1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 #include "check-integrity.h"
21 #include "rcu-string.h"
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
26 static struct bio_set *btrfs_bioset;
28 static inline bool extent_state_in_tree(const struct extent_state *state)
30 return !RB_EMPTY_NODE(&state->rb_node);
33 #ifdef CONFIG_BTRFS_DEBUG
34 static LIST_HEAD(buffers);
35 static LIST_HEAD(states);
37 static DEFINE_SPINLOCK(leak_lock);
40 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
44 spin_lock_irqsave(&leak_lock, flags);
46 spin_unlock_irqrestore(&leak_lock, flags);
50 void btrfs_leak_debug_del(struct list_head *entry)
54 spin_lock_irqsave(&leak_lock, flags);
56 spin_unlock_irqrestore(&leak_lock, flags);
60 void btrfs_leak_debug_check(void)
62 struct extent_state *state;
63 struct extent_buffer *eb;
65 while (!list_empty(&states)) {
66 state = list_entry(states.next, struct extent_state, leak_list);
67 pr_err("BTRFS: state leak: start %llu end %llu state %lu in tree %d refs %d\n",
68 state->start, state->end, state->state,
69 extent_state_in_tree(state),
70 atomic_read(&state->refs));
71 list_del(&state->leak_list);
72 kmem_cache_free(extent_state_cache, state);
75 while (!list_empty(&buffers)) {
76 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
77 printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu "
79 eb->start, eb->len, atomic_read(&eb->refs));
80 list_del(&eb->leak_list);
81 kmem_cache_free(extent_buffer_cache, eb);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
88 struct extent_io_tree *tree, u64 start, u64 end)
96 inode = tree->mapping->host;
97 isize = i_size_read(inode);
98 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
99 printk_ratelimited(KERN_DEBUG
100 "BTRFS: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
101 caller, btrfs_ino(inode), isize, start, end);
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry) do {} while (0)
107 #define btrfs_leak_debug_check() do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
111 #define BUFFER_LRU_MAX 64
116 struct rb_node rb_node;
119 struct extent_page_data {
121 struct extent_io_tree *tree;
122 get_extent_t *get_extent;
123 unsigned long bio_flags;
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
128 unsigned int extent_locked:1;
130 /* tells the submit_bio code to use a WRITE_SYNC */
131 unsigned int sync_io:1;
134 static noinline void flush_write_bio(void *data);
135 static inline struct btrfs_fs_info *
136 tree_fs_info(struct extent_io_tree *tree)
140 return btrfs_sb(tree->mapping->host->i_sb);
143 int __init extent_io_init(void)
145 extent_state_cache = kmem_cache_create("btrfs_extent_state",
146 sizeof(struct extent_state), 0,
147 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
148 if (!extent_state_cache)
151 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
152 sizeof(struct extent_buffer), 0,
153 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
154 if (!extent_buffer_cache)
155 goto free_state_cache;
157 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
158 offsetof(struct btrfs_io_bio, bio));
160 goto free_buffer_cache;
162 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
168 bioset_free(btrfs_bioset);
172 kmem_cache_destroy(extent_buffer_cache);
173 extent_buffer_cache = NULL;
176 kmem_cache_destroy(extent_state_cache);
177 extent_state_cache = NULL;
181 void extent_io_exit(void)
183 btrfs_leak_debug_check();
186 * Make sure all delayed rcu free are flushed before we
190 if (extent_state_cache)
191 kmem_cache_destroy(extent_state_cache);
192 if (extent_buffer_cache)
193 kmem_cache_destroy(extent_buffer_cache);
195 bioset_free(btrfs_bioset);
198 void extent_io_tree_init(struct extent_io_tree *tree,
199 struct address_space *mapping)
201 tree->state = RB_ROOT;
203 tree->dirty_bytes = 0;
204 spin_lock_init(&tree->lock);
205 tree->mapping = mapping;
208 static struct extent_state *alloc_extent_state(gfp_t mask)
210 struct extent_state *state;
212 state = kmem_cache_alloc(extent_state_cache, mask);
217 RB_CLEAR_NODE(&state->rb_node);
218 btrfs_leak_debug_add(&state->leak_list, &states);
219 atomic_set(&state->refs, 1);
220 init_waitqueue_head(&state->wq);
221 trace_alloc_extent_state(state, mask, _RET_IP_);
225 void free_extent_state(struct extent_state *state)
229 if (atomic_dec_and_test(&state->refs)) {
230 WARN_ON(extent_state_in_tree(state));
231 btrfs_leak_debug_del(&state->leak_list);
232 trace_free_extent_state(state, _RET_IP_);
233 kmem_cache_free(extent_state_cache, state);
237 static struct rb_node *tree_insert(struct rb_root *root,
238 struct rb_node *search_start,
240 struct rb_node *node,
241 struct rb_node ***p_in,
242 struct rb_node **parent_in)
245 struct rb_node *parent = NULL;
246 struct tree_entry *entry;
248 if (p_in && parent_in) {
254 p = search_start ? &search_start : &root->rb_node;
257 entry = rb_entry(parent, struct tree_entry, rb_node);
259 if (offset < entry->start)
261 else if (offset > entry->end)
268 rb_link_node(node, parent, p);
269 rb_insert_color(node, root);
273 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
274 struct rb_node **prev_ret,
275 struct rb_node **next_ret,
276 struct rb_node ***p_ret,
277 struct rb_node **parent_ret)
279 struct rb_root *root = &tree->state;
280 struct rb_node **n = &root->rb_node;
281 struct rb_node *prev = NULL;
282 struct rb_node *orig_prev = NULL;
283 struct tree_entry *entry;
284 struct tree_entry *prev_entry = NULL;
288 entry = rb_entry(prev, struct tree_entry, rb_node);
291 if (offset < entry->start)
293 else if (offset > entry->end)
306 while (prev && offset > prev_entry->end) {
307 prev = rb_next(prev);
308 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
315 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
316 while (prev && offset < prev_entry->start) {
317 prev = rb_prev(prev);
318 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
325 static inline struct rb_node *
326 tree_search_for_insert(struct extent_io_tree *tree,
328 struct rb_node ***p_ret,
329 struct rb_node **parent_ret)
331 struct rb_node *prev = NULL;
334 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
340 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
343 return tree_search_for_insert(tree, offset, NULL, NULL);
346 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
347 struct extent_state *other)
349 if (tree->ops && tree->ops->merge_extent_hook)
350 tree->ops->merge_extent_hook(tree->mapping->host, new,
355 * utility function to look for merge candidates inside a given range.
356 * Any extents with matching state are merged together into a single
357 * extent in the tree. Extents with EXTENT_IO in their state field
358 * are not merged because the end_io handlers need to be able to do
359 * operations on them without sleeping (or doing allocations/splits).
361 * This should be called with the tree lock held.
363 static void merge_state(struct extent_io_tree *tree,
364 struct extent_state *state)
366 struct extent_state *other;
367 struct rb_node *other_node;
369 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
372 other_node = rb_prev(&state->rb_node);
374 other = rb_entry(other_node, struct extent_state, rb_node);
375 if (other->end == state->start - 1 &&
376 other->state == state->state) {
377 merge_cb(tree, state, other);
378 state->start = other->start;
379 rb_erase(&other->rb_node, &tree->state);
380 RB_CLEAR_NODE(&other->rb_node);
381 free_extent_state(other);
384 other_node = rb_next(&state->rb_node);
386 other = rb_entry(other_node, struct extent_state, rb_node);
387 if (other->start == state->end + 1 &&
388 other->state == state->state) {
389 merge_cb(tree, state, other);
390 state->end = other->end;
391 rb_erase(&other->rb_node, &tree->state);
392 RB_CLEAR_NODE(&other->rb_node);
393 free_extent_state(other);
398 static void set_state_cb(struct extent_io_tree *tree,
399 struct extent_state *state, unsigned long *bits)
401 if (tree->ops && tree->ops->set_bit_hook)
402 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
405 static void clear_state_cb(struct extent_io_tree *tree,
406 struct extent_state *state, unsigned long *bits)
408 if (tree->ops && tree->ops->clear_bit_hook)
409 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
412 static void set_state_bits(struct extent_io_tree *tree,
413 struct extent_state *state, unsigned long *bits);
416 * insert an extent_state struct into the tree. 'bits' are set on the
417 * struct before it is inserted.
419 * This may return -EEXIST if the extent is already there, in which case the
420 * state struct is freed.
422 * The tree lock is not taken internally. This is a utility function and
423 * probably isn't what you want to call (see set/clear_extent_bit).
425 static int insert_state(struct extent_io_tree *tree,
426 struct extent_state *state, u64 start, u64 end,
428 struct rb_node **parent,
431 struct rb_node *node;
434 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
436 state->start = start;
439 set_state_bits(tree, state, bits);
441 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
443 struct extent_state *found;
444 found = rb_entry(node, struct extent_state, rb_node);
445 printk(KERN_ERR "BTRFS: found node %llu %llu on insert of "
447 found->start, found->end, start, end);
450 merge_state(tree, state);
454 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
457 if (tree->ops && tree->ops->split_extent_hook)
458 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
462 * split a given extent state struct in two, inserting the preallocated
463 * struct 'prealloc' as the newly created second half. 'split' indicates an
464 * offset inside 'orig' where it should be split.
467 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
468 * are two extent state structs in the tree:
469 * prealloc: [orig->start, split - 1]
470 * orig: [ split, orig->end ]
472 * The tree locks are not taken by this function. They need to be held
475 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
476 struct extent_state *prealloc, u64 split)
478 struct rb_node *node;
480 split_cb(tree, orig, split);
482 prealloc->start = orig->start;
483 prealloc->end = split - 1;
484 prealloc->state = orig->state;
487 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
488 &prealloc->rb_node, NULL, NULL);
490 free_extent_state(prealloc);
496 static struct extent_state *next_state(struct extent_state *state)
498 struct rb_node *next = rb_next(&state->rb_node);
500 return rb_entry(next, struct extent_state, rb_node);
506 * utility function to clear some bits in an extent state struct.
507 * it will optionally wake up any one waiting on this state (wake == 1).
509 * If no bits are set on the state struct after clearing things, the
510 * struct is freed and removed from the tree
512 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
513 struct extent_state *state,
514 unsigned long *bits, int wake)
516 struct extent_state *next;
517 unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
519 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
520 u64 range = state->end - state->start + 1;
521 WARN_ON(range > tree->dirty_bytes);
522 tree->dirty_bytes -= range;
524 clear_state_cb(tree, state, bits);
525 state->state &= ~bits_to_clear;
528 if (state->state == 0) {
529 next = next_state(state);
530 if (extent_state_in_tree(state)) {
531 rb_erase(&state->rb_node, &tree->state);
532 RB_CLEAR_NODE(&state->rb_node);
533 free_extent_state(state);
538 merge_state(tree, state);
539 next = next_state(state);
544 static struct extent_state *
545 alloc_extent_state_atomic(struct extent_state *prealloc)
548 prealloc = alloc_extent_state(GFP_ATOMIC);
553 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
555 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
556 "Extent tree was modified by another "
557 "thread while locked.");
561 * clear some bits on a range in the tree. This may require splitting
562 * or inserting elements in the tree, so the gfp mask is used to
563 * indicate which allocations or sleeping are allowed.
565 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
566 * the given range from the tree regardless of state (ie for truncate).
568 * the range [start, end] is inclusive.
570 * This takes the tree lock, and returns 0 on success and < 0 on error.
572 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
573 unsigned long bits, int wake, int delete,
574 struct extent_state **cached_state,
577 struct extent_state *state;
578 struct extent_state *cached;
579 struct extent_state *prealloc = NULL;
580 struct rb_node *node;
585 btrfs_debug_check_extent_io_range(tree, start, end);
587 if (bits & EXTENT_DELALLOC)
588 bits |= EXTENT_NORESERVE;
591 bits |= ~EXTENT_CTLBITS;
592 bits |= EXTENT_FIRST_DELALLOC;
594 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
597 if (!prealloc && (mask & __GFP_WAIT)) {
598 prealloc = alloc_extent_state(mask);
603 spin_lock(&tree->lock);
605 cached = *cached_state;
608 *cached_state = NULL;
612 if (cached && extent_state_in_tree(cached) &&
613 cached->start <= start && cached->end > start) {
615 atomic_dec(&cached->refs);
620 free_extent_state(cached);
623 * this search will find the extents that end after
626 node = tree_search(tree, start);
629 state = rb_entry(node, struct extent_state, rb_node);
631 if (state->start > end)
633 WARN_ON(state->end < start);
634 last_end = state->end;
636 /* the state doesn't have the wanted bits, go ahead */
637 if (!(state->state & bits)) {
638 state = next_state(state);
643 * | ---- desired range ---- |
645 * | ------------- state -------------- |
647 * We need to split the extent we found, and may flip
648 * bits on second half.
650 * If the extent we found extends past our range, we
651 * just split and search again. It'll get split again
652 * the next time though.
654 * If the extent we found is inside our range, we clear
655 * the desired bit on it.
658 if (state->start < start) {
659 prealloc = alloc_extent_state_atomic(prealloc);
661 err = split_state(tree, state, prealloc, start);
663 extent_io_tree_panic(tree, err);
668 if (state->end <= end) {
669 state = clear_state_bit(tree, state, &bits, wake);
675 * | ---- desired range ---- |
677 * We need to split the extent, and clear the bit
680 if (state->start <= end && state->end > end) {
681 prealloc = alloc_extent_state_atomic(prealloc);
683 err = split_state(tree, state, prealloc, end + 1);
685 extent_io_tree_panic(tree, err);
690 clear_state_bit(tree, prealloc, &bits, wake);
696 state = clear_state_bit(tree, state, &bits, wake);
698 if (last_end == (u64)-1)
700 start = last_end + 1;
701 if (start <= end && state && !need_resched())
706 spin_unlock(&tree->lock);
708 free_extent_state(prealloc);
715 spin_unlock(&tree->lock);
716 if (mask & __GFP_WAIT)
721 static void wait_on_state(struct extent_io_tree *tree,
722 struct extent_state *state)
723 __releases(tree->lock)
724 __acquires(tree->lock)
727 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
728 spin_unlock(&tree->lock);
730 spin_lock(&tree->lock);
731 finish_wait(&state->wq, &wait);
735 * waits for one or more bits to clear on a range in the state tree.
