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);
1956 * this bypasses the standard btrfs submit functions deliberately, as
1957 * the standard behavior is to write all copies in a raid setup. here we only
1958 * want to write the one bad copy. so we do the mapping for ourselves and issue
1959 * submit_bio directly.
1960 * to avoid any synchronization issues, wait for the data after writing, which
1961 * actually prevents the read that triggered the error from finishing.
1962 * currently, there can be no more than two copies of every data bit. thus,
1963 * exactly one rewrite is required.
1965 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
1966 u64 length, u64 logical, struct page *page,
1970 struct btrfs_device *dev;
1973 struct btrfs_bio *bbio = NULL;
1974 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1977 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
1978 BUG_ON(!mirror_num);
1980 /* we can't repair anything in raid56 yet */
1981 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
1984 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1987 bio->bi_iter.bi_size = 0;
1988 map_length = length;
1990 ret = btrfs_map_block(fs_info, WRITE, logical,
1991 &map_length, &bbio, mirror_num);
1996 BUG_ON(mirror_num != bbio->mirror_num);
1997 sector = bbio->stripes[mirror_num-1].physical >> 9;
1998 bio->bi_iter.bi_sector = sector;
1999 dev = bbio->stripes[mirror_num-1].dev;
2001 if (!dev || !dev->bdev || !dev->writeable) {
2005 bio->bi_bdev = dev->bdev;
2006 bio_add_page(bio, page, length, start - page_offset(page));
2008 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
2009 /* try to remap that extent elsewhere? */
2011 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2015 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
2016 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2017 start, rcu_str_deref(dev->name), sector);
2023 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2026 u64 start = eb->start;
2027 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2030 if (root->fs_info->sb->s_flags & MS_RDONLY)
2033 for (i = 0; i < num_pages; i++) {
2034 struct page *p = extent_buffer_page(eb, i);
2035 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2036 start, p, mirror_num);
2039 start += PAGE_CACHE_SIZE;
2046 * each time an IO finishes, we do a fast check in the IO failure tree
2047 * to see if we need to process or clean up an io_failure_record
2049 static int clean_io_failure(u64 start, struct page *page)
2052 u64 private_failure;
2053 struct io_failure_record *failrec;
2054 struct inode *inode = page->mapping->host;
2055 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2056 struct extent_state *state;
2062 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2063 (u64)-1, 1, EXTENT_DIRTY, 0);
2067 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2072 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2073 BUG_ON(!failrec->this_mirror);
2075 if (failrec->in_validation) {
2076 /* there was no real error, just free the record */
2077 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2082 if (fs_info->sb->s_flags & MS_RDONLY)
2085 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2086 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2089 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2091 if (state && state->start <= failrec->start &&
2092 state->end >= failrec->start + failrec->len - 1) {
2093 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2095 if (num_copies > 1) {
2096 ret = repair_io_failure(fs_info, start, failrec->len,
2097 failrec->logical, page,
2098 failrec->failed_mirror);
2106 ret = free_io_failure(inode, failrec, did_repair);
2112 * this is a generic handler for readpage errors (default
2113 * readpage_io_failed_hook). if other copies exist, read those and write back
2114 * good data to the failed position. does not investigate in remapping the
2115 * failed extent elsewhere, hoping the device will be smart enough to do this as
2119 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2120 struct page *page, u64 start, u64 end,
2123 struct io_failure_record *failrec = NULL;
2125 struct extent_map *em;
2126 struct inode *inode = page->mapping->host;
2127 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2128 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2129 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2131 struct btrfs_io_bio *btrfs_failed_bio;
2132 struct btrfs_io_bio *btrfs_bio;
2138 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2140 ret = get_state_private(failure_tree, start, &private);
2142 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2145 failrec->start = start;
2146 failrec->len = end - start + 1;
2147 failrec->this_mirror = 0;
2148 failrec->bio_flags = 0;
2149 failrec->in_validation = 0;
2151 read_lock(&em_tree->lock);
2152 em = lookup_extent_mapping(em_tree, start, failrec->len);
2154 read_unlock(&em_tree->lock);
2159 if (em->start > start || em->start + em->len < start) {
2160 free_extent_map(em);
2163 read_unlock(&em_tree->lock);
2169 logical = start - em->start;
2170 logical = em->block_start + logical;
2171 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2172 logical = em->block_start;
2173 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2174 extent_set_compress_type(&failrec->bio_flags,
2177 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2178 "len=%llu\n", logical, start, failrec->len);
2179 failrec->logical = logical;
2180 free_extent_map(em);
2182 /* set the bits in the private failure tree */
2183 ret = set_extent_bits(failure_tree, start, end,
2184 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2186 ret = set_state_private(failure_tree, start,
2187 (u64)(unsigned long)failrec);
2188 /* set the bits in the inode's tree */
2190 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2197 failrec = (struct io_failure_record *)(unsigned long)private;
2198 pr_debug("bio_readpage_error: (found) logical=%llu, "
2199 "start=%llu, len=%llu, validation=%d\n",
2200 failrec->logical, failrec->start, failrec->len,
2201 failrec->in_validation);
2203 * when data can be on disk more than twice, add to failrec here
2204 * (e.g. with a list for failed_mirror) to make
2205 * clean_io_failure() clean all those errors at once.
2208 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2209 failrec->logical, failrec->len);
2210 if (num_copies == 1) {
2212 * we only have a single copy of the data, so don't bother with
2213 * all the retry and error correction code that follows. no
2214 * matter what the error is, it is very likely to persist.
2216 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2217 num_copies, failrec->this_mirror, failed_mirror);
2218 free_io_failure(inode, failrec, 0);
2223 * there are two premises:
2224 * a) deliver good data to the caller
2225 * b) correct the bad sectors on disk
2227 if (failed_bio->bi_vcnt > 1) {
2229 * to fulfill b), we need to know the exact failing sectors, as
2230 * we don't want to rewrite any more than the failed ones. thus,
2231 * we need separate read requests for the failed bio
2233 * if the following BUG_ON triggers, our validation request got
2234 * merged. we need separate requests for our algorithm to work.
2236 BUG_ON(failrec->in_validation);
2237 failrec->in_validation = 1;
2238 failrec->this_mirror = failed_mirror;
2239 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2242 * we're ready to fulfill a) and b) alongside. get a good copy
2243 * of the failed sector and if we succeed, we have setup
2244 * everything for repair_io_failure to do the rest for us.
2246 if (failrec->in_validation) {
2247 BUG_ON(failrec->this_mirror != failed_mirror);
2248 failrec->in_validation = 0;
2249 failrec->this_mirror = 0;
2251 failrec->failed_mirror = failed_mirror;
2252 failrec->this_mirror++;
2253 if (failrec->this_mirror == failed_mirror)
2254 failrec->this_mirror++;
2255 read_mode = READ_SYNC;
2258 if (failrec->this_mirror > num_copies) {
2259 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2260 num_copies, failrec->this_mirror, failed_mirror);
2261 free_io_failure(inode, failrec, 0);
2265 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2267 free_io_failure(inode, failrec, 0);
2270 bio->bi_end_io = failed_bio->bi_end_io;
2271 bio->bi_iter.bi_sector = failrec->logical >> 9;
2272 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2273 bio->bi_iter.bi_size = 0;
2275 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2276 if (btrfs_failed_bio->csum) {
2277 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2278 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2280 btrfs_bio = btrfs_io_bio(bio);
2281 btrfs_bio->csum = btrfs_bio->csum_inline;
2282 phy_offset >>= inode->i_sb->s_blocksize_bits;
2283 phy_offset *= csum_size;
2284 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset,
2288 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2290 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2291 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2292 failrec->this_mirror, num_copies, failrec->in_validation);
2294 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2295 failrec->this_mirror,
2296 failrec->bio_flags, 0);
2300 /* lots and lots of room for performance fixes in the end_bio funcs */
2302 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2304 int uptodate = (err == 0);
2305 struct extent_io_tree *tree;
2308 tree = &BTRFS_I(page->mapping->host)->io_tree;
2310 if (tree->ops && tree->ops->writepage_end_io_hook) {
2311 ret = tree->ops->writepage_end_io_hook(page, start,
2312 end, NULL, uptodate);
2318 ClearPageUptodate(page);
2325 * after a writepage IO is done, we need to:
2326 * clear the uptodate bits on error
2327 * clear the writeback bits in the extent tree for this IO
2328 * end_page_writeback if the page has no more pending IO
2330 * Scheduling is not allowed, so the extent state tree is expected
2331 * to have one and only one object corresponding to this IO.
