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)) {
599 * Don't care for allocation failure here because we might end
600 * up not needing the pre-allocated extent state at all, which
601 * is the case if we only have in the tree extent states that
602 * cover our input range and don't cover too any other range.
603 * If we end up needing a new extent state we allocate it later.
605 prealloc = alloc_extent_state(mask);
608 spin_lock(&tree->lock);
610 cached = *cached_state;
613 *cached_state = NULL;
617 if (cached && extent_state_in_tree(cached) &&
618 cached->start <= start && cached->end > start) {
620 atomic_dec(&cached->refs);
625 free_extent_state(cached);
628 * this search will find the extents that end after
631 node = tree_search(tree, start);
634 state = rb_entry(node, struct extent_state, rb_node);
636 if (state->start > end)
638 WARN_ON(state->end < start);
639 last_end = state->end;
641 /* the state doesn't have the wanted bits, go ahead */
642 if (!(state->state & bits)) {
643 state = next_state(state);
648 * | ---- desired range ---- |
650 * | ------------- state -------------- |
652 * We need to split the extent we found, and may flip
653 * bits on second half.
655 * If the extent we found extends past our range, we
656 * just split and search again. It'll get split again
657 * the next time though.
659 * If the extent we found is inside our range, we clear
660 * the desired bit on it.
663 if (state->start < start) {
664 prealloc = alloc_extent_state_atomic(prealloc);
666 err = split_state(tree, state, prealloc, start);
668 extent_io_tree_panic(tree, err);
673 if (state->end <= end) {
674 state = clear_state_bit(tree, state, &bits, wake);
680 * | ---- desired range ---- |
682 * We need to split the extent, and clear the bit
685 if (state->start <= end && state->end > end) {
686 prealloc = alloc_extent_state_atomic(prealloc);
688 err = split_state(tree, state, prealloc, end + 1);
690 extent_io_tree_panic(tree, err);
695 clear_state_bit(tree, prealloc, &bits, wake);
701 state = clear_state_bit(tree, state, &bits, wake);
703 if (last_end == (u64)-1)
705 start = last_end + 1;
706 if (start <= end && state && !need_resched())
711 spin_unlock(&tree->lock);
713 free_extent_state(prealloc);
720 spin_unlock(&tree->lock);
721 if (mask & __GFP_WAIT)
726 static void wait_on_state(struct extent_io_tree *tree,
727 struct extent_state *state)
728 __releases(tree->lock)
729 __acquires(tree->lock)
732 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
733 spin_unlock(&tree->lock);
735 spin_lock(&tree->lock);
736 finish_wait(&state->wq, &wait);
740 * waits for one or more bits to clear on a range in the state tree.
741 * The range [start, end] is inclusive.
742 * The tree lock is taken by this function
744 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
747 struct extent_state *state;
748 struct rb_node *node;
750 btrfs_debug_check_extent_io_range(tree, start, end);
752 spin_lock(&tree->lock);
756 * this search will find all the extents that end after
759 node = tree_search(tree, start);
764 state = rb_entry(node, struct extent_state, rb_node);
766 if (state->start > end)
769 if (state->state & bits) {
770 start = state->start;
771 atomic_inc(&state->refs);
772 wait_on_state(tree, state);
773 free_extent_state(state);
776 start = state->end + 1;
781 if (!cond_resched_lock(&tree->lock)) {
782 node = rb_next(node);
787 spin_unlock(&tree->lock);
790 static void set_state_bits(struct extent_io_tree *tree,
791 struct extent_state *state,
794 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
796 set_state_cb(tree, state, bits);
797 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
798 u64 range = state->end - state->start + 1;
799 tree->dirty_bytes += range;
801 state->state |= bits_to_set;
804 static void cache_state_if_flags(struct extent_state *state,
805 struct extent_state **cached_ptr,
808 if (cached_ptr && !(*cached_ptr)) {
809 if (!flags || (state->state & flags)) {
811 atomic_inc(&state->refs);
816 static void cache_state(struct extent_state *state,
817 struct extent_state **cached_ptr)
819 return cache_state_if_flags(state, cached_ptr,
820 EXTENT_IOBITS | EXTENT_BOUNDARY);
824 * set some bits on a range in the tree. This may require allocations or
825 * sleeping, so the gfp mask is used to indicate what is allowed.
827 * If any of the exclusive bits are set, this will fail with -EEXIST if some
828 * part of the range already has the desired bits set. The start of the
829 * existing range is returned in failed_start in this case.
831 * [start, end] is inclusive This takes the tree lock.
834 static int __must_check
835 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
836 unsigned long bits, unsigned long exclusive_bits,
837 u64 *failed_start, struct extent_state **cached_state,
840 struct extent_state *state;
841 struct extent_state *prealloc = NULL;
842 struct rb_node *node;
844 struct rb_node *parent;
849 btrfs_debug_check_extent_io_range(tree, start, end);
851 bits |= EXTENT_FIRST_DELALLOC;
853 if (!prealloc && (mask & __GFP_WAIT)) {
854 prealloc = alloc_extent_state(mask);
858 spin_lock(&tree->lock);
859 if (cached_state && *cached_state) {
860 state = *cached_state;
861 if (state->start <= start && state->end > start &&
862 extent_state_in_tree(state)) {
863 node = &state->rb_node;
868 * this search will find all the extents that end after
871 node = tree_search_for_insert(tree, start, &p, &parent);
873 prealloc = alloc_extent_state_atomic(prealloc);
875 err = insert_state(tree, prealloc, start, end,
878 extent_io_tree_panic(tree, err);
880 cache_state(prealloc, cached_state);
884 state = rb_entry(node, struct extent_state, rb_node);
886 last_start = state->start;
887 last_end = state->end;
890 * | ---- desired range ---- |
893 * Just lock what we found and keep going
895 if (state->start == start && state->end <= end) {
896 if (state->state & exclusive_bits) {
897 *failed_start = state->start;
902 set_state_bits(tree, state, &bits);
903 cache_state(state, cached_state);
904 merge_state(tree, state);
905 if (last_end == (u64)-1)
907 start = last_end + 1;
908 state = next_state(state);
909 if (start < end && state && state->start == start &&
916 * | ---- desired range ---- |
919 * | ------------- state -------------- |
921 * We need to split the extent we found, and may flip bits on
924 * If the extent we found extends past our
925 * range, we just split and search again. It'll get split
926 * again the next time though.
928 * If the extent we found is inside our range, we set the
931 if (state->start < start) {
932 if (state->state & exclusive_bits) {
933 *failed_start = start;
938 prealloc = alloc_extent_state_atomic(prealloc);
940 err = split_state(tree, state, prealloc, start);
942 extent_io_tree_panic(tree, err);
947 if (state->end <= end) {
948 set_state_bits(tree, state, &bits);
949 cache_state(state, cached_state);
950 merge_state(tree, state);
951 if (last_end == (u64)-1)
953 start = last_end + 1;
954 state = next_state(state);
955 if (start < end && state && state->start == start &&
962 * | ---- desired range ---- |
963 * | state | or | state |
965 * There's a hole, we need to insert something in it and
966 * ignore the extent we found.
968 if (state->start > start) {
970 if (end < last_start)
973 this_end = last_start - 1;
975 prealloc = alloc_extent_state_atomic(prealloc);
979 * Avoid to free 'prealloc' if it can be merged with
982 err = insert_state(tree, prealloc, start, this_end,
985 extent_io_tree_panic(tree, err);
987 cache_state(prealloc, cached_state);
989 start = this_end + 1;
993 * | ---- desired range ---- |
995 * We need to split the extent, and set the bit
998 if (state->start <= end && state->end > end) {
999 if (state->state & exclusive_bits) {
1000 *failed_start = start;
1005 prealloc = alloc_extent_state_atomic(prealloc);
1007 err = split_state(tree, state, prealloc, end + 1);
1009 extent_io_tree_panic(tree, err);
1011 set_state_bits(tree, prealloc, &bits);
1012 cache_state(prealloc, cached_state);
1013 merge_state(tree, prealloc);
1021 spin_unlock(&tree->lock);
1023 free_extent_state(prealloc);
1030 spin_unlock(&tree->lock);
1031 if (mask & __GFP_WAIT)
1036 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1037 unsigned long bits, u64 * failed_start,
1038 struct extent_state **cached_state, gfp_t mask)
1040 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1041 cached_state, mask);
1046 * convert_extent_bit - convert all bits in a given range from one bit to
1048 * @tree: the io tree to search
1049 * @start: the start offset in bytes
1050 * @end: the end offset in bytes (inclusive)
1051 * @bits: the bits to set in this range
1052 * @clear_bits: the bits to clear in this range
1053 * @cached_state: state that we're going to cache
1054 * @mask: the allocation mask
1056 * This will go through and set bits for the given range. If any states exist
1057 * already in this range they are set with the given bit and cleared of the
1058 * clear_bits. This is only meant to be used by things that are mergeable, ie
1059 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1060 * boundary bits like LOCK.
1062 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1063 unsigned long bits, unsigned long clear_bits,
1064 struct extent_state **cached_state, gfp_t mask)
1066 struct extent_state *state;
1067 struct extent_state *prealloc = NULL;
1068 struct rb_node *node;
1070 struct rb_node *parent;
1074 bool first_iteration = true;
1076 btrfs_debug_check_extent_io_range(tree, start, end);
1079 if (!prealloc && (mask & __GFP_WAIT)) {
1081 * Best effort, don't worry if extent state allocation fails
1082 * here for the first iteration. We might have a cached state
1083 * that matches exactly the target range, in which case no
1084 * extent state allocations are needed. We'll only know this
1085 * after locking the tree.
1087 prealloc = alloc_extent_state(mask);
1088 if (!prealloc && !first_iteration)
1092 spin_lock(&tree->lock);
1093 if (cached_state && *cached_state) {
1094 state = *cached_state;
1095 if (state->start <= start && state->end > start &&
1096 extent_state_in_tree(state)) {
1097 node = &state->rb_node;
1103 * this search will find all the extents that end after
1106 node = tree_search_for_insert(tree, start, &p, &parent);
1108 prealloc = alloc_extent_state_atomic(prealloc);
1113 err = insert_state(tree, prealloc, start, end,
1114 &p, &parent, &bits);
1116 extent_io_tree_panic(tree, err);
1117 cache_state(prealloc, cached_state);
1121 state = rb_entry(node, struct extent_state, rb_node);
1123 last_start = state->start;
1124 last_end = state->end;
1127 * | ---- desired range ---- |
1130 * Just lock what we found and keep going
1132 if (state->start == start && state->end <= end) {
1133 set_state_bits(tree, state, &bits);
1134 cache_state(state, cached_state);
1135 state = clear_state_bit(tree, state, &clear_bits, 0);
1136 if (last_end == (u64)-1)
1138 start = last_end + 1;
1139 if (start < end && state && state->start == start &&
1146 * | ---- desired range ---- |
1149 * | ------------- state -------------- |
1151 * We need to split the extent we found, and may flip bits on
1154 * If the extent we found extends past our
1155 * range, we just split and search again. It'll get split
1156 * again the next time though.
1158 * If the extent we found is inside our range, we set the
1159 * desired bit on it.
1161 if (state->start < start) {
1162 prealloc = alloc_extent_state_atomic(prealloc);
1167 err = split_state(tree, state, prealloc, start);
1169 extent_io_tree_panic(tree, err);
1173 if (state->end <= end) {
1174 set_state_bits(tree, state, &bits);
1175 cache_state(state, cached_state);
1176 state = clear_state_bit(tree, state, &clear_bits, 0);
1177 if (last_end == (u64)-1)
1179 start = last_end + 1;
1180 if (start < end && state && state->start == start &&
1187 * | ---- desired range ---- |
1188 * | state | or | state |
1190 * There's a hole, we need to insert something in it and
1191 * ignore the extent we found.
