1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
25 static struct kmem_cache *extent_state_cache;
26 static struct kmem_cache *extent_buffer_cache;
28 static LIST_HEAD(buffers);
29 static LIST_HEAD(states);
33 static DEFINE_SPINLOCK(leak_lock);
36 #define BUFFER_LRU_MAX 64
41 struct rb_node rb_node;
44 struct extent_page_data {
46 struct extent_io_tree *tree;
47 get_extent_t *get_extent;
48 unsigned long bio_flags;
50 /* tells writepage not to lock the state bits for this range
51 * it still does the unlocking
53 unsigned int extent_locked:1;
55 /* tells the submit_bio code to use a WRITE_SYNC */
56 unsigned int sync_io:1;
59 static noinline void flush_write_bio(void *data);
60 static inline struct btrfs_fs_info *
61 tree_fs_info(struct extent_io_tree *tree)
63 return btrfs_sb(tree->mapping->host->i_sb);
66 int __init extent_io_init(void)
68 extent_state_cache = kmem_cache_create("btrfs_extent_state",
69 sizeof(struct extent_state), 0,
70 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
71 if (!extent_state_cache)
74 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
75 sizeof(struct extent_buffer), 0,
76 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
77 if (!extent_buffer_cache)
78 goto free_state_cache;
82 kmem_cache_destroy(extent_state_cache);
86 void extent_io_exit(void)
88 struct extent_state *state;
89 struct extent_buffer *eb;
91 while (!list_empty(&states)) {
92 state = list_entry(states.next, struct extent_state, leak_list);
93 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
94 "state %lu in tree %p refs %d\n",
95 (unsigned long long)state->start,
96 (unsigned long long)state->end,
97 state->state, state->tree, atomic_read(&state->refs));
98 list_del(&state->leak_list);
99 kmem_cache_free(extent_state_cache, state);
103 while (!list_empty(&buffers)) {
104 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
105 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
106 "refs %d\n", (unsigned long long)eb->start,
107 eb->len, atomic_read(&eb->refs));
108 list_del(&eb->leak_list);
109 kmem_cache_free(extent_buffer_cache, eb);
111 if (extent_state_cache)
112 kmem_cache_destroy(extent_state_cache);
113 if (extent_buffer_cache)
114 kmem_cache_destroy(extent_buffer_cache);
117 void extent_io_tree_init(struct extent_io_tree *tree,
118 struct address_space *mapping)
120 tree->state = RB_ROOT;
121 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
123 tree->dirty_bytes = 0;
124 spin_lock_init(&tree->lock);
125 spin_lock_init(&tree->buffer_lock);
126 tree->mapping = mapping;
129 static struct extent_state *alloc_extent_state(gfp_t mask)
131 struct extent_state *state;
136 state = kmem_cache_alloc(extent_state_cache, mask);
143 spin_lock_irqsave(&leak_lock, flags);
144 list_add(&state->leak_list, &states);
145 spin_unlock_irqrestore(&leak_lock, flags);
147 atomic_set(&state->refs, 1);
148 init_waitqueue_head(&state->wq);
149 trace_alloc_extent_state(state, mask, _RET_IP_);
153 void free_extent_state(struct extent_state *state)
157 if (atomic_dec_and_test(&state->refs)) {
161 WARN_ON(state->tree);
163 spin_lock_irqsave(&leak_lock, flags);
164 list_del(&state->leak_list);
165 spin_unlock_irqrestore(&leak_lock, flags);
167 trace_free_extent_state(state, _RET_IP_);
168 kmem_cache_free(extent_state_cache, state);
172 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
173 struct rb_node *node)
175 struct rb_node **p = &root->rb_node;
176 struct rb_node *parent = NULL;
177 struct tree_entry *entry;
181 entry = rb_entry(parent, struct tree_entry, rb_node);
183 if (offset < entry->start)
185 else if (offset > entry->end)
191 rb_link_node(node, parent, p);
192 rb_insert_color(node, root);
196 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
197 struct rb_node **prev_ret,
198 struct rb_node **next_ret)
200 struct rb_root *root = &tree->state;
201 struct rb_node *n = root->rb_node;
202 struct rb_node *prev = NULL;
203 struct rb_node *orig_prev = NULL;
204 struct tree_entry *entry;
205 struct tree_entry *prev_entry = NULL;
208 entry = rb_entry(n, struct tree_entry, rb_node);
212 if (offset < entry->start)
214 else if (offset > entry->end)
222 while (prev && offset > prev_entry->end) {
223 prev = rb_next(prev);
224 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
231 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
232 while (prev && offset < prev_entry->start) {
233 prev = rb_prev(prev);
234 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
241 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
244 struct rb_node *prev = NULL;
247 ret = __etree_search(tree, offset, &prev, NULL);
253 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
254 struct extent_state *other)
256 if (tree->ops && tree->ops->merge_extent_hook)
257 tree->ops->merge_extent_hook(tree->mapping->host, new,
262 * utility function to look for merge candidates inside a given range.
263 * Any extents with matching state are merged together into a single
264 * extent in the tree. Extents with EXTENT_IO in their state field
265 * are not merged because the end_io handlers need to be able to do
266 * operations on them without sleeping (or doing allocations/splits).
268 * This should be called with the tree lock held.
270 static void merge_state(struct extent_io_tree *tree,
271 struct extent_state *state)
273 struct extent_state *other;
274 struct rb_node *other_node;
276 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
279 other_node = rb_prev(&state->rb_node);
281 other = rb_entry(other_node, struct extent_state, rb_node);
282 if (other->end == state->start - 1 &&
283 other->state == state->state) {
284 merge_cb(tree, state, other);
285 state->start = other->start;
287 rb_erase(&other->rb_node, &tree->state);
288 free_extent_state(other);
291 other_node = rb_next(&state->rb_node);
293 other = rb_entry(other_node, struct extent_state, rb_node);
294 if (other->start == state->end + 1 &&
295 other->state == state->state) {
296 merge_cb(tree, state, other);
297 state->end = other->end;
299 rb_erase(&other->rb_node, &tree->state);
300 free_extent_state(other);
305 static void set_state_cb(struct extent_io_tree *tree,
306 struct extent_state *state, int *bits)
308 if (tree->ops && tree->ops->set_bit_hook)
309 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
312 static void clear_state_cb(struct extent_io_tree *tree,
313 struct extent_state *state, int *bits)
315 if (tree->ops && tree->ops->clear_bit_hook)
316 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
319 static void set_state_bits(struct extent_io_tree *tree,
320 struct extent_state *state, int *bits);
323 * insert an extent_state struct into the tree. 'bits' are set on the
324 * struct before it is inserted.
326 * This may return -EEXIST if the extent is already there, in which case the
327 * state struct is freed.
329 * The tree lock is not taken internally. This is a utility function and
330 * probably isn't what you want to call (see set/clear_extent_bit).
332 static int insert_state(struct extent_io_tree *tree,
333 struct extent_state *state, u64 start, u64 end,
336 struct rb_node *node;
339 printk(KERN_ERR "btrfs end < start %llu %llu\n",
340 (unsigned long long)end,
341 (unsigned long long)start);
344 state->start = start;
347 set_state_bits(tree, state, bits);
349 node = tree_insert(&tree->state, end, &state->rb_node);
351 struct extent_state *found;
352 found = rb_entry(node, struct extent_state, rb_node);
353 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
354 "%llu %llu\n", (unsigned long long)found->start,
355 (unsigned long long)found->end,
356 (unsigned long long)start, (unsigned long long)end);
360 merge_state(tree, state);
364 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
367 if (tree->ops && tree->ops->split_extent_hook)
368 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
372 * split a given extent state struct in two, inserting the preallocated
373 * struct 'prealloc' as the newly created second half. 'split' indicates an
374 * offset inside 'orig' where it should be split.
377 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
378 * are two extent state structs in the tree:
379 * prealloc: [orig->start, split - 1]
380 * orig: [ split, orig->end ]
382 * The tree locks are not taken by this function. They need to be held
385 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
386 struct extent_state *prealloc, u64 split)
388 struct rb_node *node;
390 split_cb(tree, orig, split);
392 prealloc->start = orig->start;
393 prealloc->end = split - 1;
394 prealloc->state = orig->state;
397 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
399 free_extent_state(prealloc);
402 prealloc->tree = tree;
406 static struct extent_state *next_state(struct extent_state *state)
408 struct rb_node *next = rb_next(&state->rb_node);
410 return rb_entry(next, struct extent_state, rb_node);
416 * utility function to clear some bits in an extent state struct.
417 * it will optionally wake up any one waiting on this state (wake == 1).
419 * If no bits are set on the state struct after clearing things, the
420 * struct is freed and removed from the tree
422 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
423 struct extent_state *state,
426 struct extent_state *next;
427 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
429 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
430 u64 range = state->end - state->start + 1;
431 WARN_ON(range > tree->dirty_bytes);
432 tree->dirty_bytes -= range;
434 clear_state_cb(tree, state, bits);
435 state->state &= ~bits_to_clear;
438 if (state->state == 0) {
439 next = next_state(state);
441 rb_erase(&state->rb_node, &tree->state);
443 free_extent_state(state);
448 merge_state(tree, state);
449 next = next_state(state);
454 static struct extent_state *
455 alloc_extent_state_atomic(struct extent_state *prealloc)
458 prealloc = alloc_extent_state(GFP_ATOMIC);
463 void extent_io_tree_panic(struct extent_io_tree *tree, int err)
465 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
466 "Extent tree was modified by another "
467 "thread while locked.");
471 * clear some bits on a range in the tree. This may require splitting
472 * or inserting elements in the tree, so the gfp mask is used to
473 * indicate which allocations or sleeping are allowed.
475 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
476 * the given range from the tree regardless of state (ie for truncate).
478 * the range [start, end] is inclusive.
480 * This takes the tree lock, and returns 0 on success and < 0 on error.
482 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
483 int bits, int wake, int delete,
484 struct extent_state **cached_state,
487 struct extent_state *state;
488 struct extent_state *cached;
489 struct extent_state *prealloc = NULL;
490 struct rb_node *node;
496 bits |= ~EXTENT_CTLBITS;
497 bits |= EXTENT_FIRST_DELALLOC;
499 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
502 if (!prealloc && (mask & __GFP_WAIT)) {
503 prealloc = alloc_extent_state(mask);
508 spin_lock(&tree->lock);
510 cached = *cached_state;
513 *cached_state = NULL;
517 if (cached && cached->tree && cached->start <= start &&
518 cached->end > start) {
520 atomic_dec(&cached->refs);
525 free_extent_state(cached);
528 * this search will find the extents that end after
531 node = tree_search(tree, start);
534 state = rb_entry(node, struct extent_state, rb_node);
536 if (state->start > end)
538 WARN_ON(state->end < start);
539 last_end = state->end;
541 /* the state doesn't have the wanted bits, go ahead */
542 if (!(state->state & bits)) {
543 state = next_state(state);
548 * | ---- desired range ---- |
550 * | ------------- state -------------- |
552 * We need to split the extent we found, and may flip
553 * bits on second half.
555 * If the extent we found extends past our range, we
556 * just split and search again. It'll get split again
557 * the next time though.
559 * If the extent we found is inside our range, we clear
560 * the desired bit on it.
563 if (state->start < start) {
564 prealloc = alloc_extent_state_atomic(prealloc);
566 err = split_state(tree, state, prealloc, start);
568 extent_io_tree_panic(tree, err);
573 if (state->end <= end) {
574 state = clear_state_bit(tree, state, &bits, wake);
580 * | ---- desired range ---- |
582 * We need to split the extent, and clear the bit
585 if (state->start <= end && state->end > end) {
586 prealloc = alloc_extent_state_atomic(prealloc);
588 err = split_state(tree, state, prealloc, end + 1);
590 extent_io_tree_panic(tree, err);
595 clear_state_bit(tree, prealloc, &bits, wake);
601 state = clear_state_bit(tree, state, &bits, wake);
603 if (last_end == (u64)-1)
605 start = last_end + 1;
606 if (start <= end && state && !need_resched())
611 spin_unlock(&tree->lock);
613 free_extent_state(prealloc);
620 spin_unlock(&tree->lock);
621 if (mask & __GFP_WAIT)
626 static void wait_on_state(struct extent_io_tree *tree,
627 struct extent_state *state)
628 __releases(tree->lock)
629 __acquires(tree->lock)
632 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
633 spin_unlock(&tree->lock);
635 spin_lock(&tree->lock);
636 finish_wait(&state->wq, &wait);
640 * waits for one or more bits to clear on a range in the state tree.
641 * The range [start, end] is inclusive.
