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"
18 #include "btrfs_inode.h"
20 #include "check-integrity.h"
22 #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 #ifdef CONFIG_BTRFS_DEBUG
29 static LIST_HEAD(buffers);
30 static LIST_HEAD(states);
32 static DEFINE_SPINLOCK(leak_lock);
35 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
39 spin_lock_irqsave(&leak_lock, flags);
41 spin_unlock_irqrestore(&leak_lock, flags);
45 void btrfs_leak_debug_del(struct list_head *entry)
49 spin_lock_irqsave(&leak_lock, flags);
51 spin_unlock_irqrestore(&leak_lock, flags);
55 void btrfs_leak_debug_check(void)
57 struct extent_state *state;
58 struct extent_buffer *eb;
60 while (!list_empty(&states)) {
61 state = list_entry(states.next, struct extent_state, leak_list);
62 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
63 "state %lu in tree %p refs %d\n",
64 (unsigned long long)state->start,
65 (unsigned long long)state->end,
66 state->state, state->tree, atomic_read(&state->refs));
67 list_del(&state->leak_list);
68 kmem_cache_free(extent_state_cache, state);
71 while (!list_empty(&buffers)) {
72 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
73 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
74 "refs %d\n", (unsigned long long)eb->start,
75 eb->len, atomic_read(&eb->refs));
76 list_del(&eb->leak_list);
77 kmem_cache_free(extent_buffer_cache, eb);
81 #define btrfs_debug_check_extent_io_range(inode, start, end) \
82 __btrfs_debug_check_extent_io_range(__func__, (inode), (start), (end))
83 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
84 struct inode *inode, u64 start, u64 end)
86 u64 isize = i_size_read(inode);
88 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
89 printk_ratelimited(KERN_DEBUG
90 "btrfs: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
92 (unsigned long long)btrfs_ino(inode),
93 (unsigned long long)isize,
94 (unsigned long long)start,
95 (unsigned long long)end);
99 #define btrfs_leak_debug_add(new, head) do {} while (0)
100 #define btrfs_leak_debug_del(entry) do {} while (0)
101 #define btrfs_leak_debug_check() do {} while (0)
102 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
105 #define BUFFER_LRU_MAX 64
110 struct rb_node rb_node;
113 struct extent_page_data {
115 struct extent_io_tree *tree;
116 get_extent_t *get_extent;
117 unsigned long bio_flags;
119 /* tells writepage not to lock the state bits for this range
120 * it still does the unlocking
122 unsigned int extent_locked:1;
124 /* tells the submit_bio code to use a WRITE_SYNC */
125 unsigned int sync_io:1;
128 static noinline void flush_write_bio(void *data);
129 static inline struct btrfs_fs_info *
130 tree_fs_info(struct extent_io_tree *tree)
132 return btrfs_sb(tree->mapping->host->i_sb);
135 int __init extent_io_init(void)
137 extent_state_cache = kmem_cache_create("btrfs_extent_state",
138 sizeof(struct extent_state), 0,
139 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
140 if (!extent_state_cache)
143 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
144 sizeof(struct extent_buffer), 0,
145 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
146 if (!extent_buffer_cache)
147 goto free_state_cache;
149 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
150 offsetof(struct btrfs_io_bio, bio));
152 goto free_buffer_cache;
156 kmem_cache_destroy(extent_buffer_cache);
157 extent_buffer_cache = NULL;
160 kmem_cache_destroy(extent_state_cache);
161 extent_state_cache = NULL;
165 void extent_io_exit(void)
167 btrfs_leak_debug_check();
170 * Make sure all delayed rcu free are flushed before we
174 if (extent_state_cache)
175 kmem_cache_destroy(extent_state_cache);
176 if (extent_buffer_cache)
177 kmem_cache_destroy(extent_buffer_cache);
179 bioset_free(btrfs_bioset);
182 void extent_io_tree_init(struct extent_io_tree *tree,
183 struct address_space *mapping)
185 tree->state = RB_ROOT;
186 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
188 tree->dirty_bytes = 0;
189 spin_lock_init(&tree->lock);
190 spin_lock_init(&tree->buffer_lock);
191 tree->mapping = mapping;
194 static struct extent_state *alloc_extent_state(gfp_t mask)
196 struct extent_state *state;
198 state = kmem_cache_alloc(extent_state_cache, mask);
204 btrfs_leak_debug_add(&state->leak_list, &states);
205 atomic_set(&state->refs, 1);
206 init_waitqueue_head(&state->wq);
207 trace_alloc_extent_state(state, mask, _RET_IP_);
211 void free_extent_state(struct extent_state *state)
215 if (atomic_dec_and_test(&state->refs)) {
216 WARN_ON(state->tree);
217 btrfs_leak_debug_del(&state->leak_list);
218 trace_free_extent_state(state, _RET_IP_);
219 kmem_cache_free(extent_state_cache, state);
223 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
224 struct rb_node *node)
226 struct rb_node **p = &root->rb_node;
227 struct rb_node *parent = NULL;
228 struct tree_entry *entry;
232 entry = rb_entry(parent, struct tree_entry, rb_node);
234 if (offset < entry->start)
236 else if (offset > entry->end)
242 rb_link_node(node, parent, p);
243 rb_insert_color(node, root);
247 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
248 struct rb_node **prev_ret,
249 struct rb_node **next_ret)
251 struct rb_root *root = &tree->state;
252 struct rb_node *n = root->rb_node;
253 struct rb_node *prev = NULL;
254 struct rb_node *orig_prev = NULL;
255 struct tree_entry *entry;
256 struct tree_entry *prev_entry = NULL;
259 entry = rb_entry(n, struct tree_entry, rb_node);
263 if (offset < entry->start)
265 else if (offset > entry->end)
273 while (prev && offset > prev_entry->end) {
274 prev = rb_next(prev);
275 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
282 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
283 while (prev && offset < prev_entry->start) {
284 prev = rb_prev(prev);
285 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
292 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
295 struct rb_node *prev = NULL;
298 ret = __etree_search(tree, offset, &prev, NULL);
304 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
305 struct extent_state *other)
307 if (tree->ops && tree->ops->merge_extent_hook)
308 tree->ops->merge_extent_hook(tree->mapping->host, new,
313 * utility function to look for merge candidates inside a given range.
314 * Any extents with matching state are merged together into a single
315 * extent in the tree. Extents with EXTENT_IO in their state field
316 * are not merged because the end_io handlers need to be able to do
317 * operations on them without sleeping (or doing allocations/splits).
319 * This should be called with the tree lock held.
321 static void merge_state(struct extent_io_tree *tree,
322 struct extent_state *state)
324 struct extent_state *other;
325 struct rb_node *other_node;
327 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
330 other_node = rb_prev(&state->rb_node);
332 other = rb_entry(other_node, struct extent_state, rb_node);
333 if (other->end == state->start - 1 &&
334 other->state == state->state) {
335 merge_cb(tree, state, other);
336 state->start = other->start;
338 rb_erase(&other->rb_node, &tree->state);
339 free_extent_state(other);
342 other_node = rb_next(&state->rb_node);
344 other = rb_entry(other_node, struct extent_state, rb_node);
345 if (other->start == state->end + 1 &&
346 other->state == state->state) {
347 merge_cb(tree, state, other);
348 state->end = other->end;
350 rb_erase(&other->rb_node, &tree->state);
351 free_extent_state(other);
356 static void set_state_cb(struct extent_io_tree *tree,
357 struct extent_state *state, unsigned long *bits)
359 if (tree->ops && tree->ops->set_bit_hook)
360 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
363 static void clear_state_cb(struct extent_io_tree *tree,
364 struct extent_state *state, unsigned long *bits)
366 if (tree->ops && tree->ops->clear_bit_hook)
367 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
370 static void set_state_bits(struct extent_io_tree *tree,
371 struct extent_state *state, unsigned long *bits);
374 * insert an extent_state struct into the tree. 'bits' are set on the
375 * struct before it is inserted.
377 * This may return -EEXIST if the extent is already there, in which case the
378 * state struct is freed.
380 * The tree lock is not taken internally. This is a utility function and
381 * probably isn't what you want to call (see set/clear_extent_bit).
383 static int insert_state(struct extent_io_tree *tree,
384 struct extent_state *state, u64 start, u64 end,
387 struct rb_node *node;
390 WARN(1, KERN_ERR "btrfs end < start %llu %llu\n",
391 (unsigned long long)end,
392 (unsigned long long)start);
393 state->start = start;
396 set_state_bits(tree, state, bits);
398 node = tree_insert(&tree->state, end, &state->rb_node);
400 struct extent_state *found;
401 found = rb_entry(node, struct extent_state, rb_node);
402 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
403 "%llu %llu\n", (unsigned long long)found->start,
404 (unsigned long long)found->end,
405 (unsigned long long)start, (unsigned long long)end);
409 merge_state(tree, state);
413 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
416 if (tree->ops && tree->ops->split_extent_hook)
417 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
421 * split a given extent state struct in two, inserting the preallocated
422 * struct 'prealloc' as the newly created second half. 'split' indicates an
423 * offset inside 'orig' where it should be split.
426 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
427 * are two extent state structs in the tree:
428 * prealloc: [orig->start, split - 1]
429 * orig: [ split, orig->end ]
431 * The tree locks are not taken by this function. They need to be held
434 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
435 struct extent_state *prealloc, u64 split)
437 struct rb_node *node;
439 split_cb(tree, orig, split);
441 prealloc->start = orig->start;
442 prealloc->end = split - 1;
443 prealloc->state = orig->state;
446 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
448 free_extent_state(prealloc);
451 prealloc->tree = tree;
455 static struct extent_state *next_state(struct extent_state *state)
457 struct rb_node *next = rb_next(&state->rb_node);
459 return rb_entry(next, struct extent_state, rb_node);
465 * utility function to clear some bits in an extent state struct.
466 * it will optionally wake up any one waiting on this state (wake == 1).
468 * If no bits are set on the state struct after clearing things, the
469 * struct is freed and removed from the tree
471 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
472 struct extent_state *state,
473 unsigned long *bits, int wake)
475 struct extent_state *next;
476 unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
478 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
479 u64 range = state->end - state->start + 1;
480 WARN_ON(range > tree->dirty_bytes);
481 tree->dirty_bytes -= range;
483 clear_state_cb(tree, state, bits);
484 state->state &= ~bits_to_clear;
487 if (state->state == 0) {
488 next = next_state(state);
490 rb_erase(&state->rb_node, &tree->state);
492 free_extent_state(state);
497 merge_state(tree, state);
498 next = next_state(state);
503 static struct extent_state *
504 alloc_extent_state_atomic(struct extent_state *prealloc)
507 prealloc = alloc_extent_state(GFP_ATOMIC);
512 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
514 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
515 "Extent tree was modified by another "
516 "thread while locked.");
520 * clear some bits on a range in the tree. This may require splitting
521 * or inserting elements in the tree, so the gfp mask is used to
522 * indicate which allocations or sleeping are allowed.
524 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
525 * the given range from the tree regardless of state (ie for truncate).
527 * the range [start, end] is inclusive.
529 * This takes the tree lock, and returns 0 on success and < 0 on error.
531 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
532 unsigned long bits, int wake, int delete,
533 struct extent_state **cached_state,
536 struct extent_state *state;
537 struct extent_state *cached;
538 struct extent_state *prealloc = NULL;
539 struct rb_node *node;
544 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
546 if (bits & EXTENT_DELALLOC)
547 bits |= EXTENT_NORESERVE;
550 bits |= ~EXTENT_CTLBITS;
551 bits |= EXTENT_FIRST_DELALLOC;
553 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
556 if (!prealloc && (mask & __GFP_WAIT)) {
557 prealloc = alloc_extent_state(mask);
562 spin_lock(&tree->lock);
564 cached = *cached_state;
567 *cached_state = NULL;
571 if (cached && cached->tree && cached->start <= start &&
572 cached->end > start) {
574 atomic_dec(&cached->refs);
579 free_extent_state(cached);
582 * this search will find the extents that end after
585 node = tree_search(tree, start);
588 state = rb_entry(node, struct extent_state, rb_node);
590 if (state->start > end)
592 WARN_ON(state->end < start);
593 last_end = state->end;
595 /* the state doesn't have the wanted bits, go ahead */
596 if (!(state->state & bits)) {
597 state = next_state(state);
602 * | ---- desired range ---- |
604 * | ------------- state -------------- |
606 * We need to split the extent we found, and may flip
607 * bits on second half.
609 * If the extent we found extends past our range, we
610 * just split and search again. It'll get split again
611 * the next time though.
613 * If the extent we found is inside our range, we clear
614 * the desired bit on it.
