1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
26 static struct bio_set *btrfs_bioset;
28 static inline bool extent_state_in_tree(const struct extent_state *state)
30 return !RB_EMPTY_NODE(&state->rb_node);
33 #ifdef CONFIG_BTRFS_DEBUG
34 static LIST_HEAD(buffers);
35 static LIST_HEAD(states);
37 static DEFINE_SPINLOCK(leak_lock);
40 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
44 spin_lock_irqsave(&leak_lock, flags);
46 spin_unlock_irqrestore(&leak_lock, flags);
50 void btrfs_leak_debug_del(struct list_head *entry)
54 spin_lock_irqsave(&leak_lock, flags);
56 spin_unlock_irqrestore(&leak_lock, flags);
60 void btrfs_leak_debug_check(void)
62 struct extent_state *state;
63 struct extent_buffer *eb;
65 while (!list_empty(&states)) {
66 state = list_entry(states.next, struct extent_state, leak_list);
67 pr_err("BTRFS: state leak: start %llu end %llu state %lu in tree %d refs %d\n",
68 state->start, state->end, state->state,
69 extent_state_in_tree(state),
70 atomic_read(&state->refs));
71 list_del(&state->leak_list);
72 kmem_cache_free(extent_state_cache, state);
75 while (!list_empty(&buffers)) {
76 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
77 printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu "
79 eb->start, eb->len, atomic_read(&eb->refs));
80 list_del(&eb->leak_list);
81 kmem_cache_free(extent_buffer_cache, eb);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
88 struct extent_io_tree *tree, u64 start, u64 end)
96 inode = tree->mapping->host;
97 isize = i_size_read(inode);
98 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
99 printk_ratelimited(KERN_DEBUG
100 "BTRFS: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
101 caller, btrfs_ino(inode), isize, start, end);
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry) do {} while (0)
107 #define btrfs_leak_debug_check() do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
111 #define BUFFER_LRU_MAX 64
116 struct rb_node rb_node;
119 struct extent_page_data {
121 struct extent_io_tree *tree;
122 get_extent_t *get_extent;
123 unsigned long bio_flags;
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
128 unsigned int extent_locked:1;
130 /* tells the submit_bio code to use a WRITE_SYNC */
131 unsigned int sync_io:1;
134 static noinline void flush_write_bio(void *data);
135 static inline struct btrfs_fs_info *
136 tree_fs_info(struct extent_io_tree *tree)
140 return btrfs_sb(tree->mapping->host->i_sb);
143 int __init extent_io_init(void)
145 extent_state_cache = kmem_cache_create("btrfs_extent_state",
146 sizeof(struct extent_state), 0,
147 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
148 if (!extent_state_cache)
151 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
152 sizeof(struct extent_buffer), 0,
153 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
154 if (!extent_buffer_cache)
155 goto free_state_cache;
157 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
158 offsetof(struct btrfs_io_bio, bio));
160 goto free_buffer_cache;
162 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
168 bioset_free(btrfs_bioset);
172 kmem_cache_destroy(extent_buffer_cache);
173 extent_buffer_cache = NULL;
176 kmem_cache_destroy(extent_state_cache);
177 extent_state_cache = NULL;
181 void extent_io_exit(void)
183 btrfs_leak_debug_check();
186 * Make sure all delayed rcu free are flushed before we
190 if (extent_state_cache)
191 kmem_cache_destroy(extent_state_cache);
192 if (extent_buffer_cache)
193 kmem_cache_destroy(extent_buffer_cache);
195 bioset_free(btrfs_bioset);
198 void extent_io_tree_init(struct extent_io_tree *tree,
199 struct address_space *mapping)
201 tree->state = RB_ROOT;
203 tree->dirty_bytes = 0;
204 spin_lock_init(&tree->lock);
205 tree->mapping = mapping;
208 static struct extent_state *alloc_extent_state(gfp_t mask)
210 struct extent_state *state;
212 state = kmem_cache_alloc(extent_state_cache, mask);
217 RB_CLEAR_NODE(&state->rb_node);
218 btrfs_leak_debug_add(&state->leak_list, &states);
219 atomic_set(&state->refs, 1);
220 init_waitqueue_head(&state->wq);
221 trace_alloc_extent_state(state, mask, _RET_IP_);
225 void free_extent_state(struct extent_state *state)
229 if (atomic_dec_and_test(&state->refs)) {
230 WARN_ON(extent_state_in_tree(state));
231 btrfs_leak_debug_del(&state->leak_list);
232 trace_free_extent_state(state, _RET_IP_);
233 kmem_cache_free(extent_state_cache, state);
237 static struct rb_node *tree_insert(struct rb_root *root,
238 struct rb_node *search_start,
240 struct rb_node *node,
241 struct rb_node ***p_in,
242 struct rb_node **parent_in)
245 struct rb_node *parent = NULL;
246 struct tree_entry *entry;
248 if (p_in && parent_in) {
254 p = search_start ? &search_start : &root->rb_node;
257 entry = rb_entry(parent, struct tree_entry, rb_node);
259 if (offset < entry->start)
261 else if (offset > entry->end)
268 rb_link_node(node, parent, p);
269 rb_insert_color(node, root);
273 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
274 struct rb_node **prev_ret,
275 struct rb_node **next_ret,
276 struct rb_node ***p_ret,
277 struct rb_node **parent_ret)
279 struct rb_root *root = &tree->state;
280 struct rb_node **n = &root->rb_node;
281 struct rb_node *prev = NULL;
282 struct rb_node *orig_prev = NULL;
283 struct tree_entry *entry;
284 struct tree_entry *prev_entry = NULL;
288 entry = rb_entry(prev, struct tree_entry, rb_node);
291 if (offset < entry->start)
293 else if (offset > entry->end)
306 while (prev && offset > prev_entry->end) {
307 prev = rb_next(prev);
308 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
315 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
316 while (prev && offset < prev_entry->start) {
317 prev = rb_prev(prev);
318 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
325 static inline struct rb_node *
326 tree_search_for_insert(struct extent_io_tree *tree,
328 struct rb_node ***p_ret,
329 struct rb_node **parent_ret)
331 struct rb_node *prev = NULL;
334 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
340 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
343 return tree_search_for_insert(tree, offset, NULL, NULL);
346 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
347 struct extent_state *other)
349 if (tree->ops && tree->ops->merge_extent_hook)
350 tree->ops->merge_extent_hook(tree->mapping->host, new,
355 * utility function to look for merge candidates inside a given range.
356 * Any extents with matching state are merged together into a single
357 * extent in the tree. Extents with EXTENT_IO in their state field
358 * are not merged because the end_io handlers need to be able to do
359 * operations on them without sleeping (or doing allocations/splits).
361 * This should be called with the tree lock held.
363 static void merge_state(struct extent_io_tree *tree,
364 struct extent_state *state)
366 struct extent_state *other;
367 struct rb_node *other_node;
369 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
372 other_node = rb_prev(&state->rb_node);
374 other = rb_entry(other_node, struct extent_state, rb_node);
375 if (other->end == state->start - 1 &&
376 other->state == state->state) {
377 merge_cb(tree, state, other);
378 state->start = other->start;
379 rb_erase(&other->rb_node, &tree->state);
380 RB_CLEAR_NODE(&other->rb_node);
381 free_extent_state(other);
384 other_node = rb_next(&state->rb_node);
386 other = rb_entry(other_node, struct extent_state, rb_node);
387 if (other->start == state->end + 1 &&
388 other->state == state->state) {
389 merge_cb(tree, state, other);
390 state->end = other->end;
391 rb_erase(&other->rb_node, &tree->state);
392 RB_CLEAR_NODE(&other->rb_node);
393 free_extent_state(other);
398 static void set_state_cb(struct extent_io_tree *tree,
399 struct extent_state *state, unsigned long *bits)
401 if (tree->ops && tree->ops->set_bit_hook)
402 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
405 static void clear_state_cb(struct extent_io_tree *tree,
406 struct extent_state *state, unsigned long *bits)
408 if (tree->ops && tree->ops->clear_bit_hook)
409 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
412 static void set_state_bits(struct extent_io_tree *tree,
413 struct extent_state *state, unsigned long *bits);
416 * insert an extent_state struct into the tree. 'bits' are set on the
417 * struct before it is inserted.
419 * This may return -EEXIST if the extent is already there, in which case the
420 * state struct is freed.
422 * The tree lock is not taken internally. This is a utility function and
423 * probably isn't what you want to call (see set/clear_extent_bit).
425 static int insert_state(struct extent_io_tree *tree,
426 struct extent_state *state, u64 start, u64 end,
428 struct rb_node **parent,
431 struct rb_node *node;
434 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
436 state->start = start;
439 set_state_bits(tree, state, bits);
441 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
443 struct extent_state *found;
444 found = rb_entry(node, struct extent_state, rb_node);
445 printk(KERN_ERR "BTRFS: found node %llu %llu on insert of "
447 found->start, found->end, start, end);
450 merge_state(tree, state);
454 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
457 if (tree->ops && tree->ops->split_extent_hook)
458 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
462 * split a given extent state struct in two, inserting the preallocated
463 * struct 'prealloc' as the newly created second half. 'split' indicates an
464 * offset inside 'orig' where it should be split.
467 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
468 * are two extent state structs in the tree:
469 * prealloc: [orig->start, split - 1]
470 * orig: [ split, orig->end ]
472 * The tree locks are not taken by this function. They need to be held
475 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
476 struct extent_state *prealloc, u64 split)
478 struct rb_node *node;
480 split_cb(tree, orig, split);
482 prealloc->start = orig->start;
483 prealloc->end = split - 1;
484 prealloc->state = orig->state;
487 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
488 &prealloc->rb_node, NULL, NULL);
490 free_extent_state(prealloc);
496 static struct extent_state *next_state(struct extent_state *state)
498 struct rb_node *next = rb_next(&state->rb_node);
500 return rb_entry(next, struct extent_state, rb_node);
506 * utility function to clear some bits in an extent state struct.
507 * it will optionally wake up any one waiting on this state (wake == 1).
509 * If no bits are set on the state struct after clearing things, the
510 * struct is freed and removed from the tree
512 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
513 struct extent_state *state,
514 unsigned long *bits, int wake)
516 struct extent_state *next;
517 unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
519 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
520 u64 range = state->end - state->start + 1;
521 WARN_ON(range > tree->dirty_bytes);
522 tree->dirty_bytes -= range;
524 clear_state_cb(tree, state, bits);
525 state->state &= ~bits_to_clear;
528 if (state->state == 0) {
529 next = next_state(state);
530 if (extent_state_in_tree(state)) {
531 rb_erase(&state->rb_node, &tree->state);
532 RB_CLEAR_NODE(&state->rb_node);
533 free_extent_state(state);
538 merge_state(tree, state);
539 next = next_state(state);
544 static struct extent_state *
545 alloc_extent_state_atomic(struct extent_state *prealloc)
548 prealloc = alloc_extent_state(GFP_ATOMIC);
553 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
555 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
556 "Extent tree was modified by another "
557 "thread while locked.");
561 * clear some bits on a range in the tree. This may require splitting
562 * or inserting elements in the tree, so the gfp mask is used to
563 * indicate which allocations or sleeping are allowed.
565 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
566 * the given range from the tree regardless of state (ie for truncate).
568 * the range [start, end] is inclusive.
570 * This takes the tree lock, and returns 0 on success and < 0 on error.
572 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
573 unsigned long bits, int wake, int delete,
574 struct extent_state **cached_state,
577 struct extent_state *state;
578 struct extent_state *cached;
579 struct extent_state *prealloc = NULL;
580 struct rb_node *node;
585 btrfs_debug_check_extent_io_range(tree, start, end);
587 if (bits & EXTENT_DELALLOC)
588 bits |= EXTENT_NORESERVE;
591 bits |= ~EXTENT_CTLBITS;
592 bits |= EXTENT_FIRST_DELALLOC;
594 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
597 if (!prealloc && (mask & __GFP_WAIT)) {
598 prealloc = alloc_extent_state(mask);
603 spin_lock(&tree->lock);
605 cached = *cached_state;
608 *cached_state = NULL;
612 if (cached && extent_state_in_tree(cached) &&
613 cached->start <= start && cached->end > start) {
615 atomic_dec(&cached->refs);
620 free_extent_state(cached);
623 * this search will find the extents that end after
626 node = tree_search(tree, start);
629 state = rb_entry(node, struct extent_state, rb_node);
631 if (state->start > end)
633 WARN_ON(state->end < start);
634 last_end = state->end;
636 /* the state doesn't have the wanted bits, go ahead */
637 if (!(state->state & bits)) {
638 state = next_state(state);
643 * | ---- desired range ---- |
645 * | ------------- state -------------- |
647 * We need to split the extent we found, and may flip
648 * bits on second half.
650 * If the extent we found extends past our range, we
651 * just split and search again. It'll get split again
652 * the next time though.
654 * If the extent we found is inside our range, we clear
655 * the desired bit on it.
