*)page
->private;
3628 struct btrfs_inode
*btree_ino
= BTRFS_I(eb
->fs_info
->btree_inode
);
3631 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
))
3635 * If writeback for a btree extent that doesn't belong to a log tree
3636 * failed, increment the counter transaction->eb_write_errors.
3637 * We do this because while the transaction is running and before it's
3638 * committing (when we call filemap_fdata[write|wait]_range against
3639 * the btree inode), we might have
3640 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3641 * returns an error or an error happens during writeback, when we're
3642 * committing the transaction we wouldn't know about it, since the pages
3643 * can be no longer dirty nor marked anymore for writeback (if a
3644 * subsequent modification to the extent buffer didn't happen before the
3645 * transaction commit), which makes filemap_fdata[write|wait]_range not
3646 * able to find the pages tagged with SetPageError at transaction
3647 * commit time. So if this happens we must abort the transaction,
3648 * otherwise we commit a super block with btree roots that point to
3649 * btree nodes/leafs whose content on disk is invalid - either garbage
3650 * or the content of some node/leaf from a past generation that got
3651 * cowed or deleted and is no longer valid.
3653 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3654 * not be enough - we need to distinguish between log tree extents vs
3655 * non-log tree extents, and the next filemap_fdatawait_range() call
3656 * will catch and clear such errors in the mapping - and that call might
3657 * be from a log sync and not from a transaction commit. Also, checking
3658 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3659 * not done and would not be reliable - the eb might have been released
3660 * from memory and reading it back again means that flag would not be
3661 * set (since it's a runtime flag, not persisted on disk).
3663 * Using the flags below in the btree inode also makes us achieve the
3664 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3665 * writeback for all dirty pages and before filemap_fdatawait_range()
3666 * is called, the writeback for all dirty pages had already finished
3667 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3668 * filemap_fdatawait_range() would return success, as it could not know
3669 * that writeback errors happened (the pages were no longer tagged for
3672 switch (eb
->log_index
) {
3674 set_bit(BTRFS_INODE_BTREE_ERR
, &btree_ino
->runtime_flags
);
3677 set_bit(BTRFS_INODE_BTREE_LOG1_ERR
, &btree_ino
->runtime_flags
);
3680 set_bit(BTRFS_INODE_BTREE_LOG2_ERR
, &btree_ino
->runtime_flags
);
3683 BUG(); /* unexpected, logic error */
3687 static void end_bio_extent_buffer_writepage(struct bio
*bio
)
3689 struct bio_vec
*bvec
;
3690 struct extent_buffer
*eb
;
3693 bio_for_each_segment_all(bvec
, bio
, i
) {
3694 struct page
*page
= bvec
->bv_page
;
3696 eb
= (struct extent_buffer
*)page
->private;
3698 done
= atomic_dec_and_test(&eb
->io_pages
);
3700 if (bio
->bi_error
||
3701 test_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
)) {
3702 ClearPageUptodate(page
);
3703 set_btree_ioerr(page
);
3706 end_page_writeback(page
);
3711 end_extent_buffer_writeback(eb
);
3717 static noinline_for_stack
int write_one_eb(struct extent_buffer
*eb
,
3718 struct btrfs_fs_info
*fs_info
,
3719 struct writeback_control
*wbc
,
3720 struct extent_page_data
*epd
)
3722 struct block_device
*bdev
= fs_info
->fs_devices
->latest_bdev
;
3723 struct extent_io_tree
*tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
3724 u64 offset
= eb
->start
;
3725 unsigned long i
, num_pages
;
3726 unsigned long bio_flags
= 0;
3727 int write_flags
= (epd
->sync_io
? WRITE_SYNC
: 0) | REQ_META
;
3730 clear_bit(EXTENT_BUFFER_WRITE_ERR
, &eb
->bflags
);
3731 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
3732 atomic_set(&eb
->io_pages
, num_pages
);
3733 if (btrfs_header_owner(eb
) == BTRFS_TREE_LOG_OBJECTID
)
3734 bio_flags
= EXTENT_BIO_TREE_LOG
;
3736 for (i
= 0; i
< num_pages
; i
++) {
3737 struct page
*p
= eb
->pages
[i
];
3739 clear_page_dirty_for_io(p
);
3740 set_page_writeback(p
);
3741 ret
= submit_extent_page(REQ_OP_WRITE
, write_flags
, tree
, wbc
,
3742 p
, offset
>> 9, PAGE_SIZE
, 0, bdev
,
3744 end_bio_extent_buffer_writepage
,
3745 0, epd
->bio_flags
, bio_flags
, false);
3746 epd
->bio_flags
= bio_flags
;
3749 end_page_writeback(p
);
3750 if (atomic_sub_and_test(num_pages
- i
, &eb
->io_pages
))
3751 end_extent_buffer_writeback(eb
);
3755 offset
+= PAGE_SIZE
;
3756 update_nr_written(p
, wbc
, 1);
3760 if (unlikely(ret
)) {
3761 for (; i
< num_pages
; i
++) {
3762 struct page
*p
= eb
->pages
[i
];
3763 clear_page_dirty_for_io(p
);
3771 int btree_write_cache_pages(struct address_space
*mapping
,
3772 struct writeback_control
*wbc
)
3774 struct extent_io_tree
*tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3775 struct btrfs_fs_info
*fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
3776 struct extent_buffer
*eb
, *prev_eb
= NULL
;
3777 struct extent_page_data epd
= {
3781 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
3786 int nr_to_write_done
= 0;
3787 struct pagevec pvec
;
3790 pgoff_t end
; /* Inclusive */
3794 pagevec_init(&pvec
, 0);
3795 if (wbc
->range_cyclic
) {
3796 index
= mapping
->writeback_index
; /* Start from prev offset */
3799 index
= wbc
->range_start
>> PAGE_SHIFT
;
3800 end
= wbc
->range_end
>> PAGE_SHIFT
;
3803 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3804 tag
= PAGECACHE_TAG_TOWRITE
;
3806 tag
= PAGECACHE_TAG_DIRTY
;
3808 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3809 tag_pages_for_writeback(mapping
, index
, end
);
3810 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3811 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3812 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3816 for (i
= 0; i
< nr_pages
; i
++) {
3817 struct page
*page
= pvec
.pages
[i
];
3819 if (!PagePrivate(page
))
3822 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3827 spin_lock(&mapping
->private_lock
);
3828 if (!PagePrivate(page
)) {
3829 spin_unlock(&mapping
->private_lock
);
3833 eb
= (struct extent_buffer
*)page
->private;
3836 * Shouldn't happen and normally this would be a BUG_ON
3837 * but no sense in crashing the users box for something
3838 * we can survive anyway.