617 if (state->start < start) {
618 prealloc = alloc_extent_state_atomic(prealloc);
620 err = split_state(tree, state, prealloc, start);
622 extent_io_tree_panic(tree, err);
627 if (state->end <= end) {
628 state = clear_state_bit(tree, state, &bits, wake);
634 * | ---- desired range ---- |
636 * We need to split the extent, and clear the bit
639 if (state->start <= end && state->end > end) {
640 prealloc = alloc_extent_state_atomic(prealloc);
642 err = split_state(tree, state, prealloc, end + 1);
644 extent_io_tree_panic(tree, err);
649 clear_state_bit(tree, prealloc, &bits, wake);
655 state = clear_state_bit(tree, state, &bits, wake);
657 if (last_end == (u64)-1)
659 start = last_end + 1;
660 if (start <= end && state && !need_resched())
665 spin_unlock(&tree->lock);
667 free_extent_state(prealloc);
674 spin_unlock(&tree->lock);
675 if (mask & __GFP_WAIT)
680 static void wait_on_state(struct extent_io_tree *tree,
681 struct extent_state *state)
682 __releases(tree->lock)
683 __acquires(tree->lock)
686 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
687 spin_unlock(&tree->lock);
689 spin_lock(&tree->lock);
690 finish_wait(&state->wq, &wait);
694 * waits for one or more bits to clear on a range in the state tree.
695 * The range [start, end] is inclusive.
696 * The tree lock is taken by this function
698 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
701 struct extent_state *state;
702 struct rb_node *node;
704 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
706 spin_lock(&tree->lock);
710 * this search will find all the extents that end after
713 node = tree_search(tree, start);
717 state = rb_entry(node, struct extent_state, rb_node);
719 if (state->start > end)
722 if (state->state & bits) {
723 start = state->start;
724 atomic_inc(&state->refs);
725 wait_on_state(tree, state);
726 free_extent_state(state);
729 start = state->end + 1;
734 cond_resched_lock(&tree->lock);
737 spin_unlock(&tree->lock);
740 static void set_state_bits(struct extent_io_tree *tree,
741 struct extent_state *state,
744 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
746 set_state_cb(tree, state, bits);
747 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
748 u64 range = state->end - state->start + 1;
749 tree->dirty_bytes += range;
751 state->state |= bits_to_set;
754 static void cache_state(struct extent_state *state,
755 struct extent_state **cached_ptr)
757 if (cached_ptr && !(*cached_ptr)) {
758 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
760 atomic_inc(&state->refs);
765 static void uncache_state(struct extent_state **cached_ptr)
767 if (cached_ptr && (*cached_ptr)) {
768 struct extent_state *state = *cached_ptr;
770 free_extent_state(state);
775 * set some bits on a range in the tree. This may require allocations or
776 * sleeping, so the gfp mask is used to indicate what is allowed.
778 * If any of the exclusive bits are set, this will fail with -EEXIST if some
779 * part of the range already has the desired bits set. The start of the
780 * existing range is returned in failed_start in this case.
782 * [start, end] is inclusive This takes the tree lock.
785 static int __must_check
786 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
787 unsigned long bits, unsigned long exclusive_bits,
788 u64 *failed_start, struct extent_state **cached_state,
791 struct extent_state *state;
792 struct extent_state *prealloc = NULL;
793 struct rb_node *node;
798 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
800 bits |= EXTENT_FIRST_DELALLOC;
802 if (!prealloc && (mask & __GFP_WAIT)) {
803 prealloc = alloc_extent_state(mask);
807 spin_lock(&tree->lock);
808 if (cached_state && *cached_state) {
809 state = *cached_state;
810 if (state->start <= start && state->end > start &&
812 node = &state->rb_node;
817 * this search will find all the extents that end after
820 node = tree_search(tree, start);
822 prealloc = alloc_extent_state_atomic(prealloc);
824 err = insert_state(tree, prealloc, start, end, &bits);
826 extent_io_tree_panic(tree, err);
831 state = rb_entry(node, struct extent_state, rb_node);
833 last_start = state->start;
834 last_end = state->end;
837 * | ---- desired range ---- |
840 * Just lock what we found and keep going
842 if (state->start == start && state->end <= end) {
843 if (state->state & exclusive_bits) {
844 *failed_start = state->start;
849 set_state_bits(tree, state, &bits);
850 cache_state(state, cached_state);
851 merge_state(tree, state);
852 if (last_end == (u64)-1)
854 start = last_end + 1;
855 state = next_state(state);
856 if (start < end && state && state->start == start &&
863 * | ---- desired range ---- |
866 * | ------------- state -------------- |
868 * We need to split the extent we found, and may flip bits on
871 * If the extent we found extends past our
872 * range, we just split and search again. It'll get split
873 * again the next time though.
875 * If the extent we found is inside our range, we set the
878 if (state->start < start) {
879 if (state->state & exclusive_bits) {
880 *failed_start = start;
885 prealloc = alloc_extent_state_atomic(prealloc);
887 err = split_state(tree, state, prealloc, start);
889 extent_io_tree_panic(tree, err);
894 if (state->end <= end) {
895 set_state_bits(tree, state, &bits);
896 cache_state(state, cached_state);
897 merge_state(tree, state);
898 if (last_end == (u64)-1)
900 start = last_end + 1;
901 state = next_state(state);
902 if (start < end && state && state->start == start &&
909 * | ---- desired range ---- |
910 * | state | or | state |
912 * There's a hole, we need to insert something in it and
913 * ignore the extent we found.
915 if (state->start > start) {
917 if (end < last_start)
920 this_end = last_start - 1;
922 prealloc = alloc_extent_state_atomic(prealloc);
926 * Avoid to free 'prealloc' if it can be merged with
929 err = insert_state(tree, prealloc, start, this_end,
932 extent_io_tree_panic(tree, err);
934 cache_state(prealloc, cached_state);
936 start = this_end + 1;
940 * | ---- desired range ---- |
942 * We need to split the extent, and set the bit
945 if (state->start <= end && state->end > end) {
946 if (state->state & exclusive_bits) {
947 *failed_start = start;
952 prealloc = alloc_extent_state_atomic(prealloc);
954 err = split_state(tree, state, prealloc, end + 1);
956 extent_io_tree_panic(tree, err);
958 set_state_bits(tree, prealloc, &bits);
959 cache_state(prealloc, cached_state);
960 merge_state(tree, prealloc);
968 spin_unlock(&tree->lock);
970 free_extent_state(prealloc);
977 spin_unlock(&tree->lock);
978 if (mask & __GFP_WAIT)
983 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
984 unsigned long bits, u64 * failed_start,
985 struct extent_state **cached_state, gfp_t mask)
987 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
993 * convert_extent_bit - convert all bits in a given range from one bit to
995 * @tree: the io tree to search
996 * @start: the start offset in bytes
997 * @end: the end offset in bytes (inclusive)
998 * @bits: the bits to set in this range
999 * @clear_bits: the bits to clear in this range
1000 * @cached_state: state that we're going to cache
1001 * @mask: the allocation mask
1003 * This will go through and set bits for the given range. If any states exist
1004 * already in this range they are set with the given bit and cleared of the
1005 * clear_bits. This is only meant to be used by things that are mergeable, ie
1006 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1007 * boundary bits like LOCK.
1009 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1010 unsigned long bits, unsigned long clear_bits,
1011 struct extent_state **cached_state, gfp_t mask)
1013 struct extent_state *state;
1014 struct extent_state *prealloc = NULL;
1015 struct rb_node *node;
1020 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
1023 if (!prealloc && (mask & __GFP_WAIT)) {
1024 prealloc = alloc_extent_state(mask);
1029 spin_lock(&tree->lock);
1030 if (cached_state && *cached_state) {
1031 state = *cached_state;
1032 if (state->start <= start && state->end > start &&
1034 node = &state->rb_node;
1040 * this search will find all the extents that end after
1043 node = tree_search(tree, start);
1045 prealloc = alloc_extent_state_atomic(prealloc);
1050 err = insert_state(tree, prealloc, start, end, &bits);
1053 extent_io_tree_panic(tree, err);
1056 state = rb_entry(node, struct extent_state, rb_node);
1058 last_start = state->start;
1059 last_end = state->end;
1062 * | ---- desired range ---- |
1065 * Just lock what we found and keep going
1067 if (state->start == start && state->end <= end) {
1068 set_state_bits(tree, state, &bits);
1069 cache_state(state, cached_state);
1070 state = clear_state_bit(tree, state, &clear_bits, 0);
1071 if (last_end == (u64)-1)
1073 start = last_end + 1;
1074 if (start < end && state && state->start == start &&
1081 * | ---- desired range ---- |
1084 * | ------------- state -------------- |
1086 * We need to split the extent we found, and may flip bits on
1089 * If the extent we found extends past our
1090 * range, we just split and search again. It'll get split
1091 * again the next time though.
1093 * If the extent we found is inside our range, we set the
1094 * desired bit on it.
1096 if (state->start < start) {
1097 prealloc = alloc_extent_state_atomic(prealloc);
1102 err = split_state(tree, state, prealloc, start);
1104 extent_io_tree_panic(tree, err);
1108 if (state->end <= end) {
1109 set_state_bits(tree, state, &bits);
1110 cache_state(state, cached_state);
1111 state = clear_state_bit(tree, state, &clear_bits, 0);
1112 if (last_end == (u64)-1)
1114 start = last_end + 1;
1115 if (start < end && state && state->start == start &&
1122 * | ---- desired range ---- |
1123 * | state | or | state |
1125 * There's a hole, we need to insert something in it and
1126 * ignore the extent we found.
1128 if (state->start > start) {
1130 if (end < last_start)
1133 this_end = last_start - 1;
1135 prealloc = alloc_extent_state_atomic(prealloc);
1142 * Avoid to free 'prealloc' if it can be merged with
1145 err = insert_state(tree, prealloc, start, this_end,
1148 extent_io_tree_panic(tree, err);
1149 cache_state(prealloc, cached_state);
1151 start = this_end + 1;
1155 * | ---- desired range ---- |
1157 * We need to split the extent, and set the bit
1160 if (state->start <= end && state->end > end) {
1161 prealloc = alloc_extent_state_atomic(prealloc);
1167 err = split_state(tree, state, prealloc, end + 1);
1169 extent_io_tree_panic(tree, err);
1171 set_state_bits(tree, prealloc, &bits);
1172 cache_state(prealloc, cached_state);
1173 clear_state_bit(tree, prealloc, &clear_bits, 0);
1181 spin_unlock(&tree->lock);
1183 free_extent_state(prealloc);
1190 spin_unlock(&tree->lock);
1191 if (mask & __GFP_WAIT)
1196 /* wrappers around set/clear extent bit */
1197 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1200 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1204 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1205 unsigned long bits, gfp_t mask)
1207 return set_extent_bit(tree, start, end, bits, NULL,
1211 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1212 unsigned long bits, gfp_t mask)
1214 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1217 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1218 struct extent_state **cached_state, gfp_t mask)
1220 return set_extent_bit(tree, start, end,
1221 EXTENT_DELALLOC | EXTENT_UPTODATE,
1222 NULL, cached_state, mask);
1225 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1226 struct extent_state **cached_state, gfp_t mask)
1228 return set_extent_bit(tree, start, end,
1229 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1230 NULL, cached_state, mask);
1233 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1236 return clear_extent_bit(tree, start, end,
1237 EXTENT_DIRTY | EXTENT_DELALLOC |
1238 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1241 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1244 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1248 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1249 struct extent_state **cached_state, gfp_t mask)
1251 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1252 cached_state, mask);
1255 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1256 struct extent_state **cached_state, gfp_t mask)
1258 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1259 cached_state, mask);
1263 * either insert or lock state struct between start and end use mask to tell
1264 * us if waiting is desired.
1266 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1267 unsigned long bits, struct extent_state **cached_state)
1272 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1273 EXTENT_LOCKED, &failed_start,
1274 cached_state, GFP_NOFS);
1275 if (err == -EEXIST) {
1276 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1277 start = failed_start;
1280 WARN_ON(start > end);
1285 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1287 return lock_extent_bits(tree, start, end, 0, NULL);
1290 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1295 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1296 &failed_start, NULL, GFP_NOFS);
1297 if (err == -EEXIST) {
1298 if (failed_start > start)
1299 clear_extent_bit(tree, start, failed_start - 1,
1300 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1306 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1307 struct extent_state **cached, gfp_t mask)
1309 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1313 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1315 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1319 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1321 unsigned long index = start >> PAGE_CACHE_SHIFT;
1322 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1325 while (index <= end_index) {
1326 page = find_get_page(inode->i_mapping, index);
1327 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1328 clear_page_dirty_for_io(page);
1329 page_cache_release(page);
1335 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1337 unsigned long index = start >> PAGE_CACHE_SHIFT;
1338 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1341 while (index <= end_index) {
1342 page = find_get_page(inode->i_mapping, index);
1343 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1344 account_page_redirty(page);
1345 __set_page_dirty_nobuffers(page);
1346 page_cache_release(page);
1353 * helper function to set both pages and extents in the tree writeback
1355 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1357 unsigned long index = start >> PAGE_CACHE_SHIFT;
1358 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1361 while (index <= end_index) {
1362 page = find_get_page(tree->mapping, index);
1363 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1364 set_page_writeback(page);
1365 page_cache_release(page);
1371 /* find the first state struct with 'bits' set after 'start', and
1372 * return it. tree->lock must be held. NULL will returned if
1373 * nothing was found after 'start'
1375 static struct extent_state *
1376 find_first_extent_bit_state(struct extent_io_tree *tree,
1377 u64 start, unsigned long bits)
1379 struct rb_node *node;
1380 struct extent_state *state;
1383 * this search will find all the extents that end after
1386 node = tree_search(tree, start);
1391 state = rb_entry(node, struct extent_state, rb_node);
1392 if (state->end >= start && (state->state & bits))
1395 node = rb_next(node);
1404 * find the first offset in the io tree with 'bits' set. zero is
1405 * returned if we find something, and *start_ret and *end_ret are
1406 * set to reflect the state struct that was found.
