1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 #include "check-integrity.h"
22 #include "rcu-string.h"
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
26 static struct bio_set *btrfs_bioset;
28 #ifdef CONFIG_BTRFS_DEBUG
29 static LIST_HEAD(buffers);
30 static LIST_HEAD(states);
32 static DEFINE_SPINLOCK(leak_lock);
35 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
39 spin_lock_irqsave(&leak_lock, flags);
41 spin_unlock_irqrestore(&leak_lock, flags);
45 void btrfs_leak_debug_del(struct list_head *entry)
49 spin_lock_irqsave(&leak_lock, flags);
51 spin_unlock_irqrestore(&leak_lock, flags);
55 void btrfs_leak_debug_check(void)
57 struct extent_state *state;
58 struct extent_buffer *eb;
60 while (!list_empty(&states)) {
61 state = list_entry(states.next, struct extent_state, leak_list);
62 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
63 "state %lu in tree %p refs %d\n",
64 (unsigned long long)state->start,
65 (unsigned long long)state->end,
66 state->state, state->tree, atomic_read(&state->refs));
67 list_del(&state->leak_list);
68 kmem_cache_free(extent_state_cache, state);
71 while (!list_empty(&buffers)) {
72 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
73 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
74 "refs %d\n", (unsigned long long)eb->start,
75 eb->len, atomic_read(&eb->refs));
76 list_del(&eb->leak_list);
77 kmem_cache_free(extent_buffer_cache, eb);
81 #define btrfs_leak_debug_add(new, head) do {} while (0)
82 #define btrfs_leak_debug_del(entry) do {} while (0)
83 #define btrfs_leak_debug_check() do {} while (0)
86 #define BUFFER_LRU_MAX 64
91 struct rb_node rb_node;
94 struct extent_page_data {
96 struct extent_io_tree *tree;
97 get_extent_t *get_extent;
98 unsigned long bio_flags;
100 /* tells writepage not to lock the state bits for this range
101 * it still does the unlocking
103 unsigned int extent_locked:1;
105 /* tells the submit_bio code to use a WRITE_SYNC */
106 unsigned int sync_io:1;
109 static noinline void flush_write_bio(void *data);
110 static inline struct btrfs_fs_info *
111 tree_fs_info(struct extent_io_tree *tree)
113 return btrfs_sb(tree->mapping->host->i_sb);
116 int __init extent_io_init(void)
118 extent_state_cache = kmem_cache_create("btrfs_extent_state",
119 sizeof(struct extent_state), 0,
120 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
121 if (!extent_state_cache)
124 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
125 sizeof(struct extent_buffer), 0,
126 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
127 if (!extent_buffer_cache)
128 goto free_state_cache;
130 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
131 offsetof(struct btrfs_io_bio, bio));
133 goto free_buffer_cache;
137 kmem_cache_destroy(extent_buffer_cache);
138 extent_buffer_cache = NULL;
141 kmem_cache_destroy(extent_state_cache);
142 extent_state_cache = NULL;
146 void extent_io_exit(void)
148 btrfs_leak_debug_check();
151 * Make sure all delayed rcu free are flushed before we
155 if (extent_state_cache)
156 kmem_cache_destroy(extent_state_cache);
157 if (extent_buffer_cache)
158 kmem_cache_destroy(extent_buffer_cache);
160 bioset_free(btrfs_bioset);
163 void extent_io_tree_init(struct extent_io_tree *tree,
164 struct address_space *mapping)
166 tree->state = RB_ROOT;
167 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
169 tree->dirty_bytes = 0;
170 spin_lock_init(&tree->lock);
171 spin_lock_init(&tree->buffer_lock);
172 tree->mapping = mapping;
175 static struct extent_state *alloc_extent_state(gfp_t mask)
177 struct extent_state *state;
179 state = kmem_cache_alloc(extent_state_cache, mask);
185 btrfs_leak_debug_add(&state->leak_list, &states);
186 atomic_set(&state->refs, 1);
187 init_waitqueue_head(&state->wq);
188 trace_alloc_extent_state(state, mask, _RET_IP_);
192 void free_extent_state(struct extent_state *state)
196 if (atomic_dec_and_test(&state->refs)) {
197 WARN_ON(state->tree);
198 btrfs_leak_debug_del(&state->leak_list);
199 trace_free_extent_state(state, _RET_IP_);
200 kmem_cache_free(extent_state_cache, state);
204 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
205 struct rb_node *node)
207 struct rb_node **p = &root->rb_node;
208 struct rb_node *parent = NULL;
209 struct tree_entry *entry;
213 entry = rb_entry(parent, struct tree_entry, rb_node);
215 if (offset < entry->start)
217 else if (offset > entry->end)
223 rb_link_node(node, parent, p);
224 rb_insert_color(node, root);
228 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
229 struct rb_node **prev_ret,
230 struct rb_node **next_ret)
232 struct rb_root *root = &tree->state;
233 struct rb_node *n = root->rb_node;
234 struct rb_node *prev = NULL;
235 struct rb_node *orig_prev = NULL;
236 struct tree_entry *entry;
237 struct tree_entry *prev_entry = NULL;
240 entry = rb_entry(n, struct tree_entry, rb_node);
244 if (offset < entry->start)
246 else if (offset > entry->end)
254 while (prev && offset > prev_entry->end) {
255 prev = rb_next(prev);
256 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
263 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
264 while (prev && offset < prev_entry->start) {
265 prev = rb_prev(prev);
266 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
273 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
276 struct rb_node *prev = NULL;
279 ret = __etree_search(tree, offset, &prev, NULL);
285 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
286 struct extent_state *other)
288 if (tree->ops && tree->ops->merge_extent_hook)
289 tree->ops->merge_extent_hook(tree->mapping->host, new,
294 * utility function to look for merge candidates inside a given range.
295 * Any extents with matching state are merged together into a single
296 * extent in the tree. Extents with EXTENT_IO in their state field
297 * are not merged because the end_io handlers need to be able to do
298 * operations on them without sleeping (or doing allocations/splits).
300 * This should be called with the tree lock held.
302 static void merge_state(struct extent_io_tree *tree,
303 struct extent_state *state)
305 struct extent_state *other;
306 struct rb_node *other_node;
308 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
311 other_node = rb_prev(&state->rb_node);
313 other = rb_entry(other_node, struct extent_state, rb_node);
314 if (other->end == state->start - 1 &&
315 other->state == state->state) {
316 merge_cb(tree, state, other);
317 state->start = other->start;
319 rb_erase(&other->rb_node, &tree->state);
320 free_extent_state(other);
323 other_node = rb_next(&state->rb_node);
325 other = rb_entry(other_node, struct extent_state, rb_node);
326 if (other->start == state->end + 1 &&
327 other->state == state->state) {
328 merge_cb(tree, state, other);
329 state->end = other->end;
331 rb_erase(&other->rb_node, &tree->state);
332 free_extent_state(other);
337 static void set_state_cb(struct extent_io_tree *tree,
338 struct extent_state *state, unsigned long *bits)
340 if (tree->ops && tree->ops->set_bit_hook)
341 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
344 static void clear_state_cb(struct extent_io_tree *tree,
345 struct extent_state *state, unsigned long *bits)
347 if (tree->ops && tree->ops->clear_bit_hook)
348 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
351 static void set_state_bits(struct extent_io_tree *tree,
352 struct extent_state *state, unsigned long *bits);
355 * insert an extent_state struct into the tree. 'bits' are set on the
356 * struct before it is inserted.
358 * This may return -EEXIST if the extent is already there, in which case the
359 * state struct is freed.
361 * The tree lock is not taken internally. This is a utility function and
362 * probably isn't what you want to call (see set/clear_extent_bit).
364 static int insert_state(struct extent_io_tree *tree,
365 struct extent_state *state, u64 start, u64 end,
368 struct rb_node *node;
371 WARN(1, KERN_ERR "btrfs end < start %llu %llu\n",
372 (unsigned long long)end,
373 (unsigned long long)start);
374 state->start = start;
377 set_state_bits(tree, state, bits);
379 node = tree_insert(&tree->state, end, &state->rb_node);
381 struct extent_state *found;
382 found = rb_entry(node, struct extent_state, rb_node);
383 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
384 "%llu %llu\n", (unsigned long long)found->start,
385 (unsigned long long)found->end,
386 (unsigned long long)start, (unsigned long long)end);
390 merge_state(tree, state);
394 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
397 if (tree->ops && tree->ops->split_extent_hook)
398 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
402 * split a given extent state struct in two, inserting the preallocated
403 * struct 'prealloc' as the newly created second half. 'split' indicates an
404 * offset inside 'orig' where it should be split.
407 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
408 * are two extent state structs in the tree:
409 * prealloc: [orig->start, split - 1]
410 * orig: [ split, orig->end ]
412 * The tree locks are not taken by this function. They need to be held
415 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
416 struct extent_state *prealloc, u64 split)
418 struct rb_node *node;
420 split_cb(tree, orig, split);
422 prealloc->start = orig->start;
423 prealloc->end = split - 1;
424 prealloc->state = orig->state;
427 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
429 free_extent_state(prealloc);
432 prealloc->tree = tree;
436 static struct extent_state *next_state(struct extent_state *state)
438 struct rb_node *next = rb_next(&state->rb_node);
440 return rb_entry(next, struct extent_state, rb_node);
446 * utility function to clear some bits in an extent state struct.
447 * it will optionally wake up any one waiting on this state (wake == 1).
449 * If no bits are set on the state struct after clearing things, the
450 * struct is freed and removed from the tree
452 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
453 struct extent_state *state,
454 unsigned long *bits, int wake)
456 struct extent_state *next;
457 unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
459 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
460 u64 range = state->end - state->start + 1;
461 WARN_ON(range > tree->dirty_bytes);
462 tree->dirty_bytes -= range;
464 clear_state_cb(tree, state, bits);
465 state->state &= ~bits_to_clear;
468 if (state->state == 0) {
469 next = next_state(state);
471 rb_erase(&state->rb_node, &tree->state);
473 free_extent_state(state);
478 merge_state(tree, state);
479 next = next_state(state);
484 static struct extent_state *
485 alloc_extent_state_atomic(struct extent_state *prealloc)
488 prealloc = alloc_extent_state(GFP_ATOMIC);
493 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
495 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
496 "Extent tree was modified by another "
497 "thread while locked.");
501 * clear some bits on a range in the tree. This may require splitting
502 * or inserting elements in the tree, so the gfp mask is used to
503 * indicate which allocations or sleeping are allowed.
505 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
506 * the given range from the tree regardless of state (ie for truncate).
508 * the range [start, end] is inclusive.
510 * This takes the tree lock, and returns 0 on success and < 0 on error.
512 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
513 unsigned long bits, int wake, int delete,
514 struct extent_state **cached_state,
517 struct extent_state *state;
518 struct extent_state *cached;
519 struct extent_state *prealloc = NULL;
520 struct rb_node *node;
526 bits |= ~EXTENT_CTLBITS;
527 bits |= EXTENT_FIRST_DELALLOC;
529 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
532 if (!prealloc && (mask & __GFP_WAIT)) {
533 prealloc = alloc_extent_state(mask);
538 spin_lock(&tree->lock);
540 cached = *cached_state;
543 *cached_state = NULL;
547 if (cached && cached->tree && cached->start <= start &&
548 cached->end > start) {
550 atomic_dec(&cached->refs);
555 free_extent_state(cached);
558 * this search will find the extents that end after
561 node = tree_search(tree, start);
564 state = rb_entry(node, struct extent_state, rb_node);
566 if (state->start > end)
568 WARN_ON(state->end < start);
569 last_end = state->end;
571 /* the state doesn't have the wanted bits, go ahead */
572 if (!(state->state & bits)) {
573 state = next_state(state);
578 * | ---- desired range ---- |
580 * | ------------- state -------------- |
582 * We need to split the extent we found, and may flip
583 * bits on second half.
585 * If the extent we found extends past our range, we
586 * just split and search again. It'll get split again
587 * the next time though.
589 * If the extent we found is inside our range, we clear
590 * the desired bit on it.
593 if (state->start < start) {
594 prealloc = alloc_extent_state_atomic(prealloc);
596 err = split_state(tree, state, prealloc, start);
598 extent_io_tree_panic(tree, err);
603 if (state->end <= end) {
604 state = clear_state_bit(tree, state, &bits, wake);
610 * | ---- desired range ---- |
612 * We need to split the extent, and clear the bit
615 if (state->start <= end && state->end > end) {
616 prealloc = alloc_extent_state_atomic(prealloc);
618 err = split_state(tree, state, prealloc, end + 1);
620 extent_io_tree_panic(tree, err);
625 clear_state_bit(tree, prealloc, &bits, wake);
631 state = clear_state_bit(tree, state, &bits, wake);
633 if (last_end == (u64)-1)
635 start = last_end + 1;
636 if (start <= end && state && !need_resched())
641 spin_unlock(&tree->lock);
643 free_extent_state(prealloc);
650 spin_unlock(&tree->lock);
651 if (mask & __GFP_WAIT)
656 static void wait_on_state(struct extent_io_tree *tree,
657 struct extent_state *state)
658 __releases(tree->lock)
659 __acquires(tree->lock)
662 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
663 spin_unlock(&tree->lock);
665 spin_lock(&tree->lock);
666 finish_wait(&state->wq, &wait);
670 * waits for one or more bits to clear on a range in the state tree.