3841 spin_unlock(&mapping
->private_lock
);
3845 if (eb
== prev_eb
) {
3846 spin_unlock(&mapping
->private_lock
);
3850 ret
= atomic_inc_not_zero(&eb
->refs
);
3851 spin_unlock(&mapping
->private_lock
);
3856 ret
= lock_extent_buffer_for_io(eb
, fs_info
, &epd
);
3858 free_extent_buffer(eb
);
3862 ret
= write_one_eb(eb
, fs_info
, wbc
, &epd
);
3865 free_extent_buffer(eb
);
3868 free_extent_buffer(eb
);
3871 * the filesystem may choose to bump up nr_to_write.
3872 * We have to make sure to honor the new nr_to_write
3875 nr_to_write_done
= wbc
->nr_to_write
<= 0;
3877 pagevec_release(&pvec
);
3880 if (!scanned
&& !done
) {
3882 * We hit the last page and there is more work to be done: wrap
3883 * back to the start of the file
3889 flush_write_bio(&epd
);
3894 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3895 * @mapping: address space structure to write
3896 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3897 * @writepage: function called for each page
3898 * @data: data passed to writepage function
3900 * If a page is already under I/O, write_cache_pages() skips it, even
3901 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3902 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3903 * and msync() need to guarantee that all the data which was dirty at the time
3904 * the call was made get new I/O started against them. If wbc->sync_mode is
3905 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3906 * existing IO to complete.
3908 static int extent_write_cache_pages(struct extent_io_tree
*tree
,
3909 struct address_space
*mapping
,
3910 struct writeback_control
*wbc
,
3911 writepage_t writepage
, void *data
,
3912 void (*flush_fn
)(void *))
3914 struct inode
*inode
= mapping
->host
;
3917 int nr_to_write_done
= 0;
3918 struct pagevec pvec
;
3921 pgoff_t end
; /* Inclusive */
3923 int range_whole
= 0;
3928 * We have to hold onto the inode so that ordered extents can do their
3929 * work when the IO finishes. The alternative to this is failing to add
3930 * an ordered extent if the igrab() fails there and that is a huge pain
3931 * to deal with, so instead just hold onto the inode throughout the
3932 * writepages operation. If it fails here we are freeing up the inode
3933 * anyway and we'd rather not waste our time writing out stuff that is
3934 * going to be truncated anyway.
3939 pagevec_init(&pvec
, 0);
3940 if (wbc
->range_cyclic
) {
3941 index
= mapping
->writeback_index
; /* Start from prev offset */
3944 index
= wbc
->range_start
>> PAGE_SHIFT
;
3945 end
= wbc
->range_end
>> PAGE_SHIFT
;
3946 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
3950 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3951 tag
= PAGECACHE_TAG_TOWRITE
;
3953 tag
= PAGECACHE_TAG_DIRTY
;
3955 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3956 tag_pages_for_writeback(mapping
, index
, end
);
3958 while (!done
&& !nr_to_write_done
&& (index
<= end
) &&
3959 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
3960 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
3964 for (i
= 0; i
< nr_pages
; i
++) {
3965 struct page
*page
= pvec
.pages
[i
];
3967 done_index
= page
->index
;
3969 * At this point we hold neither mapping->tree_lock nor
3970 * lock on the page itself: the page may be truncated or
3971 * invalidated (changing page->mapping to NULL), or even
3972 * swizzled back from swapper_space to tmpfs file
3975 if (!trylock_page(page
)) {
3980 if (unlikely(page
->mapping
!= mapping
)) {
3985 if (!wbc
->range_cyclic
&& page
->index
> end
) {
3991 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3992 if (PageWriteback(page
))
3994 wait_on_page_writeback(page
);
3997 if (PageWriteback(page
) ||
3998 !clear_page_dirty_for_io(page
)) {
4003 ret
= (*writepage
)(page
, wbc
, data
);
4005 if (unlikely(ret
== AOP_WRITEPAGE_ACTIVATE
)) {
4011 * done_index is set past this page,
4012 * so media errors will not choke
4013 * background writeout for the entire
4014 * file. This has consequences for
4015 * range_cyclic semantics (ie. it may
4016 * not be suitable for data integrity
4019 done_index
= page
->index
+ 1;
4025 * the filesystem may choose to bump up nr_to_write.