642 * The tree lock is taken by this function
644 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
646 struct extent_state *state;
647 struct rb_node *node;
649 spin_lock(&tree->lock);
653 * this search will find all the extents that end after
656 node = tree_search(tree, start);
660 state = rb_entry(node, struct extent_state, rb_node);
662 if (state->start > end)
665 if (state->state & bits) {
666 start = state->start;
667 atomic_inc(&state->refs);
668 wait_on_state(tree, state);
669 free_extent_state(state);
672 start = state->end + 1;
677 cond_resched_lock(&tree->lock);
680 spin_unlock(&tree->lock);
683 static void set_state_bits(struct extent_io_tree *tree,
684 struct extent_state *state,
687 int bits_to_set = *bits & ~EXTENT_CTLBITS;
689 set_state_cb(tree, state, bits);
690 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
691 u64 range = state->end - state->start + 1;
692 tree->dirty_bytes += range;
694 state->state |= bits_to_set;
697 static void cache_state(struct extent_state *state,
698 struct extent_state **cached_ptr)
700 if (cached_ptr && !(*cached_ptr)) {
701 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
703 atomic_inc(&state->refs);
708 static void uncache_state(struct extent_state **cached_ptr)
710 if (cached_ptr && (*cached_ptr)) {
711 struct extent_state *state = *cached_ptr;
713 free_extent_state(state);
718 * set some bits on a range in the tree. This may require allocations or
719 * sleeping, so the gfp mask is used to indicate what is allowed.
721 * If any of the exclusive bits are set, this will fail with -EEXIST if some
722 * part of the range already has the desired bits set. The start of the
723 * existing range is returned in failed_start in this case.
725 * [start, end] is inclusive This takes the tree lock.
728 static int __must_check
729 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
730 int bits, int exclusive_bits, u64 *failed_start,
731 struct extent_state **cached_state, gfp_t mask)
733 struct extent_state *state;
734 struct extent_state *prealloc = NULL;
735 struct rb_node *node;
740 bits |= EXTENT_FIRST_DELALLOC;
742 if (!prealloc && (mask & __GFP_WAIT)) {
743 prealloc = alloc_extent_state(mask);
747 spin_lock(&tree->lock);
748 if (cached_state && *cached_state) {
749 state = *cached_state;
750 if (state->start <= start && state->end > start &&
752 node = &state->rb_node;
757 * this search will find all the extents that end after
760 node = tree_search(tree, start);
762 prealloc = alloc_extent_state_atomic(prealloc);
764 err = insert_state(tree, prealloc, start, end, &bits);
766 extent_io_tree_panic(tree, err);
771 state = rb_entry(node, struct extent_state, rb_node);
773 last_start = state->start;
774 last_end = state->end;
777 * | ---- desired range ---- |
780 * Just lock what we found and keep going
782 if (state->start == start && state->end <= end) {
783 if (state->state & exclusive_bits) {
784 *failed_start = state->start;
789 set_state_bits(tree, state, &bits);
790 cache_state(state, cached_state);
791 merge_state(tree, state);
792 if (last_end == (u64)-1)
794 start = last_end + 1;
795 state = next_state(state);
796 if (start < end && state && state->start == start &&
803 * | ---- desired range ---- |
806 * | ------------- state -------------- |
808 * We need to split the extent we found, and may flip bits on
811 * If the extent we found extends past our
812 * range, we just split and search again. It'll get split
813 * again the next time though.
815 * If the extent we found is inside our range, we set the
818 if (state->start < start) {
819 if (state->state & exclusive_bits) {
820 *failed_start = start;
825 prealloc = alloc_extent_state_atomic(prealloc);
827 err = split_state(tree, state, prealloc, start);
829 extent_io_tree_panic(tree, err);
834 if (state->end <= end) {
835 set_state_bits(tree, state, &bits);
836 cache_state(state, cached_state);
837 merge_state(tree, state);
838 if (last_end == (u64)-1)
840 start = last_end + 1;
841 state = next_state(state);
842 if (start < end && state && state->start == start &&
849 * | ---- desired range ---- |
850 * | state | or | state |
852 * There's a hole, we need to insert something in it and
853 * ignore the extent we found.
855 if (state->start > start) {
857 if (end < last_start)
860 this_end = last_start - 1;
862 prealloc = alloc_extent_state_atomic(prealloc);
866 * Avoid to free 'prealloc' if it can be merged with
869 err = insert_state(tree, prealloc, start, this_end,
872 extent_io_tree_panic(tree, err);
874 cache_state(prealloc, cached_state);
876 start = this_end + 1;
880 * | ---- desired range ---- |
882 * We need to split the extent, and set the bit
885 if (state->start <= end && state->end > end) {
886 if (state->state & exclusive_bits) {
887 *failed_start = start;
892 prealloc = alloc_extent_state_atomic(prealloc);
894 err = split_state(tree, state, prealloc, end + 1);
896 extent_io_tree_panic(tree, err);
898 set_state_bits(tree, prealloc, &bits);
899 cache_state(prealloc, cached_state);
900 merge_state(tree, prealloc);
908 spin_unlock(&tree->lock);
910 free_extent_state(prealloc);
917 spin_unlock(&tree->lock);
918 if (mask & __GFP_WAIT)
923 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
924 u64 *failed_start, struct extent_state **cached_state,
927 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
933 * convert_extent_bit - convert all bits in a given range from one bit to
935 * @tree: the io tree to search
936 * @start: the start offset in bytes
937 * @end: the end offset in bytes (inclusive)
938 * @bits: the bits to set in this range
939 * @clear_bits: the bits to clear in this range
940 * @mask: the allocation mask
942 * This will go through and set bits for the given range. If any states exist
943 * already in this range they are set with the given bit and cleared of the
944 * clear_bits. This is only meant to be used by things that are mergeable, ie
945 * converting from say DELALLOC to DIRTY. This is not meant to be used with
946 * boundary bits like LOCK.
948 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
949 int bits, int clear_bits, gfp_t mask)
951 struct extent_state *state;
952 struct extent_state *prealloc = NULL;
953 struct rb_node *node;
959 if (!prealloc && (mask & __GFP_WAIT)) {
960 prealloc = alloc_extent_state(mask);
965 spin_lock(&tree->lock);
967 * this search will find all the extents that end after
970 node = tree_search(tree, start);
972 prealloc = alloc_extent_state_atomic(prealloc);
977 err = insert_state(tree, prealloc, start, end, &bits);
980 extent_io_tree_panic(tree, err);
983 state = rb_entry(node, struct extent_state, rb_node);
985 last_start = state->start;
986 last_end = state->end;
989 * | ---- desired range ---- |
992 * Just lock what we found and keep going
994 if (state->start == start && state->end <= end) {
995 set_state_bits(tree, state, &bits);
996 state = clear_state_bit(tree, state, &clear_bits, 0);
997 if (last_end == (u64)-1)
999 start = last_end + 1;
1000 if (start < end && state && state->start == start &&
1007 * | ---- desired range ---- |
1010 * | ------------- state -------------- |
1012 * We need to split the extent we found, and may flip bits on
1015 * If the extent we found extends past our
1016 * range, we just split and search again. It'll get split
1017 * again the next time though.
1019 * If the extent we found is inside our range, we set the
1020 * desired bit on it.
1022 if (state->start < start) {
1023 prealloc = alloc_extent_state_atomic(prealloc);
1028 err = split_state(tree, state, prealloc, start);
1030 extent_io_tree_panic(tree, err);
1034 if (state->end <= end) {
1035 set_state_bits(tree, state, &bits);
1036 state = clear_state_bit(tree, state, &clear_bits, 0);
1037 if (last_end == (u64)-1)
1039 start = last_end + 1;
1040 if (start < end && state && state->start == start &&
1047 * | ---- desired range ---- |
1048 * | state | or | state |
1050 * There's a hole, we need to insert something in it and
1051 * ignore the extent we found.
1053 if (state->start > start) {
1055 if (end < last_start)
1058 this_end = last_start - 1;
1060 prealloc = alloc_extent_state_atomic(prealloc);
1067 * Avoid to free 'prealloc' if it can be merged with
1070 err = insert_state(tree, prealloc, start, this_end,
1073 extent_io_tree_panic(tree, err);
1075 start = this_end + 1;
1079 * | ---- desired range ---- |
1081 * We need to split the extent, and set the bit
1084 if (state->start <= end && state->end > end) {
1085 prealloc = alloc_extent_state_atomic(prealloc);
1091 err = split_state(tree, state, prealloc, end + 1);
1093 extent_io_tree_panic(tree, err);
1095 set_state_bits(tree, prealloc, &bits);
1096 clear_state_bit(tree, prealloc, &clear_bits, 0);
1104 spin_unlock(&tree->lock);
1106 free_extent_state(prealloc);
1113 spin_unlock(&tree->lock);
1114 if (mask & __GFP_WAIT)
1119 /* wrappers around set/clear extent bit */
1120 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1123 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1127 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1128 int bits, gfp_t mask)
1130 return set_extent_bit(tree, start, end, bits, NULL,
1134 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1135 int bits, gfp_t mask)
1137 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1140 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1141 struct extent_state **cached_state, gfp_t mask)
1143 return set_extent_bit(tree, start, end,
1144 EXTENT_DELALLOC | EXTENT_UPTODATE,
1145 NULL, cached_state, mask);
1148 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1149 struct extent_state **cached_state, gfp_t mask)
1151 return set_extent_bit(tree, start, end,
1152 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1153 NULL, cached_state, mask);
1156 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1159 return clear_extent_bit(tree, start, end,
1160 EXTENT_DIRTY | EXTENT_DELALLOC |
1161 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1164 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1167 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1171 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1172 struct extent_state **cached_state, gfp_t mask)
1174 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1175 cached_state, mask);
1178 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1179 struct extent_state **cached_state, gfp_t mask)
1181 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1182 cached_state, mask);
1186 * either insert or lock state struct between start and end use mask to tell
1187 * us if waiting is desired.
1189 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1190 int bits, struct extent_state **cached_state)
1195 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1196 EXTENT_LOCKED, &failed_start,
1197 cached_state, GFP_NOFS);
1198 if (err == -EEXIST) {
1199 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1200 start = failed_start;
1203 WARN_ON(start > end);
1208 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1210 return lock_extent_bits(tree, start, end, 0, NULL);
1213 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1218 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1219 &failed_start, NULL, GFP_NOFS);
1220 if (err == -EEXIST) {
1221 if (failed_start > start)
1222 clear_extent_bit(tree, start, failed_start - 1,
1223 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1229 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1230 struct extent_state **cached, gfp_t mask)
1232 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1236 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1238 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1243 * helper function to set both pages and extents in the tree writeback
1245 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1247 unsigned long index = start >> PAGE_CACHE_SHIFT;
1248 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1251 while (index <= end_index) {
1252 page = find_get_page(tree->mapping, index);
1253 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1254 set_page_writeback(page);
1255 page_cache_release(page);
1261 /* find the first state struct with 'bits' set after 'start', and
1262 * return it. tree->lock must be held. NULL will returned if
1263 * nothing was found after 'start'
1265 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1266 u64 start, int bits)
1268 struct rb_node *node;
1269 struct extent_state *state;
1272 * this search will find all the extents that end after
1275 node = tree_search(tree, start);
1280 state = rb_entry(node, struct extent_state, rb_node);
1281 if (state->end >= start && (state->state & bits))
1284 node = rb_next(node);
1293 * find the first offset in the io tree with 'bits' set. zero is
1294 * returned if we find something, and *start_ret and *end_ret are
1295 * set to reflect the state struct that was found.
1297 * If nothing was found, 1 is returned. If found something, return 0.
