1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
26 static struct bio_set *btrfs_bioset;
28 #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 state->start, state->end, state->state, state->tree,
65 atomic_read(&state->refs));
66 list_del(&state->leak_list);
67 kmem_cache_free(extent_state_cache, state);
70 while (!list_empty(&buffers)) {
71 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
72 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
74 eb->start, eb->len, atomic_read(&eb->refs));
75 list_del(&eb->leak_list);
76 kmem_cache_free(extent_buffer_cache, eb);
80 #define btrfs_debug_check_extent_io_range(inode, start, end) \
81 __btrfs_debug_check_extent_io_range(__func__, (inode), (start), (end))
82 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
83 struct inode *inode, u64 start, u64 end)
85 u64 isize = i_size_read(inode);
87 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
88 printk_ratelimited(KERN_DEBUG
89 "btrfs: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
90 caller, btrfs_ino(inode), isize, start, end);
94 #define btrfs_leak_debug_add(new, head) do {} while (0)
95 #define btrfs_leak_debug_del(entry) do {} while (0)
96 #define btrfs_leak_debug_check() do {} while (0)
97 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
100 #define BUFFER_LRU_MAX 64
105 struct rb_node rb_node;
108 struct extent_page_data {
110 struct extent_io_tree *tree;
111 get_extent_t *get_extent;
112 unsigned long bio_flags;
114 /* tells writepage not to lock the state bits for this range
115 * it still does the unlocking
117 unsigned int extent_locked:1;
119 /* tells the submit_bio code to use a WRITE_SYNC */
120 unsigned int sync_io:1;
123 static noinline void flush_write_bio(void *data);
124 static inline struct btrfs_fs_info *
125 tree_fs_info(struct extent_io_tree *tree)
127 return btrfs_sb(tree->mapping->host->i_sb);
130 int __init extent_io_init(void)
132 extent_state_cache = kmem_cache_create("btrfs_extent_state",
133 sizeof(struct extent_state), 0,
134 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
135 if (!extent_state_cache)
138 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
139 sizeof(struct extent_buffer), 0,
140 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
141 if (!extent_buffer_cache)
142 goto free_state_cache;
144 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
145 offsetof(struct btrfs_io_bio, bio));
147 goto free_buffer_cache;
149 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
155 bioset_free(btrfs_bioset);
159 kmem_cache_destroy(extent_buffer_cache);
160 extent_buffer_cache = NULL;
163 kmem_cache_destroy(extent_state_cache);
164 extent_state_cache = NULL;
168 void extent_io_exit(void)
170 btrfs_leak_debug_check();
173 * Make sure all delayed rcu free are flushed before we
177 if (extent_state_cache)
178 kmem_cache_destroy(extent_state_cache);
179 if (extent_buffer_cache)
180 kmem_cache_destroy(extent_buffer_cache);
182 bioset_free(btrfs_bioset);
185 void extent_io_tree_init(struct extent_io_tree *tree,
186 struct address_space *mapping)
188 tree->state = RB_ROOT;
189 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
191 tree->dirty_bytes = 0;
192 spin_lock_init(&tree->lock);
193 spin_lock_init(&tree->buffer_lock);
194 tree->mapping = mapping;
197 static struct extent_state *alloc_extent_state(gfp_t mask)
199 struct extent_state *state;
201 state = kmem_cache_alloc(extent_state_cache, mask);
207 btrfs_leak_debug_add(&state->leak_list, &states);
208 atomic_set(&state->refs, 1);
209 init_waitqueue_head(&state->wq);
210 trace_alloc_extent_state(state, mask, _RET_IP_);
214 void free_extent_state(struct extent_state *state)
218 if (atomic_dec_and_test(&state->refs)) {
219 WARN_ON(state->tree);
220 btrfs_leak_debug_del(&state->leak_list);
221 trace_free_extent_state(state, _RET_IP_);
222 kmem_cache_free(extent_state_cache, state);
226 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
227 struct rb_node *node)
229 struct rb_node **p = &root->rb_node;
230 struct rb_node *parent = NULL;
231 struct tree_entry *entry;
235 entry = rb_entry(parent, struct tree_entry, rb_node);
237 if (offset < entry->start)
239 else if (offset > entry->end)
245 rb_link_node(node, parent, p);
246 rb_insert_color(node, root);
250 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
251 struct rb_node **prev_ret,
252 struct rb_node **next_ret)
254 struct rb_root *root = &tree->state;
255 struct rb_node *n = root->rb_node;
256 struct rb_node *prev = NULL;
257 struct rb_node *orig_prev = NULL;
258 struct tree_entry *entry;
259 struct tree_entry *prev_entry = NULL;
262 entry = rb_entry(n, struct tree_entry, rb_node);
266 if (offset < entry->start)
268 else if (offset > entry->end)
276 while (prev && offset > prev_entry->end) {
277 prev = rb_next(prev);
278 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
285 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
286 while (prev && offset < prev_entry->start) {
287 prev = rb_prev(prev);
288 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
295 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
298 struct rb_node *prev = NULL;
301 ret = __etree_search(tree, offset, &prev, NULL);
307 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
308 struct extent_state *other)
310 if (tree->ops && tree->ops->merge_extent_hook)
311 tree->ops->merge_extent_hook(tree->mapping->host, new,
316 * utility function to look for merge candidates inside a given range.
317 * Any extents with matching state are merged together into a single
318 * extent in the tree. Extents with EXTENT_IO in their state field
319 * are not merged because the end_io handlers need to be able to do
320 * operations on them without sleeping (or doing allocations/splits).
322 * This should be called with the tree lock held.
324 static void merge_state(struct extent_io_tree *tree,
325 struct extent_state *state)
327 struct extent_state *other;
328 struct rb_node *other_node;
330 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
333 other_node = rb_prev(&state->rb_node);
335 other = rb_entry(other_node, struct extent_state, rb_node);
336 if (other->end == state->start - 1 &&
337 other->state == state->state) {
338 merge_cb(tree, state, other);
339 state->start = other->start;
341 rb_erase(&other->rb_node, &tree->state);
342 free_extent_state(other);
345 other_node = rb_next(&state->rb_node);
347 other = rb_entry(other_node, struct extent_state, rb_node);
348 if (other->start == state->end + 1 &&
349 other->state == state->state) {
350 merge_cb(tree, state, other);
351 state->end = other->end;
353 rb_erase(&other->rb_node, &tree->state);
354 free_extent_state(other);
359 static void set_state_cb(struct extent_io_tree *tree,
360 struct extent_state *state, unsigned long *bits)
362 if (tree->ops && tree->ops->set_bit_hook)
363 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
366 static void clear_state_cb(struct extent_io_tree *tree,
367 struct extent_state *state, unsigned long *bits)
369 if (tree->ops && tree->ops->clear_bit_hook)
370 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
373 static void set_state_bits(struct extent_io_tree *tree,
374 struct extent_state *state, unsigned long *bits);
377 * insert an extent_state struct into the tree. 'bits' are set on the
378 * struct before it is inserted.
380 * This may return -EEXIST if the extent is already there, in which case the
381 * state struct is freed.
383 * The tree lock is not taken internally. This is a utility function and
384 * probably isn't what you want to call (see set/clear_extent_bit).
386 static int insert_state(struct extent_io_tree *tree,
387 struct extent_state *state, u64 start, u64 end,
390 struct rb_node *node;
393 WARN(1, KERN_ERR "btrfs end < start %llu %llu\n",
395 state->start = start;
398 set_state_bits(tree, state, bits);
400 node = tree_insert(&tree->state, end, &state->rb_node);
402 struct extent_state *found;
403 found = rb_entry(node, struct extent_state, rb_node);
404 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
406 found->start, found->end, start, end);
410 merge_state(tree, state);
414 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
417 if (tree->ops && tree->ops->split_extent_hook)
418 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
422 * split a given extent state struct in two, inserting the preallocated
423 * struct 'prealloc' as the newly created second half. 'split' indicates an
424 * offset inside 'orig' where it should be split.
427 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
428 * are two extent state structs in the tree:
429 * prealloc: [orig->start, split - 1]
430 * orig: [ split, orig->end ]
432 * The tree locks are not taken by this function. They need to be held
435 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
436 struct extent_state *prealloc, u64 split)
438 struct rb_node *node;
440 split_cb(tree, orig, split);
442 prealloc->start = orig->start;
443 prealloc->end = split - 1;
444 prealloc->state = orig->state;
447 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
449 free_extent_state(prealloc);
452 prealloc->tree = tree;
456 static struct extent_state *next_state(struct extent_state *state)
458 struct rb_node *next = rb_next(&state->rb_node);
460 return rb_entry(next, struct extent_state, rb_node);
466 * utility function to clear some bits in an extent state struct.
467 * it will optionally wake up any one waiting on this state (wake == 1).
469 * If no bits are set on the state struct after clearing things, the
470 * struct is freed and removed from the tree
472 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
473 struct extent_state *state,
474 unsigned long *bits, int wake)
476 struct extent_state *next;
477 unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
479 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
480 u64 range = state->end - state->start + 1;
481 WARN_ON(range > tree->dirty_bytes);
482 tree->dirty_bytes -= range;
484 clear_state_cb(tree, state, bits);
485 state->state &= ~bits_to_clear;
488 if (state->state == 0) {
489 next = next_state(state);
491 rb_erase(&state->rb_node, &tree->state);
493 free_extent_state(state);
498 merge_state(tree, state);
499 next = next_state(state);
504 static struct extent_state *
505 alloc_extent_state_atomic(struct extent_state *prealloc)
508 prealloc = alloc_extent_state(GFP_ATOMIC);
513 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
515 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
516 "Extent tree was modified by another "
517 "thread while locked.");
521 * clear some bits on a range in the tree. This may require splitting
522 * or inserting elements in the tree, so the gfp mask is used to
523 * indicate which allocations or sleeping are allowed.
525 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
526 * the given range from the tree regardless of state (ie for truncate).
528 * the range [start, end] is inclusive.
530 * This takes the tree lock, and returns 0 on success and < 0 on error.
532 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
533 unsigned long bits, int wake, int delete,
534 struct extent_state **cached_state,
537 struct extent_state *state;
538 struct extent_state *cached;
539 struct extent_state *prealloc = NULL;
540 struct rb_node *node;
545 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
547 if (bits & EXTENT_DELALLOC)
548 bits |= EXTENT_NORESERVE;
551 bits |= ~EXTENT_CTLBITS;
552 bits |= EXTENT_FIRST_DELALLOC;
554 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
557 if (!prealloc && (mask & __GFP_WAIT)) {
558 prealloc = alloc_extent_state(mask);
563 spin_lock(&tree->lock);
565 cached = *cached_state;
568 *cached_state = NULL;
572 if (cached && cached->tree && cached->start <= start &&
573 cached->end > start) {
575 atomic_dec(&cached->refs);
580 free_extent_state(cached);
583 * this search will find the extents that end after
586 node = tree_search(tree, start);
589 state = rb_entry(node, struct extent_state, rb_node);
591 if (state->start > end)
593 WARN_ON(state->end < start);
594 last_end = state->end;
596 /* the state doesn't have the wanted bits, go ahead */
597 if (!(state->state & bits)) {
598 state = next_state(state);
603 * | ---- desired range ---- |
605 * | ------------- state -------------- |
607 * We need to split the extent we found, and may flip
608 * bits on second half.
610 * If the extent we found extends past our range, we
611 * just split and search again. It'll get split again
612 * the next time though.
614 * If the extent we found is inside our range, we clear
615 * the desired bit on it.
618 if (state->start < start) {
619 prealloc = alloc_extent_state_atomic(prealloc);
621 err = split_state(tree, state, prealloc, start);
623 extent_io_tree_panic(tree, err);
628 if (state->end <= end) {
629 state = clear_state_bit(tree, state, &bits, wake);
635 * | ---- desired range ---- |
637 * We need to split the extent, and clear the bit
640 if (state->start <= end && state->end > end) {
641 prealloc = alloc_extent_state_atomic(prealloc);
643 err = split_state(tree, state, prealloc, end + 1);
645 extent_io_tree_panic(tree, err);
650 clear_state_bit(tree, prealloc, &bits, wake);
656 state = clear_state_bit(tree, state, &bits, wake);
658 if (last_end == (u64)-1)
660 start = last_end + 1;
661 if (start <= end && state && !need_resched())
666 spin_unlock(&tree->lock);
668 free_extent_state(prealloc);
675 spin_unlock(&tree->lock);
676 if (mask & __GFP_WAIT)
681 static void wait_on_state(struct extent_io_tree *tree,
682 struct extent_state *state)
683 __releases(tree->lock)
684 __acquires(tree->lock)
687 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
688 spin_unlock(&tree->lock);
690 spin_lock(&tree->lock);
691 finish_wait(&state->wq, &wait);
695 * waits for one or more bits to clear on a range in the state tree.
696 * The range [start, end] is inclusive.