1408 * If nothing was found, 1 is returned. If found something, return 0.
1410 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1411 u64 *start_ret, u64 *end_ret, unsigned long bits,
1412 struct extent_state **cached_state)
1414 struct extent_state *state;
1418 spin_lock(&tree->lock);
1419 if (cached_state && *cached_state) {
1420 state = *cached_state;
1421 if (state->end == start - 1 && state->tree) {
1422 n = rb_next(&state->rb_node);
1424 state = rb_entry(n, struct extent_state,
1426 if (state->state & bits)
1430 free_extent_state(*cached_state);
1431 *cached_state = NULL;
1434 free_extent_state(*cached_state);
1435 *cached_state = NULL;
1438 state = find_first_extent_bit_state(tree, start, bits);
1441 cache_state(state, cached_state);
1442 *start_ret = state->start;
1443 *end_ret = state->end;
1447 spin_unlock(&tree->lock);
1452 * find a contiguous range of bytes in the file marked as delalloc, not
1453 * more than 'max_bytes'. start and end are used to return the range,
1455 * 1 is returned if we find something, 0 if nothing was in the tree
1457 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1458 u64 *start, u64 *end, u64 max_bytes,
1459 struct extent_state **cached_state)
1461 struct rb_node *node;
1462 struct extent_state *state;
1463 u64 cur_start = *start;
1465 u64 total_bytes = 0;
1467 spin_lock(&tree->lock);
1470 * this search will find all the extents that end after
1473 node = tree_search(tree, cur_start);
1481 state = rb_entry(node, struct extent_state, rb_node);
1482 if (found && (state->start != cur_start ||
1483 (state->state & EXTENT_BOUNDARY))) {
1486 if (!(state->state & EXTENT_DELALLOC)) {
1492 *start = state->start;
1493 *cached_state = state;
1494 atomic_inc(&state->refs);
1498 cur_start = state->end + 1;
1499 node = rb_next(node);
1502 total_bytes += state->end - state->start + 1;
1503 if (total_bytes >= max_bytes)
1507 spin_unlock(&tree->lock);
1511 static noinline void __unlock_for_delalloc(struct inode *inode,
1512 struct page *locked_page,
1516 struct page *pages[16];
1517 unsigned long index = start >> PAGE_CACHE_SHIFT;
1518 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1519 unsigned long nr_pages = end_index - index + 1;
1522 if (index == locked_page->index && end_index == index)
1525 while (nr_pages > 0) {
1526 ret = find_get_pages_contig(inode->i_mapping, index,
1527 min_t(unsigned long, nr_pages,
1528 ARRAY_SIZE(pages)), pages);
1529 for (i = 0; i < ret; i++) {
1530 if (pages[i] != locked_page)
1531 unlock_page(pages[i]);
1532 page_cache_release(pages[i]);
1540 static noinline int lock_delalloc_pages(struct inode *inode,
1541 struct page *locked_page,
1545 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1546 unsigned long start_index = index;
1547 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1548 unsigned long pages_locked = 0;
1549 struct page *pages[16];
1550 unsigned long nrpages;
1554 /* the caller is responsible for locking the start index */
1555 if (index == locked_page->index && index == end_index)
1558 /* skip the page at the start index */
1559 nrpages = end_index - index + 1;
1560 while (nrpages > 0) {
1561 ret = find_get_pages_contig(inode->i_mapping, index,
1562 min_t(unsigned long,
1563 nrpages, ARRAY_SIZE(pages)), pages);
1568 /* now we have an array of pages, lock them all */
1569 for (i = 0; i < ret; i++) {
1571 * the caller is taking responsibility for
1574 if (pages[i] != locked_page) {
1575 lock_page(pages[i]);
1576 if (!PageDirty(pages[i]) ||
1577 pages[i]->mapping != inode->i_mapping) {
1579 unlock_page(pages[i]);
1580 page_cache_release(pages[i]);
1584 page_cache_release(pages[i]);
1593 if (ret && pages_locked) {
1594 __unlock_for_delalloc(inode, locked_page,
1596 ((u64)(start_index + pages_locked - 1)) <<
1603 * find a contiguous range of bytes in the file marked as delalloc, not
1604 * more than 'max_bytes'. start and end are used to return the range,
1606 * 1 is returned if we find something, 0 if nothing was in the tree
1608 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1609 struct extent_io_tree *tree,
1610 struct page *locked_page,
1611 u64 *start, u64 *end,
1617 struct extent_state *cached_state = NULL;
1622 /* step one, find a bunch of delalloc bytes starting at start */
1623 delalloc_start = *start;
1625 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1626 max_bytes, &cached_state);
1627 if (!found || delalloc_end <= *start) {
1628 *start = delalloc_start;
1629 *end = delalloc_end;
1630 free_extent_state(cached_state);
1635 * start comes from the offset of locked_page. We have to lock
1636 * pages in order, so we can't process delalloc bytes before
1639 if (delalloc_start < *start)
1640 delalloc_start = *start;
1643 * make sure to limit the number of pages we try to lock down
1646 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1647 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1649 /* step two, lock all the pages after the page that has start */
1650 ret = lock_delalloc_pages(inode, locked_page,
1651 delalloc_start, delalloc_end);
1652 if (ret == -EAGAIN) {
1653 /* some of the pages are gone, lets avoid looping by
1654 * shortening the size of the delalloc range we're searching
1656 free_extent_state(cached_state);
1658 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1659 max_bytes = PAGE_CACHE_SIZE - offset;
1667 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1669 /* step three, lock the state bits for the whole range */
1670 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1672 /* then test to make sure it is all still delalloc */
1673 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1674 EXTENT_DELALLOC, 1, cached_state);
1676 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1677 &cached_state, GFP_NOFS);
1678 __unlock_for_delalloc(inode, locked_page,
1679 delalloc_start, delalloc_end);
1683 free_extent_state(cached_state);
1684 *start = delalloc_start;
1685 *end = delalloc_end;
1690 int extent_clear_unlock_delalloc(struct inode *inode,
1691 struct extent_io_tree *tree,
1692 u64 start, u64 end, struct page *locked_page,
1696 struct page *pages[16];
1697 unsigned long index = start >> PAGE_CACHE_SHIFT;
1698 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1699 unsigned long nr_pages = end_index - index + 1;
1701 unsigned long clear_bits = 0;
1703 if (op & EXTENT_CLEAR_UNLOCK)
1704 clear_bits |= EXTENT_LOCKED;
1705 if (op & EXTENT_CLEAR_DIRTY)
1706 clear_bits |= EXTENT_DIRTY;
1708 if (op & EXTENT_CLEAR_DELALLOC)
1709 clear_bits |= EXTENT_DELALLOC;
1711 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1712 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1713 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1714 EXTENT_SET_PRIVATE2)))
1717 while (nr_pages > 0) {
1718 ret = find_get_pages_contig(inode->i_mapping, index,
1719 min_t(unsigned long,
1720 nr_pages, ARRAY_SIZE(pages)), pages);
1721 for (i = 0; i < ret; i++) {
1723 if (op & EXTENT_SET_PRIVATE2)
1724 SetPagePrivate2(pages[i]);
1726 if (pages[i] == locked_page) {
1727 page_cache_release(pages[i]);
1730 if (op & EXTENT_CLEAR_DIRTY)
1731 clear_page_dirty_for_io(pages[i]);
1732 if (op & EXTENT_SET_WRITEBACK)
1733 set_page_writeback(pages[i]);
1734 if (op & EXTENT_END_WRITEBACK)
1735 end_page_writeback(pages[i]);
1736 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1737 unlock_page(pages[i]);
1738 page_cache_release(pages[i]);
1748 * count the number of bytes in the tree that have a given bit(s)
1749 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1750 * cached. The total number found is returned.
1752 u64 count_range_bits(struct extent_io_tree *tree,
1753 u64 *start, u64 search_end, u64 max_bytes,
1754 unsigned long bits, int contig)
1756 struct rb_node *node;
1757 struct extent_state *state;
1758 u64 cur_start = *start;
1759 u64 total_bytes = 0;
1763 if (search_end <= cur_start) {
1768 spin_lock(&tree->lock);
1769 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1770 total_bytes = tree->dirty_bytes;
1774 * this search will find all the extents that end after
1777 node = tree_search(tree, cur_start);
1782 state = rb_entry(node, struct extent_state, rb_node);
1783 if (state->start > search_end)
1785 if (contig && found && state->start > last + 1)
1787 if (state->end >= cur_start && (state->state & bits) == bits) {
1788 total_bytes += min(search_end, state->end) + 1 -
1789 max(cur_start, state->start);
1790 if (total_bytes >= max_bytes)
1793 *start = max(cur_start, state->start);
1797 } else if (contig && found) {
1800 node = rb_next(node);
1805 spin_unlock(&tree->lock);
1810 * set the private field for a given byte offset in the tree. If there isn't
1811 * an extent_state there already, this does nothing.
1813 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1815 struct rb_node *node;
1816 struct extent_state *state;
1819 spin_lock(&tree->lock);
1821 * this search will find all the extents that end after
1824 node = tree_search(tree, start);
1829 state = rb_entry(node, struct extent_state, rb_node);
1830 if (state->start != start) {
1834 state->private = private;
1836 spin_unlock(&tree->lock);
1840 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1842 struct rb_node *node;
1843 struct extent_state *state;
1846 spin_lock(&tree->lock);
1848 * this search will find all the extents that end after
1851 node = tree_search(tree, start);
1856 state = rb_entry(node, struct extent_state, rb_node);
1857 if (state->start != start) {
1861 *private = state->private;
1863 spin_unlock(&tree->lock);
1868 * searches a range in the state tree for a given mask.
1869 * If 'filled' == 1, this returns 1 only if every extent in the tree
1870 * has the bits set. Otherwise, 1 is returned if any bit in the
1871 * range is found set.
1873 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1874 unsigned long bits, int filled, struct extent_state *cached)
1876 struct extent_state *state = NULL;
1877 struct rb_node *node;
1880 spin_lock(&tree->lock);
1881 if (cached && cached->tree && cached->start <= start &&
1882 cached->end > start)
1883 node = &cached->rb_node;
1885 node = tree_search(tree, start);
1886 while (node && start <= end) {
1887 state = rb_entry(node, struct extent_state, rb_node);
1889 if (filled && state->start > start) {
1894 if (state->start > end)
1897 if (state->state & bits) {
1901 } else if (filled) {
1906 if (state->end == (u64)-1)
1909 start = state->end + 1;
1912 node = rb_next(node);
1919 spin_unlock(&tree->lock);
1924 * helper function to set a given page up to date if all the
1925 * extents in the tree for that page are up to date
1927 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1929 u64 start = page_offset(page);
1930 u64 end = start + PAGE_CACHE_SIZE - 1;
1931 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1932 SetPageUptodate(page);
1936 * When IO fails, either with EIO or csum verification fails, we
1937 * try other mirrors that might have a good copy of the data. This
1938 * io_failure_record is used to record state as we go through all the
1939 * mirrors. If another mirror has good data, the page is set up to date
1940 * and things continue. If a good mirror can't be found, the original
1941 * bio end_io callback is called to indicate things have failed.
1943 struct io_failure_record {
1948 unsigned long bio_flags;
1954 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1959 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1961 set_state_private(failure_tree, rec->start, 0);
1962 ret = clear_extent_bits(failure_tree, rec->start,
1963 rec->start + rec->len - 1,
1964 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1968 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1969 rec->start + rec->len - 1,
1970 EXTENT_DAMAGED, GFP_NOFS);
1978 static void repair_io_failure_callback(struct bio *bio, int err)
1980 complete(bio->bi_private);
1984 * this bypasses the standard btrfs submit functions deliberately, as
1985 * the standard behavior is to write all copies in a raid setup. here we only
1986 * want to write the one bad copy. so we do the mapping for ourselves and issue
1987 * submit_bio directly.
1988 * to avoid any synchronization issues, wait for the data after writing, which
1989 * actually prevents the read that triggered the error from finishing.
1990 * currently, there can be no more than two copies of every data bit. thus,
1991 * exactly one rewrite is required.