1299 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1300 u64 *start_ret, u64 *end_ret, int bits)
1302 struct extent_state *state;
1305 spin_lock(&tree->lock);
1306 state = find_first_extent_bit_state(tree, start, bits);
1308 *start_ret = state->start;
1309 *end_ret = state->end;
1312 spin_unlock(&tree->lock);
1317 * find a contiguous range of bytes in the file marked as delalloc, not
1318 * more than 'max_bytes'. start and end are used to return the range,
1320 * 1 is returned if we find something, 0 if nothing was in the tree
1322 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1323 u64 *start, u64 *end, u64 max_bytes,
1324 struct extent_state **cached_state)
1326 struct rb_node *node;
1327 struct extent_state *state;
1328 u64 cur_start = *start;
1330 u64 total_bytes = 0;
1332 spin_lock(&tree->lock);
1335 * this search will find all the extents that end after
1338 node = tree_search(tree, cur_start);
1346 state = rb_entry(node, struct extent_state, rb_node);
1347 if (found && (state->start != cur_start ||
1348 (state->state & EXTENT_BOUNDARY))) {
1351 if (!(state->state & EXTENT_DELALLOC)) {
1357 *start = state->start;
1358 *cached_state = state;
1359 atomic_inc(&state->refs);
1363 cur_start = state->end + 1;
1364 node = rb_next(node);
1367 total_bytes += state->end - state->start + 1;
1368 if (total_bytes >= max_bytes)
1372 spin_unlock(&tree->lock);
1376 static noinline void __unlock_for_delalloc(struct inode *inode,
1377 struct page *locked_page,
1381 struct page *pages[16];
1382 unsigned long index = start >> PAGE_CACHE_SHIFT;
1383 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1384 unsigned long nr_pages = end_index - index + 1;
1387 if (index == locked_page->index && end_index == index)
1390 while (nr_pages > 0) {
1391 ret = find_get_pages_contig(inode->i_mapping, index,
1392 min_t(unsigned long, nr_pages,
1393 ARRAY_SIZE(pages)), pages);
1394 for (i = 0; i < ret; i++) {
1395 if (pages[i] != locked_page)
1396 unlock_page(pages[i]);
1397 page_cache_release(pages[i]);
1405 static noinline int lock_delalloc_pages(struct inode *inode,
1406 struct page *locked_page,
1410 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1411 unsigned long start_index = index;
1412 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1413 unsigned long pages_locked = 0;
1414 struct page *pages[16];
1415 unsigned long nrpages;
1419 /* the caller is responsible for locking the start index */
1420 if (index == locked_page->index && index == end_index)
1423 /* skip the page at the start index */
1424 nrpages = end_index - index + 1;
1425 while (nrpages > 0) {
1426 ret = find_get_pages_contig(inode->i_mapping, index,
1427 min_t(unsigned long,
1428 nrpages, ARRAY_SIZE(pages)), pages);
1433 /* now we have an array of pages, lock them all */
1434 for (i = 0; i < ret; i++) {
1436 * the caller is taking responsibility for
1439 if (pages[i] != locked_page) {
1440 lock_page(pages[i]);
1441 if (!PageDirty(pages[i]) ||
1442 pages[i]->mapping != inode->i_mapping) {
1444 unlock_page(pages[i]);
1445 page_cache_release(pages[i]);
1449 page_cache_release(pages[i]);
1458 if (ret && pages_locked) {
1459 __unlock_for_delalloc(inode, locked_page,
1461 ((u64)(start_index + pages_locked - 1)) <<
1468 * find a contiguous range of bytes in the file marked as delalloc, not
1469 * more than 'max_bytes'. start and end are used to return the range,
1471 * 1 is returned if we find something, 0 if nothing was in the tree
1473 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1474 struct extent_io_tree *tree,
1475 struct page *locked_page,
1476 u64 *start, u64 *end,
1482 struct extent_state *cached_state = NULL;
1487 /* step one, find a bunch of delalloc bytes starting at start */
1488 delalloc_start = *start;
1490 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1491 max_bytes, &cached_state);
1492 if (!found || delalloc_end <= *start) {
1493 *start = delalloc_start;
1494 *end = delalloc_end;
1495 free_extent_state(cached_state);
1500 * start comes from the offset of locked_page. We have to lock
1501 * pages in order, so we can't process delalloc bytes before
1504 if (delalloc_start < *start)
1505 delalloc_start = *start;
1508 * make sure to limit the number of pages we try to lock down
1511 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1512 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1514 /* step two, lock all the pages after the page that has start */
1515 ret = lock_delalloc_pages(inode, locked_page,
1516 delalloc_start, delalloc_end);
1517 if (ret == -EAGAIN) {
1518 /* some of the pages are gone, lets avoid looping by
1519 * shortening the size of the delalloc range we're searching
1521 free_extent_state(cached_state);
1523 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1524 max_bytes = PAGE_CACHE_SIZE - offset;
1532 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1534 /* step three, lock the state bits for the whole range */
1535 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1537 /* then test to make sure it is all still delalloc */
1538 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1539 EXTENT_DELALLOC, 1, cached_state);
1541 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1542 &cached_state, GFP_NOFS);
1543 __unlock_for_delalloc(inode, locked_page,
1544 delalloc_start, delalloc_end);
1548 free_extent_state(cached_state);
1549 *start = delalloc_start;
1550 *end = delalloc_end;
1555 int extent_clear_unlock_delalloc(struct inode *inode,
1556 struct extent_io_tree *tree,
1557 u64 start, u64 end, struct page *locked_page,
1561 struct page *pages[16];
1562 unsigned long index = start >> PAGE_CACHE_SHIFT;
1563 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1564 unsigned long nr_pages = end_index - index + 1;
1568 if (op & EXTENT_CLEAR_UNLOCK)
1569 clear_bits |= EXTENT_LOCKED;
1570 if (op & EXTENT_CLEAR_DIRTY)
1571 clear_bits |= EXTENT_DIRTY;
1573 if (op & EXTENT_CLEAR_DELALLOC)
1574 clear_bits |= EXTENT_DELALLOC;
1576 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1577 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1578 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1579 EXTENT_SET_PRIVATE2)))
1582 while (nr_pages > 0) {
1583 ret = find_get_pages_contig(inode->i_mapping, index,
1584 min_t(unsigned long,
1585 nr_pages, ARRAY_SIZE(pages)), pages);
1586 for (i = 0; i < ret; i++) {
1588 if (op & EXTENT_SET_PRIVATE2)
1589 SetPagePrivate2(pages[i]);
1591 if (pages[i] == locked_page) {
1592 page_cache_release(pages[i]);
1595 if (op & EXTENT_CLEAR_DIRTY)
1596 clear_page_dirty_for_io(pages[i]);
1597 if (op & EXTENT_SET_WRITEBACK)
1598 set_page_writeback(pages[i]);
1599 if (op & EXTENT_END_WRITEBACK)
1600 end_page_writeback(pages[i]);
1601 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1602 unlock_page(pages[i]);
1603 page_cache_release(pages[i]);
1613 * count the number of bytes in the tree that have a given bit(s)
1614 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1615 * cached. The total number found is returned.
1617 u64 count_range_bits(struct extent_io_tree *tree,
1618 u64 *start, u64 search_end, u64 max_bytes,
1619 unsigned long bits, int contig)
1621 struct rb_node *node;
1622 struct extent_state *state;
1623 u64 cur_start = *start;
1624 u64 total_bytes = 0;
1628 if (search_end <= cur_start) {
1633 spin_lock(&tree->lock);
1634 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1635 total_bytes = tree->dirty_bytes;
1639 * this search will find all the extents that end after
1642 node = tree_search(tree, cur_start);
1647 state = rb_entry(node, struct extent_state, rb_node);
1648 if (state->start > search_end)
1650 if (contig && found && state->start > last + 1)
1652 if (state->end >= cur_start && (state->state & bits) == bits) {
1653 total_bytes += min(search_end, state->end) + 1 -
1654 max(cur_start, state->start);
1655 if (total_bytes >= max_bytes)
1658 *start = max(cur_start, state->start);
1662 } else if (contig && found) {
1665 node = rb_next(node);
1670 spin_unlock(&tree->lock);
1675 * set the private field for a given byte offset in the tree. If there isn't
1676 * an extent_state there already, this does nothing.
1678 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1680 struct rb_node *node;
1681 struct extent_state *state;
1684 spin_lock(&tree->lock);
1686 * this search will find all the extents that end after
1689 node = tree_search(tree, start);
1694 state = rb_entry(node, struct extent_state, rb_node);
1695 if (state->start != start) {
1699 state->private = private;
1701 spin_unlock(&tree->lock);
1705 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1707 struct rb_node *node;
1708 struct extent_state *state;
1711 spin_lock(&tree->lock);
1713 * this search will find all the extents that end after
1716 node = tree_search(tree, start);
1721 state = rb_entry(node, struct extent_state, rb_node);
1722 if (state->start != start) {
1726 *private = state->private;
1728 spin_unlock(&tree->lock);
1733 * searches a range in the state tree for a given mask.
1734 * If 'filled' == 1, this returns 1 only if every extent in the tree
1735 * has the bits set. Otherwise, 1 is returned if any bit in the
1736 * range is found set.
1738 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1739 int bits, int filled, struct extent_state *cached)
1741 struct extent_state *state = NULL;
1742 struct rb_node *node;
1745 spin_lock(&tree->lock);
1746 if (cached && cached->tree && cached->start <= start &&
1747 cached->end > start)
1748 node = &cached->rb_node;
1750 node = tree_search(tree, start);
1751 while (node && start <= end) {
1752 state = rb_entry(node, struct extent_state, rb_node);
1754 if (filled && state->start > start) {
1759 if (state->start > end)
1762 if (state->state & bits) {
1766 } else if (filled) {
1771 if (state->end == (u64)-1)
1774 start = state->end + 1;
1777 node = rb_next(node);
1784 spin_unlock(&tree->lock);
1789 * helper function to set a given page up to date if all the
1790 * extents in the tree for that page are up to date
1792 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1794 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1795 u64 end = start + PAGE_CACHE_SIZE - 1;
1796 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1797 SetPageUptodate(page);
1801 * helper function to unlock a page if all the extents in the tree
1802 * for that page are unlocked
1804 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1806 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1807 u64 end = start + PAGE_CACHE_SIZE - 1;
1808 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1813 * helper function to end page writeback if all the extents
1814 * in the tree for that page are done with writeback
1816 static void check_page_writeback(struct extent_io_tree *tree,
1819 end_page_writeback(page);
1823 * When IO fails, either with EIO or csum verification fails, we
1824 * try other mirrors that might have a good copy of the data. This
1825 * io_failure_record is used to record state as we go through all the
1826 * mirrors. If another mirror has good data, the page is set up to date
1827 * and things continue. If a good mirror can't be found, the original
1828 * bio end_io callback is called to indicate things have failed.
1830 struct io_failure_record {
1835 unsigned long bio_flags;
1841 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1846 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1848 set_state_private(failure_tree, rec->start, 0);
1849 ret = clear_extent_bits(failure_tree, rec->start,
1850 rec->start + rec->len - 1,
1851 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1856 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1857 rec->start + rec->len - 1,
1858 EXTENT_DAMAGED, GFP_NOFS);
1867 static void repair_io_failure_callback(struct bio *bio, int err)
1869 complete(bio->bi_private);
1873 * this bypasses the standard btrfs submit functions deliberately, as
1874 * the standard behavior is to write all copies in a raid setup. here we only
1875 * want to write the one bad copy. so we do the mapping for ourselves and issue
1876 * submit_bio directly.
1877 * to avoid any synchonization issues, wait for the data after writing, which
1878 * actually prevents the read that triggered the error from finishing.
1879 * currently, there can be no more than two copies of every data bit. thus,
1880 * exactly one rewrite is required.