4026 * We have to make sure to honor the new nr_to_write
4029 nr_to_write_done
= wbc
->nr_to_write
<= 0;
4031 pagevec_release(&pvec
);
4034 if (!scanned
&& !done
) {
4036 * We hit the last page and there is more work to be done: wrap
4037 * back to the start of the file
4044 if (wbc
->range_cyclic
|| (wbc
->nr_to_write
> 0 && range_whole
))
4045 mapping
->writeback_index
= done_index
;
4047 btrfs_add_delayed_iput(inode
);
4051 static void flush_epd_write_bio(struct extent_page_data
*epd
)
4056 bio_set_op_attrs(epd
->bio
, REQ_OP_WRITE
,
4057 epd
->sync_io
? WRITE_SYNC
: 0);
4059 ret
= submit_one_bio(epd
->bio
, 0, epd
->bio_flags
);
4060 BUG_ON(ret
< 0); /* -ENOMEM */
4065 static noinline
void flush_write_bio(void *data
)
4067 struct extent_page_data
*epd
= data
;
4068 flush_epd_write_bio(epd
);
4071 int extent_write_full_page(struct extent_io_tree
*tree
, struct page
*page
,
4072 get_extent_t
*get_extent
,
4073 struct writeback_control
*wbc
)
4076 struct extent_page_data epd
= {
4079 .get_extent
= get_extent
,
4081 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4085 ret
= __extent_writepage(page
, wbc
, &epd
);
4087 flush_epd_write_bio(&epd
);
4091 int extent_write_locked_range(struct extent_io_tree
*tree
, struct inode
*inode
,
4092 u64 start
, u64 end
, get_extent_t
*get_extent
,
4096 struct address_space
*mapping
= inode
->i_mapping
;
4098 unsigned long nr_pages
= (end
- start
+ PAGE_SIZE
) >>
4101 struct extent_page_data epd
= {
4104 .get_extent
= get_extent
,
4106 .sync_io
= mode
== WB_SYNC_ALL
,
4109 struct writeback_control wbc_writepages
= {
4111 .nr_to_write
= nr_pages
* 2,
4112 .range_start
= start
,
4113 .range_end
= end
+ 1,
4116 while (start
<= end
) {
4117 page
= find_get_page(mapping
, start
>> PAGE_SHIFT
);
4118 if (clear_page_dirty_for_io(page
))
4119 ret
= __extent_writepage(page
, &wbc_writepages
, &epd
);
4121 if (tree
->ops
&& tree
->ops
->writepage_end_io_hook
)
4122 tree
->ops
->writepage_end_io_hook(page
, start
,
4123 start
+ PAGE_SIZE
- 1,
4131 flush_epd_write_bio(&epd
);
4135 int extent_writepages(struct extent_io_tree
*tree
,
4136 struct address_space
*mapping
,
4137 get_extent_t
*get_extent
,
4138 struct writeback_control
*wbc
)
4141 struct extent_page_data epd
= {
4144 .get_extent
= get_extent
,
4146 .sync_io
= wbc
->sync_mode
== WB_SYNC_ALL
,
4150 ret
= extent_write_cache_pages(tree
, mapping
, wbc
,
4151 __extent_writepage
, &epd
,
4153 flush_epd_write_bio(&epd
);
4157 int extent_readpages(struct extent_io_tree
*tree
,
4158 struct address_space
*mapping
,
4159 struct list_head
*pages
, unsigned nr_pages
,
4160 get_extent_t get_extent
)
4162 struct bio
*bio
= NULL
;
4164 unsigned long bio_flags
= 0;
4165 struct page
*pagepool
[16];
4167 struct extent_map
*em_cached
= NULL
;
4169 u64 prev_em_start
= (u64
)-1;
4171 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
4172 page
= list_entry(pages
->prev
, struct page
, lru
);
4174 prefetchw(&page
->flags
);
4175 list_del(&page
->lru
);
4176 if (add_to_page_cache_lru(page
, mapping
,
4178 readahead_gfp_mask(mapping
))) {
4183 pagepool
[nr
++] = page
;
4184 if (nr
< ARRAY_SIZE(pagepool
))
4186 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4187 &bio
, 0, &bio_flags
, &prev_em_start
);
4191 __extent_readpages(tree
, pagepool
, nr
, get_extent
, &em_cached
,
4192 &bio
, 0, &bio_flags
, &prev_em_start
);
4195 free_extent_map(em_cached
);
4197 BUG_ON(!list_empty(pages
));
4199 return submit_one_bio(bio
, 0, bio_flags
);
4204 * basic invalidatepage code, this waits on any locked or writeback
4205 * ranges corresponding to the page, and then deletes any extent state
4206 * records from the tree
4208 int extent_invalidatepage(struct extent_io_tree
*tree
,
4209 struct page
*page
, unsigned long offset
)
4211 struct extent_state
*cached_state
= NULL
;
4212 u64 start
= page_offset(page
);
4213 u64 end
= start
+ PAGE_SIZE
- 1;
4214 size_t blocksize
= page
->mapping
->host
->i_sb
->s_blocksize
;
4216 start
+= ALIGN(offset
, blocksize
);
4220 lock_extent_bits(tree
, start
, end
, &cached_state
);
4221 wait_on_page_writeback(page
);
4222 clear_extent_bit(tree
, start
, end
,
4223 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4224 EXTENT_DO_ACCOUNTING
,
4225 1, 1, &cached_state
, GFP_NOFS
);
4230 * a helper for releasepage, this tests for areas of the page that
4231 * are locked or under IO and drops the related state bits if it is safe
4234 static int try_release_extent_state(struct extent_map_tree
*map
,
4235 struct extent_io_tree
*tree
,
4236 struct page
*page
, gfp_t mask
)
4238 u64 start
= page_offset(page
);
4239 u64 end
= start
+ PAGE_SIZE
- 1;
4242 if (test_range_bit(tree
, start
, end
,
4243 EXTENT_IOBITS
, 0, NULL
))
4246 if ((mask
& GFP_NOFS
) == GFP_NOFS
)
4249 * at this point we can safely clear everything except the
4250 * locked bit and the nodatasum bit
4252 ret
= clear_extent_bit(tree
, start
, end
,
4253 ~(EXTENT_LOCKED
| EXTENT_NODATASUM
),
4256 /* if clear_extent_bit failed for enomem reasons,
4257 * we can't allow the release to continue.