736 * The range [start, end] is inclusive.
737 * The tree lock is taken by this function
739 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
742 struct extent_state *state;
743 struct rb_node *node;
745 btrfs_debug_check_extent_io_range(tree, start, end);
747 spin_lock(&tree->lock);
751 * this search will find all the extents that end after
754 node = tree_search(tree, start);
759 state = rb_entry(node, struct extent_state, rb_node);
761 if (state->start > end)
764 if (state->state & bits) {
765 start = state->start;
766 atomic_inc(&state->refs);
767 wait_on_state(tree, state);
768 free_extent_state(state);
771 start = state->end + 1;
776 if (!cond_resched_lock(&tree->lock)) {
777 node = rb_next(node);
782 spin_unlock(&tree->lock);
785 static void set_state_bits(struct extent_io_tree *tree,
786 struct extent_state *state,
789 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
791 set_state_cb(tree, state, bits);
792 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
793 u64 range = state->end - state->start + 1;
794 tree->dirty_bytes += range;
796 state->state |= bits_to_set;
799 static void cache_state(struct extent_state *state,
800 struct extent_state **cached_ptr)
802 if (cached_ptr && !(*cached_ptr)) {
803 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
805 atomic_inc(&state->refs);
811 * set some bits on a range in the tree. This may require allocations or
812 * sleeping, so the gfp mask is used to indicate what is allowed.
814 * If any of the exclusive bits are set, this will fail with -EEXIST if some
815 * part of the range already has the desired bits set. The start of the
816 * existing range is returned in failed_start in this case.
818 * [start, end] is inclusive This takes the tree lock.
821 static int __must_check
822 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
823 unsigned long bits, unsigned long exclusive_bits,
824 u64 *failed_start, struct extent_state **cached_state,
827 struct extent_state *state;
828 struct extent_state *prealloc = NULL;
829 struct rb_node *node;
831 struct rb_node *parent;
836 btrfs_debug_check_extent_io_range(tree, start, end);
838 bits |= EXTENT_FIRST_DELALLOC;
840 if (!prealloc && (mask & __GFP_WAIT)) {
841 prealloc = alloc_extent_state(mask);
845 spin_lock(&tree->lock);
846 if (cached_state && *cached_state) {
847 state = *cached_state;
848 if (state->start <= start && state->end > start &&
849 extent_state_in_tree(state)) {
850 node = &state->rb_node;
855 * this search will find all the extents that end after
858 node = tree_search_for_insert(tree, start, &p, &parent);
860 prealloc = alloc_extent_state_atomic(prealloc);
862 err = insert_state(tree, prealloc, start, end,
865 extent_io_tree_panic(tree, err);
867 cache_state(prealloc, cached_state);
871 state = rb_entry(node, struct extent_state, rb_node);
873 last_start = state->start;
874 last_end = state->end;
877 * | ---- desired range ---- |
880 * Just lock what we found and keep going
882 if (state->start == start && state->end <= end) {
883 if (state->state & exclusive_bits) {
884 *failed_start = state->start;
889 set_state_bits(tree, state, &bits);
890 cache_state(state, cached_state);
891 merge_state(tree, state);
892 if (last_end == (u64)-1)
894 start = last_end + 1;
895 state = next_state(state);
896 if (start < end && state && state->start == start &&
903 * | ---- desired range ---- |
906 * | ------------- state -------------- |
908 * We need to split the extent we found, and may flip bits on
911 * If the extent we found extends past our
912 * range, we just split and search again. It'll get split
913 * again the next time though.
915 * If the extent we found is inside our range, we set the
918 if (state->start < start) {
919 if (state->state & exclusive_bits) {
920 *failed_start = start;
925 prealloc = alloc_extent_state_atomic(prealloc);
927 err = split_state(tree, state, prealloc, start);
929 extent_io_tree_panic(tree, err);
934 if (state->end <= end) {
935 set_state_bits(tree, state, &bits);
936 cache_state(state, cached_state);
937 merge_state(tree, state);
938 if (last_end == (u64)-1)
940 start = last_end + 1;
941 state = next_state(state);
942 if (start < end && state && state->start == start &&
949 * | ---- desired range ---- |
950 * | state | or | state |
952 * There's a hole, we need to insert something in it and
953 * ignore the extent we found.
955 if (state->start > start) {
957 if (end < last_start)
960 this_end = last_start - 1;
962 prealloc = alloc_extent_state_atomic(prealloc);
966 * Avoid to free 'prealloc' if it can be merged with
969 err = insert_state(tree, prealloc, start, this_end,
972 extent_io_tree_panic(tree, err);
974 cache_state(prealloc, cached_state);
976 start = this_end + 1;
980 * | ---- desired range ---- |
982 * We need to split the extent, and set the bit
985 if (state->start <= end && state->end > end) {
986 if (state->state & exclusive_bits) {
987 *failed_start = start;
992 prealloc = alloc_extent_state_atomic(prealloc);
994 err = split_state(tree, state, prealloc, end + 1);
996 extent_io_tree_panic(tree, err);
998 set_state_bits(tree, prealloc, &bits);
999 cache_state(prealloc, cached_state);
1000 merge_state(tree, prealloc);
1008 spin_unlock(&tree->lock);
1010 free_extent_state(prealloc);
1017 spin_unlock(&tree->lock);
1018 if (mask & __GFP_WAIT)
1023 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1024 unsigned long bits, u64 * failed_start,
1025 struct extent_state **cached_state, gfp_t mask)
1027 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1028 cached_state, mask);
1033 * convert_extent_bit - convert all bits in a given range from one bit to
1035 * @tree: the io tree to search
1036 * @start: the start offset in bytes
1037 * @end: the end offset in bytes (inclusive)
1038 * @bits: the bits to set in this range
1039 * @clear_bits: the bits to clear in this range
1040 * @cached_state: state that we're going to cache
1041 * @mask: the allocation mask
1043 * This will go through and set bits for the given range. If any states exist
1044 * already in this range they are set with the given bit and cleared of the
1045 * clear_bits. This is only meant to be used by things that are mergeable, ie
1046 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1047 * boundary bits like LOCK.
1049 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1050 unsigned long bits, unsigned long clear_bits,
1051 struct extent_state **cached_state, gfp_t mask)
1053 struct extent_state *state;
1054 struct extent_state *prealloc = NULL;
1055 struct rb_node *node;
1057 struct rb_node *parent;
1062 btrfs_debug_check_extent_io_range(tree, start, end);
1065 if (!prealloc && (mask & __GFP_WAIT)) {
1066 prealloc = alloc_extent_state(mask);
1071 spin_lock(&tree->lock);
1072 if (cached_state && *cached_state) {
1073 state = *cached_state;
1074 if (state->start <= start && state->end > start &&
1075 extent_state_in_tree(state)) {
1076 node = &state->rb_node;
1082 * this search will find all the extents that end after
1085 node = tree_search_for_insert(tree, start, &p, &parent);
1087 prealloc = alloc_extent_state_atomic(prealloc);
1092 err = insert_state(tree, prealloc, start, end,
1093 &p, &parent, &bits);
1095 extent_io_tree_panic(tree, err);
1096 cache_state(prealloc, cached_state);
1100 state = rb_entry(node, struct extent_state, rb_node);
1102 last_start = state->start;
1103 last_end = state->end;
1106 * | ---- desired range ---- |
1109 * Just lock what we found and keep going
1111 if (state->start == start && state->end <= end) {
1112 set_state_bits(tree, state, &bits);
1113 cache_state(state, cached_state);
1114 state = clear_state_bit(tree, state, &clear_bits, 0);
1115 if (last_end == (u64)-1)
1117 start = last_end + 1;
1118 if (start < end && state && state->start == start &&
1125 * | ---- desired range ---- |
1128 * | ------------- state -------------- |
1130 * We need to split the extent we found, and may flip bits on
1133 * If the extent we found extends past our
1134 * range, we just split and search again. It'll get split
1135 * again the next time though.
1137 * If the extent we found is inside our range, we set the
1138 * desired bit on it.
1140 if (state->start < start) {
1141 prealloc = alloc_extent_state_atomic(prealloc);
1146 err = split_state(tree, state, prealloc, start);
1148 extent_io_tree_panic(tree, err);
1152 if (state->end <= end) {
1153 set_state_bits(tree, state, &bits);
1154 cache_state(state, cached_state);
1155 state = clear_state_bit(tree, state, &clear_bits, 0);
1156 if (last_end == (u64)-1)
1158 start = last_end + 1;
1159 if (start < end && state && state->start == start &&
1166 * | ---- desired range ---- |
1167 * | state | or | state |
1169 * There's a hole, we need to insert something in it and
1170 * ignore the extent we found.
1172 if (state->start > start) {
1174 if (end < last_start)
1177 this_end = last_start - 1;
1179 prealloc = alloc_extent_state_atomic(prealloc);
1186 * Avoid to free 'prealloc' if it can be merged with
1189 err = insert_state(tree, prealloc, start, this_end,
1192 extent_io_tree_panic(tree, err);
1193 cache_state(prealloc, cached_state);
1195 start = this_end + 1;
1199 * | ---- desired range ---- |
1201 * We need to split the extent, and set the bit
1204 if (state->start <= end && state->end > end) {
1205 prealloc = alloc_extent_state_atomic(prealloc);
1211 err = split_state(tree, state, prealloc, end + 1);
1213 extent_io_tree_panic(tree, err);
1215 set_state_bits(tree, prealloc, &bits);
1216 cache_state(prealloc, cached_state);
1217 clear_state_bit(tree, prealloc, &clear_bits, 0);
1225 spin_unlock(&tree->lock);
1227 free_extent_state(prealloc);
1234 spin_unlock(&tree->lock);
1235 if (mask & __GFP_WAIT)
1240 /* wrappers around set/clear extent bit */
1241 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1244 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1248 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1249 unsigned long bits, gfp_t mask)
1251 return set_extent_bit(tree, start, end, bits, NULL,
1255 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1256 unsigned long bits, gfp_t mask)
1258 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1261 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1262 struct extent_state **cached_state, gfp_t mask)
1264 return set_extent_bit(tree, start, end,
1265 EXTENT_DELALLOC | EXTENT_UPTODATE,
1266 NULL, cached_state, mask);
1269 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1270 struct extent_state **cached_state, gfp_t mask)
1272 return set_extent_bit(tree, start, end,
1273 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1274 NULL, cached_state, mask);
1277 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1280 return clear_extent_bit(tree, start, end,
1281 EXTENT_DIRTY | EXTENT_DELALLOC |
1282 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1285 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1288 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1292 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1293 struct extent_state **cached_state, gfp_t mask)
1295 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1296 cached_state, mask);
1299 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1300 struct extent_state **cached_state, gfp_t mask)
1302 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1303 cached_state, mask);
1307 * either insert or lock state struct between start and end use mask to tell
1308 * us if waiting is desired.
1310 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1311 unsigned long bits, struct extent_state **cached_state)
1316 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1317 EXTENT_LOCKED, &failed_start,
1318 cached_state, GFP_NOFS);
1319 if (err == -EEXIST) {
1320 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1321 start = failed_start;
1324 WARN_ON(start > end);
1329 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1331 return lock_extent_bits(tree, start, end, 0, NULL);
1334 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1339 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1340 &failed_start, NULL, GFP_NOFS);
1341 if (err == -EEXIST) {
1342 if (failed_start > start)
1343 clear_extent_bit(tree, start, failed_start - 1,
1344 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1350 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1351 struct extent_state **cached, gfp_t mask)
1353 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1357 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1359 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1363 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1365 unsigned long index = start >> PAGE_CACHE_SHIFT;
1366 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1369 while (index <= end_index) {
1370 page = find_get_page(inode->i_mapping, index);
1371 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1372 clear_page_dirty_for_io(page);
1373 page_cache_release(page);
1379 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1381 unsigned long index = start >> PAGE_CACHE_SHIFT;
1382 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1385 while (index <= end_index) {
1386 page = find_get_page(inode->i_mapping, index);
1387 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1388 account_page_redirty(page);
1389 __set_page_dirty_nobuffers(page);
1390 page_cache_release(page);
1397 * helper function to set both pages and extents in the tree writeback
1399 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1401 unsigned long index = start >> PAGE_CACHE_SHIFT;
1402 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1405 while (index <= end_index) {
1406 page = find_get_page(tree->mapping, index);
1407 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1408 set_page_writeback(page);
1409 page_cache_release(page);
1415 /* find the first state struct with 'bits' set after 'start', and
1416 * return it. tree->lock must be held. NULL will returned if
1417 * nothing was found after 'start'
1419 static struct extent_state *
1420 find_first_extent_bit_state(struct extent_io_tree *tree,
1421 u64 start, unsigned long bits)
1423 struct rb_node *node;
1424 struct extent_state *state;
1427 * this search will find all the extents that end after
1430 node = tree_search(tree, start);
1435 state = rb_entry(node, struct extent_state, rb_node);
1436 if (state->end >= start && (state->state & bits))
1439 node = rb_next(node);
1448 * find the first offset in the io tree with 'bits' set. zero is
1449 * returned if we find something, and *start_ret and *end_ret are
1450 * set to reflect the state struct that was found.
1452 * If nothing was found, 1 is returned. If found something, return 0.