1882 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1883 u64 length, u64 logical, struct page *page,
1887 struct btrfs_device *dev;
1888 DECLARE_COMPLETION_ONSTACK(compl);
1891 struct btrfs_bio *bbio = NULL;
1894 BUG_ON(!mirror_num);
1896 bio = bio_alloc(GFP_NOFS, 1);
1899 bio->bi_private = &compl;
1900 bio->bi_end_io = repair_io_failure_callback;
1902 map_length = length;
1904 ret = btrfs_map_block(map_tree, WRITE, logical,
1905 &map_length, &bbio, mirror_num);
1910 BUG_ON(mirror_num != bbio->mirror_num);
1911 sector = bbio->stripes[mirror_num-1].physical >> 9;
1912 bio->bi_sector = sector;
1913 dev = bbio->stripes[mirror_num-1].dev;
1915 if (!dev || !dev->bdev || !dev->writeable) {
1919 bio->bi_bdev = dev->bdev;
1920 bio_add_page(bio, page, length, start-page_offset(page));
1921 btrfsic_submit_bio(WRITE_SYNC, bio);
1922 wait_for_completion(&compl);
1924 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1925 /* try to remap that extent elsewhere? */
1927 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
1931 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
1932 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
1933 start, rcu_str_deref(dev->name), sector);
1939 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
1942 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1943 u64 start = eb->start;
1944 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
1947 for (i = 0; i < num_pages; i++) {
1948 struct page *p = extent_buffer_page(eb, i);
1949 ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE,
1950 start, p, mirror_num);
1953 start += PAGE_CACHE_SIZE;
1960 * each time an IO finishes, we do a fast check in the IO failure tree
1961 * to see if we need to process or clean up an io_failure_record
1963 static int clean_io_failure(u64 start, struct page *page)
1966 u64 private_failure;
1967 struct io_failure_record *failrec;
1968 struct btrfs_mapping_tree *map_tree;
1969 struct extent_state *state;
1973 struct inode *inode = page->mapping->host;
1976 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1977 (u64)-1, 1, EXTENT_DIRTY, 0);
1981 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1986 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1987 BUG_ON(!failrec->this_mirror);
1989 if (failrec->in_validation) {
1990 /* there was no real error, just free the record */
1991 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1997 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1998 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2001 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2003 if (state && state->start == failrec->start) {
2004 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
2005 num_copies = btrfs_num_copies(map_tree, failrec->logical,
2007 if (num_copies > 1) {
2008 ret = repair_io_failure(map_tree, start, failrec->len,
2009 failrec->logical, page,
2010 failrec->failed_mirror);
2017 ret = free_io_failure(inode, failrec, did_repair);
2023 * this is a generic handler for readpage errors (default
2024 * readpage_io_failed_hook). if other copies exist, read those and write back
2025 * good data to the failed position. does not investigate in remapping the
2026 * failed extent elsewhere, hoping the device will be smart enough to do this as
2030 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2031 u64 start, u64 end, int failed_mirror,
2032 struct extent_state *state)
2034 struct io_failure_record *failrec = NULL;
2036 struct extent_map *em;
2037 struct inode *inode = page->mapping->host;
2038 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2039 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2040 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2047 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2049 ret = get_state_private(failure_tree, start, &private);
2051 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2054 failrec->start = start;
2055 failrec->len = end - start + 1;
2056 failrec->this_mirror = 0;
2057 failrec->bio_flags = 0;
2058 failrec->in_validation = 0;
2060 read_lock(&em_tree->lock);
2061 em = lookup_extent_mapping(em_tree, start, failrec->len);
2063 read_unlock(&em_tree->lock);
2068 if (em->start > start || em->start + em->len < start) {
2069 free_extent_map(em);
2072 read_unlock(&em_tree->lock);
2074 if (!em || IS_ERR(em)) {
2078 logical = start - em->start;
2079 logical = em->block_start + logical;
2080 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2081 logical = em->block_start;
2082 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2083 extent_set_compress_type(&failrec->bio_flags,
2086 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2087 "len=%llu\n", logical, start, failrec->len);
2088 failrec->logical = logical;
2089 free_extent_map(em);
2091 /* set the bits in the private failure tree */
2092 ret = set_extent_bits(failure_tree, start, end,
2093 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2095 ret = set_state_private(failure_tree, start,
2096 (u64)(unsigned long)failrec);
2097 /* set the bits in the inode's tree */
2099 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2106 failrec = (struct io_failure_record *)(unsigned long)private;
2107 pr_debug("bio_readpage_error: (found) logical=%llu, "
2108 "start=%llu, len=%llu, validation=%d\n",
2109 failrec->logical, failrec->start, failrec->len,
2110 failrec->in_validation);
2112 * when data can be on disk more than twice, add to failrec here
2113 * (e.g. with a list for failed_mirror) to make
2114 * clean_io_failure() clean all those errors at once.
2117 num_copies = btrfs_num_copies(
2118 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2119 failrec->logical, failrec->len);
2120 if (num_copies == 1) {
2122 * we only have a single copy of the data, so don't bother with
2123 * all the retry and error correction code that follows. no
2124 * matter what the error is, it is very likely to persist.
2126 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2127 "state=%p, num_copies=%d, next_mirror %d, "
2128 "failed_mirror %d\n", state, num_copies,
2129 failrec->this_mirror, failed_mirror);
2130 free_io_failure(inode, failrec, 0);
2135 spin_lock(&tree->lock);
2136 state = find_first_extent_bit_state(tree, failrec->start,
2138 if (state && state->start != failrec->start)
2140 spin_unlock(&tree->lock);
2144 * there are two premises:
2145 * a) deliver good data to the caller
2146 * b) correct the bad sectors on disk
2148 if (failed_bio->bi_vcnt > 1) {
2150 * to fulfill b), we need to know the exact failing sectors, as
2151 * we don't want to rewrite any more than the failed ones. thus,
2152 * we need separate read requests for the failed bio
2154 * if the following BUG_ON triggers, our validation request got
2155 * merged. we need separate requests for our algorithm to work.
2157 BUG_ON(failrec->in_validation);
2158 failrec->in_validation = 1;
2159 failrec->this_mirror = failed_mirror;
2160 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2163 * we're ready to fulfill a) and b) alongside. get a good copy
2164 * of the failed sector and if we succeed, we have setup
2165 * everything for repair_io_failure to do the rest for us.
2167 if (failrec->in_validation) {
2168 BUG_ON(failrec->this_mirror != failed_mirror);
2169 failrec->in_validation = 0;
2170 failrec->this_mirror = 0;
2172 failrec->failed_mirror = failed_mirror;
2173 failrec->this_mirror++;
2174 if (failrec->this_mirror == failed_mirror)
2175 failrec->this_mirror++;
2176 read_mode = READ_SYNC;
2179 if (!state || failrec->this_mirror > num_copies) {
2180 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2181 "next_mirror %d, failed_mirror %d\n", state,
2182 num_copies, failrec->this_mirror, failed_mirror);
2183 free_io_failure(inode, failrec, 0);
2187 bio = bio_alloc(GFP_NOFS, 1);
2189 free_io_failure(inode, failrec, 0);
2192 bio->bi_private = state;
2193 bio->bi_end_io = failed_bio->bi_end_io;
2194 bio->bi_sector = failrec->logical >> 9;
2195 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2198 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2200 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2201 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2202 failrec->this_mirror, num_copies, failrec->in_validation);
2204 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2205 failrec->this_mirror,
2206 failrec->bio_flags, 0);
2210 /* lots and lots of room for performance fixes in the end_bio funcs */
2212 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2214 int uptodate = (err == 0);
2215 struct extent_io_tree *tree;
2218 tree = &BTRFS_I(page->mapping->host)->io_tree;
2220 if (tree->ops && tree->ops->writepage_end_io_hook) {
2221 ret = tree->ops->writepage_end_io_hook(page, start,
2222 end, NULL, uptodate);
2228 ClearPageUptodate(page);
2235 * after a writepage IO is done, we need to:
2236 * clear the uptodate bits on error
2237 * clear the writeback bits in the extent tree for this IO
2238 * end_page_writeback if the page has no more pending IO
2240 * Scheduling is not allowed, so the extent state tree is expected
2241 * to have one and only one object corresponding to this IO.
2243 static void end_bio_extent_writepage(struct bio *bio, int err)
2245 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2246 struct extent_io_tree *tree;
2252 struct page *page = bvec->bv_page;
2253 tree = &BTRFS_I(page->mapping->host)->io_tree;
2255 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2257 end = start + bvec->bv_len - 1;
2259 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2264 if (--bvec >= bio->bi_io_vec)
2265 prefetchw(&bvec->bv_page->flags);
2267 if (end_extent_writepage(page, err, start, end))
2271 end_page_writeback(page);
2273 check_page_writeback(tree, page);
2274 } while (bvec >= bio->bi_io_vec);
2280 * after a readpage IO is done, we need to:
2281 * clear the uptodate bits on error
2282 * set the uptodate bits if things worked
2283 * set the page up to date if all extents in the tree are uptodate
2284 * clear the lock bit in the extent tree
2285 * unlock the page if there are no other extents locked for it
2287 * Scheduling is not allowed, so the extent state tree is expected
2288 * to have one and only one object corresponding to this IO.
2290 static void end_bio_extent_readpage(struct bio *bio, int err)
2292 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2293 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2294 struct bio_vec *bvec = bio->bi_io_vec;
2295 struct extent_io_tree *tree;
2306 struct page *page = bvec->bv_page;
2307 struct extent_state *cached = NULL;
2308 struct extent_state *state;
2310 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2311 "mirror=%ld\n", (u64)bio->bi_sector, err,
2312 (long int)bio->bi_bdev);
2313 tree = &BTRFS_I(page->mapping->host)->io_tree;
2315 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2317 end = start + bvec->bv_len - 1;
2319 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2324 if (++bvec <= bvec_end)
2325 prefetchw(&bvec->bv_page->flags);
2327 spin_lock(&tree->lock);
2328 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2329 if (state && state->start == start) {
2331 * take a reference on the state, unlock will drop
2334 cache_state(state, &cached);
2336 spin_unlock(&tree->lock);
2338 mirror = (int)(unsigned long)bio->bi_bdev;
2339 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2340 ret = tree->ops->readpage_end_io_hook(page, start, end,
2345 clean_io_failure(start, page);
2348 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2349 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2351 test_bit(BIO_UPTODATE, &bio->bi_flags))
2353 } else if (!uptodate) {
2355 * The generic bio_readpage_error handles errors the
2356 * following way: If possible, new read requests are
2357 * created and submitted and will end up in
2358 * end_bio_extent_readpage as well (if we're lucky, not
2359 * in the !uptodate case). In that case it returns 0 and
2360 * we just go on with the next page in our bio. If it
2361 * can't handle the error it will return -EIO and we
2362 * remain responsible for that page.
2364 ret = bio_readpage_error(bio, page, start, end, mirror, NULL);
2367 test_bit(BIO_UPTODATE, &bio->bi_flags);
2370 uncache_state(&cached);
2375 if (uptodate && tree->track_uptodate) {
2376 set_extent_uptodate(tree, start, end, &cached,
2379 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2383 SetPageUptodate(page);
2385 ClearPageUptodate(page);
2391 check_page_uptodate(tree, page);
2393 ClearPageUptodate(page);
2396 check_page_locked(tree, page);
2398 } while (bvec <= bvec_end);
2404 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2409 bio = bio_alloc(gfp_flags, nr_vecs);
2411 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2412 while (!bio && (nr_vecs /= 2))
2413 bio = bio_alloc(gfp_flags, nr_vecs);
2418 bio->bi_bdev = bdev;
2419 bio->bi_sector = first_sector;
2425 * Since writes are async, they will only return -ENOMEM.
2426 * Reads can return the full range of I/O error conditions.