671 * The range [start, end] is inclusive.
672 * The tree lock is taken by this function
674 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
677 struct extent_state *state;
678 struct rb_node *node;
680 spin_lock(&tree->lock);
684 * this search will find all the extents that end after
687 node = tree_search(tree, start);
691 state = rb_entry(node, struct extent_state, rb_node);
693 if (state->start > end)
696 if (state->state & bits) {
697 start = state->start;
698 atomic_inc(&state->refs);
699 wait_on_state(tree, state);
700 free_extent_state(state);
703 start = state->end + 1;
708 cond_resched_lock(&tree->lock);
711 spin_unlock(&tree->lock);
714 static void set_state_bits(struct extent_io_tree *tree,
715 struct extent_state *state,
718 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
720 set_state_cb(tree, state, bits);
721 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
722 u64 range = state->end - state->start + 1;
723 tree->dirty_bytes += range;
725 state->state |= bits_to_set;
728 static void cache_state(struct extent_state *state,
729 struct extent_state **cached_ptr)
731 if (cached_ptr && !(*cached_ptr)) {
732 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
734 atomic_inc(&state->refs);
739 static void uncache_state(struct extent_state **cached_ptr)
741 if (cached_ptr && (*cached_ptr)) {
742 struct extent_state *state = *cached_ptr;
744 free_extent_state(state);
749 * set some bits on a range in the tree. This may require allocations or
750 * sleeping, so the gfp mask is used to indicate what is allowed.
752 * If any of the exclusive bits are set, this will fail with -EEXIST if some
753 * part of the range already has the desired bits set. The start of the
754 * existing range is returned in failed_start in this case.
756 * [start, end] is inclusive This takes the tree lock.
759 static int __must_check
760 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
761 unsigned long bits, unsigned long exclusive_bits,
762 u64 *failed_start, struct extent_state **cached_state,
765 struct extent_state *state;
766 struct extent_state *prealloc = NULL;
767 struct rb_node *node;
772 bits |= EXTENT_FIRST_DELALLOC;
774 if (!prealloc && (mask & __GFP_WAIT)) {
775 prealloc = alloc_extent_state(mask);
779 spin_lock(&tree->lock);
780 if (cached_state && *cached_state) {
781 state = *cached_state;
782 if (state->start <= start && state->end > start &&
784 node = &state->rb_node;
789 * this search will find all the extents that end after
792 node = tree_search(tree, start);
794 prealloc = alloc_extent_state_atomic(prealloc);
796 err = insert_state(tree, prealloc, start, end, &bits);
798 extent_io_tree_panic(tree, err);
803 state = rb_entry(node, struct extent_state, rb_node);
805 last_start = state->start;
806 last_end = state->end;
809 * | ---- desired range ---- |
812 * Just lock what we found and keep going
814 if (state->start == start && state->end <= end) {
815 if (state->state & exclusive_bits) {
816 *failed_start = state->start;
821 set_state_bits(tree, state, &bits);
822 cache_state(state, cached_state);
823 merge_state(tree, state);
824 if (last_end == (u64)-1)
826 start = last_end + 1;
827 state = next_state(state);
828 if (start < end && state && state->start == start &&
835 * | ---- desired range ---- |
838 * | ------------- state -------------- |
840 * We need to split the extent we found, and may flip bits on
843 * If the extent we found extends past our
844 * range, we just split and search again. It'll get split
845 * again the next time though.
847 * If the extent we found is inside our range, we set the
850 if (state->start < start) {
851 if (state->state & exclusive_bits) {
852 *failed_start = start;
857 prealloc = alloc_extent_state_atomic(prealloc);
859 err = split_state(tree, state, prealloc, start);
861 extent_io_tree_panic(tree, err);
866 if (state->end <= end) {
867 set_state_bits(tree, state, &bits);
868 cache_state(state, cached_state);
869 merge_state(tree, state);
870 if (last_end == (u64)-1)
872 start = last_end + 1;
873 state = next_state(state);
874 if (start < end && state && state->start == start &&
881 * | ---- desired range ---- |
882 * | state | or | state |
884 * There's a hole, we need to insert something in it and
885 * ignore the extent we found.
887 if (state->start > start) {
889 if (end < last_start)
892 this_end = last_start - 1;
894 prealloc = alloc_extent_state_atomic(prealloc);
898 * Avoid to free 'prealloc' if it can be merged with
901 err = insert_state(tree, prealloc, start, this_end,
904 extent_io_tree_panic(tree, err);
906 cache_state(prealloc, cached_state);
908 start = this_end + 1;
912 * | ---- desired range ---- |
914 * We need to split the extent, and set the bit
917 if (state->start <= end && state->end > end) {
918 if (state->state & exclusive_bits) {
919 *failed_start = start;
924 prealloc = alloc_extent_state_atomic(prealloc);
926 err = split_state(tree, state, prealloc, end + 1);
928 extent_io_tree_panic(tree, err);
930 set_state_bits(tree, prealloc, &bits);
931 cache_state(prealloc, cached_state);
932 merge_state(tree, prealloc);
940 spin_unlock(&tree->lock);
942 free_extent_state(prealloc);
949 spin_unlock(&tree->lock);
950 if (mask & __GFP_WAIT)
955 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
956 unsigned long bits, u64 * failed_start,
957 struct extent_state **cached_state, gfp_t mask)
959 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
965 * convert_extent_bit - convert all bits in a given range from one bit to
967 * @tree: the io tree to search
968 * @start: the start offset in bytes
969 * @end: the end offset in bytes (inclusive)
970 * @bits: the bits to set in this range
971 * @clear_bits: the bits to clear in this range
972 * @cached_state: state that we're going to cache
973 * @mask: the allocation mask
975 * This will go through and set bits for the given range. If any states exist
976 * already in this range they are set with the given bit and cleared of the
977 * clear_bits. This is only meant to be used by things that are mergeable, ie
978 * converting from say DELALLOC to DIRTY. This is not meant to be used with
979 * boundary bits like LOCK.
981 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
982 unsigned long bits, unsigned long clear_bits,
983 struct extent_state **cached_state, gfp_t mask)
985 struct extent_state *state;
986 struct extent_state *prealloc = NULL;
987 struct rb_node *node;
993 if (!prealloc && (mask & __GFP_WAIT)) {
994 prealloc = alloc_extent_state(mask);
999 spin_lock(&tree->lock);
1000 if (cached_state && *cached_state) {
1001 state = *cached_state;
1002 if (state->start <= start && state->end > start &&
1004 node = &state->rb_node;
1010 * this search will find all the extents that end after
1013 node = tree_search(tree, start);
1015 prealloc = alloc_extent_state_atomic(prealloc);
1020 err = insert_state(tree, prealloc, start, end, &bits);
1023 extent_io_tree_panic(tree, err);
1026 state = rb_entry(node, struct extent_state, rb_node);
1028 last_start = state->start;
1029 last_end = state->end;
1032 * | ---- desired range ---- |
1035 * Just lock what we found and keep going
1037 if (state->start == start && state->end <= end) {
1038 set_state_bits(tree, state, &bits);
1039 cache_state(state, cached_state);
1040 state = clear_state_bit(tree, state, &clear_bits, 0);
1041 if (last_end == (u64)-1)
1043 start = last_end + 1;
1044 if (start < end && state && state->start == start &&
1051 * | ---- desired range ---- |
1054 * | ------------- state -------------- |
1056 * We need to split the extent we found, and may flip bits on
1059 * If the extent we found extends past our
1060 * range, we just split and search again. It'll get split
1061 * again the next time though.
1063 * If the extent we found is inside our range, we set the
1064 * desired bit on it.
1066 if (state->start < start) {
1067 prealloc = alloc_extent_state_atomic(prealloc);
1072 err = split_state(tree, state, prealloc, start);
1074 extent_io_tree_panic(tree, err);
1078 if (state->end <= end) {
1079 set_state_bits(tree, state, &bits);
1080 cache_state(state, cached_state);
1081 state = clear_state_bit(tree, state, &clear_bits, 0);
1082 if (last_end == (u64)-1)
1084 start = last_end + 1;
1085 if (start < end && state && state->start == start &&
1092 * | ---- desired range ---- |
1093 * | state | or | state |
1095 * There's a hole, we need to insert something in it and
1096 * ignore the extent we found.
1098 if (state->start > start) {
1100 if (end < last_start)
1103 this_end = last_start - 1;
1105 prealloc = alloc_extent_state_atomic(prealloc);
1112 * Avoid to free 'prealloc' if it can be merged with
1115 err = insert_state(tree, prealloc, start, this_end,
1118 extent_io_tree_panic(tree, err);
1119 cache_state(prealloc, cached_state);
1121 start = this_end + 1;
1125 * | ---- desired range ---- |
1127 * We need to split the extent, and set the bit
1130 if (state->start <= end && state->end > end) {
1131 prealloc = alloc_extent_state_atomic(prealloc);
1137 err = split_state(tree, state, prealloc, end + 1);
1139 extent_io_tree_panic(tree, err);
1141 set_state_bits(tree, prealloc, &bits);
1142 cache_state(prealloc, cached_state);
1143 clear_state_bit(tree, prealloc, &clear_bits, 0);
1151 spin_unlock(&tree->lock);
1153 free_extent_state(prealloc);
1160 spin_unlock(&tree->lock);
1161 if (mask & __GFP_WAIT)
1166 /* wrappers around set/clear extent bit */
1167 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1170 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1174 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1175 unsigned long bits, gfp_t mask)
1177 return set_extent_bit(tree, start, end, bits, NULL,
1181 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1182 unsigned long bits, gfp_t mask)
1184 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1187 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1188 struct extent_state **cached_state, gfp_t mask)
1190 return set_extent_bit(tree, start, end,
1191 EXTENT_DELALLOC | EXTENT_UPTODATE,
1192 NULL, cached_state, mask);
1195 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1196 struct extent_state **cached_state, gfp_t mask)
1198 return set_extent_bit(tree, start, end,
1199 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1200 NULL, cached_state, mask);
1203 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1206 return clear_extent_bit(tree, start, end,
1207 EXTENT_DIRTY | EXTENT_DELALLOC |
1208 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1211 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1214 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1218 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1219 struct extent_state **cached_state, gfp_t mask)
1221 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1222 cached_state, mask);
1225 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1226 struct extent_state **cached_state, gfp_t mask)
1228 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1229 cached_state, mask);
1233 * either insert or lock state struct between start and end use mask to tell
1234 * us if waiting is desired.
1236 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1237 unsigned long bits, struct extent_state **cached_state)
1242 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1243 EXTENT_LOCKED, &failed_start,
1244 cached_state, GFP_NOFS);
1245 if (err == -EEXIST) {
1246 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1247 start = failed_start;
1250 WARN_ON(start > end);
1255 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1257 return lock_extent_bits(tree, start, end, 0, NULL);
1260 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1265 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1266 &failed_start, NULL, GFP_NOFS);
1267 if (err == -EEXIST) {
1268 if (failed_start > start)
1269 clear_extent_bit(tree, start, failed_start - 1,
1270 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1276 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1277 struct extent_state **cached, gfp_t mask)
1279 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1283 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1285 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1289 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1291 unsigned long index = start >> PAGE_CACHE_SHIFT;
1292 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1295 while (index <= end_index) {
1296 page = find_get_page(inode->i_mapping, index);
1297 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1298 clear_page_dirty_for_io(page);
1299 page_cache_release(page);
1305 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1307 unsigned long index = start >> PAGE_CACHE_SHIFT;
1308 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1311 while (index <= end_index) {
1312 page = find_get_page(inode->i_mapping, index);
1313 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1314 account_page_redirty(page);
1315 __set_page_dirty_nobuffers(page);
1316 page_cache_release(page);
1323 * helper function to set both pages and extents in the tree writeback
1325 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1327 unsigned long index = start >> PAGE_CACHE_SHIFT;
1328 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1331 while (index <= end_index) {
1332 page = find_get_page(tree->mapping, index);
1333 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1334 set_page_writeback(page);
1335 page_cache_release(page);
1341 /* find the first state struct with 'bits' set after 'start', and
1342 * return it. tree->lock must be held. NULL will returned if
1343 * nothing was found after 'start'
1345 static struct extent_state *
1346 find_first_extent_bit_state(struct extent_io_tree *tree,
1347 u64 start, unsigned long bits)
1349 struct rb_node *node;
1350 struct extent_state *state;
1353 * this search will find all the extents that end after
1356 node = tree_search(tree, start);
1361 state = rb_entry(node, struct extent_state, rb_node);
1362 if (state->end >= start && (state->state & bits))
1365 node = rb_next(node);
1374 * find the first offset in the io tree with 'bits' set. zero is
1375 * returned if we find something, and *start_ret and *end_ret are
1376 * set to reflect the state struct that was found.