4268 * a helper for releasepage. As long as there are no locked extents
4269 * in the range corresponding to the page, both state records and extent
4270 * map records are removed
4272 int try_release_extent_mapping(struct extent_map_tree
*map
,
4273 struct extent_io_tree
*tree
, struct page
*page
,
4276 struct extent_map
*em
;
4277 u64 start
= page_offset(page
);
4278 u64 end
= start
+ PAGE_SIZE
- 1;
4280 if (gfpflags_allow_blocking(mask
) &&
4281 page
->mapping
->host
->i_size
> SZ_16M
) {
4283 while (start
<= end
) {
4284 len
= end
- start
+ 1;
4285 write_lock(&map
->lock
);
4286 em
= lookup_extent_mapping(map
, start
, len
);
4288 write_unlock(&map
->lock
);
4291 if (test_bit(EXTENT_FLAG_PINNED
, &em
->flags
) ||
4292 em
->start
!= start
) {
4293 write_unlock(&map
->lock
);
4294 free_extent_map(em
);
4297 if (!test_range_bit(tree
, em
->start
,
4298 extent_map_end(em
) - 1,
4299 EXTENT_LOCKED
| EXTENT_WRITEBACK
,
4301 remove_extent_mapping(map
, em
);
4302 /* once for the rb tree */
4303 free_extent_map(em
);
4305 start
= extent_map_end(em
);
4306 write_unlock(&map
->lock
);
4309 free_extent_map(em
);
4312 return try_release_extent_state(map
, tree
, page
, mask
);
4316 * helper function for fiemap, which doesn't want to see any holes.
4317 * This maps until we find something past 'last'
4319 static struct extent_map
*get_extent_skip_holes(struct inode
*inode
,
4322 get_extent_t
*get_extent
)
4324 u64 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
4325 struct extent_map
*em
;
4332 len
= last
- offset
;
4335 len
= ALIGN(len
, sectorsize
);
4336 em
= get_extent(inode
, NULL
, 0, offset
, len
, 0);
4337 if (IS_ERR_OR_NULL(em
))
4340 /* if this isn't a hole return it */
4341 if (!test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
) &&
4342 em
->block_start
!= EXTENT_MAP_HOLE
) {
4346 /* this is a hole, advance to the next extent */
4347 offset
= extent_map_end(em
);
4348 free_extent_map(em
);
4355 int extent_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4356 __u64 start
, __u64 len
, get_extent_t
*get_extent
)
4360 u64 max
= start
+ len
;
4364 u64 last_for_get_extent
= 0;
4366 u64 isize
= i_size_read(inode
);
4367 struct btrfs_key found_key
;
4368 struct extent_map
*em
= NULL
;
4369 struct extent_state
*cached_state
= NULL
;
4370 struct btrfs_path
*path
;
4371 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4380 path
= btrfs_alloc_path();
4383 path
->leave_spinning
= 1;
4385 start
= round_down(start
, BTRFS_I(inode
)->root
->sectorsize
);
4386 len
= round_up(max
, BTRFS_I(inode
)->root
->sectorsize
) - start
;
4389 * lookup the last file extent. We're not using i_size here
4390 * because there might be preallocation past i_size
4392 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
), -1,
4395 btrfs_free_path(path
);
4404 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
, path
->slots
[0]);
4405 found_type
= found_key
.type
;
4407 /* No extents, but there might be delalloc bits */
4408 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4409 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4410 /* have to trust i_size as the end */
4412 last_for_get_extent
= isize
;
4415 * remember the start of the last extent. There are a
4416 * bunch of different factors that go into the length of the
4417 * extent, so its much less complex to remember where it started
4419 last
= found_key
.offset
;
4420 last_for_get_extent
= last
+ 1;
4422 btrfs_release_path(path
);
4425 * we might have some extents allocated but more delalloc past those
4426 * extents. so, we trust isize unless the start of the last extent is
4431 last_for_get_extent
= isize
;
4434 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4437 em
= get_extent_skip_holes(inode
, start
, last_for_get_extent
,
4447 u64 offset_in_extent
= 0;
4449 /* break if the extent we found is outside the range */
4450 if (em
->start
>= max
|| extent_map_end(em
) < off
)
4454 * get_extent may return an extent that starts before our
4455 * requested range. We have to make sure the ranges
4456 * we return to fiemap always move forward and don't
4457 * overlap, so adjust the offsets here
4459 em_start
= max(em
->start
, off
);
4462 * record the offset from the start of the extent
4463 * for adjusting the disk offset below. Only do this if the
4464 * extent isn't compressed since our in ram offset may be past
4465 * what we have actually allocated on disk.