2428 static int __must_check submit_one_bio(int rw, struct bio *bio,
2429 int mirror_num, unsigned long bio_flags)
2432 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2433 struct page *page = bvec->bv_page;
2434 struct extent_io_tree *tree = bio->bi_private;
2437 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2439 bio->bi_private = NULL;
2443 if (tree->ops && tree->ops->submit_bio_hook)
2444 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2445 mirror_num, bio_flags, start);
2447 btrfsic_submit_bio(rw, bio);
2449 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2455 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2456 unsigned long offset, size_t size, struct bio *bio,
2457 unsigned long bio_flags)
2460 if (tree->ops && tree->ops->merge_bio_hook)
2461 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2468 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2469 struct page *page, sector_t sector,
2470 size_t size, unsigned long offset,
2471 struct block_device *bdev,
2472 struct bio **bio_ret,
2473 unsigned long max_pages,
2474 bio_end_io_t end_io_func,
2476 unsigned long prev_bio_flags,
2477 unsigned long bio_flags)
2483 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2484 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2485 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2487 if (bio_ret && *bio_ret) {
2490 contig = bio->bi_sector == sector;
2492 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2495 if (prev_bio_flags != bio_flags || !contig ||
2496 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2497 bio_add_page(bio, page, page_size, offset) < page_size) {
2498 ret = submit_one_bio(rw, bio, mirror_num,
2507 if (this_compressed)
2510 nr = bio_get_nr_vecs(bdev);
2512 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2516 bio_add_page(bio, page, page_size, offset);
2517 bio->bi_end_io = end_io_func;
2518 bio->bi_private = tree;
2523 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2528 void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
2530 if (!PagePrivate(page)) {
2531 SetPagePrivate(page);
2532 page_cache_get(page);
2533 set_page_private(page, (unsigned long)eb);
2535 WARN_ON(page->private != (unsigned long)eb);
2539 void set_page_extent_mapped(struct page *page)
2541 if (!PagePrivate(page)) {
2542 SetPagePrivate(page);
2543 page_cache_get(page);
2544 set_page_private(page, EXTENT_PAGE_PRIVATE);
2549 * basic readpage implementation. Locked extent state structs are inserted
2550 * into the tree that are removed when the IO is done (by the end_io
2552 * XXX JDM: This needs looking at to ensure proper page locking
2554 static int __extent_read_full_page(struct extent_io_tree *tree,
2556 get_extent_t *get_extent,
2557 struct bio **bio, int mirror_num,
2558 unsigned long *bio_flags)
2560 struct inode *inode = page->mapping->host;
2561 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2562 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2566 u64 last_byte = i_size_read(inode);
2570 struct extent_map *em;
2571 struct block_device *bdev;
2572 struct btrfs_ordered_extent *ordered;
2575 size_t pg_offset = 0;
2577 size_t disk_io_size;
2578 size_t blocksize = inode->i_sb->s_blocksize;
2579 unsigned long this_bio_flag = 0;
2581 set_page_extent_mapped(page);
2583 if (!PageUptodate(page)) {
2584 if (cleancache_get_page(page) == 0) {
2585 BUG_ON(blocksize != PAGE_SIZE);
2592 lock_extent(tree, start, end);
2593 ordered = btrfs_lookup_ordered_extent(inode, start);
2596 unlock_extent(tree, start, end);
2597 btrfs_start_ordered_extent(inode, ordered, 1);
2598 btrfs_put_ordered_extent(ordered);
2601 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2603 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2606 iosize = PAGE_CACHE_SIZE - zero_offset;
2607 userpage = kmap_atomic(page);
2608 memset(userpage + zero_offset, 0, iosize);
2609 flush_dcache_page(page);
2610 kunmap_atomic(userpage);
2613 while (cur <= end) {
2614 if (cur >= last_byte) {
2616 struct extent_state *cached = NULL;
2618 iosize = PAGE_CACHE_SIZE - pg_offset;
2619 userpage = kmap_atomic(page);
2620 memset(userpage + pg_offset, 0, iosize);
2621 flush_dcache_page(page);
2622 kunmap_atomic(userpage);
2623 set_extent_uptodate(tree, cur, cur + iosize - 1,
2625 unlock_extent_cached(tree, cur, cur + iosize - 1,
2629 em = get_extent(inode, page, pg_offset, cur,
2631 if (IS_ERR_OR_NULL(em)) {
2633 unlock_extent(tree, cur, end);
2636 extent_offset = cur - em->start;
2637 BUG_ON(extent_map_end(em) <= cur);
2640 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2641 this_bio_flag = EXTENT_BIO_COMPRESSED;
2642 extent_set_compress_type(&this_bio_flag,
2646 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2647 cur_end = min(extent_map_end(em) - 1, end);
2648 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2649 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2650 disk_io_size = em->block_len;
2651 sector = em->block_start >> 9;
2653 sector = (em->block_start + extent_offset) >> 9;
2654 disk_io_size = iosize;
2657 block_start = em->block_start;
2658 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2659 block_start = EXTENT_MAP_HOLE;
2660 free_extent_map(em);
2663 /* we've found a hole, just zero and go on */
2664 if (block_start == EXTENT_MAP_HOLE) {
2666 struct extent_state *cached = NULL;
2668 userpage = kmap_atomic(page);
2669 memset(userpage + pg_offset, 0, iosize);
2670 flush_dcache_page(page);
2671 kunmap_atomic(userpage);
2673 set_extent_uptodate(tree, cur, cur + iosize - 1,
2675 unlock_extent_cached(tree, cur, cur + iosize - 1,
2678 pg_offset += iosize;
2681 /* the get_extent function already copied into the page */
2682 if (test_range_bit(tree, cur, cur_end,
2683 EXTENT_UPTODATE, 1, NULL)) {
2684 check_page_uptodate(tree, page);
2685 unlock_extent(tree, cur, cur + iosize - 1);
2687 pg_offset += iosize;
2690 /* we have an inline extent but it didn't get marked up
2691 * to date. Error out
2693 if (block_start == EXTENT_MAP_INLINE) {
2695 unlock_extent(tree, cur, cur + iosize - 1);
2697 pg_offset += iosize;
2702 if (tree->ops && tree->ops->readpage_io_hook) {
2703 ret = tree->ops->readpage_io_hook(page, cur,
2707 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2709 ret = submit_extent_page(READ, tree, page,
2710 sector, disk_io_size, pg_offset,
2712 end_bio_extent_readpage, mirror_num,
2715 BUG_ON(ret == -ENOMEM);
2717 *bio_flags = this_bio_flag;
2722 pg_offset += iosize;
2726 if (!PageError(page))
2727 SetPageUptodate(page);
2733 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2734 get_extent_t *get_extent, int mirror_num)
2736 struct bio *bio = NULL;
2737 unsigned long bio_flags = 0;
2740 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2743 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2747 static noinline void update_nr_written(struct page *page,
2748 struct writeback_control *wbc,
2749 unsigned long nr_written)
2751 wbc->nr_to_write -= nr_written;
2752 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2753 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2754 page->mapping->writeback_index = page->index + nr_written;
2758 * the writepage semantics are similar to regular writepage. extent
2759 * records are inserted to lock ranges in the tree, and as dirty areas
2760 * are found, they are marked writeback. Then the lock bits are removed
2761 * and the end_io handler clears the writeback ranges
2763 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2766 struct inode *inode = page->mapping->host;
2767 struct extent_page_data *epd = data;
2768 struct extent_io_tree *tree = epd->tree;
2769 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2771 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2775 u64 last_byte = i_size_read(inode);
2779 struct extent_state *cached_state = NULL;
2780 struct extent_map *em;
2781 struct block_device *bdev;
2784 size_t pg_offset = 0;
2786 loff_t i_size = i_size_read(inode);
2787 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2793 unsigned long nr_written = 0;
2794 bool fill_delalloc = true;
2796 if (wbc->sync_mode == WB_SYNC_ALL)
2797 write_flags = WRITE_SYNC;
2799 write_flags = WRITE;
2801 trace___extent_writepage(page, inode, wbc);
2803 WARN_ON(!PageLocked(page));
2805 ClearPageError(page);
2807 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2808 if (page->index > end_index ||
2809 (page->index == end_index && !pg_offset)) {
2810 page->mapping->a_ops->invalidatepage(page, 0);
2815 if (page->index == end_index) {
2818 userpage = kmap_atomic(page);
2819 memset(userpage + pg_offset, 0,
2820 PAGE_CACHE_SIZE - pg_offset);
2821 kunmap_atomic(userpage);
2822 flush_dcache_page(page);
2826 set_page_extent_mapped(page);
2828 if (!tree->ops || !tree->ops->fill_delalloc)
2829 fill_delalloc = false;
2831 delalloc_start = start;
2834 if (!epd->extent_locked && fill_delalloc) {
2835 u64 delalloc_to_write = 0;
2837 * make sure the wbc mapping index is at least updated
2840 update_nr_written(page, wbc, 0);
2842 while (delalloc_end < page_end) {
2843 nr_delalloc = find_lock_delalloc_range(inode, tree,
2848 if (nr_delalloc == 0) {
2849 delalloc_start = delalloc_end + 1;
2852 ret = tree->ops->fill_delalloc(inode, page,
2857 /* File system has been set read-only */
2863 * delalloc_end is already one less than the total
2864 * length, so we don't subtract one from
2867 delalloc_to_write += (delalloc_end - delalloc_start +
2870 delalloc_start = delalloc_end + 1;
2872 if (wbc->nr_to_write < delalloc_to_write) {
2875 if (delalloc_to_write < thresh * 2)
2876 thresh = delalloc_to_write;
2877 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2881 /* did the fill delalloc function already unlock and start
2887 * we've unlocked the page, so we can't update
2888 * the mapping's writeback index, just update
2891 wbc->nr_to_write -= nr_written;
2895 if (tree->ops && tree->ops->writepage_start_hook) {
2896 ret = tree->ops->writepage_start_hook(page, start,
2899 /* Fixup worker will requeue */
2901 wbc->pages_skipped++;
2903 redirty_page_for_writepage(wbc, page);
2904 update_nr_written(page, wbc, nr_written);
2912 * we don't want to touch the inode after unlocking the page,
2913 * so we update the mapping writeback index now
2915 update_nr_written(page, wbc, nr_written + 1);
2918 if (last_byte <= start) {
2919 if (tree->ops && tree->ops->writepage_end_io_hook)
2920 tree->ops->writepage_end_io_hook(page, start,
2925 blocksize = inode->i_sb->s_blocksize;
2927 while (cur <= end) {
2928 if (cur >= last_byte) {
2929 if (tree->ops && tree->ops->writepage_end_io_hook)
2930 tree->ops->writepage_end_io_hook(page, cur,
2934 em = epd->get_extent(inode, page, pg_offset, cur,
2936 if (IS_ERR_OR_NULL(em)) {
2941 extent_offset = cur - em->start;
2942 BUG_ON(extent_map_end(em) <= cur);
2944 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2945 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2946 sector = (em->block_start + extent_offset) >> 9;
2948 block_start = em->block_start;
2949 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2950 free_extent_map(em);
2954 * compressed and inline extents are written through other
2957 if (compressed || block_start == EXTENT_MAP_HOLE ||
2958 block_start == EXTENT_MAP_INLINE) {
2960 * end_io notification does not happen here for
2961 * compressed extents
2963 if (!compressed && tree->ops &&
2964 tree->ops->writepage_end_io_hook)
2965 tree->ops->writepage_end_io_hook(page, cur,
2968 else if (compressed) {
2969 /* we don't want to end_page_writeback on
2970 * a compressed extent. this happens
2977 pg_offset += iosize;
2980 /* leave this out until we have a page_mkwrite call */
2981 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2982 EXTENT_DIRTY, 0, NULL)) {
2984 pg_offset += iosize;
2988 if (tree->ops && tree->ops->writepage_io_hook) {
2989 ret = tree->ops->writepage_io_hook(page, cur,
2997 unsigned long max_nr = end_index + 1;
2999 set_range_writeback(tree, cur, cur + iosize - 1);
3000 if (!PageWriteback(page)) {
3001 printk(KERN_ERR "btrfs warning page %lu not "
3002 "writeback, cur %llu end %llu\n",
3003 page->index, (unsigned long long)cur,
3004 (unsigned long long)end);
3007 ret = submit_extent_page(write_flags, tree, page,
3008 sector, iosize, pg_offset,
3009 bdev, &epd->bio, max_nr,
3010 end_bio_extent_writepage,
3016 pg_offset += iosize;
3021 /* make sure the mapping tag for page dirty gets cleared */
3022 set_page_writeback(page);
3023 end_page_writeback(page);
3029 /* drop our reference on any cached states */
3030 free_extent_state(cached_state);
3034 static int eb_wait(void *word)
3040 static void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3042 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3043 TASK_UNINTERRUPTIBLE);
3046 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3047 struct btrfs_fs_info *fs_info,
3048 struct extent_page_data *epd)
3050 unsigned long i, num_pages;
3054 if (!btrfs_try_tree_write_lock(eb)) {
3056 flush_write_bio(epd);
3057 btrfs_tree_lock(eb);
3060 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3061 btrfs_tree_unlock(eb);
3065 flush_write_bio(epd);
3069 wait_on_extent_buffer_writeback(eb);
3070 btrfs_tree_lock(eb);
3071 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3073 btrfs_tree_unlock(eb);
3078 * We need to do this to prevent races in people who check if the eb is
3079 * under IO since we can end up having no IO bits set for a short period
3082 spin_lock(&eb->refs_lock);
3083 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3084 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3085 spin_unlock(&eb->refs_lock);
3086 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3087 spin_lock(&fs_info->delalloc_lock);
3088 if (fs_info->dirty_metadata_bytes >= eb->len)
3089 fs_info->dirty_metadata_bytes -= eb->len;
3092 spin_unlock(&fs_info->delalloc_lock);
3095 spin_unlock(&eb->refs_lock);
3098 btrfs_tree_unlock(eb);
3103 num_pages = num_extent_pages(eb->start, eb->len);
3104 for (i = 0; i < num_pages; i++) {
3105 struct page *p = extent_buffer_page(eb, i);
3107 if (!trylock_page(p)) {
3109 flush_write_bio(epd);
3119 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3121 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3122 smp_mb__after_clear_bit();
3123 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3126 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3128 int uptodate = err == 0;
3129 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3130 struct extent_buffer *eb;
3134 struct page *page = bvec->bv_page;
3137 eb = (struct extent_buffer *)page->private;
3139 done = atomic_dec_and_test(&eb->io_pages);
3141 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3142 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3143 ClearPageUptodate(page);
3147 end_page_writeback(page);
3152 end_extent_buffer_writeback(eb);
3153 } while (bvec >= bio->bi_io_vec);
3159 static int write_one_eb(struct extent_buffer *eb,
3160 struct btrfs_fs_info *fs_info,
3161 struct writeback_control *wbc,
3162 struct extent_page_data *epd)
3164 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3165 u64 offset = eb->start;
3166 unsigned long i, num_pages;
3167 unsigned long bio_flags = 0;
3168 int rw = (epd->sync_io ? WRITE_SYNC : WRITE);
3171 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3172 num_pages = num_extent_pages(eb->start, eb->len);
3173 atomic_set(&eb->io_pages, num_pages);
3174 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3175 bio_flags = EXTENT_BIO_TREE_LOG;
3177 for (i = 0; i < num_pages; i++) {
3178 struct page *p = extent_buffer_page(eb, i);
3180 clear_page_dirty_for_io(p);
3181 set_page_writeback(p);
3182 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3183 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3184 -1, end_bio_extent_buffer_writepage,
3185 0, epd->bio_flags, bio_flags);
3186 epd->bio_flags = bio_flags;
3188 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3190 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3191 end_extent_buffer_writeback(eb);
3195 offset += PAGE_CACHE_SIZE;
3196 update_nr_written(p, wbc, 1);
3200 if (unlikely(ret)) {
3201 for (; i < num_pages; i++) {
3202 struct page *p = extent_buffer_page(eb, i);
3210 int btree_write_cache_pages(struct address_space *mapping,
3211 struct writeback_control *wbc)
3213 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3214 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3215 struct extent_buffer *eb, *prev_eb = NULL;
3216 struct extent_page_data epd = {
3220 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3225 int nr_to_write_done = 0;
3226 struct pagevec pvec;
3229 pgoff_t end; /* Inclusive */
3233 pagevec_init(&pvec, 0);
3234 if (wbc->range_cyclic) {
3235 index = mapping->writeback_index; /* Start from prev offset */
3238 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3239 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3242 if (wbc->sync_mode == WB_SYNC_ALL)
3243 tag = PAGECACHE_TAG_TOWRITE;
3245 tag = PAGECACHE_TAG_DIRTY;
3247 if (wbc->sync_mode == WB_SYNC_ALL)
3248 tag_pages_for_writeback(mapping, index, end);
3249 while (!done && !nr_to_write_done && (index <= end) &&
3250 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3251 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3255 for (i = 0; i < nr_pages; i++) {
3256 struct page *page = pvec.pages[i];
3258 if (!PagePrivate(page))
3261 if (!wbc->range_cyclic && page->index > end) {
3266 spin_lock(&mapping->private_lock);
3267 if (!PagePrivate(page)) {
3268 spin_unlock(&mapping->private_lock);
3272 eb = (struct extent_buffer *)page->private;
3275 * Shouldn't happen and normally this would be a BUG_ON
3276 * but no sense in crashing the users box for something
3277 * we can survive anyway.