1454 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1455 u64 *start_ret, u64 *end_ret, unsigned long bits,
1456 struct extent_state **cached_state)
1458 struct extent_state *state;
1462 spin_lock(&tree->lock);
1463 if (cached_state && *cached_state) {
1464 state = *cached_state;
1465 if (state->end == start - 1 && extent_state_in_tree(state)) {
1466 n = rb_next(&state->rb_node);
1468 state = rb_entry(n, struct extent_state,
1470 if (state->state & bits)
1474 free_extent_state(*cached_state);
1475 *cached_state = NULL;
1478 free_extent_state(*cached_state);
1479 *cached_state = NULL;
1482 state = find_first_extent_bit_state(tree, start, bits);
1485 cache_state(state, cached_state);
1486 *start_ret = state->start;
1487 *end_ret = state->end;
1491 spin_unlock(&tree->lock);
1496 * find a contiguous range of bytes in the file marked as delalloc, not
1497 * more than 'max_bytes'. start and end are used to return the range,
1499 * 1 is returned if we find something, 0 if nothing was in the tree
1501 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1502 u64 *start, u64 *end, u64 max_bytes,
1503 struct extent_state **cached_state)
1505 struct rb_node *node;
1506 struct extent_state *state;
1507 u64 cur_start = *start;
1509 u64 total_bytes = 0;
1511 spin_lock(&tree->lock);
1514 * this search will find all the extents that end after
1517 node = tree_search(tree, cur_start);
1525 state = rb_entry(node, struct extent_state, rb_node);
1526 if (found && (state->start != cur_start ||
1527 (state->state & EXTENT_BOUNDARY))) {
1530 if (!(state->state & EXTENT_DELALLOC)) {
1536 *start = state->start;
1537 *cached_state = state;
1538 atomic_inc(&state->refs);
1542 cur_start = state->end + 1;
1543 node = rb_next(node);
1544 total_bytes += state->end - state->start + 1;
1545 if (total_bytes >= max_bytes)
1551 spin_unlock(&tree->lock);
1555 static noinline void __unlock_for_delalloc(struct inode *inode,
1556 struct page *locked_page,
1560 struct page *pages[16];
1561 unsigned long index = start >> PAGE_CACHE_SHIFT;
1562 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1563 unsigned long nr_pages = end_index - index + 1;
1566 if (index == locked_page->index && end_index == index)
1569 while (nr_pages > 0) {
1570 ret = find_get_pages_contig(inode->i_mapping, index,
1571 min_t(unsigned long, nr_pages,
1572 ARRAY_SIZE(pages)), pages);
1573 for (i = 0; i < ret; i++) {
1574 if (pages[i] != locked_page)
1575 unlock_page(pages[i]);
1576 page_cache_release(pages[i]);
1584 static noinline int lock_delalloc_pages(struct inode *inode,
1585 struct page *locked_page,
1589 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1590 unsigned long start_index = index;
1591 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1592 unsigned long pages_locked = 0;
1593 struct page *pages[16];
1594 unsigned long nrpages;
1598 /* the caller is responsible for locking the start index */
1599 if (index == locked_page->index && index == end_index)
1602 /* skip the page at the start index */
1603 nrpages = end_index - index + 1;
1604 while (nrpages > 0) {
1605 ret = find_get_pages_contig(inode->i_mapping, index,
1606 min_t(unsigned long,
1607 nrpages, ARRAY_SIZE(pages)), pages);
1612 /* now we have an array of pages, lock them all */
1613 for (i = 0; i < ret; i++) {
1615 * the caller is taking responsibility for
1618 if (pages[i] != locked_page) {
1619 lock_page(pages[i]);
1620 if (!PageDirty(pages[i]) ||
1621 pages[i]->mapping != inode->i_mapping) {
1623 unlock_page(pages[i]);
1624 page_cache_release(pages[i]);
1628 page_cache_release(pages[i]);
1637 if (ret && pages_locked) {
1638 __unlock_for_delalloc(inode, locked_page,
1640 ((u64)(start_index + pages_locked - 1)) <<
1647 * find a contiguous range of bytes in the file marked as delalloc, not
1648 * more than 'max_bytes'. start and end are used to return the range,
1650 * 1 is returned if we find something, 0 if nothing was in the tree
1652 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1653 struct extent_io_tree *tree,
1654 struct page *locked_page, u64 *start,
1655 u64 *end, u64 max_bytes)
1660 struct extent_state *cached_state = NULL;
1665 /* step one, find a bunch of delalloc bytes starting at start */
1666 delalloc_start = *start;
1668 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1669 max_bytes, &cached_state);
1670 if (!found || delalloc_end <= *start) {
1671 *start = delalloc_start;
1672 *end = delalloc_end;
1673 free_extent_state(cached_state);
1678 * start comes from the offset of locked_page. We have to lock
1679 * pages in order, so we can't process delalloc bytes before
1682 if (delalloc_start < *start)
1683 delalloc_start = *start;
1686 * make sure to limit the number of pages we try to lock down
1688 if (delalloc_end + 1 - delalloc_start > max_bytes)
1689 delalloc_end = delalloc_start + max_bytes - 1;
1691 /* step two, lock all the pages after the page that has start */
1692 ret = lock_delalloc_pages(inode, locked_page,
1693 delalloc_start, delalloc_end);
1694 if (ret == -EAGAIN) {
1695 /* some of the pages are gone, lets avoid looping by
1696 * shortening the size of the delalloc range we're searching
1698 free_extent_state(cached_state);
1699 cached_state = NULL;
1701 max_bytes = PAGE_CACHE_SIZE;
1709 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1711 /* step three, lock the state bits for the whole range */
1712 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1714 /* then test to make sure it is all still delalloc */
1715 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1716 EXTENT_DELALLOC, 1, cached_state);
1718 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1719 &cached_state, GFP_NOFS);
1720 __unlock_for_delalloc(inode, locked_page,
1721 delalloc_start, delalloc_end);
1725 free_extent_state(cached_state);
1726 *start = delalloc_start;
1727 *end = delalloc_end;
1732 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1733 struct page *locked_page,
1734 unsigned long clear_bits,
1735 unsigned long page_ops)
1737 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1739 struct page *pages[16];
1740 unsigned long index = start >> PAGE_CACHE_SHIFT;
1741 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1742 unsigned long nr_pages = end_index - index + 1;
1745 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1749 while (nr_pages > 0) {
1750 ret = find_get_pages_contig(inode->i_mapping, index,
1751 min_t(unsigned long,
1752 nr_pages, ARRAY_SIZE(pages)), pages);
1753 for (i = 0; i < ret; i++) {
1755 if (page_ops & PAGE_SET_PRIVATE2)
1756 SetPagePrivate2(pages[i]);
1758 if (pages[i] == locked_page) {
1759 page_cache_release(pages[i]);
1762 if (page_ops & PAGE_CLEAR_DIRTY)
1763 clear_page_dirty_for_io(pages[i]);
1764 if (page_ops & PAGE_SET_WRITEBACK)
1765 set_page_writeback(pages[i]);
1766 if (page_ops & PAGE_END_WRITEBACK)
1767 end_page_writeback(pages[i]);
1768 if (page_ops & PAGE_UNLOCK)
1769 unlock_page(pages[i]);
1770 page_cache_release(pages[i]);
1780 * count the number of bytes in the tree that have a given bit(s)
1781 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1782 * cached. The total number found is returned.
1784 u64 count_range_bits(struct extent_io_tree *tree,
1785 u64 *start, u64 search_end, u64 max_bytes,
1786 unsigned long bits, int contig)
1788 struct rb_node *node;
1789 struct extent_state *state;
1790 u64 cur_start = *start;
1791 u64 total_bytes = 0;
1795 if (WARN_ON(search_end <= cur_start))
1798 spin_lock(&tree->lock);
1799 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1800 total_bytes = tree->dirty_bytes;
1804 * this search will find all the extents that end after
1807 node = tree_search(tree, cur_start);
1812 state = rb_entry(node, struct extent_state, rb_node);
1813 if (state->start > search_end)
1815 if (contig && found && state->start > last + 1)
1817 if (state->end >= cur_start && (state->state & bits) == bits) {
1818 total_bytes += min(search_end, state->end) + 1 -
1819 max(cur_start, state->start);
1820 if (total_bytes >= max_bytes)
1823 *start = max(cur_start, state->start);
1827 } else if (contig && found) {
1830 node = rb_next(node);
1835 spin_unlock(&tree->lock);
1840 * set the private field for a given byte offset in the tree. If there isn't
1841 * an extent_state there already, this does nothing.
1843 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1845 struct rb_node *node;
1846 struct extent_state *state;
1849 spin_lock(&tree->lock);
1851 * this search will find all the extents that end after
1854 node = tree_search(tree, start);
1859 state = rb_entry(node, struct extent_state, rb_node);
1860 if (state->start != start) {
1864 state->private = private;
1866 spin_unlock(&tree->lock);
1870 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1872 struct rb_node *node;
1873 struct extent_state *state;
1876 spin_lock(&tree->lock);
1878 * this search will find all the extents that end after
1881 node = tree_search(tree, start);
1886 state = rb_entry(node, struct extent_state, rb_node);
1887 if (state->start != start) {
1891 *private = state->private;
1893 spin_unlock(&tree->lock);
1898 * searches a range in the state tree for a given mask.
1899 * If 'filled' == 1, this returns 1 only if every extent in the tree
1900 * has the bits set. Otherwise, 1 is returned if any bit in the
1901 * range is found set.
1903 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1904 unsigned long bits, int filled, struct extent_state *cached)
1906 struct extent_state *state = NULL;
1907 struct rb_node *node;
1910 spin_lock(&tree->lock);
1911 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1912 cached->end > start)
1913 node = &cached->rb_node;
1915 node = tree_search(tree, start);
1916 while (node && start <= end) {
1917 state = rb_entry(node, struct extent_state, rb_node);
1919 if (filled && state->start > start) {
1924 if (state->start > end)
1927 if (state->state & bits) {
1931 } else if (filled) {
1936 if (state->end == (u64)-1)
1939 start = state->end + 1;
1942 node = rb_next(node);
1949 spin_unlock(&tree->lock);
1954 * helper function to set a given page up to date if all the
1955 * extents in the tree for that page are up to date
1957 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1959 u64 start = page_offset(page);
1960 u64 end = start + PAGE_CACHE_SIZE - 1;
1961 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1962 SetPageUptodate(page);
1965 int free_io_failure(struct inode *inode, struct io_failure_record *rec)
1969 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1971 set_state_private(failure_tree, rec->start, 0);
1972 ret = clear_extent_bits(failure_tree, rec->start,
1973 rec->start + rec->len - 1,
1974 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1978 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1979 rec->start + rec->len - 1,
1980 EXTENT_DAMAGED, GFP_NOFS);
1989 * this bypasses the standard btrfs submit functions deliberately, as
1990 * the standard behavior is to write all copies in a raid setup. here we only
1991 * want to write the one bad copy. so we do the mapping for ourselves and issue
1992 * submit_bio directly.
1993 * to avoid any synchronization issues, wait for the data after writing, which
1994 * actually prevents the read that triggered the error from finishing.
1995 * currently, there can be no more than two copies of every data bit. thus,
1996 * exactly one rewrite is required.
1998 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
1999 struct page *page, unsigned int pg_offset, int mirror_num)
2001 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2003 struct btrfs_device *dev;
2006 struct btrfs_bio *bbio = NULL;
2007 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2010 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2011 BUG_ON(!mirror_num);
2013 /* we can't repair anything in raid56 yet */
2014 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2017 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2020 bio->bi_iter.bi_size = 0;
2021 map_length = length;
2023 ret = btrfs_map_block(fs_info, WRITE, logical,
2024 &map_length, &bbio, mirror_num);
2029 BUG_ON(mirror_num != bbio->mirror_num);
2030 sector = bbio->stripes[mirror_num-1].physical >> 9;
2031 bio->bi_iter.bi_sector = sector;
2032 dev = bbio->stripes[mirror_num-1].dev;
2034 if (!dev || !dev->bdev || !dev->writeable) {
2038 bio->bi_bdev = dev->bdev;
2039 bio_add_page(bio, page, length, pg_offset);
2041 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
2042 /* try to remap that extent elsewhere? */
2044 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2048 printk_ratelimited_in_rcu(KERN_INFO
2049 "BTRFS: read error corrected: ino %llu off %llu (dev %s sector %llu)\n",
2050 btrfs_ino(inode), start,
2051 rcu_str_deref(dev->name), sector);
2056 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2059 u64 start = eb->start;
2060 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2063 if (root->fs_info->sb->s_flags & MS_RDONLY)
2066 for (i = 0; i < num_pages; i++) {
2067 struct page *p = eb->pages[i];
2069 ret = repair_io_failure(root->fs_info->btree_inode, start,
2070 PAGE_CACHE_SIZE, start, p,
2071 start - page_offset(p), mirror_num);
2074 start += PAGE_CACHE_SIZE;
2081 * each time an IO finishes, we do a fast check in the IO failure tree
2082 * to see if we need to process or clean up an io_failure_record
2084 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
2085 unsigned int pg_offset)
2088 u64 private_failure;
2089 struct io_failure_record *failrec;
2090 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2091 struct extent_state *state;
2096 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2097 (u64)-1, 1, EXTENT_DIRTY, 0);
2101 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2106 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2107 BUG_ON(!failrec->this_mirror);
2109 if (failrec->in_validation) {
2110 /* there was no real error, just free the record */
2111 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2115 if (fs_info->sb->s_flags & MS_RDONLY)
2118 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2119 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2122 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2124 if (state && state->start <= failrec->start &&
2125 state->end >= failrec->start + failrec->len - 1) {
2126 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2128 if (num_copies > 1) {
2129 repair_io_failure(inode, start, failrec->len,
2130 failrec->logical, page,
2131 pg_offset, failrec->failed_mirror);
2136 free_io_failure(inode, failrec);
2142 * Can be called when
2143 * - hold extent lock
2144 * - under ordered extent
2145 * - the inode is freeing
2147 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end)
2149 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2150 struct io_failure_record *failrec;
2151 struct extent_state *state, *next;
2153 if (RB_EMPTY_ROOT(&failure_tree->state))
2156 spin_lock(&failure_tree->lock);
2157 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2159 if (state->start > end)
2162 ASSERT(state->end <= end);
2164 next = next_state(state);
2166 failrec = (struct io_failure_record *)state->private;
2167 free_extent_state(state);
2172 spin_unlock(&failure_tree->lock);
2175 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2176 struct io_failure_record **failrec_ret)
2178 struct io_failure_record *failrec;
2180 struct extent_map *em;
2181 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2182 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2183 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2187 ret = get_state_private(failure_tree, start, &private);
2189 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2193 failrec->start = start;
2194 failrec->len = end - start + 1;
2195 failrec->this_mirror = 0;
2196 failrec->bio_flags = 0;
2197 failrec->in_validation = 0;
2199 read_lock(&em_tree->lock);
2200 em = lookup_extent_mapping(em_tree, start, failrec->len);
2202 read_unlock(&em_tree->lock);
2207 if (em->start > start || em->start + em->len <= start) {
2208 free_extent_map(em);
2211 read_unlock(&em_tree->lock);
2217 logical = start - em->start;
2218 logical = em->block_start + logical;
2219 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2220 logical = em->block_start;
2221 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2222 extent_set_compress_type(&failrec->bio_flags,
2226 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2227 logical, start, failrec->len);
2229 failrec->logical = logical;
2230 free_extent_map(em);
2232 /* set the bits in the private failure tree */
2233 ret = set_extent_bits(failure_tree, start, end,
2234 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2236 ret = set_state_private(failure_tree, start,
2237 (u64)(unsigned long)failrec);
2238 /* set the bits in the inode's tree */
2240 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2247 failrec = (struct io_failure_record *)(unsigned long)private;
2248 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2249 failrec->logical, failrec->start, failrec->len,
2250 failrec->in_validation);
2252 * when data can be on disk more than twice, add to failrec here
2253 * (e.g. with a list for failed_mirror) to make
2254 * clean_io_failure() clean all those errors at once.