1378 * If nothing was found, 1 is returned. If found something, return 0.
1380 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1381 u64 *start_ret, u64 *end_ret, unsigned long bits,
1382 struct extent_state **cached_state)
1384 struct extent_state *state;
1388 spin_lock(&tree->lock);
1389 if (cached_state && *cached_state) {
1390 state = *cached_state;
1391 if (state->end == start - 1 && state->tree) {
1392 n = rb_next(&state->rb_node);
1394 state = rb_entry(n, struct extent_state,
1396 if (state->state & bits)
1400 free_extent_state(*cached_state);
1401 *cached_state = NULL;
1404 free_extent_state(*cached_state);
1405 *cached_state = NULL;
1408 state = find_first_extent_bit_state(tree, start, bits);
1411 cache_state(state, cached_state);
1412 *start_ret = state->start;
1413 *end_ret = state->end;
1417 spin_unlock(&tree->lock);
1422 * find a contiguous range of bytes in the file marked as delalloc, not
1423 * more than 'max_bytes'. start and end are used to return the range,
1425 * 1 is returned if we find something, 0 if nothing was in the tree
1427 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1428 u64 *start, u64 *end, u64 max_bytes,
1429 struct extent_state **cached_state)
1431 struct rb_node *node;
1432 struct extent_state *state;
1433 u64 cur_start = *start;
1435 u64 total_bytes = 0;
1437 spin_lock(&tree->lock);
1440 * this search will find all the extents that end after
1443 node = tree_search(tree, cur_start);
1451 state = rb_entry(node, struct extent_state, rb_node);
1452 if (found && (state->start != cur_start ||
1453 (state->state & EXTENT_BOUNDARY))) {
1456 if (!(state->state & EXTENT_DELALLOC)) {
1462 *start = state->start;
1463 *cached_state = state;
1464 atomic_inc(&state->refs);
1468 cur_start = state->end + 1;
1469 node = rb_next(node);
1472 total_bytes += state->end - state->start + 1;
1473 if (total_bytes >= max_bytes)
1477 spin_unlock(&tree->lock);
1481 static noinline void __unlock_for_delalloc(struct inode *inode,
1482 struct page *locked_page,
1486 struct page *pages[16];
1487 unsigned long index = start >> PAGE_CACHE_SHIFT;
1488 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1489 unsigned long nr_pages = end_index - index + 1;
1492 if (index == locked_page->index && end_index == index)
1495 while (nr_pages > 0) {
1496 ret = find_get_pages_contig(inode->i_mapping, index,
1497 min_t(unsigned long, nr_pages,
1498 ARRAY_SIZE(pages)), pages);
1499 for (i = 0; i < ret; i++) {
1500 if (pages[i] != locked_page)
1501 unlock_page(pages[i]);
1502 page_cache_release(pages[i]);
1510 static noinline int lock_delalloc_pages(struct inode *inode,
1511 struct page *locked_page,
1515 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1516 unsigned long start_index = index;
1517 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1518 unsigned long pages_locked = 0;
1519 struct page *pages[16];
1520 unsigned long nrpages;
1524 /* the caller is responsible for locking the start index */
1525 if (index == locked_page->index && index == end_index)
1528 /* skip the page at the start index */
1529 nrpages = end_index - index + 1;
1530 while (nrpages > 0) {
1531 ret = find_get_pages_contig(inode->i_mapping, index,
1532 min_t(unsigned long,
1533 nrpages, ARRAY_SIZE(pages)), pages);
1538 /* now we have an array of pages, lock them all */
1539 for (i = 0; i < ret; i++) {
1541 * the caller is taking responsibility for
1544 if (pages[i] != locked_page) {
1545 lock_page(pages[i]);
1546 if (!PageDirty(pages[i]) ||
1547 pages[i]->mapping != inode->i_mapping) {
1549 unlock_page(pages[i]);
1550 page_cache_release(pages[i]);
1554 page_cache_release(pages[i]);
1563 if (ret && pages_locked) {
1564 __unlock_for_delalloc(inode, locked_page,
1566 ((u64)(start_index + pages_locked - 1)) <<
1573 * find a contiguous range of bytes in the file marked as delalloc, not
1574 * more than 'max_bytes'. start and end are used to return the range,
1576 * 1 is returned if we find something, 0 if nothing was in the tree
1578 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1579 struct extent_io_tree *tree,
1580 struct page *locked_page,
1581 u64 *start, u64 *end,
1587 struct extent_state *cached_state = NULL;
1592 /* step one, find a bunch of delalloc bytes starting at start */
1593 delalloc_start = *start;
1595 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1596 max_bytes, &cached_state);
1597 if (!found || delalloc_end <= *start) {
1598 *start = delalloc_start;
1599 *end = delalloc_end;
1600 free_extent_state(cached_state);
1605 * start comes from the offset of locked_page. We have to lock
1606 * pages in order, so we can't process delalloc bytes before
1609 if (delalloc_start < *start)
1610 delalloc_start = *start;
1613 * make sure to limit the number of pages we try to lock down
1616 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1617 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1619 /* step two, lock all the pages after the page that has start */
1620 ret = lock_delalloc_pages(inode, locked_page,
1621 delalloc_start, delalloc_end);
1622 if (ret == -EAGAIN) {
1623 /* some of the pages are gone, lets avoid looping by
1624 * shortening the size of the delalloc range we're searching
1626 free_extent_state(cached_state);
1628 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1629 max_bytes = PAGE_CACHE_SIZE - offset;
1637 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1639 /* step three, lock the state bits for the whole range */
1640 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1642 /* then test to make sure it is all still delalloc */
1643 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1644 EXTENT_DELALLOC, 1, cached_state);
1646 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1647 &cached_state, GFP_NOFS);
1648 __unlock_for_delalloc(inode, locked_page,
1649 delalloc_start, delalloc_end);
1653 free_extent_state(cached_state);
1654 *start = delalloc_start;
1655 *end = delalloc_end;
1660 int extent_clear_unlock_delalloc(struct inode *inode,
1661 struct extent_io_tree *tree,
1662 u64 start, u64 end, struct page *locked_page,
1666 struct page *pages[16];
1667 unsigned long index = start >> PAGE_CACHE_SHIFT;
1668 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1669 unsigned long nr_pages = end_index - index + 1;
1671 unsigned long clear_bits = 0;
1673 if (op & EXTENT_CLEAR_UNLOCK)
1674 clear_bits |= EXTENT_LOCKED;
1675 if (op & EXTENT_CLEAR_DIRTY)
1676 clear_bits |= EXTENT_DIRTY;
1678 if (op & EXTENT_CLEAR_DELALLOC)
1679 clear_bits |= EXTENT_DELALLOC;
1681 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1682 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1683 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1684 EXTENT_SET_PRIVATE2)))
1687 while (nr_pages > 0) {
1688 ret = find_get_pages_contig(inode->i_mapping, index,
1689 min_t(unsigned long,
1690 nr_pages, ARRAY_SIZE(pages)), pages);
1691 for (i = 0; i < ret; i++) {
1693 if (op & EXTENT_SET_PRIVATE2)
1694 SetPagePrivate2(pages[i]);
1696 if (pages[i] == locked_page) {
1697 page_cache_release(pages[i]);
1700 if (op & EXTENT_CLEAR_DIRTY)
1701 clear_page_dirty_for_io(pages[i]);
1702 if (op & EXTENT_SET_WRITEBACK)
1703 set_page_writeback(pages[i]);
1704 if (op & EXTENT_END_WRITEBACK)
1705 end_page_writeback(pages[i]);
1706 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1707 unlock_page(pages[i]);
1708 page_cache_release(pages[i]);
1718 * count the number of bytes in the tree that have a given bit(s)
1719 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1720 * cached. The total number found is returned.
1722 u64 count_range_bits(struct extent_io_tree *tree,
1723 u64 *start, u64 search_end, u64 max_bytes,
1724 unsigned long bits, int contig)
1726 struct rb_node *node;
1727 struct extent_state *state;
1728 u64 cur_start = *start;
1729 u64 total_bytes = 0;
1733 if (search_end <= cur_start) {
1738 spin_lock(&tree->lock);
1739 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1740 total_bytes = tree->dirty_bytes;
1744 * this search will find all the extents that end after
1747 node = tree_search(tree, cur_start);
1752 state = rb_entry(node, struct extent_state, rb_node);
1753 if (state->start > search_end)
1755 if (contig && found && state->start > last + 1)
1757 if (state->end >= cur_start && (state->state & bits) == bits) {
1758 total_bytes += min(search_end, state->end) + 1 -
1759 max(cur_start, state->start);
1760 if (total_bytes >= max_bytes)
1763 *start = max(cur_start, state->start);
1767 } else if (contig && found) {
1770 node = rb_next(node);
1775 spin_unlock(&tree->lock);
1780 * set the private field for a given byte offset in the tree. If there isn't
1781 * an extent_state there already, this does nothing.
1783 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1785 struct rb_node *node;
1786 struct extent_state *state;
1789 spin_lock(&tree->lock);
1791 * this search will find all the extents that end after
1794 node = tree_search(tree, start);
1799 state = rb_entry(node, struct extent_state, rb_node);
1800 if (state->start != start) {
1804 state->private = private;
1806 spin_unlock(&tree->lock);
1810 void extent_cache_csums_dio(struct extent_io_tree *tree, u64 start, u32 csums[],
1813 struct rb_node *node;
1814 struct extent_state *state;
1816 spin_lock(&tree->lock);
1818 * this search will find all the extents that end after
1821 node = tree_search(tree, start);
1824 state = rb_entry(node, struct extent_state, rb_node);
1825 BUG_ON(state->start != start);
1828 state->private = *csums++;
1830 state = next_state(state);
1832 spin_unlock(&tree->lock);
1835 static inline u64 __btrfs_get_bio_offset(struct bio *bio, int bio_index)
1837 struct bio_vec *bvec = bio->bi_io_vec + bio_index;
1839 return page_offset(bvec->bv_page) + bvec->bv_offset;
1842 void extent_cache_csums(struct extent_io_tree *tree, struct bio *bio, int bio_index,
1843 u32 csums[], int count)
1845 struct rb_node *node;
1846 struct extent_state *state = NULL;
1849 spin_lock(&tree->lock);
1851 start = __btrfs_get_bio_offset(bio, bio_index);
1852 if (state == NULL || state->start != start) {
1853 node = tree_search(tree, start);
1856 state = rb_entry(node, struct extent_state, rb_node);
1857 BUG_ON(state->start != start);
1859 state->private = *csums++;
1863 state = next_state(state);
1865 spin_unlock(&tree->lock);
1868 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1870 struct rb_node *node;
1871 struct extent_state *state;
1874 spin_lock(&tree->lock);
1876 * this search will find all the extents that end after
1879 node = tree_search(tree, start);
1884 state = rb_entry(node, struct extent_state, rb_node);
1885 if (state->start != start) {
1889 *private = state->private;
1891 spin_unlock(&tree->lock);
1896 * searches a range in the state tree for a given mask.
1897 * If 'filled' == 1, this returns 1 only if every extent in the tree
1898 * has the bits set. Otherwise, 1 is returned if any bit in the
1899 * range is found set.
1901 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1902 unsigned long bits, int filled, struct extent_state *cached)
1904 struct extent_state *state = NULL;
1905 struct rb_node *node;
1908 spin_lock(&tree->lock);
1909 if (cached && cached->tree && cached->start <= start &&
1910 cached->end > start)
1911 node = &cached->rb_node;
1913 node = tree_search(tree, start);
1914 while (node && start <= end) {
1915 state = rb_entry(node, struct extent_state, rb_node);
1917 if (filled && state->start > start) {
1922 if (state->start > end)
1925 if (state->state & bits) {
1929 } else if (filled) {
1934 if (state->end == (u64)-1)
1937 start = state->end + 1;
1940 node = rb_next(node);
1947 spin_unlock(&tree->lock);
1952 * helper function to set a given page up to date if all the
1953 * extents in the tree for that page are up to date
1955 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1957 u64 start = page_offset(page);
1958 u64 end = start + PAGE_CACHE_SIZE - 1;
1959 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1960 SetPageUptodate(page);
1964 * When IO fails, either with EIO or csum verification fails, we
1965 * try other mirrors that might have a good copy of the data. This
1966 * io_failure_record is used to record state as we go through all the
1967 * mirrors. If another mirror has good data, the page is set up to date
1968 * and things continue. If a good mirror can't be found, the original
1969 * bio end_io callback is called to indicate things have failed.