2333 static void end_bio_extent_writepage(struct bio *bio, int err)
2335 struct bio_vec *bvec;
2336 struct extent_io_tree *tree;
2341 bio_for_each_segment_all(bvec, bio, i) {
2342 struct page *page = bvec->bv_page;
2343 tree = &BTRFS_I(page->mapping->host)->io_tree;
2345 /* We always issue full-page reads, but if some block
2346 * in a page fails to read, blk_update_request() will
2347 * advance bv_offset and adjust bv_len to compensate.
2348 * Print a warning for nonzero offsets, and an error
2349 * if they don't add up to a full page. */
2350 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2351 printk("%s page write in btrfs with offset %u and length %u\n",
2352 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2353 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2354 bvec->bv_offset, bvec->bv_len);
2356 start = page_offset(page);
2357 end = start + bvec->bv_offset + bvec->bv_len - 1;
2359 if (end_extent_writepage(page, err, start, end))
2362 end_page_writeback(page);
2369 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2372 struct extent_state *cached = NULL;
2373 u64 end = start + len - 1;
2375 if (uptodate && tree->track_uptodate)
2376 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2377 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2381 * after a readpage IO is done, we need to:
2382 * clear the uptodate bits on error
2383 * set the uptodate bits if things worked
2384 * set the page up to date if all extents in the tree are uptodate
2385 * clear the lock bit in the extent tree
2386 * unlock the page if there are no other extents locked for it
2388 * Scheduling is not allowed, so the extent state tree is expected
2389 * to have one and only one object corresponding to this IO.
2391 static void end_bio_extent_readpage(struct bio *bio, int err)
2393 struct bio_vec *bvec;
2394 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2395 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2396 struct extent_io_tree *tree;
2401 u64 extent_start = 0;
2410 bio_for_each_segment_all(bvec, bio, i) {
2411 struct page *page = bvec->bv_page;
2412 struct inode *inode = page->mapping->host;
2414 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2415 "mirror=%lu\n", (u64)bio->bi_iter.bi_sector, err,
2416 io_bio->mirror_num);
2417 tree = &BTRFS_I(inode)->io_tree;
2419 /* We always issue full-page reads, but if some block
2420 * in a page fails to read, blk_update_request() will
2421 * advance bv_offset and adjust bv_len to compensate.
2422 * Print a warning for nonzero offsets, and an error
2423 * if they don't add up to a full page. */
2424 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2425 printk("%s page read in btrfs with offset %u and length %u\n",
2426 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2427 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2428 bvec->bv_offset, bvec->bv_len);
2430 start = page_offset(page);
2431 end = start + bvec->bv_offset + bvec->bv_len - 1;
2434 mirror = io_bio->mirror_num;
2435 if (likely(uptodate && tree->ops &&
2436 tree->ops->readpage_end_io_hook)) {
2437 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2443 clean_io_failure(start, page);
2446 if (likely(uptodate))
2449 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2450 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2452 test_bit(BIO_UPTODATE, &bio->bi_flags))
2456 * The generic bio_readpage_error handles errors the
2457 * following way: If possible, new read requests are
2458 * created and submitted and will end up in
2459 * end_bio_extent_readpage as well (if we're lucky, not
2460 * in the !uptodate case). In that case it returns 0 and
2461 * we just go on with the next page in our bio. If it
2462 * can't handle the error it will return -EIO and we
2463 * remain responsible for that page.
2465 ret = bio_readpage_error(bio, offset, page, start, end,
2469 test_bit(BIO_UPTODATE, &bio->bi_flags);
2476 if (likely(uptodate)) {
2477 loff_t i_size = i_size_read(inode);
2478 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2481 /* Zero out the end if this page straddles i_size */
2482 offset = i_size & (PAGE_CACHE_SIZE-1);
2483 if (page->index == end_index && offset)
2484 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2485 SetPageUptodate(page);
2487 ClearPageUptodate(page);
2493 if (unlikely(!uptodate)) {
2495 endio_readpage_release_extent(tree,
2501 endio_readpage_release_extent(tree, start,
2502 end - start + 1, 0);
2503 } else if (!extent_len) {
2504 extent_start = start;
2505 extent_len = end + 1 - start;
2506 } else if (extent_start + extent_len == start) {
2507 extent_len += end + 1 - start;
2509 endio_readpage_release_extent(tree, extent_start,
2510 extent_len, uptodate);
2511 extent_start = start;
2512 extent_len = end + 1 - start;
2517 endio_readpage_release_extent(tree, extent_start, extent_len,
2520 io_bio->end_io(io_bio, err);
2525 * this allocates from the btrfs_bioset. We're returning a bio right now
2526 * but you can call btrfs_io_bio for the appropriate container_of magic
2529 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2532 struct btrfs_io_bio *btrfs_bio;
2535 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2537 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2538 while (!bio && (nr_vecs /= 2)) {
2539 bio = bio_alloc_bioset(gfp_flags,
2540 nr_vecs, btrfs_bioset);
2545 bio->bi_bdev = bdev;
2546 bio->bi_iter.bi_sector = first_sector;
2547 btrfs_bio = btrfs_io_bio(bio);
2548 btrfs_bio->csum = NULL;
2549 btrfs_bio->csum_allocated = NULL;
2550 btrfs_bio->end_io = NULL;
2555 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2557 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2561 /* this also allocates from the btrfs_bioset */
2562 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2564 struct btrfs_io_bio *btrfs_bio;
2567 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2569 btrfs_bio = btrfs_io_bio(bio);
2570 btrfs_bio->csum = NULL;
2571 btrfs_bio->csum_allocated = NULL;
2572 btrfs_bio->end_io = NULL;
2578 static int __must_check submit_one_bio(int rw, struct bio *bio,
2579 int mirror_num, unsigned long bio_flags)
2582 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2583 struct page *page = bvec->bv_page;
2584 struct extent_io_tree *tree = bio->bi_private;
2587 start = page_offset(page) + bvec->bv_offset;
2589 bio->bi_private = NULL;
2593 if (tree->ops && tree->ops->submit_bio_hook)
2594 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2595 mirror_num, bio_flags, start);
2597 btrfsic_submit_bio(rw, bio);
2599 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2605 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2606 unsigned long offset, size_t size, struct bio *bio,
2607 unsigned long bio_flags)
2610 if (tree->ops && tree->ops->merge_bio_hook)
2611 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2618 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2619 struct page *page, sector_t sector,
2620 size_t size, unsigned long offset,
2621 struct block_device *bdev,
2622 struct bio **bio_ret,
2623 unsigned long max_pages,
2624 bio_end_io_t end_io_func,
2626 unsigned long prev_bio_flags,
2627 unsigned long bio_flags)
2633 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2634 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2635 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2637 if (bio_ret && *bio_ret) {
2640 contig = bio->bi_iter.bi_sector == sector;
2642 contig = bio_end_sector(bio) == sector;
2644 if (prev_bio_flags != bio_flags || !contig ||
2645 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2646 bio_add_page(bio, page, page_size, offset) < page_size) {
2647 ret = submit_one_bio(rw, bio, mirror_num,
2656 if (this_compressed)
2659 nr = bio_get_nr_vecs(bdev);
2661 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2665 bio_add_page(bio, page, page_size, offset);
2666 bio->bi_end_io = end_io_func;
2667 bio->bi_private = tree;
2672 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2677 static void attach_extent_buffer_page(struct extent_buffer *eb,
2680 if (!PagePrivate(page)) {
2681 SetPagePrivate(page);
2682 page_cache_get(page);
2683 set_page_private(page, (unsigned long)eb);
2685 WARN_ON(page->private != (unsigned long)eb);
2689 void set_page_extent_mapped(struct page *page)
2691 if (!PagePrivate(page)) {
2692 SetPagePrivate(page);
2693 page_cache_get(page);
2694 set_page_private(page, EXTENT_PAGE_PRIVATE);
2698 static struct extent_map *
2699 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2700 u64 start, u64 len, get_extent_t *get_extent,
2701 struct extent_map **em_cached)
2703 struct extent_map *em;
2705 if (em_cached && *em_cached) {
2707 if (em->in_tree && start >= em->start &&
2708 start < extent_map_end(em)) {
2709 atomic_inc(&em->refs);
2713 free_extent_map(em);