658 if (state->start < start) {
659 prealloc = alloc_extent_state_atomic(prealloc);
661 err = split_state(tree, state, prealloc, start);
663 extent_io_tree_panic(tree, err);
668 if (state->end <= end) {
669 state = clear_state_bit(tree, state, &bits, wake);
675 * | ---- desired range ---- |
677 * We need to split the extent, and clear the bit
680 if (state->start <= end && state->end > end) {
681 prealloc = alloc_extent_state_atomic(prealloc);
683 err = split_state(tree, state, prealloc, end + 1);
685 extent_io_tree_panic(tree, err);
690 clear_state_bit(tree, prealloc, &bits, wake);
696 state = clear_state_bit(tree, state, &bits, wake);
698 if (last_end == (u64)-1)
700 start = last_end + 1;
701 if (start <= end && state && !need_resched())
706 spin_unlock(&tree->lock);
708 free_extent_state(prealloc);
715 spin_unlock(&tree->lock);
716 if (mask & __GFP_WAIT)
721 static void wait_on_state(struct extent_io_tree *tree,
722 struct extent_state *state)
723 __releases(tree->lock)
724 __acquires(tree->lock)
727 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
728 spin_unlock(&tree->lock);
730 spin_lock(&tree->lock);
731 finish_wait(&state->wq, &wait);
735 * waits for one or more bits to clear on a range in the state tree.
736 * The range [start, end] is inclusive.
737 * The tree lock is taken by this function
739 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
742 struct extent_state *state;
743 struct rb_node *node;
745 btrfs_debug_check_extent_io_range(tree, start, end);
747 spin_lock(&tree->lock);
751 * this search will find all the extents that end after
754 node = tree_search(tree, start);
759 state = rb_entry(node, struct extent_state, rb_node);
761 if (state->start > end)
764 if (state->state & bits) {
765 start = state->start;
766 atomic_inc(&state->refs);
767 wait_on_state(tree, state);
768 free_extent_state(state);
771 start = state->end + 1;
776 if (!cond_resched_lock(&tree->lock)) {
777 node = rb_next(node);
782 spin_unlock(&tree->lock);
785 static void set_state_bits(struct extent_io_tree *tree,
786 struct extent_state *state,
789 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
791 set_state_cb(tree, state, bits);
792 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
793 u64 range = state->end - state->start + 1;
794 tree->dirty_bytes += range;
796 state->state |= bits_to_set;
799 static void cache_state(struct extent_state *state,
800 struct extent_state **cached_ptr)
802 if (cached_ptr && !(*cached_ptr)) {
803 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
805 atomic_inc(&state->refs);
811 * set some bits on a range in the tree. This may require allocations or
812 * sleeping, so the gfp mask is used to indicate what is allowed.
814 * If any of the exclusive bits are set, this will fail with -EEXIST if some
815 * part of the range already has the desired bits set. The start of the
816 * existing range is returned in failed_start in this case.
818 * [start, end] is inclusive This takes the tree lock.
821 static int __must_check
822 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
823 unsigned long bits, unsigned long exclusive_bits,
824 u64 *failed_start, struct extent_state **cached_state,
827 struct extent_state *state;
828 struct extent_state *prealloc = NULL;
829 struct rb_node *node;
831 struct rb_node *parent;
836 btrfs_debug_check_extent_io_range(tree, start, end);
838 bits |= EXTENT_FIRST_DELALLOC;
840 if (!prealloc && (mask & __GFP_WAIT)) {
841 prealloc = alloc_extent_state(mask);
845 spin_lock(&tree->lock);
846 if (cached_state && *cached_state) {
847 state = *cached_state;
848 if (state->start <= start && state->end > start &&
849 extent_state_in_tree(state)) {
850 node = &state->rb_node;
855 * this search will find all the extents that end after
858 node = tree_search_for_insert(tree, start, &p, &parent);
860 prealloc = alloc_extent_state_atomic(prealloc);
862 err = insert_state(tree, prealloc, start, end,
865 extent_io_tree_panic(tree, err);
867 cache_state(prealloc, cached_state);
871 state = rb_entry(node, struct extent_state, rb_node);
873 last_start = state->start;
874 last_end = state->end;
877 * | ---- desired range ---- |
880 * Just lock what we found and keep going
882 if (state->start == start && state->end <= end) {
883 if (state->state & exclusive_bits) {
884 *failed_start = state->start;
889 set_state_bits(tree, state, &bits);
890 cache_state(state, cached_state);
891 merge_state(tree, state);
892 if (last_end == (u64)-1)
894 start = last_end + 1;
895 state = next_state(state);
896 if (start < end && state && state->start == start &&
903 * | ---- desired range ---- |
906 * | ------------- state -------------- |
908 * We need to split the extent we found, and may flip bits on
911 * If the extent we found extends past our
912 * range, we just split and search again. It'll get split
913 * again the next time though.
915 * If the extent we found is inside our range, we set the
918 if (state->start < start) {
919 if (state->state & exclusive_bits) {
920 *failed_start = start;
925 prealloc = alloc_extent_state_atomic(prealloc);
927 err = split_state(tree, state, prealloc, start);
929 extent_io_tree_panic(tree, err);
934 if (state->end <= end) {
935 set_state_bits(tree, state, &bits);
936 cache_state(state, cached_state);
937 merge_state(tree, state);
938 if (last_end == (u64)-1)
940 start = last_end + 1;
941 state = next_state(state);
942 if (start < end && state && state->start == start &&
949 * | ---- desired range ---- |
950 * | state | or | state |
952 * There's a hole, we need to insert something in it and
953 * ignore the extent we found.
955 if (state->start > start) {
957 if (end < last_start)
960 this_end = last_start - 1;
962 prealloc = alloc_extent_state_atomic(prealloc);
966 * Avoid to free 'prealloc' if it can be merged with
969 err = insert_state(tree, prealloc, start, this_end,
972 extent_io_tree_panic(tree, err);
974 cache_state(prealloc, cached_state);
976 start = this_end + 1;
980 * | ---- desired range ---- |
982 * We need to split the extent, and set the bit
985 if (state->start <= end && state->end > end) {
986 if (state->state & exclusive_bits) {
987 *failed_start = start;
992 prealloc = alloc_extent_state_atomic(prealloc);
994 err = split_state(tree, state, prealloc, end + 1);
996 extent_io_tree_panic(tree, err);
998 set_state_bits(tree, prealloc, &bits);
999 cache_state(prealloc, cached_state);
1000 merge_state(tree, prealloc);
1008 spin_unlock(&tree->lock);
1010 free_extent_state(prealloc);
1017 spin_unlock(&tree->lock);
1018 if (mask & __GFP_WAIT)
1023 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1024 unsigned long bits, u64 * failed_start,
1025 struct extent_state **cached_state, gfp_t mask)
1027 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1028 cached_state, mask);
1033 * convert_extent_bit - convert all bits in a given range from one bit to
1035 * @tree: the io tree to search
1036 * @start: the start offset in bytes
1037 * @end: the end offset in bytes (inclusive)
1038 * @bits: the bits to set in this range
1039 * @clear_bits: the bits to clear in this range
1040 * @cached_state: state that we're going to cache
1041 * @mask: the allocation mask
1043 * This will go through and set bits for the given range. If any states exist
1044 * already in this range they are set with the given bit and cleared of the
1045 * clear_bits. This is only meant to be used by things that are mergeable, ie
1046 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1047 * boundary bits like LOCK.
1049 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1050 unsigned long bits, unsigned long clear_bits,
1051 struct extent_state **cached_state, gfp_t mask)
1053 struct extent_state *state;
1054 struct extent_state *prealloc = NULL;
1055 struct rb_node *node;
1057 struct rb_node *parent;
1062 btrfs_debug_check_extent_io_range(tree, start, end);
1065 if (!prealloc && (mask & __GFP_WAIT)) {
1066 prealloc = alloc_extent_state(mask);
1071 spin_lock(&tree->lock);
1072 if (cached_state && *cached_state) {
1073 state = *cached_state;
1074 if (state->start <= start && state->end > start &&
1075 extent_state_in_tree(state)) {
1076 node = &state->rb_node;
1082 * this search will find all the extents that end after
1085 node = tree_search_for_insert(tree, start, &p, &parent);
1087 prealloc = alloc_extent_state_atomic(prealloc);
1092 err = insert_state(tree, prealloc, start, end,
1093 &p, &parent, &bits);
1095 extent_io_tree_panic(tree, err);
1096 cache_state(prealloc, cached_state);
1100 state = rb_entry(node, struct extent_state, rb_node);
1102 last_start = state->start;
1103 last_end = state->end;
1106 * | ---- desired range ---- |
1109 * Just lock what we found and keep going
1111 if (state->start == start && state->end <= end) {
1112 set_state_bits(tree, state, &bits);
1113 cache_state(state, cached_state);
1114 state = clear_state_bit(tree, state, &clear_bits, 0);
1115 if (last_end == (u64)-1)
1117 start = last_end + 1;
1118 if (start < end && state && state->start == start &&
1125 * | ---- desired range ---- |
1128 * | ------------- state -------------- |
1130 * We need to split the extent we found, and may flip bits on
1133 * If the extent we found extends past our
1134 * range, we just split and search again. It'll get split
1135 * again the next time though.
1137 * If the extent we found is inside our range, we set the
1138 * desired bit on it.
1140 if (state->start < start) {
1141 prealloc = alloc_extent_state_atomic(prealloc);
1146 err = split_state(tree, state, prealloc, start);
1148 extent_io_tree_panic(tree, err);
1152 if (state->end <= end) {
1153 set_state_bits(tree, state, &bits);
1154 cache_state(state, cached_state);
1155 state = clear_state_bit(tree, state, &clear_bits, 0);
1156 if (last_end == (u64)-1)
1158 start = last_end + 1;
1159 if (start < end && state && state->start == start &&
1166 * | ---- desired range ---- |
1167 * | state | or | state |
1169 * There's a hole, we need to insert something in it and
1170 * ignore the extent we found.
1172 if (state->start > start) {
1174 if (end < last_start)
1177 this_end = last_start - 1;
1179 prealloc = alloc_extent_state_atomic(prealloc);
1186 * Avoid to free 'prealloc' if it can be merged with
1189 err = insert_state(tree, prealloc, start, this_end,
1192 extent_io_tree_panic(tree, err);
1193 cache_state(prealloc, cached_state);
1195 start = this_end + 1;
1199 * | ---- desired range ---- |
1201 * We need to split the extent, and set the bit
1204 if (state->start <= end && state->end > end) {
1205 prealloc = alloc_extent_state_atomic(prealloc);
1211 err = split_state(tree, state, prealloc, end + 1);
1213 extent_io_tree_panic(tree, err);
1215 set_state_bits(tree, prealloc, &bits);
1216 cache_state(prealloc, cached_state);
1217 clear_state_bit(tree, prealloc, &clear_bits, 0);
1225 spin_unlock(&tree->lock);
1227 free_extent_state(prealloc);
1234 spin_unlock(&tree->lock);
1235 if (mask & __GFP_WAIT)
1240 /* wrappers around set/clear extent bit */
1241 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1244 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1248 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1249 unsigned long bits, gfp_t mask)
1251 return set_extent_bit(tree, start, end, bits, NULL,
1255 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1256 unsigned long bits, gfp_t mask)
1258 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1261 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1262 struct extent_state **cached_state, gfp_t mask)
1264 return set_extent_bit(tree, start, end,
1265 EXTENT_DELALLOC | EXTENT_UPTODATE,
1266 NULL, cached_state, mask);
1269 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1270 struct extent_state **cached_state, gfp_t mask)
1272 return set_extent_bit(tree, start, end,
1273 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1274 NULL, cached_state, mask);
1277 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1280 return clear_extent_bit(tree, start, end,
1281 EXTENT_DIRTY | EXTENT_DELALLOC |
1282 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1285 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1288 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1292 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1293 struct extent_state **cached_state, gfp_t mask)
1295 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1296 cached_state, mask);
1299 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1300 struct extent_state **cached_state, gfp_t mask)
1302 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1303 cached_state, mask);
1307 * either insert or lock state struct between start and end use mask to tell
1308 * us if waiting is desired.