1193 if (state->start > start) {
1195 if (end < last_start)
1198 this_end = last_start - 1;
1200 prealloc = alloc_extent_state_atomic(prealloc);
1207 * Avoid to free 'prealloc' if it can be merged with
1210 err = insert_state(tree, prealloc, start, this_end,
1213 extent_io_tree_panic(tree, err);
1214 cache_state(prealloc, cached_state);
1216 start = this_end + 1;
1220 * | ---- desired range ---- |
1222 * We need to split the extent, and set the bit
1225 if (state->start <= end && state->end > end) {
1226 prealloc = alloc_extent_state_atomic(prealloc);
1232 err = split_state(tree, state, prealloc, end + 1);
1234 extent_io_tree_panic(tree, err);
1236 set_state_bits(tree, prealloc, &bits);
1237 cache_state(prealloc, cached_state);
1238 clear_state_bit(tree, prealloc, &clear_bits, 0);
1246 spin_unlock(&tree->lock);
1248 free_extent_state(prealloc);
1255 spin_unlock(&tree->lock);
1256 if (mask & __GFP_WAIT)
1258 first_iteration = false;
1262 /* wrappers around set/clear extent bit */
1263 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1266 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1270 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1271 unsigned long bits, gfp_t mask)
1273 return set_extent_bit(tree, start, end, bits, NULL,
1277 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1278 unsigned long bits, gfp_t mask)
1280 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1283 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1284 struct extent_state **cached_state, gfp_t mask)
1286 return set_extent_bit(tree, start, end,
1287 EXTENT_DELALLOC | EXTENT_UPTODATE,
1288 NULL, cached_state, mask);
1291 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1292 struct extent_state **cached_state, gfp_t mask)
1294 return set_extent_bit(tree, start, end,
1295 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1296 NULL, cached_state, mask);
1299 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1302 return clear_extent_bit(tree, start, end,
1303 EXTENT_DIRTY | EXTENT_DELALLOC |
1304 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1307 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1310 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1314 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1315 struct extent_state **cached_state, gfp_t mask)
1317 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1318 cached_state, mask);
1321 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1322 struct extent_state **cached_state, gfp_t mask)
1324 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1325 cached_state, mask);
1329 * either insert or lock state struct between start and end use mask to tell
1330 * us if waiting is desired.
1332 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1333 unsigned long bits, struct extent_state **cached_state)
1338 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1339 EXTENT_LOCKED, &failed_start,
1340 cached_state, GFP_NOFS);
1341 if (err == -EEXIST) {
1342 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1343 start = failed_start;
1346 WARN_ON(start > end);
1351 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1353 return lock_extent_bits(tree, start, end, 0, NULL);
1356 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1361 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1362 &failed_start, NULL, GFP_NOFS);
1363 if (err == -EEXIST) {
1364 if (failed_start > start)
1365 clear_extent_bit(tree, start, failed_start - 1,
1366 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1372 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1373 struct extent_state **cached, gfp_t mask)
1375 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1379 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1381 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1385 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1387 unsigned long index = start >> PAGE_CACHE_SHIFT;
1388 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1391 while (index <= end_index) {
1392 page = find_get_page(inode->i_mapping, index);
1393 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1394 clear_page_dirty_for_io(page);
1395 page_cache_release(page);
1401 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1403 unsigned long index = start >> PAGE_CACHE_SHIFT;
1404 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1407 while (index <= end_index) {
1408 page = find_get_page(inode->i_mapping, index);
1409 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1410 account_page_redirty(page);
1411 __set_page_dirty_nobuffers(page);
1412 page_cache_release(page);
1419 * helper function to set both pages and extents in the tree writeback
1421 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1423 unsigned long index = start >> PAGE_CACHE_SHIFT;
1424 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1427 while (index <= end_index) {
1428 page = find_get_page(tree->mapping, index);
1429 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1430 set_page_writeback(page);
1431 page_cache_release(page);
1437 /* find the first state struct with 'bits' set after 'start', and
1438 * return it. tree->lock must be held. NULL will returned if
1439 * nothing was found after 'start'
1441 static struct extent_state *
1442 find_first_extent_bit_state(struct extent_io_tree *tree,
1443 u64 start, unsigned long bits)
1445 struct rb_node *node;
1446 struct extent_state *state;
1449 * this search will find all the extents that end after
1452 node = tree_search(tree, start);
1457 state = rb_entry(node, struct extent_state, rb_node);
1458 if (state->end >= start && (state->state & bits))
1461 node = rb_next(node);
1470 * find the first offset in the io tree with 'bits' set. zero is
1471 * returned if we find something, and *start_ret and *end_ret are
1472 * set to reflect the state struct that was found.
1474 * If nothing was found, 1 is returned. If found something, return 0.
1476 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1477 u64 *start_ret, u64 *end_ret, unsigned long bits,
1478 struct extent_state **cached_state)
1480 struct extent_state *state;
1484 spin_lock(&tree->lock);
1485 if (cached_state && *cached_state) {
1486 state = *cached_state;
1487 if (state->end == start - 1 && extent_state_in_tree(state)) {
1488 n = rb_next(&state->rb_node);
1490 state = rb_entry(n, struct extent_state,
1492 if (state->state & bits)
1496 free_extent_state(*cached_state);
1497 *cached_state = NULL;
1500 free_extent_state(*cached_state);
1501 *cached_state = NULL;
1504 state = find_first_extent_bit_state(tree, start, bits);
1507 cache_state_if_flags(state, cached_state, 0);
1508 *start_ret = state->start;
1509 *end_ret = state->end;
1513 spin_unlock(&tree->lock);
1518 * find a contiguous range of bytes in the file marked as delalloc, not
1519 * more than 'max_bytes'. start and end are used to return the range,
1521 * 1 is returned if we find something, 0 if nothing was in the tree
1523 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1524 u64 *start, u64 *end, u64 max_bytes,
1525 struct extent_state **cached_state)
1527 struct rb_node *node;
1528 struct extent_state *state;
1529 u64 cur_start = *start;
1531 u64 total_bytes = 0;
1533 spin_lock(&tree->lock);
1536 * this search will find all the extents that end after
1539 node = tree_search(tree, cur_start);
1547 state = rb_entry(node, struct extent_state, rb_node);
1548 if (found && (state->start != cur_start ||
1549 (state->state & EXTENT_BOUNDARY))) {
1552 if (!(state->state & EXTENT_DELALLOC)) {
1558 *start = state->start;
1559 *cached_state = state;
1560 atomic_inc(&state->refs);
1564 cur_start = state->end + 1;
1565 node = rb_next(node);
1566 total_bytes += state->end - state->start + 1;
1567 if (total_bytes >= max_bytes)
1573 spin_unlock(&tree->lock);
1577 static noinline void __unlock_for_delalloc(struct inode *inode,
1578 struct page *locked_page,
1582 struct page *pages[16];
1583 unsigned long index = start >> PAGE_CACHE_SHIFT;
1584 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1585 unsigned long nr_pages = end_index - index + 1;
1588 if (index == locked_page->index && end_index == index)
1591 while (nr_pages > 0) {
1592 ret = find_get_pages_contig(inode->i_mapping, index,
1593 min_t(unsigned long, nr_pages,
1594 ARRAY_SIZE(pages)), pages);
1595 for (i = 0; i < ret; i++) {
1596 if (pages[i] != locked_page)
1597 unlock_page(pages[i]);
1598 page_cache_release(pages[i]);
1606 static noinline int lock_delalloc_pages(struct inode *inode,
1607 struct page *locked_page,
1611 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1612 unsigned long start_index = index;
1613 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1614 unsigned long pages_locked = 0;
1615 struct page *pages[16];
1616 unsigned long nrpages;
1620 /* the caller is responsible for locking the start index */
1621 if (index == locked_page->index && index == end_index)
1624 /* skip the page at the start index */
1625 nrpages = end_index - index + 1;
1626 while (nrpages > 0) {
1627 ret = find_get_pages_contig(inode->i_mapping, index,
1628 min_t(unsigned long,
1629 nrpages, ARRAY_SIZE(pages)), pages);
1634 /* now we have an array of pages, lock them all */
1635 for (i = 0; i < ret; i++) {
1637 * the caller is taking responsibility for
1640 if (pages[i] != locked_page) {
1641 lock_page(pages[i]);
1642 if (!PageDirty(pages[i]) ||
1643 pages[i]->mapping != inode->i_mapping) {
1645 unlock_page(pages[i]);
1646 page_cache_release(pages[i]);
1650 page_cache_release(pages[i]);
1659 if (ret && pages_locked) {
1660 __unlock_for_delalloc(inode, locked_page,
1662 ((u64)(start_index + pages_locked - 1)) <<
1669 * find a contiguous range of bytes in the file marked as delalloc, not
1670 * more than 'max_bytes'. start and end are used to return the range,
1672 * 1 is returned if we find something, 0 if nothing was in the tree
1674 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1675 struct extent_io_tree *tree,
1676 struct page *locked_page, u64 *start,
1677 u64 *end, u64 max_bytes)
1682 struct extent_state *cached_state = NULL;
1687 /* step one, find a bunch of delalloc bytes starting at start */
1688 delalloc_start = *start;
1690 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1691 max_bytes, &cached_state);
1692 if (!found || delalloc_end <= *start) {
1693 *start = delalloc_start;
1694 *end = delalloc_end;
1695 free_extent_state(cached_state);
1700 * start comes from the offset of locked_page. We have to lock
1701 * pages in order, so we can't process delalloc bytes before
1704 if (delalloc_start < *start)
1705 delalloc_start = *start;
1708 * make sure to limit the number of pages we try to lock down
1710 if (delalloc_end + 1 - delalloc_start > max_bytes)
1711 delalloc_end = delalloc_start + max_bytes - 1;
1713 /* step two, lock all the pages after the page that has start */
1714 ret = lock_delalloc_pages(inode, locked_page,
1715 delalloc_start, delalloc_end);
1716 if (ret == -EAGAIN) {
1717 /* some of the pages are gone, lets avoid looping by
1718 * shortening the size of the delalloc range we're searching
1720 free_extent_state(cached_state);
1721 cached_state = NULL;
1723 max_bytes = PAGE_CACHE_SIZE;
1731 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1733 /* step three, lock the state bits for the whole range */
1734 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1736 /* then test to make sure it is all still delalloc */
1737 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1738 EXTENT_DELALLOC, 1, cached_state);
1740 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1741 &cached_state, GFP_NOFS);
1742 __unlock_for_delalloc(inode, locked_page,
1743 delalloc_start, delalloc_end);
1747 free_extent_state(cached_state);
1748 *start = delalloc_start;
1749 *end = delalloc_end;
1754 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1755 struct page *locked_page,
1756 unsigned long clear_bits,
1757 unsigned long page_ops)
1759 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1761 struct page *pages[16];
1762 unsigned long index = start >> PAGE_CACHE_SHIFT;
1763 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1764 unsigned long nr_pages = end_index - index + 1;
1767 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1771 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1772 mapping_set_error(inode->i_mapping, -EIO);
1774 while (nr_pages > 0) {
1775 ret = find_get_pages_contig(inode->i_mapping, index,
1776 min_t(unsigned long,
1777 nr_pages, ARRAY_SIZE(pages)), pages);
1778 for (i = 0; i < ret; i++) {
1780 if (page_ops & PAGE_SET_PRIVATE2)
1781 SetPagePrivate2(pages[i]);
1783 if (pages[i] == locked_page) {
1784 page_cache_release(pages[i]);
1787 if (page_ops & PAGE_CLEAR_DIRTY)
1788 clear_page_dirty_for_io(pages[i]);
1789 if (page_ops & PAGE_SET_WRITEBACK)
1790 set_page_writeback(pages[i]);
1791 if (page_ops & PAGE_SET_ERROR)
1792 SetPageError(pages[i]);
1793 if (page_ops & PAGE_END_WRITEBACK)
1794 end_page_writeback(pages[i]);
1795 if (page_ops & PAGE_UNLOCK)
1796 unlock_page(pages[i]);
1797 page_cache_release(pages[i]);
1807 * count the number of bytes in the tree that have a given bit(s)
1808 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1809 * cached. The total number found is returned.
1811 u64 count_range_bits(struct extent_io_tree *tree,
1812 u64 *start, u64 search_end, u64 max_bytes,
1813 unsigned long bits, int contig)
1815 struct rb_node *node;
1816 struct extent_state *state;
1817 u64 cur_start = *start;
1818 u64 total_bytes = 0;
1822 if (WARN_ON(search_end <= cur_start))
1825 spin_lock(&tree->lock);
1826 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1827 total_bytes = tree->dirty_bytes;
1831 * this search will find all the extents that end after
1834 node = tree_search(tree, cur_start);
1839 state = rb_entry(node, struct extent_state, rb_node);
1840 if (state->start > search_end)
1842 if (contig && found && state->start > last + 1)
1844 if (state->end >= cur_start && (state->state & bits) == bits) {
1845 total_bytes += min(search_end, state->end) + 1 -
1846 max(cur_start, state->start);
1847 if (total_bytes >= max_bytes)
1850 *start = max(cur_start, state->start);
1854 } else if (contig && found) {
1857 node = rb_next(node);
1862 spin_unlock(&tree->lock);
1867 * set the private field for a given byte offset in the tree. If there isn't
1868 * an extent_state there already, this does nothing.
1870 static int set_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 state->private = private;
1893 spin_unlock(&tree->lock);
1897 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1899 struct rb_node *node;
1900 struct extent_state *state;
1903 spin_lock(&tree->lock);
1905 * this search will find all the extents that end after
1908 node = tree_search(tree, start);
1913 state = rb_entry(node, struct extent_state, rb_node);
1914 if (state->start != start) {
1918 *private = state->private;
1920 spin_unlock(&tree->lock);
1925 * searches a range in the state tree for a given mask.