2717 em = get_extent(inode, page, pg_offset, start, len, 0);
2718 if (em_cached && !IS_ERR_OR_NULL(em)) {
2720 atomic_inc(&em->refs);
2726 * basic readpage implementation. Locked extent state structs are inserted
2727 * into the tree that are removed when the IO is done (by the end_io
2729 * XXX JDM: This needs looking at to ensure proper page locking
2731 static int __do_readpage(struct extent_io_tree *tree,
2733 get_extent_t *get_extent,
2734 struct extent_map **em_cached,
2735 struct bio **bio, int mirror_num,
2736 unsigned long *bio_flags, int rw)
2738 struct inode *inode = page->mapping->host;
2739 u64 start = page_offset(page);
2740 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2744 u64 last_byte = i_size_read(inode);
2748 struct extent_map *em;
2749 struct block_device *bdev;
2752 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2753 size_t pg_offset = 0;
2755 size_t disk_io_size;
2756 size_t blocksize = inode->i_sb->s_blocksize;
2757 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2759 set_page_extent_mapped(page);
2762 if (!PageUptodate(page)) {
2763 if (cleancache_get_page(page) == 0) {
2764 BUG_ON(blocksize != PAGE_SIZE);
2765 unlock_extent(tree, start, end);
2770 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2772 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2775 iosize = PAGE_CACHE_SIZE - zero_offset;
2776 userpage = kmap_atomic(page);
2777 memset(userpage + zero_offset, 0, iosize);
2778 flush_dcache_page(page);
2779 kunmap_atomic(userpage);
2782 while (cur <= end) {
2783 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2785 if (cur >= last_byte) {
2787 struct extent_state *cached = NULL;
2789 iosize = PAGE_CACHE_SIZE - pg_offset;
2790 userpage = kmap_atomic(page);
2791 memset(userpage + pg_offset, 0, iosize);
2792 flush_dcache_page(page);
2793 kunmap_atomic(userpage);
2794 set_extent_uptodate(tree, cur, cur + iosize - 1,
2797 unlock_extent_cached(tree, cur,
2802 em = __get_extent_map(inode, page, pg_offset, cur,
2803 end - cur + 1, get_extent, em_cached);
2804 if (IS_ERR_OR_NULL(em)) {
2807 unlock_extent(tree, cur, end);
2810 extent_offset = cur - em->start;
2811 BUG_ON(extent_map_end(em) <= cur);
2814 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2815 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2816 extent_set_compress_type(&this_bio_flag,
2820 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2821 cur_end = min(extent_map_end(em) - 1, end);
2822 iosize = ALIGN(iosize, blocksize);
2823 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2824 disk_io_size = em->block_len;
2825 sector = em->block_start >> 9;
2827 sector = (em->block_start + extent_offset) >> 9;
2828 disk_io_size = iosize;
2831 block_start = em->block_start;
2832 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2833 block_start = EXTENT_MAP_HOLE;
2834 free_extent_map(em);
2837 /* we've found a hole, just zero and go on */
2838 if (block_start == EXTENT_MAP_HOLE) {
2840 struct extent_state *cached = NULL;
2842 userpage = kmap_atomic(page);
2843 memset(userpage + pg_offset, 0, iosize);
2844 flush_dcache_page(page);
2845 kunmap_atomic(userpage);
2847 set_extent_uptodate(tree, cur, cur + iosize - 1,
2849 unlock_extent_cached(tree, cur, cur + iosize - 1,
2852 pg_offset += iosize;
2855 /* the get_extent function already copied into the page */
2856 if (test_range_bit(tree, cur, cur_end,
2857 EXTENT_UPTODATE, 1, NULL)) {
2858 check_page_uptodate(tree, page);
2860 unlock_extent(tree, cur, cur + iosize - 1);
2862 pg_offset += iosize;
2865 /* we have an inline extent but it didn't get marked up
2866 * to date. Error out
2868 if (block_start == EXTENT_MAP_INLINE) {
2871 unlock_extent(tree, cur, cur + iosize - 1);
2873 pg_offset += iosize;
2878 ret = submit_extent_page(rw, tree, page,
2879 sector, disk_io_size, pg_offset,
2881 end_bio_extent_readpage, mirror_num,
2886 *bio_flags = this_bio_flag;
2890 unlock_extent(tree, cur, cur + iosize - 1);
2893 pg_offset += iosize;
2897 if (!PageError(page))
2898 SetPageUptodate(page);
2904 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
2905 struct page *pages[], int nr_pages,
2907 get_extent_t *get_extent,
2908 struct extent_map **em_cached,
2909 struct bio **bio, int mirror_num,
2910 unsigned long *bio_flags, int rw)
2912 struct inode *inode;
2913 struct btrfs_ordered_extent *ordered;
2916 inode = pages[0]->mapping->host;
2918 lock_extent(tree, start, end);
2919 ordered = btrfs_lookup_ordered_range(inode, start,
2923 unlock_extent(tree, start, end);
2924 btrfs_start_ordered_extent(inode, ordered, 1);
2925 btrfs_put_ordered_extent(ordered);
2928 for (index = 0; index < nr_pages; index++) {
2929 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
2930 mirror_num, bio_flags, rw);
2931 page_cache_release(pages[index]);
2935 static void __extent_readpages(struct extent_io_tree *tree,
2936 struct page *pages[],
2937 int nr_pages, get_extent_t *get_extent,
2938 struct extent_map **em_cached,
2939 struct bio **bio, int mirror_num,
2940 unsigned long *bio_flags, int rw)
2946 int first_index = 0;
2948 for (index = 0; index < nr_pages; index++) {
2949 page_start = page_offset(pages[index]);
2952 end = start + PAGE_CACHE_SIZE - 1;
2953 first_index = index;
2954 } else if (end + 1 == page_start) {
2955 end += PAGE_CACHE_SIZE;
2957 __do_contiguous_readpages(tree, &pages[first_index],
2958 index - first_index, start,
2959 end, get_extent, em_cached,
2960 bio, mirror_num, bio_flags,
2963 end = start + PAGE_CACHE_SIZE - 1;
2964 first_index = index;
2969 __do_contiguous_readpages(tree, &pages[first_index],
2970 index - first_index, start,
2971 end, get_extent, em_cached, bio,
2972 mirror_num, bio_flags, rw);
2975 static int __extent_read_full_page(struct extent_io_tree *tree,
2977 get_extent_t *get_extent,
2978 struct bio **bio, int mirror_num,
2979 unsigned long *bio_flags, int rw)
2981 struct inode *inode = page->mapping->host;
2982 struct btrfs_ordered_extent *ordered;
2983 u64 start = page_offset(page);
2984 u64 end = start + PAGE_CACHE_SIZE - 1;
2988 lock_extent(tree, start, end);
2989 ordered = btrfs_lookup_ordered_extent(inode, start);
2992 unlock_extent(tree, start, end);
2993 btrfs_start_ordered_extent(inode, ordered, 1);
2994 btrfs_put_ordered_extent(ordered);
2997 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3002 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3003 get_extent_t *get_extent, int mirror_num)
3005 struct bio *bio = NULL;
3006 unsigned long bio_flags = 0;
3009 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3012 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3016 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3017 get_extent_t *get_extent, int mirror_num)
3019 struct bio *bio = NULL;
3020 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3023 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3026 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3030 static noinline void update_nr_written(struct page *page,
3031 struct writeback_control *wbc,
3032 unsigned long nr_written)
3034 wbc->nr_to_write -= nr_written;
3035 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3036 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3037 page->mapping->writeback_index = page->index + nr_written;
3041 * the writepage semantics are similar to regular writepage. extent
3042 * records are inserted to lock ranges in the tree, and as dirty areas
3043 * are found, they are marked writeback. Then the lock bits are removed
3044 * and the end_io handler clears the writeback ranges
3046 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3049 struct inode *inode = page->mapping->host;
3050 struct extent_page_data *epd = data;
3051 struct extent_io_tree *tree = epd->tree;
3052 u64 start = page_offset(page);
3054 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3058 u64 last_byte = i_size_read(inode);
3062 struct extent_state *cached_state = NULL;
3063 struct extent_map *em;
3064 struct block_device *bdev;
3067 size_t pg_offset = 0;
3069 loff_t i_size = i_size_read(inode);
3070 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3076 unsigned long nr_written = 0;
3077 bool fill_delalloc = true;
3079 if (wbc->sync_mode == WB_SYNC_ALL)
3080 write_flags = WRITE_SYNC;
3082 write_flags = WRITE;
3084 trace___extent_writepage(page, inode, wbc);
3086 WARN_ON(!PageLocked(page));
3088 ClearPageError(page);
3090 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3091 if (page->index > end_index ||
3092 (page->index == end_index && !pg_offset)) {
3093 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3098 if (page->index == end_index) {
3101 userpage = kmap_atomic(page);
3102 memset(userpage + pg_offset, 0,
3103 PAGE_CACHE_SIZE - pg_offset);
3104 kunmap_atomic(userpage);
3105 flush_dcache_page(page);
3109 set_page_extent_mapped(page);