1310 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1311 unsigned long bits, struct extent_state **cached_state)
1316 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1317 EXTENT_LOCKED, &failed_start,
1318 cached_state, GFP_NOFS);
1319 if (err == -EEXIST) {
1320 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1321 start = failed_start;
1324 WARN_ON(start > end);
1329 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1331 return lock_extent_bits(tree, start, end, 0, NULL);
1334 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1339 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1340 &failed_start, NULL, GFP_NOFS);
1341 if (err == -EEXIST) {
1342 if (failed_start > start)
1343 clear_extent_bit(tree, start, failed_start - 1,
1344 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1350 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1351 struct extent_state **cached, gfp_t mask)
1353 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1357 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1359 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1363 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1365 unsigned long index = start >> PAGE_CACHE_SHIFT;
1366 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1369 while (index <= end_index) {
1370 page = find_get_page(inode->i_mapping, index);
1371 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1372 clear_page_dirty_for_io(page);
1373 page_cache_release(page);
1379 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1381 unsigned long index = start >> PAGE_CACHE_SHIFT;
1382 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1385 while (index <= end_index) {
1386 page = find_get_page(inode->i_mapping, index);
1387 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1388 account_page_redirty(page);
1389 __set_page_dirty_nobuffers(page);
1390 page_cache_release(page);
1397 * helper function to set both pages and extents in the tree writeback
1399 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1401 unsigned long index = start >> PAGE_CACHE_SHIFT;
1402 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1405 while (index <= end_index) {
1406 page = find_get_page(tree->mapping, index);
1407 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1408 set_page_writeback(page);
1409 page_cache_release(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 long 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 long 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(state, cached_state);
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_CACHE_SHIFT;
1562 unsigned long end_index = end >> PAGE_CACHE_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]);
1576 page_cache_release(pages[i]);
1584 static noinline int lock_delalloc_pages(struct inode *inode,
1585 struct page *locked_page,
1589 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1590 unsigned long start_index = index;
1591 unsigned long end_index = delalloc_end >> PAGE_CACHE_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]);
1624 page_cache_release(pages[i]);
1628 page_cache_release(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_CACHE_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, 0, &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 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1733 struct page *locked_page,
1734 unsigned long 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_CACHE_SHIFT;
1741 unsigned long end_index = end >> PAGE_CACHE_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 while (nr_pages > 0) {
1750 ret = find_get_pages_contig(inode->i_mapping, index,
1751 min_t(unsigned long,
1752 nr_pages, ARRAY_SIZE(pages)), pages);
1753 for (i = 0; i < ret; i++) {
1755 if (page_ops & PAGE_SET_PRIVATE2)
1756 SetPagePrivate2(pages[i]);
1758 if (pages[i] == locked_page) {
1759 page_cache_release(pages[i]);
1762 if (page_ops & PAGE_CLEAR_DIRTY)
1763 clear_page_dirty_for_io(pages[i]);
1764 if (page_ops & PAGE_SET_WRITEBACK)
1765 set_page_writeback(pages[i]);
1766 if (page_ops & PAGE_END_WRITEBACK)
1767 end_page_writeback(pages[i]);
1768 if (page_ops & PAGE_UNLOCK)
1769 unlock_page(pages[i]);
1770 page_cache_release(pages[i]);
1780 * count the number of bytes in the tree that have a given bit(s)
1781 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1782 * cached. The total number found is returned.
1784 u64 count_range_bits(struct extent_io_tree *tree,
1785 u64 *start, u64 search_end, u64 max_bytes,
1786 unsigned long bits, int contig)
1788 struct rb_node *node;
1789 struct extent_state *state;
1790 u64 cur_start = *start;
1791 u64 total_bytes = 0;
1795 if (WARN_ON(search_end <= cur_start))
1798 spin_lock(&tree->lock);
1799 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1800 total_bytes = tree->dirty_bytes;
1804 * this search will find all the extents that end after
1807 node = tree_search(tree, cur_start);
1812 state = rb_entry(node, struct extent_state, rb_node);
1813 if (state->start > search_end)
1815 if (contig && found && state->start > last + 1)
1817 if (state->end >= cur_start && (state->state & bits) == bits) {
1818 total_bytes += min(search_end, state->end) + 1 -
1819 max(cur_start, state->start);
1820 if (total_bytes >= max_bytes)
1823 *start = max(cur_start, state->start);
1827 } else if (contig && found) {
1830 node = rb_next(node);
1835 spin_unlock(&tree->lock);
1840 * set the private field for a given byte offset in the tree. If there isn't
1841 * an extent_state there already, this does nothing.
1843 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1845 struct rb_node *node;
1846 struct extent_state *state;
1849 spin_lock(&tree->lock);
1851 * this search will find all the extents that end after
1854 node = tree_search(tree, start);
1859 state = rb_entry(node, struct extent_state, rb_node);
1860 if (state->start != start) {
1864 state->private = private;
1866 spin_unlock(&tree->lock);
1870 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1872 struct rb_node *node;
1873 struct extent_state *state;
1876 spin_lock(&tree->lock);
1878 * this search will find all the extents that end after
1881 node = tree_search(tree, start);
1886 state = rb_entry(node, struct extent_state, rb_node);
1887 if (state->start != start) {
1891 *private = state->private;
1893 spin_unlock(&tree->lock);
1898 * searches a range in the state tree for a given mask.
1899 * If 'filled' == 1, this returns 1 only if every extent in the tree
1900 * has the bits set. Otherwise, 1 is returned if any bit in the
1901 * range is found set.
1903 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1904 unsigned long bits, int filled, struct extent_state *cached)
1906 struct extent_state *state = NULL;
1907 struct rb_node *node;
1910 spin_lock(&tree->lock);
1911 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1912 cached->end > start)
1913 node = &cached->rb_node;
1915 node = tree_search(tree, start);
1916 while (node && start <= end) {
1917 state = rb_entry(node, struct extent_state, rb_node);
1919 if (filled && state->start > start) {
1924 if (state->start > end)
1927 if (state->state & bits) {
1931 } else if (filled) {
1936 if (state->end == (u64)-1)
1939 start = state->end + 1;
1942 node = rb_next(node);
1949 spin_unlock(&tree->lock);
1954 * helper function to set a given page up to date if all the
1955 * extents in the tree for that page are up to date
1957 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1959 u64 start = page_offset(page);
1960 u64 end = start + PAGE_CACHE_SIZE - 1;
1961 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1962 SetPageUptodate(page);
1966 * When IO fails, either with EIO or csum verification fails, we
1967 * try other mirrors that might have a good copy of the data. This
1968 * io_failure_record is used to record state as we go through all the
1969 * mirrors. If another mirror has good data, the page is set up to date
1970 * and things continue. If a good mirror can't be found, the original
1971 * bio end_io callback is called to indicate things have failed.
1973 struct io_failure_record {
1978 unsigned long bio_flags;
1984 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1989 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1991 set_state_private(failure_tree, rec->start, 0);
1992 ret = clear_extent_bits(failure_tree, rec->start,
1993 rec->start + rec->len - 1,
1994 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1998 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1999 rec->start + rec->len - 1,
2000 EXTENT_DAMAGED, GFP_NOFS);
2009 * this bypasses the standard btrfs submit functions deliberately, as
2010 * the standard behavior is to write all copies in a raid setup. here we only
2011 * want to write the one bad copy. so we do the mapping for ourselves and issue
2012 * submit_bio directly.
2013 * to avoid any synchronization issues, wait for the data after writing, which
2014 * actually prevents the read that triggered the error from finishing.
2015 * currently, there can be no more than two copies of every data bit. thus,
2016 * exactly one rewrite is required.
2018 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
2019 u64 length, u64 logical, struct page *page,
2023 struct btrfs_device *dev;
2026 struct btrfs_bio *bbio = NULL;
2027 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2030 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2031 BUG_ON(!mirror_num);
2033 /* we can't repair anything in raid56 yet */
2034 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2037 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2040 bio->bi_iter.bi_size = 0;
2041 map_length = length;
2043 ret = btrfs_map_block(fs_info, WRITE, logical,
2044 &map_length, &bbio, mirror_num);
2049 BUG_ON(mirror_num != bbio->mirror_num);
2050 sector = bbio->stripes[mirror_num-1].physical >> 9;
2051 bio->bi_iter.bi_sector = sector;
2052 dev = bbio->stripes[mirror_num-1].dev;
2054 if (!dev || !dev->bdev || !dev->writeable) {
2058 bio->bi_bdev = dev->bdev;
2059 bio_add_page(bio, page, length, start - page_offset(page));
2061 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
2062 /* try to remap that extent elsewhere? */
2064 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2068 printk_ratelimited_in_rcu(KERN_INFO
2069 "BTRFS: read error corrected: ino %lu off %llu "
2070 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2071 start, rcu_str_deref(dev->name), sector);
2077 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2080 u64 start = eb->start;
2081 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2084 if (root->fs_info->sb->s_flags & MS_RDONLY)
2087 for (i = 0; i < num_pages; i++) {
2088 struct page *p = extent_buffer_page(eb, i);
2089 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2090 start, p, mirror_num);
2093 start += PAGE_CACHE_SIZE;
2100 * each time an IO finishes, we do a fast check in the IO failure tree
2101 * to see if we need to process or clean up an io_failure_record
2103 static int clean_io_failure(u64 start, struct page *page)
2106 u64 private_failure;
2107 struct io_failure_record *failrec;
2108 struct inode *inode = page->mapping->host;
2109 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2110 struct extent_state *state;
2116 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2117 (u64)-1, 1, EXTENT_DIRTY, 0);
2121 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2126 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2127 BUG_ON(!failrec->this_mirror);
2129 if (failrec->in_validation) {
2130 /* there was no real error, just free the record */
2131 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2136 if (fs_info->sb->s_flags & MS_RDONLY)
2139 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2140 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2143 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2145 if (state && state->start <= failrec->start &&
2146 state->end >= failrec->start + failrec->len - 1) {
2147 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2149 if (num_copies > 1) {
2150 ret = repair_io_failure(fs_info, start, failrec->len,
2151 failrec->logical, page,
2152 failrec->failed_mirror);
2160 ret = free_io_failure(inode, failrec, did_repair);
2166 * this is a generic handler for readpage errors (default
2167 * readpage_io_failed_hook). if other copies exist, read those and write back
2168 * good data to the failed position. does not investigate in remapping the
2169 * failed extent elsewhere, hoping the device will be smart enough to do this as
2173 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2174 struct page *page, u64 start, u64 end,
2177 struct io_failure_record *failrec = NULL;
2179 struct extent_map *em;
2180 struct inode *inode = page->mapping->host;
2181 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2182 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2183 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2185 struct btrfs_io_bio *btrfs_failed_bio;
2186 struct btrfs_io_bio *btrfs_bio;
2192 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2194 ret = get_state_private(failure_tree, start, &private);
2196 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2199 failrec->start = start;
2200 failrec->len = end - start + 1;
2201 failrec->this_mirror = 0;
2202 failrec->bio_flags = 0;
2203 failrec->in_validation = 0;
2205 read_lock(&em_tree->lock);
2206 em = lookup_extent_mapping(em_tree, start, failrec->len);
2208 read_unlock(&em_tree->lock);
2213 if (em->start > start || em->start + em->len <= start) {
2214 free_extent_map(em);
2217 read_unlock(&em_tree->lock);
2223 logical = start - em->start;
2224 logical = em->block_start + logical;
2225 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2226 logical = em->block_start;
2227 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2228 extent_set_compress_type(&failrec->bio_flags,
2231 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2232 "len=%llu\n", logical, start, failrec->len);
2233 failrec->logical = logical;
2234 free_extent_map(em);
2236 /* set the bits in the private failure tree */
2237 ret = set_extent_bits(failure_tree, start, end,
2238 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2240 ret = set_state_private(failure_tree, start,
2241 (u64)(unsigned long)failrec);
2242 /* set the bits in the inode's tree */
2244 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2251 failrec = (struct io_failure_record *)(unsigned long)private;
2252 pr_debug("bio_readpage_error: (found) logical=%llu, "
2253 "start=%llu, len=%llu, validation=%d\n",
2254 failrec->logical, failrec->start, failrec->len,
2255 failrec->in_validation);
2257 * when data can be on disk more than twice, add to failrec here
2258 * (e.g. with a list for failed_mirror) to make
2259 * clean_io_failure() clean all those errors at once.
2262 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2263 failrec->logical, failrec->len);
2264 if (num_copies == 1) {
2266 * we only have a single copy of the data, so don't bother with
2267 * all the retry and error correction code that follows. no
2268 * matter what the error is, it is very likely to persist.
2270 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2271 num_copies, failrec->this_mirror, failed_mirror);
2272 free_io_failure(inode, failrec, 0);
2277 * there are two premises:
2278 * a) deliver good data to the caller
2279 * b) correct the bad sectors on disk
2281 if (failed_bio->bi_vcnt > 1) {
2283 * to fulfill b), we need to know the exact failing sectors, as
2284 * we don't want to rewrite any more than the failed ones. thus,
2285 * we need separate read requests for the failed bio
2287 * if the following BUG_ON triggers, our validation request got
2288 * merged. we need separate requests for our algorithm to work.
2290 BUG_ON(failrec->in_validation);
2291 failrec->in_validation = 1;
2292 failrec->this_mirror = failed_mirror;
2293 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2296 * we're ready to fulfill a) and b) alongside. get a good copy
2297 * of the failed sector and if we succeed, we have setup
2298 * everything for repair_io_failure to do the rest for us.