3280 spin_unlock(&mapping->private_lock);
3285 if (eb == prev_eb) {
3286 spin_unlock(&mapping->private_lock);
3290 ret = atomic_inc_not_zero(&eb->refs);
3291 spin_unlock(&mapping->private_lock);
3296 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3298 free_extent_buffer(eb);
3302 ret = write_one_eb(eb, fs_info, wbc, &epd);
3305 free_extent_buffer(eb);
3308 free_extent_buffer(eb);
3311 * the filesystem may choose to bump up nr_to_write.
3312 * We have to make sure to honor the new nr_to_write
3315 nr_to_write_done = wbc->nr_to_write <= 0;
3317 pagevec_release(&pvec);
3320 if (!scanned && !done) {
3322 * We hit the last page and there is more work to be done: wrap
3323 * back to the start of the file
3329 flush_write_bio(&epd);
3334 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3335 * @mapping: address space structure to write
3336 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3337 * @writepage: function called for each page
3338 * @data: data passed to writepage function
3340 * If a page is already under I/O, write_cache_pages() skips it, even
3341 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3342 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3343 * and msync() need to guarantee that all the data which was dirty at the time
3344 * the call was made get new I/O started against them. If wbc->sync_mode is
3345 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3346 * existing IO to complete.
3348 static int extent_write_cache_pages(struct extent_io_tree *tree,
3349 struct address_space *mapping,
3350 struct writeback_control *wbc,
3351 writepage_t writepage, void *data,
3352 void (*flush_fn)(void *))
3354 struct inode *inode = mapping->host;
3357 int nr_to_write_done = 0;
3358 struct pagevec pvec;
3361 pgoff_t end; /* Inclusive */
3366 * We have to hold onto the inode so that ordered extents can do their
3367 * work when the IO finishes. The alternative to this is failing to add
3368 * an ordered extent if the igrab() fails there and that is a huge pain
3369 * to deal with, so instead just hold onto the inode throughout the
3370 * writepages operation. If it fails here we are freeing up the inode
3371 * anyway and we'd rather not waste our time writing out stuff that is
3372 * going to be truncated anyway.
3377 pagevec_init(&pvec, 0);
3378 if (wbc->range_cyclic) {
3379 index = mapping->writeback_index; /* Start from prev offset */
3382 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3383 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3386 if (wbc->sync_mode == WB_SYNC_ALL)
3387 tag = PAGECACHE_TAG_TOWRITE;
3389 tag = PAGECACHE_TAG_DIRTY;
3391 if (wbc->sync_mode == WB_SYNC_ALL)
3392 tag_pages_for_writeback(mapping, index, end);
3393 while (!done && !nr_to_write_done && (index <= end) &&
3394 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3395 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3399 for (i = 0; i < nr_pages; i++) {
3400 struct page *page = pvec.pages[i];
3403 * At this point we hold neither mapping->tree_lock nor
3404 * lock on the page itself: the page may be truncated or
3405 * invalidated (changing page->mapping to NULL), or even
3406 * swizzled back from swapper_space to tmpfs file
3410 tree->ops->write_cache_pages_lock_hook) {
3411 tree->ops->write_cache_pages_lock_hook(page,
3414 if (!trylock_page(page)) {
3420 if (unlikely(page->mapping != mapping)) {
3425 if (!wbc->range_cyclic && page->index > end) {
3431 if (wbc->sync_mode != WB_SYNC_NONE) {
3432 if (PageWriteback(page))
3434 wait_on_page_writeback(page);
3437 if (PageWriteback(page) ||
3438 !clear_page_dirty_for_io(page)) {
3443 ret = (*writepage)(page, wbc, data);
3445 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3453 * the filesystem may choose to bump up nr_to_write.
3454 * We have to make sure to honor the new nr_to_write
3457 nr_to_write_done = wbc->nr_to_write <= 0;
3459 pagevec_release(&pvec);
3462 if (!scanned && !done) {
3464 * We hit the last page and there is more work to be done: wrap
3465 * back to the start of the file
3471 btrfs_add_delayed_iput(inode);
3475 static void flush_epd_write_bio(struct extent_page_data *epd)
3484 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3485 BUG_ON(ret < 0); /* -ENOMEM */
3490 static noinline void flush_write_bio(void *data)
3492 struct extent_page_data *epd = data;
3493 flush_epd_write_bio(epd);
3496 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3497 get_extent_t *get_extent,
3498 struct writeback_control *wbc)
3501 struct extent_page_data epd = {
3504 .get_extent = get_extent,
3506 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3510 ret = __extent_writepage(page, wbc, &epd);
3512 flush_epd_write_bio(&epd);
3516 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3517 u64 start, u64 end, get_extent_t *get_extent,
3521 struct address_space *mapping = inode->i_mapping;
3523 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3526 struct extent_page_data epd = {
3529 .get_extent = get_extent,
3531 .sync_io = mode == WB_SYNC_ALL,
3534 struct writeback_control wbc_writepages = {
3536 .nr_to_write = nr_pages * 2,
3537 .range_start = start,
3538 .range_end = end + 1,
3541 while (start <= end) {
3542 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3543 if (clear_page_dirty_for_io(page))
3544 ret = __extent_writepage(page, &wbc_writepages, &epd);
3546 if (tree->ops && tree->ops->writepage_end_io_hook)
3547 tree->ops->writepage_end_io_hook(page, start,
3548 start + PAGE_CACHE_SIZE - 1,
3552 page_cache_release(page);
3553 start += PAGE_CACHE_SIZE;
3556 flush_epd_write_bio(&epd);
3560 int extent_writepages(struct extent_io_tree *tree,
3561 struct address_space *mapping,
3562 get_extent_t *get_extent,
3563 struct writeback_control *wbc)
3566 struct extent_page_data epd = {
3569 .get_extent = get_extent,
3571 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3575 ret = extent_write_cache_pages(tree, mapping, wbc,
3576 __extent_writepage, &epd,
3578 flush_epd_write_bio(&epd);
3582 int extent_readpages(struct extent_io_tree *tree,
3583 struct address_space *mapping,
3584 struct list_head *pages, unsigned nr_pages,
3585 get_extent_t get_extent)
3587 struct bio *bio = NULL;
3589 unsigned long bio_flags = 0;
3590 struct page *pagepool[16];
3595 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3596 page = list_entry(pages->prev, struct page, lru);
3598 prefetchw(&page->flags);
3599 list_del(&page->lru);
3600 if (add_to_page_cache_lru(page, mapping,
3601 page->index, GFP_NOFS)) {
3602 page_cache_release(page);
3606 pagepool[nr++] = page;
3607 if (nr < ARRAY_SIZE(pagepool))
3609 for (i = 0; i < nr; i++) {
3610 __extent_read_full_page(tree, pagepool[i], get_extent,
3611 &bio, 0, &bio_flags);
3612 page_cache_release(pagepool[i]);
3616 for (i = 0; i < nr; i++) {
3617 __extent_read_full_page(tree, pagepool[i], get_extent,
3618 &bio, 0, &bio_flags);
3619 page_cache_release(pagepool[i]);
3622 BUG_ON(!list_empty(pages));
3624 return submit_one_bio(READ, bio, 0, bio_flags);
3629 * basic invalidatepage code, this waits on any locked or writeback
3630 * ranges corresponding to the page, and then deletes any extent state
3631 * records from the tree
3633 int extent_invalidatepage(struct extent_io_tree *tree,
3634 struct page *page, unsigned long offset)
3636 struct extent_state *cached_state = NULL;
3637 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3638 u64 end = start + PAGE_CACHE_SIZE - 1;
3639 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3641 start += (offset + blocksize - 1) & ~(blocksize - 1);
3645 lock_extent_bits(tree, start, end, 0, &cached_state);
3646 wait_on_page_writeback(page);
3647 clear_extent_bit(tree, start, end,
3648 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3649 EXTENT_DO_ACCOUNTING,
3650 1, 1, &cached_state, GFP_NOFS);
3655 * a helper for releasepage, this tests for areas of the page that
3656 * are locked or under IO and drops the related state bits if it is safe
3659 int try_release_extent_state(struct extent_map_tree *map,
3660 struct extent_io_tree *tree, struct page *page,
3663 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3664 u64 end = start + PAGE_CACHE_SIZE - 1;
3667 if (test_range_bit(tree, start, end,
3668 EXTENT_IOBITS, 0, NULL))
3671 if ((mask & GFP_NOFS) == GFP_NOFS)
3674 * at this point we can safely clear everything except the
3675 * locked bit and the nodatasum bit
3677 ret = clear_extent_bit(tree, start, end,
3678 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3681 /* if clear_extent_bit failed for enomem reasons,
3682 * we can't allow the release to continue.
3693 * a helper for releasepage. As long as there are no locked extents
3694 * in the range corresponding to the page, both state records and extent
3695 * map records are removed
3697 int try_release_extent_mapping(struct extent_map_tree *map,
3698 struct extent_io_tree *tree, struct page *page,
3701 struct extent_map *em;
3702 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3703 u64 end = start + PAGE_CACHE_SIZE - 1;
3705 if ((mask & __GFP_WAIT) &&
3706 page->mapping->host->i_size > 16 * 1024 * 1024) {
3708 while (start <= end) {
3709 len = end - start + 1;
3710 write_lock(&map->lock);
3711 em = lookup_extent_mapping(map, start, len);
3713 write_unlock(&map->lock);
3716 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3717 em->start != start) {
3718 write_unlock(&map->lock);
3719 free_extent_map(em);
3722 if (!test_range_bit(tree, em->start,
3723 extent_map_end(em) - 1,
3724 EXTENT_LOCKED | EXTENT_WRITEBACK,
3726 remove_extent_mapping(map, em);
3727 /* once for the rb tree */
3728 free_extent_map(em);
3730 start = extent_map_end(em);
3731 write_unlock(&map->lock);
3734 free_extent_map(em);
3737 return try_release_extent_state(map, tree, page, mask);
3741 * helper function for fiemap, which doesn't want to see any holes.