4467 if (!test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4468 offset_in_extent
= em_start
- em
->start
;
4469 em_end
= extent_map_end(em
);
4470 em_len
= em_end
- em_start
;
4475 * bump off for our next call to get_extent
4477 off
= extent_map_end(em
);
4481 if (em
->block_start
== EXTENT_MAP_LAST_BYTE
) {
4483 flags
|= FIEMAP_EXTENT_LAST
;
4484 } else if (em
->block_start
== EXTENT_MAP_INLINE
) {
4485 flags
|= (FIEMAP_EXTENT_DATA_INLINE
|
4486 FIEMAP_EXTENT_NOT_ALIGNED
);
4487 } else if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
4488 flags
|= (FIEMAP_EXTENT_DELALLOC
|
4489 FIEMAP_EXTENT_UNKNOWN
);
4490 } else if (fieinfo
->fi_extents_max
) {
4491 struct btrfs_trans_handle
*trans
;
4493 u64 bytenr
= em
->block_start
-
4494 (em
->start
- em
->orig_start
);
4496 disko
= em
->block_start
+ offset_in_extent
;
4499 * We need a trans handle to get delayed refs
4501 trans
= btrfs_join_transaction(root
);
4503 * It's OK if we can't start a trans we can still check
4510 * As btrfs supports shared space, this information
4511 * can be exported to userspace tools via
4512 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4513 * then we're just getting a count and we can skip the
4516 ret
= btrfs_check_shared(trans
, root
->fs_info
,
4518 btrfs_ino(inode
), bytenr
);
4520 btrfs_end_transaction(trans
, root
);
4524 flags
|= FIEMAP_EXTENT_SHARED
;
4527 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
4528 flags
|= FIEMAP_EXTENT_ENCODED
;
4529 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4530 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4532 free_extent_map(em
);
4534 if ((em_start
>= last
) || em_len
== (u64
)-1 ||
4535 (last
== (u64
)-1 && isize
<= em_end
)) {
4536 flags
|= FIEMAP_EXTENT_LAST
;
4540 /* now scan forward to see if this is really the last extent. */
4541 em
= get_extent_skip_holes(inode
, off
, last_for_get_extent
,
4548 flags
|= FIEMAP_EXTENT_LAST
;
4551 ret
= fiemap_fill_next_extent(fieinfo
, em_start
, disko
,
4560 free_extent_map(em
);
4562 btrfs_free_path(path
);
4563 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
4564 &cached_state
, GFP_NOFS
);
4568 static void __free_extent_buffer(struct extent_buffer
*eb
)
4570 btrfs_leak_debug_del(&eb
->leak_list
);
4571 kmem_cache_free(extent_buffer_cache
, eb
);
4574 int extent_buffer_under_io(struct extent_buffer
*eb
)
4576 return (atomic_read(&eb
->io_pages
) ||
4577 test_bit(EXTENT_BUFFER_WRITEBACK
, &eb
->bflags
) ||
4578 test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4582 * Helper for releasing extent buffer page.
4584 static void btrfs_release_extent_buffer_page(struct extent_buffer
*eb
)
4586 unsigned long index
;
4588 int mapped
= !test_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4590 BUG_ON(extent_buffer_under_io(eb
));
4592 index
= num_extent_pages(eb
->start
, eb
->len
);
4598 page
= eb
->pages
[index
];
4602 spin_lock(&page
->mapping
->private_lock
);
4604 * We do this since we'll remove the pages after we've
4605 * removed the eb from the radix tree, so we could race
4606 * and have this page now attached to the new eb. So
4607 * only clear page_private if it's still connected to
4610 if (PagePrivate(page
) &&
4611 page
->private == (unsigned long)eb
) {
4612 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
4613 BUG_ON(PageDirty(page
));
4614 BUG_ON(PageWriteback(page
));
4616 * We need to make sure we haven't be attached
4619 ClearPagePrivate(page
);
4620 set_page_private(page
, 0);
4621 /* One for the page private */
4626 spin_unlock(&page
->mapping
->private_lock
);
4628 /* One for when we allocated the page */
4630 } while (index
!= 0);
4634 * Helper for releasing the extent buffer.
4636 static inline void btrfs_release_extent_buffer(struct extent_buffer
*eb
)
4638 btrfs_release_extent_buffer_page(eb
);
4639 __free_extent_buffer(eb
);
4642 static struct extent_buffer
*
4643 __alloc_extent_buffer(struct btrfs_fs_info
*fs_info
, u64 start
,
4646 struct extent_buffer
*eb
= NULL
;
4648 eb
= kmem_cache_zalloc(extent_buffer_cache
, GFP_NOFS
|__GFP_NOFAIL
);
4651 eb
->fs_info
= fs_info
;
4653 rwlock_init(&eb
->lock
);
4654 atomic_set(&eb
->write_locks
, 0);
4655 atomic_set(&eb
->read_locks
, 0);
4656 atomic_set(&eb
->blocking_readers
, 0);
4657 atomic_set(&eb
->blocking_writers
, 0);
4658 atomic_set(&eb
->spinning_readers
, 0);
4659 atomic_set(&eb
->spinning_writers
, 0);
4660 eb
->lock_nested
= 0;
4661 init_waitqueue_head(&eb
->write_lock_wq
);
4662 init_waitqueue_head(&eb
->read_lock_wq
);
4664 btrfs_leak_debug_add(&eb
->leak_list
, &buffers
);
4666 spin_lock_init(&eb
->refs_lock
);
4667 atomic_set(&eb
->refs
, 1);
4668 atomic_set(&eb
->io_pages
, 0);
4671 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4673 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4674 > MAX_INLINE_EXTENT_BUFFER_SIZE
);
4675 BUG_ON(len
> MAX_INLINE_EXTENT_BUFFER_SIZE