2258 *failrec_ret = failrec;
2263 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2264 struct io_failure_record *failrec, int failed_mirror)
2268 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2269 failrec->logical, failrec->len);
2270 if (num_copies == 1) {
2272 * we only have a single copy of the data, so don't bother with
2273 * all the retry and error correction code that follows. no
2274 * matter what the error is, it is very likely to persist.
2276 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2277 num_copies, failrec->this_mirror, failed_mirror);
2282 * there are two premises:
2283 * a) deliver good data to the caller
2284 * b) correct the bad sectors on disk
2286 if (failed_bio->bi_vcnt > 1) {
2288 * to fulfill b), we need to know the exact failing sectors, as
2289 * we don't want to rewrite any more than the failed ones. thus,
2290 * we need separate read requests for the failed bio
2292 * if the following BUG_ON triggers, our validation request got
2293 * merged. we need separate requests for our algorithm to work.
2295 BUG_ON(failrec->in_validation);
2296 failrec->in_validation = 1;
2297 failrec->this_mirror = failed_mirror;
2300 * we're ready to fulfill a) and b) alongside. get a good copy
2301 * of the failed sector and if we succeed, we have setup
2302 * everything for repair_io_failure to do the rest for us.
2304 if (failrec->in_validation) {
2305 BUG_ON(failrec->this_mirror != failed_mirror);
2306 failrec->in_validation = 0;
2307 failrec->this_mirror = 0;
2309 failrec->failed_mirror = failed_mirror;
2310 failrec->this_mirror++;
2311 if (failrec->this_mirror == failed_mirror)
2312 failrec->this_mirror++;
2315 if (failrec->this_mirror > num_copies) {
2316 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2317 num_copies, failrec->this_mirror, failed_mirror);
2325 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2326 struct io_failure_record *failrec,
2327 struct page *page, int pg_offset, int icsum,
2328 bio_end_io_t *endio_func, void *data)
2331 struct btrfs_io_bio *btrfs_failed_bio;
2332 struct btrfs_io_bio *btrfs_bio;
2334 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2338 bio->bi_end_io = endio_func;
2339 bio->bi_iter.bi_sector = failrec->logical >> 9;
2340 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2341 bio->bi_iter.bi_size = 0;
2342 bio->bi_private = data;
2344 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2345 if (btrfs_failed_bio->csum) {
2346 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2347 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2349 btrfs_bio = btrfs_io_bio(bio);
2350 btrfs_bio->csum = btrfs_bio->csum_inline;
2352 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2356 bio_add_page(bio, page, failrec->len, pg_offset);
2362 * this is a generic handler for readpage errors (default
2363 * readpage_io_failed_hook). if other copies exist, read those and write back
2364 * good data to the failed position. does not investigate in remapping the
2365 * failed extent elsewhere, hoping the device will be smart enough to do this as
2369 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2370 struct page *page, u64 start, u64 end,
2373 struct io_failure_record *failrec;
2374 struct inode *inode = page->mapping->host;
2375 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2380 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2382 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2386 ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror);
2388 free_io_failure(inode, failrec);
2392 if (failed_bio->bi_vcnt > 1)
2393 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2395 read_mode = READ_SYNC;
2397 phy_offset >>= inode->i_sb->s_blocksize_bits;
2398 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2399 start - page_offset(page),
2400 (int)phy_offset, failed_bio->bi_end_io,
2403 free_io_failure(inode, failrec);
2407 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2408 read_mode, failrec->this_mirror, failrec->in_validation);
2410 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2411 failrec->this_mirror,
2412 failrec->bio_flags, 0);
2414 free_io_failure(inode, failrec);
2421 /* lots and lots of room for performance fixes in the end_bio funcs */
2423 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2425 int uptodate = (err == 0);
2426 struct extent_io_tree *tree;
2429 tree = &BTRFS_I(page->mapping->host)->io_tree;
2431 if (tree->ops && tree->ops->writepage_end_io_hook) {
2432 ret = tree->ops->writepage_end_io_hook(page, start,
2433 end, NULL, uptodate);
2439 ClearPageUptodate(page);
2441 ret = ret < 0 ? ret : -EIO;
2442 mapping_set_error(page->mapping, ret);
2448 * after a writepage IO is done, we need to:
2449 * clear the uptodate bits on error
2450 * clear the writeback bits in the extent tree for this IO
2451 * end_page_writeback if the page has no more pending IO
2453 * Scheduling is not allowed, so the extent state tree is expected
2454 * to have one and only one object corresponding to this IO.
2456 static void end_bio_extent_writepage(struct bio *bio, int err)
2458 struct bio_vec *bvec;
2463 bio_for_each_segment_all(bvec, bio, i) {
2464 struct page *page = bvec->bv_page;
2466 /* We always issue full-page reads, but if some block
2467 * in a page fails to read, blk_update_request() will
2468 * advance bv_offset and adjust bv_len to compensate.
2469 * Print a warning for nonzero offsets, and an error
2470 * if they don't add up to a full page. */
2471 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2472 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2473 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2474 "partial page write in btrfs with offset %u and length %u",
2475 bvec->bv_offset, bvec->bv_len);
2477 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2478 "incomplete page write in btrfs with offset %u and "
2480 bvec->bv_offset, bvec->bv_len);
2483 start = page_offset(page);
2484 end = start + bvec->bv_offset + bvec->bv_len - 1;
2486 if (end_extent_writepage(page, err, start, end))
2489 end_page_writeback(page);
2496 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2499 struct extent_state *cached = NULL;
2500 u64 end = start + len - 1;
2502 if (uptodate && tree->track_uptodate)
2503 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2504 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2508 * after a readpage IO is done, we need to:
2509 * clear the uptodate bits on error
2510 * set the uptodate bits if things worked
2511 * set the page up to date if all extents in the tree are uptodate
2512 * clear the lock bit in the extent tree
2513 * unlock the page if there are no other extents locked for it
2515 * Scheduling is not allowed, so the extent state tree is expected
2516 * to have one and only one object corresponding to this IO.
2518 static void end_bio_extent_readpage(struct bio *bio, int err)
2520 struct bio_vec *bvec;
2521 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2522 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2523 struct extent_io_tree *tree;
2528 u64 extent_start = 0;
2537 bio_for_each_segment_all(bvec, bio, i) {
2538 struct page *page = bvec->bv_page;
2539 struct inode *inode = page->mapping->host;
2541 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2542 "mirror=%u\n", (u64)bio->bi_iter.bi_sector, err,
2543 io_bio->mirror_num);
2544 tree = &BTRFS_I(inode)->io_tree;
2546 /* We always issue full-page reads, but if some block
2547 * in a page fails to read, blk_update_request() will
2548 * advance bv_offset and adjust bv_len to compensate.
2549 * Print a warning for nonzero offsets, and an error
2550 * if they don't add up to a full page. */
2551 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2552 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2553 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2554 "partial page read in btrfs with offset %u and length %u",
2555 bvec->bv_offset, bvec->bv_len);
2557 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2558 "incomplete page read in btrfs with offset %u and "
2560 bvec->bv_offset, bvec->bv_len);
2563 start = page_offset(page);
2564 end = start + bvec->bv_offset + bvec->bv_len - 1;
2567 mirror = io_bio->mirror_num;
2568 if (likely(uptodate && tree->ops &&
2569 tree->ops->readpage_end_io_hook)) {
2570 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2576 clean_io_failure(inode, start, page, 0);
2579 if (likely(uptodate))
2582 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2583 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2585 test_bit(BIO_UPTODATE, &bio->bi_flags))
2589 * The generic bio_readpage_error handles errors the
2590 * following way: If possible, new read requests are
2591 * created and submitted and will end up in
2592 * end_bio_extent_readpage as well (if we're lucky, not
2593 * in the !uptodate case). In that case it returns 0 and
2594 * we just go on with the next page in our bio. If it
2595 * can't handle the error it will return -EIO and we
2596 * remain responsible for that page.
2598 ret = bio_readpage_error(bio, offset, page, start, end,
2602 test_bit(BIO_UPTODATE, &bio->bi_flags);
2610 if (likely(uptodate)) {
2611 loff_t i_size = i_size_read(inode);
2612 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2615 /* Zero out the end if this page straddles i_size */
2616 off = i_size & (PAGE_CACHE_SIZE-1);
2617 if (page->index == end_index && off)
2618 zero_user_segment(page, off, PAGE_CACHE_SIZE);
2619 SetPageUptodate(page);
2621 ClearPageUptodate(page);
2627 if (unlikely(!uptodate)) {
2629 endio_readpage_release_extent(tree,
2635 endio_readpage_release_extent(tree, start,
2636 end - start + 1, 0);
2637 } else if (!extent_len) {
2638 extent_start = start;
2639 extent_len = end + 1 - start;
2640 } else if (extent_start + extent_len == start) {
2641 extent_len += end + 1 - start;
2643 endio_readpage_release_extent(tree, extent_start,
2644 extent_len, uptodate);
2645 extent_start = start;
2646 extent_len = end + 1 - start;
2651 endio_readpage_release_extent(tree, extent_start, extent_len,
2654 io_bio->end_io(io_bio, err);
2659 * this allocates from the btrfs_bioset. We're returning a bio right now
2660 * but you can call btrfs_io_bio for the appropriate container_of magic
2663 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2666 struct btrfs_io_bio *btrfs_bio;
2669 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2671 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2672 while (!bio && (nr_vecs /= 2)) {
2673 bio = bio_alloc_bioset(gfp_flags,
2674 nr_vecs, btrfs_bioset);
2679 bio->bi_bdev = bdev;
2680 bio->bi_iter.bi_sector = first_sector;
2681 btrfs_bio = btrfs_io_bio(bio);
2682 btrfs_bio->csum = NULL;
2683 btrfs_bio->csum_allocated = NULL;
2684 btrfs_bio->end_io = NULL;
2689 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2691 struct btrfs_io_bio *btrfs_bio;
2694 new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2696 btrfs_bio = btrfs_io_bio(new);
2697 btrfs_bio->csum = NULL;
2698 btrfs_bio->csum_allocated = NULL;
2699 btrfs_bio->end_io = NULL;
2704 /* this also allocates from the btrfs_bioset */
2705 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2707 struct btrfs_io_bio *btrfs_bio;
2710 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2712 btrfs_bio = btrfs_io_bio(bio);
2713 btrfs_bio->csum = NULL;
2714 btrfs_bio->csum_allocated = NULL;
2715 btrfs_bio->end_io = NULL;
2721 static int __must_check submit_one_bio(int rw, struct bio *bio,
2722 int mirror_num, unsigned long bio_flags)
2725 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2726 struct page *page = bvec->bv_page;
2727 struct extent_io_tree *tree = bio->bi_private;
2730 start = page_offset(page) + bvec->bv_offset;
2732 bio->bi_private = NULL;
2736 if (tree->ops && tree->ops->submit_bio_hook)
2737 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2738 mirror_num, bio_flags, start);
2740 btrfsic_submit_bio(rw, bio);
2742 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2748 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2749 unsigned long offset, size_t size, struct bio *bio,
2750 unsigned long bio_flags)
2753 if (tree->ops && tree->ops->merge_bio_hook)
2754 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2761 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2762 struct page *page, sector_t sector,
2763 size_t size, unsigned long offset,
2764 struct block_device *bdev,
2765 struct bio **bio_ret,
2766 unsigned long max_pages,
2767 bio_end_io_t end_io_func,
2769 unsigned long prev_bio_flags,
2770 unsigned long bio_flags)
2776 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2777 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2778 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2780 if (bio_ret && *bio_ret) {
2783 contig = bio->bi_iter.bi_sector == sector;
2785 contig = bio_end_sector(bio) == sector;
2787 if (prev_bio_flags != bio_flags || !contig ||
2788 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2789 bio_add_page(bio, page, page_size, offset) < page_size) {
2790 ret = submit_one_bio(rw, bio, mirror_num,
2799 if (this_compressed)
2802 nr = bio_get_nr_vecs(bdev);
2804 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2808 bio_add_page(bio, page, page_size, offset);
2809 bio->bi_end_io = end_io_func;
2810 bio->bi_private = tree;
2815 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2820 static void attach_extent_buffer_page(struct extent_buffer *eb,
2823 if (!PagePrivate(page)) {
2824 SetPagePrivate(page);
2825 page_cache_get(page);
2826 set_page_private(page, (unsigned long)eb);
2828 WARN_ON(page->private != (unsigned long)eb);
2832 void set_page_extent_mapped(struct page *page)
2834 if (!PagePrivate(page)) {
2835 SetPagePrivate(page);
2836 page_cache_get(page);
2837 set_page_private(page, EXTENT_PAGE_PRIVATE);
2841 static struct extent_map *
2842 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2843 u64 start, u64 len, get_extent_t *get_extent,
2844 struct extent_map **em_cached)
2846 struct extent_map *em;
2848 if (em_cached && *em_cached) {
2850 if (extent_map_in_tree(em) && start >= em->start &&
2851 start < extent_map_end(em)) {
2852 atomic_inc(&em->refs);
2856 free_extent_map(em);
2860 em = get_extent(inode, page, pg_offset, start, len, 0);
2861 if (em_cached && !IS_ERR_OR_NULL(em)) {
2863 atomic_inc(&em->refs);
2869 * basic readpage implementation. Locked extent state structs are inserted
2870 * into the tree that are removed when the IO is done (by the end_io
2872 * XXX JDM: This needs looking at to ensure proper page locking
2874 static int __do_readpage(struct extent_io_tree *tree,
2876 get_extent_t *get_extent,
2877 struct extent_map **em_cached,
2878 struct bio **bio, int mirror_num,
2879 unsigned long *bio_flags, int rw)
2881 struct inode *inode = page->mapping->host;
2882 u64 start = page_offset(page);
2883 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2887 u64 last_byte = i_size_read(inode);
2891 struct extent_map *em;
2892 struct block_device *bdev;
2895 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2896 size_t pg_offset = 0;
2898 size_t disk_io_size;
2899 size_t blocksize = inode->i_sb->s_blocksize;