3111 if (!tree->ops || !tree->ops->fill_delalloc)
3112 fill_delalloc = false;
3114 delalloc_start = start;
3117 if (!epd->extent_locked && fill_delalloc) {
3118 u64 delalloc_to_write = 0;
3120 * make sure the wbc mapping index is at least updated
3123 update_nr_written(page, wbc, 0);
3125 while (delalloc_end < page_end) {
3126 nr_delalloc = find_lock_delalloc_range(inode, tree,
3131 if (nr_delalloc == 0) {
3132 delalloc_start = delalloc_end + 1;
3135 ret = tree->ops->fill_delalloc(inode, page,
3140 /* File system has been set read-only */
3146 * delalloc_end is already one less than the total
3147 * length, so we don't subtract one from
3150 delalloc_to_write += (delalloc_end - delalloc_start +
3153 delalloc_start = delalloc_end + 1;
3155 if (wbc->nr_to_write < delalloc_to_write) {
3158 if (delalloc_to_write < thresh * 2)
3159 thresh = delalloc_to_write;
3160 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3164 /* did the fill delalloc function already unlock and start
3170 * we've unlocked the page, so we can't update
3171 * the mapping's writeback index, just update
3174 wbc->nr_to_write -= nr_written;
3178 if (tree->ops && tree->ops->writepage_start_hook) {
3179 ret = tree->ops->writepage_start_hook(page, start,
3182 /* Fixup worker will requeue */
3184 wbc->pages_skipped++;
3186 redirty_page_for_writepage(wbc, page);
3187 update_nr_written(page, wbc, nr_written);
3195 * we don't want to touch the inode after unlocking the page,
3196 * so we update the mapping writeback index now
3198 update_nr_written(page, wbc, nr_written + 1);
3201 if (last_byte <= start) {
3202 if (tree->ops && tree->ops->writepage_end_io_hook)
3203 tree->ops->writepage_end_io_hook(page, start,
3208 blocksize = inode->i_sb->s_blocksize;
3210 while (cur <= end) {
3211 if (cur >= last_byte) {
3212 if (tree->ops && tree->ops->writepage_end_io_hook)
3213 tree->ops->writepage_end_io_hook(page, cur,
3217 em = epd->get_extent(inode, page, pg_offset, cur,
3219 if (IS_ERR_OR_NULL(em)) {
3224 extent_offset = cur - em->start;
3225 BUG_ON(extent_map_end(em) <= cur);
3227 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3228 iosize = ALIGN(iosize, blocksize);
3229 sector = (em->block_start + extent_offset) >> 9;
3231 block_start = em->block_start;
3232 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3233 free_extent_map(em);
3237 * compressed and inline extents are written through other
3240 if (compressed || block_start == EXTENT_MAP_HOLE ||
3241 block_start == EXTENT_MAP_INLINE) {
3243 * end_io notification does not happen here for
3244 * compressed extents
3246 if (!compressed && tree->ops &&
3247 tree->ops->writepage_end_io_hook)
3248 tree->ops->writepage_end_io_hook(page, cur,
3251 else if (compressed) {
3252 /* we don't want to end_page_writeback on
3253 * a compressed extent. this happens
3260 pg_offset += iosize;
3263 /* leave this out until we have a page_mkwrite call */
3264 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3265 EXTENT_DIRTY, 0, NULL)) {
3267 pg_offset += iosize;
3271 if (tree->ops && tree->ops->writepage_io_hook) {
3272 ret = tree->ops->writepage_io_hook(page, cur,
3280 unsigned long max_nr = end_index + 1;
3282 set_range_writeback(tree, cur, cur + iosize - 1);
3283 if (!PageWriteback(page)) {
3284 printk(KERN_ERR "btrfs warning page %lu not "
3285 "writeback, cur %llu end %llu\n",
3286 page->index, cur, end);
3289 ret = submit_extent_page(write_flags, tree, page,
3290 sector, iosize, pg_offset,
3291 bdev, &epd->bio, max_nr,
3292 end_bio_extent_writepage,
3298 pg_offset += iosize;
3303 /* make sure the mapping tag for page dirty gets cleared */
3304 set_page_writeback(page);
3305 end_page_writeback(page);
3311 /* drop our reference on any cached states */
3312 free_extent_state(cached_state);
3316 static int eb_wait(void *word)
3322 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3324 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3325 TASK_UNINTERRUPTIBLE);
3328 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3329 struct btrfs_fs_info *fs_info,
3330 struct extent_page_data *epd)
3332 unsigned long i, num_pages;
3336 if (!btrfs_try_tree_write_lock(eb)) {
3338 flush_write_bio(epd);
3339 btrfs_tree_lock(eb);
3342 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3343 btrfs_tree_unlock(eb);
3347 flush_write_bio(epd);
3351 wait_on_extent_buffer_writeback(eb);
3352 btrfs_tree_lock(eb);
3353 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3355 btrfs_tree_unlock(eb);
3360 * We need to do this to prevent races in people who check if the eb is
3361 * under IO since we can end up having no IO bits set for a short period
3364 spin_lock(&eb->refs_lock);
3365 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3366 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3367 spin_unlock(&eb->refs_lock);
3368 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3369 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3371 fs_info->dirty_metadata_batch);
3374 spin_unlock(&eb->refs_lock);
3377 btrfs_tree_unlock(eb);
3382 num_pages = num_extent_pages(eb->start, eb->len);
3383 for (i = 0; i < num_pages; i++) {
3384 struct page *p = extent_buffer_page(eb, i);
3386 if (!trylock_page(p)) {
3388 flush_write_bio(epd);
3398 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3400 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3401 smp_mb__after_clear_bit();
3402 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3405 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3407 struct bio_vec *bvec;
3408 struct extent_buffer *eb;
3411 bio_for_each_segment_all(bvec, bio, i) {
3412 struct page *page = bvec->bv_page;
3414 eb = (struct extent_buffer *)page->private;
3416 done = atomic_dec_and_test(&eb->io_pages);
3418 if (err || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3419 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3420 ClearPageUptodate(page);
3424 end_page_writeback(page);
3429 end_extent_buffer_writeback(eb);
3435 static int write_one_eb(struct extent_buffer *eb,
3436 struct btrfs_fs_info *fs_info,
3437 struct writeback_control *wbc,
3438 struct extent_page_data *epd)
3440 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3441 u64 offset = eb->start;
3442 unsigned long i, num_pages;
3443 unsigned long bio_flags = 0;
3444 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3447 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3448 num_pages = num_extent_pages(eb->start, eb->len);
3449 atomic_set(&eb->io_pages, num_pages);
3450 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3451 bio_flags = EXTENT_BIO_TREE_LOG;
3453 for (i = 0; i < num_pages; i++) {
3454 struct page *p = extent_buffer_page(eb, i);
3456 clear_page_dirty_for_io(p);
3457 set_page_writeback(p);
3458 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3459 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3460 -1, end_bio_extent_buffer_writepage,
3461 0, epd->bio_flags, bio_flags);
3462 epd->bio_flags = bio_flags;
3464 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3466 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3467 end_extent_buffer_writeback(eb);
3471 offset += PAGE_CACHE_SIZE;
3472 update_nr_written(p, wbc, 1);
3476 if (unlikely(ret)) {
3477 for (; i < num_pages; i++) {
3478 struct page *p = extent_buffer_page(eb, i);
3486 int btree_write_cache_pages(struct address_space *mapping,
3487 struct writeback_control *wbc)
3489 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3490 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3491 struct extent_buffer *eb, *prev_eb = NULL;
3492 struct extent_page_data epd = {
3496 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3501 int nr_to_write_done = 0;
3502 struct pagevec pvec;
3505 pgoff_t end; /* Inclusive */
3509 pagevec_init(&pvec, 0);
3510 if (wbc->range_cyclic) {
3511 index = mapping->writeback_index; /* Start from prev offset */
3514 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3515 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3518 if (wbc->sync_mode == WB_SYNC_ALL)
3519 tag = PAGECACHE_TAG_TOWRITE;
3521 tag = PAGECACHE_TAG_DIRTY;
3523 if (wbc->sync_mode == WB_SYNC_ALL)
3524 tag_pages_for_writeback(mapping, index, end);
3525 while (!done && !nr_to_write_done && (index <= end) &&
3526 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3527 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3531 for (i = 0; i < nr_pages; i++) {
3532 struct page *page = pvec.pages[i];
3534 if (!PagePrivate(page))
3537 if (!wbc->range_cyclic && page->index > end) {
3542 spin_lock(&mapping->private_lock);
3543 if (!PagePrivate(page)) {
3544 spin_unlock(&mapping->private_lock);
3548 eb = (struct extent_buffer *)page->private;
3551 * Shouldn't happen and normally this would be a BUG_ON
3552 * but no sense in crashing the users box for something
3553 * we can survive anyway.