);
4680 struct extent_buffer
*btrfs_clone_extent_buffer(struct extent_buffer
*src
)
4684 struct extent_buffer
*new;
4685 unsigned long num_pages
= num_extent_pages(src
->start
, src
->len
);
4687 new = __alloc_extent_buffer(src
->fs_info
, src
->start
, src
->len
);
4691 for (i
= 0; i
< num_pages
; i
++) {
4692 p
= alloc_page(GFP_NOFS
);
4694 btrfs_release_extent_buffer(new);
4697 attach_extent_buffer_page(new, p
);
4698 WARN_ON(PageDirty(p
));
4703 copy_extent_buffer(new, src
, 0, 0, src
->len
);
4704 set_bit(EXTENT_BUFFER_UPTODATE
, &new->bflags
);
4705 set_bit(EXTENT_BUFFER_DUMMY
, &new->bflags
);
4710 struct extent_buffer
*__alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4711 u64 start
, unsigned long len
)
4713 struct extent_buffer
*eb
;
4714 unsigned long num_pages
;
4717 num_pages
= num_extent_pages(start
, len
);
4719 eb
= __alloc_extent_buffer(fs_info
, start
, len
);
4723 for (i
= 0; i
< num_pages
; i
++) {
4724 eb
->pages
[i
] = alloc_page(GFP_NOFS
);
4728 set_extent_buffer_uptodate(eb
);
4729 btrfs_set_header_nritems(eb
, 0);
4730 set_bit(EXTENT_BUFFER_DUMMY
, &eb
->bflags
);
4735 __free_page(eb
->pages
[i
- 1]);
4736 __free_extent_buffer(eb
);
4740 struct extent_buffer
*alloc_dummy_extent_buffer(struct btrfs_fs_info
*fs_info
,
4741 u64 start
, u32 nodesize
)
4747 * Called only from tests that don't always have a fs_info
4752 len
= fs_info
->tree_root
->nodesize
;
4755 return __alloc_dummy_extent_buffer(fs_info
, start
, len
);
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
,
4846 u64 start
, u32 nodesize
)
4848 struct extent_buffer
*eb
, *exists
= NULL
;
4851 eb
= find_extent_buffer(fs_info
, start
);
4854 eb
= alloc_dummy_extent_buffer(fs_info
, start
, nodesize
);
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
->tree_root
->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
->tree_root
->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
, u64 start
, int wait
,
5193 get_extent_t
*get_extent
, int mirror_num
)
5196 unsigned long start_i
;
5200 int locked_pages
= 0;
5201 int all_uptodate
= 1;
5202 unsigned long num_pages
;
5203 unsigned long num_reads
= 0;
5204 struct bio
*bio
= NULL
;
5205 unsigned long bio_flags
= 0;
5207 if (test_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
))
5211 WARN_ON(start
< eb
->start
);
5212 start_i
= (start
>> PAGE_SHIFT
) -
5213 (eb
->start
>> PAGE_SHIFT
);
5218 num_pages
= num_extent_pages(eb
->start
, eb
->len
);
5219 for (i
= start_i
; i
< num_pages
; i
++) {
5220 page
= eb
->pages
[i
];
5221 if (wait
== WAIT_NONE
) {
5222 if (!trylock_page(page
))
5228 if (!PageUptodate(page
)) {
5235 set_bit(EXTENT_BUFFER_UPTODATE
, &eb
->bflags
);
5239 clear_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
5240 eb
->read_mirror
= 0;
5241 atomic_set(&eb
->io_pages
, num_reads
);
5242 for (i
= start_i
; i
< num_pages
; i
++) {
5243 page
= eb
->pages
[i
];
5245 if (!PageUptodate(page
)) {
5247 atomic_dec(&eb
->io_pages
);
5252 ClearPageError(page
);
5253 err
= __extent_read_full_page(tree
, page
,
5255 mirror_num
, &bio_flags
,
5260 * We use &bio in above __extent_read_full_page,
5261 * so we ensure that if it returns error, the
5262 * current page fails to add itself to bio and
5263 * it's been unlocked.
5265 * We must dec io_pages by ourselves.
5267 atomic_dec(&eb
->io_pages
);
5275 err
= submit_one_bio(bio
, mirror_num
, bio_flags
);
5280 if (ret
|| wait
!= WAIT_COMPLETE
)
5283 for (i
= start_i
; i
< num_pages
; i
++) {
5284 page
= eb
->pages
[i
];
5285 wait_on_page_locked(page
);
5286 if (!PageUptodate(page
))
5294 while (locked_pages
> 0) {
5295 page
= eb
->pages
[i
];
5303 void read_extent_buffer(struct extent_buffer
*eb
, void *dstv
,
5304 unsigned long start
,
5311 char *dst
= (char *)dstv
;
5312 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5313 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5315 WARN_ON(start
> eb
->len
);
5316 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5318 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5321 page
= eb
->pages
[i
];
5323 cur
= min(len
, (PAGE_SIZE
- offset
));
5324 kaddr
= page_address(page
);
5325 memcpy(dst
, kaddr
+ offset
, cur
);
5334 int read_extent_buffer_to_user(struct extent_buffer
*eb
, void __user
*dstv
,
5335 unsigned long start
,
5342 char __user
*dst
= (char __user
*)dstv
;
5343 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5344 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5347 WARN_ON(start
> eb
->len
);
5348 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5350 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5353 page
= eb
->pages
[i
];
5355 cur
= min(len
, (PAGE_SIZE
- offset
));
5356 kaddr
= page_address(page
);
5357 if (copy_to_user(dst
, kaddr
+ offset
, cur
)) {
5372 * return 0 if the item is found within a page.
5373 * return 1 if the item spans two pages.
5374 * return -EINVAL otherwise.