1926 * If 'filled' == 1, this returns 1 only if every extent in the tree
1927 * has the bits set. Otherwise, 1 is returned if any bit in the
1928 * range is found set.
1930 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1931 unsigned long bits, int filled, struct extent_state *cached)
1933 struct extent_state *state = NULL;
1934 struct rb_node *node;
1937 spin_lock(&tree->lock);
1938 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1939 cached->end > start)
1940 node = &cached->rb_node;
1942 node = tree_search(tree, start);
1943 while (node && start <= end) {
1944 state = rb_entry(node, struct extent_state, rb_node);
1946 if (filled && state->start > start) {
1951 if (state->start > end)
1954 if (state->state & bits) {
1958 } else if (filled) {
1963 if (state->end == (u64)-1)
1966 start = state->end + 1;
1969 node = rb_next(node);
1976 spin_unlock(&tree->lock);
1981 * helper function to set a given page up to date if all the
1982 * extents in the tree for that page are up to date
1984 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1986 u64 start = page_offset(page);
1987 u64 end = start + PAGE_CACHE_SIZE - 1;
1988 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1989 SetPageUptodate(page);
1992 int free_io_failure(struct inode *inode, struct io_failure_record *rec)
1996 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1998 set_state_private(failure_tree, rec->start, 0);
1999 ret = clear_extent_bits(failure_tree, rec->start,
2000 rec->start + rec->len - 1,
2001 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2005 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
2006 rec->start + rec->len - 1,
2007 EXTENT_DAMAGED, GFP_NOFS);
2016 * this bypasses the standard btrfs submit functions deliberately, as
2017 * the standard behavior is to write all copies in a raid setup. here we only
2018 * want to write the one bad copy. so we do the mapping for ourselves and issue
2019 * submit_bio directly.
2020 * to avoid any synchronization issues, wait for the data after writing, which
2021 * actually prevents the read that triggered the error from finishing.
2022 * currently, there can be no more than two copies of every data bit. thus,
2023 * exactly one rewrite is required.
2025 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
2026 struct page *page, unsigned int pg_offset, int mirror_num)
2028 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2030 struct btrfs_device *dev;
2033 struct btrfs_bio *bbio = NULL;
2034 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2037 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2038 BUG_ON(!mirror_num);
2040 /* we can't repair anything in raid56 yet */
2041 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2044 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2047 bio->bi_iter.bi_size = 0;
2048 map_length = length;
2050 ret = btrfs_map_block(fs_info, WRITE, logical,
2051 &map_length, &bbio, mirror_num);
2056 BUG_ON(mirror_num != bbio->mirror_num);
2057 sector = bbio->stripes[mirror_num-1].physical >> 9;
2058 bio->bi_iter.bi_sector = sector;
2059 dev = bbio->stripes[mirror_num-1].dev;
2061 if (!dev || !dev->bdev || !dev->writeable) {
2065 bio->bi_bdev = dev->bdev;
2066 bio_add_page(bio, page, length, pg_offset);
2068 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
2069 /* try to remap that extent elsewhere? */
2071 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2075 printk_ratelimited_in_rcu(KERN_INFO
2076 "BTRFS: read error corrected: ino %llu off %llu (dev %s sector %llu)\n",
2077 btrfs_ino(inode), start,
2078 rcu_str_deref(dev->name), sector);
2083 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2086 u64 start = eb->start;
2087 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2090 if (root->fs_info->sb->s_flags & MS_RDONLY)
2093 for (i = 0; i < num_pages; i++) {
2094 struct page *p = eb->pages[i];
2096 ret = repair_io_failure(root->fs_info->btree_inode, start,
2097 PAGE_CACHE_SIZE, start, p,
2098 start - page_offset(p), mirror_num);
2101 start += PAGE_CACHE_SIZE;
2108 * each time an IO finishes, we do a fast check in the IO failure tree
2109 * to see if we need to process or clean up an io_failure_record
2111 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
2112 unsigned int pg_offset)
2115 u64 private_failure;
2116 struct io_failure_record *failrec;
2117 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2118 struct extent_state *state;
2123 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2124 (u64)-1, 1, EXTENT_DIRTY, 0);
2128 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2133 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2134 BUG_ON(!failrec->this_mirror);
2136 if (failrec->in_validation) {
2137 /* there was no real error, just free the record */
2138 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2142 if (fs_info->sb->s_flags & MS_RDONLY)
2145 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2146 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2149 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2151 if (state && state->start <= failrec->start &&
2152 state->end >= failrec->start + failrec->len - 1) {
2153 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2155 if (num_copies > 1) {
2156 repair_io_failure(inode, start, failrec->len,
2157 failrec->logical, page,
2158 pg_offset, failrec->failed_mirror);
2163 free_io_failure(inode, failrec);
2169 * Can be called when
2170 * - hold extent lock
2171 * - under ordered extent
2172 * - the inode is freeing
2174 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end)
2176 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2177 struct io_failure_record *failrec;
2178 struct extent_state *state, *next;
2180 if (RB_EMPTY_ROOT(&failure_tree->state))
2183 spin_lock(&failure_tree->lock);
2184 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2186 if (state->start > end)
2189 ASSERT(state->end <= end);
2191 next = next_state(state);
2193 failrec = (struct io_failure_record *)state->private;
2194 free_extent_state(state);
2199 spin_unlock(&failure_tree->lock);
2202 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2203 struct io_failure_record **failrec_ret)
2205 struct io_failure_record *failrec;
2207 struct extent_map *em;
2208 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2209 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2210 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2214 ret = get_state_private(failure_tree, start, &private);
2216 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2220 failrec->start = start;
2221 failrec->len = end - start + 1;
2222 failrec->this_mirror = 0;
2223 failrec->bio_flags = 0;
2224 failrec->in_validation = 0;
2226 read_lock(&em_tree->lock);
2227 em = lookup_extent_mapping(em_tree, start, failrec->len);
2229 read_unlock(&em_tree->lock);
2234 if (em->start > start || em->start + em->len <= start) {
2235 free_extent_map(em);
2238 read_unlock(&em_tree->lock);
2244 logical = start - em->start;
2245 logical = em->block_start + logical;
2246 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2247 logical = em->block_start;
2248 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2249 extent_set_compress_type(&failrec->bio_flags,
2253 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2254 logical, start, failrec->len);
2256 failrec->logical = logical;
2257 free_extent_map(em);
2259 /* set the bits in the private failure tree */
2260 ret = set_extent_bits(failure_tree, start, end,
2261 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2263 ret = set_state_private(failure_tree, start,
2264 (u64)(unsigned long)failrec);
2265 /* set the bits in the inode's tree */
2267 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2274 failrec = (struct io_failure_record *)(unsigned long)private;
2275 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2276 failrec->logical, failrec->start, failrec->len,
2277 failrec->in_validation);
2279 * when data can be on disk more than twice, add to failrec here
2280 * (e.g. with a list for failed_mirror) to make
2281 * clean_io_failure() clean all those errors at once.
2285 *failrec_ret = failrec;
2290 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2291 struct io_failure_record *failrec, int failed_mirror)
2295 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2296 failrec->logical, failrec->len);
2297 if (num_copies == 1) {
2299 * we only have a single copy of the data, so don't bother with
2300 * all the retry and error correction code that follows. no
2301 * matter what the error is, it is very likely to persist.
2303 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2304 num_copies, failrec->this_mirror, failed_mirror);
2309 * there are two premises:
2310 * a) deliver good data to the caller
2311 * b) correct the bad sectors on disk
2313 if (failed_bio->bi_vcnt > 1) {
2315 * to fulfill b), we need to know the exact failing sectors, as
2316 * we don't want to rewrite any more than the failed ones. thus,
2317 * we need separate read requests for the failed bio
2319 * if the following BUG_ON triggers, our validation request got
2320 * merged. we need separate requests for our algorithm to work.
2322 BUG_ON(failrec->in_validation);
2323 failrec->in_validation = 1;
2324 failrec->this_mirror = failed_mirror;
2327 * we're ready to fulfill a) and b) alongside. get a good copy
2328 * of the failed sector and if we succeed, we have setup
2329 * everything for repair_io_failure to do the rest for us.
2331 if (failrec->in_validation) {
2332 BUG_ON(failrec->this_mirror != failed_mirror);
2333 failrec->in_validation = 0;
2334 failrec->this_mirror = 0;
2336 failrec->failed_mirror = failed_mirror;
2337 failrec->this_mirror++;
2338 if (failrec->this_mirror == failed_mirror)
2339 failrec->this_mirror++;
2342 if (failrec->this_mirror > num_copies) {
2343 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2344 num_copies, failrec->this_mirror, failed_mirror);
2352 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2353 struct io_failure_record *failrec,
2354 struct page *page, int pg_offset, int icsum,
2355 bio_end_io_t *endio_func, void *data)
2358 struct btrfs_io_bio *btrfs_failed_bio;
2359 struct btrfs_io_bio *btrfs_bio;
2361 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2365 bio->bi_end_io = endio_func;
2366 bio->bi_iter.bi_sector = failrec->logical >> 9;
2367 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2368 bio->bi_iter.bi_size = 0;
2369 bio->bi_private = data;
2371 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2372 if (btrfs_failed_bio->csum) {
2373 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2374 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2376 btrfs_bio = btrfs_io_bio(bio);
2377 btrfs_bio->csum = btrfs_bio->csum_inline;
2379 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2383 bio_add_page(bio, page, failrec->len, pg_offset);
2389 * this is a generic handler for readpage errors (default
2390 * readpage_io_failed_hook). if other copies exist, read those and write back
2391 * good data to the failed position. does not investigate in remapping the
2392 * failed extent elsewhere, hoping the device will be smart enough to do this as
2396 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2397 struct page *page, u64 start, u64 end,
2400 struct io_failure_record *failrec;
2401 struct inode *inode = page->mapping->host;
2402 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2407 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2409 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2413 ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror);
2415 free_io_failure(inode, failrec);
2419 if (failed_bio->bi_vcnt > 1)
2420 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2422 read_mode = READ_SYNC;
2424 phy_offset >>= inode->i_sb->s_blocksize_bits;
2425 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2426 start - page_offset(page),
2427 (int)phy_offset, failed_bio->bi_end_io,
2430 free_io_failure(inode, failrec);
2434 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2435 read_mode, failrec->this_mirror, failrec->in_validation);
2437 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2438 failrec->this_mirror,
2439 failrec->bio_flags, 0);
2441 free_io_failure(inode, failrec);
2448 /* lots and lots of room for performance fixes in the end_bio funcs */
2450 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2452 int uptodate = (err == 0);
2453 struct extent_io_tree *tree;
2456 tree = &BTRFS_I(page->mapping->host)->io_tree;
2458 if (tree->ops && tree->ops->writepage_end_io_hook) {
2459 ret = tree->ops->writepage_end_io_hook(page, start,
2460 end, NULL, uptodate);
2466 ClearPageUptodate(page);
2468 ret = ret < 0 ? ret : -EIO;
2469 mapping_set_error(page->mapping, ret);
2475 * after a writepage IO is done, we need to:
2476 * clear the uptodate bits on error
2477 * clear the writeback bits in the extent tree for this IO
2478 * end_page_writeback if the page has no more pending IO
2480 * Scheduling is not allowed, so the extent state tree is expected
2481 * to have one and only one object corresponding to this IO.
2483 static void end_bio_extent_writepage(struct bio *bio, int err)
2485 struct bio_vec *bvec;
2490 bio_for_each_segment_all(bvec, bio, i) {
2491 struct page *page = bvec->bv_page;
2493 /* We always issue full-page reads, but if some block
2494 * in a page fails to read, blk_update_request() will
2495 * advance bv_offset and adjust bv_len to compensate.
2496 * Print a warning for nonzero offsets, and an error
2497 * if they don't add up to a full page. */
2498 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2499 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2500 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2501 "partial page write in btrfs with offset %u and length %u",
2502 bvec->bv_offset, bvec->bv_len);
2504 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2505 "incomplete page write in btrfs with offset %u and "
2507 bvec->bv_offset, bvec->bv_len);
2510 start = page_offset(page);
2511 end = start + bvec->bv_offset + bvec->bv_len - 1;
2513 if (end_extent_writepage(page, err, start, end))
2516 end_page_writeback(page);
2523 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2526 struct extent_state *cached = NULL;
2527 u64 end = start + len - 1;
2529 if (uptodate && tree->track_uptodate)
2530 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2531 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2535 * after a readpage IO is done, we need to:
2536 * clear the uptodate bits on error
2537 * set the uptodate bits if things worked
2538 * set the page up to date if all extents in the tree are uptodate
2539 * clear the lock bit in the extent tree
2540 * unlock the page if there are no other extents locked for it
2542 * Scheduling is not allowed, so the extent state tree is expected
2543 * to have one and only one object corresponding to this IO.