3556 spin_unlock(&mapping->private_lock);
3560 if (eb == prev_eb) {
3561 spin_unlock(&mapping->private_lock);
3565 ret = atomic_inc_not_zero(&eb->refs);
3566 spin_unlock(&mapping->private_lock);
3571 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3573 free_extent_buffer(eb);
3577 ret = write_one_eb(eb, fs_info, wbc, &epd);
3580 free_extent_buffer(eb);
3583 free_extent_buffer(eb);
3586 * the filesystem may choose to bump up nr_to_write.
3587 * We have to make sure to honor the new nr_to_write
3590 nr_to_write_done = wbc->nr_to_write <= 0;
3592 pagevec_release(&pvec);
3595 if (!scanned && !done) {
3597 * We hit the last page and there is more work to be done: wrap
3598 * back to the start of the file
3604 flush_write_bio(&epd);
3609 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3610 * @mapping: address space structure to write
3611 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3612 * @writepage: function called for each page
3613 * @data: data passed to writepage function
3615 * If a page is already under I/O, write_cache_pages() skips it, even
3616 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3617 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3618 * and msync() need to guarantee that all the data which was dirty at the time
3619 * the call was made get new I/O started against them. If wbc->sync_mode is
3620 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3621 * existing IO to complete.
3623 static int extent_write_cache_pages(struct extent_io_tree *tree,
3624 struct address_space *mapping,
3625 struct writeback_control *wbc,
3626 writepage_t writepage, void *data,
3627 void (*flush_fn)(void *))
3629 struct inode *inode = mapping->host;
3632 int nr_to_write_done = 0;
3633 struct pagevec pvec;
3636 pgoff_t end; /* Inclusive */
3641 * We have to hold onto the inode so that ordered extents can do their
3642 * work when the IO finishes. The alternative to this is failing to add
3643 * an ordered extent if the igrab() fails there and that is a huge pain
3644 * to deal with, so instead just hold onto the inode throughout the
3645 * writepages operation. If it fails here we are freeing up the inode
3646 * anyway and we'd rather not waste our time writing out stuff that is
3647 * going to be truncated anyway.
3652 pagevec_init(&pvec, 0);
3653 if (wbc->range_cyclic) {
3654 index = mapping->writeback_index; /* Start from prev offset */
3657 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3658 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3661 if (wbc->sync_mode == WB_SYNC_ALL)
3662 tag = PAGECACHE_TAG_TOWRITE;
3664 tag = PAGECACHE_TAG_DIRTY;
3666 if (wbc->sync_mode == WB_SYNC_ALL)
3667 tag_pages_for_writeback(mapping, index, end);
3668 while (!done && !nr_to_write_done && (index <= end) &&
3669 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3670 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3674 for (i = 0; i < nr_pages; i++) {
3675 struct page *page = pvec.pages[i];
3678 * At this point we hold neither mapping->tree_lock nor
3679 * lock on the page itself: the page may be truncated or
3680 * invalidated (changing page->mapping to NULL), or even
3681 * swizzled back from swapper_space to tmpfs file
3684 if (!trylock_page(page)) {
3689 if (unlikely(page->mapping != mapping)) {
3694 if (!wbc->range_cyclic && page->index > end) {
3700 if (wbc->sync_mode != WB_SYNC_NONE) {
3701 if (PageWriteback(page))
3703 wait_on_page_writeback(page);
3706 if (PageWriteback(page) ||
3707 !clear_page_dirty_for_io(page)) {
3712 ret = (*writepage)(page, wbc, data);
3714 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3722 * the filesystem may choose to bump up nr_to_write.
3723 * We have to make sure to honor the new nr_to_write
3726 nr_to_write_done = wbc->nr_to_write <= 0;
3728 pagevec_release(&pvec);
3731 if (!scanned && !done) {
3733 * We hit the last page and there is more work to be done: wrap
3734 * back to the start of the file
3740 btrfs_add_delayed_iput(inode);
3744 static void flush_epd_write_bio(struct extent_page_data *epd)
3753 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3754 BUG_ON(ret < 0); /* -ENOMEM */
3759 static noinline void flush_write_bio(void *data)
3761 struct extent_page_data *epd = data;
3762 flush_epd_write_bio(epd);
3765 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3766 get_extent_t *get_extent,
3767 struct writeback_control *wbc)
3770 struct extent_page_data epd = {
3773 .get_extent = get_extent,
3775 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3779 ret = __extent_writepage(page, wbc, &epd);
3781 flush_epd_write_bio(&epd);
3785 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3786 u64 start, u64 end, get_extent_t *get_extent,
3790 struct address_space *mapping = inode->i_mapping;
3792 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3795 struct extent_page_data epd = {
3798 .get_extent = get_extent,
3800 .sync_io = mode == WB_SYNC_ALL,
3803 struct writeback_control wbc_writepages = {
3805 .nr_to_write = nr_pages * 2,
3806 .range_start = start,
3807 .range_end = end + 1,
3810 while (start <= end) {
3811 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3812 if (clear_page_dirty_for_io(page))
3813 ret = __extent_writepage(page, &wbc_writepages, &epd);
3815 if (tree->ops && tree->ops->writepage_end_io_hook)
3816 tree->ops->writepage_end_io_hook(page, start,
3817 start + PAGE_CACHE_SIZE - 1,
3821 page_cache_release(page);
3822 start += PAGE_CACHE_SIZE;
3825 flush_epd_write_bio(&epd);
3829 int extent_writepages(struct extent_io_tree *tree,
3830 struct address_space *mapping,
3831 get_extent_t *get_extent,
3832 struct writeback_control *wbc)
3835 struct extent_page_data epd = {
3838 .get_extent = get_extent,
3840 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3844 ret = extent_write_cache_pages(tree, mapping, wbc,
3845 __extent_writepage, &epd,
3847 flush_epd_write_bio(&epd);
3851 int extent_readpages(struct extent_io_tree *tree,
3852 struct address_space *mapping,
3853 struct list_head *pages, unsigned nr_pages,
3854 get_extent_t get_extent)
3856 struct bio *bio = NULL;
3858 unsigned long bio_flags = 0;
3859 struct page *pagepool[16];
3861 struct extent_map *em_cached = NULL;
3864 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3865 page = list_entry(pages->prev, struct page, lru);
3867 prefetchw(&page->flags);
3868 list_del(&page->lru);
3869 if (add_to_page_cache_lru(page, mapping,
3870 page->index, GFP_NOFS)) {
3871 page_cache_release(page);
3875 pagepool[nr++] = page;
3876 if (nr < ARRAY_SIZE(pagepool))
3878 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3879 &bio, 0, &bio_flags, READ);
3883 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3884 &bio, 0, &bio_flags, READ);
3887 free_extent_map(em_cached);
3889 BUG_ON(!list_empty(pages));
3891 return submit_one_bio(READ, bio, 0, bio_flags);
3896 * basic invalidatepage code, this waits on any locked or writeback
3897 * ranges corresponding to the page, and then deletes any extent state
3898 * records from the tree
3900 int extent_invalidatepage(struct extent_io_tree *tree,
3901 struct page *page, unsigned long offset)
3903 struct extent_state *cached_state = NULL;
3904 u64 start = page_offset(page);
3905 u64 end = start + PAGE_CACHE_SIZE - 1;
3906 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3908 start += ALIGN(offset, blocksize);
3912 lock_extent_bits(tree, start, end, 0, &cached_state);
3913 wait_on_page_writeback(page);
3914 clear_extent_bit(tree, start, end,
3915 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3916 EXTENT_DO_ACCOUNTING,
3917 1, 1, &cached_state, GFP_NOFS);
3922 * a helper for releasepage, this tests for areas of the page that
3923 * are locked or under IO and drops the related state bits if it is safe
3926 static int try_release_extent_state(struct extent_map_tree *map,
3927 struct extent_io_tree *tree,
3928 struct page *page, gfp_t mask)
3930 u64 start = page_offset(page);
3931 u64 end = start + PAGE_CACHE_SIZE - 1;
3934 if (test_range_bit(tree, start, end,
3935 EXTENT_IOBITS, 0, NULL))
3938 if ((mask & GFP_NOFS) == GFP_NOFS)
3941 * at this point we can safely clear everything except the
3942 * locked bit and the nodatasum bit
3944 ret = clear_extent_bit(tree, start, end,
3945 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3948 /* if clear_extent_bit failed for enomem reasons,
3949 * we can't allow the release to continue.