2900 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2902 set_page_extent_mapped(page);
2905 if (!PageUptodate(page)) {
2906 if (cleancache_get_page(page) == 0) {
2907 BUG_ON(blocksize != PAGE_SIZE);
2908 unlock_extent(tree, start, end);
2913 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2915 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2918 iosize = PAGE_CACHE_SIZE - zero_offset;
2919 userpage = kmap_atomic(page);
2920 memset(userpage + zero_offset, 0, iosize);
2921 flush_dcache_page(page);
2922 kunmap_atomic(userpage);
2925 while (cur <= end) {
2926 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2928 if (cur >= last_byte) {
2930 struct extent_state *cached = NULL;
2932 iosize = PAGE_CACHE_SIZE - pg_offset;
2933 userpage = kmap_atomic(page);
2934 memset(userpage + pg_offset, 0, iosize);
2935 flush_dcache_page(page);
2936 kunmap_atomic(userpage);
2937 set_extent_uptodate(tree, cur, cur + iosize - 1,
2940 unlock_extent_cached(tree, cur,
2945 em = __get_extent_map(inode, page, pg_offset, cur,
2946 end - cur + 1, get_extent, em_cached);
2947 if (IS_ERR_OR_NULL(em)) {
2950 unlock_extent(tree, cur, end);
2953 extent_offset = cur - em->start;
2954 BUG_ON(extent_map_end(em) <= cur);
2957 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2958 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2959 extent_set_compress_type(&this_bio_flag,
2963 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2964 cur_end = min(extent_map_end(em) - 1, end);
2965 iosize = ALIGN(iosize, blocksize);
2966 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2967 disk_io_size = em->block_len;
2968 sector = em->block_start >> 9;
2970 sector = (em->block_start + extent_offset) >> 9;
2971 disk_io_size = iosize;
2974 block_start = em->block_start;
2975 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2976 block_start = EXTENT_MAP_HOLE;
2977 free_extent_map(em);
2980 /* we've found a hole, just zero and go on */
2981 if (block_start == EXTENT_MAP_HOLE) {
2983 struct extent_state *cached = NULL;
2985 userpage = kmap_atomic(page);
2986 memset(userpage + pg_offset, 0, iosize);
2987 flush_dcache_page(page);
2988 kunmap_atomic(userpage);
2990 set_extent_uptodate(tree, cur, cur + iosize - 1,
2992 unlock_extent_cached(tree, cur, cur + iosize - 1,
2995 pg_offset += iosize;
2998 /* the get_extent function already copied into the page */
2999 if (test_range_bit(tree, cur, cur_end,
3000 EXTENT_UPTODATE, 1, NULL)) {
3001 check_page_uptodate(tree, page);
3003 unlock_extent(tree, cur, cur + iosize - 1);
3005 pg_offset += iosize;
3008 /* we have an inline extent but it didn't get marked up
3009 * to date. Error out
3011 if (block_start == EXTENT_MAP_INLINE) {
3014 unlock_extent(tree, cur, cur + iosize - 1);
3016 pg_offset += iosize;
3021 ret = submit_extent_page(rw, tree, page,
3022 sector, disk_io_size, pg_offset,
3024 end_bio_extent_readpage, mirror_num,
3029 *bio_flags = this_bio_flag;
3033 unlock_extent(tree, cur, cur + iosize - 1);
3036 pg_offset += iosize;
3040 if (!PageError(page))
3041 SetPageUptodate(page);
3047 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3048 struct page *pages[], int nr_pages,
3050 get_extent_t *get_extent,
3051 struct extent_map **em_cached,
3052 struct bio **bio, int mirror_num,
3053 unsigned long *bio_flags, int rw)
3055 struct inode *inode;
3056 struct btrfs_ordered_extent *ordered;
3059 inode = pages[0]->mapping->host;
3061 lock_extent(tree, start, end);
3062 ordered = btrfs_lookup_ordered_range(inode, start,
3066 unlock_extent(tree, start, end);
3067 btrfs_start_ordered_extent(inode, ordered, 1);
3068 btrfs_put_ordered_extent(ordered);
3071 for (index = 0; index < nr_pages; index++) {
3072 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3073 mirror_num, bio_flags, rw);
3074 page_cache_release(pages[index]);
3078 static void __extent_readpages(struct extent_io_tree *tree,
3079 struct page *pages[],
3080 int nr_pages, get_extent_t *get_extent,
3081 struct extent_map **em_cached,
3082 struct bio **bio, int mirror_num,
3083 unsigned long *bio_flags, int rw)
3089 int first_index = 0;
3091 for (index = 0; index < nr_pages; index++) {
3092 page_start = page_offset(pages[index]);
3095 end = start + PAGE_CACHE_SIZE - 1;
3096 first_index = index;
3097 } else if (end + 1 == page_start) {
3098 end += PAGE_CACHE_SIZE;
3100 __do_contiguous_readpages(tree, &pages[first_index],
3101 index - first_index, start,
3102 end, get_extent, em_cached,
3103 bio, mirror_num, bio_flags,
3106 end = start + PAGE_CACHE_SIZE - 1;
3107 first_index = index;
3112 __do_contiguous_readpages(tree, &pages[first_index],
3113 index - first_index, start,
3114 end, get_extent, em_cached, bio,
3115 mirror_num, bio_flags, rw);
3118 static int __extent_read_full_page(struct extent_io_tree *tree,
3120 get_extent_t *get_extent,
3121 struct bio **bio, int mirror_num,
3122 unsigned long *bio_flags, int rw)
3124 struct inode *inode = page->mapping->host;
3125 struct btrfs_ordered_extent *ordered;
3126 u64 start = page_offset(page);
3127 u64 end = start + PAGE_CACHE_SIZE - 1;
3131 lock_extent(tree, start, end);
3132 ordered = btrfs_lookup_ordered_extent(inode, start);
3135 unlock_extent(tree, start, end);
3136 btrfs_start_ordered_extent(inode, ordered, 1);
3137 btrfs_put_ordered_extent(ordered);
3140 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3145 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3146 get_extent_t *get_extent, int mirror_num)
3148 struct bio *bio = NULL;
3149 unsigned long bio_flags = 0;
3152 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3155 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3159 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3160 get_extent_t *get_extent, int mirror_num)
3162 struct bio *bio = NULL;
3163 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3166 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3169 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3173 static noinline void update_nr_written(struct page *page,
3174 struct writeback_control *wbc,
3175 unsigned long nr_written)
3177 wbc->nr_to_write -= nr_written;
3178 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3179 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3180 page->mapping->writeback_index = page->index + nr_written;
3184 * helper for __extent_writepage, doing all of the delayed allocation setup.
3186 * This returns 1 if our fill_delalloc function did all the work required
3187 * to write the page (copy into inline extent). In this case the IO has
3188 * been started and the page is already unlocked.
3190 * This returns 0 if all went well (page still locked)
3191 * This returns < 0 if there were errors (page still locked)
3193 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3194 struct page *page, struct writeback_control *wbc,
3195 struct extent_page_data *epd,
3197 unsigned long *nr_written)
3199 struct extent_io_tree *tree = epd->tree;
3200 u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1;
3202 u64 delalloc_to_write = 0;
3203 u64 delalloc_end = 0;
3205 int page_started = 0;
3207 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3210 while (delalloc_end < page_end) {
3211 nr_delalloc = find_lock_delalloc_range(inode, tree,
3216 if (nr_delalloc == 0) {
3217 delalloc_start = delalloc_end + 1;
3220 ret = tree->ops->fill_delalloc(inode, page,
3225 /* File system has been set read-only */
3228 /* fill_delalloc should be return < 0 for error
3229 * but just in case, we use > 0 here meaning the
3230 * IO is started, so we don't want to return > 0
3231 * unless things are going well.
3233 ret = ret < 0 ? ret : -EIO;
3237 * delalloc_end is already one less than the total
3238 * length, so we don't subtract one from
3241 delalloc_to_write += (delalloc_end - delalloc_start +
3244 delalloc_start = delalloc_end + 1;
3246 if (wbc->nr_to_write < delalloc_to_write) {
3249 if (delalloc_to_write < thresh * 2)
3250 thresh = delalloc_to_write;
3251 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3255 /* did the fill delalloc function already unlock and start
3260 * we've unlocked the page, so we can't update
3261 * the mapping's writeback index, just update
3264 wbc->nr_to_write -= *nr_written;
3275 * helper for __extent_writepage. This calls the writepage start hooks,
3276 * and does the loop to map the page into extents and bios.
3278 * We return 1 if the IO is started and the page is unlocked,
3279 * 0 if all went well (page still locked)
3280 * < 0 if there were errors (page still locked)
3282 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3284 struct writeback_control *wbc,
3285 struct extent_page_data *epd,
3287 unsigned long nr_written,
3288 int write_flags, int *nr_ret)
3290 struct extent_io_tree *tree = epd->tree;
3291 u64 start = page_offset(page);
3292 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3299 struct extent_state *cached_state = NULL;
3300 struct extent_map *em;
3301 struct block_device *bdev;
3302 size_t pg_offset = 0;
3308 if (tree->ops && tree->ops->writepage_start_hook) {
3309 ret = tree->ops->writepage_start_hook(page, start,
3312 /* Fixup worker will requeue */
3314 wbc->pages_skipped++;
3316 redirty_page_for_writepage(wbc, page);
3318 update_nr_written(page, wbc, nr_written);
3326 * we don't want to touch the inode after unlocking the page,
3327 * so we update the mapping writeback index now
3329 update_nr_written(page, wbc, nr_written + 1);
3332 if (i_size <= start) {
3333 if (tree->ops && tree->ops->writepage_end_io_hook)
3334 tree->ops->writepage_end_io_hook(page, start,
3339 blocksize = inode->i_sb->s_blocksize;
3341 while (cur <= end) {
3343 if (cur >= i_size) {
3344 if (tree->ops && tree->ops->writepage_end_io_hook)
3345 tree->ops->writepage_end_io_hook(page, cur,
3349 em = epd->get_extent(inode, page, pg_offset, cur,
3351 if (IS_ERR_OR_NULL(em)) {
3353 ret = PTR_ERR_OR_ZERO(em);
3357 extent_offset = cur - em->start;
3358 em_end = extent_map_end(em);
3359 BUG_ON(em_end <= cur);
3361 iosize = min(em_end - cur, end - cur + 1);
3362 iosize = ALIGN(iosize, blocksize);
3363 sector = (em->block_start + extent_offset) >> 9;
3365 block_start = em->block_start;
3366 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3367 free_extent_map(em);
3371 * compressed and inline extents are written through other
3374 if (compressed || block_start == EXTENT_MAP_HOLE ||
3375 block_start == EXTENT_MAP_INLINE) {
3377 * end_io notification does not happen here for
3378 * compressed extents
3380 if (!compressed && tree->ops &&
3381 tree->ops->writepage_end_io_hook)
3382 tree->ops->writepage_end_io_hook(page, cur,
3385 else if (compressed) {
3386 /* we don't want to end_page_writeback on
3387 * a compressed extent. this happens
3394 pg_offset += iosize;
3398 if (tree->ops && tree->ops->writepage_io_hook) {
3399 ret = tree->ops->writepage_io_hook(page, cur,
3407 unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1;
3409 set_range_writeback(tree, cur, cur + iosize - 1);
3410 if (!PageWriteback(page)) {
3411 btrfs_err(BTRFS_I(inode)->root->fs_info,
3412 "page %lu not writeback, cur %llu end %llu",
3413 page->index, cur, end);
3416 ret = submit_extent_page(write_flags, tree, page,
3417 sector, iosize, pg_offset,
3418 bdev, &epd->bio, max_nr,
3419 end_bio_extent_writepage,
3425 pg_offset += iosize;
3433 /* drop our reference on any cached states */
3434 free_extent_state(cached_state);
3439 * the writepage semantics are similar to regular writepage. extent
3440 * records are inserted to lock ranges in the tree, and as dirty areas
3441 * are found, they are marked writeback. Then the lock bits are removed
3442 * and the end_io handler clears the writeback ranges
3444 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3447 struct inode *inode = page->mapping->host;
3448 struct extent_page_data *epd = data;
3449 u64 start = page_offset(page);
3450 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3453 size_t pg_offset = 0;
3454 loff_t i_size = i_size_read(inode);
3455 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3457 unsigned long nr_written = 0;
3459 if (wbc->sync_mode == WB_SYNC_ALL)
3460 write_flags = WRITE_SYNC;
3462 write_flags = WRITE;
3464 trace___extent_writepage(page, inode, wbc);
3466 WARN_ON(!PageLocked(page));
3468 ClearPageError(page);
3470 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3471 if (page->index > end_index ||
3472 (page->index == end_index && !pg_offset)) {
3473 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3478 if (page->index == end_index) {
3481 userpage = kmap_atomic(page);
3482 memset(userpage + pg_offset, 0,
3483 PAGE_CACHE_SIZE - pg_offset);
3484 kunmap_atomic(userpage);
3485 flush_dcache_page(page);
3490 set_page_extent_mapped(page);
3492 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3498 ret = __extent_writepage_io(inode, page, wbc, epd,
3499 i_size, nr_written, write_flags, &nr);
3505 /* make sure the mapping tag for page dirty gets cleared */
3506 set_page_writeback(page);
3507 end_page_writeback(page);
3509 if (PageError(page)) {
3510 ret = ret < 0 ? ret : -EIO;
3511 end_extent_writepage(page, ret, start, page_end);
3520 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3522 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3523 TASK_UNINTERRUPTIBLE);
3526 static noinline_for_stack int
3527 lock_extent_buffer_for_io(struct extent_buffer *eb,
3528 struct btrfs_fs_info *fs_info,
3529 struct extent_page_data *epd)
3531 unsigned long i, num_pages;
3535 if (!btrfs_try_tree_write_lock(eb)) {
3537 flush_write_bio(epd);
3538 btrfs_tree_lock(eb);
3541 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3542 btrfs_tree_unlock(eb);
3546 flush_write_bio(epd);
3550 wait_on_extent_buffer_writeback(eb);
3551 btrfs_tree_lock(eb);
3552 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3554 btrfs_tree_unlock(eb);
3559 * We need to do this to prevent races in people who check if the eb is
3560 * under IO since we can end up having no IO bits set for a short period
3563 spin_lock(&eb->refs_lock);
3564 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3565 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3566 spin_unlock(&eb->refs_lock);
3567 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3568 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3570 fs_info->dirty_metadata_batch);
3573 spin_unlock(&eb->refs_lock);
3576 btrfs_tree_unlock(eb);
3581 num_pages = num_extent_pages(eb->start, eb->len);
3582 for (i = 0; i < num_pages; i++) {
3583 struct page *p = eb->pages[i];
3585 if (!trylock_page(p)) {
3587 flush_write_bio(epd);
3597 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3599 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3600 smp_mb__after_atomic();
3601 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3604 static void set_btree_ioerr(struct page *page)
3606 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3607 struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode);
3610 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3614 * If writeback for a btree extent that doesn't belong to a log tree
3615 * failed, increment the counter transaction->eb_write_errors.