2545 static void end_bio_extent_readpage(struct bio *bio, int err)
2547 struct bio_vec *bvec;
2548 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2549 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2550 struct extent_io_tree *tree;
2555 u64 extent_start = 0;
2564 bio_for_each_segment_all(bvec, bio, i) {
2565 struct page *page = bvec->bv_page;
2566 struct inode *inode = page->mapping->host;
2568 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2569 "mirror=%u\n", (u64)bio->bi_iter.bi_sector, err,
2570 io_bio->mirror_num);
2571 tree = &BTRFS_I(inode)->io_tree;
2573 /* We always issue full-page reads, but if some block
2574 * in a page fails to read, blk_update_request() will
2575 * advance bv_offset and adjust bv_len to compensate.
2576 * Print a warning for nonzero offsets, and an error
2577 * if they don't add up to a full page. */
2578 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2579 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2580 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2581 "partial page read in btrfs with offset %u and length %u",
2582 bvec->bv_offset, bvec->bv_len);
2584 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2585 "incomplete page read in btrfs with offset %u and "
2587 bvec->bv_offset, bvec->bv_len);
2590 start = page_offset(page);
2591 end = start + bvec->bv_offset + bvec->bv_len - 1;
2594 mirror = io_bio->mirror_num;
2595 if (likely(uptodate && tree->ops &&
2596 tree->ops->readpage_end_io_hook)) {
2597 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2603 clean_io_failure(inode, start, page, 0);
2606 if (likely(uptodate))
2609 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2610 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2612 test_bit(BIO_UPTODATE, &bio->bi_flags))
2616 * The generic bio_readpage_error handles errors the
2617 * following way: If possible, new read requests are
2618 * created and submitted and will end up in
2619 * end_bio_extent_readpage as well (if we're lucky, not
2620 * in the !uptodate case). In that case it returns 0 and
2621 * we just go on with the next page in our bio. If it
2622 * can't handle the error it will return -EIO and we
2623 * remain responsible for that page.
2625 ret = bio_readpage_error(bio, offset, page, start, end,
2629 test_bit(BIO_UPTODATE, &bio->bi_flags);
2637 if (likely(uptodate)) {
2638 loff_t i_size = i_size_read(inode);
2639 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2642 /* Zero out the end if this page straddles i_size */
2643 off = i_size & (PAGE_CACHE_SIZE-1);
2644 if (page->index == end_index && off)
2645 zero_user_segment(page, off, PAGE_CACHE_SIZE);
2646 SetPageUptodate(page);
2648 ClearPageUptodate(page);
2654 if (unlikely(!uptodate)) {
2656 endio_readpage_release_extent(tree,
2662 endio_readpage_release_extent(tree, start,
2663 end - start + 1, 0);
2664 } else if (!extent_len) {
2665 extent_start = start;
2666 extent_len = end + 1 - start;
2667 } else if (extent_start + extent_len == start) {
2668 extent_len += end + 1 - start;
2670 endio_readpage_release_extent(tree, extent_start,
2671 extent_len, uptodate);
2672 extent_start = start;
2673 extent_len = end + 1 - start;
2678 endio_readpage_release_extent(tree, extent_start, extent_len,
2681 io_bio->end_io(io_bio, err);
2686 * this allocates from the btrfs_bioset. We're returning a bio right now
2687 * but you can call btrfs_io_bio for the appropriate container_of magic
2690 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2693 struct btrfs_io_bio *btrfs_bio;
2696 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2698 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2699 while (!bio && (nr_vecs /= 2)) {
2700 bio = bio_alloc_bioset(gfp_flags,
2701 nr_vecs, btrfs_bioset);
2706 bio->bi_bdev = bdev;
2707 bio->bi_iter.bi_sector = first_sector;
2708 btrfs_bio = btrfs_io_bio(bio);
2709 btrfs_bio->csum = NULL;
2710 btrfs_bio->csum_allocated = NULL;
2711 btrfs_bio->end_io = NULL;
2716 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2718 struct btrfs_io_bio *btrfs_bio;
2721 new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2723 btrfs_bio = btrfs_io_bio(new);
2724 btrfs_bio->csum = NULL;
2725 btrfs_bio->csum_allocated = NULL;
2726 btrfs_bio->end_io = NULL;
2731 /* this also allocates from the btrfs_bioset */
2732 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2734 struct btrfs_io_bio *btrfs_bio;
2737 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2739 btrfs_bio = btrfs_io_bio(bio);
2740 btrfs_bio->csum = NULL;
2741 btrfs_bio->csum_allocated = NULL;
2742 btrfs_bio->end_io = NULL;
2748 static int __must_check submit_one_bio(int rw, struct bio *bio,
2749 int mirror_num, unsigned long bio_flags)
2752 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2753 struct page *page = bvec->bv_page;
2754 struct extent_io_tree *tree = bio->bi_private;
2757 start = page_offset(page) + bvec->bv_offset;
2759 bio->bi_private = NULL;
2763 if (tree->ops && tree->ops->submit_bio_hook)
2764 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2765 mirror_num, bio_flags, start);
2767 btrfsic_submit_bio(rw, bio);
2769 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2775 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2776 unsigned long offset, size_t size, struct bio *bio,
2777 unsigned long bio_flags)
2780 if (tree->ops && tree->ops->merge_bio_hook)
2781 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2788 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2789 struct page *page, sector_t sector,
2790 size_t size, unsigned long offset,
2791 struct block_device *bdev,
2792 struct bio **bio_ret,
2793 unsigned long max_pages,
2794 bio_end_io_t end_io_func,
2796 unsigned long prev_bio_flags,
2797 unsigned long bio_flags)
2803 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2804 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2805 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2807 if (bio_ret && *bio_ret) {
2810 contig = bio->bi_iter.bi_sector == sector;
2812 contig = bio_end_sector(bio) == sector;
2814 if (prev_bio_flags != bio_flags || !contig ||
2815 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2816 bio_add_page(bio, page, page_size, offset) < page_size) {
2817 ret = submit_one_bio(rw, bio, mirror_num,
2826 if (this_compressed)
2829 nr = bio_get_nr_vecs(bdev);
2831 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2835 bio_add_page(bio, page, page_size, offset);
2836 bio->bi_end_io = end_io_func;
2837 bio->bi_private = tree;
2842 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2847 static void attach_extent_buffer_page(struct extent_buffer *eb,
2850 if (!PagePrivate(page)) {
2851 SetPagePrivate(page);
2852 page_cache_get(page);
2853 set_page_private(page, (unsigned long)eb);
2855 WARN_ON(page->private != (unsigned long)eb);
2859 void set_page_extent_mapped(struct page *page)
2861 if (!PagePrivate(page)) {
2862 SetPagePrivate(page);
2863 page_cache_get(page);
2864 set_page_private(page, EXTENT_PAGE_PRIVATE);
2868 static struct extent_map *
2869 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2870 u64 start, u64 len, get_extent_t *get_extent,
2871 struct extent_map **em_cached)
2873 struct extent_map *em;
2875 if (em_cached && *em_cached) {
2877 if (extent_map_in_tree(em) && start >= em->start &&
2878 start < extent_map_end(em)) {
2879 atomic_inc(&em->refs);
2883 free_extent_map(em);
2887 em = get_extent(inode, page, pg_offset, start, len, 0);
2888 if (em_cached && !IS_ERR_OR_NULL(em)) {
2890 atomic_inc(&em->refs);
2896 * basic readpage implementation. Locked extent state structs are inserted
2897 * into the tree that are removed when the IO is done (by the end_io
2899 * XXX JDM: This needs looking at to ensure proper page locking
2901 static int __do_readpage(struct extent_io_tree *tree,
2903 get_extent_t *get_extent,
2904 struct extent_map **em_cached,
2905 struct bio **bio, int mirror_num,
2906 unsigned long *bio_flags, int rw)
2908 struct inode *inode = page->mapping->host;
2909 u64 start = page_offset(page);
2910 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2914 u64 last_byte = i_size_read(inode);
2918 struct extent_map *em;
2919 struct block_device *bdev;
2922 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2923 size_t pg_offset = 0;
2925 size_t disk_io_size;
2926 size_t blocksize = inode->i_sb->s_blocksize;
2927 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2929 set_page_extent_mapped(page);
2932 if (!PageUptodate(page)) {
2933 if (cleancache_get_page(page) == 0) {
2934 BUG_ON(blocksize != PAGE_SIZE);
2935 unlock_extent(tree, start, end);
2940 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2942 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2945 iosize = PAGE_CACHE_SIZE - zero_offset;
2946 userpage = kmap_atomic(page);
2947 memset(userpage + zero_offset, 0, iosize);
2948 flush_dcache_page(page);
2949 kunmap_atomic(userpage);
2952 while (cur <= end) {
2953 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2955 if (cur >= last_byte) {
2957 struct extent_state *cached = NULL;
2959 iosize = PAGE_CACHE_SIZE - pg_offset;
2960 userpage = kmap_atomic(page);
2961 memset(userpage + pg_offset, 0, iosize);
2962 flush_dcache_page(page);
2963 kunmap_atomic(userpage);
2964 set_extent_uptodate(tree, cur, cur + iosize - 1,
2967 unlock_extent_cached(tree, cur,
2972 em = __get_extent_map(inode, page, pg_offset, cur,
2973 end - cur + 1, get_extent, em_cached);
2974 if (IS_ERR_OR_NULL(em)) {
2977 unlock_extent(tree, cur, end);
2980 extent_offset = cur - em->start;
2981 BUG_ON(extent_map_end(em) <= cur);
2984 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2985 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2986 extent_set_compress_type(&this_bio_flag,
2990 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2991 cur_end = min(extent_map_end(em) - 1, end);
2992 iosize = ALIGN(iosize, blocksize);
2993 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2994 disk_io_size = em->block_len;
2995 sector = em->block_start >> 9;
2997 sector = (em->block_start + extent_offset) >> 9;
2998 disk_io_size = iosize;
3001 block_start = em->block_start;
3002 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3003 block_start = EXTENT_MAP_HOLE;
3004 free_extent_map(em);
3007 /* we've found a hole, just zero and go on */
3008 if (block_start == EXTENT_MAP_HOLE) {
3010 struct extent_state *cached = NULL;
3012 userpage = kmap_atomic(page);
3013 memset(userpage + pg_offset, 0, iosize);
3014 flush_dcache_page(page);
3015 kunmap_atomic(userpage);
3017 set_extent_uptodate(tree, cur, cur + iosize - 1,
3019 unlock_extent_cached(tree, cur, cur + iosize - 1,
3022 pg_offset += iosize;
3025 /* the get_extent function already copied into the page */
3026 if (test_range_bit(tree, cur, cur_end,
3027 EXTENT_UPTODATE, 1, NULL)) {
3028 check_page_uptodate(tree, page);
3030 unlock_extent(tree, cur, cur + iosize - 1);
3032 pg_offset += iosize;
3035 /* we have an inline extent but it didn't get marked up
3036 * to date. Error out
3038 if (block_start == EXTENT_MAP_INLINE) {
3041 unlock_extent(tree, cur, cur + iosize - 1);
3043 pg_offset += iosize;
3048 ret = submit_extent_page(rw, tree, page,
3049 sector, disk_io_size, pg_offset,
3051 end_bio_extent_readpage, mirror_num,
3056 *bio_flags = this_bio_flag;
3060 unlock_extent(tree, cur, cur + iosize - 1);
3063 pg_offset += iosize;
3067 if (!PageError(page))
3068 SetPageUptodate(page);
3074 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3075 struct page *pages[], int nr_pages,
3077 get_extent_t *get_extent,
3078 struct extent_map **em_cached,
3079 struct bio **bio, int mirror_num,
3080 unsigned long *bio_flags, int rw)
3082 struct inode *inode;
3083 struct btrfs_ordered_extent *ordered;
3086 inode = pages[0]->mapping->host;
3088 lock_extent(tree, start, end);
3089 ordered = btrfs_lookup_ordered_range(inode, start,
3093 unlock_extent(tree, start, end);
3094 btrfs_start_ordered_extent(inode, ordered, 1);
3095 btrfs_put_ordered_extent(ordered);
3098 for (index = 0; index < nr_pages; index++) {
3099 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3100 mirror_num, bio_flags, rw);
3101 page_cache_release(pages[index]);
3105 static void __extent_readpages(struct extent_io_tree *tree,
3106 struct page *pages[],
3107 int nr_pages, get_extent_t *get_extent,
3108 struct extent_map **em_cached,
3109 struct bio **bio, int mirror_num,
3110 unsigned long *bio_flags, int rw)
3116 int first_index = 0;
3118 for (index = 0; index < nr_pages; index++) {
3119 page_start = page_offset(pages[index]);
3122 end = start + PAGE_CACHE_SIZE - 1;
3123 first_index = index;
3124 } else if (end + 1 == page_start) {
3125 end += PAGE_CACHE_SIZE;
3127 __do_contiguous_readpages(tree, &pages[first_index],
3128 index - first_index, start,
3129 end, get_extent, em_cached,
3130 bio, mirror_num, bio_flags,
3133 end = start + PAGE_CACHE_SIZE - 1;
3134 first_index = index;
3139 __do_contiguous_readpages(tree, &pages[first_index],
3140 index - first_index, start,
3141 end, get_extent, em_cached, bio,
3142 mirror_num, bio_flags, rw);
3145 static int __extent_read_full_page(struct extent_io_tree *tree,
3147 get_extent_t *get_extent,
3148 struct bio **bio, int mirror_num,
3149 unsigned long *bio_flags, int rw)
3151 struct inode *inode = page->mapping->host;
3152 struct btrfs_ordered_extent *ordered;
3153 u64 start = page_offset(page);
3154 u64 end = start + PAGE_CACHE_SIZE - 1;
3158 lock_extent(tree, start, end);
3159 ordered = btrfs_lookup_ordered_extent(inode, start);
3162 unlock_extent(tree, start, end);
3163 btrfs_start_ordered_extent(inode, ordered, 1);
3164 btrfs_put_ordered_extent(ordered);
3167 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3172 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3173 get_extent_t *get_extent, int mirror_num)
3175 struct bio *bio = NULL;
3176 unsigned long bio_flags = 0;
3179 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3182 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3186 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3187 get_extent_t *get_extent, int mirror_num)
3189 struct bio *bio = NULL;
3190 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3193 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3196 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3200 static noinline void update_nr_written(struct page *page,
3201 struct writeback_control *wbc,
3202 unsigned long nr_written)
3204 wbc->nr_to_write -= nr_written;
3205 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3206 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3207 page->mapping->writeback_index = page->index + nr_written;
3211 * helper for __extent_writepage, doing all of the delayed allocation setup.