697 * The tree lock is taken by this function
699 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
702 struct extent_state *state;
703 struct rb_node *node;
705 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
707 spin_lock(&tree->lock);
711 * this search will find all the extents that end after
714 node = tree_search(tree, start);
718 state = rb_entry(node, struct extent_state, rb_node);
720 if (state->start > end)
723 if (state->state & bits) {
724 start = state->start;
725 atomic_inc(&state->refs);
726 wait_on_state(tree, state);
727 free_extent_state(state);
730 start = state->end + 1;
735 cond_resched_lock(&tree->lock);
738 spin_unlock(&tree->lock);
741 static void set_state_bits(struct extent_io_tree *tree,
742 struct extent_state *state,
745 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
747 set_state_cb(tree, state, bits);
748 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
749 u64 range = state->end - state->start + 1;
750 tree->dirty_bytes += range;
752 state->state |= bits_to_set;
755 static void cache_state(struct extent_state *state,
756 struct extent_state **cached_ptr)
758 if (cached_ptr && !(*cached_ptr)) {
759 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
761 atomic_inc(&state->refs);
767 * set some bits on a range in the tree. This may require allocations or
768 * sleeping, so the gfp mask is used to indicate what is allowed.
770 * If any of the exclusive bits are set, this will fail with -EEXIST if some
771 * part of the range already has the desired bits set. The start of the
772 * existing range is returned in failed_start in this case.
774 * [start, end] is inclusive This takes the tree lock.
777 static int __must_check
778 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
779 unsigned long bits, unsigned long exclusive_bits,
780 u64 *failed_start, struct extent_state **cached_state,
783 struct extent_state *state;
784 struct extent_state *prealloc = NULL;
785 struct rb_node *node;
790 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
792 bits |= EXTENT_FIRST_DELALLOC;
794 if (!prealloc && (mask & __GFP_WAIT)) {
795 prealloc = alloc_extent_state(mask);
799 spin_lock(&tree->lock);
800 if (cached_state && *cached_state) {
801 state = *cached_state;
802 if (state->start <= start && state->end > start &&
804 node = &state->rb_node;
809 * this search will find all the extents that end after
812 node = tree_search(tree, start);
814 prealloc = alloc_extent_state_atomic(prealloc);
816 err = insert_state(tree, prealloc, start, end, &bits);
818 extent_io_tree_panic(tree, err);
823 state = rb_entry(node, struct extent_state, rb_node);
825 last_start = state->start;
826 last_end = state->end;
829 * | ---- desired range ---- |
832 * Just lock what we found and keep going
834 if (state->start == start && state->end <= end) {
835 if (state->state & exclusive_bits) {
836 *failed_start = state->start;
841 set_state_bits(tree, state, &bits);
842 cache_state(state, cached_state);
843 merge_state(tree, state);
844 if (last_end == (u64)-1)
846 start = last_end + 1;
847 state = next_state(state);
848 if (start < end && state && state->start == start &&
855 * | ---- desired range ---- |
858 * | ------------- state -------------- |
860 * We need to split the extent we found, and may flip bits on
863 * If the extent we found extends past our
864 * range, we just split and search again. It'll get split
865 * again the next time though.
867 * If the extent we found is inside our range, we set the
870 if (state->start < start) {
871 if (state->state & exclusive_bits) {
872 *failed_start = start;
877 prealloc = alloc_extent_state_atomic(prealloc);
879 err = split_state(tree, state, prealloc, start);
881 extent_io_tree_panic(tree, err);
886 if (state->end <= end) {
887 set_state_bits(tree, state, &bits);
888 cache_state(state, cached_state);
889 merge_state(tree, state);
890 if (last_end == (u64)-1)
892 start = last_end + 1;
893 state = next_state(state);
894 if (start < end && state && state->start == start &&
901 * | ---- desired range ---- |
902 * | state | or | state |
904 * There's a hole, we need to insert something in it and
905 * ignore the extent we found.
907 if (state->start > start) {
909 if (end < last_start)
912 this_end = last_start - 1;
914 prealloc = alloc_extent_state_atomic(prealloc);
918 * Avoid to free 'prealloc' if it can be merged with
921 err = insert_state(tree, prealloc, start, this_end,
924 extent_io_tree_panic(tree, err);
926 cache_state(prealloc, cached_state);
928 start = this_end + 1;
932 * | ---- desired range ---- |
934 * We need to split the extent, and set the bit
937 if (state->start <= end && state->end > end) {
938 if (state->state & exclusive_bits) {
939 *failed_start = start;
944 prealloc = alloc_extent_state_atomic(prealloc);
946 err = split_state(tree, state, prealloc, end + 1);
948 extent_io_tree_panic(tree, err);
950 set_state_bits(tree, prealloc, &bits);
951 cache_state(prealloc, cached_state);
952 merge_state(tree, prealloc);
960 spin_unlock(&tree->lock);
962 free_extent_state(prealloc);
969 spin_unlock(&tree->lock);
970 if (mask & __GFP_WAIT)
975 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
976 unsigned long bits, u64 * failed_start,
977 struct extent_state **cached_state, gfp_t mask)
979 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
985 * convert_extent_bit - convert all bits in a given range from one bit to
987 * @tree: the io tree to search
988 * @start: the start offset in bytes
989 * @end: the end offset in bytes (inclusive)
990 * @bits: the bits to set in this range
991 * @clear_bits: the bits to clear in this range
992 * @cached_state: state that we're going to cache
993 * @mask: the allocation mask
995 * This will go through and set bits for the given range. If any states exist
996 * already in this range they are set with the given bit and cleared of the
997 * clear_bits. This is only meant to be used by things that are mergeable, ie
998 * converting from say DELALLOC to DIRTY. This is not meant to be used with
999 * boundary bits like LOCK.
1001 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1002 unsigned long bits, unsigned long clear_bits,
1003 struct extent_state **cached_state, gfp_t mask)
1005 struct extent_state *state;
1006 struct extent_state *prealloc = NULL;
1007 struct rb_node *node;
1012 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
1015 if (!prealloc && (mask & __GFP_WAIT)) {
1016 prealloc = alloc_extent_state(mask);
1021 spin_lock(&tree->lock);
1022 if (cached_state && *cached_state) {
1023 state = *cached_state;
1024 if (state->start <= start && state->end > start &&
1026 node = &state->rb_node;
1032 * this search will find all the extents that end after
1035 node = tree_search(tree, start);
1037 prealloc = alloc_extent_state_atomic(prealloc);
1042 err = insert_state(tree, prealloc, start, end, &bits);
1045 extent_io_tree_panic(tree, err);
1048 state = rb_entry(node, struct extent_state, rb_node);
1050 last_start = state->start;
1051 last_end = state->end;
1054 * | ---- desired range ---- |
1057 * Just lock what we found and keep going
1059 if (state->start == start && state->end <= end) {
1060 set_state_bits(tree, state, &bits);
1061 cache_state(state, cached_state);
1062 state = clear_state_bit(tree, state, &clear_bits, 0);
1063 if (last_end == (u64)-1)
1065 start = last_end + 1;
1066 if (start < end && state && state->start == start &&
1073 * | ---- desired range ---- |
1076 * | ------------- state -------------- |
1078 * We need to split the extent we found, and may flip bits on
1081 * If the extent we found extends past our
1082 * range, we just split and search again. It'll get split
1083 * again the next time though.
1085 * If the extent we found is inside our range, we set the
1086 * desired bit on it.
1088 if (state->start < start) {
1089 prealloc = alloc_extent_state_atomic(prealloc);
1094 err = split_state(tree, state, prealloc, start);
1096 extent_io_tree_panic(tree, err);
1100 if (state->end <= end) {
1101 set_state_bits(tree, state, &bits);
1102 cache_state(state, cached_state);
1103 state = clear_state_bit(tree, state, &clear_bits, 0);
1104 if (last_end == (u64)-1)
1106 start = last_end + 1;
1107 if (start < end && state && state->start == start &&
1114 * | ---- desired range ---- |
1115 * | state | or | state |
1117 * There's a hole, we need to insert something in it and
1118 * ignore the extent we found.
1120 if (state->start > start) {
1122 if (end < last_start)
1125 this_end = last_start - 1;
1127 prealloc = alloc_extent_state_atomic(prealloc);
1134 * Avoid to free 'prealloc' if it can be merged with
1137 err = insert_state(tree, prealloc, start, this_end,
1140 extent_io_tree_panic(tree, err);
1141 cache_state(prealloc, cached_state);
1143 start = this_end + 1;
1147 * | ---- desired range ---- |
1149 * We need to split the extent, and set the bit
1152 if (state->start <= end && state->end > end) {
1153 prealloc = alloc_extent_state_atomic(prealloc);
1159 err = split_state(tree, state, prealloc, end + 1);
1161 extent_io_tree_panic(tree, err);
1163 set_state_bits(tree, prealloc, &bits);
1164 cache_state(prealloc, cached_state);
1165 clear_state_bit(tree, prealloc, &clear_bits, 0);
1173 spin_unlock(&tree->lock);
1175 free_extent_state(prealloc);
1182 spin_unlock(&tree->lock);
1183 if (mask & __GFP_WAIT)
1188 /* wrappers around set/clear extent bit */
1189 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1192 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1196 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1197 unsigned long bits, gfp_t mask)
1199 return set_extent_bit(tree, start, end, bits, NULL,
1203 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1204 unsigned long bits, gfp_t mask)
1206 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1209 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1210 struct extent_state **cached_state, gfp_t mask)
1212 return set_extent_bit(tree, start, end,
1213 EXTENT_DELALLOC | EXTENT_UPTODATE,
1214 NULL, cached_state, mask);
1217 int set_extent_defrag(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 | EXTENT_DEFRAG,
1222 NULL, cached_state, mask);
1225 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1228 return clear_extent_bit(tree, start, end,
1229 EXTENT_DIRTY | EXTENT_DELALLOC |
1230 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1233 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1236 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1240 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1241 struct extent_state **cached_state, gfp_t mask)
1243 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1244 cached_state, mask);
1247 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1248 struct extent_state **cached_state, gfp_t mask)
1250 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1251 cached_state, mask);
1255 * either insert or lock state struct between start and end use mask to tell
1256 * us if waiting is desired.
1258 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1259 unsigned long bits, struct extent_state **cached_state)
1264 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1265 EXTENT_LOCKED, &failed_start,
1266 cached_state, GFP_NOFS);
1267 if (err == -EEXIST) {
1268 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1269 start = failed_start;
1272 WARN_ON(start > end);
1277 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1279 return lock_extent_bits(tree, start, end, 0, NULL);
1282 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1287 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1288 &failed_start, NULL, GFP_NOFS);
1289 if (err == -EEXIST) {
1290 if (failed_start > start)
1291 clear_extent_bit(tree, start, failed_start - 1,
1292 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1298 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1299 struct extent_state **cached, gfp_t mask)
1301 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1305 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1307 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1311 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1313 unsigned long index = start >> PAGE_CACHE_SHIFT;
1314 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1317 while (index <= end_index) {
1318 page = find_get_page(inode->i_mapping, index);
1319 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1320 clear_page_dirty_for_io(page);
1321 page_cache_release(page);
1327 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1329 unsigned long index = start >> PAGE_CACHE_SHIFT;
1330 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1333 while (index <= end_index) {
1334 page = find_get_page(inode->i_mapping, index);
1335 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1336 account_page_redirty(page);
1337 __set_page_dirty_nobuffers(page);
1338 page_cache_release(page);
1345 * helper function to set both pages and extents in the tree writeback
1347 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1349 unsigned long index = start >> PAGE_CACHE_SHIFT;
1350 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1353 while (index <= end_index) {
1354 page = find_get_page(tree->mapping, index);
1355 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1356 set_page_writeback(page);
1357 page_cache_release(page);
1363 /* find the first state struct with 'bits' set after 'start', and
1364 * return it. tree->lock must be held. NULL will returned if
1365 * nothing was found after 'start'
1367 static struct extent_state *
1368 find_first_extent_bit_state(struct extent_io_tree *tree,
1369 u64 start, unsigned long bits)
1371 struct rb_node *node;
1372 struct extent_state *state;
1375 * this search will find all the extents that end after
1378 node = tree_search(tree, start);
1383 state = rb_entry(node, struct extent_state, rb_node);
1384 if (state->end >= start && (state->state & bits))
1387 node = rb_next(node);
1396 * find the first offset in the io tree with 'bits' set. zero is
1397 * returned if we find something, and *start_ret and *end_ret are
1398 * set to reflect the state struct that was found.
1400 * If nothing was found, 1 is returned. If found something, return 0.