2300 if (failrec->in_validation) {
2301 BUG_ON(failrec->this_mirror != failed_mirror);
2302 failrec->in_validation = 0;
2303 failrec->this_mirror = 0;
2305 failrec->failed_mirror = failed_mirror;
2306 failrec->this_mirror++;
2307 if (failrec->this_mirror == failed_mirror)
2308 failrec->this_mirror++;
2309 read_mode = READ_SYNC;
2312 if (failrec->this_mirror > num_copies) {
2313 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2314 num_copies, failrec->this_mirror, failed_mirror);
2315 free_io_failure(inode, failrec, 0);
2319 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2321 free_io_failure(inode, failrec, 0);
2324 bio->bi_end_io = failed_bio->bi_end_io;
2325 bio->bi_iter.bi_sector = failrec->logical >> 9;
2326 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2327 bio->bi_iter.bi_size = 0;
2329 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2330 if (btrfs_failed_bio->csum) {
2331 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2332 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2334 btrfs_bio = btrfs_io_bio(bio);
2335 btrfs_bio->csum = btrfs_bio->csum_inline;
2336 phy_offset >>= inode->i_sb->s_blocksize_bits;
2337 phy_offset *= csum_size;
2338 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset,
2342 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2344 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2345 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2346 failrec->this_mirror, num_copies, failrec->in_validation);
2348 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2349 failrec->this_mirror,
2350 failrec->bio_flags, 0);
2354 /* lots and lots of room for performance fixes in the end_bio funcs */
2356 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2358 int uptodate = (err == 0);
2359 struct extent_io_tree *tree;
2362 tree = &BTRFS_I(page->mapping->host)->io_tree;
2364 if (tree->ops && tree->ops->writepage_end_io_hook) {
2365 ret = tree->ops->writepage_end_io_hook(page, start,
2366 end, NULL, uptodate);
2372 ClearPageUptodate(page);
2374 ret = ret < 0 ? ret : -EIO;
2375 mapping_set_error(page->mapping, ret);
2381 * after a writepage IO is done, we need to:
2382 * clear the uptodate bits on error
2383 * clear the writeback bits in the extent tree for this IO
2384 * end_page_writeback if the page has no more pending IO
2386 * Scheduling is not allowed, so the extent state tree is expected
2387 * to have one and only one object corresponding to this IO.
2389 static void end_bio_extent_writepage(struct bio *bio, int err)
2391 struct bio_vec *bvec;
2396 bio_for_each_segment_all(bvec, bio, i) {
2397 struct page *page = bvec->bv_page;
2399 /* We always issue full-page reads, but if some block
2400 * in a page fails to read, blk_update_request() will
2401 * advance bv_offset and adjust bv_len to compensate.
2402 * Print a warning for nonzero offsets, and an error
2403 * if they don't add up to a full page. */
2404 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2405 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2406 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2407 "partial page write in btrfs with offset %u and length %u",
2408 bvec->bv_offset, bvec->bv_len);
2410 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2411 "incomplete page write in btrfs with offset %u and "
2413 bvec->bv_offset, bvec->bv_len);
2416 start = page_offset(page);
2417 end = start + bvec->bv_offset + bvec->bv_len - 1;
2419 if (end_extent_writepage(page, err, start, end))
2422 end_page_writeback(page);
2429 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2432 struct extent_state *cached = NULL;
2433 u64 end = start + len - 1;
2435 if (uptodate && tree->track_uptodate)
2436 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2437 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2441 * after a readpage IO is done, we need to:
2442 * clear the uptodate bits on error
2443 * set the uptodate bits if things worked
2444 * set the page up to date if all extents in the tree are uptodate
2445 * clear the lock bit in the extent tree
2446 * unlock the page if there are no other extents locked for it
2448 * Scheduling is not allowed, so the extent state tree is expected
2449 * to have one and only one object corresponding to this IO.
2451 static void end_bio_extent_readpage(struct bio *bio, int err)
2453 struct bio_vec *bvec;
2454 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2455 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2456 struct extent_io_tree *tree;
2461 u64 extent_start = 0;
2470 bio_for_each_segment_all(bvec, bio, i) {
2471 struct page *page = bvec->bv_page;
2472 struct inode *inode = page->mapping->host;
2474 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2475 "mirror=%lu\n", (u64)bio->bi_iter.bi_sector, err,
2476 io_bio->mirror_num);
2477 tree = &BTRFS_I(inode)->io_tree;
2479 /* We always issue full-page reads, but if some block
2480 * in a page fails to read, blk_update_request() will
2481 * advance bv_offset and adjust bv_len to compensate.
2482 * Print a warning for nonzero offsets, and an error
2483 * if they don't add up to a full page. */
2484 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2485 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2486 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2487 "partial page read in btrfs with offset %u and length %u",
2488 bvec->bv_offset, bvec->bv_len);
2490 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2491 "incomplete page read in btrfs with offset %u and "
2493 bvec->bv_offset, bvec->bv_len);
2496 start = page_offset(page);
2497 end = start + bvec->bv_offset + bvec->bv_len - 1;
2500 mirror = io_bio->mirror_num;
2501 if (likely(uptodate && tree->ops &&
2502 tree->ops->readpage_end_io_hook)) {
2503 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2509 clean_io_failure(start, page);
2512 if (likely(uptodate))
2515 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2516 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2518 test_bit(BIO_UPTODATE, &bio->bi_flags))
2522 * The generic bio_readpage_error handles errors the
2523 * following way: If possible, new read requests are
2524 * created and submitted and will end up in
2525 * end_bio_extent_readpage as well (if we're lucky, not
2526 * in the !uptodate case). In that case it returns 0 and
2527 * we just go on with the next page in our bio. If it
2528 * can't handle the error it will return -EIO and we
2529 * remain responsible for that page.
2531 ret = bio_readpage_error(bio, offset, page, start, end,
2535 test_bit(BIO_UPTODATE, &bio->bi_flags);
2543 if (likely(uptodate)) {
2544 loff_t i_size = i_size_read(inode);
2545 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2548 /* Zero out the end if this page straddles i_size */
2549 off = i_size & (PAGE_CACHE_SIZE-1);
2550 if (page->index == end_index && off)
2551 zero_user_segment(page, off, PAGE_CACHE_SIZE);
2552 SetPageUptodate(page);
2554 ClearPageUptodate(page);
2560 if (unlikely(!uptodate)) {
2562 endio_readpage_release_extent(tree,
2568 endio_readpage_release_extent(tree, start,
2569 end - start + 1, 0);
2570 } else if (!extent_len) {
2571 extent_start = start;
2572 extent_len = end + 1 - start;
2573 } else if (extent_start + extent_len == start) {
2574 extent_len += end + 1 - start;
2576 endio_readpage_release_extent(tree, extent_start,
2577 extent_len, uptodate);
2578 extent_start = start;
2579 extent_len = end + 1 - start;
2584 endio_readpage_release_extent(tree, extent_start, extent_len,
2587 io_bio->end_io(io_bio, err);
2592 * this allocates from the btrfs_bioset. We're returning a bio right now
2593 * but you can call btrfs_io_bio for the appropriate container_of magic
2596 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2599 struct btrfs_io_bio *btrfs_bio;
2602 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2604 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2605 while (!bio && (nr_vecs /= 2)) {
2606 bio = bio_alloc_bioset(gfp_flags,
2607 nr_vecs, btrfs_bioset);
2612 bio->bi_bdev = bdev;
2613 bio->bi_iter.bi_sector = first_sector;
2614 btrfs_bio = btrfs_io_bio(bio);
2615 btrfs_bio->csum = NULL;
2616 btrfs_bio->csum_allocated = NULL;
2617 btrfs_bio->end_io = NULL;
2622 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2624 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2628 /* this also allocates from the btrfs_bioset */
2629 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2631 struct btrfs_io_bio *btrfs_bio;
2634 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2636 btrfs_bio = btrfs_io_bio(bio);
2637 btrfs_bio->csum = NULL;
2638 btrfs_bio->csum_allocated = NULL;
2639 btrfs_bio->end_io = NULL;
2645 static int __must_check submit_one_bio(int rw, struct bio *bio,
2646 int mirror_num, unsigned long bio_flags)
2649 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2650 struct page *page = bvec->bv_page;
2651 struct extent_io_tree *tree = bio->bi_private;
2654 start = page_offset(page) + bvec->bv_offset;
2656 bio->bi_private = NULL;
2660 if (tree->ops && tree->ops->submit_bio_hook)
2661 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2662 mirror_num, bio_flags, start);
2664 btrfsic_submit_bio(rw, bio);
2666 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2672 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2673 unsigned long offset, size_t size, struct bio *bio,
2674 unsigned long bio_flags)
2677 if (tree->ops && tree->ops->merge_bio_hook)
2678 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2685 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2686 struct page *page, sector_t sector,
2687 size_t size, unsigned long offset,
2688 struct block_device *bdev,
2689 struct bio **bio_ret,
2690 unsigned long max_pages,
2691 bio_end_io_t end_io_func,
2693 unsigned long prev_bio_flags,
2694 unsigned long bio_flags)
2700 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2701 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2702 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2704 if (bio_ret && *bio_ret) {
2707 contig = bio->bi_iter.bi_sector == sector;
2709 contig = bio_end_sector(bio) == sector;
2711 if (prev_bio_flags != bio_flags || !contig ||
2712 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2713 bio_add_page(bio, page, page_size, offset) < page_size) {
2714 ret = submit_one_bio(rw, bio, mirror_num,
2723 if (this_compressed)
2726 nr = bio_get_nr_vecs(bdev);
2728 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2732 bio_add_page(bio, page, page_size, offset);
2733 bio->bi_end_io = end_io_func;
2734 bio->bi_private = tree;
2739 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2744 static void attach_extent_buffer_page(struct extent_buffer *eb,
2747 if (!PagePrivate(page)) {
2748 SetPagePrivate(page);
2749 page_cache_get(page);
2750 set_page_private(page, (unsigned long)eb);
2752 WARN_ON(page->private != (unsigned long)eb);
2756 void set_page_extent_mapped(struct page *page)
2758 if (!PagePrivate(page)) {
2759 SetPagePrivate(page);
2760 page_cache_get(page);
2761 set_page_private(page, EXTENT_PAGE_PRIVATE);
2765 static struct extent_map *
2766 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2767 u64 start, u64 len, get_extent_t *get_extent,
2768 struct extent_map **em_cached)
2770 struct extent_map *em;
2772 if (em_cached && *em_cached) {
2774 if (extent_map_in_tree(em) && start >= em->start &&
2775 start < extent_map_end(em)) {
2776 atomic_inc(&em->refs);
2780 free_extent_map(em);
2784 em = get_extent(inode, page, pg_offset, start, len, 0);
2785 if (em_cached && !IS_ERR_OR_NULL(em)) {
2787 atomic_inc(&em->refs);
2793 * basic readpage implementation. Locked extent state structs are inserted
2794 * into the tree that are removed when the IO is done (by the end_io
2796 * XXX JDM: This needs looking at to ensure proper page locking
2798 static int __do_readpage(struct extent_io_tree *tree,
2800 get_extent_t *get_extent,
2801 struct extent_map **em_cached,
2802 struct bio **bio, int mirror_num,
2803 unsigned long *bio_flags, int rw)
2805 struct inode *inode = page->mapping->host;
2806 u64 start = page_offset(page);
2807 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2811 u64 last_byte = i_size_read(inode);
2815 struct extent_map *em;
2816 struct block_device *bdev;
2819 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2820 size_t pg_offset = 0;
2822 size_t disk_io_size;
2823 size_t blocksize = inode->i_sb->s_blocksize;
2824 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2826 set_page_extent_mapped(page);
2829 if (!PageUptodate(page)) {
2830 if (cleancache_get_page(page) == 0) {
2831 BUG_ON(blocksize != PAGE_SIZE);
2832 unlock_extent(tree, start, end);
2837 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2839 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2842 iosize = PAGE_CACHE_SIZE - zero_offset;
2843 userpage = kmap_atomic(page);
2844 memset(userpage + zero_offset, 0, iosize);
2845 flush_dcache_page(page);
2846 kunmap_atomic(userpage);
2849 while (cur <= end) {
2850 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2852 if (cur >= last_byte) {
2854 struct extent_state *cached = NULL;
2856 iosize = PAGE_CACHE_SIZE - pg_offset;
2857 userpage = kmap_atomic(page);
2858 memset(userpage + pg_offset, 0, iosize);
2859 flush_dcache_page(page);
2860 kunmap_atomic(userpage);
2861 set_extent_uptodate(tree, cur, cur + iosize - 1,
2864 unlock_extent_cached(tree, cur,
2869 em = __get_extent_map(inode, page, pg_offset, cur,
2870 end - cur + 1, get_extent, em_cached);
2871 if (IS_ERR_OR_NULL(em)) {
2874 unlock_extent(tree, cur, end);
2877 extent_offset = cur - em->start;
2878 BUG_ON(extent_map_end(em) <= cur);
2881 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2882 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2883 extent_set_compress_type(&this_bio_flag,
2887 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2888 cur_end = min(extent_map_end(em) - 1, end);
2889 iosize = ALIGN(iosize, blocksize);
2890 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2891 disk_io_size = em->block_len;
2892 sector = em->block_start >> 9;
2894 sector = (em->block_start + extent_offset) >> 9;
2895 disk_io_size = iosize;
2898 block_start = em->block_start;
2899 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2900 block_start = EXTENT_MAP_HOLE;
2901 free_extent_map(em);
2904 /* we've found a hole, just zero and go on */
2905 if (block_start == EXTENT_MAP_HOLE) {
2907 struct extent_state *cached = NULL;
2909 userpage = kmap_atomic(page);
2910 memset(userpage + pg_offset, 0, iosize);
2911 flush_dcache_page(page);
2912 kunmap_atomic(userpage);
2914 set_extent_uptodate(tree, cur, cur + iosize - 1,
2916 unlock_extent_cached(tree, cur, cur + iosize - 1,
2919 pg_offset += iosize;
2922 /* the get_extent function already copied into the page */
2923 if (test_range_bit(tree, cur, cur_end,
2924 EXTENT_UPTODATE, 1, NULL)) {
2925 check_page_uptodate(tree, page);
2927 unlock_extent(tree, cur, cur + iosize - 1);
2929 pg_offset += iosize;
2932 /* we have an inline extent but it didn't get marked up
2933 * to date. Error out
2935 if (block_start == EXTENT_MAP_INLINE) {
2938 unlock_extent(tree, cur, cur + iosize - 1);
2940 pg_offset += iosize;
2945 ret = submit_extent_page(rw, tree, page,
2946 sector, disk_io_size, pg_offset,
2948 end_bio_extent_readpage, mirror_num,