1971 struct io_failure_record {
1976 unsigned long bio_flags;
1982 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1987 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1989 set_state_private(failure_tree, rec->start, 0);
1990 ret = clear_extent_bits(failure_tree, rec->start,
1991 rec->start + rec->len - 1,
1992 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1996 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1997 rec->start + rec->len - 1,
1998 EXTENT_DAMAGED, GFP_NOFS);
2006 static void repair_io_failure_callback(struct bio *bio, int err)
2008 complete(bio->bi_private);
2012 * this bypasses the standard btrfs submit functions deliberately, as
2013 * the standard behavior is to write all copies in a raid setup. here we only
2014 * want to write the one bad copy. so we do the mapping for ourselves and issue
2015 * submit_bio directly.
2016 * to avoid any synchronization issues, wait for the data after writing, which
2017 * actually prevents the read that triggered the error from finishing.
2018 * currently, there can be no more than two copies of every data bit. thus,
2019 * exactly one rewrite is required.
2021 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
2022 u64 length, u64 logical, struct page *page,
2026 struct btrfs_device *dev;
2027 DECLARE_COMPLETION_ONSTACK(compl);
2030 struct btrfs_bio *bbio = NULL;
2031 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2034 BUG_ON(!mirror_num);
2036 /* we can't repair anything in raid56 yet */
2037 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2040 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2043 bio->bi_private = &compl;
2044 bio->bi_end_io = repair_io_failure_callback;
2046 map_length = length;
2048 ret = btrfs_map_block(fs_info, WRITE, logical,
2049 &map_length, &bbio, mirror_num);
2054 BUG_ON(mirror_num != bbio->mirror_num);
2055 sector = bbio->stripes[mirror_num-1].physical >> 9;
2056 bio->bi_sector = sector;
2057 dev = bbio->stripes[mirror_num-1].dev;
2059 if (!dev || !dev->bdev || !dev->writeable) {
2063 bio->bi_bdev = dev->bdev;
2064 bio_add_page(bio, page, length, start - page_offset(page));
2065 btrfsic_submit_bio(WRITE_SYNC, bio);
2066 wait_for_completion(&compl);
2068 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2069 /* try to remap that extent elsewhere? */
2071 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2075 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
2076 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2077 start, rcu_str_deref(dev->name), sector);
2083 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2086 u64 start = eb->start;
2087 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2090 for (i = 0; i < num_pages; i++) {
2091 struct page *p = extent_buffer_page(eb, i);
2092 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2093 start, p, mirror_num);
2096 start += PAGE_CACHE_SIZE;
2103 * each time an IO finishes, we do a fast check in the IO failure tree
2104 * to see if we need to process or clean up an io_failure_record
2106 static int clean_io_failure(u64 start, struct page *page)
2109 u64 private_failure;
2110 struct io_failure_record *failrec;
2111 struct btrfs_fs_info *fs_info;
2112 struct extent_state *state;
2116 struct inode *inode = page->mapping->host;
2119 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2120 (u64)-1, 1, EXTENT_DIRTY, 0);
2124 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2129 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2130 BUG_ON(!failrec->this_mirror);
2132 if (failrec->in_validation) {
2133 /* there was no real error, just free the record */
2134 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2140 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2141 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2144 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2146 if (state && state->start == failrec->start) {
2147 fs_info = BTRFS_I(inode)->root->fs_info;
2148 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2150 if (num_copies > 1) {
2151 ret = repair_io_failure(fs_info, start, failrec->len,
2152 failrec->logical, page,
2153 failrec->failed_mirror);
2161 ret = free_io_failure(inode, failrec, did_repair);
2167 * this is a generic handler for readpage errors (default
2168 * readpage_io_failed_hook). if other copies exist, read those and write back
2169 * good data to the failed position. does not investigate in remapping the
2170 * failed extent elsewhere, hoping the device will be smart enough to do this as
2174 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2175 u64 start, u64 end, int failed_mirror,
2176 struct extent_state *state)
2178 struct io_failure_record *failrec = NULL;
2180 struct extent_map *em;
2181 struct inode *inode = page->mapping->host;
2182 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2183 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2184 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2191 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2193 ret = get_state_private(failure_tree, start, &private);
2195 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2198 failrec->start = start;
2199 failrec->len = end - start + 1;
2200 failrec->this_mirror = 0;
2201 failrec->bio_flags = 0;
2202 failrec->in_validation = 0;
2204 read_lock(&em_tree->lock);
2205 em = lookup_extent_mapping(em_tree, start, failrec->len);
2207 read_unlock(&em_tree->lock);
2212 if (em->start > start || em->start + em->len < start) {
2213 free_extent_map(em);
2216 read_unlock(&em_tree->lock);
2222 logical = start - em->start;
2223 logical = em->block_start + logical;
2224 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2225 logical = em->block_start;
2226 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2227 extent_set_compress_type(&failrec->bio_flags,
2230 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2231 "len=%llu\n", logical, start, failrec->len);
2232 failrec->logical = logical;
2233 free_extent_map(em);
2235 /* set the bits in the private failure tree */
2236 ret = set_extent_bits(failure_tree, start, end,
2237 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2239 ret = set_state_private(failure_tree, start,
2240 (u64)(unsigned long)failrec);
2241 /* set the bits in the inode's tree */
2243 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2250 failrec = (struct io_failure_record *)(unsigned long)private;
2251 pr_debug("bio_readpage_error: (found) logical=%llu, "
2252 "start=%llu, len=%llu, validation=%d\n",
2253 failrec->logical, failrec->start, failrec->len,
2254 failrec->in_validation);
2256 * when data can be on disk more than twice, add to failrec here
2257 * (e.g. with a list for failed_mirror) to make
2258 * clean_io_failure() clean all those errors at once.
2261 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2262 failrec->logical, failrec->len);
2263 if (num_copies == 1) {
2265 * we only have a single copy of the data, so don't bother with
2266 * all the retry and error correction code that follows. no
2267 * matter what the error is, it is very likely to persist.
2269 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2270 "state=%p, num_copies=%d, next_mirror %d, "
2271 "failed_mirror %d\n", state, num_copies,
2272 failrec->this_mirror, failed_mirror);
2273 free_io_failure(inode, failrec, 0);
2278 spin_lock(&tree->lock);
2279 state = find_first_extent_bit_state(tree, failrec->start,
2281 if (state && state->start != failrec->start)
2283 spin_unlock(&tree->lock);
2287 * there are two premises:
2288 * a) deliver good data to the caller
2289 * b) correct the bad sectors on disk
2291 if (failed_bio->bi_vcnt > 1) {
2293 * to fulfill b), we need to know the exact failing sectors, as
2294 * we don't want to rewrite any more than the failed ones. thus,
2295 * we need separate read requests for the failed bio
2297 * if the following BUG_ON triggers, our validation request got
2298 * merged. we need separate requests for our algorithm to work.
2300 BUG_ON(failrec->in_validation);
2301 failrec->in_validation = 1;
2302 failrec->this_mirror = failed_mirror;
2303 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2306 * we're ready to fulfill a) and b) alongside. get a good copy
2307 * of the failed sector and if we succeed, we have setup
2308 * everything for repair_io_failure to do the rest for us.
2310 if (failrec->in_validation) {
2311 BUG_ON(failrec->this_mirror != failed_mirror);
2312 failrec->in_validation = 0;
2313 failrec->this_mirror = 0;
2315 failrec->failed_mirror = failed_mirror;
2316 failrec->this_mirror++;
2317 if (failrec->this_mirror == failed_mirror)
2318 failrec->this_mirror++;
2319 read_mode = READ_SYNC;
2322 if (!state || failrec->this_mirror > num_copies) {
2323 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2324 "next_mirror %d, failed_mirror %d\n", state,
2325 num_copies, failrec->this_mirror, failed_mirror);
2326 free_io_failure(inode, failrec, 0);
2330 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2332 free_io_failure(inode, failrec, 0);
2335 bio->bi_private = state;
2336 bio->bi_end_io = failed_bio->bi_end_io;
2337 bio->bi_sector = failrec->logical >> 9;
2338 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2341 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2343 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2344 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2345 failrec->this_mirror, num_copies, failrec->in_validation);
2347 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2348 failrec->this_mirror,
2349 failrec->bio_flags, 0);
2353 /* lots and lots of room for performance fixes in the end_bio funcs */
2355 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2357 int uptodate = (err == 0);
2358 struct extent_io_tree *tree;
2361 tree = &BTRFS_I(page->mapping->host)->io_tree;
2363 if (tree->ops && tree->ops->writepage_end_io_hook) {
2364 ret = tree->ops->writepage_end_io_hook(page, start,
2365 end, NULL, uptodate);
2371 ClearPageUptodate(page);
2378 * after a writepage IO is done, we need to:
2379 * clear the uptodate bits on error
2380 * clear the writeback bits in the extent tree for this IO
2381 * end_page_writeback if the page has no more pending IO
2383 * Scheduling is not allowed, so the extent state tree is expected
2384 * to have one and only one object corresponding to this IO.
2386 static void end_bio_extent_writepage(struct bio *bio, int err)
2388 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2389 struct extent_io_tree *tree;
2394 struct page *page = bvec->bv_page;
2395 tree = &BTRFS_I(page->mapping->host)->io_tree;
2397 /* We always issue full-page reads, but if some block
2398 * in a page fails to read, blk_update_request() will
2399 * advance bv_offset and adjust bv_len to compensate.
2400 * Print a warning for nonzero offsets, and an error
2401 * if they don't add up to a full page. */
2402 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2403 printk("%s page write in btrfs with offset %u and length %u\n",
2404 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2405 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2406 bvec->bv_offset, bvec->bv_len);
2408 start = page_offset(page);
2409 end = start + bvec->bv_offset + bvec->bv_len - 1;
2411 if (--bvec >= bio->bi_io_vec)
2412 prefetchw(&bvec->bv_page->flags);
2414 if (end_extent_writepage(page, err, start, end))
2417 end_page_writeback(page);
2418 } while (bvec >= bio->bi_io_vec);
2424 * after a readpage IO is done, we need to:
2425 * clear the uptodate bits on error
2426 * set the uptodate bits if things worked
2427 * set the page up to date if all extents in the tree are uptodate
2428 * clear the lock bit in the extent tree
2429 * unlock the page if there are no other extents locked for it
2431 * Scheduling is not allowed, so the extent state tree is expected
2432 * to have one and only one object corresponding to this IO.
2434 static void end_bio_extent_readpage(struct bio *bio, int err)
2436 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2437 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2438 struct bio_vec *bvec = bio->bi_io_vec;
2439 struct extent_io_tree *tree;
2449 struct page *page = bvec->bv_page;
2450 struct extent_state *cached = NULL;
2451 struct extent_state *state;
2452 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2454 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2455 "mirror=%lu\n", (u64)bio->bi_sector, err,
2456 io_bio->mirror_num);
2457 tree = &BTRFS_I(page->mapping->host)->io_tree;
2459 /* We always issue full-page reads, but if some block
2460 * in a page fails to read, blk_update_request() will
2461 * advance bv_offset and adjust bv_len to compensate.
2462 * Print a warning for nonzero offsets, and an error
2463 * if they don't add up to a full page. */
2464 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2465 printk("%s page read in btrfs with offset %u and length %u\n",
2466 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2467 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2468 bvec->bv_offset, bvec->bv_len);
2470 start = page_offset(page);
2471 end = start + bvec->bv_offset + bvec->bv_len - 1;
2473 if (++bvec <= bvec_end)
2474 prefetchw(&bvec->bv_page->flags);
2476 spin_lock(&tree->lock);
2477 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2478 if (state && state->start == start) {
2480 * take a reference on the state, unlock will drop
2483 cache_state(state, &cached);
2485 spin_unlock(&tree->lock);
2487 mirror = io_bio->mirror_num;
2488 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2489 ret = tree->ops->readpage_end_io_hook(page, start, end,
2494 clean_io_failure(start, page);
2497 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2498 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2500 test_bit(BIO_UPTODATE, &bio->bi_flags))
2502 } else if (!uptodate) {
2504 * The generic bio_readpage_error handles errors the
2505 * following way: If possible, new read requests are
2506 * created and submitted and will end up in
2507 * end_bio_extent_readpage as well (if we're lucky, not
2508 * in the !uptodate case). In that case it returns 0 and
2509 * we just go on with the next page in our bio. If it
2510 * can't handle the error it will return -EIO and we
2511 * remain responsible for that page.