3213 * This returns 1 if our fill_delalloc function did all the work required
3214 * to write the page (copy into inline extent). In this case the IO has
3215 * been started and the page is already unlocked.
3217 * This returns 0 if all went well (page still locked)
3218 * This returns < 0 if there were errors (page still locked)
3220 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3221 struct page *page, struct writeback_control *wbc,
3222 struct extent_page_data *epd,
3224 unsigned long *nr_written)
3226 struct extent_io_tree *tree = epd->tree;
3227 u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1;
3229 u64 delalloc_to_write = 0;
3230 u64 delalloc_end = 0;
3232 int page_started = 0;
3234 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3237 while (delalloc_end < page_end) {
3238 nr_delalloc = find_lock_delalloc_range(inode, tree,
3243 if (nr_delalloc == 0) {
3244 delalloc_start = delalloc_end + 1;
3247 ret = tree->ops->fill_delalloc(inode, page,
3252 /* File system has been set read-only */
3255 /* fill_delalloc should be return < 0 for error
3256 * but just in case, we use > 0 here meaning the
3257 * IO is started, so we don't want to return > 0
3258 * unless things are going well.
3260 ret = ret < 0 ? ret : -EIO;
3264 * delalloc_end is already one less than the total
3265 * length, so we don't subtract one from
3268 delalloc_to_write += (delalloc_end - delalloc_start +
3271 delalloc_start = delalloc_end + 1;
3273 if (wbc->nr_to_write < delalloc_to_write) {
3276 if (delalloc_to_write < thresh * 2)
3277 thresh = delalloc_to_write;
3278 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3282 /* did the fill delalloc function already unlock and start
3287 * we've unlocked the page, so we can't update
3288 * the mapping's writeback index, just update
3291 wbc->nr_to_write -= *nr_written;
3302 * helper for __extent_writepage. This calls the writepage start hooks,
3303 * and does the loop to map the page into extents and bios.
3305 * We return 1 if the IO is started and the page is unlocked,
3306 * 0 if all went well (page still locked)
3307 * < 0 if there were errors (page still locked)
3309 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3311 struct writeback_control *wbc,
3312 struct extent_page_data *epd,
3314 unsigned long nr_written,
3315 int write_flags, int *nr_ret)
3317 struct extent_io_tree *tree = epd->tree;
3318 u64 start = page_offset(page);
3319 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3326 struct extent_state *cached_state = NULL;
3327 struct extent_map *em;
3328 struct block_device *bdev;
3329 size_t pg_offset = 0;
3335 if (tree->ops && tree->ops->writepage_start_hook) {
3336 ret = tree->ops->writepage_start_hook(page, start,
3339 /* Fixup worker will requeue */
3341 wbc->pages_skipped++;
3343 redirty_page_for_writepage(wbc, page);
3345 update_nr_written(page, wbc, nr_written);
3353 * we don't want to touch the inode after unlocking the page,
3354 * so we update the mapping writeback index now
3356 update_nr_written(page, wbc, nr_written + 1);
3359 if (i_size <= start) {
3360 if (tree->ops && tree->ops->writepage_end_io_hook)
3361 tree->ops->writepage_end_io_hook(page, start,
3366 blocksize = inode->i_sb->s_blocksize;
3368 while (cur <= end) {
3370 if (cur >= i_size) {
3371 if (tree->ops && tree->ops->writepage_end_io_hook)
3372 tree->ops->writepage_end_io_hook(page, cur,
3376 em = epd->get_extent(inode, page, pg_offset, cur,
3378 if (IS_ERR_OR_NULL(em)) {
3380 ret = PTR_ERR_OR_ZERO(em);
3384 extent_offset = cur - em->start;
3385 em_end = extent_map_end(em);
3386 BUG_ON(em_end <= cur);
3388 iosize = min(em_end - cur, end - cur + 1);
3389 iosize = ALIGN(iosize, blocksize);
3390 sector = (em->block_start + extent_offset) >> 9;
3392 block_start = em->block_start;
3393 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3394 free_extent_map(em);
3398 * compressed and inline extents are written through other
3401 if (compressed || block_start == EXTENT_MAP_HOLE ||
3402 block_start == EXTENT_MAP_INLINE) {
3404 * end_io notification does not happen here for
3405 * compressed extents
3407 if (!compressed && tree->ops &&
3408 tree->ops->writepage_end_io_hook)
3409 tree->ops->writepage_end_io_hook(page, cur,
3412 else if (compressed) {
3413 /* we don't want to end_page_writeback on
3414 * a compressed extent. this happens
3421 pg_offset += iosize;
3425 if (tree->ops && tree->ops->writepage_io_hook) {
3426 ret = tree->ops->writepage_io_hook(page, cur,
3434 unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1;
3436 set_range_writeback(tree, cur, cur + iosize - 1);
3437 if (!PageWriteback(page)) {
3438 btrfs_err(BTRFS_I(inode)->root->fs_info,
3439 "page %lu not writeback, cur %llu end %llu",
3440 page->index, cur, end);
3443 ret = submit_extent_page(write_flags, tree, page,
3444 sector, iosize, pg_offset,
3445 bdev, &epd->bio, max_nr,
3446 end_bio_extent_writepage,
3452 pg_offset += iosize;
3460 /* drop our reference on any cached states */
3461 free_extent_state(cached_state);
3466 * the writepage semantics are similar to regular writepage. extent
3467 * records are inserted to lock ranges in the tree, and as dirty areas
3468 * are found, they are marked writeback. Then the lock bits are removed
3469 * and the end_io handler clears the writeback ranges
3471 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3474 struct inode *inode = page->mapping->host;
3475 struct extent_page_data *epd = data;
3476 u64 start = page_offset(page);
3477 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3480 size_t pg_offset = 0;
3481 loff_t i_size = i_size_read(inode);
3482 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3484 unsigned long nr_written = 0;
3486 if (wbc->sync_mode == WB_SYNC_ALL)
3487 write_flags = WRITE_SYNC;
3489 write_flags = WRITE;
3491 trace___extent_writepage(page, inode, wbc);
3493 WARN_ON(!PageLocked(page));
3495 ClearPageError(page);
3497 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3498 if (page->index > end_index ||
3499 (page->index == end_index && !pg_offset)) {
3500 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3505 if (page->index == end_index) {
3508 userpage = kmap_atomic(page);
3509 memset(userpage + pg_offset, 0,
3510 PAGE_CACHE_SIZE - pg_offset);
3511 kunmap_atomic(userpage);
3512 flush_dcache_page(page);
3517 set_page_extent_mapped(page);
3519 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3525 ret = __extent_writepage_io(inode, page, wbc, epd,
3526 i_size, nr_written, write_flags, &nr);
3532 /* make sure the mapping tag for page dirty gets cleared */
3533 set_page_writeback(page);
3534 end_page_writeback(page);
3536 if (PageError(page)) {
3537 ret = ret < 0 ? ret : -EIO;
3538 end_extent_writepage(page, ret, start, page_end);
3547 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3549 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3550 TASK_UNINTERRUPTIBLE);
3553 static noinline_for_stack int
3554 lock_extent_buffer_for_io(struct extent_buffer *eb,
3555 struct btrfs_fs_info *fs_info,
3556 struct extent_page_data *epd)
3558 unsigned long i, num_pages;
3562 if (!btrfs_try_tree_write_lock(eb)) {
3564 flush_write_bio(epd);
3565 btrfs_tree_lock(eb);
3568 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3569 btrfs_tree_unlock(eb);
3573 flush_write_bio(epd);
3577 wait_on_extent_buffer_writeback(eb);
3578 btrfs_tree_lock(eb);
3579 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3581 btrfs_tree_unlock(eb);
3586 * We need to do this to prevent races in people who check if the eb is
3587 * under IO since we can end up having no IO bits set for a short period
3590 spin_lock(&eb->refs_lock);
3591 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3592 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3593 spin_unlock(&eb->refs_lock);
3594 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3595 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3597 fs_info->dirty_metadata_batch);
3600 spin_unlock(&eb->refs_lock);
3603 btrfs_tree_unlock(eb);
3608 num_pages = num_extent_pages(eb->start, eb->len);
3609 for (i = 0; i < num_pages; i++) {
3610 struct page *p = eb->pages[i];
3612 if (!trylock_page(p)) {
3614 flush_write_bio(epd);
3624 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3626 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3627 smp_mb__after_atomic();
3628 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3631 static void set_btree_ioerr(struct page *page)
3633 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3634 struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode);
3637 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3641 * If writeback for a btree extent that doesn't belong to a log tree
3642 * failed, increment the counter transaction->eb_write_errors.
3643 * We do this because while the transaction is running and before it's
3644 * committing (when we call filemap_fdata[write|wait]_range against
3645 * the btree inode), we might have
3646 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3647 * returns an error or an error happens during writeback, when we're
3648 * committing the transaction we wouldn't know about it, since the pages
3649 * can be no longer dirty nor marked anymore for writeback (if a
3650 * subsequent modification to the extent buffer didn't happen before the
3651 * transaction commit), which makes filemap_fdata[write|wait]_range not
3652 * able to find the pages tagged with SetPageError at transaction
3653 * commit time. So if this happens we must abort the transaction,
3654 * otherwise we commit a super block with btree roots that point to
3655 * btree nodes/leafs whose content on disk is invalid - either garbage
3656 * or the content of some node/leaf from a past generation that got
3657 * cowed or deleted and is no longer valid.
3659 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3660 * not be enough - we need to distinguish between log tree extents vs
3661 * non-log tree extents, and the next filemap_fdatawait_range() call
3662 * will catch and clear such errors in the mapping - and that call might
3663 * be from a log sync and not from a transaction commit. Also, checking
3664 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3665 * not done and would not be reliable - the eb might have been released
3666 * from memory and reading it back again means that flag would not be
3667 * set (since it's a runtime flag, not persisted on disk).