2953 *bio_flags = this_bio_flag;
2957 unlock_extent(tree, cur, cur + iosize - 1);
2960 pg_offset += iosize;
2964 if (!PageError(page))
2965 SetPageUptodate(page);
2971 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
2972 struct page *pages[], int nr_pages,
2974 get_extent_t *get_extent,
2975 struct extent_map **em_cached,
2976 struct bio **bio, int mirror_num,
2977 unsigned long *bio_flags, int rw)
2979 struct inode *inode;
2980 struct btrfs_ordered_extent *ordered;
2983 inode = pages[0]->mapping->host;
2985 lock_extent(tree, start, end);
2986 ordered = btrfs_lookup_ordered_range(inode, start,
2990 unlock_extent(tree, start, end);
2991 btrfs_start_ordered_extent(inode, ordered, 1);
2992 btrfs_put_ordered_extent(ordered);
2995 for (index = 0; index < nr_pages; index++) {
2996 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
2997 mirror_num, bio_flags, rw);
2998 page_cache_release(pages[index]);
3002 static void __extent_readpages(struct extent_io_tree *tree,
3003 struct page *pages[],
3004 int nr_pages, get_extent_t *get_extent,
3005 struct extent_map **em_cached,
3006 struct bio **bio, int mirror_num,
3007 unsigned long *bio_flags, int rw)
3013 int first_index = 0;
3015 for (index = 0; index < nr_pages; index++) {
3016 page_start = page_offset(pages[index]);
3019 end = start + PAGE_CACHE_SIZE - 1;
3020 first_index = index;
3021 } else if (end + 1 == page_start) {
3022 end += PAGE_CACHE_SIZE;
3024 __do_contiguous_readpages(tree, &pages[first_index],
3025 index - first_index, start,
3026 end, get_extent, em_cached,
3027 bio, mirror_num, bio_flags,
3030 end = start + PAGE_CACHE_SIZE - 1;
3031 first_index = index;
3036 __do_contiguous_readpages(tree, &pages[first_index],
3037 index - first_index, start,
3038 end, get_extent, em_cached, bio,
3039 mirror_num, bio_flags, rw);
3042 static int __extent_read_full_page(struct extent_io_tree *tree,
3044 get_extent_t *get_extent,
3045 struct bio **bio, int mirror_num,
3046 unsigned long *bio_flags, int rw)
3048 struct inode *inode = page->mapping->host;
3049 struct btrfs_ordered_extent *ordered;
3050 u64 start = page_offset(page);
3051 u64 end = start + PAGE_CACHE_SIZE - 1;
3055 lock_extent(tree, start, end);
3056 ordered = btrfs_lookup_ordered_extent(inode, start);
3059 unlock_extent(tree, start, end);
3060 btrfs_start_ordered_extent(inode, ordered, 1);
3061 btrfs_put_ordered_extent(ordered);
3064 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3069 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3070 get_extent_t *get_extent, int mirror_num)
3072 struct bio *bio = NULL;
3073 unsigned long bio_flags = 0;
3076 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3079 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3083 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3084 get_extent_t *get_extent, int mirror_num)
3086 struct bio *bio = NULL;
3087 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3090 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3093 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3097 static noinline void update_nr_written(struct page *page,
3098 struct writeback_control *wbc,
3099 unsigned long nr_written)
3101 wbc->nr_to_write -= nr_written;
3102 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3103 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3104 page->mapping->writeback_index = page->index + nr_written;
3108 * helper for __extent_writepage, doing all of the delayed allocation setup.
3110 * This returns 1 if our fill_delalloc function did all the work required
3111 * to write the page (copy into inline extent). In this case the IO has
3112 * been started and the page is already unlocked.
3114 * This returns 0 if all went well (page still locked)
3115 * This returns < 0 if there were errors (page still locked)
3117 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3118 struct page *page, struct writeback_control *wbc,
3119 struct extent_page_data *epd,
3121 unsigned long *nr_written)
3123 struct extent_io_tree *tree = epd->tree;
3124 u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1;
3126 u64 delalloc_to_write = 0;
3127 u64 delalloc_end = 0;
3129 int page_started = 0;
3131 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3134 while (delalloc_end < page_end) {
3135 nr_delalloc = find_lock_delalloc_range(inode, tree,
3140 if (nr_delalloc == 0) {
3141 delalloc_start = delalloc_end + 1;
3144 ret = tree->ops->fill_delalloc(inode, page,
3149 /* File system has been set read-only */
3152 /* fill_delalloc should be return < 0 for error
3153 * but just in case, we use > 0 here meaning the
3154 * IO is started, so we don't want to return > 0
3155 * unless things are going well.
3157 ret = ret < 0 ? ret : -EIO;
3161 * delalloc_end is already one less than the total
3162 * length, so we don't subtract one from
3165 delalloc_to_write += (delalloc_end - delalloc_start +
3168 delalloc_start = delalloc_end + 1;
3170 if (wbc->nr_to_write < delalloc_to_write) {
3173 if (delalloc_to_write < thresh * 2)
3174 thresh = delalloc_to_write;
3175 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3179 /* did the fill delalloc function already unlock and start
3184 * we've unlocked the page, so we can't update
3185 * the mapping's writeback index, just update
3188 wbc->nr_to_write -= *nr_written;
3199 * helper for __extent_writepage. This calls the writepage start hooks,
3200 * and does the loop to map the page into extents and bios.
3202 * We return 1 if the IO is started and the page is unlocked,
3203 * 0 if all went well (page still locked)
3204 * < 0 if there were errors (page still locked)
3206 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3208 struct writeback_control *wbc,
3209 struct extent_page_data *epd,
3211 unsigned long nr_written,
3212 int write_flags, int *nr_ret)
3214 struct extent_io_tree *tree = epd->tree;
3215 u64 start = page_offset(page);
3216 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3223 struct extent_state *cached_state = NULL;
3224 struct extent_map *em;
3225 struct block_device *bdev;
3226 size_t pg_offset = 0;
3232 if (tree->ops && tree->ops->writepage_start_hook) {
3233 ret = tree->ops->writepage_start_hook(page, start,
3236 /* Fixup worker will requeue */
3238 wbc->pages_skipped++;
3240 redirty_page_for_writepage(wbc, page);
3242 update_nr_written(page, wbc, nr_written);
3250 * we don't want to touch the inode after unlocking the page,
3251 * so we update the mapping writeback index now
3253 update_nr_written(page, wbc, nr_written + 1);
3256 if (i_size <= start) {
3257 if (tree->ops && tree->ops->writepage_end_io_hook)
3258 tree->ops->writepage_end_io_hook(page, start,
3263 blocksize = inode->i_sb->s_blocksize;
3265 while (cur <= end) {
3267 if (cur >= i_size) {
3268 if (tree->ops && tree->ops->writepage_end_io_hook)
3269 tree->ops->writepage_end_io_hook(page, cur,
3273 em = epd->get_extent(inode, page, pg_offset, cur,
3275 if (IS_ERR_OR_NULL(em)) {
3277 ret = PTR_ERR_OR_ZERO(em);
3281 extent_offset = cur - em->start;
3282 em_end = extent_map_end(em);
3283 BUG_ON(em_end <= cur);
3285 iosize = min(em_end - cur, end - cur + 1);
3286 iosize = ALIGN(iosize, blocksize);
3287 sector = (em->block_start + extent_offset) >> 9;
3289 block_start = em->block_start;
3290 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3291 free_extent_map(em);
3295 * compressed and inline extents are written through other
3298 if (compressed || block_start == EXTENT_MAP_HOLE ||
3299 block_start == EXTENT_MAP_INLINE) {
3301 * end_io notification does not happen here for
3302 * compressed extents
3304 if (!compressed && tree->ops &&
3305 tree->ops->writepage_end_io_hook)
3306 tree->ops->writepage_end_io_hook(page, cur,
3309 else if (compressed) {
3310 /* we don't want to end_page_writeback on
3311 * a compressed extent. this happens
3318 pg_offset += iosize;
3322 if (tree->ops && tree->ops->writepage_io_hook) {
3323 ret = tree->ops->writepage_io_hook(page, cur,
3331 unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1;
3333 set_range_writeback(tree, cur, cur + iosize - 1);
3334 if (!PageWriteback(page)) {
3335 btrfs_err(BTRFS_I(inode)->root->fs_info,
3336 "page %lu not writeback, cur %llu end %llu",
3337 page->index, cur, end);
3340 ret = submit_extent_page(write_flags, tree, page,
3341 sector, iosize, pg_offset,
3342 bdev, &epd->bio, max_nr,
3343 end_bio_extent_writepage,
3349 pg_offset += iosize;
3357 /* drop our reference on any cached states */
3358 free_extent_state(cached_state);
3363 * the writepage semantics are similar to regular writepage. extent
3364 * records are inserted to lock ranges in the tree, and as dirty areas
3365 * are found, they are marked writeback. Then the lock bits are removed
3366 * and the end_io handler clears the writeback ranges
3368 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3371 struct inode *inode = page->mapping->host;
3372 struct extent_page_data *epd = data;
3373 u64 start = page_offset(page);
3374 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3377 size_t pg_offset = 0;
3378 loff_t i_size = i_size_read(inode);
3379 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3381 unsigned long nr_written = 0;
3383 if (wbc->sync_mode == WB_SYNC_ALL)
3384 write_flags = WRITE_SYNC;
3386 write_flags = WRITE;
3388 trace___extent_writepage(page, inode, wbc);
3390 WARN_ON(!PageLocked(page));
3392 ClearPageError(page);
3394 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3395 if (page->index > end_index ||
3396 (page->index == end_index && !pg_offset)) {
3397 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3402 if (page->index == end_index) {
3405 userpage = kmap_atomic(page);
3406 memset(userpage + pg_offset, 0,
3407 PAGE_CACHE_SIZE - pg_offset);
3408 kunmap_atomic(userpage);
3409 flush_dcache_page(page);
3414 set_page_extent_mapped(page);
3416 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3422 ret = __extent_writepage_io(inode, page, wbc, epd,
3423 i_size, nr_written, write_flags, &nr);
3429 /* make sure the mapping tag for page dirty gets cleared */
3430 set_page_writeback(page);
3431 end_page_writeback(page);
3433 if (PageError(page)) {
3434 ret = ret < 0 ? ret : -EIO;
3435 end_extent_writepage(page, ret, start, page_end);
3444 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3446 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3447 TASK_UNINTERRUPTIBLE);
3450 static noinline_for_stack int
3451 lock_extent_buffer_for_io(struct extent_buffer *eb,
3452 struct btrfs_fs_info *fs_info,
3453 struct extent_page_data *epd)
3455 unsigned long i, num_pages;
3459 if (!btrfs_try_tree_write_lock(eb)) {
3461 flush_write_bio(epd);
3462 btrfs_tree_lock(eb);
3465 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3466 btrfs_tree_unlock(eb);
3470 flush_write_bio(epd);
3474 wait_on_extent_buffer_writeback(eb);
3475 btrfs_tree_lock(eb);
3476 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3478 btrfs_tree_unlock(eb);
3483 * We need to do this to prevent races in people who check if the eb is
3484 * under IO since we can end up having no IO bits set for a short period
3487 spin_lock(&eb->refs_lock);
3488 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3489 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3490 spin_unlock(&eb->refs_lock);
3491 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3492 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3494 fs_info->dirty_metadata_batch);
3497 spin_unlock(&eb->refs_lock);
3500 btrfs_tree_unlock(eb);
3505 num_pages = num_extent_pages(eb->start, eb->len);
3506 for (i = 0; i < num_pages; i++) {
3507 struct page *p = extent_buffer_page(eb, i);
3509 if (!trylock_page(p)) {
3511 flush_write_bio(epd);
3521 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3523 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3524 smp_mb__after_atomic();
3525 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3528 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3530 struct bio_vec *bvec;
3531 struct extent_buffer *eb;
3534 bio_for_each_segment_all(bvec, bio, i) {
3535 struct page *page = bvec->bv_page;
3537 eb = (struct extent_buffer *)page->private;
3539 done = atomic_dec_and_test(&eb->io_pages);
3541 if (err || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3542 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3543 ClearPageUptodate(page);
3547 end_page_writeback(page);
3552 end_extent_buffer_writeback(eb);
3558 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3559 struct btrfs_fs_info *fs_info,
3560 struct writeback_control *wbc,
3561 struct extent_page_data *epd)
3563 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3564 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3565 u64 offset = eb->start;
3566 unsigned long i, num_pages;
3567 unsigned long bio_flags = 0;
3568 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3571 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3572 num_pages = num_extent_pages(eb->start, eb->len);
3573 atomic_set(&eb->io_pages, num_pages);
3574 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3575 bio_flags = EXTENT_BIO_TREE_LOG;
3577 for (i = 0; i < num_pages; i++) {
3578 struct page *p = extent_buffer_page(eb, i);
3580 clear_page_dirty_for_io(p);
3581 set_page_writeback(p);
3582 ret = submit_extent_page(rw, tree, p, offset >> 9,
3583 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3584 -1, end_bio_extent_buffer_writepage,
3585 0, epd->bio_flags, bio_flags);
3586 epd->bio_flags = bio_flags;
3588 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3590 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3591 end_extent_buffer_writeback(eb);
3595 offset += PAGE_CACHE_SIZE;
3596 update_nr_written(p, wbc, 1);
3600 if (unlikely(ret)) {
3601 for (; i < num_pages; i++) {
3602 struct page *p = extent_buffer_page(eb, i);
3610 int btree_write_cache_pages(struct address_space *mapping,
3611 struct writeback_control *wbc)
3613 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3614 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3615 struct extent_buffer *eb, *prev_eb = NULL;
3616 struct extent_page_data epd = {
3620 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3625 int nr_to_write_done = 0;
3626 struct pagevec pvec;
3629 pgoff_t end; /* Inclusive */
3633 pagevec_init(&pvec, 0);
3634 if (wbc->range_cyclic) {
3635 index = mapping->writeback_index; /* Start from prev offset */
3638 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3639 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3642 if (wbc->sync_mode == WB_SYNC_ALL)
3643 tag = PAGECACHE_TAG_TOWRITE;
3645 tag = PAGECACHE_TAG_DIRTY;
3647 if (wbc->sync_mode == WB_SYNC_ALL)
3648 tag_pages_for_writeback(mapping, index, end);
3649 while (!done && !nr_to_write_done && (index <= end) &&
3650 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3651 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3655 for (i = 0; i < nr_pages; i++) {
3656 struct page *page = pvec.pages[i];
3658 if (!PagePrivate(page))
3661 if (!wbc->range_cyclic && page->index > end) {
3666 spin_lock(&mapping->private_lock);
3667 if (!PagePrivate(page)) {
3668 spin_unlock(&mapping->private_lock);
3672 eb = (struct extent_buffer *)page->private;
3675 * Shouldn't happen and normally this would be a BUG_ON
3676 * but no sense in crashing the users box for something
3677 * we can survive anyway.