5376 int map_private_extent_buffer(struct extent_buffer
*eb
, unsigned long start
,
5377 unsigned long min_len
, char **map
,
5378 unsigned long *map_start
,
5379 unsigned long *map_len
)
5381 size_t offset
= start
& (PAGE_SIZE
- 1);
5384 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5385 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5386 unsigned long end_i
= (start_offset
+ start
+ min_len
- 1) >>
5393 offset
= start_offset
;
5397 *map_start
= ((u64
)i
<< PAGE_SHIFT
) - start_offset
;
5400 if (start
+ min_len
> eb
->len
) {
5401 WARN(1, KERN_ERR
"btrfs bad mapping eb start %llu len %lu, "
5403 eb
->start
, eb
->len
, start
, min_len
);
5408 kaddr
= page_address(p
);
5409 *map
= kaddr
+ offset
;
5410 *map_len
= PAGE_SIZE
- offset
;
5414 int memcmp_extent_buffer(struct extent_buffer
*eb
, const void *ptrv
,
5415 unsigned long start
,
5422 char *ptr
= (char *)ptrv
;
5423 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5424 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5427 WARN_ON(start
> eb
->len
);
5428 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5430 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5433 page
= eb
->pages
[i
];
5435 cur
= min(len
, (PAGE_SIZE
- offset
));
5437 kaddr
= page_address(page
);
5438 ret
= memcmp(ptr
, kaddr
+ offset
, cur
);
5450 void write_extent_buffer(struct extent_buffer
*eb
, const void *srcv
,
5451 unsigned long start
, unsigned long len
)
5457 char *src
= (char *)srcv
;
5458 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5459 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5461 WARN_ON(start
> eb
->len
);
5462 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5464 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5467 page
= eb
->pages
[i
];
5468 WARN_ON(!PageUptodate(page
));
5470 cur
= min(len
, PAGE_SIZE
- offset
);
5471 kaddr
= page_address(page
);
5472 memcpy(kaddr
+ offset
, src
, cur
);
5481 void memset_extent_buffer(struct extent_buffer
*eb
, char c
,
5482 unsigned long start
, unsigned long len
)
5488 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5489 unsigned long i
= (start_offset
+ start
) >> PAGE_SHIFT
;
5491 WARN_ON(start
> eb
->len
);
5492 WARN_ON(start
+ len
> eb
->start
+ eb
->len
);
5494 offset
= (start_offset
+ start
) & (PAGE_SIZE
- 1);
5497 page
= eb
->pages
[i
];
5498 WARN_ON(!PageUptodate(page
));
5500 cur
= min(len
, PAGE_SIZE
- offset
);
5501 kaddr
= page_address(page
);
5502 memset(kaddr
+ offset
, c
, cur
);
5510 void copy_extent_buffer(struct extent_buffer
*dst
, struct extent_buffer
*src
,
5511 unsigned long dst_offset
, unsigned long src_offset
,
5514 u64 dst_len
= dst
->len
;
5519 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5520 unsigned long i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5522 WARN_ON(src
->len
!= dst_len
);
5524 offset
= (start_offset
+ dst_offset
) &
5528 page
= dst
->pages
[i
];
5529 WARN_ON(!PageUptodate(page
));
5531 cur
= min(len
, (unsigned long)(PAGE_SIZE
- offset
));
5533 kaddr
= page_address(page
);
5534 read_extent_buffer(src
, kaddr
+ offset
, src_offset
, cur
);
5544 * The extent buffer bitmap operations are done with byte granularity because
5545 * bitmap items are not guaranteed to be aligned to a word and therefore a
5546 * single word in a bitmap may straddle two pages in the extent buffer.
5548 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
5549 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
5550 #define BITMAP_FIRST_BYTE_MASK(start) \
5551 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
5552 #define BITMAP_LAST_BYTE_MASK(nbits) \
5553 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
5556 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5558 * @eb: the extent buffer
5559 * @start: offset of the bitmap item in the extent buffer
5561 * @page_index: return index of the page in the extent buffer that contains the
5563 * @page_offset: return offset into the page given by page_index
5565 * This helper hides the ugliness of finding the byte in an extent buffer which
5566 * contains a given bit.
5568 static inline void eb_bitmap_offset(struct extent_buffer
*eb
,
5569 unsigned long start
, unsigned long nr
,
5570 unsigned long *page_index
,
5571 size_t *page_offset
)
5573 size_t start_offset
= eb
->start
& ((u64
)PAGE_SIZE
- 1);
5574 size_t byte_offset
= BIT_BYTE(nr
);
5578 * The byte we want is the offset of the extent buffer + the offset of
5579 * the bitmap item in the extent buffer + the offset of the byte in the
5582 offset
= start_offset
+ start
+ byte_offset
;
5584 *page_index
= offset
>> PAGE_SHIFT
;
5585 *page_offset
= offset
& (PAGE_SIZE
- 1);
5589 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5590 * @eb: the extent buffer
5591 * @start: offset of the bitmap item in the extent buffer
5592 * @nr: bit number to test
5594 int extent_buffer_test_bit(struct extent_buffer
*eb
, unsigned long start
,
5602 eb_bitmap_offset(eb
, start
, nr
, &i
, &offset
);
5603 page
= eb
->pages
[i
];
5604 WARN_ON(!PageUptodate(page
));
5605 kaddr
= page_address(page
);
5606 return 1U & (kaddr
[offset
] >> (nr
& (BITS_PER_BYTE
- 1)));
5610 * extent_buffer_bitmap_set - set an area of a bitmap
5611 * @eb: the extent buffer
5612 * @start: offset of the bitmap item in the extent buffer
5613 * @pos: bit number of the first bit
5614 * @len: number of bits to set
5616 void extent_buffer_bitmap_set(struct extent_buffer
*eb
, unsigned long start
,
5617 unsigned long pos
, unsigned long len
)
5623 const unsigned int size
= pos
+ len
;
5624 int bits_to_set
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5625 unsigned int mask_to_set
= BITMAP_FIRST_BYTE_MASK(pos
);
5627 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5628 page