1993 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
1994 u64 length, u64 logical, struct page *page,
1998 struct btrfs_device *dev;
1999 DECLARE_COMPLETION_ONSTACK(compl);
2002 struct btrfs_bio *bbio = NULL;
2003 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2006 BUG_ON(!mirror_num);
2008 /* we can't repair anything in raid56 yet */
2009 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2012 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2015 bio->bi_private = &compl;
2016 bio->bi_end_io = repair_io_failure_callback;
2018 map_length = length;
2020 ret = btrfs_map_block(fs_info, WRITE, logical,
2021 &map_length, &bbio, mirror_num);
2026 BUG_ON(mirror_num != bbio->mirror_num);
2027 sector = bbio->stripes[mirror_num-1].physical >> 9;
2028 bio->bi_sector = sector;
2029 dev = bbio->stripes[mirror_num-1].dev;
2031 if (!dev || !dev->bdev || !dev->writeable) {
2035 bio->bi_bdev = dev->bdev;
2036 bio_add_page(bio, page, length, start - page_offset(page));
2037 btrfsic_submit_bio(WRITE_SYNC, bio);
2038 wait_for_completion(&compl);
2040 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2041 /* try to remap that extent elsewhere? */
2043 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2047 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
2048 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2049 start, rcu_str_deref(dev->name), sector);
2055 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2058 u64 start = eb->start;
2059 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2062 for (i = 0; i < num_pages; i++) {
2063 struct page *p = extent_buffer_page(eb, i);
2064 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2065 start, p, mirror_num);
2068 start += PAGE_CACHE_SIZE;
2075 * each time an IO finishes, we do a fast check in the IO failure tree
2076 * to see if we need to process or clean up an io_failure_record
2078 static int clean_io_failure(u64 start, struct page *page)
2081 u64 private_failure;
2082 struct io_failure_record *failrec;
2083 struct btrfs_fs_info *fs_info;
2084 struct extent_state *state;
2088 struct inode *inode = page->mapping->host;
2091 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2092 (u64)-1, 1, EXTENT_DIRTY, 0);
2096 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2101 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2102 BUG_ON(!failrec->this_mirror);
2104 if (failrec->in_validation) {
2105 /* there was no real error, just free the record */
2106 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2112 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2113 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2116 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2118 if (state && state->start == failrec->start) {
2119 fs_info = BTRFS_I(inode)->root->fs_info;
2120 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2122 if (num_copies > 1) {
2123 ret = repair_io_failure(fs_info, start, failrec->len,
2124 failrec->logical, page,
2125 failrec->failed_mirror);
2133 ret = free_io_failure(inode, failrec, did_repair);
2139 * this is a generic handler for readpage errors (default
2140 * readpage_io_failed_hook). if other copies exist, read those and write back
2141 * good data to the failed position. does not investigate in remapping the
2142 * failed extent elsewhere, hoping the device will be smart enough to do this as
2146 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2147 struct page *page, u64 start, u64 end,
2150 struct io_failure_record *failrec = NULL;
2152 struct extent_map *em;
2153 struct inode *inode = page->mapping->host;
2154 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2155 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2156 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2158 struct btrfs_io_bio *btrfs_failed_bio;
2159 struct btrfs_io_bio *btrfs_bio;
2165 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2167 ret = get_state_private(failure_tree, start, &private);
2169 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2172 failrec->start = start;
2173 failrec->len = end - start + 1;
2174 failrec->this_mirror = 0;
2175 failrec->bio_flags = 0;
2176 failrec->in_validation = 0;
2178 read_lock(&em_tree->lock);
2179 em = lookup_extent_mapping(em_tree, start, failrec->len);
2181 read_unlock(&em_tree->lock);
2186 if (em->start > start || em->start + em->len < start) {
2187 free_extent_map(em);
2190 read_unlock(&em_tree->lock);
2196 logical = start - em->start;
2197 logical = em->block_start + logical;
2198 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2199 logical = em->block_start;
2200 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2201 extent_set_compress_type(&failrec->bio_flags,
2204 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2205 "len=%llu\n", logical, start, failrec->len);
2206 failrec->logical = logical;
2207 free_extent_map(em);
2209 /* set the bits in the private failure tree */
2210 ret = set_extent_bits(failure_tree, start, end,
2211 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2213 ret = set_state_private(failure_tree, start,
2214 (u64)(unsigned long)failrec);
2215 /* set the bits in the inode's tree */
2217 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2224 failrec = (struct io_failure_record *)(unsigned long)private;
2225 pr_debug("bio_readpage_error: (found) logical=%llu, "
2226 "start=%llu, len=%llu, validation=%d\n",
2227 failrec->logical, failrec->start, failrec->len,
2228 failrec->in_validation);
2230 * when data can be on disk more than twice, add to failrec here
2231 * (e.g. with a list for failed_mirror) to make
2232 * clean_io_failure() clean all those errors at once.
2235 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2236 failrec->logical, failrec->len);
2237 if (num_copies == 1) {
2239 * we only have a single copy of the data, so don't bother with
2240 * all the retry and error correction code that follows. no
2241 * matter what the error is, it is very likely to persist.
2243 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2244 num_copies, failrec->this_mirror, failed_mirror);
2245 free_io_failure(inode, failrec, 0);
2250 * there are two premises:
2251 * a) deliver good data to the caller
2252 * b) correct the bad sectors on disk
2254 if (failed_bio->bi_vcnt > 1) {
2256 * to fulfill b), we need to know the exact failing sectors, as
2257 * we don't want to rewrite any more than the failed ones. thus,
2258 * we need separate read requests for the failed bio
2260 * if the following BUG_ON triggers, our validation request got
2261 * merged. we need separate requests for our algorithm to work.
2263 BUG_ON(failrec->in_validation);
2264 failrec->in_validation = 1;
2265 failrec->this_mirror = failed_mirror;
2266 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2269 * we're ready to fulfill a) and b) alongside. get a good copy
2270 * of the failed sector and if we succeed, we have setup
2271 * everything for repair_io_failure to do the rest for us.
2273 if (failrec->in_validation) {
2274 BUG_ON(failrec->this_mirror != failed_mirror);
2275 failrec->in_validation = 0;
2276 failrec->this_mirror = 0;
2278 failrec->failed_mirror = failed_mirror;
2279 failrec->this_mirror++;
2280 if (failrec->this_mirror == failed_mirror)
2281 failrec->this_mirror++;
2282 read_mode = READ_SYNC;
2285 if (failrec->this_mirror > num_copies) {
2286 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2287 num_copies, failrec->this_mirror, failed_mirror);
2288 free_io_failure(inode, failrec, 0);
2292 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2294 free_io_failure(inode, failrec, 0);
2297 bio->bi_end_io = failed_bio->bi_end_io;
2298 bio->bi_sector = failrec->logical >> 9;
2299 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2302 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2303 if (btrfs_failed_bio->csum) {
2304 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2305 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2307 btrfs_bio = btrfs_io_bio(bio);
2308 btrfs_bio->csum = btrfs_bio->csum_inline;
2309 phy_offset >>= inode->i_sb->s_blocksize_bits;
2310 phy_offset *= csum_size;
2311 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset,
2315 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2317 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2318 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2319 failrec->this_mirror, num_copies, failrec->in_validation);
2321 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2322 failrec->this_mirror,
2323 failrec->bio_flags, 0);
2327 /* lots and lots of room for performance fixes in the end_bio funcs */
2329 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2331 int uptodate = (err == 0);
2332 struct extent_io_tree *tree;
2335 tree = &BTRFS_I(page->mapping->host)->io_tree;
2337 if (tree->ops && tree->ops->writepage_end_io_hook) {
2338 ret = tree->ops->writepage_end_io_hook(page, start,
2339 end, NULL, uptodate);
2345 ClearPageUptodate(page);
2352 * after a writepage IO is done, we need to:
2353 * clear the uptodate bits on error
2354 * clear the writeback bits in the extent tree for this IO
2355 * end_page_writeback if the page has no more pending IO
2357 * Scheduling is not allowed, so the extent state tree is expected
2358 * to have one and only one object corresponding to this IO.
2360 static void end_bio_extent_writepage(struct bio *bio, int err)
2362 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2363 struct extent_io_tree *tree;
2368 struct page *page = bvec->bv_page;
2369 tree = &BTRFS_I(page->mapping->host)->io_tree;
2371 /* We always issue full-page reads, but if some block
2372 * in a page fails to read, blk_update_request() will
2373 * advance bv_offset and adjust bv_len to compensate.
2374 * Print a warning for nonzero offsets, and an error
2375 * if they don't add up to a full page. */
2376 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2377 printk("%s page write in btrfs with offset %u and length %u\n",
2378 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2379 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2380 bvec->bv_offset, bvec->bv_len);
2382 start = page_offset(page);
2383 end = start + bvec->bv_offset + bvec->bv_len - 1;
2385 if (--bvec >= bio->bi_io_vec)
2386 prefetchw(&bvec->bv_page->flags);
2388 if (end_extent_writepage(page, err, start, end))
2391 end_page_writeback(page);
2392 } while (bvec >= bio->bi_io_vec);
2398 * after a readpage IO is done, we need to:
2399 * clear the uptodate bits on error
2400 * set the uptodate bits if things worked
2401 * set the page up to date if all extents in the tree are uptodate
2402 * clear the lock bit in the extent tree
2403 * unlock the page if there are no other extents locked for it
2405 * Scheduling is not allowed, so the extent state tree is expected
2406 * to have one and only one object corresponding to this IO.
2408 static void end_bio_extent_readpage(struct bio *bio, int err)
2410 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2411 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2412 struct bio_vec *bvec = bio->bi_io_vec;
2413 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2414 struct extent_io_tree *tree;
2426 struct page *page = bvec->bv_page;
2427 struct extent_state *cached = NULL;
2428 struct extent_state *state;
2429 struct inode *inode = page->mapping->host;
2431 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2432 "mirror=%lu\n", (u64)bio->bi_sector, err,
2433 io_bio->mirror_num);
2434 tree = &BTRFS_I(inode)->io_tree;
2436 /* We always issue full-page reads, but if some block
2437 * in a page fails to read, blk_update_request() will
2438 * advance bv_offset and adjust bv_len to compensate.
2439 * Print a warning for nonzero offsets, and an error
2440 * if they don't add up to a full page. */
2441 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2442 printk("%s page read in btrfs with offset %u and length %u\n",
2443 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2444 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2445 bvec->bv_offset, bvec->bv_len);
2447 start = page_offset(page);
2448 end = start + bvec->bv_offset + bvec->bv_len - 1;
2451 if (++bvec <= bvec_end)
2452 prefetchw(&bvec->bv_page->flags);
2454 spin_lock(&tree->lock);
2455 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2456 if (likely(state && state->start == start)) {
2458 * take a reference on the state, unlock will drop
2461 cache_state(state, &cached);
2463 spin_unlock(&tree->lock);
2465 mirror = io_bio->mirror_num;
2466 if (likely(uptodate && tree->ops &&
2467 tree->ops->readpage_end_io_hook)) {
2468 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2474 clean_io_failure(start, page);
2477 if (likely(uptodate))
2480 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2481 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2483 test_bit(BIO_UPTODATE, &bio->bi_flags))
2487 * The generic bio_readpage_error handles errors the
2488 * following way: If possible, new read requests are
2489 * created and submitted and will end up in
2490 * end_bio_extent_readpage as well (if we're lucky, not
2491 * in the !uptodate case). In that case it returns 0 and
2492 * we just go on with the next page in our bio. If it
2493 * can't handle the error it will return -EIO and we
2494 * remain responsible for that page.