3680 spin_unlock(&mapping->private_lock);
3684 if (eb == prev_eb) {
3685 spin_unlock(&mapping->private_lock);
3689 ret = atomic_inc_not_zero(&eb->refs);
3690 spin_unlock(&mapping->private_lock);
3695 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3697 free_extent_buffer(eb);
3701 ret = write_one_eb(eb, fs_info, wbc, &epd);
3704 free_extent_buffer(eb);
3707 free_extent_buffer(eb);
3710 * the filesystem may choose to bump up nr_to_write.
3711 * We have to make sure to honor the new nr_to_write
3714 nr_to_write_done = wbc->nr_to_write <= 0;
3716 pagevec_release(&pvec);
3719 if (!scanned && !done) {
3721 * We hit the last page and there is more work to be done: wrap
3722 * back to the start of the file
3728 flush_write_bio(&epd);
3733 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3734 * @mapping: address space structure to write
3735 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3736 * @writepage: function called for each page
3737 * @data: data passed to writepage function
3739 * If a page is already under I/O, write_cache_pages() skips it, even
3740 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3741 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3742 * and msync() need to guarantee that all the data which was dirty at the time
3743 * the call was made get new I/O started against them. If wbc->sync_mode is
3744 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3745 * existing IO to complete.
3747 static int extent_write_cache_pages(struct extent_io_tree *tree,
3748 struct address_space *mapping,
3749 struct writeback_control *wbc,
3750 writepage_t writepage, void *data,
3751 void (*flush_fn)(void *))
3753 struct inode *inode = mapping->host;
3757 int nr_to_write_done = 0;
3758 struct pagevec pvec;
3761 pgoff_t end; /* Inclusive */
3766 * We have to hold onto the inode so that ordered extents can do their
3767 * work when the IO finishes. The alternative to this is failing to add
3768 * an ordered extent if the igrab() fails there and that is a huge pain
3769 * to deal with, so instead just hold onto the inode throughout the
3770 * writepages operation. If it fails here we are freeing up the inode
3771 * anyway and we'd rather not waste our time writing out stuff that is
3772 * going to be truncated anyway.
3777 pagevec_init(&pvec, 0);
3778 if (wbc->range_cyclic) {
3779 index = mapping->writeback_index; /* Start from prev offset */
3782 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3783 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3786 if (wbc->sync_mode == WB_SYNC_ALL)
3787 tag = PAGECACHE_TAG_TOWRITE;
3789 tag = PAGECACHE_TAG_DIRTY;
3791 if (wbc->sync_mode == WB_SYNC_ALL)
3792 tag_pages_for_writeback(mapping, index, end);
3793 while (!done && !nr_to_write_done && (index <= end) &&
3794 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3795 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3799 for (i = 0; i < nr_pages; i++) {
3800 struct page *page = pvec.pages[i];
3803 * At this point we hold neither mapping->tree_lock nor
3804 * lock on the page itself: the page may be truncated or
3805 * invalidated (changing page->mapping to NULL), or even
3806 * swizzled back from swapper_space to tmpfs file
3809 if (!trylock_page(page)) {
3814 if (unlikely(page->mapping != mapping)) {
3819 if (!wbc->range_cyclic && page->index > end) {
3825 if (wbc->sync_mode != WB_SYNC_NONE) {
3826 if (PageWriteback(page))
3828 wait_on_page_writeback(page);
3831 if (PageWriteback(page) ||
3832 !clear_page_dirty_for_io(page)) {
3837 ret = (*writepage)(page, wbc, data);
3839 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3843 if (!err && ret < 0)
3847 * the filesystem may choose to bump up nr_to_write.
3848 * We have to make sure to honor the new nr_to_write
3851 nr_to_write_done = wbc->nr_to_write <= 0;
3853 pagevec_release(&pvec);
3856 if (!scanned && !done && !err) {
3858 * We hit the last page and there is more work to be done: wrap
3859 * back to the start of the file
3865 btrfs_add_delayed_iput(inode);
3869 static void flush_epd_write_bio(struct extent_page_data *epd)
3878 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3879 BUG_ON(ret < 0); /* -ENOMEM */
3884 static noinline void flush_write_bio(void *data)
3886 struct extent_page_data *epd = data;
3887 flush_epd_write_bio(epd);
3890 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3891 get_extent_t *get_extent,
3892 struct writeback_control *wbc)
3895 struct extent_page_data epd = {
3898 .get_extent = get_extent,
3900 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3904 ret = __extent_writepage(page, wbc, &epd);
3906 flush_epd_write_bio(&epd);
3910 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3911 u64 start, u64 end, get_extent_t *get_extent,
3915 struct address_space *mapping = inode->i_mapping;
3917 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3920 struct extent_page_data epd = {
3923 .get_extent = get_extent,
3925 .sync_io = mode == WB_SYNC_ALL,
3928 struct writeback_control wbc_writepages = {
3930 .nr_to_write = nr_pages * 2,
3931 .range_start = start,
3932 .range_end = end + 1,
3935 while (start <= end) {
3936 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3937 if (clear_page_dirty_for_io(page))
3938 ret = __extent_writepage(page, &wbc_writepages, &epd);
3940 if (tree->ops && tree->ops->writepage_end_io_hook)
3941 tree->ops->writepage_end_io_hook(page, start,
3942 start + PAGE_CACHE_SIZE - 1,
3946 page_cache_release(page);
3947 start += PAGE_CACHE_SIZE;
3950 flush_epd_write_bio(&epd);
3954 int extent_writepages(struct extent_io_tree *tree,
3955 struct address_space *mapping,
3956 get_extent_t *get_extent,
3957 struct writeback_control *wbc)
3960 struct extent_page_data epd = {
3963 .get_extent = get_extent,
3965 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3969 ret = extent_write_cache_pages(tree, mapping, wbc,
3970 __extent_writepage, &epd,
3972 flush_epd_write_bio(&epd);
3976 int extent_readpages(struct extent_io_tree *tree,
3977 struct address_space *mapping,
3978 struct list_head *pages, unsigned nr_pages,
3979 get_extent_t get_extent)
3981 struct bio *bio = NULL;
3983 unsigned long bio_flags = 0;
3984 struct page *pagepool[16];
3986 struct extent_map *em_cached = NULL;
3989 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3990 page = list_entry(pages->prev, struct page, lru);
3992 prefetchw(&page->flags);
3993 list_del(&page->lru);
3994 if (add_to_page_cache_lru(page, mapping,
3995 page->index, GFP_NOFS)) {
3996 page_cache_release(page);
4000 pagepool[nr++] = page;
4001 if (nr < ARRAY_SIZE(pagepool))
4003 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4004 &bio, 0, &bio_flags, READ);
4008 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4009 &bio, 0, &bio_flags, READ);
4012 free_extent_map(em_cached);
4014 BUG_ON(!list_empty(pages));
4016 return submit_one_bio(READ, bio, 0, bio_flags);
4021 * basic invalidatepage code, this waits on any locked or writeback
4022 * ranges corresponding to the page, and then deletes any extent state
4023 * records from the tree
4025 int extent_invalidatepage(struct extent_io_tree *tree,
4026 struct page *page, unsigned long offset)
4028 struct extent_state *cached_state = NULL;
4029 u64 start = page_offset(page);
4030 u64 end = start + PAGE_CACHE_SIZE - 1;
4031 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4033 start += ALIGN(offset, blocksize);
4037 lock_extent_bits(tree, start, end, 0, &cached_state);
4038 wait_on_page_writeback(page);
4039 clear_extent_bit(tree, start, end,
4040 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4041 EXTENT_DO_ACCOUNTING,
4042 1, 1, &cached_state, GFP_NOFS);
4047 * a helper for releasepage, this tests for areas of the page that
4048 * are locked or under IO and drops the related state bits if it is safe
4051 static int try_release_extent_state(struct extent_map_tree *map,
4052 struct extent_io_tree *tree,
4053 struct page *page, gfp_t mask)
4055 u64 start = page_offset(page);
4056 u64 end = start + PAGE_CACHE_SIZE - 1;
4059 if (test_range_bit(tree, start, end,
4060 EXTENT_IOBITS, 0, NULL))
4063 if ((mask & GFP_NOFS) == GFP_NOFS)
4066 * at this point we can safely clear everything except the
4067 * locked bit and the nodatasum bit
4069 ret = clear_extent_bit(tree, start, end,
4070 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4073 /* if clear_extent_bit failed for enomem reasons,
4074 * we can't allow the release to continue.
4085 * a helper for releasepage. As long as there are no locked extents
4086 * in the range corresponding to the page, both state records and extent
4087 * map records are removed
4089 int try_release_extent_mapping(struct extent_map_tree *map,
4090 struct extent_io_tree *tree, struct page *page,
4093 struct extent_map *em;
4094 u64 start = page_offset(page);
4095 u64 end = start + PAGE_CACHE_SIZE - 1;
4097 if ((mask & __GFP_WAIT) &&
4098 page->mapping->host->i_size > 16 * 1024 * 1024) {
4100 while (start <= end) {
4101 len = end - start + 1;
4102 write_lock(&map->lock);
4103 em = lookup_extent_mapping(map, start, len);
4105 write_unlock(&map->lock);
4108 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4109 em->start != start) {
4110 write_unlock(&map->lock);
4111 free_extent_map(em);
4114 if (!test_range_bit(tree, em->start,
4115 extent_map_end(em) - 1,
4116 EXTENT_LOCKED | EXTENT_WRITEBACK,
4118 remove_extent_mapping(map, em);
4119 /* once for the rb tree */
4120 free_extent_map(em);
4122 start = extent_map_end(em);
4123 write_unlock(&map->lock);
4126 free_extent_map(em);
4129 return try_release_extent_state(map, tree, page, mask);
4133 * helper function for fiemap, which doesn't want to see any holes.