1402 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1403 u64 *start_ret, u64 *end_ret, unsigned long bits,
1404 struct extent_state **cached_state)
1406 struct extent_state *state;
1410 spin_lock(&tree->lock);
1411 if (cached_state && *cached_state) {
1412 state = *cached_state;
1413 if (state->end == start - 1 && state->tree) {
1414 n = rb_next(&state->rb_node);
1416 state = rb_entry(n, struct extent_state,
1418 if (state->state & bits)
1422 free_extent_state(*cached_state);
1423 *cached_state = NULL;
1426 free_extent_state(*cached_state);
1427 *cached_state = NULL;
1430 state = find_first_extent_bit_state(tree, start, bits);
1433 cache_state(state, cached_state);
1434 *start_ret = state->start;
1435 *end_ret = state->end;
1439 spin_unlock(&tree->lock);
1444 * find a contiguous range of bytes in the file marked as delalloc, not
1445 * more than 'max_bytes'. start and end are used to return the range,
1447 * 1 is returned if we find something, 0 if nothing was in the tree
1449 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1450 u64 *start, u64 *end, u64 max_bytes,
1451 struct extent_state **cached_state)
1453 struct rb_node *node;
1454 struct extent_state *state;
1455 u64 cur_start = *start;
1457 u64 total_bytes = 0;
1459 spin_lock(&tree->lock);
1462 * this search will find all the extents that end after
1465 node = tree_search(tree, cur_start);
1473 state = rb_entry(node, struct extent_state, rb_node);
1474 if (found && (state->start != cur_start ||
1475 (state->state & EXTENT_BOUNDARY))) {
1478 if (!(state->state & EXTENT_DELALLOC)) {
1484 *start = state->start;
1485 *cached_state = state;
1486 atomic_inc(&state->refs);
1490 cur_start = state->end + 1;
1491 node = rb_next(node);
1492 total_bytes += state->end - state->start + 1;
1493 if (total_bytes >= max_bytes)
1499 spin_unlock(&tree->lock);
1503 static noinline void __unlock_for_delalloc(struct inode *inode,
1504 struct page *locked_page,
1508 struct page *pages[16];
1509 unsigned long index = start >> PAGE_CACHE_SHIFT;
1510 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1511 unsigned long nr_pages = end_index - index + 1;
1514 if (index == locked_page->index && end_index == index)
1517 while (nr_pages > 0) {
1518 ret = find_get_pages_contig(inode->i_mapping, index,
1519 min_t(unsigned long, nr_pages,
1520 ARRAY_SIZE(pages)), pages);
1521 for (i = 0; i < ret; i++) {
1522 if (pages[i] != locked_page)
1523 unlock_page(pages[i]);
1524 page_cache_release(pages[i]);
1532 static noinline int lock_delalloc_pages(struct inode *inode,
1533 struct page *locked_page,
1537 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1538 unsigned long start_index = index;
1539 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1540 unsigned long pages_locked = 0;
1541 struct page *pages[16];
1542 unsigned long nrpages;
1546 /* the caller is responsible for locking the start index */
1547 if (index == locked_page->index && index == end_index)
1550 /* skip the page at the start index */
1551 nrpages = end_index - index + 1;
1552 while (nrpages > 0) {
1553 ret = find_get_pages_contig(inode->i_mapping, index,
1554 min_t(unsigned long,
1555 nrpages, ARRAY_SIZE(pages)), pages);
1560 /* now we have an array of pages, lock them all */
1561 for (i = 0; i < ret; i++) {
1563 * the caller is taking responsibility for
1566 if (pages[i] != locked_page) {
1567 lock_page(pages[i]);
1568 if (!PageDirty(pages[i]) ||
1569 pages[i]->mapping != inode->i_mapping) {
1571 unlock_page(pages[i]);
1572 page_cache_release(pages[i]);
1576 page_cache_release(pages[i]);
1585 if (ret && pages_locked) {
1586 __unlock_for_delalloc(inode, locked_page,
1588 ((u64)(start_index + pages_locked - 1)) <<
1595 * find a contiguous range of bytes in the file marked as delalloc, not
1596 * more than 'max_bytes'. start and end are used to return the range,
1598 * 1 is returned if we find something, 0 if nothing was in the tree
1600 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1601 struct extent_io_tree *tree,
1602 struct page *locked_page, u64 *start,
1603 u64 *end, u64 max_bytes)
1608 struct extent_state *cached_state = NULL;
1613 /* step one, find a bunch of delalloc bytes starting at start */
1614 delalloc_start = *start;
1616 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1617 max_bytes, &cached_state);
1618 if (!found || delalloc_end <= *start) {
1619 *start = delalloc_start;
1620 *end = delalloc_end;
1621 free_extent_state(cached_state);
1626 * start comes from the offset of locked_page. We have to lock
1627 * pages in order, so we can't process delalloc bytes before
1630 if (delalloc_start < *start)
1631 delalloc_start = *start;
1634 * make sure to limit the number of pages we try to lock down
1636 if (delalloc_end + 1 - delalloc_start > max_bytes)
1637 delalloc_end = delalloc_start + max_bytes - 1;
1639 /* step two, lock all the pages after the page that has start */
1640 ret = lock_delalloc_pages(inode, locked_page,
1641 delalloc_start, delalloc_end);
1642 if (ret == -EAGAIN) {
1643 /* some of the pages are gone, lets avoid looping by
1644 * shortening the size of the delalloc range we're searching
1646 free_extent_state(cached_state);
1648 max_bytes = PAGE_CACHE_SIZE;
1656 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1658 /* step three, lock the state bits for the whole range */
1659 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1661 /* then test to make sure it is all still delalloc */
1662 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1663 EXTENT_DELALLOC, 1, cached_state);
1665 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1666 &cached_state, GFP_NOFS);
1667 __unlock_for_delalloc(inode, locked_page,
1668 delalloc_start, delalloc_end);
1672 free_extent_state(cached_state);
1673 *start = delalloc_start;
1674 *end = delalloc_end;
1679 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1680 struct page *locked_page,
1681 unsigned long clear_bits,
1682 unsigned long page_ops)
1684 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1686 struct page *pages[16];
1687 unsigned long index = start >> PAGE_CACHE_SHIFT;
1688 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1689 unsigned long nr_pages = end_index - index + 1;
1692 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1696 while (nr_pages > 0) {
1697 ret = find_get_pages_contig(inode->i_mapping, index,
1698 min_t(unsigned long,
1699 nr_pages, ARRAY_SIZE(pages)), pages);
1700 for (i = 0; i < ret; i++) {
1702 if (page_ops & PAGE_SET_PRIVATE2)
1703 SetPagePrivate2(pages[i]);
1705 if (pages[i] == locked_page) {
1706 page_cache_release(pages[i]);
1709 if (page_ops & PAGE_CLEAR_DIRTY)
1710 clear_page_dirty_for_io(pages[i]);
1711 if (page_ops & PAGE_SET_WRITEBACK)
1712 set_page_writeback(pages[i]);
1713 if (page_ops & PAGE_END_WRITEBACK)
1714 end_page_writeback(pages[i]);
1715 if (page_ops & PAGE_UNLOCK)
1716 unlock_page(pages[i]);
1717 page_cache_release(pages[i]);
1727 * count the number of bytes in the tree that have a given bit(s)
1728 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1729 * cached. The total number found is returned.
1731 u64 count_range_bits(struct extent_io_tree *tree,
1732 u64 *start, u64 search_end, u64 max_bytes,
1733 unsigned long bits, int contig)
1735 struct rb_node *node;
1736 struct extent_state *state;
1737 u64 cur_start = *start;
1738 u64 total_bytes = 0;
1742 if (WARN_ON(search_end <= cur_start))
1745 spin_lock(&tree->lock);
1746 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1747 total_bytes = tree->dirty_bytes;
1751 * this search will find all the extents that end after
1754 node = tree_search(tree, cur_start);
1759 state = rb_entry(node, struct extent_state, rb_node);
1760 if (state->start > search_end)
1762 if (contig && found && state->start > last + 1)
1764 if (state->end >= cur_start && (state->state & bits) == bits) {
1765 total_bytes += min(search_end, state->end) + 1 -
1766 max(cur_start, state->start);
1767 if (total_bytes >= max_bytes)
1770 *start = max(cur_start, state->start);
1774 } else if (contig && found) {
1777 node = rb_next(node);
1782 spin_unlock(&tree->lock);
1787 * set the private field for a given byte offset in the tree. If there isn't
1788 * an extent_state there already, this does nothing.
1790 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1792 struct rb_node *node;
1793 struct extent_state *state;
1796 spin_lock(&tree->lock);
1798 * this search will find all the extents that end after
1801 node = tree_search(tree, start);
1806 state = rb_entry(node, struct extent_state, rb_node);
1807 if (state->start != start) {
1811 state->private = private;
1813 spin_unlock(&tree->lock);
1817 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1819 struct rb_node *node;
1820 struct extent_state *state;
1823 spin_lock(&tree->lock);
1825 * this search will find all the extents that end after
1828 node = tree_search(tree, start);
1833 state = rb_entry(node, struct extent_state, rb_node);
1834 if (state->start != start) {
1838 *private = state->private;
1840 spin_unlock(&tree->lock);
1845 * searches a range in the state tree for a given mask.
1846 * If 'filled' == 1, this returns 1 only if every extent in the tree
1847 * has the bits set. Otherwise, 1 is returned if any bit in the
1848 * range is found set.
1850 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1851 unsigned long bits, int filled, struct extent_state *cached)
1853 struct extent_state *state = NULL;
1854 struct rb_node *node;
1857 spin_lock(&tree->lock);
1858 if (cached && cached->tree && cached->start <= start &&
1859 cached->end > start)
1860 node = &cached->rb_node;
1862 node = tree_search(tree, start);
1863 while (node && start <= end) {
1864 state = rb_entry(node, struct extent_state, rb_node);
1866 if (filled && state->start > start) {
1871 if (state->start > end)
1874 if (state->state & bits) {
1878 } else if (filled) {
1883 if (state->end == (u64)-1)
1886 start = state->end + 1;
1889 node = rb_next(node);
1896 spin_unlock(&tree->lock);
1901 * helper function to set a given page up to date if all the
1902 * extents in the tree for that page are up to date
1904 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1906 u64 start = page_offset(page);
1907 u64 end = start + PAGE_CACHE_SIZE - 1;
1908 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1909 SetPageUptodate(page);
1913 * When IO fails, either with EIO or csum verification fails, we
1914 * try other mirrors that might have a good copy of the data. This
1915 * io_failure_record is used to record state as we go through all the
1916 * mirrors. If another mirror has good data, the page is set up to date
1917 * and things continue. If a good mirror can't be found, the original
1918 * bio end_io callback is called to indicate things have failed.
1920 struct io_failure_record {
1925 unsigned long bio_flags;
1931 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1936 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1938 set_state_private(failure_tree, rec->start, 0);
1939 ret = clear_extent_bits(failure_tree, rec->start,
1940 rec->start + rec->len - 1,
1941 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1945 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1946 rec->start + rec->len - 1,
1947 EXTENT_DAMAGED, GFP_NOFS);
1955 static void repair_io_failure_callback(struct bio *bio, int err)
1957 complete(bio->bi_private);
1961 * this bypasses the standard btrfs submit functions deliberately, as
1962 * the standard behavior is to write all copies in a raid setup. here we only
1963 * want to write the one bad copy. so we do the mapping for ourselves and issue
1964 * submit_bio directly.
1965 * to avoid any synchronization issues, wait for the data after writing, which
1966 * actually prevents the read that triggered the error from finishing.
1967 * currently, there can be no more than two copies of every data bit. thus,
1968 * exactly one rewrite is required.