3616 * We do this because while the transaction is running and before it's
3617 * committing (when we call filemap_fdata[write|wait]_range against
3618 * the btree inode), we might have
3619 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3620 * returns an error or an error happens during writeback, when we're
3621 * committing the transaction we wouldn't know about it, since the pages
3622 * can be no longer dirty nor marked anymore for writeback (if a
3623 * subsequent modification to the extent buffer didn't happen before the
3624 * transaction commit), which makes filemap_fdata[write|wait]_range not
3625 * able to find the pages tagged with SetPageError at transaction
3626 * commit time. So if this happens we must abort the transaction,
3627 * otherwise we commit a super block with btree roots that point to
3628 * btree nodes/leafs whose content on disk is invalid - either garbage
3629 * or the content of some node/leaf from a past generation that got
3630 * cowed or deleted and is no longer valid.
3632 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3633 * not be enough - we need to distinguish between log tree extents vs
3634 * non-log tree extents, and the next filemap_fdatawait_range() call
3635 * will catch and clear such errors in the mapping - and that call might
3636 * be from a log sync and not from a transaction commit. Also, checking
3637 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3638 * not done and would not be reliable - the eb might have been released
3639 * from memory and reading it back again means that flag would not be
3640 * set (since it's a runtime flag, not persisted on disk).
3642 * Using the flags below in the btree inode also makes us achieve the
3643 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3644 * writeback for all dirty pages and before filemap_fdatawait_range()
3645 * is called, the writeback for all dirty pages had already finished
3646 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3647 * filemap_fdatawait_range() would return success, as it could not know
3648 * that writeback errors happened (the pages were no longer tagged for
3651 switch (eb->log_index) {
3653 set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags);
3656 set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
3659 set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
3662 BUG(); /* unexpected, logic error */
3666 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3668 struct bio_vec *bvec;
3669 struct extent_buffer *eb;
3672 bio_for_each_segment_all(bvec, bio, i) {
3673 struct page *page = bvec->bv_page;
3675 eb = (struct extent_buffer *)page->private;
3677 done = atomic_dec_and_test(&eb->io_pages);
3679 if (err || test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3680 ClearPageUptodate(page);
3681 set_btree_ioerr(page);
3684 end_page_writeback(page);
3689 end_extent_buffer_writeback(eb);
3695 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3696 struct btrfs_fs_info *fs_info,
3697 struct writeback_control *wbc,
3698 struct extent_page_data *epd)
3700 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3701 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3702 u64 offset = eb->start;
3703 unsigned long i, num_pages;
3704 unsigned long bio_flags = 0;
3705 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3708 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3709 num_pages = num_extent_pages(eb->start, eb->len);
3710 atomic_set(&eb->io_pages, num_pages);
3711 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3712 bio_flags = EXTENT_BIO_TREE_LOG;
3714 for (i = 0; i < num_pages; i++) {
3715 struct page *p = eb->pages[i];
3717 clear_page_dirty_for_io(p);
3718 set_page_writeback(p);
3719 ret = submit_extent_page(rw, tree, p, offset >> 9,
3720 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3721 -1, end_bio_extent_buffer_writepage,
3722 0, epd->bio_flags, bio_flags);
3723 epd->bio_flags = bio_flags;
3726 end_page_writeback(p);
3727 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3728 end_extent_buffer_writeback(eb);
3732 offset += PAGE_CACHE_SIZE;
3733 update_nr_written(p, wbc, 1);
3737 if (unlikely(ret)) {
3738 for (; i < num_pages; i++) {
3739 struct page *p = eb->pages[i];
3740 clear_page_dirty_for_io(p);
3748 int btree_write_cache_pages(struct address_space *mapping,
3749 struct writeback_control *wbc)
3751 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3752 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3753 struct extent_buffer *eb, *prev_eb = NULL;
3754 struct extent_page_data epd = {
3758 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3763 int nr_to_write_done = 0;
3764 struct pagevec pvec;
3767 pgoff_t end; /* Inclusive */
3771 pagevec_init(&pvec, 0);
3772 if (wbc->range_cyclic) {
3773 index = mapping->writeback_index; /* Start from prev offset */
3776 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3777 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3780 if (wbc->sync_mode == WB_SYNC_ALL)
3781 tag = PAGECACHE_TAG_TOWRITE;
3783 tag = PAGECACHE_TAG_DIRTY;
3785 if (wbc->sync_mode == WB_SYNC_ALL)
3786 tag_pages_for_writeback(mapping, index, end);
3787 while (!done && !nr_to_write_done && (index <= end) &&
3788 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3789 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3793 for (i = 0; i < nr_pages; i++) {
3794 struct page *page = pvec.pages[i];
3796 if (!PagePrivate(page))
3799 if (!wbc->range_cyclic && page->index > end) {
3804 spin_lock(&mapping->private_lock);
3805 if (!PagePrivate(page)) {
3806 spin_unlock(&mapping->private_lock);
3810 eb = (struct extent_buffer *)page->private;
3813 * Shouldn't happen and normally this would be a BUG_ON
3814 * but no sense in crashing the users box for something
3815 * we can survive anyway.
3818 spin_unlock(&mapping->private_lock);
3822 if (eb == prev_eb) {
3823 spin_unlock(&mapping->private_lock);
3827 ret = atomic_inc_not_zero(&eb->refs);
3828 spin_unlock(&mapping->private_lock);
3833 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3835 free_extent_buffer(eb);
3839 ret = write_one_eb(eb, fs_info, wbc, &epd);
3842 free_extent_buffer(eb);
3845 free_extent_buffer(eb);
3848 * the filesystem may choose to bump up nr_to_write.
3849 * We have to make sure to honor the new nr_to_write
3852 nr_to_write_done = wbc->nr_to_write <= 0;
3854 pagevec_release(&pvec);
3857 if (!scanned && !done) {
3859 * We hit the last page and there is more work to be done: wrap
3860 * back to the start of the file
3866 flush_write_bio(&epd);
3871 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3872 * @mapping: address space structure to write
3873 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3874 * @writepage: function called for each page
3875 * @data: data passed to writepage function
3877 * If a page is already under I/O, write_cache_pages() skips it, even
3878 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3879 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3880 * and msync() need to guarantee that all the data which was dirty at the time
3881 * the call was made get new I/O started against them. If wbc->sync_mode is
3882 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3883 * existing IO to complete.
3885 static int extent_write_cache_pages(struct extent_io_tree *tree,
3886 struct address_space *mapping,
3887 struct writeback_control *wbc,
3888 writepage_t writepage, void *data,
3889 void (*flush_fn)(void *))
3891 struct inode *inode = mapping->host;
3895 int nr_to_write_done = 0;
3896 struct pagevec pvec;
3899 pgoff_t end; /* Inclusive */
3904 * We have to hold onto the inode so that ordered extents can do their
3905 * work when the IO finishes. The alternative to this is failing to add
3906 * an ordered extent if the igrab() fails there and that is a huge pain
3907 * to deal with, so instead just hold onto the inode throughout the
3908 * writepages operation. If it fails here we are freeing up the inode
3909 * anyway and we'd rather not waste our time writing out stuff that is
3910 * going to be truncated anyway.
3915 pagevec_init(&pvec, 0);
3916 if (wbc->range_cyclic) {
3917 index = mapping->writeback_index; /* Start from prev offset */
3920 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3921 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3924 if (wbc->sync_mode == WB_SYNC_ALL)
3925 tag = PAGECACHE_TAG_TOWRITE;
3927 tag = PAGECACHE_TAG_DIRTY;
3929 if (wbc->sync_mode == WB_SYNC_ALL)
3930 tag_pages_for_writeback(mapping, index, end);
3931 while (!done && !nr_to_write_done && (index <= end) &&
3932 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3933 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3937 for (i = 0; i < nr_pages; i++) {
3938 struct page *page = pvec.pages[i];
3941 * At this point we hold neither mapping->tree_lock nor
3942 * lock on the page itself: the page may be truncated or
3943 * invalidated (changing page->mapping to NULL), or even
3944 * swizzled back from swapper_space to tmpfs file
3947 if (!trylock_page(page)) {
3952 if (unlikely(page->mapping != mapping)) {
3957 if (!wbc->range_cyclic && page->index > end) {
3963 if (wbc->sync_mode != WB_SYNC_NONE) {
3964 if (PageWriteback(page))
3966 wait_on_page_writeback(page);
3969 if (PageWriteback(page) ||
3970 !clear_page_dirty_for_io(page)) {
3975 ret = (*writepage)(page, wbc, data);
3977 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3981 if (!err && ret < 0)
3985 * the filesystem may choose to bump up nr_to_write.
3986 * We have to make sure to honor the new nr_to_write
3989 nr_to_write_done = wbc->nr_to_write <= 0;
3991 pagevec_release(&pvec);
3994 if (!scanned && !done && !err) {
3996 * We hit the last page and there is more work to be done: wrap
3997 * back to the start of the file
4003 btrfs_add_delayed_iput(inode);
4007 static void flush_epd_write_bio(struct extent_page_data *epd)
4016 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
4017 BUG_ON(ret < 0); /* -ENOMEM */
4022 static noinline void flush_write_bio(void *data)
4024 struct extent_page_data *epd = data;
4025 flush_epd_write_bio(epd);
4028 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4029 get_extent_t *get_extent,
4030 struct writeback_control *wbc)
4033 struct extent_page_data epd = {
4036 .get_extent = get_extent,
4038 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4042 ret = __extent_writepage(page, wbc, &epd);
4044 flush_epd_write_bio(&epd);
4048 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4049 u64 start, u64 end, get_extent_t *get_extent,
4053 struct address_space *mapping = inode->i_mapping;
4055 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
4058 struct extent_page_data epd = {
4061 .get_extent = get_extent,
4063 .sync_io = mode == WB_SYNC_ALL,
4066 struct writeback_control wbc_writepages = {
4068 .nr_to_write = nr_pages * 2,
4069 .range_start = start,
4070 .range_end = end + 1,
4073 while (start <= end) {
4074 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
4075 if (clear_page_dirty_for_io(page))
4076 ret = __extent_writepage(page, &wbc_writepages, &epd);
4078 if (tree->ops && tree->ops->writepage_end_io_hook)
4079 tree->ops->writepage_end_io_hook(page, start,
4080 start + PAGE_CACHE_SIZE - 1,
4084 page_cache_release(page);
4085 start += PAGE_CACHE_SIZE;
4088 flush_epd_write_bio(&epd);
4092 int extent_writepages(struct extent_io_tree *tree,
4093 struct address_space *mapping,
4094 get_extent_t *get_extent,
4095 struct writeback_control *wbc)
4098 struct extent_page_data epd = {
4101 .get_extent = get_extent,
4103 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4107 ret = extent_write_cache_pages(tree, mapping, wbc,
4108 __extent_writepage, &epd,
4110 flush_epd_write_bio(&epd);
4114 int extent_readpages(struct extent_io_tree *tree,
4115 struct address_space *mapping,
4116 struct list_head *pages, unsigned nr_pages,
4117 get_extent_t get_extent)
4119 struct bio *bio = NULL;
4121 unsigned long bio_flags = 0;
4122 struct page *pagepool[16];
4124 struct extent_map *em_cached = NULL;
4127 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4128 page = list_entry(pages->prev, struct page, lru);
4130 prefetchw(&page->flags);
4131 list_del(&page->lru);
4132 if (add_to_page_cache_lru(page, mapping,
4133 page->index, GFP_NOFS)) {
4134 page_cache_release(page);
4138 pagepool[nr++] = page;
4139 if (nr < ARRAY_SIZE(pagepool))
4141 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4142 &bio, 0, &bio_flags, READ);
4146 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4147 &bio, 0, &bio_flags, READ);
4150 free_extent_map(em_cached);
4152 BUG_ON(!list_empty(pages));
4154 return submit_one_bio(READ, bio, 0, bio_flags);
4159 * basic invalidatepage code, this waits on any locked or writeback
4160 * ranges corresponding to the page, and then deletes any extent state
4161 * records from the tree
4163 int extent_invalidatepage(struct extent_io_tree *tree,
4164 struct page *page, unsigned long offset)
4166 struct extent_state *cached_state = NULL;
4167 u64 start = page_offset(page);
4168 u64 end = start + PAGE_CACHE_SIZE - 1;
4169 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4171 start += ALIGN(offset, blocksize);
4175 lock_extent_bits(tree, start, end, 0, &cached_state);
4176 wait_on_page_writeback(page);
4177 clear_extent_bit(tree, start, end,
4178 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4179 EXTENT_DO_ACCOUNTING,
4180 1, 1, &cached_state, GFP_NOFS);
4185 * a helper for releasepage, this tests for areas of the page that
4186 * are locked or under IO and drops the related state bits if it is safe
4189 static int try_release_extent_state(struct extent_map_tree *map,
4190 struct extent_io_tree *tree,
4191 struct page *page, gfp_t mask)
4193 u64 start = page_offset(page);
4194 u64 end = start + PAGE_CACHE_SIZE - 1;
4197 if (test_range_bit(tree, start, end,
4198 EXTENT_IOBITS, 0, NULL))
4201 if ((mask & GFP_NOFS) == GFP_NOFS)
4204 * at this point we can safely clear everything except the
4205 * locked bit and the nodatasum bit
4207 ret = clear_extent_bit(tree, start, end,
4208 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4211 /* if clear_extent_bit failed for enomem reasons,
4212 * we can't allow the release to continue.