2513 ret = bio_readpage_error(bio, page, start, end, mirror, NULL);
2516 test_bit(BIO_UPTODATE, &bio->bi_flags);
2519 uncache_state(&cached);
2524 if (uptodate && tree->track_uptodate) {
2525 set_extent_uptodate(tree, start, end, &cached,
2528 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2531 SetPageUptodate(page);
2533 ClearPageUptodate(page);
2537 } while (bvec <= bvec_end);
2543 * this allocates from the btrfs_bioset. We're returning a bio right now
2544 * but you can call btrfs_io_bio for the appropriate container_of magic
2547 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2552 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2554 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2555 while (!bio && (nr_vecs /= 2)) {
2556 bio = bio_alloc_bioset(gfp_flags,
2557 nr_vecs, btrfs_bioset);
2563 bio->bi_bdev = bdev;
2564 bio->bi_sector = first_sector;
2569 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2571 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2575 /* this also allocates from the btrfs_bioset */
2576 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2578 return bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2582 static int __must_check submit_one_bio(int rw, struct bio *bio,
2583 int mirror_num, unsigned long bio_flags)
2586 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2587 struct page *page = bvec->bv_page;
2588 struct extent_io_tree *tree = bio->bi_private;
2591 start = page_offset(page) + bvec->bv_offset;
2593 bio->bi_private = NULL;
2597 if (tree->ops && tree->ops->submit_bio_hook)
2598 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2599 mirror_num, bio_flags, start);
2601 btrfsic_submit_bio(rw, bio);
2603 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2609 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2610 unsigned long offset, size_t size, struct bio *bio,
2611 unsigned long bio_flags)
2614 if (tree->ops && tree->ops->merge_bio_hook)
2615 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2622 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2623 struct page *page, sector_t sector,
2624 size_t size, unsigned long offset,
2625 struct block_device *bdev,
2626 struct bio **bio_ret,
2627 unsigned long max_pages,
2628 bio_end_io_t end_io_func,
2630 unsigned long prev_bio_flags,
2631 unsigned long bio_flags)
2637 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2638 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2639 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2641 if (bio_ret && *bio_ret) {
2644 contig = bio->bi_sector == sector;
2646 contig = bio_end_sector(bio) == sector;
2648 if (prev_bio_flags != bio_flags || !contig ||
2649 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2650 bio_add_page(bio, page, page_size, offset) < page_size) {
2651 ret = submit_one_bio(rw, bio, mirror_num,
2660 if (this_compressed)
2663 nr = bio_get_nr_vecs(bdev);
2665 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2669 bio_add_page(bio, page, page_size, offset);
2670 bio->bi_end_io = end_io_func;
2671 bio->bi_private = tree;
2676 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2681 static void attach_extent_buffer_page(struct extent_buffer *eb,
2684 if (!PagePrivate(page)) {
2685 SetPagePrivate(page);
2686 page_cache_get(page);
2687 set_page_private(page, (unsigned long)eb);
2689 WARN_ON(page->private != (unsigned long)eb);
2693 void set_page_extent_mapped(struct page *page)
2695 if (!PagePrivate(page)) {
2696 SetPagePrivate(page);
2697 page_cache_get(page);
2698 set_page_private(page, EXTENT_PAGE_PRIVATE);
2703 * basic readpage implementation. Locked extent state structs are inserted
2704 * into the tree that are removed when the IO is done (by the end_io
2706 * XXX JDM: This needs looking at to ensure proper page locking
2708 static int __extent_read_full_page(struct extent_io_tree *tree,
2710 get_extent_t *get_extent,
2711 struct bio **bio, int mirror_num,
2712 unsigned long *bio_flags, int rw)
2714 struct inode *inode = page->mapping->host;
2715 u64 start = page_offset(page);
2716 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2720 u64 last_byte = i_size_read(inode);
2724 struct extent_map *em;
2725 struct block_device *bdev;
2726 struct btrfs_ordered_extent *ordered;
2729 size_t pg_offset = 0;
2731 size_t disk_io_size;
2732 size_t blocksize = inode->i_sb->s_blocksize;
2733 unsigned long this_bio_flag = 0;
2735 set_page_extent_mapped(page);
2737 if (!PageUptodate(page)) {
2738 if (cleancache_get_page(page) == 0) {
2739 BUG_ON(blocksize != PAGE_SIZE);
2746 lock_extent(tree, start, end);
2747 ordered = btrfs_lookup_ordered_extent(inode, start);
2750 unlock_extent(tree, start, end);
2751 btrfs_start_ordered_extent(inode, ordered, 1);
2752 btrfs_put_ordered_extent(ordered);
2755 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2757 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2760 iosize = PAGE_CACHE_SIZE - zero_offset;
2761 userpage = kmap_atomic(page);
2762 memset(userpage + zero_offset, 0, iosize);
2763 flush_dcache_page(page);
2764 kunmap_atomic(userpage);
2767 while (cur <= end) {
2768 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2770 if (cur >= last_byte) {
2772 struct extent_state *cached = NULL;
2774 iosize = PAGE_CACHE_SIZE - pg_offset;
2775 userpage = kmap_atomic(page);
2776 memset(userpage + pg_offset, 0, iosize);
2777 flush_dcache_page(page);
2778 kunmap_atomic(userpage);
2779 set_extent_uptodate(tree, cur, cur + iosize - 1,
2781 unlock_extent_cached(tree, cur, cur + iosize - 1,
2785 em = get_extent(inode, page, pg_offset, cur,
2787 if (IS_ERR_OR_NULL(em)) {
2789 unlock_extent(tree, cur, end);
2792 extent_offset = cur - em->start;
2793 BUG_ON(extent_map_end(em) <= cur);
2796 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2797 this_bio_flag = EXTENT_BIO_COMPRESSED;
2798 extent_set_compress_type(&this_bio_flag,
2802 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2803 cur_end = min(extent_map_end(em) - 1, end);
2804 iosize = ALIGN(iosize, blocksize);
2805 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2806 disk_io_size = em->block_len;
2807 sector = em->block_start >> 9;
2809 sector = (em->block_start + extent_offset) >> 9;
2810 disk_io_size = iosize;
2813 block_start = em->block_start;
2814 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2815 block_start = EXTENT_MAP_HOLE;
2816 free_extent_map(em);
2819 /* we've found a hole, just zero and go on */
2820 if (block_start == EXTENT_MAP_HOLE) {
2822 struct extent_state *cached = NULL;
2824 userpage = kmap_atomic(page);
2825 memset(userpage + pg_offset, 0, iosize);
2826 flush_dcache_page(page);
2827 kunmap_atomic(userpage);
2829 set_extent_uptodate(tree, cur, cur + iosize - 1,
2831 unlock_extent_cached(tree, cur, cur + iosize - 1,
2834 pg_offset += iosize;
2837 /* the get_extent function already copied into the page */
2838 if (test_range_bit(tree, cur, cur_end,
2839 EXTENT_UPTODATE, 1, NULL)) {
2840 check_page_uptodate(tree, page);
2841 unlock_extent(tree, cur, cur + iosize - 1);
2843 pg_offset += iosize;
2846 /* we have an inline extent but it didn't get marked up
2847 * to date. Error out
2849 if (block_start == EXTENT_MAP_INLINE) {
2851 unlock_extent(tree, cur, cur + iosize - 1);
2853 pg_offset += iosize;
2858 ret = submit_extent_page(rw, tree, page,
2859 sector, disk_io_size, pg_offset,
2861 end_bio_extent_readpage, mirror_num,
2866 *bio_flags = this_bio_flag;
2869 unlock_extent(tree, cur, cur + iosize - 1);
2872 pg_offset += iosize;
2876 if (!PageError(page))
2877 SetPageUptodate(page);
2883 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2884 get_extent_t *get_extent, int mirror_num)
2886 struct bio *bio = NULL;
2887 unsigned long bio_flags = 0;
2890 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2893 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2897 static noinline void update_nr_written(struct page *page,
2898 struct writeback_control *wbc,
2899 unsigned long nr_written)
2901 wbc->nr_to_write -= nr_written;
2902 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2903 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2904 page->mapping->writeback_index = page->index + nr_written;
2908 * the writepage semantics are similar to regular writepage. extent
2909 * records are inserted to lock ranges in the tree, and as dirty areas
2910 * are found, they are marked writeback. Then the lock bits are removed
2911 * and the end_io handler clears the writeback ranges
2913 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2916 struct inode *inode = page->mapping->host;
2917 struct extent_page_data *epd = data;
2918 struct extent_io_tree *tree = epd->tree;
2919 u64 start = page_offset(page);
2921 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2925 u64 last_byte = i_size_read(inode);
2929 struct extent_state *cached_state = NULL;
2930 struct extent_map *em;
2931 struct block_device *bdev;
2934 size_t pg_offset = 0;
2936 loff_t i_size = i_size_read(inode);
2937 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2943 unsigned long nr_written = 0;
2944 bool fill_delalloc = true;
2946 if (wbc->sync_mode == WB_SYNC_ALL)
2947 write_flags = WRITE_SYNC;
2949 write_flags = WRITE;
2951 trace___extent_writepage(page, inode, wbc);
2953 WARN_ON(!PageLocked(page));
2955 ClearPageError(page);
2957 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2958 if (page->index > end_index ||
2959 (page->index == end_index && !pg_offset)) {
2960 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
2965 if (page->index == end_index) {
2968 userpage = kmap_atomic(page);
2969 memset(userpage + pg_offset, 0,
2970 PAGE_CACHE_SIZE - pg_offset);
2971 kunmap_atomic(userpage);
2972 flush_dcache_page(page);
2976 set_page_extent_mapped(page);
2978 if (!tree->ops || !tree->ops->fill_delalloc)
2979 fill_delalloc = false;
2981 delalloc_start = start;
2984 if (!epd->extent_locked && fill_delalloc) {
2985 u64 delalloc_to_write = 0;
2987 * make sure the wbc mapping index is at least updated
2990 update_nr_written(page, wbc, 0);
2992 while (delalloc_end < page_end) {
2993 nr_delalloc = find_lock_delalloc_range(inode, tree,
2998 if (nr_delalloc == 0) {
2999 delalloc_start = delalloc_end + 1;
3002 ret = tree->ops->fill_delalloc(inode, page,
3007 /* File system has been set read-only */
3013 * delalloc_end is already one less than the total
3014 * length, so we don't subtract one from
3017 delalloc_to_write += (delalloc_end - delalloc_start +
3020 delalloc_start = delalloc_end + 1;
3022 if (wbc->nr_to_write < delalloc_to_write) {
3025 if (delalloc_to_write < thresh * 2)
3026 thresh = delalloc_to_write;
3027 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3031 /* did the fill delalloc function already unlock and start
3037 * we've unlocked the page, so we can't update
3038 * the mapping's writeback index, just update
3041 wbc->nr_to_write -= nr_written;
3045 if (tree->ops && tree->ops->writepage_start_hook) {
3046 ret = tree->ops->writepage_start_hook(page, start,
3049 /* Fixup worker will requeue */
3051 wbc->pages_skipped++;
3053 redirty_page_for_writepage(wbc, page);
3054 update_nr_written(page, wbc, nr_written);
3062 * we don't want to touch the inode after unlocking the page,
3063 * so we update the mapping writeback index now
3065 update_nr_written(page, wbc, nr_written + 1);
3068 if (last_byte <= start) {
3069 if (tree->ops && tree->ops->writepage_end_io_hook)
3070 tree->ops->writepage_end_io_hook(page, start,
3075 blocksize = inode->i_sb->s_blocksize;
3077 while (cur <= end) {
3078 if (cur >= last_byte) {
3079 if (tree->ops && tree->ops->writepage_end_io_hook)
3080 tree->ops->writepage_end_io_hook(page, cur,
3084 em = epd->get_extent(inode, page, pg_offset, cur,
3086 if (IS_ERR_OR_NULL(em)) {
3091 extent_offset = cur - em->start;
3092 BUG_ON(extent_map_end(em) <= cur);
3094 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3095 iosize = ALIGN(iosize, blocksize);
3096 sector = (em->block_start + extent_offset) >> 9;
3098 block_start = em->block_start;
3099 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3100 free_extent_map(em);
3104 * compressed and inline extents are written through other
3107 if (compressed || block_start == EXTENT_MAP_HOLE ||
3108 block_start == EXTENT_MAP_INLINE) {
3110 * end_io notification does not happen here for
3111 * compressed extents
3113 if (!compressed && tree->ops &&
3114 tree->ops->writepage_end_io_hook)
3115 tree->ops->writepage_end_io_hook(page, cur,
3118 else if (compressed) {
3119 /* we don't want to end_page_writeback on
3120 * a compressed extent. this happens
3127 pg_offset += iosize;
3130 /* leave this out until we have a page_mkwrite call */
3131 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3132 EXTENT_DIRTY, 0, NULL)) {
3134 pg_offset += iosize;
3138 if (tree->ops && tree->ops->writepage_io_hook) {
3139 ret = tree->ops->writepage_io_hook(page, cur,
3147 unsigned long max_nr = end_index + 1;
3149 set_range_writeback(tree, cur, cur + iosize - 1);
3150 if (!PageWriteback(page)) {
3151 printk(KERN_ERR "btrfs warning page %lu not "
3152 "writeback, cur %llu end %llu\n",
3153 page->index, (unsigned long long)cur,
3154 (unsigned long long)end);
3157 ret = submit_extent_page(write_flags, tree, page,
3158 sector, iosize, pg_offset,
3159 bdev, &epd->bio, max_nr,
3160 end_bio_extent_writepage,
3166 pg_offset += iosize;
3171 /* make sure the mapping tag for page dirty gets cleared */
3172 set_page_writeback(page);
3173 end_page_writeback(page);
3179 /* drop our reference on any cached states */
3180 free_extent_state(cached_state);