3960 * a helper for releasepage. As long as there are no locked extents
3961 * in the range corresponding to the page, both state records and extent
3962 * map records are removed
3964 int try_release_extent_mapping(struct extent_map_tree *map,
3965 struct extent_io_tree *tree, struct page *page,
3968 struct extent_map *em;
3969 u64 start = page_offset(page);
3970 u64 end = start + PAGE_CACHE_SIZE - 1;
3972 if ((mask & __GFP_WAIT) &&
3973 page->mapping->host->i_size > 16 * 1024 * 1024) {
3975 while (start <= end) {
3976 len = end - start + 1;
3977 write_lock(&map->lock);
3978 em = lookup_extent_mapping(map, start, len);
3980 write_unlock(&map->lock);
3983 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3984 em->start != start) {
3985 write_unlock(&map->lock);
3986 free_extent_map(em);
3989 if (!test_range_bit(tree, em->start,
3990 extent_map_end(em) - 1,
3991 EXTENT_LOCKED | EXTENT_WRITEBACK,
3993 remove_extent_mapping(map, em);
3994 /* once for the rb tree */
3995 free_extent_map(em);
3997 start = extent_map_end(em);
3998 write_unlock(&map->lock);
4001 free_extent_map(em);
4004 return try_release_extent_state(map, tree, page, mask);
4008 * helper function for fiemap, which doesn't want to see any holes.
4009 * This maps until we find something past 'last'
4011 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4014 get_extent_t *get_extent)
4016 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4017 struct extent_map *em;
4024 len = last - offset;
4027 len = ALIGN(len, sectorsize);
4028 em = get_extent(inode, NULL, 0, offset, len, 0);
4029 if (IS_ERR_OR_NULL(em))
4032 /* if this isn't a hole return it */
4033 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4034 em->block_start != EXTENT_MAP_HOLE) {
4038 /* this is a hole, advance to the next extent */
4039 offset = extent_map_end(em);
4040 free_extent_map(em);
4047 static noinline int count_ext_ref(u64 inum, u64 offset, u64 root_id, void *ctx)
4049 unsigned long cnt = *((unsigned long *)ctx);
4052 *((unsigned long *)ctx) = cnt;
4054 /* Now we're sure that the extent is shared. */
4060 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4061 __u64 start, __u64 len, get_extent_t *get_extent)
4065 u64 max = start + len;
4069 u64 last_for_get_extent = 0;
4071 u64 isize = i_size_read(inode);
4072 struct btrfs_key found_key;
4073 struct extent_map *em = NULL;
4074 struct extent_state *cached_state = NULL;
4075 struct btrfs_path *path;
4076 struct btrfs_file_extent_item *item;
4081 unsigned long emflags;
4086 path = btrfs_alloc_path();
4089 path->leave_spinning = 1;
4091 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
4092 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
4095 * lookup the last file extent. We're not using i_size here
4096 * because there might be preallocation past i_size
4098 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
4099 path, btrfs_ino(inode), -1, 0);
4101 btrfs_free_path(path);
4106 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4107 struct btrfs_file_extent_item);
4108 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4109 found_type = btrfs_key_type(&found_key);
4111 /* No extents, but there might be delalloc bits */
4112 if (found_key.objectid != btrfs_ino(inode) ||
4113 found_type != BTRFS_EXTENT_DATA_KEY) {
4114 /* have to trust i_size as the end */
4116 last_for_get_extent = isize;
4119 * remember the start of the last extent. There are a
4120 * bunch of different factors that go into the length of the
4121 * extent, so its much less complex to remember where it started
4123 last = found_key.offset;
4124 last_for_get_extent = last + 1;
4126 btrfs_release_path(path);
4129 * we might have some extents allocated but more delalloc past those
4130 * extents. so, we trust isize unless the start of the last extent is
4135 last_for_get_extent = isize;
4138 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4141 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4151 u64 offset_in_extent = 0;
4153 /* break if the extent we found is outside the range */
4154 if (em->start >= max || extent_map_end(em) < off)
4158 * get_extent may return an extent that starts before our
4159 * requested range. We have to make sure the ranges
4160 * we return to fiemap always move forward and don't
4161 * overlap, so adjust the offsets here
4163 em_start = max(em->start, off);
4166 * record the offset from the start of the extent
4167 * for adjusting the disk offset below. Only do this if the
4168 * extent isn't compressed since our in ram offset may be past
4169 * what we have actually allocated on disk.
4171 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4172 offset_in_extent = em_start - em->start;
4173 em_end = extent_map_end(em);
4174 em_len = em_end - em_start;
4175 emflags = em->flags;
4180 * bump off for our next call to get_extent
4182 off = extent_map_end(em);
4186 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4188 flags |= FIEMAP_EXTENT_LAST;
4189 } else if (em->block_start == EXTENT_MAP_INLINE) {
4190 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4191 FIEMAP_EXTENT_NOT_ALIGNED);
4192 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4193 flags |= (FIEMAP_EXTENT_DELALLOC |
4194 FIEMAP_EXTENT_UNKNOWN);
4196 unsigned long ref_cnt = 0;
4198 disko = em->block_start + offset_in_extent;
4201 * As btrfs supports shared space, this information
4202 * can be exported to userspace tools via
4203 * flag FIEMAP_EXTENT_SHARED.
4205 ret = iterate_inodes_from_logical(
4207 BTRFS_I(inode)->root->fs_info,
4208 path, count_ext_ref, &ref_cnt);
4209 if (ret < 0 && ret != -ENOENT)
4213 flags |= FIEMAP_EXTENT_SHARED;
4215 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4216 flags |= FIEMAP_EXTENT_ENCODED;
4218 free_extent_map(em);
4220 if ((em_start >= last) || em_len == (u64)-1 ||
4221 (last == (u64)-1 && isize <= em_end)) {
4222 flags |= FIEMAP_EXTENT_LAST;
4226 /* now scan forward to see if this is really the last extent. */
4227 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4234 flags |= FIEMAP_EXTENT_LAST;
4237 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4243 free_extent_map(em);
4245 btrfs_free_path(path);
4246 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4247 &cached_state, GFP_NOFS);
4251 static void __free_extent_buffer(struct extent_buffer *eb)
4253 btrfs_leak_debug_del(&eb->leak_list);
4254 kmem_cache_free(extent_buffer_cache, eb);
4257 static int extent_buffer_under_io(struct extent_buffer *eb)
4259 return (atomic_read(&eb->io_pages) ||
4260 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4261 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4265 * Helper for releasing extent buffer page.
4267 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4268 unsigned long start_idx)
4270 unsigned long index;
4271 unsigned long num_pages;
4273 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4275 BUG_ON(extent_buffer_under_io(eb));
4277 num_pages = num_extent_pages(eb->start, eb->len);
4278 index = start_idx + num_pages;
4279 if (start_idx >= index)
4284 page = extent_buffer_page(eb, index);
4285 if (page && mapped) {
4286 spin_lock(&page->mapping->private_lock);
4288 * We do this since we'll remove the pages after we've
4289 * removed the eb from the radix tree, so we could race
4290 * and have this page now attached to the new eb. So
4291 * only clear page_private if it's still connected to
4294 if (PagePrivate(page) &&
4295 page->private == (unsigned long)eb) {
4296 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4297 BUG_ON(PageDirty(page));
4298 BUG_ON(PageWriteback(page));
4300 * We need to make sure we haven't be attached
4303 ClearPagePrivate(page);
4304 set_page_private(page, 0);
4305 /* One for the page private */
4306 page_cache_release(page);
4308 spin_unlock(&page->mapping->private_lock);
4312 /* One for when we alloced the page */
4313 page_cache_release(page);
4315 } while (index != start_idx);
4319 * Helper for releasing the extent buffer.