1970 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
1971 u64 length, u64 logical, struct page *page,
1975 struct btrfs_device *dev;
1976 DECLARE_COMPLETION_ONSTACK(compl);
1979 struct btrfs_bio *bbio = NULL;
1980 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1983 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
1984 BUG_ON(!mirror_num);
1986 /* we can't repair anything in raid56 yet */
1987 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
1990 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1993 bio->bi_private = &compl;
1994 bio->bi_end_io = repair_io_failure_callback;
1996 map_length = length;
1998 ret = btrfs_map_block(fs_info, WRITE, logical,
1999 &map_length, &bbio, mirror_num);
2004 BUG_ON(mirror_num != bbio->mirror_num);
2005 sector = bbio->stripes[mirror_num-1].physical >> 9;
2006 bio->bi_sector = sector;
2007 dev = bbio->stripes[mirror_num-1].dev;
2009 if (!dev || !dev->bdev || !dev->writeable) {
2013 bio->bi_bdev = dev->bdev;
2014 bio_add_page(bio, page, length, start - page_offset(page));
2015 btrfsic_submit_bio(WRITE_SYNC, bio);
2016 wait_for_completion(&compl);
2018 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2019 /* try to remap that extent elsewhere? */
2021 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2025 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
2026 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2027 start, rcu_str_deref(dev->name), sector);
2033 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2036 u64 start = eb->start;
2037 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2040 if (root->fs_info->sb->s_flags & MS_RDONLY)
2043 for (i = 0; i < num_pages; i++) {
2044 struct page *p = extent_buffer_page(eb, i);
2045 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2046 start, p, mirror_num);
2049 start += PAGE_CACHE_SIZE;
2056 * each time an IO finishes, we do a fast check in the IO failure tree
2057 * to see if we need to process or clean up an io_failure_record
2059 static int clean_io_failure(u64 start, struct page *page)
2062 u64 private_failure;
2063 struct io_failure_record *failrec;
2064 struct inode *inode = page->mapping->host;
2065 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2066 struct extent_state *state;
2072 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2073 (u64)-1, 1, EXTENT_DIRTY, 0);
2077 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2082 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2083 BUG_ON(!failrec->this_mirror);
2085 if (failrec->in_validation) {
2086 /* there was no real error, just free the record */
2087 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2092 if (fs_info->sb->s_flags & MS_RDONLY)
2095 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2096 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2099 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2101 if (state && state->start <= failrec->start &&
2102 state->end >= failrec->start + failrec->len - 1) {
2103 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2105 if (num_copies > 1) {
2106 ret = repair_io_failure(fs_info, start, failrec->len,
2107 failrec->logical, page,
2108 failrec->failed_mirror);
2116 ret = free_io_failure(inode, failrec, did_repair);
2122 * this is a generic handler for readpage errors (default
2123 * readpage_io_failed_hook). if other copies exist, read those and write back
2124 * good data to the failed position. does not investigate in remapping the
2125 * failed extent elsewhere, hoping the device will be smart enough to do this as
2129 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2130 struct page *page, u64 start, u64 end,
2133 struct io_failure_record *failrec = NULL;
2135 struct extent_map *em;
2136 struct inode *inode = page->mapping->host;
2137 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2138 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2139 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2141 struct btrfs_io_bio *btrfs_failed_bio;
2142 struct btrfs_io_bio *btrfs_bio;
2148 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2150 ret = get_state_private(failure_tree, start, &private);
2152 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2155 failrec->start = start;
2156 failrec->len = end - start + 1;
2157 failrec->this_mirror = 0;
2158 failrec->bio_flags = 0;
2159 failrec->in_validation = 0;
2161 read_lock(&em_tree->lock);
2162 em = lookup_extent_mapping(em_tree, start, failrec->len);
2164 read_unlock(&em_tree->lock);
2169 if (em->start > start || em->start + em->len < start) {
2170 free_extent_map(em);
2173 read_unlock(&em_tree->lock);
2179 logical = start - em->start;
2180 logical = em->block_start + logical;
2181 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2182 logical = em->block_start;
2183 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2184 extent_set_compress_type(&failrec->bio_flags,
2187 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2188 "len=%llu\n", logical, start, failrec->len);
2189 failrec->logical = logical;
2190 free_extent_map(em);
2192 /* set the bits in the private failure tree */
2193 ret = set_extent_bits(failure_tree, start, end,
2194 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2196 ret = set_state_private(failure_tree, start,
2197 (u64)(unsigned long)failrec);
2198 /* set the bits in the inode's tree */
2200 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2207 failrec = (struct io_failure_record *)(unsigned long)private;
2208 pr_debug("bio_readpage_error: (found) logical=%llu, "
2209 "start=%llu, len=%llu, validation=%d\n",
2210 failrec->logical, failrec->start, failrec->len,
2211 failrec->in_validation);
2213 * when data can be on disk more than twice, add to failrec here
2214 * (e.g. with a list for failed_mirror) to make
2215 * clean_io_failure() clean all those errors at once.
2218 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2219 failrec->logical, failrec->len);
2220 if (num_copies == 1) {
2222 * we only have a single copy of the data, so don't bother with
2223 * all the retry and error correction code that follows. no
2224 * matter what the error is, it is very likely to persist.
2226 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2227 num_copies, failrec->this_mirror, failed_mirror);
2228 free_io_failure(inode, failrec, 0);
2233 * there are two premises:
2234 * a) deliver good data to the caller
2235 * b) correct the bad sectors on disk
2237 if (failed_bio->bi_vcnt > 1) {
2239 * to fulfill b), we need to know the exact failing sectors, as
2240 * we don't want to rewrite any more than the failed ones. thus,
2241 * we need separate read requests for the failed bio
2243 * if the following BUG_ON triggers, our validation request got
2244 * merged. we need separate requests for our algorithm to work.
2246 BUG_ON(failrec->in_validation);
2247 failrec->in_validation = 1;
2248 failrec->this_mirror = failed_mirror;
2249 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2252 * we're ready to fulfill a) and b) alongside. get a good copy
2253 * of the failed sector and if we succeed, we have setup
2254 * everything for repair_io_failure to do the rest for us.
2256 if (failrec->in_validation) {
2257 BUG_ON(failrec->this_mirror != failed_mirror);
2258 failrec->in_validation = 0;
2259 failrec->this_mirror = 0;
2261 failrec->failed_mirror = failed_mirror;
2262 failrec->this_mirror++;
2263 if (failrec->this_mirror == failed_mirror)
2264 failrec->this_mirror++;
2265 read_mode = READ_SYNC;
2268 if (failrec->this_mirror > num_copies) {
2269 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2270 num_copies, failrec->this_mirror, failed_mirror);
2271 free_io_failure(inode, failrec, 0);
2275 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2277 free_io_failure(inode, failrec, 0);
2280 bio->bi_end_io = failed_bio->bi_end_io;
2281 bio->bi_sector = failrec->logical >> 9;
2282 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2285 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2286 if (btrfs_failed_bio->csum) {
2287 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2288 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2290 btrfs_bio = btrfs_io_bio(bio);
2291 btrfs_bio->csum = btrfs_bio->csum_inline;
2292 phy_offset >>= inode->i_sb->s_blocksize_bits;
2293 phy_offset *= csum_size;
2294 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset,
2298 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2300 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2301 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2302 failrec->this_mirror, num_copies, failrec->in_validation);
2304 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2305 failrec->this_mirror,
2306 failrec->bio_flags, 0);
2310 /* lots and lots of room for performance fixes in the end_bio funcs */
2312 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2314 int uptodate = (err == 0);
2315 struct extent_io_tree *tree;
2318 tree = &BTRFS_I(page->mapping->host)->io_tree;
2320 if (tree->ops && tree->ops->writepage_end_io_hook) {
2321 ret = tree->ops->writepage_end_io_hook(page, start,
2322 end, NULL, uptodate);
2328 ClearPageUptodate(page);
2335 * after a writepage IO is done, we need to:
2336 * clear the uptodate bits on error
2337 * clear the writeback bits in the extent tree for this IO
2338 * end_page_writeback if the page has no more pending IO
2340 * Scheduling is not allowed, so the extent state tree is expected
2341 * to have one and only one object corresponding to this IO.
2343 static void end_bio_extent_writepage(struct bio *bio, int err)
2345 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2346 struct extent_io_tree *tree;
2351 struct page *page = bvec->bv_page;
2352 tree = &BTRFS_I(page->mapping->host)->io_tree;
2354 /* We always issue full-page reads, but if some block
2355 * in a page fails to read, blk_update_request() will
2356 * advance bv_offset and adjust bv_len to compensate.
2357 * Print a warning for nonzero offsets, and an error
2358 * if they don't add up to a full page. */
2359 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2360 printk("%s page write in btrfs with offset %u and length %u\n",
2361 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2362 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2363 bvec->bv_offset, bvec->bv_len);
2365 start = page_offset(page);
2366 end = start + bvec->bv_offset + bvec->bv_len - 1;
2368 if (--bvec >= bio->bi_io_vec)
2369 prefetchw(&bvec->bv_page->flags);
2371 if (end_extent_writepage(page, err, start, end))
2374 end_page_writeback(page);
2375 } while (bvec >= bio->bi_io_vec);
2381 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2384 struct extent_state *cached = NULL;
2385 u64 end = start + len - 1;
2387 if (uptodate && tree->track_uptodate)
2388 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2389 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2393 * after a readpage IO is done, we need to:
2394 * clear the uptodate bits on error
2395 * set the uptodate bits if things worked
2396 * set the page up to date if all extents in the tree are uptodate
2397 * clear the lock bit in the extent tree
2398 * unlock the page if there are no other extents locked for it
2400 * Scheduling is not allowed, so the extent state tree is expected
2401 * to have one and only one object corresponding to this IO.
2403 static void end_bio_extent_readpage(struct bio *bio, int err)
2405 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2406 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2407 struct bio_vec *bvec = bio->bi_io_vec;
2408 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2409 struct extent_io_tree *tree;
2414 u64 extent_start = 0;
2423 struct page *page = bvec->bv_page;
2424 struct inode *inode = page->mapping->host;
2426 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2427 "mirror=%lu\n", (u64)bio->bi_sector, err,
2428 io_bio->mirror_num);
2429 tree = &BTRFS_I(inode)->io_tree;
2431 /* We always issue full-page reads, but if some block
2432 * in a page fails to read, blk_update_request() will
2433 * advance bv_offset and adjust bv_len to compensate.
2434 * Print a warning for nonzero offsets, and an error
2435 * if they don't add up to a full page. */
2436 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2437 printk("%s page read in btrfs with offset %u and length %u\n",
2438 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2439 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2440 bvec->bv_offset, bvec->bv_len);
2442 start = page_offset(page);
2443 end = start + bvec->bv_offset + bvec->bv_len - 1;
2446 if (++bvec <= bvec_end)
2447 prefetchw(&bvec->bv_page->flags);
2449 mirror = io_bio->mirror_num;
2450 if (likely(uptodate && tree->ops &&
2451 tree->ops->readpage_end_io_hook)) {
2452 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2458 clean_io_failure(start, page);
2461 if (likely(uptodate))
2464 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2465 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2467 test_bit(BIO_UPTODATE, &bio->bi_flags))
2471 * The generic bio_readpage_error handles errors the
2472 * following way: If possible, new read requests are
2473 * created and submitted and will end up in
2474 * end_bio_extent_readpage as well (if we're lucky, not
2475 * in the !uptodate case). In that case it returns 0 and
2476 * we just go on with the next page in our bio. If it
2477 * can't handle the error it will return -EIO and we
2478 * remain responsible for that page.