3184 static int eb_wait(void *word)
3190 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3192 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3193 TASK_UNINTERRUPTIBLE);
3196 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3197 struct btrfs_fs_info *fs_info,
3198 struct extent_page_data *epd)
3200 unsigned long i, num_pages;
3204 if (!btrfs_try_tree_write_lock(eb)) {
3206 flush_write_bio(epd);
3207 btrfs_tree_lock(eb);
3210 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3211 btrfs_tree_unlock(eb);
3215 flush_write_bio(epd);
3219 wait_on_extent_buffer_writeback(eb);
3220 btrfs_tree_lock(eb);
3221 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3223 btrfs_tree_unlock(eb);
3228 * We need to do this to prevent races in people who check if the eb is
3229 * under IO since we can end up having no IO bits set for a short period
3232 spin_lock(&eb->refs_lock);
3233 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3234 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3235 spin_unlock(&eb->refs_lock);
3236 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3237 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3239 fs_info->dirty_metadata_batch);
3242 spin_unlock(&eb->refs_lock);
3245 btrfs_tree_unlock(eb);
3250 num_pages = num_extent_pages(eb->start, eb->len);
3251 for (i = 0; i < num_pages; i++) {
3252 struct page *p = extent_buffer_page(eb, i);
3254 if (!trylock_page(p)) {
3256 flush_write_bio(epd);
3266 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3268 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3269 smp_mb__after_clear_bit();
3270 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3273 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3275 int uptodate = err == 0;
3276 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3277 struct extent_buffer *eb;
3281 struct page *page = bvec->bv_page;
3284 eb = (struct extent_buffer *)page->private;
3286 done = atomic_dec_and_test(&eb->io_pages);
3288 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3289 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3290 ClearPageUptodate(page);
3294 end_page_writeback(page);
3299 end_extent_buffer_writeback(eb);
3300 } while (bvec >= bio->bi_io_vec);
3306 static int write_one_eb(struct extent_buffer *eb,
3307 struct btrfs_fs_info *fs_info,
3308 struct writeback_control *wbc,
3309 struct extent_page_data *epd)
3311 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3312 u64 offset = eb->start;
3313 unsigned long i, num_pages;
3314 unsigned long bio_flags = 0;
3315 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3318 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3319 num_pages = num_extent_pages(eb->start, eb->len);
3320 atomic_set(&eb->io_pages, num_pages);
3321 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3322 bio_flags = EXTENT_BIO_TREE_LOG;
3324 for (i = 0; i < num_pages; i++) {
3325 struct page *p = extent_buffer_page(eb, i);
3327 clear_page_dirty_for_io(p);
3328 set_page_writeback(p);
3329 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3330 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3331 -1, end_bio_extent_buffer_writepage,
3332 0, epd->bio_flags, bio_flags);
3333 epd->bio_flags = bio_flags;
3335 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3337 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3338 end_extent_buffer_writeback(eb);
3342 offset += PAGE_CACHE_SIZE;
3343 update_nr_written(p, wbc, 1);
3347 if (unlikely(ret)) {
3348 for (; i < num_pages; i++) {
3349 struct page *p = extent_buffer_page(eb, i);
3357 int btree_write_cache_pages(struct address_space *mapping,
3358 struct writeback_control *wbc)
3360 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3361 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3362 struct extent_buffer *eb, *prev_eb = NULL;
3363 struct extent_page_data epd = {
3367 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3372 int nr_to_write_done = 0;
3373 struct pagevec pvec;
3376 pgoff_t end; /* Inclusive */
3380 pagevec_init(&pvec, 0);
3381 if (wbc->range_cyclic) {
3382 index = mapping->writeback_index; /* Start from prev offset */
3385 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3386 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3389 if (wbc->sync_mode == WB_SYNC_ALL)
3390 tag = PAGECACHE_TAG_TOWRITE;
3392 tag = PAGECACHE_TAG_DIRTY;
3394 if (wbc->sync_mode == WB_SYNC_ALL)
3395 tag_pages_for_writeback(mapping, index, end);
3396 while (!done && !nr_to_write_done && (index <= end) &&
3397 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3398 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3402 for (i = 0; i < nr_pages; i++) {
3403 struct page *page = pvec.pages[i];
3405 if (!PagePrivate(page))
3408 if (!wbc->range_cyclic && page->index > end) {
3413 spin_lock(&mapping->private_lock);
3414 if (!PagePrivate(page)) {
3415 spin_unlock(&mapping->private_lock);
3419 eb = (struct extent_buffer *)page->private;
3422 * Shouldn't happen and normally this would be a BUG_ON
3423 * but no sense in crashing the users box for something
3424 * we can survive anyway.
3427 spin_unlock(&mapping->private_lock);
3432 if (eb == prev_eb) {
3433 spin_unlock(&mapping->private_lock);
3437 ret = atomic_inc_not_zero(&eb->refs);
3438 spin_unlock(&mapping->private_lock);
3443 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3445 free_extent_buffer(eb);
3449 ret = write_one_eb(eb, fs_info, wbc, &epd);
3452 free_extent_buffer(eb);
3455 free_extent_buffer(eb);
3458 * the filesystem may choose to bump up nr_to_write.
3459 * We have to make sure to honor the new nr_to_write
3462 nr_to_write_done = wbc->nr_to_write <= 0;
3464 pagevec_release(&pvec);
3467 if (!scanned && !done) {
3469 * We hit the last page and there is more work to be done: wrap
3470 * back to the start of the file
3476 flush_write_bio(&epd);
3481 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3482 * @mapping: address space structure to write
3483 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3484 * @writepage: function called for each page
3485 * @data: data passed to writepage function
3487 * If a page is already under I/O, write_cache_pages() skips it, even
3488 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3489 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3490 * and msync() need to guarantee that all the data which was dirty at the time
3491 * the call was made get new I/O started against them. If wbc->sync_mode is
3492 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3493 * existing IO to complete.
3495 static int extent_write_cache_pages(struct extent_io_tree *tree,
3496 struct address_space *mapping,
3497 struct writeback_control *wbc,
3498 writepage_t writepage, void *data,
3499 void (*flush_fn)(void *))
3501 struct inode *inode = mapping->host;
3504 int nr_to_write_done = 0;
3505 struct pagevec pvec;
3508 pgoff_t end; /* Inclusive */
3513 * We have to hold onto the inode so that ordered extents can do their
3514 * work when the IO finishes. The alternative to this is failing to add
3515 * an ordered extent if the igrab() fails there and that is a huge pain
3516 * to deal with, so instead just hold onto the inode throughout the
3517 * writepages operation. If it fails here we are freeing up the inode
3518 * anyway and we'd rather not waste our time writing out stuff that is
3519 * going to be truncated anyway.
3524 pagevec_init(&pvec, 0);
3525 if (wbc->range_cyclic) {
3526 index = mapping->writeback_index; /* Start from prev offset */
3529 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3530 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3533 if (wbc->sync_mode == WB_SYNC_ALL)
3534 tag = PAGECACHE_TAG_TOWRITE;
3536 tag = PAGECACHE_TAG_DIRTY;
3538 if (wbc->sync_mode == WB_SYNC_ALL)
3539 tag_pages_for_writeback(mapping, index, end);
3540 while (!done && !nr_to_write_done && (index <= end) &&
3541 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3542 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3546 for (i = 0; i < nr_pages; i++) {
3547 struct page *page = pvec.pages[i];
3550 * At this point we hold neither mapping->tree_lock nor
3551 * lock on the page itself: the page may be truncated or
3552 * invalidated (changing page->mapping to NULL), or even
3553 * swizzled back from swapper_space to tmpfs file
3556 if (!trylock_page(page)) {
3561 if (unlikely(page->mapping != mapping)) {
3566 if (!wbc->range_cyclic && page->index > end) {
3572 if (wbc->sync_mode != WB_SYNC_NONE) {
3573 if (PageWriteback(page))
3575 wait_on_page_writeback(page);
3578 if (PageWriteback(page) ||
3579 !clear_page_dirty_for_io(page)) {
3584 ret = (*writepage)(page, wbc, data);
3586 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3594 * the filesystem may choose to bump up nr_to_write.
3595 * We have to make sure to honor the new nr_to_write
3598 nr_to_write_done = wbc->nr_to_write <= 0;
3600 pagevec_release(&pvec);
3603 if (!scanned && !done) {
3605 * We hit the last page and there is more work to be done: wrap
3606 * back to the start of the file
3612 btrfs_add_delayed_iput(inode);
3616 static void flush_epd_write_bio(struct extent_page_data *epd)
3625 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3626 BUG_ON(ret < 0); /* -ENOMEM */
3631 static noinline void flush_write_bio(void *data)
3633 struct extent_page_data *epd = data;
3634 flush_epd_write_bio(epd);
3637 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3638 get_extent_t *get_extent,
3639 struct writeback_control *wbc)
3642 struct extent_page_data epd = {
3645 .get_extent = get_extent,
3647 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3651 ret = __extent_writepage(page, wbc, &epd);
3653 flush_epd_write_bio(&epd);
3657 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3658 u64 start, u64 end, get_extent_t *get_extent,
3662 struct address_space *mapping = inode->i_mapping;
3664 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3667 struct extent_page_data epd = {
3670 .get_extent = get_extent,
3672 .sync_io = mode == WB_SYNC_ALL,
3675 struct writeback_control wbc_writepages = {
3677 .nr_to_write = nr_pages * 2,
3678 .range_start = start,
3679 .range_end = end + 1,
3682 while (start <= end) {
3683 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3684 if (clear_page_dirty_for_io(page))
3685 ret = __extent_writepage(page, &wbc_writepages, &epd);
3687 if (tree->ops && tree->ops->writepage_end_io_hook)
3688 tree->ops->writepage_end_io_hook(page, start,
3689 start + PAGE_CACHE_SIZE - 1,
3693 page_cache_release(page);
3694 start += PAGE_CACHE_SIZE;
3697 flush_epd_write_bio(&epd);
3701 int extent_writepages(struct extent_io_tree *tree,
3702 struct address_space *mapping,
3703 get_extent_t *get_extent,
3704 struct writeback_control *wbc)
3707 struct extent_page_data epd = {
3710 .get_extent = get_extent,
3712 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3716 ret = extent_write_cache_pages(tree, mapping, wbc,
3717 __extent_writepage, &epd,
3719 flush_epd_write_bio(&epd);
3723 int extent_readpages(struct extent_io_tree *tree,
3724 struct address_space *mapping,
3725 struct list_head *pages, unsigned nr_pages,
3726 get_extent_t get_extent)
3728 struct bio *bio = NULL;
3730 unsigned long bio_flags = 0;
3731 struct page *pagepool[16];
3736 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3737 page = list_entry(pages->prev, struct page, lru);
3739 prefetchw(&page->flags);
3740 list_del(&page->lru);
3741 if (add_to_page_cache_lru(page, mapping,
3742 page->index, GFP_NOFS)) {
3743 page_cache_release(page);
3747 pagepool[nr++] = page;
3748 if (nr < ARRAY_SIZE(pagepool))
3750 for (i = 0; i < nr; i++) {
3751 __extent_read_full_page(tree, pagepool[i], get_extent,
3752 &bio, 0, &bio_flags, READ);
3753 page_cache_release(pagepool[i]);
3757 for (i = 0; i < nr; i++) {
3758 __extent_read_full_page(tree, pagepool[i], get_extent,
3759 &bio, 0, &bio_flags, READ);
3760 page_cache_release(pagepool[i]);
3763 BUG_ON(!list_empty(pages));
3765 return submit_one_bio(READ, bio, 0, bio_flags);
3770 * basic invalidatepage code, this waits on any locked or writeback
3771 * ranges corresponding to the page, and then deletes any extent state
3772 * records from the tree
3774 int extent_invalidatepage(struct extent_io_tree *tree,
3775 struct page *page, unsigned long offset)
3777 struct extent_state *cached_state = NULL;
3778 u64 start = page_offset(page);
3779 u64 end = start + PAGE_CACHE_SIZE - 1;
3780 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3782 start += ALIGN(offset, blocksize);
3786 lock_extent_bits(tree, start, end, 0, &cached_state);
3787 wait_on_page_writeback(page);
3788 clear_extent_bit(tree, start, end,
3789 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3790 EXTENT_DO_ACCOUNTING,
3791 1, 1, &cached_state, GFP_NOFS);
3796 * a helper for releasepage, this tests for areas of the page that
3797 * are locked or under IO and drops the related state bits if it is safe
3800 static int try_release_extent_state(struct extent_map_tree *map,
3801 struct extent_io_tree *tree,
3802 struct page *page, gfp_t mask)
3804 u64 start = page_offset(page);
3805 u64 end = start + PAGE_CACHE_SIZE - 1;
3808 if (test_range_bit(tree, start, end,
3809 EXTENT_IOBITS, 0, NULL))
3812 if ((mask & GFP_NOFS) == GFP_NOFS)
3815 * at this point we can safely clear everything except the
3816 * locked bit and the nodatasum bit
3818 ret = clear_extent_bit(tree, start, end,
3819 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3822 /* if clear_extent_bit failed for enomem reasons,
3823 * we can't allow the release to continue.