2496 ret = bio_readpage_error(bio, offset, page, start, end,
2500 test_bit(BIO_UPTODATE, &bio->bi_flags);
2503 uncache_state(&cached);
2508 if (uptodate && tree->track_uptodate) {
2509 set_extent_uptodate(tree, start, end, &cached,
2512 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2515 loff_t i_size = i_size_read(inode);
2516 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2519 /* Zero out the end if this page straddles i_size */
2520 offset = i_size & (PAGE_CACHE_SIZE-1);
2521 if (page->index == end_index && offset)
2522 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2523 SetPageUptodate(page);
2525 ClearPageUptodate(page);
2530 } while (bvec <= bvec_end);
2533 io_bio->end_io(io_bio, err);
2538 * this allocates from the btrfs_bioset. We're returning a bio right now
2539 * but you can call btrfs_io_bio for the appropriate container_of magic
2542 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2545 struct btrfs_io_bio *btrfs_bio;
2548 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2550 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2551 while (!bio && (nr_vecs /= 2)) {
2552 bio = bio_alloc_bioset(gfp_flags,
2553 nr_vecs, btrfs_bioset);
2559 bio->bi_bdev = bdev;
2560 bio->bi_sector = first_sector;
2561 btrfs_bio = btrfs_io_bio(bio);
2562 btrfs_bio->csum = NULL;
2563 btrfs_bio->csum_allocated = NULL;
2564 btrfs_bio->end_io = NULL;
2569 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2571 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2575 /* this also allocates from the btrfs_bioset */
2576 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2578 struct btrfs_io_bio *btrfs_bio;
2581 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2583 btrfs_bio = btrfs_io_bio(bio);
2584 btrfs_bio->csum = NULL;
2585 btrfs_bio->csum_allocated = NULL;
2586 btrfs_bio->end_io = NULL;
2592 static int __must_check submit_one_bio(int rw, struct bio *bio,
2593 int mirror_num, unsigned long bio_flags)
2596 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2597 struct page *page = bvec->bv_page;
2598 struct extent_io_tree *tree = bio->bi_private;
2601 start = page_offset(page) + bvec->bv_offset;
2603 bio->bi_private = NULL;
2607 if (tree->ops && tree->ops->submit_bio_hook)
2608 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2609 mirror_num, bio_flags, start);
2611 btrfsic_submit_bio(rw, bio);
2613 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2619 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2620 unsigned long offset, size_t size, struct bio *bio,
2621 unsigned long bio_flags)
2624 if (tree->ops && tree->ops->merge_bio_hook)
2625 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2632 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2633 struct page *page, sector_t sector,
2634 size_t size, unsigned long offset,
2635 struct block_device *bdev,
2636 struct bio **bio_ret,
2637 unsigned long max_pages,
2638 bio_end_io_t end_io_func,
2640 unsigned long prev_bio_flags,
2641 unsigned long bio_flags)
2647 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2648 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2649 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2651 if (bio_ret && *bio_ret) {
2654 contig = bio->bi_sector == sector;
2656 contig = bio_end_sector(bio) == sector;
2658 if (prev_bio_flags != bio_flags || !contig ||
2659 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2660 bio_add_page(bio, page, page_size, offset) < page_size) {
2661 ret = submit_one_bio(rw, bio, mirror_num,
2670 if (this_compressed)
2673 nr = bio_get_nr_vecs(bdev);
2675 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2679 bio_add_page(bio, page, page_size, offset);
2680 bio->bi_end_io = end_io_func;
2681 bio->bi_private = tree;
2686 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2691 static void attach_extent_buffer_page(struct extent_buffer *eb,
2694 if (!PagePrivate(page)) {
2695 SetPagePrivate(page);
2696 page_cache_get(page);
2697 set_page_private(page, (unsigned long)eb);
2699 WARN_ON(page->private != (unsigned long)eb);
2703 void set_page_extent_mapped(struct page *page)
2705 if (!PagePrivate(page)) {
2706 SetPagePrivate(page);
2707 page_cache_get(page);
2708 set_page_private(page, EXTENT_PAGE_PRIVATE);
2713 * basic readpage implementation. Locked extent state structs are inserted
2714 * into the tree that are removed when the IO is done (by the end_io
2716 * XXX JDM: This needs looking at to ensure proper page locking
2718 static int __extent_read_full_page(struct extent_io_tree *tree,
2720 get_extent_t *get_extent,
2721 struct bio **bio, int mirror_num,
2722 unsigned long *bio_flags, int rw)
2724 struct inode *inode = page->mapping->host;
2725 u64 start = page_offset(page);
2726 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2730 u64 last_byte = i_size_read(inode);
2734 struct extent_map *em;
2735 struct block_device *bdev;
2736 struct btrfs_ordered_extent *ordered;
2739 size_t pg_offset = 0;
2741 size_t disk_io_size;
2742 size_t blocksize = inode->i_sb->s_blocksize;
2743 unsigned long this_bio_flag = 0;
2745 set_page_extent_mapped(page);
2747 if (!PageUptodate(page)) {
2748 if (cleancache_get_page(page) == 0) {
2749 BUG_ON(blocksize != PAGE_SIZE);
2756 lock_extent(tree, start, end);
2757 ordered = btrfs_lookup_ordered_extent(inode, start);
2760 unlock_extent(tree, start, end);
2761 btrfs_start_ordered_extent(inode, ordered, 1);
2762 btrfs_put_ordered_extent(ordered);
2765 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2767 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2770 iosize = PAGE_CACHE_SIZE - zero_offset;
2771 userpage = kmap_atomic(page);
2772 memset(userpage + zero_offset, 0, iosize);
2773 flush_dcache_page(page);
2774 kunmap_atomic(userpage);
2777 while (cur <= end) {
2778 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2780 if (cur >= last_byte) {
2782 struct extent_state *cached = NULL;
2784 iosize = PAGE_CACHE_SIZE - pg_offset;
2785 userpage = kmap_atomic(page);
2786 memset(userpage + pg_offset, 0, iosize);
2787 flush_dcache_page(page);
2788 kunmap_atomic(userpage);
2789 set_extent_uptodate(tree, cur, cur + iosize - 1,
2791 unlock_extent_cached(tree, cur, cur + iosize - 1,
2795 em = get_extent(inode, page, pg_offset, cur,
2797 if (IS_ERR_OR_NULL(em)) {
2799 unlock_extent(tree, cur, end);
2802 extent_offset = cur - em->start;
2803 BUG_ON(extent_map_end(em) <= cur);
2806 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2807 this_bio_flag = EXTENT_BIO_COMPRESSED;
2808 extent_set_compress_type(&this_bio_flag,
2812 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2813 cur_end = min(extent_map_end(em) - 1, end);
2814 iosize = ALIGN(iosize, blocksize);
2815 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2816 disk_io_size = em->block_len;
2817 sector = em->block_start >> 9;
2819 sector = (em->block_start + extent_offset) >> 9;
2820 disk_io_size = iosize;
2823 block_start = em->block_start;
2824 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2825 block_start = EXTENT_MAP_HOLE;
2826 free_extent_map(em);
2829 /* we've found a hole, just zero and go on */
2830 if (block_start == EXTENT_MAP_HOLE) {
2832 struct extent_state *cached = NULL;
2834 userpage = kmap_atomic(page);
2835 memset(userpage + pg_offset, 0, iosize);
2836 flush_dcache_page(page);
2837 kunmap_atomic(userpage);
2839 set_extent_uptodate(tree, cur, cur + iosize - 1,
2841 unlock_extent_cached(tree, cur, cur + iosize - 1,
2844 pg_offset += iosize;
2847 /* the get_extent function already copied into the page */
2848 if (test_range_bit(tree, cur, cur_end,
2849 EXTENT_UPTODATE, 1, NULL)) {
2850 check_page_uptodate(tree, page);
2851 unlock_extent(tree, cur, cur + iosize - 1);
2853 pg_offset += iosize;
2856 /* we have an inline extent but it didn't get marked up
2857 * to date. Error out
2859 if (block_start == EXTENT_MAP_INLINE) {
2861 unlock_extent(tree, cur, cur + iosize - 1);
2863 pg_offset += iosize;
2868 ret = submit_extent_page(rw, tree, page,
2869 sector, disk_io_size, pg_offset,
2871 end_bio_extent_readpage, mirror_num,
2876 *bio_flags = this_bio_flag;
2879 unlock_extent(tree, cur, cur + iosize - 1);
2882 pg_offset += iosize;
2886 if (!PageError(page))
2887 SetPageUptodate(page);
2893 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2894 get_extent_t *get_extent, int mirror_num)
2896 struct bio *bio = NULL;
2897 unsigned long bio_flags = 0;
2900 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2903 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2907 static noinline void update_nr_written(struct page *page,
2908 struct writeback_control *wbc,
2909 unsigned long nr_written)
2911 wbc->nr_to_write -= nr_written;
2912 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2913 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2914 page->mapping->writeback_index = page->index + nr_written;
2918 * the writepage semantics are similar to regular writepage. extent
2919 * records are inserted to lock ranges in the tree, and as dirty areas
2920 * are found, they are marked writeback. Then the lock bits are removed
2921 * and the end_io handler clears the writeback ranges
2923 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2926 struct inode *inode = page->mapping->host;
2927 struct extent_page_data *epd = data;
2928 struct extent_io_tree *tree = epd->tree;
2929 u64 start = page_offset(page);
2931 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2935 u64 last_byte = i_size_read(inode);
2939 struct extent_state *cached_state = NULL;
2940 struct extent_map *em;
2941 struct block_device *bdev;
2944 size_t pg_offset = 0;
2946 loff_t i_size = i_size_read(inode);
2947 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2953 unsigned long nr_written = 0;
2954 bool fill_delalloc = true;
2956 if (wbc->sync_mode == WB_SYNC_ALL)
2957 write_flags = WRITE_SYNC;
2959 write_flags = WRITE;
2961 trace___extent_writepage(page, inode, wbc);
2963 WARN_ON(!PageLocked(page));
2965 ClearPageError(page);
2967 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2968 if (page->index > end_index ||
2969 (page->index == end_index && !pg_offset)) {
2970 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
2975 if (page->index == end_index) {
2978 userpage = kmap_atomic(page);
2979 memset(userpage + pg_offset, 0,
2980 PAGE_CACHE_SIZE - pg_offset);
2981 kunmap_atomic(userpage);
2982 flush_dcache_page(page);
2986 set_page_extent_mapped(page);
2988 if (!tree->ops || !tree->ops->fill_delalloc)
2989 fill_delalloc = false;
2991 delalloc_start = start;
2994 if (!epd->extent_locked && fill_delalloc) {
2995 u64 delalloc_to_write = 0;
2997 * make sure the wbc mapping index is at least updated
3000 update_nr_written(page, wbc, 0);
3002 while (delalloc_end < page_end) {
3003 nr_delalloc = find_lock_delalloc_range(inode, tree,
3008 if (nr_delalloc == 0) {
3009 delalloc_start = delalloc_end + 1;
3012 ret = tree->ops->fill_delalloc(inode, page,
3017 /* File system has been set read-only */
3023 * delalloc_end is already one less than the total
3024 * length, so we don't subtract one from
3027 delalloc_to_write += (delalloc_end - delalloc_start +
3030 delalloc_start = delalloc_end + 1;
3032 if (wbc->nr_to_write < delalloc_to_write) {
3035 if (delalloc_to_write < thresh * 2)
3036 thresh = delalloc_to_write;
3037 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3041 /* did the fill delalloc function already unlock and start
3047 * we've unlocked the page, so we can't update
3048 * the mapping's writeback index, just update
3051 wbc->nr_to_write -= nr_written;
3055 if (tree->ops && tree->ops->writepage_start_hook) {
3056 ret = tree->ops->writepage_start_hook(page, start,
3059 /* Fixup worker will requeue */
3061 wbc->pages_skipped++;
3063 redirty_page_for_writepage(wbc, page);
3064 update_nr_written(page, wbc, nr_written);
3072 * we don't want to touch the inode after unlocking the page,
3073 * so we update the mapping writeback index now
3075 update_nr_written(page, wbc, nr_written + 1);
3078 if (last_byte <= start) {
3079 if (tree->ops && tree->ops->writepage_end_io_hook)
3080 tree->ops->writepage_end_io_hook(page, start,
3085 blocksize = inode->i_sb->s_blocksize;
3087 while (cur <= end) {
3088 if (cur >= last_byte) {
3089 if (tree->ops && tree->ops->writepage_end_io_hook)
3090 tree->ops->writepage_end_io_hook(page, cur,
3094 em = epd->get_extent(inode, page, pg_offset, cur,
3096 if (IS_ERR_OR_NULL(em)) {
3101 extent_offset = cur - em->start;
3102 BUG_ON(extent_map_end(em) <= cur);
3104 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3105 iosize = ALIGN(iosize, blocksize);
3106 sector = (em->block_start + extent_offset) >> 9;
3108 block_start = em->block_start;
3109 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3110 free_extent_map(em);
3114 * compressed and inline extents are written through other
3117 if (compressed || block_start == EXTENT_MAP_HOLE ||
3118 block_start == EXTENT_MAP_INLINE) {
3120 * end_io notification does not happen here for
3121 * compressed extents
3123 if (!compressed && tree->ops &&
3124 tree->ops->writepage_end_io_hook)
3125 tree->ops->writepage_end_io_hook(page, cur,
3128 else if (compressed) {
3129 /* we don't want to end_page_writeback on
3130 * a compressed extent. this happens
3137 pg_offset += iosize;
3140 /* leave this out until we have a page_mkwrite call */
3141 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3142 EXTENT_DIRTY, 0, NULL)) {
3144 pg_offset += iosize;
3148 if (tree->ops && tree->ops->writepage_io_hook) {
3149 ret = tree->ops->writepage_io_hook(page, cur,
3157 unsigned long max_nr = end_index + 1;
3159 set_range_writeback(tree, cur, cur + iosize - 1);
3160 if (!PageWriteback(page)) {
3161 printk(KERN_ERR "btrfs warning page %lu not "
3162 "writeback, cur %llu end %llu\n",
3163 page->index, (unsigned long long)cur,
3164 (unsigned long long)end);
3167 ret = submit_extent_page(write_flags, tree, page,