2480 ret = bio_readpage_error(bio, offset, page, start, end,
2484 test_bit(BIO_UPTODATE, &bio->bi_flags);
2491 if (likely(uptodate)) {
2492 loff_t i_size = i_size_read(inode);
2493 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2496 /* Zero out the end if this page straddles i_size */
2497 offset = i_size & (PAGE_CACHE_SIZE-1);
2498 if (page->index == end_index && offset)
2499 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2500 SetPageUptodate(page);
2502 ClearPageUptodate(page);
2508 if (unlikely(!uptodate)) {
2510 endio_readpage_release_extent(tree,
2516 endio_readpage_release_extent(tree, start,
2517 end - start + 1, 0);
2518 } else if (!extent_len) {
2519 extent_start = start;
2520 extent_len = end + 1 - start;
2521 } else if (extent_start + extent_len == start) {
2522 extent_len += end + 1 - start;
2524 endio_readpage_release_extent(tree, extent_start,
2525 extent_len, uptodate);
2526 extent_start = start;
2527 extent_len = end + 1 - start;
2529 } while (bvec <= bvec_end);
2532 endio_readpage_release_extent(tree, extent_start, extent_len,
2535 io_bio->end_io(io_bio, err);
2540 * this allocates from the btrfs_bioset. We're returning a bio right now
2541 * but you can call btrfs_io_bio for the appropriate container_of magic
2544 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2547 struct btrfs_io_bio *btrfs_bio;
2550 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2552 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2553 while (!bio && (nr_vecs /= 2)) {
2554 bio = bio_alloc_bioset(gfp_flags,
2555 nr_vecs, btrfs_bioset);
2561 bio->bi_bdev = bdev;
2562 bio->bi_sector = first_sector;
2563 btrfs_bio = btrfs_io_bio(bio);
2564 btrfs_bio->csum = NULL;
2565 btrfs_bio->csum_allocated = NULL;
2566 btrfs_bio->end_io = NULL;
2571 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2573 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2577 /* this also allocates from the btrfs_bioset */
2578 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2580 struct btrfs_io_bio *btrfs_bio;
2583 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2585 btrfs_bio = btrfs_io_bio(bio);
2586 btrfs_bio->csum = NULL;
2587 btrfs_bio->csum_allocated = NULL;
2588 btrfs_bio->end_io = NULL;
2594 static int __must_check submit_one_bio(int rw, struct bio *bio,
2595 int mirror_num, unsigned long bio_flags)
2598 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2599 struct page *page = bvec->bv_page;
2600 struct extent_io_tree *tree = bio->bi_private;
2603 start = page_offset(page) + bvec->bv_offset;
2605 bio->bi_private = NULL;
2609 if (tree->ops && tree->ops->submit_bio_hook)
2610 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2611 mirror_num, bio_flags, start);
2613 btrfsic_submit_bio(rw, bio);
2615 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2621 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2622 unsigned long offset, size_t size, struct bio *bio,
2623 unsigned long bio_flags)
2626 if (tree->ops && tree->ops->merge_bio_hook)
2627 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2634 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2635 struct page *page, sector_t sector,
2636 size_t size, unsigned long offset,
2637 struct block_device *bdev,
2638 struct bio **bio_ret,
2639 unsigned long max_pages,
2640 bio_end_io_t end_io_func,
2642 unsigned long prev_bio_flags,
2643 unsigned long bio_flags)
2649 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2650 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2651 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2653 if (bio_ret && *bio_ret) {
2656 contig = bio->bi_sector == sector;
2658 contig = bio_end_sector(bio) == sector;
2660 if (prev_bio_flags != bio_flags || !contig ||
2661 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2662 bio_add_page(bio, page, page_size, offset) < page_size) {
2663 ret = submit_one_bio(rw, bio, mirror_num,
2672 if (this_compressed)
2675 nr = bio_get_nr_vecs(bdev);
2677 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2681 bio_add_page(bio, page, page_size, offset);
2682 bio->bi_end_io = end_io_func;
2683 bio->bi_private = tree;
2688 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2693 static void attach_extent_buffer_page(struct extent_buffer *eb,
2696 if (!PagePrivate(page)) {
2697 SetPagePrivate(page);
2698 page_cache_get(page);
2699 set_page_private(page, (unsigned long)eb);
2701 WARN_ON(page->private != (unsigned long)eb);
2705 void set_page_extent_mapped(struct page *page)
2707 if (!PagePrivate(page)) {
2708 SetPagePrivate(page);
2709 page_cache_get(page);
2710 set_page_private(page, EXTENT_PAGE_PRIVATE);
2714 static struct extent_map *
2715 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2716 u64 start, u64 len, get_extent_t *get_extent,
2717 struct extent_map **em_cached)
2719 struct extent_map *em;
2721 if (em_cached && *em_cached) {
2723 if (em->in_tree && start >= em->start &&
2724 start < extent_map_end(em)) {
2725 atomic_inc(&em->refs);
2729 free_extent_map(em);
2733 em = get_extent(inode, page, pg_offset, start, len, 0);
2734 if (em_cached && !IS_ERR_OR_NULL(em)) {
2736 atomic_inc(&em->refs);
2742 * basic readpage implementation. Locked extent state structs are inserted
2743 * into the tree that are removed when the IO is done (by the end_io
2745 * XXX JDM: This needs looking at to ensure proper page locking
2747 static int __do_readpage(struct extent_io_tree *tree,
2749 get_extent_t *get_extent,
2750 struct extent_map **em_cached,
2751 struct bio **bio, int mirror_num,
2752 unsigned long *bio_flags, int rw)
2754 struct inode *inode = page->mapping->host;
2755 u64 start = page_offset(page);
2756 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2760 u64 last_byte = i_size_read(inode);
2764 struct extent_map *em;
2765 struct block_device *bdev;
2768 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2769 size_t pg_offset = 0;
2771 size_t disk_io_size;
2772 size_t blocksize = inode->i_sb->s_blocksize;
2773 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2775 set_page_extent_mapped(page);
2778 if (!PageUptodate(page)) {
2779 if (cleancache_get_page(page) == 0) {
2780 BUG_ON(blocksize != PAGE_SIZE);
2781 unlock_extent(tree, start, end);
2786 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2788 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2791 iosize = PAGE_CACHE_SIZE - zero_offset;
2792 userpage = kmap_atomic(page);
2793 memset(userpage + zero_offset, 0, iosize);
2794 flush_dcache_page(page);
2795 kunmap_atomic(userpage);
2798 while (cur <= end) {
2799 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2801 if (cur >= last_byte) {
2803 struct extent_state *cached = NULL;
2805 iosize = PAGE_CACHE_SIZE - pg_offset;
2806 userpage = kmap_atomic(page);
2807 memset(userpage + pg_offset, 0, iosize);
2808 flush_dcache_page(page);
2809 kunmap_atomic(userpage);
2810 set_extent_uptodate(tree, cur, cur + iosize - 1,
2813 unlock_extent_cached(tree, cur,
2818 em = __get_extent_map(inode, page, pg_offset, cur,
2819 end - cur + 1, get_extent, em_cached);
2820 if (IS_ERR_OR_NULL(em)) {
2823 unlock_extent(tree, cur, end);
2826 extent_offset = cur - em->start;
2827 BUG_ON(extent_map_end(em) <= cur);
2830 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2831 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2832 extent_set_compress_type(&this_bio_flag,
2836 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2837 cur_end = min(extent_map_end(em) - 1, end);
2838 iosize = ALIGN(iosize, blocksize);
2839 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2840 disk_io_size = em->block_len;
2841 sector = em->block_start >> 9;
2843 sector = (em->block_start + extent_offset) >> 9;
2844 disk_io_size = iosize;
2847 block_start = em->block_start;
2848 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2849 block_start = EXTENT_MAP_HOLE;
2850 free_extent_map(em);
2853 /* we've found a hole, just zero and go on */
2854 if (block_start == EXTENT_MAP_HOLE) {
2856 struct extent_state *cached = NULL;
2858 userpage = kmap_atomic(page);
2859 memset(userpage + pg_offset, 0, iosize);
2860 flush_dcache_page(page);
2861 kunmap_atomic(userpage);
2863 set_extent_uptodate(tree, cur, cur + iosize - 1,
2865 unlock_extent_cached(tree, cur, cur + iosize - 1,
2868 pg_offset += iosize;
2871 /* the get_extent function already copied into the page */
2872 if (test_range_bit(tree, cur, cur_end,
2873 EXTENT_UPTODATE, 1, NULL)) {
2874 check_page_uptodate(tree, page);
2876 unlock_extent(tree, cur, cur + iosize - 1);
2878 pg_offset += iosize;
2881 /* we have an inline extent but it didn't get marked up
2882 * to date. Error out
2884 if (block_start == EXTENT_MAP_INLINE) {
2887 unlock_extent(tree, cur, cur + iosize - 1);
2889 pg_offset += iosize;
2894 ret = submit_extent_page(rw, tree, page,
2895 sector, disk_io_size, pg_offset,
2897 end_bio_extent_readpage, mirror_num,
2902 *bio_flags = this_bio_flag;
2906 unlock_extent(tree, cur, cur + iosize - 1);
2909 pg_offset += iosize;
2913 if (!PageError(page))
2914 SetPageUptodate(page);
2920 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
2921 struct page *pages[], int nr_pages,
2923 get_extent_t *get_extent,
2924 struct extent_map **em_cached,
2925 struct bio **bio, int mirror_num,
2926 unsigned long *bio_flags, int rw)
2928 struct inode *inode;
2929 struct btrfs_ordered_extent *ordered;
2932 inode = pages[0]->mapping->host;
2934 lock_extent(tree, start, end);
2935 ordered = btrfs_lookup_ordered_range(inode, start,
2939 unlock_extent(tree, start, end);
2940 btrfs_start_ordered_extent(inode, ordered, 1);
2941 btrfs_put_ordered_extent(ordered);
2944 for (index = 0; index < nr_pages; index++) {
2945 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
2946 mirror_num, bio_flags, rw);
2947 page_cache_release(pages[index]);
2951 static void __extent_readpages(struct extent_io_tree *tree,
2952 struct page *pages[],
2953 int nr_pages, get_extent_t *get_extent,
2954 struct extent_map **em_cached,
2955 struct bio **bio, int mirror_num,
2956 unsigned long *bio_flags, int rw)
2962 int first_index = 0;
2964 for (index = 0; index < nr_pages; index++) {
2965 page_start = page_offset(pages[index]);
2968 end = start + PAGE_CACHE_SIZE - 1;
2969 first_index = index;
2970 } else if (end + 1 == page_start) {
2971 end += PAGE_CACHE_SIZE;
2973 __do_contiguous_readpages(tree, &pages[first_index],
2974 index - first_index, start,
2975 end, get_extent, em_cached,
2976 bio, mirror_num, bio_flags,
2979 end = start + PAGE_CACHE_SIZE - 1;
2980 first_index = index;
2985 __do_contiguous_readpages(tree, &pages[first_index],
2986 index - first_index, start,
2987 end, get_extent, em_cached, bio,
2988 mirror_num, bio_flags, rw);
2991 static int __extent_read_full_page(struct extent_io_tree *tree,
2993 get_extent_t *get_extent,
2994 struct bio **bio, int mirror_num,
2995 unsigned long *bio_flags, int rw)
2997 struct inode *inode = page->mapping->host;
2998 struct btrfs_ordered_extent *ordered;
2999 u64 start = page_offset(page);
3000 u64 end = start + PAGE_CACHE_SIZE - 1;
3004 lock_extent(tree, start, end);
3005 ordered = btrfs_lookup_ordered_extent(inode, start);
3008 unlock_extent(tree, start, end);
3009 btrfs_start_ordered_extent(inode, ordered, 1);
3010 btrfs_put_ordered_extent(ordered);
3013 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3018 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3019 get_extent_t *get_extent, int mirror_num)
3021 struct bio *bio = NULL;
3022 unsigned long bio_flags = 0;
3025 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3028 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3032 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3033 get_extent_t *get_extent, int mirror_num)
3035 struct bio *bio = NULL;
3036 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3039 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3042 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3046 static noinline void update_nr_written(struct page *page,
3047 struct writeback_control *wbc,
3048 unsigned long nr_written)
3050 wbc->nr_to_write -= nr_written;
3051 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3052 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3053 page->mapping->writeback_index = page->index + nr_written;
3057 * the writepage semantics are similar to regular writepage. extent
3058 * records are inserted to lock ranges in the tree, and as dirty areas
3059 * are found, they are marked writeback. Then the lock bits are removed
3060 * and the end_io handler clears the writeback ranges
3062 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3065 struct inode *inode = page->mapping->host;
3066 struct extent_page_data *epd = data;
3067 struct extent_io_tree *tree = epd->tree;
3068 u64 start = page_offset(page);
3070 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3074 u64 last_byte = i_size_read(inode);
3078 struct extent_state *cached_state = NULL;
3079 struct extent_map *em;
3080 struct block_device *bdev;
3083 size_t pg_offset = 0;
3085 loff_t i_size = i_size_read(inode);
3086 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3092 unsigned long nr_written = 0;
3093 bool fill_delalloc = true;
3095 if (wbc->sync_mode == WB_SYNC_ALL)
3096 write_flags = WRITE_SYNC;
3098 write_flags = WRITE;
3100 trace___extent_writepage(page, inode, wbc);
3102 WARN_ON(!PageLocked(page));
3104 ClearPageError(page);
3106 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3107 if (page->index > end_index ||
3108 (page->index == end_index && !pg_offset)) {
3109 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3114 if (page->index == end_index) {
3117 userpage = kmap_atomic(page);
3118 memset(userpage + pg_offset, 0,
3119 PAGE_CACHE_SIZE - pg_offset);
3120 kunmap_atomic(userpage);
3121 flush_dcache_page(page);
3125 set_page_extent_mapped(page);
3127 if (!tree->ops || !tree->ops->fill_delalloc)
3128 fill_delalloc = false;
3130 delalloc_start = start;
3133 if (!epd->extent_locked && fill_delalloc) {
3134 u64 delalloc_to_write = 0;
3136 * make sure the wbc mapping index is at least updated
3139 update_nr_written(page, wbc, 0);
3141 while (delalloc_end < page_end) {
3142 nr_delalloc = find_lock_delalloc_range(inode, tree,
3147 if (nr_delalloc == 0) {
3148 delalloc_start = delalloc_end + 1;
3151 ret = tree->ops->fill_delalloc(inode, page,
3156 /* File system has been set read-only */
3162 * delalloc_end is already one less than the total
3163 * length, so we don't subtract one from
3166 delalloc_to_write += (delalloc_end - delalloc_start +
3169 delalloc_start = delalloc_end + 1;
3171 if (wbc->nr_to_write < delalloc_to_write) {
3174 if (delalloc_to_write < thresh * 2)
3175 thresh = delalloc_to_write;
3176 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3180 /* did the fill delalloc function already unlock and start
3186 * we've unlocked the page, so we can't update
3187 * the mapping's writeback index, just update
3190 wbc->nr_to_write -= nr_written;
3194 if (tree->ops && tree->ops->writepage_start_hook) {
3195 ret = tree->ops->writepage_start_hook(page, start,
3198 /* Fixup worker will requeue */
3200 wbc->pages_skipped++;
3202 redirty_page_for_writepage(wbc, page);
3203 update_nr_written(page, wbc, nr_written);
3211 * we don't want to touch the inode after unlocking the page,
3212 * so we update the mapping writeback index now
3214 update_nr_written(page, wbc, nr_written + 1);
3217 if (last_byte <= start) {
3218 if (tree->ops && tree->ops->writepage_end_io_hook)
3219 tree->ops->writepage_end_io_hook(page, start,
3224 blocksize = inode->i_sb->s_blocksize;