3834 * a helper for releasepage. As long as there are no locked extents
3835 * in the range corresponding to the page, both state records and extent
3836 * map records are removed
3838 int try_release_extent_mapping(struct extent_map_tree *map,
3839 struct extent_io_tree *tree, struct page *page,
3842 struct extent_map *em;
3843 u64 start = page_offset(page);
3844 u64 end = start + PAGE_CACHE_SIZE - 1;
3846 if ((mask & __GFP_WAIT) &&
3847 page->mapping->host->i_size > 16 * 1024 * 1024) {
3849 while (start <= end) {
3850 len = end - start + 1;
3851 write_lock(&map->lock);
3852 em = lookup_extent_mapping(map, start, len);
3854 write_unlock(&map->lock);
3857 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3858 em->start != start) {
3859 write_unlock(&map->lock);
3860 free_extent_map(em);
3863 if (!test_range_bit(tree, em->start,
3864 extent_map_end(em) - 1,
3865 EXTENT_LOCKED | EXTENT_WRITEBACK,
3867 remove_extent_mapping(map, em);
3868 /* once for the rb tree */
3869 free_extent_map(em);
3871 start = extent_map_end(em);
3872 write_unlock(&map->lock);
3875 free_extent_map(em);
3878 return try_release_extent_state(map, tree, page, mask);
3882 * helper function for fiemap, which doesn't want to see any holes.
3883 * This maps until we find something past 'last'
3885 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3888 get_extent_t *get_extent)
3890 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3891 struct extent_map *em;
3898 len = last - offset;
3901 len = ALIGN(len, sectorsize);
3902 em = get_extent(inode, NULL, 0, offset, len, 0);
3903 if (IS_ERR_OR_NULL(em))
3906 /* if this isn't a hole return it */
3907 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3908 em->block_start != EXTENT_MAP_HOLE) {
3912 /* this is a hole, advance to the next extent */
3913 offset = extent_map_end(em);
3914 free_extent_map(em);
3921 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3922 __u64 start, __u64 len, get_extent_t *get_extent)
3926 u64 max = start + len;
3930 u64 last_for_get_extent = 0;
3932 u64 isize = i_size_read(inode);
3933 struct btrfs_key found_key;
3934 struct extent_map *em = NULL;
3935 struct extent_state *cached_state = NULL;
3936 struct btrfs_path *path;
3937 struct btrfs_file_extent_item *item;
3942 unsigned long emflags;
3947 path = btrfs_alloc_path();
3950 path->leave_spinning = 1;
3952 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3953 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3956 * lookup the last file extent. We're not using i_size here
3957 * because there might be preallocation past i_size
3959 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3960 path, btrfs_ino(inode), -1, 0);
3962 btrfs_free_path(path);
3967 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3968 struct btrfs_file_extent_item);
3969 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3970 found_type = btrfs_key_type(&found_key);
3972 /* No extents, but there might be delalloc bits */
3973 if (found_key.objectid != btrfs_ino(inode) ||
3974 found_type != BTRFS_EXTENT_DATA_KEY) {
3975 /* have to trust i_size as the end */
3977 last_for_get_extent = isize;
3980 * remember the start of the last extent. There are a
3981 * bunch of different factors that go into the length of the
3982 * extent, so its much less complex to remember where it started
3984 last = found_key.offset;
3985 last_for_get_extent = last + 1;
3987 btrfs_free_path(path);
3990 * we might have some extents allocated but more delalloc past those
3991 * extents. so, we trust isize unless the start of the last extent is
3996 last_for_get_extent = isize;
3999 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4002 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4012 u64 offset_in_extent;
4014 /* break if the extent we found is outside the range */
4015 if (em->start >= max || extent_map_end(em) < off)
4019 * get_extent may return an extent that starts before our
4020 * requested range. We have to make sure the ranges
4021 * we return to fiemap always move forward and don't
4022 * overlap, so adjust the offsets here
4024 em_start = max(em->start, off);
4027 * record the offset from the start of the extent
4028 * for adjusting the disk offset below
4030 offset_in_extent = em_start - em->start;
4031 em_end = extent_map_end(em);
4032 em_len = em_end - em_start;
4033 emflags = em->flags;
4038 * bump off for our next call to get_extent
4040 off = extent_map_end(em);
4044 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4046 flags |= FIEMAP_EXTENT_LAST;
4047 } else if (em->block_start == EXTENT_MAP_INLINE) {
4048 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4049 FIEMAP_EXTENT_NOT_ALIGNED);
4050 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4051 flags |= (FIEMAP_EXTENT_DELALLOC |
4052 FIEMAP_EXTENT_UNKNOWN);
4054 disko = em->block_start + offset_in_extent;
4056 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4057 flags |= FIEMAP_EXTENT_ENCODED;
4059 free_extent_map(em);
4061 if ((em_start >= last) || em_len == (u64)-1 ||
4062 (last == (u64)-1 && isize <= em_end)) {
4063 flags |= FIEMAP_EXTENT_LAST;
4067 /* now scan forward to see if this is really the last extent. */
4068 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4075 flags |= FIEMAP_EXTENT_LAST;
4078 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4084 free_extent_map(em);
4086 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4087 &cached_state, GFP_NOFS);
4091 static void __free_extent_buffer(struct extent_buffer *eb)
4093 btrfs_leak_debug_del(&eb->leak_list);
4094 kmem_cache_free(extent_buffer_cache, eb);
4097 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
4102 struct extent_buffer *eb = NULL;
4104 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4111 rwlock_init(&eb->lock);
4112 atomic_set(&eb->write_locks, 0);
4113 atomic_set(&eb->read_locks, 0);
4114 atomic_set(&eb->blocking_readers, 0);
4115 atomic_set(&eb->blocking_writers, 0);
4116 atomic_set(&eb->spinning_readers, 0);
4117 atomic_set(&eb->spinning_writers, 0);
4118 eb->lock_nested = 0;
4119 init_waitqueue_head(&eb->write_lock_wq);
4120 init_waitqueue_head(&eb->read_lock_wq);
4122 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4124 spin_lock_init(&eb->refs_lock);
4125 atomic_set(&eb->refs, 1);
4126 atomic_set(&eb->io_pages, 0);
4129 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4131 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4132 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4133 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4138 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4142 struct extent_buffer *new;
4143 unsigned long num_pages = num_extent_pages(src->start, src->len);
4145 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4149 for (i = 0; i < num_pages; i++) {
4150 p = alloc_page(GFP_ATOMIC);
4152 attach_extent_buffer_page(new, p);
4153 WARN_ON(PageDirty(p));
4158 copy_extent_buffer(new, src, 0, 0, src->len);
4159 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4160 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4165 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4167 struct extent_buffer *eb;
4168 unsigned long num_pages = num_extent_pages(0, len);
4171 eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4175 for (i = 0; i < num_pages; i++) {
4176 eb->pages[i] = alloc_page(GFP_ATOMIC);
4180 set_extent_buffer_uptodate(eb);
4181 btrfs_set_header_nritems(eb, 0);
4182 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4187 __free_page(eb->pages[i - 1]);
4188 __free_extent_buffer(eb);
4192 static int extent_buffer_under_io(struct extent_buffer *eb)
4194 return (atomic_read(&eb->io_pages) ||
4195 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4196 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4200 * Helper for releasing extent buffer page.
4202 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4203 unsigned long start_idx)
4205 unsigned long index;
4206 unsigned long num_pages;
4208 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4210 BUG_ON(extent_buffer_under_io(eb));
4212 num_pages = num_extent_pages(eb->start, eb->len);
4213 index = start_idx + num_pages;
4214 if (start_idx >= index)
4219 page = extent_buffer_page(eb, index);
4220 if (page && mapped) {
4221 spin_lock(&page->mapping->private_lock);
4223 * We do this since we'll remove the pages after we've
4224 * removed the eb from the radix tree, so we could race
4225 * and have this page now attached to the new eb. So
4226 * only clear page_private if it's still connected to
4229 if (PagePrivate(page) &&
4230 page->private == (unsigned long)eb) {
4231 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4232 BUG_ON(PageDirty(page));
4233 BUG_ON(PageWriteback(page));
4235 * We need to make sure we haven't be attached
4238 ClearPagePrivate(page);
4239 set_page_private(page, 0);
4240 /* One for the page private */
4241 page_cache_release(page);
4243 spin_unlock(&page->mapping->private_lock);
4247 /* One for when we alloced the page */
4248 page_cache_release(page);
4250 } while (index != start_idx);
4254 * Helper for releasing the extent buffer.
4256 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4258 btrfs_release_extent_buffer_page(eb, 0);
4259 __free_extent_buffer(eb);
4262 static void check_buffer_tree_ref(struct extent_buffer *eb)
4265 /* the ref bit is tricky. We have to make sure it is set
4266 * if we have the buffer dirty. Otherwise the
4267 * code to free a buffer can end up dropping a dirty
4270 * Once the ref bit is set, it won't go away while the
4271 * buffer is dirty or in writeback, and it also won't
4272 * go away while we have the reference count on the
4275 * We can't just set the ref bit without bumping the
4276 * ref on the eb because free_extent_buffer might
4277 * see the ref bit and try to clear it. If this happens
4278 * free_extent_buffer might end up dropping our original
4279 * ref by mistake and freeing the page before we are able
4280 * to add one more ref.
4282 * So bump the ref count first, then set the bit. If someone
4283 * beat us to it, drop the ref we added.
4285 refs = atomic_read(&eb->refs);
4286 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4289 spin_lock(&eb->refs_lock);
4290 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4291 atomic_inc(&eb->refs);
4292 spin_unlock(&eb->refs_lock);
4295 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4297 unsigned long num_pages, i;
4299 check_buffer_tree_ref(eb);
4301 num_pages = num_extent_pages(eb->start, eb->len);
4302 for (i = 0; i < num_pages; i++) {
4303 struct page *p = extent_buffer_page(eb, i);
4304 mark_page_accessed(p);
4308 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4309 u64 start, unsigned long len)
4311 unsigned long num_pages = num_extent_pages(start, len);
4313 unsigned long index = start >> PAGE_CACHE_SHIFT;
4314 struct extent_buffer *eb;
4315 struct extent_buffer *exists = NULL;
4317 struct address_space *mapping = tree->mapping;
4322 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4323 if (eb && atomic_inc_not_zero(&eb->refs)) {
4325 mark_extent_buffer_accessed(eb);
4330 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4334 for (i = 0; i < num_pages; i++, index++) {
4335 p = find_or_create_page(mapping, index, GFP_NOFS);
4339 spin_lock(&mapping->private_lock);
4340 if (PagePrivate(p)) {
4342 * We could have already allocated an eb for this page
4343 * and attached one so lets see if we can get a ref on
4344 * the existing eb, and if we can we know it's good and
4345 * we can just return that one, else we know we can just
4346 * overwrite page->private.
4348 exists = (struct extent_buffer *)p->private;
4349 if (atomic_inc_not_zero(&exists->refs)) {
4350 spin_unlock(&mapping->private_lock);
4352 page_cache_release(p);
4353 mark_extent_buffer_accessed(exists);
4358 * Do this so attach doesn't complain and we need to
4359 * drop the ref the old guy had.
4361 ClearPagePrivate(p);
4362 WARN_ON(PageDirty(p));
4363 page_cache_release(p);
4365 attach_extent_buffer_page(eb, p);
4366 spin_unlock(&mapping->private_lock);
4367 WARN_ON(PageDirty(p));
4368 mark_page_accessed(p);
4370 if (!PageUptodate(p))
4374 * see below about how we avoid a nasty race with release page
4375 * and why we unlock later
4379 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4381 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4385 spin_lock(&tree->buffer_lock);
4386 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4387 if (ret == -EEXIST) {
4388 exists = radix_tree_lookup(&tree->buffer,
4389 start >> PAGE_CACHE_SHIFT);
4390 if (!atomic_inc_not_zero(&exists->refs)) {
4391 spin_unlock(&tree->buffer_lock);
4392 radix_tree_preload_end();
4396 spin_unlock(&tree->buffer_lock);
4397 radix_tree_preload_end();
4398 mark_extent_buffer_accessed(exists);
4401 /* add one reference for the tree */
4402 check_buffer_tree_ref(eb);
4403 spin_unlock(&tree->buffer_lock);
4404 radix_tree_preload_end();
4407 * there is a race where release page may have
4408 * tried to find this extent buffer in the radix
4409 * but failed. It will tell the VM it is safe to
4410 * reclaim the, and it will clear the page private bit.