3168 sector, iosize, pg_offset,
3169 bdev, &epd->bio, max_nr,
3170 end_bio_extent_writepage,
3176 pg_offset += iosize;
3181 /* make sure the mapping tag for page dirty gets cleared */
3182 set_page_writeback(page);
3183 end_page_writeback(page);
3189 /* drop our reference on any cached states */
3190 free_extent_state(cached_state);
3194 static int eb_wait(void *word)
3200 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3202 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3203 TASK_UNINTERRUPTIBLE);
3206 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3207 struct btrfs_fs_info *fs_info,
3208 struct extent_page_data *epd)
3210 unsigned long i, num_pages;
3214 if (!btrfs_try_tree_write_lock(eb)) {
3216 flush_write_bio(epd);
3217 btrfs_tree_lock(eb);
3220 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3221 btrfs_tree_unlock(eb);
3225 flush_write_bio(epd);
3229 wait_on_extent_buffer_writeback(eb);
3230 btrfs_tree_lock(eb);
3231 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3233 btrfs_tree_unlock(eb);
3238 * We need to do this to prevent races in people who check if the eb is
3239 * under IO since we can end up having no IO bits set for a short period
3242 spin_lock(&eb->refs_lock);
3243 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3244 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3245 spin_unlock(&eb->refs_lock);
3246 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3247 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3249 fs_info->dirty_metadata_batch);
3252 spin_unlock(&eb->refs_lock);
3255 btrfs_tree_unlock(eb);
3260 num_pages = num_extent_pages(eb->start, eb->len);
3261 for (i = 0; i < num_pages; i++) {
3262 struct page *p = extent_buffer_page(eb, i);
3264 if (!trylock_page(p)) {
3266 flush_write_bio(epd);
3276 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3278 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3279 smp_mb__after_clear_bit();
3280 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3283 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3285 int uptodate = err == 0;
3286 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3287 struct extent_buffer *eb;
3291 struct page *page = bvec->bv_page;
3294 eb = (struct extent_buffer *)page->private;
3296 done = atomic_dec_and_test(&eb->io_pages);
3298 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3299 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3300 ClearPageUptodate(page);
3304 end_page_writeback(page);
3309 end_extent_buffer_writeback(eb);
3310 } while (bvec >= bio->bi_io_vec);
3316 static int write_one_eb(struct extent_buffer *eb,
3317 struct btrfs_fs_info *fs_info,
3318 struct writeback_control *wbc,
3319 struct extent_page_data *epd)
3321 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3322 u64 offset = eb->start;
3323 unsigned long i, num_pages;
3324 unsigned long bio_flags = 0;
3325 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3328 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3329 num_pages = num_extent_pages(eb->start, eb->len);
3330 atomic_set(&eb->io_pages, num_pages);
3331 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3332 bio_flags = EXTENT_BIO_TREE_LOG;
3334 for (i = 0; i < num_pages; i++) {
3335 struct page *p = extent_buffer_page(eb, i);
3337 clear_page_dirty_for_io(p);
3338 set_page_writeback(p);
3339 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3340 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3341 -1, end_bio_extent_buffer_writepage,
3342 0, epd->bio_flags, bio_flags);
3343 epd->bio_flags = bio_flags;
3345 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3347 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3348 end_extent_buffer_writeback(eb);
3352 offset += PAGE_CACHE_SIZE;
3353 update_nr_written(p, wbc, 1);
3357 if (unlikely(ret)) {
3358 for (; i < num_pages; i++) {
3359 struct page *p = extent_buffer_page(eb, i);
3367 int btree_write_cache_pages(struct address_space *mapping,
3368 struct writeback_control *wbc)
3370 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3371 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3372 struct extent_buffer *eb, *prev_eb = NULL;
3373 struct extent_page_data epd = {
3377 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3382 int nr_to_write_done = 0;
3383 struct pagevec pvec;
3386 pgoff_t end; /* Inclusive */
3390 pagevec_init(&pvec, 0);
3391 if (wbc->range_cyclic) {
3392 index = mapping->writeback_index; /* Start from prev offset */
3395 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3396 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3399 if (wbc->sync_mode == WB_SYNC_ALL)
3400 tag = PAGECACHE_TAG_TOWRITE;
3402 tag = PAGECACHE_TAG_DIRTY;
3404 if (wbc->sync_mode == WB_SYNC_ALL)
3405 tag_pages_for_writeback(mapping, index, end);
3406 while (!done && !nr_to_write_done && (index <= end) &&
3407 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3408 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3412 for (i = 0; i < nr_pages; i++) {
3413 struct page *page = pvec.pages[i];
3415 if (!PagePrivate(page))
3418 if (!wbc->range_cyclic && page->index > end) {
3423 spin_lock(&mapping->private_lock);
3424 if (!PagePrivate(page)) {
3425 spin_unlock(&mapping->private_lock);
3429 eb = (struct extent_buffer *)page->private;
3432 * Shouldn't happen and normally this would be a BUG_ON
3433 * but no sense in crashing the users box for something
3434 * we can survive anyway.
3437 spin_unlock(&mapping->private_lock);
3442 if (eb == prev_eb) {
3443 spin_unlock(&mapping->private_lock);
3447 ret = atomic_inc_not_zero(&eb->refs);
3448 spin_unlock(&mapping->private_lock);
3453 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3455 free_extent_buffer(eb);
3459 ret = write_one_eb(eb, fs_info, wbc, &epd);
3462 free_extent_buffer(eb);
3465 free_extent_buffer(eb);
3468 * the filesystem may choose to bump up nr_to_write.
3469 * We have to make sure to honor the new nr_to_write
3472 nr_to_write_done = wbc->nr_to_write <= 0;
3474 pagevec_release(&pvec);
3477 if (!scanned && !done) {
3479 * We hit the last page and there is more work to be done: wrap
3480 * back to the start of the file
3486 flush_write_bio(&epd);
3491 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3492 * @mapping: address space structure to write
3493 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3494 * @writepage: function called for each page
3495 * @data: data passed to writepage function
3497 * If a page is already under I/O, write_cache_pages() skips it, even
3498 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3499 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3500 * and msync() need to guarantee that all the data which was dirty at the time
3501 * the call was made get new I/O started against them. If wbc->sync_mode is
3502 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3503 * existing IO to complete.
3505 static int extent_write_cache_pages(struct extent_io_tree *tree,
3506 struct address_space *mapping,
3507 struct writeback_control *wbc,
3508 writepage_t writepage, void *data,
3509 void (*flush_fn)(void *))
3511 struct inode *inode = mapping->host;
3514 int nr_to_write_done = 0;
3515 struct pagevec pvec;
3518 pgoff_t end; /* Inclusive */
3523 * We have to hold onto the inode so that ordered extents can do their
3524 * work when the IO finishes. The alternative to this is failing to add
3525 * an ordered extent if the igrab() fails there and that is a huge pain
3526 * to deal with, so instead just hold onto the inode throughout the
3527 * writepages operation. If it fails here we are freeing up the inode
3528 * anyway and we'd rather not waste our time writing out stuff that is
3529 * going to be truncated anyway.
3534 pagevec_init(&pvec, 0);
3535 if (wbc->range_cyclic) {
3536 index = mapping->writeback_index; /* Start from prev offset */
3539 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3540 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3543 if (wbc->sync_mode == WB_SYNC_ALL)
3544 tag = PAGECACHE_TAG_TOWRITE;
3546 tag = PAGECACHE_TAG_DIRTY;
3548 if (wbc->sync_mode == WB_SYNC_ALL)
3549 tag_pages_for_writeback(mapping, index, end);
3550 while (!done && !nr_to_write_done && (index <= end) &&
3551 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3552 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3556 for (i = 0; i < nr_pages; i++) {
3557 struct page *page = pvec.pages[i];
3560 * At this point we hold neither mapping->tree_lock nor
3561 * lock on the page itself: the page may be truncated or
3562 * invalidated (changing page->mapping to NULL), or even
3563 * swizzled back from swapper_space to tmpfs file
3566 if (!trylock_page(page)) {
3571 if (unlikely(page->mapping != mapping)) {
3576 if (!wbc->range_cyclic && page->index > end) {
3582 if (wbc->sync_mode != WB_SYNC_NONE) {
3583 if (PageWriteback(page))
3585 wait_on_page_writeback(page);
3588 if (PageWriteback(page) ||
3589 !clear_page_dirty_for_io(page)) {
3594 ret = (*writepage)(page, wbc, data);
3596 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3604 * the filesystem may choose to bump up nr_to_write.
3605 * We have to make sure to honor the new nr_to_write
3608 nr_to_write_done = wbc->nr_to_write <= 0;
3610 pagevec_release(&pvec);
3613 if (!scanned && !done) {
3615 * We hit the last page and there is more work to be done: wrap
3616 * back to the start of the file
3622 btrfs_add_delayed_iput(inode);
3626 static void flush_epd_write_bio(struct extent_page_data *epd)
3635 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3636 BUG_ON(ret < 0); /* -ENOMEM */
3641 static noinline void flush_write_bio(void *data)
3643 struct extent_page_data *epd = data;
3644 flush_epd_write_bio(epd);
3647 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3648 get_extent_t *get_extent,
3649 struct writeback_control *wbc)
3652 struct extent_page_data epd = {
3655 .get_extent = get_extent,
3657 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3661 ret = __extent_writepage(page, wbc, &epd);
3663 flush_epd_write_bio(&epd);
3667 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3668 u64 start, u64 end, get_extent_t *get_extent,
3672 struct address_space *mapping = inode->i_mapping;
3674 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3677 struct extent_page_data epd = {
3680 .get_extent = get_extent,
3682 .sync_io = mode == WB_SYNC_ALL,
3685 struct writeback_control wbc_writepages = {
3687 .nr_to_write = nr_pages * 2,
3688 .range_start = start,
3689 .range_end = end + 1,
3692 while (start <= end) {
3693 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3694 if (clear_page_dirty_for_io(page))
3695 ret = __extent_writepage(page, &wbc_writepages, &epd);
3697 if (tree->ops && tree->ops->writepage_end_io_hook)
3698 tree->ops->writepage_end_io_hook(page, start,
3699 start + PAGE_CACHE_SIZE - 1,
3703 page_cache_release(page);
3704 start += PAGE_CACHE_SIZE;
3707 flush_epd_write_bio(&epd);
3711 int extent_writepages(struct extent_io_tree *tree,
3712 struct address_space *mapping,
3713 get_extent_t *get_extent,
3714 struct writeback_control *wbc)
3717 struct extent_page_data epd = {
3720 .get_extent = get_extent,
3722 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3726 ret = extent_write_cache_pages(tree, mapping, wbc,
3727 __extent_writepage, &epd,
3729 flush_epd_write_bio(&epd);
3733 int extent_readpages(struct extent_io_tree *tree,
3734 struct address_space *mapping,
3735 struct list_head *pages, unsigned nr_pages,
3736 get_extent_t get_extent)
3738 struct bio *bio = NULL;
3740 unsigned long bio_flags = 0;
3741 struct page *pagepool[16];
3746 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3747 page = list_entry(pages->prev, struct page, lru);
3749 prefetchw(&page->flags);
3750 list_del(&page->lru);
3751 if (add_to_page_cache_lru(page, mapping,
3752 page->index, GFP_NOFS)) {
3753 page_cache_release(page);
3757 pagepool[nr++] = page;
3758 if (nr < ARRAY_SIZE(pagepool))
3760 for (i = 0; i < nr; i++) {
3761 __extent_read_full_page(tree, pagepool[i], get_extent,
3762 &bio, 0, &bio_flags, READ);
3763 page_cache_release(pagepool[i]);
3767 for (i = 0; i < nr; i++) {
3768 __extent_read_full_page(tree, pagepool[i], get_extent,
3769 &bio, 0, &bio_flags, READ);
3770 page_cache_release(pagepool[i]);
3773 BUG_ON(!list_empty(pages));
3775 return submit_one_bio(READ, bio, 0, bio_flags);
3780 * basic invalidatepage code, this waits on any locked or writeback
3781 * ranges corresponding to the page, and then deletes any extent state
3782 * records from the tree
3784 int extent_invalidatepage(struct extent_io_tree *tree,
3785 struct page *page, unsigned long offset)
3787 struct extent_state *cached_state = NULL;
3788 u64 start = page_offset(page);
3789 u64 end = start + PAGE_CACHE_SIZE - 1;
3790 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3792 start += ALIGN(offset, blocksize);
3796 lock_extent_bits(tree, start, end, 0, &cached_state);
3797 wait_on_page_writeback(page);
3798 clear_extent_bit(tree, start, end,
3799 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3800 EXTENT_DO_ACCOUNTING,
3801 1, 1, &cached_state, GFP_NOFS);
3806 * a helper for releasepage, this tests for areas of the page that
3807 * are locked or under IO and drops the related state bits if it is safe
3810 static int try_release_extent_state(struct extent_map_tree *map,
3811 struct extent_io_tree *tree,
3812 struct page *page, gfp_t mask)
3814 u64 start = page_offset(page);
3815 u64 end = start + PAGE_CACHE_SIZE - 1;
3818 if (test_range_bit(tree, start, end,
3819 EXTENT_IOBITS, 0, NULL))
3822 if ((mask & GFP_NOFS) == GFP_NOFS)
3825 * at this point we can safely clear everything except the
3826 * locked bit and the nodatasum bit
3828 ret = clear_extent_bit(tree, start, end,
3829 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3832 /* if clear_extent_bit failed for enomem reasons,
3833 * we can't allow the release to continue.