3669 * Using the flags below in the btree inode also makes us achieve the
3670 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3671 * writeback for all dirty pages and before filemap_fdatawait_range()
3672 * is called, the writeback for all dirty pages had already finished
3673 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3674 * filemap_fdatawait_range() would return success, as it could not know
3675 * that writeback errors happened (the pages were no longer tagged for
3678 switch (eb->log_index) {
3680 set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags);
3683 set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
3686 set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
3689 BUG(); /* unexpected, logic error */
3693 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3695 struct bio_vec *bvec;
3696 struct extent_buffer *eb;
3699 bio_for_each_segment_all(bvec, bio, i) {
3700 struct page *page = bvec->bv_page;
3702 eb = (struct extent_buffer *)page->private;
3704 done = atomic_dec_and_test(&eb->io_pages);
3706 if (err || test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3707 ClearPageUptodate(page);
3708 set_btree_ioerr(page);
3711 end_page_writeback(page);
3716 end_extent_buffer_writeback(eb);
3722 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3723 struct btrfs_fs_info *fs_info,
3724 struct writeback_control *wbc,
3725 struct extent_page_data *epd)
3727 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3728 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3729 u64 offset = eb->start;
3730 unsigned long i, num_pages;
3731 unsigned long bio_flags = 0;
3732 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3735 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3736 num_pages = num_extent_pages(eb->start, eb->len);
3737 atomic_set(&eb->io_pages, num_pages);
3738 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3739 bio_flags = EXTENT_BIO_TREE_LOG;
3741 for (i = 0; i < num_pages; i++) {
3742 struct page *p = eb->pages[i];
3744 clear_page_dirty_for_io(p);
3745 set_page_writeback(p);
3746 ret = submit_extent_page(rw, tree, p, offset >> 9,
3747 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3748 -1, end_bio_extent_buffer_writepage,
3749 0, epd->bio_flags, bio_flags);
3750 epd->bio_flags = bio_flags;
3753 end_page_writeback(p);
3754 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3755 end_extent_buffer_writeback(eb);
3759 offset += PAGE_CACHE_SIZE;
3760 update_nr_written(p, wbc, 1);
3764 if (unlikely(ret)) {
3765 for (; i < num_pages; i++) {
3766 struct page *p = eb->pages[i];
3767 clear_page_dirty_for_io(p);
3775 int btree_write_cache_pages(struct address_space *mapping,
3776 struct writeback_control *wbc)
3778 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3779 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3780 struct extent_buffer *eb, *prev_eb = NULL;
3781 struct extent_page_data epd = {
3785 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3790 int nr_to_write_done = 0;
3791 struct pagevec pvec;
3794 pgoff_t end; /* Inclusive */
3798 pagevec_init(&pvec, 0);
3799 if (wbc->range_cyclic) {
3800 index = mapping->writeback_index; /* Start from prev offset */
3803 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3804 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3807 if (wbc->sync_mode == WB_SYNC_ALL)
3808 tag = PAGECACHE_TAG_TOWRITE;
3810 tag = PAGECACHE_TAG_DIRTY;
3812 if (wbc->sync_mode == WB_SYNC_ALL)
3813 tag_pages_for_writeback(mapping, index, end);
3814 while (!done && !nr_to_write_done && (index <= end) &&
3815 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3816 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3820 for (i = 0; i < nr_pages; i++) {
3821 struct page *page = pvec.pages[i];
3823 if (!PagePrivate(page))
3826 if (!wbc->range_cyclic && page->index > end) {
3831 spin_lock(&mapping->private_lock);
3832 if (!PagePrivate(page)) {
3833 spin_unlock(&mapping->private_lock);
3837 eb = (struct extent_buffer *)page->private;
3840 * Shouldn't happen and normally this would be a BUG_ON
3841 * but no sense in crashing the users box for something
3842 * we can survive anyway.
3845 spin_unlock(&mapping->private_lock);
3849 if (eb == prev_eb) {
3850 spin_unlock(&mapping->private_lock);
3854 ret = atomic_inc_not_zero(&eb->refs);
3855 spin_unlock(&mapping->private_lock);
3860 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3862 free_extent_buffer(eb);
3866 ret = write_one_eb(eb, fs_info, wbc, &epd);
3869 free_extent_buffer(eb);
3872 free_extent_buffer(eb);
3875 * the filesystem may choose to bump up nr_to_write.
3876 * We have to make sure to honor the new nr_to_write
3879 nr_to_write_done = wbc->nr_to_write <= 0;
3881 pagevec_release(&pvec);
3884 if (!scanned && !done) {
3886 * We hit the last page and there is more work to be done: wrap
3887 * back to the start of the file
3893 flush_write_bio(&epd);
3898 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3899 * @mapping: address space structure to write
3900 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3901 * @writepage: function called for each page
3902 * @data: data passed to writepage function
3904 * If a page is already under I/O, write_cache_pages() skips it, even
3905 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3906 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3907 * and msync() need to guarantee that all the data which was dirty at the time
3908 * the call was made get new I/O started against them. If wbc->sync_mode is
3909 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3910 * existing IO to complete.
3912 static int extent_write_cache_pages(struct extent_io_tree *tree,
3913 struct address_space *mapping,
3914 struct writeback_control *wbc,
3915 writepage_t writepage, void *data,
3916 void (*flush_fn)(void *))
3918 struct inode *inode = mapping->host;
3922 int nr_to_write_done = 0;
3923 struct pagevec pvec;
3926 pgoff_t end; /* Inclusive */
3931 * We have to hold onto the inode so that ordered extents can do their
3932 * work when the IO finishes. The alternative to this is failing to add
3933 * an ordered extent if the igrab() fails there and that is a huge pain
3934 * to deal with, so instead just hold onto the inode throughout the
3935 * writepages operation. If it fails here we are freeing up the inode
3936 * anyway and we'd rather not waste our time writing out stuff that is
3937 * going to be truncated anyway.
3942 pagevec_init(&pvec, 0);
3943 if (wbc->range_cyclic) {
3944 index = mapping->writeback_index; /* Start from prev offset */
3947 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3948 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3951 if (wbc->sync_mode == WB_SYNC_ALL)
3952 tag = PAGECACHE_TAG_TOWRITE;
3954 tag = PAGECACHE_TAG_DIRTY;
3956 if (wbc->sync_mode == WB_SYNC_ALL)
3957 tag_pages_for_writeback(mapping, index, end);
3958 while (!done && !nr_to_write_done && (index <= end) &&
3959 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3960 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3964 for (i = 0; i < nr_pages; i++) {
3965 struct page *page = pvec.pages[i];
3968 * At this point we hold neither mapping->tree_lock nor
3969 * lock on the page itself: the page may be truncated or
3970 * invalidated (changing page->mapping to NULL), or even
3971 * swizzled back from swapper_space to tmpfs file
3974 if (!trylock_page(page)) {
3979 if (unlikely(page->mapping != mapping)) {
3984 if (!wbc->range_cyclic && page->index > end) {
3990 if (wbc->sync_mode != WB_SYNC_NONE) {
3991 if (PageWriteback(page))
3993 wait_on_page_writeback(page);
3996 if (PageWriteback(page) ||
3997 !clear_page_dirty_for_io(page)) {
4002 ret = (*writepage)(page, wbc, data);
4004 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4008 if (!err && ret < 0)
4012 * the filesystem may choose to bump up nr_to_write.
4013 * We have to make sure to honor the new nr_to_write
4016 nr_to_write_done = wbc->nr_to_write <= 0;
4018 pagevec_release(&pvec);
4021 if (!scanned && !done && !err) {
4023 * We hit the last page and there is more work to be done: wrap
4024 * back to the start of the file
4030 btrfs_add_delayed_iput(inode);
4034 static void flush_epd_write_bio(struct extent_page_data *epd)
4043 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
4044 BUG_ON(ret < 0); /* -ENOMEM */
4049 static noinline void flush_write_bio(void *data)
4051 struct extent_page_data *epd = data;
4052 flush_epd_write_bio(epd);
4055 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4056 get_extent_t *get_extent,
4057 struct writeback_control *wbc)
4060 struct extent_page_data epd = {
4063 .get_extent = get_extent,
4065 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4069 ret = __extent_writepage(page, wbc, &epd);
4071 flush_epd_write_bio(&epd);
4075 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4076 u64 start, u64 end, get_extent_t *get_extent,
4080 struct address_space *mapping = inode->i_mapping;
4082 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
4085 struct extent_page_data epd = {
4088 .get_extent = get_extent,
4090 .sync_io = mode == WB_SYNC_ALL,
4093 struct writeback_control wbc_writepages = {
4095 .nr_to_write = nr_pages * 2,
4096 .range_start = start,
4097 .range_end = end + 1,
4100 while (start <= end) {
4101 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
4102 if (clear_page_dirty_for_io(page))
4103 ret = __extent_writepage(page, &wbc_writepages, &epd);
4105 if (tree->ops && tree->ops->writepage_end_io_hook)
4106 tree->ops->writepage_end_io_hook(page, start,
4107 start + PAGE_CACHE_SIZE - 1,
4111 page_cache_release(page);
4112 start += PAGE_CACHE_SIZE;
4115 flush_epd_write_bio(&epd);
4119 int extent_writepages(struct extent_io_tree *tree,
4120 struct address_space *mapping,
4121 get_extent_t *get_extent,
4122 struct writeback_control *wbc)
4125 struct extent_page_data epd = {
4128 .get_extent = get_extent,
4130 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4134 ret = extent_write_cache_pages(tree, mapping, wbc,
4135 __extent_writepage, &epd,
4137 flush_epd_write_bio(&epd);
4141 int extent_readpages(struct extent_io_tree *tree,
4142 struct address_space *mapping,
4143 struct list_head *pages, unsigned nr_pages,
4144 get_extent_t get_extent)
4146 struct bio *bio = NULL;
4148 unsigned long bio_flags = 0;
4149 struct page *pagepool[16];
4151 struct extent_map *em_cached = NULL;
4154 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4155 page = list_entry(pages->prev, struct page, lru);
4157 prefetchw(&page->flags);
4158 list_del(&page->lru);
4159 if (add_to_page_cache_lru(page, mapping,
4160 page->index, GFP_NOFS)) {
4161 page_cache_release(page);
4165 pagepool[nr++] = page;
4166 if (nr < ARRAY_SIZE(pagepool))
4168 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4169 &bio, 0, &bio_flags, READ);
4173 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4174 &bio, 0, &bio_flags, READ);
4177 free_extent_map(em_cached);
4179 BUG_ON(!list_empty(pages));
4181 return submit_one_bio(READ, bio, 0, bio_flags);
4186 * basic invalidatepage code, this waits on any locked or writeback
4187 * ranges corresponding to the page, and then deletes any extent state
4188 * records from the tree
4190 int extent_invalidatepage(struct extent_io_tree *tree,
4191 struct page *page, unsigned long offset)
4193 struct extent_state *cached_state = NULL;
4194 u64 start = page_offset(page);
4195 u64 end = start + PAGE_CACHE_SIZE - 1;
4196 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4198 start += ALIGN(offset, blocksize);
4202 lock_extent_bits(tree, start, end, 0, &cached_state);
4203 wait_on_page_writeback(page);
4204 clear_extent_bit(tree, start, end,
4205 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4206 EXTENT_DO_ACCOUNTING,
4207 1, 1, &cached_state, GFP_NOFS);
4212 * a helper for releasepage, this tests for areas of the page that
4213 * are locked or under IO and drops the related state bits if it is safe
4216 static int try_release_extent_state(struct extent_map_tree *map,
4217 struct extent_io_tree *tree,
4218 struct page *page, gfp_t mask)
4220 u64 start = page_offset(page);
4221 u64 end = start + PAGE_CACHE_SIZE - 1;
4224 if (test_range_bit(tree, start, end,
4225 EXTENT_IOBITS, 0, NULL))
4228 if ((mask & GFP_NOFS) == GFP_NOFS)
4231 * at this point we can safely clear everything except the
4232 * locked bit and the nodatasum bit
4234 ret = clear_extent_bit(tree, start, end,
4235 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4238 /* if clear_extent_bit failed for enomem reasons,
4239 * we can't allow the release to continue.
4250 * a helper for releasepage. As long as there are no locked extents
4251 * in the range corresponding to the page, both state records and extent
4252 * map records are removed
4254 int try_release_extent_mapping(struct extent_map_tree *map,
4255 struct extent_io_tree *tree, struct page *page,
4258 struct extent_map *em;
4259 u64 start = page_offset(page);
4260 u64 end = start + PAGE_CACHE_SIZE - 1;
4262 if ((mask & __GFP_WAIT) &&
4263 page->mapping->host->i_size > 16 * 1024 * 1024) {
4265 while (start <= end) {
4266 len = end - start + 1;
4267 write_lock(&map->lock);
4268 em = lookup_extent_mapping(map, start, len);
4270 write_unlock(&map->lock);
4273 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4274 em->start != start) {
4275 write_unlock(&map->lock);
4276 free_extent_map(em);
4279 if (!test_range_bit(tree, em->start,
4280 extent_map_end(em) - 1,
4281 EXTENT_LOCKED | EXTENT_WRITEBACK,
4283 remove_extent_mapping(map, em);
4284 /* once for the rb tree */
4285 free_extent_map(em);
4287 start = extent_map_end(em);
4288 write_unlock(&map->lock);
4291 free_extent_map(em);
4294 return try_release_extent_state(map, tree, page, mask);
4298 * helper function for fiemap, which doesn't want to see any holes.