4411 * We must make sure to set the page private bit properly
4412 * after the extent buffer is in the radix tree so
4413 * it doesn't get lost
4415 SetPageChecked(eb->pages[0]);
4416 for (i = 1; i < num_pages; i++) {
4417 p = extent_buffer_page(eb, i);
4418 ClearPageChecked(p);
4421 unlock_page(eb->pages[0]);
4425 for (i = 0; i < num_pages; i++) {
4427 unlock_page(eb->pages[i]);
4430 WARN_ON(!atomic_dec_and_test(&eb->refs));
4431 btrfs_release_extent_buffer(eb);
4435 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4436 u64 start, unsigned long len)
4438 struct extent_buffer *eb;
4441 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4442 if (eb && atomic_inc_not_zero(&eb->refs)) {
4444 mark_extent_buffer_accessed(eb);
4452 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4454 struct extent_buffer *eb =
4455 container_of(head, struct extent_buffer, rcu_head);
4457 __free_extent_buffer(eb);
4460 /* Expects to have eb->eb_lock already held */
4461 static int release_extent_buffer(struct extent_buffer *eb)
4463 WARN_ON(atomic_read(&eb->refs) == 0);
4464 if (atomic_dec_and_test(&eb->refs)) {
4465 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4466 spin_unlock(&eb->refs_lock);
4468 struct extent_io_tree *tree = eb->tree;
4470 spin_unlock(&eb->refs_lock);
4472 spin_lock(&tree->buffer_lock);
4473 radix_tree_delete(&tree->buffer,
4474 eb->start >> PAGE_CACHE_SHIFT);
4475 spin_unlock(&tree->buffer_lock);
4478 /* Should be safe to release our pages at this point */
4479 btrfs_release_extent_buffer_page(eb, 0);
4480 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4483 spin_unlock(&eb->refs_lock);
4488 void free_extent_buffer(struct extent_buffer *eb)
4496 refs = atomic_read(&eb->refs);
4499 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4504 spin_lock(&eb->refs_lock);
4505 if (atomic_read(&eb->refs) == 2 &&
4506 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4507 atomic_dec(&eb->refs);
4509 if (atomic_read(&eb->refs) == 2 &&
4510 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4511 !extent_buffer_under_io(eb) &&
4512 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4513 atomic_dec(&eb->refs);
4516 * I know this is terrible, but it's temporary until we stop tracking
4517 * the uptodate bits and such for the extent buffers.
4519 release_extent_buffer(eb);
4522 void free_extent_buffer_stale(struct extent_buffer *eb)
4527 spin_lock(&eb->refs_lock);
4528 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4530 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4531 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4532 atomic_dec(&eb->refs);
4533 release_extent_buffer(eb);
4536 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4539 unsigned long num_pages;
4542 num_pages = num_extent_pages(eb->start, eb->len);
4544 for (i = 0; i < num_pages; i++) {
4545 page = extent_buffer_page(eb, i);
4546 if (!PageDirty(page))
4550 WARN_ON(!PagePrivate(page));
4552 clear_page_dirty_for_io(page);
4553 spin_lock_irq(&page->mapping->tree_lock);
4554 if (!PageDirty(page)) {
4555 radix_tree_tag_clear(&page->mapping->page_tree,
4557 PAGECACHE_TAG_DIRTY);
4559 spin_unlock_irq(&page->mapping->tree_lock);
4560 ClearPageError(page);
4563 WARN_ON(atomic_read(&eb->refs) == 0);
4566 int set_extent_buffer_dirty(struct extent_buffer *eb)
4569 unsigned long num_pages;
4572 check_buffer_tree_ref(eb);
4574 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4576 num_pages = num_extent_pages(eb->start, eb->len);
4577 WARN_ON(atomic_read(&eb->refs) == 0);
4578 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4580 for (i = 0; i < num_pages; i++)
4581 set_page_dirty(extent_buffer_page(eb, i));
4585 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4589 unsigned long num_pages;
4591 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4592 num_pages = num_extent_pages(eb->start, eb->len);
4593 for (i = 0; i < num_pages; i++) {
4594 page = extent_buffer_page(eb, i);
4596 ClearPageUptodate(page);
4601 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4605 unsigned long num_pages;
4607 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4608 num_pages = num_extent_pages(eb->start, eb->len);
4609 for (i = 0; i < num_pages; i++) {
4610 page = extent_buffer_page(eb, i);
4611 SetPageUptodate(page);
4616 int extent_buffer_uptodate(struct extent_buffer *eb)
4618 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4621 int read_extent_buffer_pages(struct extent_io_tree *tree,
4622 struct extent_buffer *eb, u64 start, int wait,
4623 get_extent_t *get_extent, int mirror_num)
4626 unsigned long start_i;
4630 int locked_pages = 0;
4631 int all_uptodate = 1;
4632 unsigned long num_pages;
4633 unsigned long num_reads = 0;
4634 struct bio *bio = NULL;
4635 unsigned long bio_flags = 0;
4637 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4641 WARN_ON(start < eb->start);
4642 start_i = (start >> PAGE_CACHE_SHIFT) -
4643 (eb->start >> PAGE_CACHE_SHIFT);
4648 num_pages = num_extent_pages(eb->start, eb->len);
4649 for (i = start_i; i < num_pages; i++) {
4650 page = extent_buffer_page(eb, i);
4651 if (wait == WAIT_NONE) {
4652 if (!trylock_page(page))
4658 if (!PageUptodate(page)) {
4665 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4669 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4670 eb->read_mirror = 0;
4671 atomic_set(&eb->io_pages, num_reads);
4672 for (i = start_i; i < num_pages; i++) {
4673 page = extent_buffer_page(eb, i);
4674 if (!PageUptodate(page)) {
4675 ClearPageError(page);
4676 err = __extent_read_full_page(tree, page,
4678 mirror_num, &bio_flags,
4688 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4694 if (ret || wait != WAIT_COMPLETE)
4697 for (i = start_i; i < num_pages; i++) {
4698 page = extent_buffer_page(eb, i);
4699 wait_on_page_locked(page);
4700 if (!PageUptodate(page))
4708 while (locked_pages > 0) {
4709 page = extent_buffer_page(eb, i);
4717 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4718 unsigned long start,
4725 char *dst = (char *)dstv;
4726 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4727 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4729 WARN_ON(start > eb->len);
4730 WARN_ON(start + len > eb->start + eb->len);
4732 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4735 page = extent_buffer_page(eb, i);
4737 cur = min(len, (PAGE_CACHE_SIZE - offset));
4738 kaddr = page_address(page);
4739 memcpy(dst, kaddr + offset, cur);
4748 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4749 unsigned long min_len, char **map,
4750 unsigned long *map_start,
4751 unsigned long *map_len)
4753 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4756 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4757 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4758 unsigned long end_i = (start_offset + start + min_len - 1) >>
4765 offset = start_offset;
4769 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4772 if (start + min_len > eb->len) {
4773 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4774 "wanted %lu %lu\n", (unsigned long long)eb->start,
4775 eb->len, start, min_len);
4779 p = extent_buffer_page(eb, i);
4780 kaddr = page_address(p);
4781 *map = kaddr + offset;
4782 *map_len = PAGE_CACHE_SIZE - offset;
4786 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4787 unsigned long start,
4794 char *ptr = (char *)ptrv;
4795 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4796 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4799 WARN_ON(start > eb->len);
4800 WARN_ON(start + len > eb->start + eb->len);
4802 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4805 page = extent_buffer_page(eb, i);
4807 cur = min(len, (PAGE_CACHE_SIZE - offset));
4809 kaddr = page_address(page);
4810 ret = memcmp(ptr, kaddr + offset, cur);
4822 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4823 unsigned long start, unsigned long len)
4829 char *src = (char *)srcv;
4830 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4831 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4833 WARN_ON(start > eb->len);
4834 WARN_ON(start + len > eb->start + eb->len);
4836 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4839 page = extent_buffer_page(eb, i);
4840 WARN_ON(!PageUptodate(page));
4842 cur = min(len, PAGE_CACHE_SIZE - offset);
4843 kaddr = page_address(page);
4844 memcpy(kaddr + offset, src, cur);
4853 void memset_extent_buffer(struct extent_buffer *eb, char c,
4854 unsigned long start, unsigned long len)
4860 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4861 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4863 WARN_ON(start > eb->len);
4864 WARN_ON(start + len > eb->start + eb->len);
4866 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4869 page = extent_buffer_page(eb, i);
4870 WARN_ON(!PageUptodate(page));
4872 cur = min(len, PAGE_CACHE_SIZE - offset);
4873 kaddr = page_address(page);
4874 memset(kaddr + offset, c, cur);
4882 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4883 unsigned long dst_offset, unsigned long src_offset,
4886 u64 dst_len = dst->len;
4891 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4892 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4894 WARN_ON(src->len != dst_len);
4896 offset = (start_offset + dst_offset) &
4897 ((unsigned long)PAGE_CACHE_SIZE - 1);
4900 page = extent_buffer_page(dst, i);
4901 WARN_ON(!PageUptodate(page));
4903 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4905 kaddr = page_address(page);
4906 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4915 static void move_pages(struct page *dst_page, struct page *src_page,
4916 unsigned long dst_off, unsigned long src_off,
4919 char *dst_kaddr = page_address(dst_page);
4920 if (dst_page == src_page) {
4921 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4923 char *src_kaddr = page_address(src_page);
4924 char *p = dst_kaddr + dst_off + len;
4925 char *s = src_kaddr + src_off + len;
4932 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4934 unsigned long distance = (src > dst) ? src - dst : dst - src;
4935 return distance < len;
4938 static void copy_pages(struct page *dst_page, struct page *src_page,
4939 unsigned long dst_off, unsigned long src_off,
4942 char *dst_kaddr = page_address(dst_page);
4944 int must_memmove = 0;
4946 if (dst_page != src_page) {
4947 src_kaddr = page_address(src_page);
4949 src_kaddr = dst_kaddr;
4950 if (areas_overlap(src_off, dst_off, len))
4955 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4957 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4960 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4961 unsigned long src_offset, unsigned long len)
4964 size_t dst_off_in_page;
4965 size_t src_off_in_page;
4966 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4967 unsigned long dst_i;
4968 unsigned long src_i;
4970 if (src_offset + len > dst->len) {
4971 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4972 "len %lu dst len %lu\n", src_offset, len, dst->len);
4975 if (dst_offset + len > dst->len) {
4976 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4977 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4982 dst_off_in_page = (start_offset + dst_offset) &
4983 ((unsigned long)PAGE_CACHE_SIZE - 1);
4984 src_off_in_page = (start_offset + src_offset) &
4985 ((unsigned long)PAGE_CACHE_SIZE - 1);
4987 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4988 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4990 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4992 cur = min_t(unsigned long, cur,
4993 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4995 copy_pages(extent_buffer_page(dst, dst_i),
4996 extent_buffer_page(dst, src_i),
4997 dst_off_in_page, src_off_in_page, cur);
5005 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5006 unsigned long src_offset, unsigned long len)
5009 size_t dst_off_in_page;
5010 size_t src_off_in_page;
5011 unsigned long dst_end = dst_offset + len - 1;
5012 unsigned long src_end = src_offset + len - 1;
5013 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5014 unsigned long dst_i;
5015 unsigned long src_i;
5017 if (src_offset + len > dst->len) {
5018 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5019 "len %lu len %lu\n", src_offset, len, dst->len);
5022 if (dst_offset + len > dst->len) {
5023 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5024 "len %lu len %lu\n", dst_offset, len, dst->len);
5027 if (dst_offset < src_offset) {
5028 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5032 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5033 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5035 dst_off_in_page = (start_offset + dst_end) &
5036 ((unsigned long)PAGE_CACHE_SIZE - 1);
5037 src_off_in_page = (start_offset + src_end) &
5038 ((unsigned long)PAGE_CACHE_SIZE - 1);
5040 cur = min_t(unsigned long, len, src_off_in_page + 1);
5041 cur = min(cur, dst_off_in_page + 1);
5042 move_pages(extent_buffer_page(dst, dst_i),
5043 extent_buffer_page(dst, src_i),
5044 dst_off_in_page - cur + 1,
5045 src_off_in_page - cur + 1, cur);
5053 int try_release_extent_buffer(struct page *page)
5055 struct extent_buffer *eb;
5058 * We need to make sure noboody is attaching this page to an eb right
5061 spin_lock(&page->mapping->private_lock);
5062 if (!PagePrivate(page)) {
5063 spin_unlock(&page->mapping->private_lock);
5067 eb = (struct extent_buffer *)page->private;
5071 * This is a little awful but should be ok, we need to make sure that
5072 * the eb doesn't disappear out from under us while we're looking at
5075 spin_lock(&eb->refs_lock);
5076 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5077 spin_unlock(&eb->refs_lock);
5078 spin_unlock(&page->mapping->private_lock);
5081 spin_unlock(&page->mapping->private_lock);
5084 * If tree ref isn't set then we know the ref on this eb is a real ref,
5085 * so just return, this page will likely be freed soon anyway.
5087 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5088 spin_unlock(&eb->refs_lock);
5092 return release_extent_buffer(eb);