3226 while (cur <= end) {
3227 if (cur >= last_byte) {
3228 if (tree->ops && tree->ops->writepage_end_io_hook)
3229 tree->ops->writepage_end_io_hook(page, cur,
3233 em = epd->get_extent(inode, page, pg_offset, cur,
3235 if (IS_ERR_OR_NULL(em)) {
3240 extent_offset = cur - em->start;
3241 BUG_ON(extent_map_end(em) <= cur);
3243 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3244 iosize = ALIGN(iosize, blocksize);
3245 sector = (em->block_start + extent_offset) >> 9;
3247 block_start = em->block_start;
3248 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3249 free_extent_map(em);
3253 * compressed and inline extents are written through other
3256 if (compressed || block_start == EXTENT_MAP_HOLE ||
3257 block_start == EXTENT_MAP_INLINE) {
3259 * end_io notification does not happen here for
3260 * compressed extents
3262 if (!compressed && tree->ops &&
3263 tree->ops->writepage_end_io_hook)
3264 tree->ops->writepage_end_io_hook(page, cur,
3267 else if (compressed) {
3268 /* we don't want to end_page_writeback on
3269 * a compressed extent. this happens
3276 pg_offset += iosize;
3279 /* leave this out until we have a page_mkwrite call */
3280 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3281 EXTENT_DIRTY, 0, NULL)) {
3283 pg_offset += iosize;
3287 if (tree->ops && tree->ops->writepage_io_hook) {
3288 ret = tree->ops->writepage_io_hook(page, cur,
3296 unsigned long max_nr = end_index + 1;
3298 set_range_writeback(tree, cur, cur + iosize - 1);
3299 if (!PageWriteback(page)) {
3300 printk(KERN_ERR "btrfs warning page %lu not "
3301 "writeback, cur %llu end %llu\n",
3302 page->index, cur, end);
3305 ret = submit_extent_page(write_flags, tree, page,
3306 sector, iosize, pg_offset,
3307 bdev, &epd->bio, max_nr,
3308 end_bio_extent_writepage,
3314 pg_offset += iosize;
3319 /* make sure the mapping tag for page dirty gets cleared */
3320 set_page_writeback(page);
3321 end_page_writeback(page);
3327 /* drop our reference on any cached states */
3328 free_extent_state(cached_state);
3332 static int eb_wait(void *word)
3338 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3340 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3341 TASK_UNINTERRUPTIBLE);
3344 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3345 struct btrfs_fs_info *fs_info,
3346 struct extent_page_data *epd)
3348 unsigned long i, num_pages;
3352 if (!btrfs_try_tree_write_lock(eb)) {
3354 flush_write_bio(epd);
3355 btrfs_tree_lock(eb);
3358 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3359 btrfs_tree_unlock(eb);
3363 flush_write_bio(epd);
3367 wait_on_extent_buffer_writeback(eb);
3368 btrfs_tree_lock(eb);
3369 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3371 btrfs_tree_unlock(eb);
3376 * We need to do this to prevent races in people who check if the eb is
3377 * under IO since we can end up having no IO bits set for a short period
3380 spin_lock(&eb->refs_lock);
3381 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3382 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3383 spin_unlock(&eb->refs_lock);
3384 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3385 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3387 fs_info->dirty_metadata_batch);
3390 spin_unlock(&eb->refs_lock);
3393 btrfs_tree_unlock(eb);
3398 num_pages = num_extent_pages(eb->start, eb->len);
3399 for (i = 0; i < num_pages; i++) {
3400 struct page *p = extent_buffer_page(eb, i);
3402 if (!trylock_page(p)) {
3404 flush_write_bio(epd);
3414 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3416 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3417 smp_mb__after_clear_bit();
3418 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3421 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3423 int uptodate = err == 0;
3424 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3425 struct extent_buffer *eb;
3429 struct page *page = bvec->bv_page;
3432 eb = (struct extent_buffer *)page->private;
3434 done = atomic_dec_and_test(&eb->io_pages);
3436 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3437 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3438 ClearPageUptodate(page);
3442 end_page_writeback(page);
3447 end_extent_buffer_writeback(eb);
3448 } while (bvec >= bio->bi_io_vec);
3454 static int write_one_eb(struct extent_buffer *eb,
3455 struct btrfs_fs_info *fs_info,
3456 struct writeback_control *wbc,
3457 struct extent_page_data *epd)
3459 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3460 u64 offset = eb->start;
3461 unsigned long i, num_pages;
3462 unsigned long bio_flags = 0;
3463 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3466 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3467 num_pages = num_extent_pages(eb->start, eb->len);
3468 atomic_set(&eb->io_pages, num_pages);
3469 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3470 bio_flags = EXTENT_BIO_TREE_LOG;
3472 for (i = 0; i < num_pages; i++) {
3473 struct page *p = extent_buffer_page(eb, i);
3475 clear_page_dirty_for_io(p);
3476 set_page_writeback(p);
3477 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3478 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3479 -1, end_bio_extent_buffer_writepage,
3480 0, epd->bio_flags, bio_flags);
3481 epd->bio_flags = bio_flags;
3483 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3485 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3486 end_extent_buffer_writeback(eb);
3490 offset += PAGE_CACHE_SIZE;
3491 update_nr_written(p, wbc, 1);
3495 if (unlikely(ret)) {
3496 for (; i < num_pages; i++) {
3497 struct page *p = extent_buffer_page(eb, i);
3505 int btree_write_cache_pages(struct address_space *mapping,
3506 struct writeback_control *wbc)
3508 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3509 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3510 struct extent_buffer *eb, *prev_eb = NULL;
3511 struct extent_page_data epd = {
3515 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3520 int nr_to_write_done = 0;
3521 struct pagevec pvec;
3524 pgoff_t end; /* Inclusive */
3528 pagevec_init(&pvec, 0);
3529 if (wbc->range_cyclic) {
3530 index = mapping->writeback_index; /* Start from prev offset */
3533 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3534 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3537 if (wbc->sync_mode == WB_SYNC_ALL)
3538 tag = PAGECACHE_TAG_TOWRITE;
3540 tag = PAGECACHE_TAG_DIRTY;
3542 if (wbc->sync_mode == WB_SYNC_ALL)
3543 tag_pages_for_writeback(mapping, index, end);
3544 while (!done && !nr_to_write_done && (index <= end) &&
3545 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3546 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3550 for (i = 0; i < nr_pages; i++) {
3551 struct page *page = pvec.pages[i];
3553 if (!PagePrivate(page))
3556 if (!wbc->range_cyclic && page->index > end) {
3561 spin_lock(&mapping->private_lock);
3562 if (!PagePrivate(page)) {
3563 spin_unlock(&mapping->private_lock);
3567 eb = (struct extent_buffer *)page->private;
3570 * Shouldn't happen and normally this would be a BUG_ON
3571 * but no sense in crashing the users box for something
3572 * we can survive anyway.
3575 spin_unlock(&mapping->private_lock);
3579 if (eb == prev_eb) {
3580 spin_unlock(&mapping->private_lock);
3584 ret = atomic_inc_not_zero(&eb->refs);
3585 spin_unlock(&mapping->private_lock);
3590 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3592 free_extent_buffer(eb);
3596 ret = write_one_eb(eb, fs_info, wbc, &epd);
3599 free_extent_buffer(eb);
3602 free_extent_buffer(eb);
3605 * the filesystem may choose to bump up nr_to_write.
3606 * We have to make sure to honor the new nr_to_write
3609 nr_to_write_done = wbc->nr_to_write <= 0;
3611 pagevec_release(&pvec);
3614 if (!scanned && !done) {
3616 * We hit the last page and there is more work to be done: wrap
3617 * back to the start of the file
3623 flush_write_bio(&epd);
3628 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3629 * @mapping: address space structure to write
3630 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3631 * @writepage: function called for each page
3632 * @data: data passed to writepage function
3634 * If a page is already under I/O, write_cache_pages() skips it, even
3635 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3636 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3637 * and msync() need to guarantee that all the data which was dirty at the time
3638 * the call was made get new I/O started against them. If wbc->sync_mode is
3639 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3640 * existing IO to complete.
3642 static int extent_write_cache_pages(struct extent_io_tree *tree,
3643 struct address_space *mapping,
3644 struct writeback_control *wbc,
3645 writepage_t writepage, void *data,
3646 void (*flush_fn)(void *))
3648 struct inode *inode = mapping->host;
3651 int nr_to_write_done = 0;
3652 struct pagevec pvec;
3655 pgoff_t end; /* Inclusive */
3660 * We have to hold onto the inode so that ordered extents can do their
3661 * work when the IO finishes. The alternative to this is failing to add
3662 * an ordered extent if the igrab() fails there and that is a huge pain
3663 * to deal with, so instead just hold onto the inode throughout the
3664 * writepages operation. If it fails here we are freeing up the inode
3665 * anyway and we'd rather not waste our time writing out stuff that is
3666 * going to be truncated anyway.
3671 pagevec_init(&pvec, 0);
3672 if (wbc->range_cyclic) {
3673 index = mapping->writeback_index; /* Start from prev offset */
3676 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3677 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3680 if (wbc->sync_mode == WB_SYNC_ALL)
3681 tag = PAGECACHE_TAG_TOWRITE;
3683 tag = PAGECACHE_TAG_DIRTY;
3685 if (wbc->sync_mode == WB_SYNC_ALL)
3686 tag_pages_for_writeback(mapping, index, end);
3687 while (!done && !nr_to_write_done && (index <= end) &&
3688 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3689 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3693 for (i = 0; i < nr_pages; i++) {
3694 struct page *page = pvec.pages[i];
3697 * At this point we hold neither mapping->tree_lock nor
3698 * lock on the page itself: the page may be truncated or
3699 * invalidated (changing page->mapping to NULL), or even
3700 * swizzled back from swapper_space to tmpfs file
3703 if (!trylock_page(page)) {
3708 if (unlikely(page->mapping != mapping)) {
3713 if (!wbc->range_cyclic && page->index > end) {
3719 if (wbc->sync_mode != WB_SYNC_NONE) {
3720 if (PageWriteback(page))
3722 wait_on_page_writeback(page);
3725 if (PageWriteback(page) ||
3726 !clear_page_dirty_for_io(page)) {
3731 ret = (*writepage)(page, wbc, data);
3733 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3741 * the filesystem may choose to bump up nr_to_write.
3742 * We have to make sure to honor the new nr_to_write
3745 nr_to_write_done = wbc->nr_to_write <= 0;
3747 pagevec_release(&pvec);
3750 if (!scanned && !done) {
3752 * We hit the last page and there is more work to be done: wrap
3753 * back to the start of the file
3759 btrfs_add_delayed_iput(inode);
3763 static void flush_epd_write_bio(struct extent_page_data *epd)
3772 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3773 BUG_ON(ret < 0); /* -ENOMEM */
3778 static noinline void flush_write_bio(void *data)
3780 struct extent_page_data *epd = data;
3781 flush_epd_write_bio(epd);
3784 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3785 get_extent_t *get_extent,
3786 struct writeback_control *wbc)
3789 struct extent_page_data epd = {
3792 .get_extent = get_extent,
3794 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3798 ret = __extent_writepage(page, wbc, &epd);
3800 flush_epd_write_bio(&epd);
3804 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3805 u64 start, u64 end, get_extent_t *get_extent,
3809 struct address_space *mapping = inode->i_mapping;
3811 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3814 struct extent_page_data epd = {
3817 .get_extent = get_extent,
3819 .sync_io = mode == WB_SYNC_ALL,
3822 struct writeback_control wbc_writepages = {
3824 .nr_to_write = nr_pages * 2,
3825 .range_start = start,
3826 .range_end = end + 1,
3829 while (start <= end) {
3830 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3831 if (clear_page_dirty_for_io(page))
3832 ret = __extent_writepage(page, &wbc_writepages, &epd);
3834 if (tree->ops && tree->ops->writepage_end_io_hook)
3835 tree->ops->writepage_end_io_hook(page, start,
3836 start + PAGE_CACHE_SIZE - 1,
3840 page_cache_release(page);
3841 start += PAGE_CACHE_SIZE;
3844 flush_epd_write_bio(&epd);
3848 int extent_writepages(struct extent_io_tree *tree,
3849 struct address_space *mapping,
3850 get_extent_t *get_extent,
3851 struct writeback_control *wbc)
3854 struct extent_page_data epd = {
3857 .get_extent = get_extent,
3859 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3863 ret = extent_write_cache_pages(tree, mapping, wbc,
3864 __extent_writepage, &epd,
3866 flush_epd_write_bio(&epd);
3870 int extent_readpages(struct extent_io_tree *tree,
3871 struct address_space *mapping,
3872 struct list_head *pages, unsigned nr_pages,
3873 get_extent_t get_extent)
3875 struct bio *bio = NULL;
3877 unsigned long bio_flags = 0;
3878 struct page *pagepool[16];
3880 struct extent_map *em_cached = NULL;
3883 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3884 page = list_entry(pages->prev, struct page, lru);
3886 prefetchw(&page->flags);
3887 list_del(&page->lru);
3888 if (add_to_page_cache_lru(page, mapping,
3889 page->index, GFP_NOFS)) {
3890 page_cache_release(page);
3894 pagepool[nr++] = page;
3895 if (nr < ARRAY_SIZE(pagepool))
3897 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3898 &bio, 0, &bio_flags, READ);
3902 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3903 &bio, 0, &bio_flags, READ);
3906 free_extent_map(em_cached);
3908 BUG_ON(!list_empty(pages));
3910 return submit_one_bio(READ, bio, 0, bio_flags);
3915 * basic invalidatepage code, this waits on any locked or writeback
3916 * ranges corresponding to the page, and then deletes any extent state
3917 * records from the tree
3919 int extent_invalidatepage(struct extent_io_tree *tree,
3920 struct page *page, unsigned long offset)
3922 struct extent_state *cached_state = NULL;
3923 u64 start = page_offset(page);
3924 u64 end = start + PAGE_CACHE_SIZE - 1;
3925 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3927 start += ALIGN(offset, blocksize);
3931 lock_extent_bits(tree, start, end, 0, &cached_state);
3932 wait_on_page_writeback(page);
3933 clear_extent_bit(tree, start, end,
3934 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3935 EXTENT_DO_ACCOUNTING,
3936 1, 1, &cached_state, GFP_NOFS);
3941 * a helper for releasepage, this tests for areas of the page that
3942 * are locked or under IO and drops the related state bits if it is safe
3945 static int try_release_extent_state(struct extent_map_tree *map,
3946 struct extent_io_tree *tree,
3947 struct page *page, gfp_t mask)
3949 u64 start = page_offset(page);
3950 u64 end = start + PAGE_CACHE_SIZE - 1;
3953 if (test_range_bit(tree, start, end,
3954 EXTENT_IOBITS, 0, NULL))
3957 if ((mask & GFP_NOFS) == GFP_NOFS)
3960 * at this point we can safely clear everything except the
3961 * locked bit and the nodatasum bit
3963 ret = clear_extent_bit(tree, start, end,
3964 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3967 /* if clear_extent_bit failed for enomem reasons,
3968 * we can't allow the release to continue.