3742 * This maps until we find something past 'last'
3744 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3747 get_extent_t *get_extent)
3749 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3750 struct extent_map *em;
3757 len = last - offset;
3760 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3761 em = get_extent(inode, NULL, 0, offset, len, 0);
3762 if (IS_ERR_OR_NULL(em))
3765 /* if this isn't a hole return it */
3766 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3767 em->block_start != EXTENT_MAP_HOLE) {
3771 /* this is a hole, advance to the next extent */
3772 offset = extent_map_end(em);
3773 free_extent_map(em);
3780 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3781 __u64 start, __u64 len, get_extent_t *get_extent)
3785 u64 max = start + len;
3789 u64 last_for_get_extent = 0;
3791 u64 isize = i_size_read(inode);
3792 struct btrfs_key found_key;
3793 struct extent_map *em = NULL;
3794 struct extent_state *cached_state = NULL;
3795 struct btrfs_path *path;
3796 struct btrfs_file_extent_item *item;
3801 unsigned long emflags;
3806 path = btrfs_alloc_path();
3809 path->leave_spinning = 1;
3811 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3812 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3815 * lookup the last file extent. We're not using i_size here
3816 * because there might be preallocation past i_size
3818 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3819 path, btrfs_ino(inode), -1, 0);
3821 btrfs_free_path(path);
3826 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3827 struct btrfs_file_extent_item);
3828 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3829 found_type = btrfs_key_type(&found_key);
3831 /* No extents, but there might be delalloc bits */
3832 if (found_key.objectid != btrfs_ino(inode) ||
3833 found_type != BTRFS_EXTENT_DATA_KEY) {
3834 /* have to trust i_size as the end */
3836 last_for_get_extent = isize;
3839 * remember the start of the last extent. There are a
3840 * bunch of different factors that go into the length of the
3841 * extent, so its much less complex to remember where it started
3843 last = found_key.offset;
3844 last_for_get_extent = last + 1;
3846 btrfs_free_path(path);
3849 * we might have some extents allocated but more delalloc past those
3850 * extents. so, we trust isize unless the start of the last extent is
3855 last_for_get_extent = isize;
3858 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3861 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3871 u64 offset_in_extent;
3873 /* break if the extent we found is outside the range */
3874 if (em->start >= max || extent_map_end(em) < off)
3878 * get_extent may return an extent that starts before our
3879 * requested range. We have to make sure the ranges
3880 * we return to fiemap always move forward and don't
3881 * overlap, so adjust the offsets here
3883 em_start = max(em->start, off);
3886 * record the offset from the start of the extent
3887 * for adjusting the disk offset below
3889 offset_in_extent = em_start - em->start;
3890 em_end = extent_map_end(em);
3891 em_len = em_end - em_start;
3892 emflags = em->flags;
3897 * bump off for our next call to get_extent
3899 off = extent_map_end(em);
3903 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3905 flags |= FIEMAP_EXTENT_LAST;
3906 } else if (em->block_start == EXTENT_MAP_INLINE) {
3907 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3908 FIEMAP_EXTENT_NOT_ALIGNED);
3909 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3910 flags |= (FIEMAP_EXTENT_DELALLOC |
3911 FIEMAP_EXTENT_UNKNOWN);
3913 disko = em->block_start + offset_in_extent;
3915 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3916 flags |= FIEMAP_EXTENT_ENCODED;
3918 free_extent_map(em);
3920 if ((em_start >= last) || em_len == (u64)-1 ||
3921 (last == (u64)-1 && isize <= em_end)) {
3922 flags |= FIEMAP_EXTENT_LAST;
3926 /* now scan forward to see if this is really the last extent. */
3927 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3934 flags |= FIEMAP_EXTENT_LAST;
3937 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3943 free_extent_map(em);
3945 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3946 &cached_state, GFP_NOFS);
3950 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3953 return eb->pages[i];
3956 inline unsigned long num_extent_pages(u64 start, u64 len)
3958 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3959 (start >> PAGE_CACHE_SHIFT);
3962 static void __free_extent_buffer(struct extent_buffer *eb)
3965 unsigned long flags;
3966 spin_lock_irqsave(&leak_lock, flags);
3967 list_del(&eb->leak_list);
3968 spin_unlock_irqrestore(&leak_lock, flags);
3970 if (eb->pages && eb->pages != eb->inline_pages)
3972 kmem_cache_free(extent_buffer_cache, eb);
3975 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3980 struct extent_buffer *eb = NULL;
3982 unsigned long flags;
3985 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3992 rwlock_init(&eb->lock);
3993 atomic_set(&eb->write_locks, 0);
3994 atomic_set(&eb->read_locks, 0);
3995 atomic_set(&eb->blocking_readers, 0);
3996 atomic_set(&eb->blocking_writers, 0);
3997 atomic_set(&eb->spinning_readers, 0);
3998 atomic_set(&eb->spinning_writers, 0);
3999 eb->lock_nested = 0;
4000 init_waitqueue_head(&eb->write_lock_wq);
4001 init_waitqueue_head(&eb->read_lock_wq);
4004 spin_lock_irqsave(&leak_lock, flags);
4005 list_add(&eb->leak_list, &buffers);
4006 spin_unlock_irqrestore(&leak_lock, flags);
4008 spin_lock_init(&eb->refs_lock);
4009 atomic_set(&eb->refs, 1);
4010 atomic_set(&eb->io_pages, 0);
4012 if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) {
4013 struct page **pages;
4014 int num_pages = (len + PAGE_CACHE_SIZE - 1) >>
4016 pages = kzalloc(num_pages, mask);
4018 __free_extent_buffer(eb);
4023 eb->pages = eb->inline_pages;
4029 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4033 struct extent_buffer *new;
4034 unsigned long num_pages = num_extent_pages(src->start, src->len);
4036 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4040 for (i = 0; i < num_pages; i++) {
4041 p = alloc_page(GFP_ATOMIC);
4043 attach_extent_buffer_page(new, p);
4044 WARN_ON(PageDirty(p));
4049 copy_extent_buffer(new, src, 0, 0, src->len);
4050 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4051 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4056 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4058 struct extent_buffer *eb;
4059 unsigned long num_pages = num_extent_pages(0, len);
4062 eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4066 for (i = 0; i < num_pages; i++) {
4067 eb->pages[i] = alloc_page(GFP_ATOMIC);
4071 set_extent_buffer_uptodate(eb);
4072 btrfs_set_header_nritems(eb, 0);
4073 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4077 for (i--; i > 0; i--)
4078 __free_page(eb->pages[i]);
4079 __free_extent_buffer(eb);
4083 static int extent_buffer_under_io(struct extent_buffer *eb)
4085 return (atomic_read(&eb->io_pages) ||
4086 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4087 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4091 * Helper for releasing extent buffer page.
4093 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4094 unsigned long start_idx)
4096 unsigned long index;
4097 unsigned long num_pages;
4099 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4101 BUG_ON(extent_buffer_under_io(eb));
4103 num_pages = num_extent_pages(eb->start, eb->len);
4104 index = start_idx + num_pages;
4105 if (start_idx >= index)
4110 page = extent_buffer_page(eb, index);
4111 if (page && mapped) {
4112 spin_lock(&page->mapping->private_lock);
4114 * We do this since we'll remove the pages after we've
4115 * removed the eb from the radix tree, so we could race
4116 * and have this page now attached to the new eb. So
4117 * only clear page_private if it's still connected to
4120 if (PagePrivate(page) &&
4121 page->private == (unsigned long)eb) {
4122 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4123 BUG_ON(PageDirty(page));
4124 BUG_ON(PageWriteback(page));
4126 * We need to make sure we haven't be attached
4129 ClearPagePrivate(page);
4130 set_page_private(page, 0);
4131 /* One for the page private */
4132 page_cache_release(page);
4134 spin_unlock(&page->mapping->private_lock);
4138 /* One for when we alloced the page */
4139 page_cache_release(page);
4141 } while (index != start_idx);
4145 * Helper for releasing the extent buffer.
4147 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4149 btrfs_release_extent_buffer_page(eb, 0);
4150 __free_extent_buffer(eb);
4153 static void check_buffer_tree_ref(struct extent_buffer *eb)
4155 /* the ref bit is tricky. We have to make sure it is set
4156 * if we have the buffer dirty. Otherwise the
4157 * code to free a buffer can end up dropping a dirty
4160 * Once the ref bit is set, it won't go away while the
4161 * buffer is dirty or in writeback, and it also won't
4162 * go away while we have the reference count on the
4165 * We can't just set the ref bit without bumping the
4166 * ref on the eb because free_extent_buffer might
4167 * see the ref bit and try to clear it. If this happens
4168 * free_extent_buffer might end up dropping our original
4169 * ref by mistake and freeing the page before we are able
4170 * to add one more ref.
4172 * So bump the ref count first, then set the bit. If someone
4173 * beat us to it, drop the ref we added.
4175 spin_lock(&eb->refs_lock);
4176 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4177 atomic_inc(&eb->refs);
4178 spin_unlock(&eb->refs_lock);
4181 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4183 unsigned long num_pages, i;
4185 check_buffer_tree_ref(eb);
4187 num_pages = num_extent_pages(eb->start, eb->len);
4188 for (i = 0; i < num_pages; i++) {
4189 struct page *p = extent_buffer_page(eb, i);
4190 mark_page_accessed(p);
4194 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4195 u64 start, unsigned long len)
4197 unsigned long num_pages = num_extent_pages(start, len);
4199 unsigned long index = start >> PAGE_CACHE_SHIFT;
4200 struct extent_buffer *eb;
4201 struct extent_buffer *exists = NULL;
4203 struct address_space *mapping = tree->mapping;
4208 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4209 if (eb && atomic_inc_not_zero(&eb->refs)) {
4211 mark_extent_buffer_accessed(eb);
4216 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4220 for (i = 0; i < num_pages; i++, index++) {
4221 p = find_or_create_page(mapping, index, GFP_NOFS);
4227 spin_lock(&mapping->private_lock);
4228 if (PagePrivate(p)) {
4230 * We could have already allocated an eb for this page
4231 * and attached one so lets see if we can get a ref on
4232 * the existing eb, and if we can we know it's good and
4233 * we can just return that one, else we know we can just
4234 * overwrite page->private.
4236 exists = (struct extent_buffer *)p->private;
4237 if (atomic_inc_not_zero(&exists->refs)) {
4238 spin_unlock(&mapping->private_lock);
4240 page_cache_release(p);
4241 mark_extent_buffer_accessed(exists);
4246 * Do this so attach doesn't complain and we need to
4247 * drop the ref the old guy had.
4249 ClearPagePrivate(p);
4250 WARN_ON(PageDirty(p));
4251 page_cache_release(p);
4253 attach_extent_buffer_page(eb, p);
4254 spin_unlock(&mapping->private_lock);
4255 WARN_ON(PageDirty(p));
4256 mark_page_accessed(p);
4258 if (!PageUptodate(p))
4262 * see below about how we avoid a nasty race with release page
4263 * and why we unlock later
4267 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4269 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4273 spin_lock(&tree->buffer_lock);
4274 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4275 if (ret == -EEXIST) {
4276 exists = radix_tree_lookup(&tree->buffer,
4277 start >> PAGE_CACHE_SHIFT);
4278 if (!atomic_inc_not_zero(&exists->refs)) {
4279 spin_unlock(&tree->buffer_lock);
4280 radix_tree_preload_end();
4284 spin_unlock(&tree->buffer_lock);
4285 radix_tree_preload_end();
4286 mark_extent_buffer_accessed(exists);
4289 /* add one reference for the tree */
4290 check_buffer_tree_ref(eb);
4291 spin_unlock(&tree->buffer_lock);
4292 radix_tree_preload_end();
4295 * there is a race where release page may have
4296 * tried to find this extent buffer in the radix
4297 * but failed. It will tell the VM it is safe to
4298 * reclaim the, and it will clear the page private bit.
4299 * We must make sure to set the page private bit properly
4300 * after the extent buffer is in the radix tree so
4301 * it doesn't get lost
4303 SetPageChecked(eb->pages[0]);
4304 for (i = 1; i < num_pages; i++) {
4305 p = extent_buffer_page(eb, i);
4306 ClearPageChecked(p);
4309 unlock_page(eb->pages[0]);
4313 for (i = 0; i < num_pages; i++) {
4315 unlock_page(eb->pages[i]);
4318 WARN_ON(!atomic_dec_and_test(&eb->refs));
4319 btrfs_release_extent_buffer(eb);
4323 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4324 u64 start, unsigned long len)
4326 struct extent_buffer *eb;
4329 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4330 if (eb && atomic_inc_not_zero(&eb->refs)) {
4332 mark_extent_buffer_accessed(eb);
4340 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4342 struct extent_buffer *eb =
4343 container_of(head, struct extent_buffer, rcu_head);
4345 __free_extent_buffer(eb);
4348 /* Expects to have eb->eb_lock already held */
4349 static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask)
4351 WARN_ON(atomic_read(&eb->refs) == 0);
4352 if (atomic_dec_and_test(&eb->refs)) {
4353 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4354 spin_unlock(&eb->refs_lock);
4356 struct extent_io_tree *tree = eb->tree;
4358 spin_unlock(&eb->refs_lock);
4360 spin_lock(&tree->buffer_lock);
4361 radix_tree_delete(&tree->buffer,
4362 eb->start >> PAGE_CACHE_SHIFT);
4363 spin_unlock(&tree->buffer_lock);
4366 /* Should be safe to release our pages at this point */
4367 btrfs_release_extent_buffer_page(eb, 0);
4368 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4371 spin_unlock(&eb->refs_lock);
4376 void free_extent_buffer(struct extent_buffer *eb)
4381 spin_lock(&eb->refs_lock);
4382 if (atomic_read(&eb->refs) == 2 &&
4383 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4384 atomic_dec(&eb->refs);
4386 if (atomic_read(&eb->refs) == 2 &&
4387 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4388 !extent_buffer_under_io(eb) &&
4389 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4390 atomic_dec(&eb->refs);
4393 * I know this is terrible, but it's temporary until we stop tracking
4394 * the uptodate bits and such for the extent buffers.