4134 * This maps until we find something past 'last'
4136 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4139 get_extent_t *get_extent)
4141 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4142 struct extent_map *em;
4149 len = last - offset;
4152 len = ALIGN(len, sectorsize);
4153 em = get_extent(inode, NULL, 0, offset, len, 0);
4154 if (IS_ERR_OR_NULL(em))
4157 /* if this isn't a hole return it */
4158 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4159 em->block_start != EXTENT_MAP_HOLE) {
4163 /* this is a hole, advance to the next extent */
4164 offset = extent_map_end(em);
4165 free_extent_map(em);
4172 static noinline int count_ext_ref(u64 inum, u64 offset, u64 root_id, void *ctx)
4174 unsigned long cnt = *((unsigned long *)ctx);
4177 *((unsigned long *)ctx) = cnt;
4179 /* Now we're sure that the extent is shared. */
4185 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4186 __u64 start, __u64 len, get_extent_t *get_extent)
4190 u64 max = start + len;
4194 u64 last_for_get_extent = 0;
4196 u64 isize = i_size_read(inode);
4197 struct btrfs_key found_key;
4198 struct extent_map *em = NULL;
4199 struct extent_state *cached_state = NULL;
4200 struct btrfs_path *path;
4209 path = btrfs_alloc_path();
4212 path->leave_spinning = 1;
4214 start = round_down(start, BTRFS_I(inode)->root->sectorsize);
4215 len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start;
4218 * lookup the last file extent. We're not using i_size here
4219 * because there might be preallocation past i_size
4221 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
4222 path, btrfs_ino(inode), -1, 0);
4224 btrfs_free_path(path);
4229 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4230 found_type = found_key.type;
4232 /* No extents, but there might be delalloc bits */
4233 if (found_key.objectid != btrfs_ino(inode) ||
4234 found_type != BTRFS_EXTENT_DATA_KEY) {
4235 /* have to trust i_size as the end */
4237 last_for_get_extent = isize;
4240 * remember the start of the last extent. There are a
4241 * bunch of different factors that go into the length of the
4242 * extent, so its much less complex to remember where it started
4244 last = found_key.offset;
4245 last_for_get_extent = last + 1;
4247 btrfs_release_path(path);
4250 * we might have some extents allocated but more delalloc past those
4251 * extents. so, we trust isize unless the start of the last extent is
4256 last_for_get_extent = isize;
4259 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4262 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4272 u64 offset_in_extent = 0;
4274 /* break if the extent we found is outside the range */
4275 if (em->start >= max || extent_map_end(em) < off)
4279 * get_extent may return an extent that starts before our
4280 * requested range. We have to make sure the ranges
4281 * we return to fiemap always move forward and don't
4282 * overlap, so adjust the offsets here
4284 em_start = max(em->start, off);
4287 * record the offset from the start of the extent
4288 * for adjusting the disk offset below. Only do this if the
4289 * extent isn't compressed since our in ram offset may be past
4290 * what we have actually allocated on disk.
4292 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4293 offset_in_extent = em_start - em->start;
4294 em_end = extent_map_end(em);
4295 em_len = em_end - em_start;
4300 * bump off for our next call to get_extent
4302 off = extent_map_end(em);
4306 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4308 flags |= FIEMAP_EXTENT_LAST;
4309 } else if (em->block_start == EXTENT_MAP_INLINE) {
4310 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4311 FIEMAP_EXTENT_NOT_ALIGNED);
4312 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4313 flags |= (FIEMAP_EXTENT_DELALLOC |
4314 FIEMAP_EXTENT_UNKNOWN);
4316 unsigned long ref_cnt = 0;
4318 disko = em->block_start + offset_in_extent;
4321 * As btrfs supports shared space, this information
4322 * can be exported to userspace tools via
4323 * flag FIEMAP_EXTENT_SHARED.
4325 ret = iterate_inodes_from_logical(
4327 BTRFS_I(inode)->root->fs_info,
4328 path, count_ext_ref, &ref_cnt);
4329 if (ret < 0 && ret != -ENOENT)
4333 flags |= FIEMAP_EXTENT_SHARED;
4335 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4336 flags |= FIEMAP_EXTENT_ENCODED;
4338 free_extent_map(em);
4340 if ((em_start >= last) || em_len == (u64)-1 ||
4341 (last == (u64)-1 && isize <= em_end)) {
4342 flags |= FIEMAP_EXTENT_LAST;
4346 /* now scan forward to see if this is really the last extent. */
4347 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4354 flags |= FIEMAP_EXTENT_LAST;
4357 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4363 free_extent_map(em);
4365 btrfs_free_path(path);
4366 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4367 &cached_state, GFP_NOFS);
4371 static void __free_extent_buffer(struct extent_buffer *eb)
4373 btrfs_leak_debug_del(&eb->leak_list);
4374 kmem_cache_free(extent_buffer_cache, eb);
4377 int extent_buffer_under_io(struct extent_buffer *eb)
4379 return (atomic_read(&eb->io_pages) ||
4380 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4381 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4385 * Helper for releasing extent buffer page.
4387 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4388 unsigned long start_idx)
4390 unsigned long index;
4391 unsigned long num_pages;
4393 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4395 BUG_ON(extent_buffer_under_io(eb));
4397 num_pages = num_extent_pages(eb->start, eb->len);
4398 index = start_idx + num_pages;
4399 if (start_idx >= index)
4404 page = extent_buffer_page(eb, index);
4405 if (page && mapped) {
4406 spin_lock(&page->mapping->private_lock);
4408 * We do this since we'll remove the pages after we've
4409 * removed the eb from the radix tree, so we could race
4410 * and have this page now attached to the new eb. So
4411 * only clear page_private if it's still connected to
4414 if (PagePrivate(page) &&
4415 page->private == (unsigned long)eb) {
4416 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4417 BUG_ON(PageDirty(page));
4418 BUG_ON(PageWriteback(page));
4420 * We need to make sure we haven't be attached
4423 ClearPagePrivate(page);
4424 set_page_private(page, 0);
4425 /* One for the page private */
4426 page_cache_release(page);
4428 spin_unlock(&page->mapping->private_lock);
4432 /* One for when we alloced the page */
4433 page_cache_release(page);
4435 } while (index != start_idx);
4439 * Helper for releasing the extent buffer.
4441 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4443 btrfs_release_extent_buffer_page(eb, 0);
4444 __free_extent_buffer(eb);
4447 static struct extent_buffer *
4448 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4449 unsigned long len, gfp_t mask)
4451 struct extent_buffer *eb = NULL;
4453 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4458 eb->fs_info = fs_info;
4460 rwlock_init(&eb->lock);
4461 atomic_set(&eb->write_locks, 0);
4462 atomic_set(&eb->read_locks, 0);
4463 atomic_set(&eb->blocking_readers, 0);
4464 atomic_set(&eb->blocking_writers, 0);
4465 atomic_set(&eb->spinning_readers, 0);
4466 atomic_set(&eb->spinning_writers, 0);
4467 eb->lock_nested = 0;
4468 init_waitqueue_head(&eb->write_lock_wq);
4469 init_waitqueue_head(&eb->read_lock_wq);
4471 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4473 spin_lock_init(&eb->refs_lock);
4474 atomic_set(&eb->refs, 1);
4475 atomic_set(&eb->io_pages, 0);
4478 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4480 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4481 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4482 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4487 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4491 struct extent_buffer *new;
4492 unsigned long num_pages = num_extent_pages(src->start, src->len);
4494 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4498 for (i = 0; i < num_pages; i++) {
4499 p = alloc_page(GFP_NOFS);
4501 btrfs_release_extent_buffer(new);
4504 attach_extent_buffer_page(new, p);
4505 WARN_ON(PageDirty(p));
4510 copy_extent_buffer(new, src, 0, 0, src->len);
4511 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4512 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4517 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4519 struct extent_buffer *eb;
4520 unsigned long num_pages = num_extent_pages(0, len);
4523 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4527 for (i = 0; i < num_pages; i++) {
4528 eb->pages[i] = alloc_page(GFP_NOFS);
4532 set_extent_buffer_uptodate(eb);
4533 btrfs_set_header_nritems(eb, 0);
4534 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4539 __free_page(eb->pages[i - 1]);
4540 __free_extent_buffer(eb);
4544 static void check_buffer_tree_ref(struct extent_buffer *eb)
4547 /* the ref bit is tricky. We have to make sure it is set
4548 * if we have the buffer dirty. Otherwise the
4549 * code to free a buffer can end up dropping a dirty
4552 * Once the ref bit is set, it won't go away while the
4553 * buffer is dirty or in writeback, and it also won't
4554 * go away while we have the reference count on the
4557 * We can't just set the ref bit without bumping the
4558 * ref on the eb because free_extent_buffer might
4559 * see the ref bit and try to clear it. If this happens
4560 * free_extent_buffer might end up dropping our original
4561 * ref by mistake and freeing the page before we are able
4562 * to add one more ref.
4564 * So bump the ref count first, then set the bit. If someone
4565 * beat us to it, drop the ref we added.
4567 refs = atomic_read(&eb->refs);
4568 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4571 spin_lock(&eb->refs_lock);
4572 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4573 atomic_inc(&eb->refs);
4574 spin_unlock(&eb->refs_lock);
4577 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4578 struct page *accessed)
4580 unsigned long num_pages, i;
4582 check_buffer_tree_ref(eb);
4584 num_pages = num_extent_pages(eb->start, eb->len);
4585 for (i = 0; i < num_pages; i++) {
4586 struct page *p = extent_buffer_page(eb, i);
4588 mark_page_accessed(p);
4592 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4595 struct extent_buffer *eb;
4598 eb = radix_tree_lookup(&fs_info->buffer_radix,
4599 start >> PAGE_CACHE_SHIFT);
4600 if (eb && atomic_inc_not_zero(&eb->refs)) {
4602 mark_extent_buffer_accessed(eb, NULL);
4610 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4611 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4612 u64 start, unsigned long len)
4614 struct extent_buffer *eb, *exists = NULL;
4617 eb = find_extent_buffer(fs_info, start);
4620 eb = alloc_dummy_extent_buffer(start, len);
4623 eb->fs_info = fs_info;
4625 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4628 spin_lock(&fs_info->buffer_lock);
4629 ret = radix_tree_insert(&fs_info->buffer_radix,
4630 start >> PAGE_CACHE_SHIFT, eb);
4631 spin_unlock(&fs_info->buffer_lock);
4632 radix_tree_preload_end();
4633 if (ret == -EEXIST) {
4634 exists = find_extent_buffer(fs_info, start);
4640 check_buffer_tree_ref(eb);
4641 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4644 * We will free dummy extent buffer's if they come into
4645 * free_extent_buffer with a ref count of 2, but if we are using this we
4646 * want the buffers to stay in memory until we're done with them, so
4647 * bump the ref count again.
4649 atomic_inc(&eb->refs);
4652 btrfs_release_extent_buffer(eb);
4657 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4658 u64 start, unsigned long len)
4660 unsigned long num_pages = num_extent_pages(start, len);
4662 unsigned long index = start >> PAGE_CACHE_SHIFT;
4663 struct extent_buffer *eb;
4664 struct extent_buffer *exists = NULL;
4666 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4670 eb = find_extent_buffer(fs_info, start);
4674 eb = __alloc_extent_buffer(fs_info, start, len, GFP_NOFS);
4678 for (i = 0; i < num_pages; i++, index++) {
4679 p = find_or_create_page(mapping, index, GFP_NOFS);
4683 spin_lock(&mapping->private_lock);
4684 if (PagePrivate(p)) {
4686 * We could have already allocated an eb for this page
4687 * and attached one so lets see if we can get a ref on
4688 * the existing eb, and if we can we know it's good and
4689 * we can just return that one, else we know we can just
4690 * overwrite page->private.
4692 exists = (struct extent_buffer *)p->private;
4693 if (atomic_inc_not_zero(&exists->refs)) {
4694 spin_unlock(&mapping->private_lock);
4696 page_cache_release(p);
4697 mark_extent_buffer_accessed(exists, p);
4702 * Do this so attach doesn't complain and we need to
4703 * drop the ref the old guy had.
4705 ClearPagePrivate(p);
4706 WARN_ON(PageDirty(p));
4707 page_cache_release(p);
4709 attach_extent_buffer_page(eb, p);
4710 spin_unlock(&mapping->private_lock);
4711 WARN_ON(PageDirty(p));
4713 if (!PageUptodate(p))
4717 * see below about how we avoid a nasty race with release page
4718 * and why we unlock later
4722 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4724 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4728 spin_lock(&fs_info->buffer_lock);
4729 ret = radix_tree_insert(&fs_info->buffer_radix,
4730 start >> PAGE_CACHE_SHIFT, eb);
4731 spin_unlock(&fs_info->buffer_lock);
4732 radix_tree_preload_end();
4733 if (ret == -EEXIST) {
4734 exists = find_extent_buffer(fs_info, start);
4740 /* add one reference for the tree */
4741 check_buffer_tree_ref(eb);
4742 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4745 * there is a race where release page may have
4746 * tried to find this extent buffer in the radix
4747 * but failed. It will tell the VM it is safe to
4748 * reclaim the, and it will clear the page private bit.