3979 * a helper for releasepage. As long as there are no locked extents
3980 * in the range corresponding to the page, both state records and extent
3981 * map records are removed
3983 int try_release_extent_mapping(struct extent_map_tree *map,
3984 struct extent_io_tree *tree, struct page *page,
3987 struct extent_map *em;
3988 u64 start = page_offset(page);
3989 u64 end = start + PAGE_CACHE_SIZE - 1;
3991 if ((mask & __GFP_WAIT) &&
3992 page->mapping->host->i_size > 16 * 1024 * 1024) {
3994 while (start <= end) {
3995 len = end - start + 1;
3996 write_lock(&map->lock);
3997 em = lookup_extent_mapping(map, start, len);
3999 write_unlock(&map->lock);
4002 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4003 em->start != start) {
4004 write_unlock(&map->lock);
4005 free_extent_map(em);
4008 if (!test_range_bit(tree, em->start,
4009 extent_map_end(em) - 1,
4010 EXTENT_LOCKED | EXTENT_WRITEBACK,
4012 remove_extent_mapping(map, em);
4013 /* once for the rb tree */
4014 free_extent_map(em);
4016 start = extent_map_end(em);
4017 write_unlock(&map->lock);
4020 free_extent_map(em);
4023 return try_release_extent_state(map, tree, page, mask);
4027 * helper function for fiemap, which doesn't want to see any holes.
4028 * This maps until we find something past 'last'
4030 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4033 get_extent_t *get_extent)
4035 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4036 struct extent_map *em;
4043 len = last - offset;
4046 len = ALIGN(len, sectorsize);
4047 em = get_extent(inode, NULL, 0, offset, len, 0);
4048 if (IS_ERR_OR_NULL(em))
4051 /* if this isn't a hole return it */
4052 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4053 em->block_start != EXTENT_MAP_HOLE) {
4057 /* this is a hole, advance to the next extent */
4058 offset = extent_map_end(em);
4059 free_extent_map(em);
4066 static noinline int count_ext_ref(u64 inum, u64 offset, u64 root_id, void *ctx)
4068 unsigned long cnt = *((unsigned long *)ctx);
4071 *((unsigned long *)ctx) = cnt;
4073 /* Now we're sure that the extent is shared. */
4079 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4080 __u64 start, __u64 len, get_extent_t *get_extent)
4084 u64 max = start + len;
4088 u64 last_for_get_extent = 0;
4090 u64 isize = i_size_read(inode);
4091 struct btrfs_key found_key;
4092 struct extent_map *em = NULL;
4093 struct extent_state *cached_state = NULL;
4094 struct btrfs_path *path;
4095 struct btrfs_file_extent_item *item;
4100 unsigned long emflags;
4105 path = btrfs_alloc_path();
4108 path->leave_spinning = 1;
4110 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
4111 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
4114 * lookup the last file extent. We're not using i_size here
4115 * because there might be preallocation past i_size
4117 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
4118 path, btrfs_ino(inode), -1, 0);
4120 btrfs_free_path(path);
4125 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4126 struct btrfs_file_extent_item);
4127 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4128 found_type = btrfs_key_type(&found_key);
4130 /* No extents, but there might be delalloc bits */
4131 if (found_key.objectid != btrfs_ino(inode) ||
4132 found_type != BTRFS_EXTENT_DATA_KEY) {
4133 /* have to trust i_size as the end */
4135 last_for_get_extent = isize;
4138 * remember the start of the last extent. There are a
4139 * bunch of different factors that go into the length of the
4140 * extent, so its much less complex to remember where it started
4142 last = found_key.offset;
4143 last_for_get_extent = last + 1;
4145 btrfs_release_path(path);
4148 * we might have some extents allocated but more delalloc past those
4149 * extents. so, we trust isize unless the start of the last extent is
4154 last_for_get_extent = isize;
4157 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4160 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4170 u64 offset_in_extent = 0;
4172 /* break if the extent we found is outside the range */
4173 if (em->start >= max || extent_map_end(em) < off)
4177 * get_extent may return an extent that starts before our
4178 * requested range. We have to make sure the ranges
4179 * we return to fiemap always move forward and don't
4180 * overlap, so adjust the offsets here
4182 em_start = max(em->start, off);
4185 * record the offset from the start of the extent
4186 * for adjusting the disk offset below. Only do this if the
4187 * extent isn't compressed since our in ram offset may be past
4188 * what we have actually allocated on disk.
4190 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4191 offset_in_extent = em_start - em->start;
4192 em_end = extent_map_end(em);
4193 em_len = em_end - em_start;
4194 emflags = em->flags;
4199 * bump off for our next call to get_extent
4201 off = extent_map_end(em);
4205 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4207 flags |= FIEMAP_EXTENT_LAST;
4208 } else if (em->block_start == EXTENT_MAP_INLINE) {
4209 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4210 FIEMAP_EXTENT_NOT_ALIGNED);
4211 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4212 flags |= (FIEMAP_EXTENT_DELALLOC |
4213 FIEMAP_EXTENT_UNKNOWN);
4215 unsigned long ref_cnt = 0;
4217 disko = em->block_start + offset_in_extent;
4220 * As btrfs supports shared space, this information
4221 * can be exported to userspace tools via
4222 * flag FIEMAP_EXTENT_SHARED.
4224 ret = iterate_inodes_from_logical(
4226 BTRFS_I(inode)->root->fs_info,
4227 path, count_ext_ref, &ref_cnt);
4228 if (ret < 0 && ret != -ENOENT)
4232 flags |= FIEMAP_EXTENT_SHARED;
4234 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4235 flags |= FIEMAP_EXTENT_ENCODED;
4237 free_extent_map(em);
4239 if ((em_start >= last) || em_len == (u64)-1 ||
4240 (last == (u64)-1 && isize <= em_end)) {
4241 flags |= FIEMAP_EXTENT_LAST;
4245 /* now scan forward to see if this is really the last extent. */
4246 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4253 flags |= FIEMAP_EXTENT_LAST;
4256 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4262 free_extent_map(em);
4264 btrfs_free_path(path);
4265 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4266 &cached_state, GFP_NOFS);
4270 static void __free_extent_buffer(struct extent_buffer *eb)
4272 btrfs_leak_debug_del(&eb->leak_list);
4273 kmem_cache_free(extent_buffer_cache, eb);
4276 static int extent_buffer_under_io(struct extent_buffer *eb)
4278 return (atomic_read(&eb->io_pages) ||
4279 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4280 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4284 * Helper for releasing extent buffer page.
4286 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4287 unsigned long start_idx)
4289 unsigned long index;
4290 unsigned long num_pages;
4292 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4294 BUG_ON(extent_buffer_under_io(eb));
4296 num_pages = num_extent_pages(eb->start, eb->len);
4297 index = start_idx + num_pages;
4298 if (start_idx >= index)
4303 page = extent_buffer_page(eb, index);
4304 if (page && mapped) {
4305 spin_lock(&page->mapping->private_lock);
4307 * We do this since we'll remove the pages after we've
4308 * removed the eb from the radix tree, so we could race
4309 * and have this page now attached to the new eb. So
4310 * only clear page_private if it's still connected to
4313 if (PagePrivate(page) &&
4314 page->private == (unsigned long)eb) {
4315 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4316 BUG_ON(PageDirty(page));
4317 BUG_ON(PageWriteback(page));
4319 * We need to make sure we haven't be attached
4322 ClearPagePrivate(page);
4323 set_page_private(page, 0);
4324 /* One for the page private */
4325 page_cache_release(page);
4327 spin_unlock(&page->mapping->private_lock);
4331 /* One for when we alloced the page */
4332 page_cache_release(page);
4334 } while (index != start_idx);
4338 * Helper for releasing the extent buffer.
4340 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4342 btrfs_release_extent_buffer_page(eb, 0);
4343 __free_extent_buffer(eb);
4346 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
4351 struct extent_buffer *eb = NULL;
4353 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4360 rwlock_init(&eb->lock);
4361 atomic_set(&eb->write_locks, 0);
4362 atomic_set(&eb->read_locks, 0);
4363 atomic_set(&eb->blocking_readers, 0);
4364 atomic_set(&eb->blocking_writers, 0);
4365 atomic_set(&eb->spinning_readers, 0);
4366 atomic_set(&eb->spinning_writers, 0);
4367 eb->lock_nested = 0;
4368 init_waitqueue_head(&eb->write_lock_wq);
4369 init_waitqueue_head(&eb->read_lock_wq);
4371 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4373 spin_lock_init(&eb->refs_lock);
4374 atomic_set(&eb->refs, 1);
4375 atomic_set(&eb->io_pages, 0);
4378 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4380 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4381 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4382 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4387 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4391 struct extent_buffer *new;
4392 unsigned long num_pages = num_extent_pages(src->start, src->len);
4394 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4398 for (i = 0; i < num_pages; i++) {
4399 p = alloc_page(GFP_NOFS);
4401 btrfs_release_extent_buffer(new);
4404 attach_extent_buffer_page(new, p);
4405 WARN_ON(PageDirty(p));
4410 copy_extent_buffer(new, src, 0, 0, src->len);
4411 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4412 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4417 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4419 struct extent_buffer *eb;
4420 unsigned long num_pages = num_extent_pages(0, len);
4423 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4427 for (i = 0; i < num_pages; i++) {
4428 eb->pages[i] = alloc_page(GFP_NOFS);
4432 set_extent_buffer_uptodate(eb);
4433 btrfs_set_header_nritems(eb, 0);
4434 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4439 __free_page(eb->pages[i - 1]);
4440 __free_extent_buffer(eb);
4444 static void check_buffer_tree_ref(struct extent_buffer *eb)
4447 /* the ref bit is tricky. We have to make sure it is set
4448 * if we have the buffer dirty. Otherwise the
4449 * code to free a buffer can end up dropping a dirty
4452 * Once the ref bit is set, it won't go away while the
4453 * buffer is dirty or in writeback, and it also won't
4454 * go away while we have the reference count on the
4457 * We can't just set the ref bit without bumping the
4458 * ref on the eb because free_extent_buffer might
4459 * see the ref bit and try to clear it. If this happens
4460 * free_extent_buffer might end up dropping our original
4461 * ref by mistake and freeing the page before we are able
4462 * to add one more ref.
4464 * So bump the ref count first, then set the bit. If someone
4465 * beat us to it, drop the ref we added.
4467 refs = atomic_read(&eb->refs);
4468 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4471 spin_lock(&eb->refs_lock);
4472 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4473 atomic_inc(&eb->refs);
4474 spin_unlock(&eb->refs_lock);
4477 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4479 unsigned long num_pages, i;
4481 check_buffer_tree_ref(eb);
4483 num_pages = num_extent_pages(eb->start, eb->len);
4484 for (i = 0; i < num_pages; i++) {
4485 struct page *p = extent_buffer_page(eb, i);
4486 mark_page_accessed(p);
4490 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4493 struct extent_buffer *eb;
4496 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4497 if (eb && atomic_inc_not_zero(&eb->refs)) {
4499 mark_extent_buffer_accessed(eb);
4507 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4508 u64 start, unsigned long len)
4510 unsigned long num_pages = num_extent_pages(start, len);
4512 unsigned long index = start >> PAGE_CACHE_SHIFT;
4513 struct extent_buffer *eb;
4514 struct extent_buffer *exists = NULL;
4516 struct address_space *mapping = tree->mapping;
4521 eb = find_extent_buffer(tree, start);
4525 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4529 for (i = 0; i < num_pages; i++, index++) {
4530 p = find_or_create_page(mapping, index, GFP_NOFS);
4534 spin_lock(&mapping->private_lock);
4535 if (PagePrivate(p)) {
4537 * We could have already allocated an eb for this page
4538 * and attached one so lets see if we can get a ref on
4539 * the existing eb, and if we can we know it's good and
4540 * we can just return that one, else we know we can just
4541 * overwrite page->private.
4543 exists = (struct extent_buffer *)p->private;
4544 if (atomic_inc_not_zero(&exists->refs)) {
4545 spin_unlock(&mapping->private_lock);
4547 page_cache_release(p);
4548 mark_extent_buffer_accessed(exists);
4553 * Do this so attach doesn't complain and we need to
4554 * drop the ref the old guy had.