4321 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4323 btrfs_release_extent_buffer_page(eb, 0);
4324 __free_extent_buffer(eb);
4327 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
4332 struct extent_buffer *eb = NULL;
4334 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4341 rwlock_init(&eb->lock);
4342 atomic_set(&eb->write_locks, 0);
4343 atomic_set(&eb->read_locks, 0);
4344 atomic_set(&eb->blocking_readers, 0);
4345 atomic_set(&eb->blocking_writers, 0);
4346 atomic_set(&eb->spinning_readers, 0);
4347 atomic_set(&eb->spinning_writers, 0);
4348 eb->lock_nested = 0;
4349 init_waitqueue_head(&eb->write_lock_wq);
4350 init_waitqueue_head(&eb->read_lock_wq);
4352 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4354 spin_lock_init(&eb->refs_lock);
4355 atomic_set(&eb->refs, 1);
4356 atomic_set(&eb->io_pages, 0);
4359 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4361 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4362 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4363 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4368 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4372 struct extent_buffer *new;
4373 unsigned long num_pages = num_extent_pages(src->start, src->len);
4375 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4379 for (i = 0; i < num_pages; i++) {
4380 p = alloc_page(GFP_NOFS);
4382 btrfs_release_extent_buffer(new);
4385 attach_extent_buffer_page(new, p);
4386 WARN_ON(PageDirty(p));
4391 copy_extent_buffer(new, src, 0, 0, src->len);
4392 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4393 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4398 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4400 struct extent_buffer *eb;
4401 unsigned long num_pages = num_extent_pages(0, len);
4404 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4408 for (i = 0; i < num_pages; i++) {
4409 eb->pages[i] = alloc_page(GFP_NOFS);
4413 set_extent_buffer_uptodate(eb);
4414 btrfs_set_header_nritems(eb, 0);
4415 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4420 __free_page(eb->pages[i - 1]);
4421 __free_extent_buffer(eb);
4425 static void check_buffer_tree_ref(struct extent_buffer *eb)
4428 /* the ref bit is tricky. We have to make sure it is set
4429 * if we have the buffer dirty. Otherwise the
4430 * code to free a buffer can end up dropping a dirty
4433 * Once the ref bit is set, it won't go away while the
4434 * buffer is dirty or in writeback, and it also won't
4435 * go away while we have the reference count on the
4438 * We can't just set the ref bit without bumping the
4439 * ref on the eb because free_extent_buffer might
4440 * see the ref bit and try to clear it. If this happens
4441 * free_extent_buffer might end up dropping our original
4442 * ref by mistake and freeing the page before we are able
4443 * to add one more ref.
4445 * So bump the ref count first, then set the bit. If someone
4446 * beat us to it, drop the ref we added.
4448 refs = atomic_read(&eb->refs);
4449 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4452 spin_lock(&eb->refs_lock);
4453 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4454 atomic_inc(&eb->refs);
4455 spin_unlock(&eb->refs_lock);
4458 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4460 unsigned long num_pages, i;
4462 check_buffer_tree_ref(eb);
4464 num_pages = num_extent_pages(eb->start, eb->len);
4465 for (i = 0; i < num_pages; i++) {
4466 struct page *p = extent_buffer_page(eb, i);
4467 mark_page_accessed(p);
4471 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4474 struct extent_buffer *eb;
4477 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4478 if (eb && atomic_inc_not_zero(&eb->refs)) {
4480 mark_extent_buffer_accessed(eb);
4488 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4489 u64 start, unsigned long len)
4491 unsigned long num_pages = num_extent_pages(start, len);
4493 unsigned long index = start >> PAGE_CACHE_SHIFT;
4494 struct extent_buffer *eb;
4495 struct extent_buffer *exists = NULL;
4497 struct address_space *mapping = tree->mapping;
4502 eb = find_extent_buffer(tree, start);
4506 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4510 for (i = 0; i < num_pages; i++, index++) {
4511 p = find_or_create_page(mapping, index, GFP_NOFS);
4515 spin_lock(&mapping->private_lock);
4516 if (PagePrivate(p)) {
4518 * We could have already allocated an eb for this page
4519 * and attached one so lets see if we can get a ref on
4520 * the existing eb, and if we can we know it's good and
4521 * we can just return that one, else we know we can just
4522 * overwrite page->private.
4524 exists = (struct extent_buffer *)p->private;
4525 if (atomic_inc_not_zero(&exists->refs)) {
4526 spin_unlock(&mapping->private_lock);
4528 page_cache_release(p);
4529 mark_extent_buffer_accessed(exists);
4534 * Do this so attach doesn't complain and we need to
4535 * drop the ref the old guy had.
4537 ClearPagePrivate(p);
4538 WARN_ON(PageDirty(p));
4539 page_cache_release(p);
4541 attach_extent_buffer_page(eb, p);
4542 spin_unlock(&mapping->private_lock);
4543 WARN_ON(PageDirty(p));
4544 mark_page_accessed(p);
4546 if (!PageUptodate(p))
4550 * see below about how we avoid a nasty race with release page
4551 * and why we unlock later
4555 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4557 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4561 spin_lock(&tree->buffer_lock);
4562 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4563 spin_unlock(&tree->buffer_lock);
4564 radix_tree_preload_end();
4565 if (ret == -EEXIST) {
4566 exists = find_extent_buffer(tree, start);
4572 /* add one reference for the tree */
4573 check_buffer_tree_ref(eb);
4576 * there is a race where release page may have
4577 * tried to find this extent buffer in the radix
4578 * but failed. It will tell the VM it is safe to
4579 * reclaim the, and it will clear the page private bit.
4580 * We must make sure to set the page private bit properly
4581 * after the extent buffer is in the radix tree so
4582 * it doesn't get lost
4584 SetPageChecked(eb->pages[0]);
4585 for (i = 1; i < num_pages; i++) {
4586 p = extent_buffer_page(eb, i);
4587 ClearPageChecked(p);
4590 unlock_page(eb->pages[0]);
4594 for (i = 0; i < num_pages; i++) {
4596 unlock_page(eb->pages[i]);
4599 WARN_ON(!atomic_dec_and_test(&eb->refs));
4600 btrfs_release_extent_buffer(eb);
4604 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4606 struct extent_buffer *eb =
4607 container_of(head, struct extent_buffer, rcu_head);
4609 __free_extent_buffer(eb);
4612 /* Expects to have eb->eb_lock already held */
4613 static int release_extent_buffer(struct extent_buffer *eb)
4615 WARN_ON(atomic_read(&eb->refs) == 0);
4616 if (atomic_dec_and_test(&eb->refs)) {
4617 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4618 spin_unlock(&eb->refs_lock);
4620 struct extent_io_tree *tree = eb->tree;
4622 spin_unlock(&eb->refs_lock);
4624 spin_lock(&tree->buffer_lock);
4625 radix_tree_delete(&tree->buffer,
4626 eb->start >> PAGE_CACHE_SHIFT);
4627 spin_unlock(&tree->buffer_lock);
4630 /* Should be safe to release our pages at this point */
4631 btrfs_release_extent_buffer_page(eb, 0);
4632 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4635 spin_unlock(&eb->refs_lock);
4640 void free_extent_buffer(struct extent_buffer *eb)
4648 refs = atomic_read(&eb->refs);
4651 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4656 spin_lock(&eb->refs_lock);
4657 if (atomic_read(&eb->refs) == 2 &&
4658 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4659 atomic_dec(&eb->refs);
4661 if (atomic_read(&eb->refs) == 2 &&
4662 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4663 !extent_buffer_under_io(eb) &&
4664 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4665 atomic_dec(&eb->refs);
4668 * I know this is terrible, but it's temporary until we stop tracking
4669 * the uptodate bits and such for the extent buffers.