3844 * a helper for releasepage. As long as there are no locked extents
3845 * in the range corresponding to the page, both state records and extent
3846 * map records are removed
3848 int try_release_extent_mapping(struct extent_map_tree *map,
3849 struct extent_io_tree *tree, struct page *page,
3852 struct extent_map *em;
3853 u64 start = page_offset(page);
3854 u64 end = start + PAGE_CACHE_SIZE - 1;
3856 if ((mask & __GFP_WAIT) &&
3857 page->mapping->host->i_size > 16 * 1024 * 1024) {
3859 while (start <= end) {
3860 len = end - start + 1;
3861 write_lock(&map->lock);
3862 em = lookup_extent_mapping(map, start, len);
3864 write_unlock(&map->lock);
3867 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3868 em->start != start) {
3869 write_unlock(&map->lock);
3870 free_extent_map(em);
3873 if (!test_range_bit(tree, em->start,
3874 extent_map_end(em) - 1,
3875 EXTENT_LOCKED | EXTENT_WRITEBACK,
3877 remove_extent_mapping(map, em);
3878 /* once for the rb tree */
3879 free_extent_map(em);
3881 start = extent_map_end(em);
3882 write_unlock(&map->lock);
3885 free_extent_map(em);
3888 return try_release_extent_state(map, tree, page, mask);
3892 * helper function for fiemap, which doesn't want to see any holes.
3893 * This maps until we find something past 'last'
3895 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3898 get_extent_t *get_extent)
3900 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3901 struct extent_map *em;
3908 len = last - offset;
3911 len = ALIGN(len, sectorsize);
3912 em = get_extent(inode, NULL, 0, offset, len, 0);
3913 if (IS_ERR_OR_NULL(em))
3916 /* if this isn't a hole return it */
3917 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3918 em->block_start != EXTENT_MAP_HOLE) {
3922 /* this is a hole, advance to the next extent */
3923 offset = extent_map_end(em);
3924 free_extent_map(em);
3931 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3932 __u64 start, __u64 len, get_extent_t *get_extent)
3936 u64 max = start + len;
3940 u64 last_for_get_extent = 0;
3942 u64 isize = i_size_read(inode);
3943 struct btrfs_key found_key;
3944 struct extent_map *em = NULL;
3945 struct extent_state *cached_state = NULL;
3946 struct btrfs_path *path;
3947 struct btrfs_file_extent_item *item;
3952 unsigned long emflags;
3957 path = btrfs_alloc_path();
3960 path->leave_spinning = 1;
3962 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3963 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3966 * lookup the last file extent. We're not using i_size here
3967 * because there might be preallocation past i_size
3969 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3970 path, btrfs_ino(inode), -1, 0);
3972 btrfs_free_path(path);
3977 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3978 struct btrfs_file_extent_item);
3979 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3980 found_type = btrfs_key_type(&found_key);
3982 /* No extents, but there might be delalloc bits */
3983 if (found_key.objectid != btrfs_ino(inode) ||
3984 found_type != BTRFS_EXTENT_DATA_KEY) {
3985 /* have to trust i_size as the end */
3987 last_for_get_extent = isize;
3990 * remember the start of the last extent. There are a
3991 * bunch of different factors that go into the length of the
3992 * extent, so its much less complex to remember where it started
3994 last = found_key.offset;
3995 last_for_get_extent = last + 1;
3997 btrfs_free_path(path);
4000 * we might have some extents allocated but more delalloc past those
4001 * extents. so, we trust isize unless the start of the last extent is
4006 last_for_get_extent = isize;
4009 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4012 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4022 u64 offset_in_extent = 0;
4024 /* break if the extent we found is outside the range */
4025 if (em->start >= max || extent_map_end(em) < off)
4029 * get_extent may return an extent that starts before our
4030 * requested range. We have to make sure the ranges
4031 * we return to fiemap always move forward and don't
4032 * overlap, so adjust the offsets here
4034 em_start = max(em->start, off);
4037 * record the offset from the start of the extent
4038 * for adjusting the disk offset below. Only do this if the
4039 * extent isn't compressed since our in ram offset may be past
4040 * what we have actually allocated on disk.
4042 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4043 offset_in_extent = em_start - em->start;
4044 em_end = extent_map_end(em);
4045 em_len = em_end - em_start;
4046 emflags = em->flags;
4051 * bump off for our next call to get_extent
4053 off = extent_map_end(em);
4057 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4059 flags |= FIEMAP_EXTENT_LAST;
4060 } else if (em->block_start == EXTENT_MAP_INLINE) {
4061 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4062 FIEMAP_EXTENT_NOT_ALIGNED);
4063 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4064 flags |= (FIEMAP_EXTENT_DELALLOC |
4065 FIEMAP_EXTENT_UNKNOWN);
4067 disko = em->block_start + offset_in_extent;
4069 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4070 flags |= FIEMAP_EXTENT_ENCODED;
4072 free_extent_map(em);
4074 if ((em_start >= last) || em_len == (u64)-1 ||
4075 (last == (u64)-1 && isize <= em_end)) {
4076 flags |= FIEMAP_EXTENT_LAST;
4080 /* now scan forward to see if this is really the last extent. */
4081 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4088 flags |= FIEMAP_EXTENT_LAST;
4091 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4097 free_extent_map(em);
4099 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4100 &cached_state, GFP_NOFS);
4104 static void __free_extent_buffer(struct extent_buffer *eb)
4106 btrfs_leak_debug_del(&eb->leak_list);
4107 kmem_cache_free(extent_buffer_cache, eb);
4110 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
4115 struct extent_buffer *eb = NULL;
4117 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4124 rwlock_init(&eb->lock);
4125 atomic_set(&eb->write_locks, 0);
4126 atomic_set(&eb->read_locks, 0);
4127 atomic_set(&eb->blocking_readers, 0);
4128 atomic_set(&eb->blocking_writers, 0);
4129 atomic_set(&eb->spinning_readers, 0);
4130 atomic_set(&eb->spinning_writers, 0);
4131 eb->lock_nested = 0;
4132 init_waitqueue_head(&eb->write_lock_wq);
4133 init_waitqueue_head(&eb->read_lock_wq);
4135 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4137 spin_lock_init(&eb->refs_lock);
4138 atomic_set(&eb->refs, 1);
4139 atomic_set(&eb->io_pages, 0);
4142 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4144 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4145 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4146 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4151 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4155 struct extent_buffer *new;
4156 unsigned long num_pages = num_extent_pages(src->start, src->len);
4158 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4162 for (i = 0; i < num_pages; i++) {
4163 p = alloc_page(GFP_ATOMIC);
4165 attach_extent_buffer_page(new, p);
4166 WARN_ON(PageDirty(p));
4171 copy_extent_buffer(new, src, 0, 0, src->len);
4172 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4173 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4178 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4180 struct extent_buffer *eb;
4181 unsigned long num_pages = num_extent_pages(0, len);
4184 eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4188 for (i = 0; i < num_pages; i++) {
4189 eb->pages[i] = alloc_page(GFP_ATOMIC);
4193 set_extent_buffer_uptodate(eb);
4194 btrfs_set_header_nritems(eb, 0);
4195 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4200 __free_page(eb->pages[i - 1]);
4201 __free_extent_buffer(eb);
4205 static int extent_buffer_under_io(struct extent_buffer *eb)
4207 return (atomic_read(&eb->io_pages) ||
4208 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4209 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4213 * Helper for releasing extent buffer page.
4215 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4216 unsigned long start_idx)
4218 unsigned long index;
4219 unsigned long num_pages;
4221 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4223 BUG_ON(extent_buffer_under_io(eb));
4225 num_pages = num_extent_pages(eb->start, eb->len);
4226 index = start_idx + num_pages;
4227 if (start_idx >= index)
4232 page = extent_buffer_page(eb, index);
4233 if (page && mapped) {
4234 spin_lock(&page->mapping->private_lock);
4236 * We do this since we'll remove the pages after we've
4237 * removed the eb from the radix tree, so we could race
4238 * and have this page now attached to the new eb. So
4239 * only clear page_private if it's still connected to
4242 if (PagePrivate(page) &&
4243 page->private == (unsigned long)eb) {
4244 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4245 BUG_ON(PageDirty(page));
4246 BUG_ON(PageWriteback(page));
4248 * We need to make sure we haven't be attached
4251 ClearPagePrivate(page);
4252 set_page_private(page, 0);
4253 /* One for the page private */
4254 page_cache_release(page);
4256 spin_unlock(&page->mapping->private_lock);
4260 /* One for when we alloced the page */
4261 page_cache_release(page);
4263 } while (index != start_idx);
4267 * Helper for releasing the extent buffer.
4269 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4271 btrfs_release_extent_buffer_page(eb, 0);
4272 __free_extent_buffer(eb);
4275 static void check_buffer_tree_ref(struct extent_buffer *eb)
4278 /* the ref bit is tricky. We have to make sure it is set
4279 * if we have the buffer dirty. Otherwise the
4280 * code to free a buffer can end up dropping a dirty
4283 * Once the ref bit is set, it won't go away while the
4284 * buffer is dirty or in writeback, and it also won't
4285 * go away while we have the reference count on the
4288 * We can't just set the ref bit without bumping the
4289 * ref on the eb because free_extent_buffer might
4290 * see the ref bit and try to clear it. If this happens
4291 * free_extent_buffer might end up dropping our original
4292 * ref by mistake and freeing the page before we are able
4293 * to add one more ref.
4295 * So bump the ref count first, then set the bit. If someone
4296 * beat us to it, drop the ref we added.
4298 refs = atomic_read(&eb->refs);
4299 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4302 spin_lock(&eb->refs_lock);
4303 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4304 atomic_inc(&eb->refs);
4305 spin_unlock(&eb->refs_lock);
4308 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4310 unsigned long num_pages, i;
4312 check_buffer_tree_ref(eb);
4314 num_pages = num_extent_pages(eb->start, eb->len);
4315 for (i = 0; i < num_pages; i++) {
4316 struct page *p = extent_buffer_page(eb, i);
4317 mark_page_accessed(p);
4321 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4322 u64 start, unsigned long len)
4324 unsigned long num_pages = num_extent_pages(start, len);
4326 unsigned long index = start >> PAGE_CACHE_SHIFT;
4327 struct extent_buffer *eb;
4328 struct extent_buffer *exists = NULL;
4330 struct address_space *mapping = tree->mapping;
4335 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4336 if (eb && atomic_inc_not_zero(&eb->refs)) {
4338 mark_extent_buffer_accessed(eb);
4343 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4347 for (i = 0; i < num_pages; i++, index++) {
4348 p = find_or_create_page(mapping, index, GFP_NOFS);
4352 spin_lock(&mapping->private_lock);
4353 if (PagePrivate(p)) {
4355 * We could have already allocated an eb for this page
4356 * and attached one so lets see if we can get a ref on
4357 * the existing eb, and if we can we know it's good and
4358 * we can just return that one, else we know we can just
4359 * overwrite page->private.
4361 exists = (struct extent_buffer *)p->private;
4362 if (atomic_inc_not_zero(&exists->refs)) {
4363 spin_unlock(&mapping->private_lock);
4365 page_cache_release(p);
4366 mark_extent_buffer_accessed(exists);
4371 * Do this so attach doesn't complain and we need to
4372 * drop the ref the old guy had.
4374 ClearPagePrivate(p);
4375 WARN_ON(PageDirty(p));
4376 page_cache_release(p);
4378 attach_extent_buffer_page(eb, p);
4379 spin_unlock(&mapping->private_lock);
4380 WARN_ON(PageDirty(p));
4381 mark_page_accessed(p);
4383 if (!PageUptodate(p))
4387 * see below about how we avoid a nasty race with release page
4388 * and why we unlock later
4392 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4394 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4398 spin_lock(&tree->buffer_lock);
4399 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4400 if (ret == -EEXIST) {
4401 exists = radix_tree_lookup(&tree->buffer,
4402 start >> PAGE_CACHE_SHIFT);
4403 if (!atomic_inc_not_zero(&exists->refs)) {
4404 spin_unlock(&tree->buffer_lock);
4405 radix_tree_preload_end();
4409 spin_unlock(&tree->buffer_lock);
4410 radix_tree_preload_end();
4411 mark_extent_buffer_accessed(exists);
4414 /* add one reference for the tree */
4415 check_buffer_tree_ref(eb);
4416 spin_unlock(&tree->buffer_lock);
4417 radix_tree_preload_end();
4420 * there is a race where release page may have
4421 * tried to find this extent buffer in the radix
4422 * but failed. It will tell the VM it is safe to
4423 * reclaim the, and it will clear the page private bit.