4299 * This maps until we find something past 'last'
4301 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4304 get_extent_t *get_extent)
4306 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4307 struct extent_map *em;
4314 len = last - offset;
4317 len = ALIGN(len, sectorsize);
4318 em = get_extent(inode, NULL, 0, offset, len, 0);
4319 if (IS_ERR_OR_NULL(em))
4322 /* if this isn't a hole return it */
4323 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4324 em->block_start != EXTENT_MAP_HOLE) {
4328 /* this is a hole, advance to the next extent */
4329 offset = extent_map_end(em);
4330 free_extent_map(em);
4337 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4338 __u64 start, __u64 len, get_extent_t *get_extent)
4342 u64 max = start + len;
4346 u64 last_for_get_extent = 0;
4348 u64 isize = i_size_read(inode);
4349 struct btrfs_key found_key;
4350 struct extent_map *em = NULL;
4351 struct extent_state *cached_state = NULL;
4352 struct btrfs_path *path;
4353 struct btrfs_root *root = BTRFS_I(inode)->root;
4362 path = btrfs_alloc_path();
4365 path->leave_spinning = 1;
4367 start = round_down(start, BTRFS_I(inode)->root->sectorsize);
4368 len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start;
4371 * lookup the last file extent. We're not using i_size here
4372 * because there might be preallocation past i_size
4374 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4377 btrfs_free_path(path);
4382 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4383 found_type = found_key.type;
4385 /* No extents, but there might be delalloc bits */
4386 if (found_key.objectid != btrfs_ino(inode) ||
4387 found_type != BTRFS_EXTENT_DATA_KEY) {
4388 /* have to trust i_size as the end */
4390 last_for_get_extent = isize;
4393 * remember the start of the last extent. There are a
4394 * bunch of different factors that go into the length of the
4395 * extent, so its much less complex to remember where it started
4397 last = found_key.offset;
4398 last_for_get_extent = last + 1;
4400 btrfs_release_path(path);
4403 * we might have some extents allocated but more delalloc past those
4404 * extents. so, we trust isize unless the start of the last extent is
4409 last_for_get_extent = isize;
4412 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4415 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4425 u64 offset_in_extent = 0;
4427 /* break if the extent we found is outside the range */
4428 if (em->start >= max || extent_map_end(em) < off)
4432 * get_extent may return an extent that starts before our
4433 * requested range. We have to make sure the ranges
4434 * we return to fiemap always move forward and don't
4435 * overlap, so adjust the offsets here
4437 em_start = max(em->start, off);
4440 * record the offset from the start of the extent
4441 * for adjusting the disk offset below. Only do this if the
4442 * extent isn't compressed since our in ram offset may be past
4443 * what we have actually allocated on disk.
4445 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4446 offset_in_extent = em_start - em->start;
4447 em_end = extent_map_end(em);
4448 em_len = em_end - em_start;
4453 * bump off for our next call to get_extent
4455 off = extent_map_end(em);
4459 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4461 flags |= FIEMAP_EXTENT_LAST;
4462 } else if (em->block_start == EXTENT_MAP_INLINE) {
4463 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4464 FIEMAP_EXTENT_NOT_ALIGNED);
4465 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4466 flags |= (FIEMAP_EXTENT_DELALLOC |
4467 FIEMAP_EXTENT_UNKNOWN);
4468 } else if (fieinfo->fi_extents_max) {
4469 u64 bytenr = em->block_start -
4470 (em->start - em->orig_start);
4472 disko = em->block_start + offset_in_extent;
4475 * As btrfs supports shared space, this information
4476 * can be exported to userspace tools via
4477 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4478 * then we're just getting a count and we can skip the
4481 ret = btrfs_check_shared(NULL, root->fs_info,
4483 btrfs_ino(inode), bytenr);
4487 flags |= FIEMAP_EXTENT_SHARED;
4490 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4491 flags |= FIEMAP_EXTENT_ENCODED;
4493 free_extent_map(em);
4495 if ((em_start >= last) || em_len == (u64)-1 ||
4496 (last == (u64)-1 && isize <= em_end)) {
4497 flags |= FIEMAP_EXTENT_LAST;
4501 /* now scan forward to see if this is really the last extent. */
4502 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4509 flags |= FIEMAP_EXTENT_LAST;
4512 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4518 free_extent_map(em);
4520 btrfs_free_path(path);
4521 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4522 &cached_state, GFP_NOFS);
4526 static void __free_extent_buffer(struct extent_buffer *eb)
4528 btrfs_leak_debug_del(&eb->leak_list);
4529 kmem_cache_free(extent_buffer_cache, eb);
4532 int extent_buffer_under_io(struct extent_buffer *eb)
4534 return (atomic_read(&eb->io_pages) ||
4535 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4536 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4540 * Helper for releasing extent buffer page.
4542 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4544 unsigned long index;
4546 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4548 BUG_ON(extent_buffer_under_io(eb));
4550 index = num_extent_pages(eb->start, eb->len);
4556 page = eb->pages[index];
4557 if (page && mapped) {
4558 spin_lock(&page->mapping->private_lock);
4560 * We do this since we'll remove the pages after we've
4561 * removed the eb from the radix tree, so we could race
4562 * and have this page now attached to the new eb. So
4563 * only clear page_private if it's still connected to
4566 if (PagePrivate(page) &&
4567 page->private == (unsigned long)eb) {
4568 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4569 BUG_ON(PageDirty(page));
4570 BUG_ON(PageWriteback(page));
4572 * We need to make sure we haven't be attached
4575 ClearPagePrivate(page);
4576 set_page_private(page, 0);
4577 /* One for the page private */
4578 page_cache_release(page);
4580 spin_unlock(&page->mapping->private_lock);
4584 /* One for when we alloced the page */
4585 page_cache_release(page);
4587 } while (index != 0);
4591 * Helper for releasing the extent buffer.
4593 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4595 btrfs_release_extent_buffer_page(eb);
4596 __free_extent_buffer(eb);
4599 static struct extent_buffer *
4600 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4601 unsigned long len, gfp_t mask)
4603 struct extent_buffer *eb = NULL;
4605 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4610 eb->fs_info = fs_info;
4612 rwlock_init(&eb->lock);
4613 atomic_set(&eb->write_locks, 0);
4614 atomic_set(&eb->read_locks, 0);
4615 atomic_set(&eb->blocking_readers, 0);
4616 atomic_set(&eb->blocking_writers, 0);
4617 atomic_set(&eb->spinning_readers, 0);
4618 atomic_set(&eb->spinning_writers, 0);
4619 eb->lock_nested = 0;
4620 init_waitqueue_head(&eb->write_lock_wq);
4621 init_waitqueue_head(&eb->read_lock_wq);
4623 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4625 spin_lock_init(&eb->refs_lock);
4626 atomic_set(&eb->refs, 1);
4627 atomic_set(&eb->io_pages, 0);
4630 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4632 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4633 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4634 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4639 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4643 struct extent_buffer *new;
4644 unsigned long num_pages = num_extent_pages(src->start, src->len);
4646 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4650 for (i = 0; i < num_pages; i++) {
4651 p = alloc_page(GFP_NOFS);
4653 btrfs_release_extent_buffer(new);
4656 attach_extent_buffer_page(new, p);
4657 WARN_ON(PageDirty(p));
4662 copy_extent_buffer(new, src, 0, 0, src->len);
4663 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4664 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4669 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4671 struct extent_buffer *eb;
4672 unsigned long num_pages = num_extent_pages(0, len);
4675 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4679 for (i = 0; i < num_pages; i++) {
4680 eb->pages[i] = alloc_page(GFP_NOFS);
4684 set_extent_buffer_uptodate(eb);
4685 btrfs_set_header_nritems(eb, 0);
4686 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4691 __free_page(eb->pages[i - 1]);
4692 __free_extent_buffer(eb);
4696 static void check_buffer_tree_ref(struct extent_buffer *eb)
4699 /* the ref bit is tricky. We have to make sure it is set
4700 * if we have the buffer dirty. Otherwise the
4701 * code to free a buffer can end up dropping a dirty
4704 * Once the ref bit is set, it won't go away while the
4705 * buffer is dirty or in writeback, and it also won't
4706 * go away while we have the reference count on the
4709 * We can't just set the ref bit without bumping the
4710 * ref on the eb because free_extent_buffer might
4711 * see the ref bit and try to clear it. If this happens
4712 * free_extent_buffer might end up dropping our original
4713 * ref by mistake and freeing the page before we are able
4714 * to add one more ref.
4716 * So bump the ref count first, then set the bit. If someone
4717 * beat us to it, drop the ref we added.
4719 refs = atomic_read(&eb->refs);
4720 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4723 spin_lock(&eb->refs_lock);
4724 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4725 atomic_inc(&eb->refs);
4726 spin_unlock(&eb->refs_lock);
4729 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4730 struct page *accessed)
4732 unsigned long num_pages, i;
4734 check_buffer_tree_ref(eb);
4736 num_pages = num_extent_pages(eb->start, eb->len);
4737 for (i = 0; i < num_pages; i++) {
4738 struct page *p = eb->pages[i];
4741 mark_page_accessed(p);
4745 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4748 struct extent_buffer *eb;
4751 eb = radix_tree_lookup(&fs_info->buffer_radix,
4752 start >> PAGE_CACHE_SHIFT);
4753 if (eb && atomic_inc_not_zero(&eb->refs)) {
4755 mark_extent_buffer_accessed(eb, NULL);
4763 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4764 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4765 u64 start, unsigned long len)
4767 struct extent_buffer *eb, *exists = NULL;
4770 eb = find_extent_buffer(fs_info, start);
4773 eb = alloc_dummy_extent_buffer(start, len);
4776 eb->fs_info = fs_info;
4778 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4781 spin_lock(&fs_info->buffer_lock);
4782 ret = radix_tree_insert(&fs_info->buffer_radix,
4783 start >> PAGE_CACHE_SHIFT, eb);
4784 spin_unlock(&fs_info->buffer_lock);
4785 radix_tree_preload_end();
4786 if (ret == -EEXIST) {
4787 exists = find_extent_buffer(fs_info, start);
4793 check_buffer_tree_ref(eb);
4794 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4797 * We will free dummy extent buffer's if they come into
4798 * free_extent_buffer with a ref count of 2, but if we are using this we
4799 * want the buffers to stay in memory until we're done with them, so
4800 * bump the ref count again.
4802 atomic_inc(&eb->refs);
4805 btrfs_release_extent_buffer(eb);
4810 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4811 u64 start, unsigned long len)
4813 unsigned long num_pages = num_extent_pages(start, len);
4815 unsigned long index = start >> PAGE_CACHE_SHIFT;
4816 struct extent_buffer *eb;
4817 struct extent_buffer *exists = NULL;
4819 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4823 eb = find_extent_buffer(fs_info, start);
4827 eb = __alloc_extent_buffer(fs_info, start, len, GFP_NOFS);
4831 for (i = 0; i < num_pages; i++, index++) {
4832 p = find_or_create_page(mapping, index, GFP_NOFS);
4836 spin_lock(&mapping->private_lock);
4837 if (PagePrivate(p)) {
4839 * We could have already allocated an eb for this page
4840 * and attached one so lets see if we can get a ref on
4841 * the existing eb, and if we can we know it's good and
4842 * we can just return that one, else we know we can just
4843 * overwrite page->private.
4845 exists = (struct extent_buffer *)p->private;
4846 if (atomic_inc_not_zero(&exists->refs)) {
4847 spin_unlock(&mapping->private_lock);
4849 page_cache_release(p);
4850 mark_extent_buffer_accessed(exists, p);
4855 * Do this so attach doesn't complain and we need to
4856 * drop the ref the old guy had.
4858 ClearPagePrivate(p);
4859 WARN_ON(PageDirty(p));
4860 page_cache_release(p);
4862 attach_extent_buffer_page(eb, p);
4863 spin_unlock(&mapping->private_lock);
4864 WARN_ON(PageDirty(p));
4866 if (!PageUptodate(p))
4870 * see below about how we avoid a nasty race with release page
4871 * and why we unlock later
4875 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4877 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4881 spin_lock(&fs_info->buffer_lock);
4882 ret = radix_tree_insert(&fs_info->buffer_radix,
4883 start >> PAGE_CACHE_SHIFT, eb);
4884 spin_unlock(&fs_info->buffer_lock);
4885 radix_tree_preload_end();
4886 if (ret == -EEXIST) {
4887 exists = find_extent_buffer(fs_info, start);
4893 /* add one reference for the tree */
4894 check_buffer_tree_ref(eb);
4895 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4898 * there is a race where release page may have
4899 * tried to find this extent buffer in the radix
4900 * but failed. It will tell the VM it is safe to
4901 * reclaim the, and it will clear the page private bit.