4671 release_extent_buffer(eb);
4674 void free_extent_buffer_stale(struct extent_buffer *eb)
4679 spin_lock(&eb->refs_lock);
4680 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4682 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4683 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4684 atomic_dec(&eb->refs);
4685 release_extent_buffer(eb);
4688 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4691 unsigned long num_pages;
4694 num_pages = num_extent_pages(eb->start, eb->len);
4696 for (i = 0; i < num_pages; i++) {
4697 page = extent_buffer_page(eb, i);
4698 if (!PageDirty(page))
4702 WARN_ON(!PagePrivate(page));
4704 clear_page_dirty_for_io(page);
4705 spin_lock_irq(&page->mapping->tree_lock);
4706 if (!PageDirty(page)) {
4707 radix_tree_tag_clear(&page->mapping->page_tree,
4709 PAGECACHE_TAG_DIRTY);
4711 spin_unlock_irq(&page->mapping->tree_lock);
4712 ClearPageError(page);
4715 WARN_ON(atomic_read(&eb->refs) == 0);
4718 int set_extent_buffer_dirty(struct extent_buffer *eb)
4721 unsigned long num_pages;
4724 check_buffer_tree_ref(eb);
4726 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4728 num_pages = num_extent_pages(eb->start, eb->len);
4729 WARN_ON(atomic_read(&eb->refs) == 0);
4730 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4732 for (i = 0; i < num_pages; i++)
4733 set_page_dirty(extent_buffer_page(eb, i));
4737 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4741 unsigned long num_pages;
4743 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4744 num_pages = num_extent_pages(eb->start, eb->len);
4745 for (i = 0; i < num_pages; i++) {
4746 page = extent_buffer_page(eb, i);
4748 ClearPageUptodate(page);
4753 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4757 unsigned long num_pages;
4759 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4760 num_pages = num_extent_pages(eb->start, eb->len);
4761 for (i = 0; i < num_pages; i++) {
4762 page = extent_buffer_page(eb, i);
4763 SetPageUptodate(page);
4768 int extent_buffer_uptodate(struct extent_buffer *eb)
4770 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4773 int read_extent_buffer_pages(struct extent_io_tree *tree,
4774 struct extent_buffer *eb, u64 start, int wait,
4775 get_extent_t *get_extent, int mirror_num)
4778 unsigned long start_i;
4782 int locked_pages = 0;
4783 int all_uptodate = 1;
4784 unsigned long num_pages;
4785 unsigned long num_reads = 0;
4786 struct bio *bio = NULL;
4787 unsigned long bio_flags = 0;
4789 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4793 WARN_ON(start < eb->start);
4794 start_i = (start >> PAGE_CACHE_SHIFT) -
4795 (eb->start >> PAGE_CACHE_SHIFT);
4800 num_pages = num_extent_pages(eb->start, eb->len);
4801 for (i = start_i; i < num_pages; i++) {
4802 page = extent_buffer_page(eb, i);
4803 if (wait == WAIT_NONE) {
4804 if (!trylock_page(page))
4810 if (!PageUptodate(page)) {
4817 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4821 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4822 eb->read_mirror = 0;
4823 atomic_set(&eb->io_pages, num_reads);
4824 for (i = start_i; i < num_pages; i++) {
4825 page = extent_buffer_page(eb, i);
4826 if (!PageUptodate(page)) {
4827 ClearPageError(page);
4828 err = __extent_read_full_page(tree, page,
4830 mirror_num, &bio_flags,
4840 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4846 if (ret || wait != WAIT_COMPLETE)
4849 for (i = start_i; i < num_pages; i++) {
4850 page = extent_buffer_page(eb, i);
4851 wait_on_page_locked(page);
4852 if (!PageUptodate(page))
4860 while (locked_pages > 0) {
4861 page = extent_buffer_page(eb, i);
4869 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4870 unsigned long start,
4877 char *dst = (char *)dstv;
4878 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4879 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4881 WARN_ON(start > eb->len);
4882 WARN_ON(start + len > eb->start + eb->len);
4884 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4887 page = extent_buffer_page(eb, i);
4889 cur = min(len, (PAGE_CACHE_SIZE - offset));
4890 kaddr = page_address(page);
4891 memcpy(dst, kaddr + offset, cur);
4900 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4901 unsigned long min_len, char **map,
4902 unsigned long *map_start,
4903 unsigned long *map_len)
4905 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4908 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4909 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4910 unsigned long end_i = (start_offset + start + min_len - 1) >>
4917 offset = start_offset;
4921 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4924 if (start + min_len > eb->len) {
4925 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4927 eb->start, eb->len, start, min_len);
4931 p = extent_buffer_page(eb, i);
4932 kaddr = page_address(p);
4933 *map = kaddr + offset;
4934 *map_len = PAGE_CACHE_SIZE - offset;
4938 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4939 unsigned long start,
4946 char *ptr = (char *)ptrv;
4947 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4948 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4951 WARN_ON(start > eb->len);
4952 WARN_ON(start + len > eb->start + eb->len);
4954 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4957 page = extent_buffer_page(eb, i);
4959 cur = min(len, (PAGE_CACHE_SIZE - offset));
4961 kaddr = page_address(page);
4962 ret = memcmp(ptr, kaddr + offset, cur);
4974 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4975 unsigned long start, unsigned long len)
4981 char *src = (char *)srcv;
4982 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4983 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4985 WARN_ON(start > eb->len);
4986 WARN_ON(start + len > eb->start + eb->len);
4988 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4991 page = extent_buffer_page(eb, i);
4992 WARN_ON(!PageUptodate(page));
4994 cur = min(len, PAGE_CACHE_SIZE - offset);
4995 kaddr = page_address(page);
4996 memcpy(kaddr + offset, src, cur);
5005 void memset_extent_buffer(struct extent_buffer *eb, char c,
5006 unsigned long start, unsigned long len)
5012 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5013 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5015 WARN_ON(start > eb->len);
5016 WARN_ON(start + len > eb->start + eb->len);
5018 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5021 page = extent_buffer_page(eb, i);
5022 WARN_ON(!PageUptodate(page));
5024 cur = min(len, PAGE_CACHE_SIZE - offset);
5025 kaddr = page_address(page);
5026 memset(kaddr + offset, c, cur);
5034 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5035 unsigned long dst_offset, unsigned long src_offset,
5038 u64 dst_len = dst->len;
5043 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5044 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5046 WARN_ON(src->len != dst_len);
5048 offset = (start_offset + dst_offset) &
5049 (PAGE_CACHE_SIZE - 1);
5052 page = extent_buffer_page(dst, i);
5053 WARN_ON(!PageUptodate(page));
5055 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5057 kaddr = page_address(page);
5058 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5067 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5069 unsigned long distance = (src > dst) ? src - dst : dst - src;
5070 return distance < len;
5073 static void copy_pages(struct page *dst_page, struct page *src_page,
5074 unsigned long dst_off, unsigned long src_off,
5077 char *dst_kaddr = page_address(dst_page);
5079 int must_memmove = 0;
5081 if (dst_page != src_page) {
5082 src_kaddr = page_address(src_page);
5084 src_kaddr = dst_kaddr;
5085 if (areas_overlap(src_off, dst_off, len))
5090 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5092 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5095 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5096 unsigned long src_offset, unsigned long len)
5099 size_t dst_off_in_page;
5100 size_t src_off_in_page;
5101 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5102 unsigned long dst_i;
5103 unsigned long src_i;
5105 if (src_offset + len > dst->len) {
5106 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5107 "len %lu dst len %lu\n", src_offset, len, dst->len);
5110 if (dst_offset + len > dst->len) {
5111 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5112 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5117 dst_off_in_page = (start_offset + dst_offset) &
5118 (PAGE_CACHE_SIZE - 1);
5119 src_off_in_page = (start_offset + src_offset) &
5120 (PAGE_CACHE_SIZE - 1);
5122 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5123 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5125 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5127 cur = min_t(unsigned long, cur,
5128 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5130 copy_pages(extent_buffer_page(dst, dst_i),
5131 extent_buffer_page(dst, src_i),
5132 dst_off_in_page, src_off_in_page, cur);
5140 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5141 unsigned long src_offset, unsigned long len)
5144 size_t dst_off_in_page;
5145 size_t src_off_in_page;
5146 unsigned long dst_end = dst_offset + len - 1;
5147 unsigned long src_end = src_offset + len - 1;
5148 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5149 unsigned long dst_i;
5150 unsigned long src_i;
5152 if (src_offset + len > dst->len) {
5153 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5154 "len %lu len %lu\n", src_offset, len, dst->len);
5157 if (dst_offset + len > dst->len) {
5158 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5159 "len %lu len %lu\n", dst_offset, len, dst->len);
5162 if (dst_offset < src_offset) {
5163 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5167 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5168 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5170 dst_off_in_page = (start_offset + dst_end) &
5171 (PAGE_CACHE_SIZE - 1);
5172 src_off_in_page = (start_offset + src_end) &
5173 (PAGE_CACHE_SIZE - 1);
5175 cur = min_t(unsigned long, len, src_off_in_page + 1);
5176 cur = min(cur, dst_off_in_page + 1);
5177 copy_pages(extent_buffer_page(dst, dst_i),
5178 extent_buffer_page(dst, src_i),
5179 dst_off_in_page - cur + 1,
5180 src_off_in_page - cur + 1, cur);
5188 int try_release_extent_buffer(struct page *page)
5190 struct extent_buffer *eb;
5193 * We need to make sure noboody is attaching this page to an eb right
5196 spin_lock(&page->mapping->private_lock);
5197 if (!PagePrivate(page)) {
5198 spin_unlock(&page->mapping->private_lock);
5202 eb = (struct extent_buffer *)page->private;
5206 * This is a little awful but should be ok, we need to make sure that
5207 * the eb doesn't disappear out from under us while we're looking at
5210 spin_lock(&eb->refs_lock);
5211 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5212 spin_unlock(&eb->refs_lock);
5213 spin_unlock(&page->mapping->private_lock);
5216 spin_unlock(&page->mapping->private_lock);
5219 * If tree ref isn't set then we know the ref on this eb is a real ref,
5220 * so just return, this page will likely be freed soon anyway.
5222 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5223 spin_unlock(&eb->refs_lock);
5227 return release_extent_buffer(eb);