4424 * We must make sure to set the page private bit properly
4425 * after the extent buffer is in the radix tree so
4426 * it doesn't get lost
4428 SetPageChecked(eb->pages[0]);
4429 for (i = 1; i < num_pages; i++) {
4430 p = extent_buffer_page(eb, i);
4431 ClearPageChecked(p);
4434 unlock_page(eb->pages[0]);
4438 for (i = 0; i < num_pages; i++) {
4440 unlock_page(eb->pages[i]);
4443 WARN_ON(!atomic_dec_and_test(&eb->refs));
4444 btrfs_release_extent_buffer(eb);
4448 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4449 u64 start, unsigned long len)
4451 struct extent_buffer *eb;
4454 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4455 if (eb && atomic_inc_not_zero(&eb->refs)) {
4457 mark_extent_buffer_accessed(eb);
4465 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4467 struct extent_buffer *eb =
4468 container_of(head, struct extent_buffer, rcu_head);
4470 __free_extent_buffer(eb);
4473 /* Expects to have eb->eb_lock already held */
4474 static int release_extent_buffer(struct extent_buffer *eb)
4476 WARN_ON(atomic_read(&eb->refs) == 0);
4477 if (atomic_dec_and_test(&eb->refs)) {
4478 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4479 spin_unlock(&eb->refs_lock);
4481 struct extent_io_tree *tree = eb->tree;
4483 spin_unlock(&eb->refs_lock);
4485 spin_lock(&tree->buffer_lock);
4486 radix_tree_delete(&tree->buffer,
4487 eb->start >> PAGE_CACHE_SHIFT);
4488 spin_unlock(&tree->buffer_lock);
4491 /* Should be safe to release our pages at this point */
4492 btrfs_release_extent_buffer_page(eb, 0);
4493 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4496 spin_unlock(&eb->refs_lock);
4501 void free_extent_buffer(struct extent_buffer *eb)
4509 refs = atomic_read(&eb->refs);
4512 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4517 spin_lock(&eb->refs_lock);
4518 if (atomic_read(&eb->refs) == 2 &&
4519 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4520 atomic_dec(&eb->refs);
4522 if (atomic_read(&eb->refs) == 2 &&
4523 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4524 !extent_buffer_under_io(eb) &&
4525 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4526 atomic_dec(&eb->refs);
4529 * I know this is terrible, but it's temporary until we stop tracking
4530 * the uptodate bits and such for the extent buffers.
4532 release_extent_buffer(eb);
4535 void free_extent_buffer_stale(struct extent_buffer *eb)
4540 spin_lock(&eb->refs_lock);
4541 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4543 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4544 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4545 atomic_dec(&eb->refs);
4546 release_extent_buffer(eb);
4549 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4552 unsigned long num_pages;
4555 num_pages = num_extent_pages(eb->start, eb->len);
4557 for (i = 0; i < num_pages; i++) {
4558 page = extent_buffer_page(eb, i);
4559 if (!PageDirty(page))
4563 WARN_ON(!PagePrivate(page));
4565 clear_page_dirty_for_io(page);
4566 spin_lock_irq(&page->mapping->tree_lock);
4567 if (!PageDirty(page)) {
4568 radix_tree_tag_clear(&page->mapping->page_tree,
4570 PAGECACHE_TAG_DIRTY);
4572 spin_unlock_irq(&page->mapping->tree_lock);
4573 ClearPageError(page);
4576 WARN_ON(atomic_read(&eb->refs) == 0);
4579 int set_extent_buffer_dirty(struct extent_buffer *eb)
4582 unsigned long num_pages;
4585 check_buffer_tree_ref(eb);
4587 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4589 num_pages = num_extent_pages(eb->start, eb->len);
4590 WARN_ON(atomic_read(&eb->refs) == 0);
4591 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4593 for (i = 0; i < num_pages; i++)
4594 set_page_dirty(extent_buffer_page(eb, i));
4598 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4602 unsigned long num_pages;
4604 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4605 num_pages = num_extent_pages(eb->start, eb->len);
4606 for (i = 0; i < num_pages; i++) {
4607 page = extent_buffer_page(eb, i);
4609 ClearPageUptodate(page);
4614 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4618 unsigned long num_pages;
4620 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4621 num_pages = num_extent_pages(eb->start, eb->len);
4622 for (i = 0; i < num_pages; i++) {
4623 page = extent_buffer_page(eb, i);
4624 SetPageUptodate(page);
4629 int extent_buffer_uptodate(struct extent_buffer *eb)
4631 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4634 int read_extent_buffer_pages(struct extent_io_tree *tree,
4635 struct extent_buffer *eb, u64 start, int wait,
4636 get_extent_t *get_extent, int mirror_num)
4639 unsigned long start_i;
4643 int locked_pages = 0;
4644 int all_uptodate = 1;
4645 unsigned long num_pages;
4646 unsigned long num_reads = 0;
4647 struct bio *bio = NULL;
4648 unsigned long bio_flags = 0;
4650 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4654 WARN_ON(start < eb->start);
4655 start_i = (start >> PAGE_CACHE_SHIFT) -
4656 (eb->start >> PAGE_CACHE_SHIFT);
4661 num_pages = num_extent_pages(eb->start, eb->len);
4662 for (i = start_i; i < num_pages; i++) {
4663 page = extent_buffer_page(eb, i);
4664 if (wait == WAIT_NONE) {
4665 if (!trylock_page(page))
4671 if (!PageUptodate(page)) {
4678 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4682 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4683 eb->read_mirror = 0;
4684 atomic_set(&eb->io_pages, num_reads);
4685 for (i = start_i; i < num_pages; i++) {
4686 page = extent_buffer_page(eb, i);
4687 if (!PageUptodate(page)) {
4688 ClearPageError(page);
4689 err = __extent_read_full_page(tree, page,
4691 mirror_num, &bio_flags,
4701 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4707 if (ret || wait != WAIT_COMPLETE)
4710 for (i = start_i; i < num_pages; i++) {
4711 page = extent_buffer_page(eb, i);
4712 wait_on_page_locked(page);
4713 if (!PageUptodate(page))
4721 while (locked_pages > 0) {
4722 page = extent_buffer_page(eb, i);
4730 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4731 unsigned long start,
4738 char *dst = (char *)dstv;
4739 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4740 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4742 WARN_ON(start > eb->len);
4743 WARN_ON(start + len > eb->start + eb->len);
4745 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4748 page = extent_buffer_page(eb, i);
4750 cur = min(len, (PAGE_CACHE_SIZE - offset));
4751 kaddr = page_address(page);
4752 memcpy(dst, kaddr + offset, cur);
4761 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4762 unsigned long min_len, char **map,
4763 unsigned long *map_start,
4764 unsigned long *map_len)
4766 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4769 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4770 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4771 unsigned long end_i = (start_offset + start + min_len - 1) >>
4778 offset = start_offset;
4782 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4785 if (start + min_len > eb->len) {
4786 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4787 "wanted %lu %lu\n", (unsigned long long)eb->start,
4788 eb->len, start, min_len);
4792 p = extent_buffer_page(eb, i);
4793 kaddr = page_address(p);
4794 *map = kaddr + offset;
4795 *map_len = PAGE_CACHE_SIZE - offset;
4799 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4800 unsigned long start,
4807 char *ptr = (char *)ptrv;
4808 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4809 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4812 WARN_ON(start > eb->len);
4813 WARN_ON(start + len > eb->start + eb->len);
4815 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4818 page = extent_buffer_page(eb, i);
4820 cur = min(len, (PAGE_CACHE_SIZE - offset));
4822 kaddr = page_address(page);
4823 ret = memcmp(ptr, kaddr + offset, cur);
4835 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4836 unsigned long start, unsigned long len)
4842 char *src = (char *)srcv;
4843 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4844 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4846 WARN_ON(start > eb->len);
4847 WARN_ON(start + len > eb->start + eb->len);
4849 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4852 page = extent_buffer_page(eb, i);
4853 WARN_ON(!PageUptodate(page));
4855 cur = min(len, PAGE_CACHE_SIZE - offset);
4856 kaddr = page_address(page);
4857 memcpy(kaddr + offset, src, cur);
4866 void memset_extent_buffer(struct extent_buffer *eb, char c,
4867 unsigned long start, unsigned long len)
4873 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4874 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4876 WARN_ON(start > eb->len);
4877 WARN_ON(start + len > eb->start + eb->len);
4879 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4882 page = extent_buffer_page(eb, i);
4883 WARN_ON(!PageUptodate(page));
4885 cur = min(len, PAGE_CACHE_SIZE - offset);
4886 kaddr = page_address(page);
4887 memset(kaddr + offset, c, cur);
4895 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4896 unsigned long dst_offset, unsigned long src_offset,
4899 u64 dst_len = dst->len;
4904 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4905 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4907 WARN_ON(src->len != dst_len);
4909 offset = (start_offset + dst_offset) &
4910 ((unsigned long)PAGE_CACHE_SIZE - 1);
4913 page = extent_buffer_page(dst, i);
4914 WARN_ON(!PageUptodate(page));
4916 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4918 kaddr = page_address(page);
4919 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4928 static void move_pages(struct page *dst_page, struct page *src_page,
4929 unsigned long dst_off, unsigned long src_off,
4932 char *dst_kaddr = page_address(dst_page);
4933 if (dst_page == src_page) {
4934 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4936 char *src_kaddr = page_address(src_page);
4937 char *p = dst_kaddr + dst_off + len;
4938 char *s = src_kaddr + src_off + len;
4945 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4947 unsigned long distance = (src > dst) ? src - dst : dst - src;
4948 return distance < len;
4951 static void copy_pages(struct page *dst_page, struct page *src_page,
4952 unsigned long dst_off, unsigned long src_off,
4955 char *dst_kaddr = page_address(dst_page);
4957 int must_memmove = 0;
4959 if (dst_page != src_page) {
4960 src_kaddr = page_address(src_page);
4962 src_kaddr = dst_kaddr;
4963 if (areas_overlap(src_off, dst_off, len))
4968 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4970 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4973 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4974 unsigned long src_offset, unsigned long len)
4977 size_t dst_off_in_page;
4978 size_t src_off_in_page;
4979 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4980 unsigned long dst_i;
4981 unsigned long src_i;
4983 if (src_offset + len > dst->len) {
4984 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4985 "len %lu dst len %lu\n", src_offset, len, dst->len);
4988 if (dst_offset + len > dst->len) {
4989 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4990 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4995 dst_off_in_page = (start_offset + dst_offset) &
4996 ((unsigned long)PAGE_CACHE_SIZE - 1);
4997 src_off_in_page = (start_offset + src_offset) &
4998 ((unsigned long)PAGE_CACHE_SIZE - 1);
5000 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5001 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5003 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5005 cur = min_t(unsigned long, cur,
5006 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5008 copy_pages(extent_buffer_page(dst, dst_i),
5009 extent_buffer_page(dst, src_i),
5010 dst_off_in_page, src_off_in_page, cur);
5018 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5019 unsigned long src_offset, unsigned long len)
5022 size_t dst_off_in_page;
5023 size_t src_off_in_page;
5024 unsigned long dst_end = dst_offset + len - 1;
5025 unsigned long src_end = src_offset + len - 1;
5026 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5027 unsigned long dst_i;
5028 unsigned long src_i;
5030 if (src_offset + len > dst->len) {
5031 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5032 "len %lu len %lu\n", src_offset, len, dst->len);
5035 if (dst_offset + len > dst->len) {
5036 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5037 "len %lu len %lu\n", dst_offset, len, dst->len);
5040 if (dst_offset < src_offset) {
5041 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5045 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5046 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5048 dst_off_in_page = (start_offset + dst_end) &
5049 ((unsigned long)PAGE_CACHE_SIZE - 1);
5050 src_off_in_page = (start_offset + src_end) &
5051 ((unsigned long)PAGE_CACHE_SIZE - 1);
5053 cur = min_t(unsigned long, len, src_off_in_page + 1);
5054 cur = min(cur, dst_off_in_page + 1);
5055 move_pages(extent_buffer_page(dst, dst_i),
5056 extent_buffer_page(dst, src_i),
5057 dst_off_in_page - cur + 1,
5058 src_off_in_page - cur + 1, cur);
5066 int try_release_extent_buffer(struct page *page)
5068 struct extent_buffer *eb;
5071 * We need to make sure noboody is attaching this page to an eb right
5074 spin_lock(&page->mapping->private_lock);
5075 if (!PagePrivate(page)) {
5076 spin_unlock(&page->mapping->private_lock);
5080 eb = (struct extent_buffer *)page->private;
5084 * This is a little awful but should be ok, we need to make sure that
5085 * the eb doesn't disappear out from under us while we're looking at
5088 spin_lock(&eb->refs_lock);
5089 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5090 spin_unlock(&eb->refs_lock);
5091 spin_unlock(&page->mapping->private_lock);
5094 spin_unlock(&page->mapping->private_lock);
5097 * If tree ref isn't set then we know the ref on this eb is a real ref,
5098 * so just return, this page will likely be freed soon anyway.
5100 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5101 spin_unlock(&eb->refs_lock);
5105 return release_extent_buffer(eb);