= eb
->pages
[i
];
5629 WARN_ON(!PageUptodate(page
));
5630 kaddr
= page_address(page
);
5632 while (len
>= bits_to_set
) {
5633 kaddr
[offset
] |= mask_to_set
;
5635 bits_to_set
= BITS_PER_BYTE
;
5637 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5639 page
= eb
->pages
[++i
];
5640 WARN_ON(!PageUptodate(page
));
5641 kaddr
= page_address(page
);
5645 mask_to_set
&= BITMAP_LAST_BYTE_MASK(size
);
5646 kaddr
[offset
] |= mask_to_set
;
5652 * extent_buffer_bitmap_clear - clear an area of a bitmap
5653 * @eb: the extent buffer
5654 * @start: offset of the bitmap item in the extent buffer
5655 * @pos: bit number of the first bit
5656 * @len: number of bits to clear
5658 void extent_buffer_bitmap_clear(struct extent_buffer
*eb
, unsigned long start
,
5659 unsigned long pos
, unsigned long len
)
5665 const unsigned int size
= pos
+ len
;
5666 int bits_to_clear
= BITS_PER_BYTE
- (pos
% BITS_PER_BYTE
);
5667 unsigned int mask_to_clear
= BITMAP_FIRST_BYTE_MASK(pos
);
5669 eb_bitmap_offset(eb
, start
, pos
, &i
, &offset
);
5670 page
= eb
->pages
[i
];
5671 WARN_ON(!PageUptodate(page
));
5672 kaddr
= page_address(page
);
5674 while (len
>= bits_to_clear
) {
5675 kaddr
[offset
] &= ~mask_to_clear
;
5676 len
-= bits_to_clear
;
5677 bits_to_clear
= BITS_PER_BYTE
;
5678 mask_to_clear
= ~0U;
5679 if (++offset
>= PAGE_SIZE
&& len
> 0) {
5681 page
= eb
->pages
[++i
];
5682 WARN_ON(!PageUptodate(page
));
5683 kaddr
= page_address(page
);
5687 mask_to_clear
&= BITMAP_LAST_BYTE_MASK(size
);
5688 kaddr
[offset
] &= ~mask_to_clear
;
5692 static inline bool areas_overlap(unsigned long src
, unsigned long dst
, unsigned long len
)
5694 unsigned long distance
= (src
> dst
) ? src
- dst
: dst
- src
;
5695 return distance
< len
;
5698 static void copy_pages(struct page
*dst_page
, struct page
*src_page
,
5699 unsigned long dst_off
, unsigned long src_off
,
5702 char *dst_kaddr
= page_address(dst_page
);
5704 int must_memmove
= 0;
5706 if (dst_page
!= src_page
) {
5707 src_kaddr
= page_address(src_page
);
5709 src_kaddr
= dst_kaddr
;
5710 if (areas_overlap(src_off
, dst_off
, len
))
5715 memmove(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5717 memcpy(dst_kaddr
+ dst_off
, src_kaddr
+ src_off
, len
);
5720 void memcpy_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5721 unsigned long src_offset
, unsigned long len
)
5724 size_t dst_off_in_page
;
5725 size_t src_off_in_page
;
5726 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5727 unsigned long dst_i
;
5728 unsigned long src_i
;
5730 if (src_offset
+ len
> dst
->len
) {
5731 btrfs_err(dst
->fs_info
,
5732 "memmove bogus src_offset %lu move "
5733 "len %lu dst len %lu", src_offset
, len
, dst
->len
);
5736 if (dst_offset
+ len
> dst
->len
) {
5737 btrfs_err(dst
->fs_info
,
5738 "memmove bogus dst_offset %lu move "
5739 "len %lu dst len %lu", dst_offset
, len
, dst
->len
);
5744 dst_off_in_page
= (start_offset
+ dst_offset
) &
5746 src_off_in_page
= (start_offset
+ src_offset
) &
5749 dst_i
= (start_offset
+ dst_offset
) >> PAGE_SHIFT
;
5750 src_i
= (start_offset
+ src_offset
) >> PAGE_SHIFT
;
5752 cur
= min(len
, (unsigned long)(PAGE_SIZE
-
5754 cur
= min_t(unsigned long, cur
,
5755 (unsigned long)(PAGE_SIZE
- dst_off_in_page
));
5757 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5758 dst_off_in_page
, src_off_in_page
, cur
);
5766 void memmove_extent_buffer(struct extent_buffer
*dst
, unsigned long dst_offset
,
5767 unsigned long src_offset
, unsigned long len
)
5770 size_t dst_off_in_page
;
5771 size_t src_off_in_page
;
5772 unsigned long dst_end
= dst_offset
+ len
- 1;
5773 unsigned long src_end
= src_offset
+ len
- 1;
5774 size_t start_offset
= dst
->start
& ((u64
)PAGE_SIZE
- 1);
5775 unsigned long dst_i
;
5776 unsigned long src_i
;
5778 if (src_offset
+ len
> dst
->len
) {
5779 btrfs_err(dst
->fs_info
, "memmove bogus src_offset %lu move "
5780 "len %lu len %lu", src_offset
, len
, dst
->len
);
5783 if (dst_offset
+ len
> dst
->len
) {
5784 btrfs_err(dst
->fs_info
, "memmove bogus dst_offset %lu move "
5785 "len %lu len %lu", dst_offset
, len
, dst
->len
);
5788 if (dst_offset
< src_offset
) {
5789 memcpy_extent_buffer(dst
, dst_offset
, src_offset
, len
);
5793 dst_i
= (start_offset
+ dst_end
) >> PAGE_SHIFT
;
5794 src_i
= (start_offset
+ src_end
) >> PAGE_SHIFT
;
5796 dst_off_in_page
= (start_offset
+ dst_end
) &
5798 src_off_in_page
= (start_offset
+ src_end
) &
5801 cur
= min_t(unsigned long, len
, src_off_in_page
+ 1);
5802 cur
= min(cur
, dst_off_in_page
+ 1);
5803 copy_pages(dst
->pages
[dst_i
], dst
->pages
[src_i
],
5804 dst_off_in_page
- cur
+ 1,
5805 src_off_in_page
- cur
+ 1, cur
);
5813 int try_release_extent_buffer(struct page
*page
)
5815 struct extent_buffer
*eb
;
5818 * We need to make sure nobody is attaching this page to an eb right
5821 spin_lock(&page
->mapping
->private_lock
);
5822 if (!PagePrivate(page
)) {
5823 spin_unlock(&page
->mapping
->private_lock
);
5827 eb
= (struct extent_buffer
*)page
->private;
5831 * This is a little awful but should be ok, we need to make sure that
5832 * the eb doesn't disappear out from under us while we're looking at
5835 spin_lock(&eb
->refs_lock
);
5836 if (atomic_read(&eb
->refs
) != 1 || extent_buffer_under_io(eb
)) {
5837 spin_unlock(&eb
->refs_lock
);
5838 spin_unlock(&page
->mapping
->private_lock
);
5841 spin_unlock(&page
->mapping
->private_lock
);
5844 * If tree ref isn't set then we know the ref on this eb is a real ref,
5845 * so just return, this page will likely be freed soon anyway.
5847 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF
, &eb
->bflags
)) {
5848 spin_unlock(&eb
->refs_lock
);
5852 return release_extent_buffer(eb
);