4749 * We must make sure to set the page private bit properly
4750 * after the extent buffer is in the radix tree so
4751 * it doesn't get lost
4753 SetPageChecked(eb->pages[0]);
4754 for (i = 1; i < num_pages; i++) {
4755 p = extent_buffer_page(eb, i);
4756 ClearPageChecked(p);
4759 unlock_page(eb->pages[0]);
4763 for (i = 0; i < num_pages; i++) {
4765 unlock_page(eb->pages[i]);
4768 WARN_ON(!atomic_dec_and_test(&eb->refs));
4769 btrfs_release_extent_buffer(eb);
4773 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4775 struct extent_buffer *eb =
4776 container_of(head, struct extent_buffer, rcu_head);
4778 __free_extent_buffer(eb);
4781 /* Expects to have eb->eb_lock already held */
4782 static int release_extent_buffer(struct extent_buffer *eb)
4784 WARN_ON(atomic_read(&eb->refs) == 0);
4785 if (atomic_dec_and_test(&eb->refs)) {
4786 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4787 struct btrfs_fs_info *fs_info = eb->fs_info;
4789 spin_unlock(&eb->refs_lock);
4791 spin_lock(&fs_info->buffer_lock);
4792 radix_tree_delete(&fs_info->buffer_radix,
4793 eb->start >> PAGE_CACHE_SHIFT);
4794 spin_unlock(&fs_info->buffer_lock);
4796 spin_unlock(&eb->refs_lock);
4799 /* Should be safe to release our pages at this point */
4800 btrfs_release_extent_buffer_page(eb, 0);
4801 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4804 spin_unlock(&eb->refs_lock);
4809 void free_extent_buffer(struct extent_buffer *eb)
4817 refs = atomic_read(&eb->refs);
4820 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4825 spin_lock(&eb->refs_lock);
4826 if (atomic_read(&eb->refs) == 2 &&
4827 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4828 atomic_dec(&eb->refs);
4830 if (atomic_read(&eb->refs) == 2 &&
4831 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4832 !extent_buffer_under_io(eb) &&
4833 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4834 atomic_dec(&eb->refs);
4837 * I know this is terrible, but it's temporary until we stop tracking
4838 * the uptodate bits and such for the extent buffers.
4840 release_extent_buffer(eb);
4843 void free_extent_buffer_stale(struct extent_buffer *eb)
4848 spin_lock(&eb->refs_lock);
4849 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4851 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4852 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4853 atomic_dec(&eb->refs);
4854 release_extent_buffer(eb);
4857 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4860 unsigned long num_pages;
4863 num_pages = num_extent_pages(eb->start, eb->len);
4865 for (i = 0; i < num_pages; i++) {
4866 page = extent_buffer_page(eb, i);
4867 if (!PageDirty(page))
4871 WARN_ON(!PagePrivate(page));
4873 clear_page_dirty_for_io(page);
4874 spin_lock_irq(&page->mapping->tree_lock);
4875 if (!PageDirty(page)) {
4876 radix_tree_tag_clear(&page->mapping->page_tree,
4878 PAGECACHE_TAG_DIRTY);
4880 spin_unlock_irq(&page->mapping->tree_lock);
4881 ClearPageError(page);
4884 WARN_ON(atomic_read(&eb->refs) == 0);
4887 int set_extent_buffer_dirty(struct extent_buffer *eb)
4890 unsigned long num_pages;
4893 check_buffer_tree_ref(eb);
4895 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4897 num_pages = num_extent_pages(eb->start, eb->len);
4898 WARN_ON(atomic_read(&eb->refs) == 0);
4899 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4901 for (i = 0; i < num_pages; i++)
4902 set_page_dirty(extent_buffer_page(eb, i));
4906 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4910 unsigned long num_pages;
4912 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4913 num_pages = num_extent_pages(eb->start, eb->len);
4914 for (i = 0; i < num_pages; i++) {
4915 page = extent_buffer_page(eb, i);
4917 ClearPageUptodate(page);
4922 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4926 unsigned long num_pages;
4928 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4929 num_pages = num_extent_pages(eb->start, eb->len);
4930 for (i = 0; i < num_pages; i++) {
4931 page = extent_buffer_page(eb, i);
4932 SetPageUptodate(page);
4937 int extent_buffer_uptodate(struct extent_buffer *eb)
4939 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4942 int read_extent_buffer_pages(struct extent_io_tree *tree,
4943 struct extent_buffer *eb, u64 start, int wait,
4944 get_extent_t *get_extent, int mirror_num)
4947 unsigned long start_i;
4951 int locked_pages = 0;
4952 int all_uptodate = 1;
4953 unsigned long num_pages;
4954 unsigned long num_reads = 0;
4955 struct bio *bio = NULL;
4956 unsigned long bio_flags = 0;
4958 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4962 WARN_ON(start < eb->start);
4963 start_i = (start >> PAGE_CACHE_SHIFT) -
4964 (eb->start >> PAGE_CACHE_SHIFT);
4969 num_pages = num_extent_pages(eb->start, eb->len);
4970 for (i = start_i; i < num_pages; i++) {
4971 page = extent_buffer_page(eb, i);
4972 if (wait == WAIT_NONE) {
4973 if (!trylock_page(page))
4979 if (!PageUptodate(page)) {
4986 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4990 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4991 eb->read_mirror = 0;
4992 atomic_set(&eb->io_pages, num_reads);
4993 for (i = start_i; i < num_pages; i++) {
4994 page = extent_buffer_page(eb, i);
4995 if (!PageUptodate(page)) {
4996 ClearPageError(page);
4997 err = __extent_read_full_page(tree, page,
4999 mirror_num, &bio_flags,
5009 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
5015 if (ret || wait != WAIT_COMPLETE)
5018 for (i = start_i; i < num_pages; i++) {
5019 page = extent_buffer_page(eb, i);
5020 wait_on_page_locked(page);
5021 if (!PageUptodate(page))
5029 while (locked_pages > 0) {
5030 page = extent_buffer_page(eb, i);
5038 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
5039 unsigned long start,
5046 char *dst = (char *)dstv;
5047 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5048 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5050 WARN_ON(start > eb->len);
5051 WARN_ON(start + len > eb->start + eb->len);
5053 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5056 page = extent_buffer_page(eb, i);
5058 cur = min(len, (PAGE_CACHE_SIZE - offset));
5059 kaddr = page_address(page);
5060 memcpy(dst, kaddr + offset, cur);
5069 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv,
5070 unsigned long start,
5077 char __user *dst = (char __user *)dstv;
5078 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5079 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5082 WARN_ON(start > eb->len);
5083 WARN_ON(start + len > eb->start + eb->len);
5085 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5088 page = extent_buffer_page(eb, i);
5090 cur = min(len, (PAGE_CACHE_SIZE - offset));
5091 kaddr = page_address(page);
5092 if (copy_to_user(dst, kaddr + offset, cur)) {
5106 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
5107 unsigned long min_len, char **map,
5108 unsigned long *map_start,
5109 unsigned long *map_len)
5111 size_t offset = start & (PAGE_CACHE_SIZE - 1);
5114 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5115 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5116 unsigned long end_i = (start_offset + start + min_len - 1) >>
5123 offset = start_offset;
5127 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
5130 if (start + min_len > eb->len) {
5131 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
5133 eb->start, eb->len, start, min_len);
5137 p = extent_buffer_page(eb, i);
5138 kaddr = page_address(p);
5139 *map = kaddr + offset;
5140 *map_len = PAGE_CACHE_SIZE - offset;
5144 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5145 unsigned long start,
5152 char *ptr = (char *)ptrv;
5153 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5154 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5157 WARN_ON(start > eb->len);
5158 WARN_ON(start + len > eb->start + eb->len);
5160 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5163 page = extent_buffer_page(eb, i);
5165 cur = min(len, (PAGE_CACHE_SIZE - offset));
5167 kaddr = page_address(page);
5168 ret = memcmp(ptr, kaddr + offset, cur);
5180 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5181 unsigned long start, unsigned long len)
5187 char *src = (char *)srcv;
5188 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5189 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5191 WARN_ON(start > eb->len);
5192 WARN_ON(start + len > eb->start + eb->len);
5194 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5197 page = extent_buffer_page(eb, i);
5198 WARN_ON(!PageUptodate(page));
5200 cur = min(len, PAGE_CACHE_SIZE - offset);
5201 kaddr = page_address(page);
5202 memcpy(kaddr + offset, src, cur);
5211 void memset_extent_buffer(struct extent_buffer *eb, char c,
5212 unsigned long start, unsigned long len)
5218 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5219 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5221 WARN_ON(start > eb->len);
5222 WARN_ON(start + len > eb->start + eb->len);
5224 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5227 page = extent_buffer_page(eb, i);
5228 WARN_ON(!PageUptodate(page));
5230 cur = min(len, PAGE_CACHE_SIZE - offset);
5231 kaddr = page_address(page);
5232 memset(kaddr + offset, c, cur);
5240 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5241 unsigned long dst_offset, unsigned long src_offset,
5244 u64 dst_len = dst->len;
5249 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5250 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5252 WARN_ON(src->len != dst_len);
5254 offset = (start_offset + dst_offset) &
5255 (PAGE_CACHE_SIZE - 1);
5258 page = extent_buffer_page(dst, i);
5259 WARN_ON(!PageUptodate(page));
5261 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5263 kaddr = page_address(page);
5264 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5273 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5275 unsigned long distance = (src > dst) ? src - dst : dst - src;
5276 return distance < len;
5279 static void copy_pages(struct page *dst_page, struct page *src_page,
5280 unsigned long dst_off, unsigned long src_off,
5283 char *dst_kaddr = page_address(dst_page);
5285 int must_memmove = 0;
5287 if (dst_page != src_page) {
5288 src_kaddr = page_address(src_page);
5290 src_kaddr = dst_kaddr;
5291 if (areas_overlap(src_off, dst_off, len))
5296 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5298 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5301 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5302 unsigned long src_offset, unsigned long len)
5305 size_t dst_off_in_page;
5306 size_t src_off_in_page;
5307 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5308 unsigned long dst_i;
5309 unsigned long src_i;
5311 if (src_offset + len > dst->len) {
5312 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5313 "len %lu dst len %lu\n", src_offset, len, dst->len);
5316 if (dst_offset + len > dst->len) {
5317 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5318 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5323 dst_off_in_page = (start_offset + dst_offset) &
5324 (PAGE_CACHE_SIZE - 1);
5325 src_off_in_page = (start_offset + src_offset) &
5326 (PAGE_CACHE_SIZE - 1);
5328 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5329 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5331 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5333 cur = min_t(unsigned long, cur,
5334 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5336 copy_pages(extent_buffer_page(dst, dst_i),
5337 extent_buffer_page(dst, src_i),
5338 dst_off_in_page, src_off_in_page, cur);
5346 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5347 unsigned long src_offset, unsigned long len)
5350 size_t dst_off_in_page;
5351 size_t src_off_in_page;
5352 unsigned long dst_end = dst_offset + len - 1;
5353 unsigned long src_end = src_offset + len - 1;
5354 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5355 unsigned long dst_i;
5356 unsigned long src_i;
5358 if (src_offset + len > dst->len) {
5359 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5360 "len %lu len %lu\n", src_offset, len, dst->len);
5363 if (dst_offset + len > dst->len) {
5364 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5365 "len %lu len %lu\n", dst_offset, len, dst->len);
5368 if (dst_offset < src_offset) {
5369 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5373 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5374 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5376 dst_off_in_page = (start_offset + dst_end) &
5377 (PAGE_CACHE_SIZE - 1);
5378 src_off_in_page = (start_offset + src_end) &
5379 (PAGE_CACHE_SIZE - 1);
5381 cur = min_t(unsigned long, len, src_off_in_page + 1);
5382 cur = min(cur, dst_off_in_page + 1);
5383 copy_pages(extent_buffer_page(dst, dst_i),
5384 extent_buffer_page(dst, src_i),
5385 dst_off_in_page - cur + 1,
5386 src_off_in_page - cur + 1, cur);
5394 int try_release_extent_buffer(struct page *page)
5396 struct extent_buffer *eb;
5399 * We need to make sure noboody is attaching this page to an eb right
5402 spin_lock(&page->mapping->private_lock);
5403 if (!PagePrivate(page)) {
5404 spin_unlock(&page->mapping->private_lock);
5408 eb = (struct extent_buffer *)page->private;
5412 * This is a little awful but should be ok, we need to make sure that
5413 * the eb doesn't disappear out from under us while we're looking at
5416 spin_lock(&eb->refs_lock);
5417 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5418 spin_unlock(&eb->refs_lock);
5419 spin_unlock(&page->mapping->private_lock);
5422 spin_unlock(&page->mapping->private_lock);
5425 * If tree ref isn't set then we know the ref on this eb is a real ref,
5426 * so just return, this page will likely be freed soon anyway.
5428 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5429 spin_unlock(&eb->refs_lock);
5433 return release_extent_buffer(eb);