4223 * a helper for releasepage. As long as there are no locked extents
4224 * in the range corresponding to the page, both state records and extent
4225 * map records are removed
4227 int try_release_extent_mapping(struct extent_map_tree *map,
4228 struct extent_io_tree *tree, struct page *page,
4231 struct extent_map *em;
4232 u64 start = page_offset(page);
4233 u64 end = start + PAGE_CACHE_SIZE - 1;
4235 if ((mask & __GFP_WAIT) &&
4236 page->mapping->host->i_size > 16 * 1024 * 1024) {
4238 while (start <= end) {
4239 len = end - start + 1;
4240 write_lock(&map->lock);
4241 em = lookup_extent_mapping(map, start, len);
4243 write_unlock(&map->lock);
4246 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4247 em->start != start) {
4248 write_unlock(&map->lock);
4249 free_extent_map(em);
4252 if (!test_range_bit(tree, em->start,
4253 extent_map_end(em) - 1,
4254 EXTENT_LOCKED | EXTENT_WRITEBACK,
4256 remove_extent_mapping(map, em);
4257 /* once for the rb tree */
4258 free_extent_map(em);
4260 start = extent_map_end(em);
4261 write_unlock(&map->lock);
4264 free_extent_map(em);
4267 return try_release_extent_state(map, tree, page, mask);
4271 * helper function for fiemap, which doesn't want to see any holes.
4272 * This maps until we find something past 'last'
4274 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4277 get_extent_t *get_extent)
4279 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4280 struct extent_map *em;
4287 len = last - offset;
4290 len = ALIGN(len, sectorsize);
4291 em = get_extent(inode, NULL, 0, offset, len, 0);
4292 if (IS_ERR_OR_NULL(em))
4295 /* if this isn't a hole return it */
4296 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4297 em->block_start != EXTENT_MAP_HOLE) {
4301 /* this is a hole, advance to the next extent */
4302 offset = extent_map_end(em);
4303 free_extent_map(em);
4310 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4311 __u64 start, __u64 len, get_extent_t *get_extent)
4315 u64 max = start + len;
4319 u64 last_for_get_extent = 0;
4321 u64 isize = i_size_read(inode);
4322 struct btrfs_key found_key;
4323 struct extent_map *em = NULL;
4324 struct extent_state *cached_state = NULL;
4325 struct btrfs_path *path;
4326 struct btrfs_root *root = BTRFS_I(inode)->root;
4335 path = btrfs_alloc_path();
4338 path->leave_spinning = 1;
4340 start = round_down(start, BTRFS_I(inode)->root->sectorsize);
4341 len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start;
4344 * lookup the last file extent. We're not using i_size here
4345 * because there might be preallocation past i_size
4347 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4350 btrfs_free_path(path);
4355 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4356 found_type = found_key.type;
4358 /* No extents, but there might be delalloc bits */
4359 if (found_key.objectid != btrfs_ino(inode) ||
4360 found_type != BTRFS_EXTENT_DATA_KEY) {
4361 /* have to trust i_size as the end */
4363 last_for_get_extent = isize;
4366 * remember the start of the last extent. There are a
4367 * bunch of different factors that go into the length of the
4368 * extent, so its much less complex to remember where it started
4370 last = found_key.offset;
4371 last_for_get_extent = last + 1;
4373 btrfs_release_path(path);
4376 * we might have some extents allocated but more delalloc past those
4377 * extents. so, we trust isize unless the start of the last extent is
4382 last_for_get_extent = isize;
4385 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4388 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4398 u64 offset_in_extent = 0;
4400 /* break if the extent we found is outside the range */
4401 if (em->start >= max || extent_map_end(em) < off)
4405 * get_extent may return an extent that starts before our
4406 * requested range. We have to make sure the ranges
4407 * we return to fiemap always move forward and don't
4408 * overlap, so adjust the offsets here
4410 em_start = max(em->start, off);
4413 * record the offset from the start of the extent
4414 * for adjusting the disk offset below. Only do this if the
4415 * extent isn't compressed since our in ram offset may be past
4416 * what we have actually allocated on disk.
4418 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4419 offset_in_extent = em_start - em->start;
4420 em_end = extent_map_end(em);
4421 em_len = em_end - em_start;
4426 * bump off for our next call to get_extent
4428 off = extent_map_end(em);
4432 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4434 flags |= FIEMAP_EXTENT_LAST;
4435 } else if (em->block_start == EXTENT_MAP_INLINE) {
4436 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4437 FIEMAP_EXTENT_NOT_ALIGNED);
4438 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4439 flags |= (FIEMAP_EXTENT_DELALLOC |
4440 FIEMAP_EXTENT_UNKNOWN);
4441 } else if (fieinfo->fi_extents_max) {
4442 u64 bytenr = em->block_start -
4443 (em->start - em->orig_start);
4445 disko = em->block_start + offset_in_extent;
4448 * As btrfs supports shared space, this information
4449 * can be exported to userspace tools via
4450 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4451 * then we're just getting a count and we can skip the
4454 ret = btrfs_check_shared(NULL, root->fs_info,
4456 btrfs_ino(inode), bytenr);
4460 flags |= FIEMAP_EXTENT_SHARED;
4463 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4464 flags |= FIEMAP_EXTENT_ENCODED;
4466 free_extent_map(em);
4468 if ((em_start >= last) || em_len == (u64)-1 ||
4469 (last == (u64)-1 && isize <= em_end)) {
4470 flags |= FIEMAP_EXTENT_LAST;
4474 /* now scan forward to see if this is really the last extent. */
4475 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4482 flags |= FIEMAP_EXTENT_LAST;
4485 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4491 free_extent_map(em);
4493 btrfs_free_path(path);
4494 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4495 &cached_state, GFP_NOFS);
4499 static void __free_extent_buffer(struct extent_buffer *eb)
4501 btrfs_leak_debug_del(&eb->leak_list);
4502 kmem_cache_free(extent_buffer_cache, eb);
4505 int extent_buffer_under_io(struct extent_buffer *eb)
4507 return (atomic_read(&eb->io_pages) ||
4508 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4509 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4513 * Helper for releasing extent buffer page.
4515 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4517 unsigned long index;
4519 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4521 BUG_ON(extent_buffer_under_io(eb));
4523 index = num_extent_pages(eb->start, eb->len);
4529 page = eb->pages[index];
4530 if (page && mapped) {
4531 spin_lock(&page->mapping->private_lock);
4533 * We do this since we'll remove the pages after we've
4534 * removed the eb from the radix tree, so we could race
4535 * and have this page now attached to the new eb. So
4536 * only clear page_private if it's still connected to
4539 if (PagePrivate(page) &&
4540 page->private == (unsigned long)eb) {
4541 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4542 BUG_ON(PageDirty(page));
4543 BUG_ON(PageWriteback(page));
4545 * We need to make sure we haven't be attached
4548 ClearPagePrivate(page);
4549 set_page_private(page, 0);
4550 /* One for the page private */
4551 page_cache_release(page);
4553 spin_unlock(&page->mapping->private_lock);
4557 /* One for when we alloced the page */
4558 page_cache_release(page);
4560 } while (index != 0);
4564 * Helper for releasing the extent buffer.
4566 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4568 btrfs_release_extent_buffer_page(eb);
4569 __free_extent_buffer(eb);
4572 static struct extent_buffer *
4573 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4574 unsigned long len, gfp_t mask)
4576 struct extent_buffer *eb = NULL;
4578 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4583 eb->fs_info = fs_info;
4585 rwlock_init(&eb->lock);
4586 atomic_set(&eb->write_locks, 0);
4587 atomic_set(&eb->read_locks, 0);
4588 atomic_set(&eb->blocking_readers, 0);
4589 atomic_set(&eb->blocking_writers, 0);
4590 atomic_set(&eb->spinning_readers, 0);
4591 atomic_set(&eb->spinning_writers, 0);
4592 eb->lock_nested = 0;
4593 init_waitqueue_head(&eb->write_lock_wq);
4594 init_waitqueue_head(&eb->read_lock_wq);
4596 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4598 spin_lock_init(&eb->refs_lock);
4599 atomic_set(&eb->refs, 1);
4600 atomic_set(&eb->io_pages, 0);
4603 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4605 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4606 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4607 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4612 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4616 struct extent_buffer *new;
4617 unsigned long num_pages = num_extent_pages(src->start, src->len);
4619 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4623 for (i = 0; i < num_pages; i++) {
4624 p = alloc_page(GFP_NOFS);
4626 btrfs_release_extent_buffer(new);
4629 attach_extent_buffer_page(new, p);
4630 WARN_ON(PageDirty(p));
4635 copy_extent_buffer(new, src, 0, 0, src->len);
4636 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4637 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4642 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4644 struct extent_buffer *eb;
4645 unsigned long num_pages = num_extent_pages(0, len);
4648 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4652 for (i = 0; i < num_pages; i++) {
4653 eb->pages[i] = alloc_page(GFP_NOFS);
4657 set_extent_buffer_uptodate(eb);
4658 btrfs_set_header_nritems(eb, 0);
4659 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4664 __free_page(eb->pages[i - 1]);
4665 __free_extent_buffer(eb);
4669 static void check_buffer_tree_ref(struct extent_buffer *eb)
4672 /* the ref bit is tricky. We have to make sure it is set
4673 * if we have the buffer dirty. Otherwise the
4674 * code to free a buffer can end up dropping a dirty
4677 * Once the ref bit is set, it won't go away while the
4678 * buffer is dirty or in writeback, and it also won't
4679 * go away while we have the reference count on the
4682 * We can't just set the ref bit without bumping the
4683 * ref on the eb because free_extent_buffer might
4684 * see the ref bit and try to clear it. If this happens
4685 * free_extent_buffer might end up dropping our original
4686 * ref by mistake and freeing the page before we are able
4687 * to add one more ref.
4689 * So bump the ref count first, then set the bit. If someone
4690 * beat us to it, drop the ref we added.
4692 refs = atomic_read(&eb->refs);
4693 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4696 spin_lock(&eb->refs_lock);
4697 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4698 atomic_inc(&eb->refs);
4699 spin_unlock(&eb->refs_lock);
4702 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4703 struct page *accessed)
4705 unsigned long num_pages, i;
4707 check_buffer_tree_ref(eb);
4709 num_pages = num_extent_pages(eb->start, eb->len);
4710 for (i = 0; i < num_pages; i++) {
4711 struct page *p = eb->pages[i];
4714 mark_page_accessed(p);
4718 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4721 struct extent_buffer *eb;
4724 eb = radix_tree_lookup(&fs_info->buffer_radix,
4725 start >> PAGE_CACHE_SHIFT);
4726 if (eb && atomic_inc_not_zero(&eb->refs)) {
4728 mark_extent_buffer_accessed(eb, NULL);
4736 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4737 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4738 u64 start, unsigned long len)
4740 struct extent_buffer *eb, *exists = NULL;
4743 eb = find_extent_buffer(fs_info, start);
4746 eb = alloc_dummy_extent_buffer(start, len);
4749 eb->fs_info = fs_info;
4751 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4754 spin_lock(&fs_info->buffer_lock);
4755 ret = radix_tree_insert(&fs_info->buffer_radix,
4756 start >> PAGE_CACHE_SHIFT, eb);
4757 spin_unlock(&fs_info->buffer_lock);
4758 radix_tree_preload_end();
4759 if (ret == -EEXIST) {
4760 exists = find_extent_buffer(fs_info, start);
4766 check_buffer_tree_ref(eb);
4767 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4770 * We will free dummy extent buffer's if they come into
4771 * free_extent_buffer with a ref count of 2, but if we are using this we
4772 * want the buffers to stay in memory until we're done with them, so
4773 * bump the ref count again.
4775 atomic_inc(&eb->refs);
4778 btrfs_release_extent_buffer(eb);
4783 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4784 u64 start, unsigned long len)
4786 unsigned long num_pages = num_extent_pages(start, len);
4788 unsigned long index = start >> PAGE_CACHE_SHIFT;
4789 struct extent_buffer *eb;
4790 struct extent_buffer *exists = NULL;
4792 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4796 eb = find_extent_buffer(fs_info, start);
4800 eb = __alloc_extent_buffer(fs_info, start, len, GFP_NOFS);
4804 for (i = 0; i < num_pages; i++, index++) {
4805 p = find_or_create_page(mapping, index, GFP_NOFS);
4809 spin_lock(&mapping->private_lock);
4810 if (PagePrivate(p)) {
4812 * We could have already allocated an eb for this page
4813 * and attached one so lets see if we can get a ref on
4814 * the existing eb, and if we can we know it's good and
4815 * we can just return that one, else we know we can just
4816 * overwrite page->private.