4556 ClearPagePrivate(p);
4557 WARN_ON(PageDirty(p));
4558 page_cache_release(p);
4560 attach_extent_buffer_page(eb, p);
4561 spin_unlock(&mapping->private_lock);
4562 WARN_ON(PageDirty(p));
4563 mark_page_accessed(p);
4565 if (!PageUptodate(p))
4569 * see below about how we avoid a nasty race with release page
4570 * and why we unlock later
4574 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4576 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4580 spin_lock(&tree->buffer_lock);
4581 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4582 spin_unlock(&tree->buffer_lock);
4583 radix_tree_preload_end();
4584 if (ret == -EEXIST) {
4585 exists = find_extent_buffer(tree, start);
4591 /* add one reference for the tree */
4592 check_buffer_tree_ref(eb);
4595 * there is a race where release page may have
4596 * tried to find this extent buffer in the radix
4597 * but failed. It will tell the VM it is safe to
4598 * reclaim the, and it will clear the page private bit.
4599 * We must make sure to set the page private bit properly
4600 * after the extent buffer is in the radix tree so
4601 * it doesn't get lost
4603 SetPageChecked(eb->pages[0]);
4604 for (i = 1; i < num_pages; i++) {
4605 p = extent_buffer_page(eb, i);
4606 ClearPageChecked(p);
4609 unlock_page(eb->pages[0]);
4613 for (i = 0; i < num_pages; i++) {
4615 unlock_page(eb->pages[i]);
4618 WARN_ON(!atomic_dec_and_test(&eb->refs));
4619 btrfs_release_extent_buffer(eb);
4623 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4625 struct extent_buffer *eb =
4626 container_of(head, struct extent_buffer, rcu_head);
4628 __free_extent_buffer(eb);
4631 /* Expects to have eb->eb_lock already held */
4632 static int release_extent_buffer(struct extent_buffer *eb)
4634 WARN_ON(atomic_read(&eb->refs) == 0);
4635 if (atomic_dec_and_test(&eb->refs)) {
4636 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4637 spin_unlock(&eb->refs_lock);
4639 struct extent_io_tree *tree = eb->tree;
4641 spin_unlock(&eb->refs_lock);
4643 spin_lock(&tree->buffer_lock);
4644 radix_tree_delete(&tree->buffer,
4645 eb->start >> PAGE_CACHE_SHIFT);
4646 spin_unlock(&tree->buffer_lock);
4649 /* Should be safe to release our pages at this point */
4650 btrfs_release_extent_buffer_page(eb, 0);
4651 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4654 spin_unlock(&eb->refs_lock);
4659 void free_extent_buffer(struct extent_buffer *eb)
4667 refs = atomic_read(&eb->refs);
4670 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4675 spin_lock(&eb->refs_lock);
4676 if (atomic_read(&eb->refs) == 2 &&
4677 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4678 atomic_dec(&eb->refs);
4680 if (atomic_read(&eb->refs) == 2 &&
4681 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4682 !extent_buffer_under_io(eb) &&
4683 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4684 atomic_dec(&eb->refs);
4687 * I know this is terrible, but it's temporary until we stop tracking
4688 * the uptodate bits and such for the extent buffers.
4690 release_extent_buffer(eb);
4693 void free_extent_buffer_stale(struct extent_buffer *eb)
4698 spin_lock(&eb->refs_lock);
4699 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4701 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4702 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4703 atomic_dec(&eb->refs);
4704 release_extent_buffer(eb);
4707 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4710 unsigned long num_pages;
4713 num_pages = num_extent_pages(eb->start, eb->len);
4715 for (i = 0; i < num_pages; i++) {
4716 page = extent_buffer_page(eb, i);
4717 if (!PageDirty(page))
4721 WARN_ON(!PagePrivate(page));
4723 clear_page_dirty_for_io(page);
4724 spin_lock_irq(&page->mapping->tree_lock);
4725 if (!PageDirty(page)) {
4726 radix_tree_tag_clear(&page->mapping->page_tree,
4728 PAGECACHE_TAG_DIRTY);
4730 spin_unlock_irq(&page->mapping->tree_lock);
4731 ClearPageError(page);
4734 WARN_ON(atomic_read(&eb->refs) == 0);
4737 int set_extent_buffer_dirty(struct extent_buffer *eb)
4740 unsigned long num_pages;
4743 check_buffer_tree_ref(eb);
4745 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4747 num_pages = num_extent_pages(eb->start, eb->len);
4748 WARN_ON(atomic_read(&eb->refs) == 0);
4749 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4751 for (i = 0; i < num_pages; i++)
4752 set_page_dirty(extent_buffer_page(eb, i));
4756 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4760 unsigned long num_pages;
4762 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4763 num_pages = num_extent_pages(eb->start, eb->len);
4764 for (i = 0; i < num_pages; i++) {
4765 page = extent_buffer_page(eb, i);
4767 ClearPageUptodate(page);
4772 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4776 unsigned long num_pages;
4778 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4779 num_pages = num_extent_pages(eb->start, eb->len);
4780 for (i = 0; i < num_pages; i++) {
4781 page = extent_buffer_page(eb, i);
4782 SetPageUptodate(page);
4787 int extent_buffer_uptodate(struct extent_buffer *eb)
4789 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4792 int read_extent_buffer_pages(struct extent_io_tree *tree,
4793 struct extent_buffer *eb, u64 start, int wait,
4794 get_extent_t *get_extent, int mirror_num)
4797 unsigned long start_i;
4801 int locked_pages = 0;
4802 int all_uptodate = 1;
4803 unsigned long num_pages;
4804 unsigned long num_reads = 0;
4805 struct bio *bio = NULL;
4806 unsigned long bio_flags = 0;
4808 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4812 WARN_ON(start < eb->start);
4813 start_i = (start >> PAGE_CACHE_SHIFT) -
4814 (eb->start >> PAGE_CACHE_SHIFT);
4819 num_pages = num_extent_pages(eb->start, eb->len);
4820 for (i = start_i; i < num_pages; i++) {
4821 page = extent_buffer_page(eb, i);
4822 if (wait == WAIT_NONE) {
4823 if (!trylock_page(page))
4829 if (!PageUptodate(page)) {
4836 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4840 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4841 eb->read_mirror = 0;
4842 atomic_set(&eb->io_pages, num_reads);
4843 for (i = start_i; i < num_pages; i++) {
4844 page = extent_buffer_page(eb, i);
4845 if (!PageUptodate(page)) {
4846 ClearPageError(page);
4847 err = __extent_read_full_page(tree, page,
4849 mirror_num, &bio_flags,
4859 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4865 if (ret || wait != WAIT_COMPLETE)
4868 for (i = start_i; i < num_pages; i++) {
4869 page = extent_buffer_page(eb, i);
4870 wait_on_page_locked(page);
4871 if (!PageUptodate(page))
4879 while (locked_pages > 0) {
4880 page = extent_buffer_page(eb, i);
4888 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4889 unsigned long start,
4896 char *dst = (char *)dstv;
4897 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4898 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4900 WARN_ON(start > eb->len);
4901 WARN_ON(start + len > eb->start + eb->len);
4903 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4906 page = extent_buffer_page(eb, i);
4908 cur = min(len, (PAGE_CACHE_SIZE - offset));
4909 kaddr = page_address(page);
4910 memcpy(dst, kaddr + offset, cur);
4919 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4920 unsigned long min_len, char **map,
4921 unsigned long *map_start,
4922 unsigned long *map_len)
4924 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4927 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4928 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4929 unsigned long end_i = (start_offset + start + min_len - 1) >>
4936 offset = start_offset;
4940 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4943 if (start + min_len > eb->len) {
4944 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4946 eb->start, eb->len, start, min_len);
4950 p = extent_buffer_page(eb, i);
4951 kaddr = page_address(p);
4952 *map = kaddr + offset;
4953 *map_len = PAGE_CACHE_SIZE - offset;
4957 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4958 unsigned long start,
4965 char *ptr = (char *)ptrv;
4966 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4967 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4970 WARN_ON(start > eb->len);
4971 WARN_ON(start + len > eb->start + eb->len);
4973 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4976 page = extent_buffer_page(eb, i);
4978 cur = min(len, (PAGE_CACHE_SIZE - offset));
4980 kaddr = page_address(page);
4981 ret = memcmp(ptr, kaddr + offset, cur);
4993 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4994 unsigned long start, unsigned long len)
5000 char *src = (char *)srcv;
5001 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5002 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5004 WARN_ON(start > eb->len);
5005 WARN_ON(start + len > eb->start + eb->len);
5007 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5010 page = extent_buffer_page(eb, i);
5011 WARN_ON(!PageUptodate(page));
5013 cur = min(len, PAGE_CACHE_SIZE - offset);
5014 kaddr = page_address(page);
5015 memcpy(kaddr + offset, src, cur);
5024 void memset_extent_buffer(struct extent_buffer *eb, char c,
5025 unsigned long start, unsigned long len)
5031 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5032 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5034 WARN_ON(start > eb->len);
5035 WARN_ON(start + len > eb->start + eb->len);
5037 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5040 page = extent_buffer_page(eb, i);
5041 WARN_ON(!PageUptodate(page));
5043 cur = min(len, PAGE_CACHE_SIZE - offset);
5044 kaddr = page_address(page);
5045 memset(kaddr + offset, c, cur);
5053 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5054 unsigned long dst_offset, unsigned long src_offset,
5057 u64 dst_len = dst->len;
5062 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5063 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5065 WARN_ON(src->len != dst_len);
5067 offset = (start_offset + dst_offset) &
5068 (PAGE_CACHE_SIZE - 1);
5071 page = extent_buffer_page(dst, i);
5072 WARN_ON(!PageUptodate(page));
5074 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5076 kaddr = page_address(page);
5077 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5086 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5088 unsigned long distance = (src > dst) ? src - dst : dst - src;
5089 return distance < len;
5092 static void copy_pages(struct page *dst_page, struct page *src_page,
5093 unsigned long dst_off, unsigned long src_off,
5096 char *dst_kaddr = page_address(dst_page);
5098 int must_memmove = 0;
5100 if (dst_page != src_page) {
5101 src_kaddr = page_address(src_page);
5103 src_kaddr = dst_kaddr;
5104 if (areas_overlap(src_off, dst_off, len))
5109 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5111 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5114 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5115 unsigned long src_offset, unsigned long len)
5118 size_t dst_off_in_page;
5119 size_t src_off_in_page;
5120 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5121 unsigned long dst_i;
5122 unsigned long src_i;
5124 if (src_offset + len > dst->len) {
5125 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5126 "len %lu dst len %lu\n", src_offset, len, dst->len);
5129 if (dst_offset + len > dst->len) {
5130 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5131 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5136 dst_off_in_page = (start_offset + dst_offset) &
5137 (PAGE_CACHE_SIZE - 1);
5138 src_off_in_page = (start_offset + src_offset) &
5139 (PAGE_CACHE_SIZE - 1);
5141 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5142 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5144 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5146 cur = min_t(unsigned long, cur,
5147 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5149 copy_pages(extent_buffer_page(dst, dst_i),
5150 extent_buffer_page(dst, src_i),
5151 dst_off_in_page, src_off_in_page, cur);
5159 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5160 unsigned long src_offset, unsigned long len)
5163 size_t dst_off_in_page;
5164 size_t src_off_in_page;
5165 unsigned long dst_end = dst_offset + len - 1;
5166 unsigned long src_end = src_offset + len - 1;
5167 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5168 unsigned long dst_i;
5169 unsigned long src_i;
5171 if (src_offset + len > dst->len) {
5172 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5173 "len %lu len %lu\n", src_offset, len, dst->len);
5176 if (dst_offset + len > dst->len) {
5177 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5178 "len %lu len %lu\n", dst_offset, len, dst->len);
5181 if (dst_offset < src_offset) {
5182 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5186 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5187 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5189 dst_off_in_page = (start_offset + dst_end) &
5190 (PAGE_CACHE_SIZE - 1);
5191 src_off_in_page = (start_offset + src_end) &
5192 (PAGE_CACHE_SIZE - 1);
5194 cur = min_t(unsigned long, len, src_off_in_page + 1);
5195 cur = min(cur, dst_off_in_page + 1);
5196 copy_pages(extent_buffer_page(dst, dst_i),
5197 extent_buffer_page(dst, src_i),
5198 dst_off_in_page - cur + 1,
5199 src_off_in_page - cur + 1, cur);
5207 int try_release_extent_buffer(struct page *page)
5209 struct extent_buffer *eb;
5212 * We need to make sure noboody is attaching this page to an eb right
5215 spin_lock(&page->mapping->private_lock);
5216 if (!PagePrivate(page)) {
5217 spin_unlock(&page->mapping->private_lock);
5221 eb = (struct extent_buffer *)page->private;
5225 * This is a little awful but should be ok, we need to make sure that
5226 * the eb doesn't disappear out from under us while we're looking at
5229 spin_lock(&eb->refs_lock);
5230 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5231 spin_unlock(&eb->refs_lock);
5232 spin_unlock(&page->mapping->private_lock);
5235 spin_unlock(&page->mapping->private_lock);
5238 * If tree ref isn't set then we know the ref on this eb is a real ref,
5239 * so just return, this page will likely be freed soon anyway.
5241 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5242 spin_unlock(&eb->refs_lock);
5246 return release_extent_buffer(eb);