4902 * We must make sure to set the page private bit properly
4903 * after the extent buffer is in the radix tree so
4904 * it doesn't get lost
4906 SetPageChecked(eb->pages[0]);
4907 for (i = 1; i < num_pages; i++) {
4909 ClearPageChecked(p);
4912 unlock_page(eb->pages[0]);
4916 for (i = 0; i < num_pages; i++) {
4918 unlock_page(eb->pages[i]);
4921 WARN_ON(!atomic_dec_and_test(&eb->refs));
4922 btrfs_release_extent_buffer(eb);
4926 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4928 struct extent_buffer *eb =
4929 container_of(head, struct extent_buffer, rcu_head);
4931 __free_extent_buffer(eb);
4934 /* Expects to have eb->eb_lock already held */
4935 static int release_extent_buffer(struct extent_buffer *eb)
4937 WARN_ON(atomic_read(&eb->refs) == 0);
4938 if (atomic_dec_and_test(&eb->refs)) {
4939 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4940 struct btrfs_fs_info *fs_info = eb->fs_info;
4942 spin_unlock(&eb->refs_lock);
4944 spin_lock(&fs_info->buffer_lock);
4945 radix_tree_delete(&fs_info->buffer_radix,
4946 eb->start >> PAGE_CACHE_SHIFT);
4947 spin_unlock(&fs_info->buffer_lock);
4949 spin_unlock(&eb->refs_lock);
4952 /* Should be safe to release our pages at this point */
4953 btrfs_release_extent_buffer_page(eb);
4954 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4957 spin_unlock(&eb->refs_lock);
4962 void free_extent_buffer(struct extent_buffer *eb)
4970 refs = atomic_read(&eb->refs);
4973 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4978 spin_lock(&eb->refs_lock);
4979 if (atomic_read(&eb->refs) == 2 &&
4980 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4981 atomic_dec(&eb->refs);
4983 if (atomic_read(&eb->refs) == 2 &&
4984 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4985 !extent_buffer_under_io(eb) &&
4986 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4987 atomic_dec(&eb->refs);
4990 * I know this is terrible, but it's temporary until we stop tracking
4991 * the uptodate bits and such for the extent buffers.
4993 release_extent_buffer(eb);
4996 void free_extent_buffer_stale(struct extent_buffer *eb)
5001 spin_lock(&eb->refs_lock);
5002 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5004 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5005 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5006 atomic_dec(&eb->refs);
5007 release_extent_buffer(eb);
5010 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5013 unsigned long num_pages;
5016 num_pages = num_extent_pages(eb->start, eb->len);
5018 for (i = 0; i < num_pages; i++) {
5019 page = eb->pages[i];
5020 if (!PageDirty(page))
5024 WARN_ON(!PagePrivate(page));
5026 clear_page_dirty_for_io(page);
5027 spin_lock_irq(&page->mapping->tree_lock);
5028 if (!PageDirty(page)) {
5029 radix_tree_tag_clear(&page->mapping->page_tree,
5031 PAGECACHE_TAG_DIRTY);
5033 spin_unlock_irq(&page->mapping->tree_lock);
5034 ClearPageError(page);
5037 WARN_ON(atomic_read(&eb->refs) == 0);
5040 int set_extent_buffer_dirty(struct extent_buffer *eb)
5043 unsigned long num_pages;
5046 check_buffer_tree_ref(eb);
5048 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5050 num_pages = num_extent_pages(eb->start, eb->len);
5051 WARN_ON(atomic_read(&eb->refs) == 0);
5052 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5054 for (i = 0; i < num_pages; i++)
5055 set_page_dirty(eb->pages[i]);
5059 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
5063 unsigned long num_pages;
5065 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5066 num_pages = num_extent_pages(eb->start, eb->len);
5067 for (i = 0; i < num_pages; i++) {
5068 page = eb->pages[i];
5070 ClearPageUptodate(page);
5075 int set_extent_buffer_uptodate(struct extent_buffer *eb)
5079 unsigned long num_pages;
5081 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5082 num_pages = num_extent_pages(eb->start, eb->len);
5083 for (i = 0; i < num_pages; i++) {
5084 page = eb->pages[i];
5085 SetPageUptodate(page);
5090 int extent_buffer_uptodate(struct extent_buffer *eb)
5092 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5095 int read_extent_buffer_pages(struct extent_io_tree *tree,
5096 struct extent_buffer *eb, u64 start, int wait,
5097 get_extent_t *get_extent, int mirror_num)
5100 unsigned long start_i;
5104 int locked_pages = 0;
5105 int all_uptodate = 1;
5106 unsigned long num_pages;
5107 unsigned long num_reads = 0;
5108 struct bio *bio = NULL;
5109 unsigned long bio_flags = 0;
5111 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5115 WARN_ON(start < eb->start);
5116 start_i = (start >> PAGE_CACHE_SHIFT) -
5117 (eb->start >> PAGE_CACHE_SHIFT);
5122 num_pages = num_extent_pages(eb->start, eb->len);
5123 for (i = start_i; i < num_pages; i++) {
5124 page = eb->pages[i];
5125 if (wait == WAIT_NONE) {
5126 if (!trylock_page(page))
5132 if (!PageUptodate(page)) {
5139 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5143 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5144 eb->read_mirror = 0;
5145 atomic_set(&eb->io_pages, num_reads);
5146 for (i = start_i; i < num_pages; i++) {
5147 page = eb->pages[i];
5148 if (!PageUptodate(page)) {
5149 ClearPageError(page);
5150 err = __extent_read_full_page(tree, page,
5152 mirror_num, &bio_flags,
5162 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
5168 if (ret || wait != WAIT_COMPLETE)
5171 for (i = start_i; i < num_pages; i++) {
5172 page = eb->pages[i];
5173 wait_on_page_locked(page);
5174 if (!PageUptodate(page))
5182 while (locked_pages > 0) {
5183 page = eb->pages[i];
5191 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
5192 unsigned long start,
5199 char *dst = (char *)dstv;
5200 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5201 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5203 WARN_ON(start > eb->len);
5204 WARN_ON(start + len > eb->start + eb->len);
5206 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5209 page = eb->pages[i];
5211 cur = min(len, (PAGE_CACHE_SIZE - offset));
5212 kaddr = page_address(page);
5213 memcpy(dst, kaddr + offset, cur);
5222 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv,
5223 unsigned long start,
5230 char __user *dst = (char __user *)dstv;
5231 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5232 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5235 WARN_ON(start > eb->len);
5236 WARN_ON(start + len > eb->start + eb->len);
5238 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5241 page = eb->pages[i];
5243 cur = min(len, (PAGE_CACHE_SIZE - offset));
5244 kaddr = page_address(page);
5245 if (copy_to_user(dst, kaddr + offset, cur)) {
5259 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
5260 unsigned long min_len, char **map,
5261 unsigned long *map_start,
5262 unsigned long *map_len)
5264 size_t offset = start & (PAGE_CACHE_SIZE - 1);
5267 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5268 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5269 unsigned long end_i = (start_offset + start + min_len - 1) >>
5276 offset = start_offset;
5280 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
5283 if (start + min_len > eb->len) {
5284 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
5286 eb->start, eb->len, start, min_len);
5291 kaddr = page_address(p);
5292 *map = kaddr + offset;
5293 *map_len = PAGE_CACHE_SIZE - offset;
5297 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5298 unsigned long start,
5305 char *ptr = (char *)ptrv;
5306 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5307 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5310 WARN_ON(start > eb->len);
5311 WARN_ON(start + len > eb->start + eb->len);
5313 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5316 page = eb->pages[i];
5318 cur = min(len, (PAGE_CACHE_SIZE - offset));
5320 kaddr = page_address(page);
5321 ret = memcmp(ptr, kaddr + offset, cur);
5333 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5334 unsigned long start, unsigned long len)
5340 char *src = (char *)srcv;
5341 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5342 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5344 WARN_ON(start > eb->len);
5345 WARN_ON(start + len > eb->start + eb->len);
5347 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5350 page = eb->pages[i];
5351 WARN_ON(!PageUptodate(page));
5353 cur = min(len, PAGE_CACHE_SIZE - offset);
5354 kaddr = page_address(page);
5355 memcpy(kaddr + offset, src, cur);
5364 void memset_extent_buffer(struct extent_buffer *eb, char c,
5365 unsigned long start, unsigned long len)
5371 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5372 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5374 WARN_ON(start > eb->len);
5375 WARN_ON(start + len > eb->start + eb->len);
5377 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5380 page = eb->pages[i];
5381 WARN_ON(!PageUptodate(page));
5383 cur = min(len, PAGE_CACHE_SIZE - offset);
5384 kaddr = page_address(page);
5385 memset(kaddr + offset, c, cur);
5393 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5394 unsigned long dst_offset, unsigned long src_offset,
5397 u64 dst_len = dst->len;
5402 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5403 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5405 WARN_ON(src->len != dst_len);
5407 offset = (start_offset + dst_offset) &
5408 (PAGE_CACHE_SIZE - 1);
5411 page = dst->pages[i];
5412 WARN_ON(!PageUptodate(page));
5414 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5416 kaddr = page_address(page);
5417 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5426 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5428 unsigned long distance = (src > dst) ? src - dst : dst - src;
5429 return distance < len;
5432 static void copy_pages(struct page *dst_page, struct page *src_page,
5433 unsigned long dst_off, unsigned long src_off,
5436 char *dst_kaddr = page_address(dst_page);
5438 int must_memmove = 0;
5440 if (dst_page != src_page) {
5441 src_kaddr = page_address(src_page);
5443 src_kaddr = dst_kaddr;
5444 if (areas_overlap(src_off, dst_off, len))
5449 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5451 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5454 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5455 unsigned long src_offset, unsigned long len)
5458 size_t dst_off_in_page;
5459 size_t src_off_in_page;
5460 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5461 unsigned long dst_i;
5462 unsigned long src_i;
5464 if (src_offset + len > dst->len) {
5465 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5466 "len %lu dst len %lu\n", src_offset, len, dst->len);
5469 if (dst_offset + len > dst->len) {
5470 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5471 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5476 dst_off_in_page = (start_offset + dst_offset) &
5477 (PAGE_CACHE_SIZE - 1);
5478 src_off_in_page = (start_offset + src_offset) &
5479 (PAGE_CACHE_SIZE - 1);
5481 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5482 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5484 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5486 cur = min_t(unsigned long, cur,
5487 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5489 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5490 dst_off_in_page, src_off_in_page, cur);
5498 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5499 unsigned long src_offset, unsigned long len)
5502 size_t dst_off_in_page;
5503 size_t src_off_in_page;
5504 unsigned long dst_end = dst_offset + len - 1;
5505 unsigned long src_end = src_offset + len - 1;
5506 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5507 unsigned long dst_i;
5508 unsigned long src_i;
5510 if (src_offset + len > dst->len) {
5511 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5512 "len %lu len %lu\n", src_offset, len, dst->len);
5515 if (dst_offset + len > dst->len) {
5516 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5517 "len %lu len %lu\n", dst_offset, len, dst->len);
5520 if (dst_offset < src_offset) {
5521 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5525 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5526 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5528 dst_off_in_page = (start_offset + dst_end) &
5529 (PAGE_CACHE_SIZE - 1);
5530 src_off_in_page = (start_offset + src_end) &
5531 (PAGE_CACHE_SIZE - 1);
5533 cur = min_t(unsigned long, len, src_off_in_page + 1);
5534 cur = min(cur, dst_off_in_page + 1);
5535 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5536 dst_off_in_page - cur + 1,
5537 src_off_in_page - cur + 1, cur);
5545 int try_release_extent_buffer(struct page *page)
5547 struct extent_buffer *eb;
5550 * We need to make sure noboody is attaching this page to an eb right
5553 spin_lock(&page->mapping->private_lock);
5554 if (!PagePrivate(page)) {
5555 spin_unlock(&page->mapping->private_lock);
5559 eb = (struct extent_buffer *)page->private;
5563 * This is a little awful but should be ok, we need to make sure that
5564 * the eb doesn't disappear out from under us while we're looking at
5567 spin_lock(&eb->refs_lock);
5568 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5569 spin_unlock(&eb->refs_lock);
5570 spin_unlock(&page->mapping->private_lock);
5573 spin_unlock(&page->mapping->private_lock);
5576 * If tree ref isn't set then we know the ref on this eb is a real ref,
5577 * so just return, this page will likely be freed soon anyway.
5579 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5580 spin_unlock(&eb->refs_lock);
5584 return release_extent_buffer(eb);