4818 exists = (struct extent_buffer *)p->private;
4819 if (atomic_inc_not_zero(&exists->refs)) {
4820 spin_unlock(&mapping->private_lock);
4822 page_cache_release(p);
4823 mark_extent_buffer_accessed(exists, p);
4828 * Do this so attach doesn't complain and we need to
4829 * drop the ref the old guy had.
4831 ClearPagePrivate(p);
4832 WARN_ON(PageDirty(p));
4833 page_cache_release(p);
4835 attach_extent_buffer_page(eb, p);
4836 spin_unlock(&mapping->private_lock);
4837 WARN_ON(PageDirty(p));
4839 if (!PageUptodate(p))
4843 * see below about how we avoid a nasty race with release page
4844 * and why we unlock later
4848 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4850 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4854 spin_lock(&fs_info->buffer_lock);
4855 ret = radix_tree_insert(&fs_info->buffer_radix,
4856 start >> PAGE_CACHE_SHIFT, eb);
4857 spin_unlock(&fs_info->buffer_lock);
4858 radix_tree_preload_end();
4859 if (ret == -EEXIST) {
4860 exists = find_extent_buffer(fs_info, start);
4866 /* add one reference for the tree */
4867 check_buffer_tree_ref(eb);
4868 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4871 * there is a race where release page may have
4872 * tried to find this extent buffer in the radix
4873 * but failed. It will tell the VM it is safe to
4874 * reclaim the, and it will clear the page private bit.
4875 * We must make sure to set the page private bit properly
4876 * after the extent buffer is in the radix tree so
4877 * it doesn't get lost
4879 SetPageChecked(eb->pages[0]);
4880 for (i = 1; i < num_pages; i++) {
4882 ClearPageChecked(p);
4885 unlock_page(eb->pages[0]);
4889 for (i = 0; i < num_pages; i++) {
4891 unlock_page(eb->pages[i]);
4894 WARN_ON(!atomic_dec_and_test(&eb->refs));
4895 btrfs_release_extent_buffer(eb);
4899 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4901 struct extent_buffer *eb =
4902 container_of(head, struct extent_buffer, rcu_head);
4904 __free_extent_buffer(eb);
4907 /* Expects to have eb->eb_lock already held */
4908 static int release_extent_buffer(struct extent_buffer *eb)
4910 WARN_ON(atomic_read(&eb->refs) == 0);
4911 if (atomic_dec_and_test(&eb->refs)) {
4912 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4913 struct btrfs_fs_info *fs_info = eb->fs_info;
4915 spin_unlock(&eb->refs_lock);
4917 spin_lock(&fs_info->buffer_lock);
4918 radix_tree_delete(&fs_info->buffer_radix,
4919 eb->start >> PAGE_CACHE_SHIFT);
4920 spin_unlock(&fs_info->buffer_lock);
4922 spin_unlock(&eb->refs_lock);
4925 /* Should be safe to release our pages at this point */
4926 btrfs_release_extent_buffer_page(eb);
4927 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4930 spin_unlock(&eb->refs_lock);
4935 void free_extent_buffer(struct extent_buffer *eb)
4943 refs = atomic_read(&eb->refs);
4946 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4951 spin_lock(&eb->refs_lock);
4952 if (atomic_read(&eb->refs) == 2 &&
4953 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4954 atomic_dec(&eb->refs);
4956 if (atomic_read(&eb->refs) == 2 &&
4957 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4958 !extent_buffer_under_io(eb) &&
4959 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4960 atomic_dec(&eb->refs);
4963 * I know this is terrible, but it's temporary until we stop tracking
4964 * the uptodate bits and such for the extent buffers.
4966 release_extent_buffer(eb);
4969 void free_extent_buffer_stale(struct extent_buffer *eb)
4974 spin_lock(&eb->refs_lock);
4975 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4977 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4978 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4979 atomic_dec(&eb->refs);
4980 release_extent_buffer(eb);
4983 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4986 unsigned long num_pages;
4989 num_pages = num_extent_pages(eb->start, eb->len);
4991 for (i = 0; i < num_pages; i++) {
4992 page = eb->pages[i];
4993 if (!PageDirty(page))
4997 WARN_ON(!PagePrivate(page));
4999 clear_page_dirty_for_io(page);
5000 spin_lock_irq(&page->mapping->tree_lock);
5001 if (!PageDirty(page)) {
5002 radix_tree_tag_clear(&page->mapping->page_tree,
5004 PAGECACHE_TAG_DIRTY);
5006 spin_unlock_irq(&page->mapping->tree_lock);
5007 ClearPageError(page);
5010 WARN_ON(atomic_read(&eb->refs) == 0);
5013 int set_extent_buffer_dirty(struct extent_buffer *eb)
5016 unsigned long num_pages;
5019 check_buffer_tree_ref(eb);
5021 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5023 num_pages = num_extent_pages(eb->start, eb->len);
5024 WARN_ON(atomic_read(&eb->refs) == 0);
5025 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5027 for (i = 0; i < num_pages; i++)
5028 set_page_dirty(eb->pages[i]);
5032 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
5036 unsigned long num_pages;
5038 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5039 num_pages = num_extent_pages(eb->start, eb->len);
5040 for (i = 0; i < num_pages; i++) {
5041 page = eb->pages[i];
5043 ClearPageUptodate(page);
5048 int set_extent_buffer_uptodate(struct extent_buffer *eb)
5052 unsigned long num_pages;
5054 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5055 num_pages = num_extent_pages(eb->start, eb->len);
5056 for (i = 0; i < num_pages; i++) {
5057 page = eb->pages[i];
5058 SetPageUptodate(page);
5063 int extent_buffer_uptodate(struct extent_buffer *eb)
5065 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5068 int read_extent_buffer_pages(struct extent_io_tree *tree,
5069 struct extent_buffer *eb, u64 start, int wait,
5070 get_extent_t *get_extent, int mirror_num)
5073 unsigned long start_i;
5077 int locked_pages = 0;
5078 int all_uptodate = 1;
5079 unsigned long num_pages;
5080 unsigned long num_reads = 0;
5081 struct bio *bio = NULL;
5082 unsigned long bio_flags = 0;
5084 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5088 WARN_ON(start < eb->start);
5089 start_i = (start >> PAGE_CACHE_SHIFT) -
5090 (eb->start >> PAGE_CACHE_SHIFT);
5095 num_pages = num_extent_pages(eb->start, eb->len);
5096 for (i = start_i; i < num_pages; i++) {
5097 page = eb->pages[i];
5098 if (wait == WAIT_NONE) {
5099 if (!trylock_page(page))
5105 if (!PageUptodate(page)) {
5112 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5116 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5117 eb->read_mirror = 0;
5118 atomic_set(&eb->io_pages, num_reads);
5119 for (i = start_i; i < num_pages; i++) {
5120 page = eb->pages[i];
5121 if (!PageUptodate(page)) {
5122 ClearPageError(page);
5123 err = __extent_read_full_page(tree, page,
5125 mirror_num, &bio_flags,
5135 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
5141 if (ret || wait != WAIT_COMPLETE)
5144 for (i = start_i; i < num_pages; i++) {
5145 page = eb->pages[i];
5146 wait_on_page_locked(page);
5147 if (!PageUptodate(page))
5155 while (locked_pages > 0) {
5156 page = eb->pages[i];
5164 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
5165 unsigned long start,
5172 char *dst = (char *)dstv;
5173 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5174 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5176 WARN_ON(start > eb->len);
5177 WARN_ON(start + len > eb->start + eb->len);
5179 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5182 page = eb->pages[i];
5184 cur = min(len, (PAGE_CACHE_SIZE - offset));
5185 kaddr = page_address(page);
5186 memcpy(dst, kaddr + offset, cur);
5195 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv,
5196 unsigned long start,
5203 char __user *dst = (char __user *)dstv;
5204 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5205 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5208 WARN_ON(start > eb->len);
5209 WARN_ON(start + len > eb->start + eb->len);
5211 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5214 page = eb->pages[i];
5216 cur = min(len, (PAGE_CACHE_SIZE - offset));
5217 kaddr = page_address(page);
5218 if (copy_to_user(dst, kaddr + offset, cur)) {
5232 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
5233 unsigned long min_len, char **map,
5234 unsigned long *map_start,
5235 unsigned long *map_len)
5237 size_t offset = start & (PAGE_CACHE_SIZE - 1);
5240 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5241 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5242 unsigned long end_i = (start_offset + start + min_len - 1) >>
5249 offset = start_offset;
5253 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
5256 if (start + min_len > eb->len) {
5257 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
5259 eb->start, eb->len, start, min_len);
5264 kaddr = page_address(p);
5265 *map = kaddr + offset;
5266 *map_len = PAGE_CACHE_SIZE - offset;
5270 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5271 unsigned long start,
5278 char *ptr = (char *)ptrv;
5279 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5280 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5283 WARN_ON(start > eb->len);
5284 WARN_ON(start + len > eb->start + eb->len);
5286 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5289 page = eb->pages[i];
5291 cur = min(len, (PAGE_CACHE_SIZE - offset));
5293 kaddr = page_address(page);
5294 ret = memcmp(ptr, kaddr + offset, cur);
5306 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5307 unsigned long start, unsigned long len)
5313 char *src = (char *)srcv;
5314 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5315 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5317 WARN_ON(start > eb->len);
5318 WARN_ON(start + len > eb->start + eb->len);
5320 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5323 page = eb->pages[i];
5324 WARN_ON(!PageUptodate(page));
5326 cur = min(len, PAGE_CACHE_SIZE - offset);
5327 kaddr = page_address(page);
5328 memcpy(kaddr + offset, src, cur);
5337 void memset_extent_buffer(struct extent_buffer *eb, char c,
5338 unsigned long start, unsigned long len)
5344 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5345 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5347 WARN_ON(start > eb->len);
5348 WARN_ON(start + len > eb->start + eb->len);
5350 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5353 page = eb->pages[i];
5354 WARN_ON(!PageUptodate(page));
5356 cur = min(len, PAGE_CACHE_SIZE - offset);
5357 kaddr = page_address(page);
5358 memset(kaddr + offset, c, cur);
5366 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5367 unsigned long dst_offset, unsigned long src_offset,
5370 u64 dst_len = dst->len;
5375 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5376 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5378 WARN_ON(src->len != dst_len);
5380 offset = (start_offset + dst_offset) &
5381 (PAGE_CACHE_SIZE - 1);
5384 page = dst->pages[i];
5385 WARN_ON(!PageUptodate(page));
5387 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5389 kaddr = page_address(page);
5390 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5399 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5401 unsigned long distance = (src > dst) ? src - dst : dst - src;
5402 return distance < len;
5405 static void copy_pages(struct page *dst_page, struct page *src_page,
5406 unsigned long dst_off, unsigned long src_off,
5409 char *dst_kaddr = page_address(dst_page);
5411 int must_memmove = 0;
5413 if (dst_page != src_page) {
5414 src_kaddr = page_address(src_page);
5416 src_kaddr = dst_kaddr;
5417 if (areas_overlap(src_off, dst_off, len))
5422 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5424 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5427 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5428 unsigned long src_offset, unsigned long len)
5431 size_t dst_off_in_page;
5432 size_t src_off_in_page;
5433 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5434 unsigned long dst_i;
5435 unsigned long src_i;
5437 if (src_offset + len > dst->len) {
5438 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5439 "len %lu dst len %lu\n", src_offset, len, dst->len);
5442 if (dst_offset + len > dst->len) {
5443 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5444 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5449 dst_off_in_page = (start_offset + dst_offset) &
5450 (PAGE_CACHE_SIZE - 1);
5451 src_off_in_page = (start_offset + src_offset) &
5452 (PAGE_CACHE_SIZE - 1);
5454 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5455 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5457 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5459 cur = min_t(unsigned long, cur,
5460 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5462 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5463 dst_off_in_page, src_off_in_page, cur);
5471 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5472 unsigned long src_offset, unsigned long len)
5475 size_t dst_off_in_page;
5476 size_t src_off_in_page;
5477 unsigned long dst_end = dst_offset + len - 1;
5478 unsigned long src_end = src_offset + len - 1;
5479 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5480 unsigned long dst_i;
5481 unsigned long src_i;
5483 if (src_offset + len > dst->len) {
5484 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5485 "len %lu len %lu\n", src_offset, len, dst->len);
5488 if (dst_offset + len > dst->len) {
5489 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5490 "len %lu len %lu\n", dst_offset, len, dst->len);
5493 if (dst_offset < src_offset) {
5494 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5498 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5499 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5501 dst_off_in_page = (start_offset + dst_end) &
5502 (PAGE_CACHE_SIZE - 1);
5503 src_off_in_page = (start_offset + src_end) &
5504 (PAGE_CACHE_SIZE - 1);
5506 cur = min_t(unsigned long, len, src_off_in_page + 1);
5507 cur = min(cur, dst_off_in_page + 1);
5508 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5509 dst_off_in_page - cur + 1,
5510 src_off_in_page - cur + 1, cur);
5518 int try_release_extent_buffer(struct page *page)
5520 struct extent_buffer *eb;
5523 * We need to make sure noboody is attaching this page to an eb right
5526 spin_lock(&page->mapping->private_lock);
5527 if (!PagePrivate(page)) {
5528 spin_unlock(&page->mapping->private_lock);
5532 eb = (struct extent_buffer *)page->private;
5536 * This is a little awful but should be ok, we need to make sure that
5537 * the eb doesn't disappear out from under us while we're looking at
5540 spin_lock(&eb->refs_lock);
5541 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5542 spin_unlock(&eb->refs_lock);
5543 spin_unlock(&page->mapping->private_lock);
5546 spin_unlock(&page->mapping->private_lock);
5549 * If tree ref isn't set then we know the ref on this eb is a real ref,
5550 * so just return, this page will likely be freed soon anyway.
5552 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5553 spin_unlock(&eb->refs_lock);
5557 return release_extent_buffer(eb);