4396 release_extent_buffer(eb, GFP_ATOMIC);
4399 void free_extent_buffer_stale(struct extent_buffer *eb)
4404 spin_lock(&eb->refs_lock);
4405 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4407 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4408 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4409 atomic_dec(&eb->refs);
4410 release_extent_buffer(eb, GFP_NOFS);
4413 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4416 unsigned long num_pages;
4419 num_pages = num_extent_pages(eb->start, eb->len);
4421 for (i = 0; i < num_pages; i++) {
4422 page = extent_buffer_page(eb, i);
4423 if (!PageDirty(page))
4427 WARN_ON(!PagePrivate(page));
4429 clear_page_dirty_for_io(page);
4430 spin_lock_irq(&page->mapping->tree_lock);
4431 if (!PageDirty(page)) {
4432 radix_tree_tag_clear(&page->mapping->page_tree,
4434 PAGECACHE_TAG_DIRTY);
4436 spin_unlock_irq(&page->mapping->tree_lock);
4437 ClearPageError(page);
4440 WARN_ON(atomic_read(&eb->refs) == 0);
4443 int set_extent_buffer_dirty(struct extent_buffer *eb)
4446 unsigned long num_pages;
4449 check_buffer_tree_ref(eb);
4451 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4453 num_pages = num_extent_pages(eb->start, eb->len);
4454 WARN_ON(atomic_read(&eb->refs) == 0);
4455 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4457 for (i = 0; i < num_pages; i++)
4458 set_page_dirty(extent_buffer_page(eb, i));
4462 static int range_straddles_pages(u64 start, u64 len)
4464 if (len < PAGE_CACHE_SIZE)
4466 if (start & (PAGE_CACHE_SIZE - 1))
4468 if ((start + len) & (PAGE_CACHE_SIZE - 1))
4473 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4477 unsigned long num_pages;
4479 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4480 num_pages = num_extent_pages(eb->start, eb->len);
4481 for (i = 0; i < num_pages; i++) {
4482 page = extent_buffer_page(eb, i);
4484 ClearPageUptodate(page);
4489 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4493 unsigned long num_pages;
4495 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4496 num_pages = num_extent_pages(eb->start, eb->len);
4497 for (i = 0; i < num_pages; i++) {
4498 page = extent_buffer_page(eb, i);
4499 SetPageUptodate(page);
4504 int extent_range_uptodate(struct extent_io_tree *tree,
4509 int pg_uptodate = 1;
4511 unsigned long index;
4513 if (range_straddles_pages(start, end - start + 1)) {
4514 ret = test_range_bit(tree, start, end,
4515 EXTENT_UPTODATE, 1, NULL);
4519 while (start <= end) {
4520 index = start >> PAGE_CACHE_SHIFT;
4521 page = find_get_page(tree->mapping, index);
4524 uptodate = PageUptodate(page);
4525 page_cache_release(page);
4530 start += PAGE_CACHE_SIZE;
4535 int extent_buffer_uptodate(struct extent_buffer *eb)
4537 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4540 int read_extent_buffer_pages(struct extent_io_tree *tree,
4541 struct extent_buffer *eb, u64 start, int wait,
4542 get_extent_t *get_extent, int mirror_num)
4545 unsigned long start_i;
4549 int locked_pages = 0;
4550 int all_uptodate = 1;
4551 unsigned long num_pages;
4552 unsigned long num_reads = 0;
4553 struct bio *bio = NULL;
4554 unsigned long bio_flags = 0;
4556 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4560 WARN_ON(start < eb->start);
4561 start_i = (start >> PAGE_CACHE_SHIFT) -
4562 (eb->start >> PAGE_CACHE_SHIFT);
4567 num_pages = num_extent_pages(eb->start, eb->len);
4568 for (i = start_i; i < num_pages; i++) {
4569 page = extent_buffer_page(eb, i);
4570 if (wait == WAIT_NONE) {
4571 if (!trylock_page(page))
4577 if (!PageUptodate(page)) {
4584 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4588 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4589 eb->read_mirror = 0;
4590 atomic_set(&eb->io_pages, num_reads);
4591 for (i = start_i; i < num_pages; i++) {
4592 page = extent_buffer_page(eb, i);
4593 if (!PageUptodate(page)) {
4594 ClearPageError(page);
4595 err = __extent_read_full_page(tree, page,
4597 mirror_num, &bio_flags);
4606 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4611 if (ret || wait != WAIT_COMPLETE)
4614 for (i = start_i; i < num_pages; i++) {
4615 page = extent_buffer_page(eb, i);
4616 wait_on_page_locked(page);
4617 if (!PageUptodate(page))
4625 while (locked_pages > 0) {
4626 page = extent_buffer_page(eb, i);
4634 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4635 unsigned long start,
4642 char *dst = (char *)dstv;
4643 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4644 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4646 WARN_ON(start > eb->len);
4647 WARN_ON(start + len > eb->start + eb->len);
4649 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4652 page = extent_buffer_page(eb, i);
4654 cur = min(len, (PAGE_CACHE_SIZE - offset));
4655 kaddr = page_address(page);
4656 memcpy(dst, kaddr + offset, cur);
4665 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4666 unsigned long min_len, char **map,
4667 unsigned long *map_start,
4668 unsigned long *map_len)
4670 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4673 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4674 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4675 unsigned long end_i = (start_offset + start + min_len - 1) >>
4682 offset = start_offset;
4686 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4689 if (start + min_len > eb->len) {
4690 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4691 "wanted %lu %lu\n", (unsigned long long)eb->start,
4692 eb->len, start, min_len);
4697 p = extent_buffer_page(eb, i);
4698 kaddr = page_address(p);
4699 *map = kaddr + offset;
4700 *map_len = PAGE_CACHE_SIZE - offset;
4704 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4705 unsigned long start,
4712 char *ptr = (char *)ptrv;
4713 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4714 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4717 WARN_ON(start > eb->len);
4718 WARN_ON(start + len > eb->start + eb->len);
4720 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4723 page = extent_buffer_page(eb, i);
4725 cur = min(len, (PAGE_CACHE_SIZE - offset));
4727 kaddr = page_address(page);
4728 ret = memcmp(ptr, kaddr + offset, cur);
4740 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4741 unsigned long start, unsigned long len)
4747 char *src = (char *)srcv;
4748 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4749 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4751 WARN_ON(start > eb->len);
4752 WARN_ON(start + len > eb->start + eb->len);
4754 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4757 page = extent_buffer_page(eb, i);
4758 WARN_ON(!PageUptodate(page));
4760 cur = min(len, PAGE_CACHE_SIZE - offset);
4761 kaddr = page_address(page);
4762 memcpy(kaddr + offset, src, cur);
4771 void memset_extent_buffer(struct extent_buffer *eb, char c,
4772 unsigned long start, unsigned long len)
4778 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4779 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4781 WARN_ON(start > eb->len);
4782 WARN_ON(start + len > eb->start + eb->len);
4784 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4787 page = extent_buffer_page(eb, i);
4788 WARN_ON(!PageUptodate(page));
4790 cur = min(len, PAGE_CACHE_SIZE - offset);
4791 kaddr = page_address(page);
4792 memset(kaddr + offset, c, cur);
4800 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4801 unsigned long dst_offset, unsigned long src_offset,
4804 u64 dst_len = dst->len;
4809 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4810 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4812 WARN_ON(src->len != dst_len);
4814 offset = (start_offset + dst_offset) &
4815 ((unsigned long)PAGE_CACHE_SIZE - 1);
4818 page = extent_buffer_page(dst, i);
4819 WARN_ON(!PageUptodate(page));
4821 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4823 kaddr = page_address(page);
4824 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4833 static void move_pages(struct page *dst_page, struct page *src_page,
4834 unsigned long dst_off, unsigned long src_off,
4837 char *dst_kaddr = page_address(dst_page);
4838 if (dst_page == src_page) {
4839 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4841 char *src_kaddr = page_address(src_page);
4842 char *p = dst_kaddr + dst_off + len;
4843 char *s = src_kaddr + src_off + len;
4850 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4852 unsigned long distance = (src > dst) ? src - dst : dst - src;
4853 return distance < len;
4856 static void copy_pages(struct page *dst_page, struct page *src_page,
4857 unsigned long dst_off, unsigned long src_off,
4860 char *dst_kaddr = page_address(dst_page);
4862 int must_memmove = 0;
4864 if (dst_page != src_page) {
4865 src_kaddr = page_address(src_page);
4867 src_kaddr = dst_kaddr;
4868 if (areas_overlap(src_off, dst_off, len))
4873 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4875 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4878 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4879 unsigned long src_offset, unsigned long len)
4882 size_t dst_off_in_page;
4883 size_t src_off_in_page;
4884 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4885 unsigned long dst_i;
4886 unsigned long src_i;
4888 if (src_offset + len > dst->len) {
4889 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4890 "len %lu dst len %lu\n", src_offset, len, dst->len);
4893 if (dst_offset + len > dst->len) {
4894 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4895 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4900 dst_off_in_page = (start_offset + dst_offset) &
4901 ((unsigned long)PAGE_CACHE_SIZE - 1);
4902 src_off_in_page = (start_offset + src_offset) &
4903 ((unsigned long)PAGE_CACHE_SIZE - 1);
4905 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4906 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4908 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4910 cur = min_t(unsigned long, cur,
4911 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4913 copy_pages(extent_buffer_page(dst, dst_i),
4914 extent_buffer_page(dst, src_i),
4915 dst_off_in_page, src_off_in_page, cur);
4923 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4924 unsigned long src_offset, unsigned long len)
4927 size_t dst_off_in_page;
4928 size_t src_off_in_page;
4929 unsigned long dst_end = dst_offset + len - 1;
4930 unsigned long src_end = src_offset + len - 1;
4931 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4932 unsigned long dst_i;
4933 unsigned long src_i;
4935 if (src_offset + len > dst->len) {
4936 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4937 "len %lu len %lu\n", src_offset, len, dst->len);
4940 if (dst_offset + len > dst->len) {
4941 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4942 "len %lu len %lu\n", dst_offset, len, dst->len);
4945 if (dst_offset < src_offset) {
4946 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4950 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4951 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4953 dst_off_in_page = (start_offset + dst_end) &
4954 ((unsigned long)PAGE_CACHE_SIZE - 1);
4955 src_off_in_page = (start_offset + src_end) &
4956 ((unsigned long)PAGE_CACHE_SIZE - 1);
4958 cur = min_t(unsigned long, len, src_off_in_page + 1);
4959 cur = min(cur, dst_off_in_page + 1);
4960 move_pages(extent_buffer_page(dst, dst_i),
4961 extent_buffer_page(dst, src_i),
4962 dst_off_in_page - cur + 1,
4963 src_off_in_page - cur + 1, cur);
4971 int try_release_extent_buffer(struct page *page, gfp_t mask)
4973 struct extent_buffer *eb;
4976 * We need to make sure noboody is attaching this page to an eb right
4979 spin_lock(&page->mapping->private_lock);
4980 if (!PagePrivate(page)) {
4981 spin_unlock(&page->mapping->private_lock);
4985 eb = (struct extent_buffer *)page->private;
4989 * This is a little awful but should be ok, we need to make sure that
4990 * the eb doesn't disappear out from under us while we're looking at
4993 spin_lock(&eb->refs_lock);
4994 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4995 spin_unlock(&eb->refs_lock);
4996 spin_unlock(&page->mapping->private_lock);
4999 spin_unlock(&page->mapping->private_lock);
5001 if ((mask & GFP_NOFS) == GFP_NOFS)
5005 * If tree ref isn't set then we know the ref on this eb is a real ref,
5006 * so just return, this page will likely be freed soon anyway.
5008 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5009 spin_unlock(&eb->refs_lock);
5013 return release_extent_buffer(eb, mask);