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 %u 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 *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 *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 *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 *bits, int wake)
516 struct extent_state *next;
517 unsigned 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 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 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 bits, unsigned 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 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 bits, unsigned 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 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 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 bits, struct extent_state **cached_state)
1339 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1340 EXTENT_LOCKED, &failed_start,
1341 cached_state, GFP_NOFS);
1342 if (err == -EEXIST) {
1343 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1344 start = failed_start;
1347 WARN_ON(start > end);
1352 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1354 return lock_extent_bits(tree, start, end, 0, NULL);
1357 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1362 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1363 &failed_start, NULL, GFP_NOFS);
1364 if (err == -EEXIST) {
1365 if (failed_start > start)
1366 clear_extent_bit(tree, start, failed_start - 1,
1367 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1373 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1374 struct extent_state **cached, gfp_t mask)
1376 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1380 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1382 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1386 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1388 unsigned long index = start >> PAGE_CACHE_SHIFT;
1389 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1392 while (index <= end_index) {
1393 page = find_get_page(inode->i_mapping, index);
1394 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1395 clear_page_dirty_for_io(page);
1396 page_cache_release(page);
1402 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1404 unsigned long index = start >> PAGE_CACHE_SHIFT;
1405 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1408 while (index <= end_index) {
1409 page = find_get_page(inode->i_mapping, index);
1410 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1411 account_page_redirty(page);
1412 __set_page_dirty_nobuffers(page);
1413 page_cache_release(page);
1420 * helper function to set both pages and extents in the tree writeback
1422 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1424 unsigned long index = start >> PAGE_CACHE_SHIFT;
1425 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1428 while (index <= end_index) {
1429 page = find_get_page(tree->mapping, index);
1430 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1431 set_page_writeback(page);
1432 page_cache_release(page);
1438 /* find the first state struct with 'bits' set after 'start', and
1439 * return it. tree->lock must be held. NULL will returned if
1440 * nothing was found after 'start'
1442 static struct extent_state *
1443 find_first_extent_bit_state(struct extent_io_tree *tree,
1444 u64 start, unsigned bits)
1446 struct rb_node *node;
1447 struct extent_state *state;
1450 * this search will find all the extents that end after
1453 node = tree_search(tree, start);
1458 state = rb_entry(node, struct extent_state, rb_node);
1459 if (state->end >= start && (state->state & bits))
1462 node = rb_next(node);
1471 * find the first offset in the io tree with 'bits' set. zero is
1472 * returned if we find something, and *start_ret and *end_ret are
1473 * set to reflect the state struct that was found.
1475 * If nothing was found, 1 is returned. If found something, return 0.
1477 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1478 u64 *start_ret, u64 *end_ret, unsigned bits,
1479 struct extent_state **cached_state)
1481 struct extent_state *state;
1485 spin_lock(&tree->lock);
1486 if (cached_state && *cached_state) {
1487 state = *cached_state;
1488 if (state->end == start - 1 && extent_state_in_tree(state)) {
1489 n = rb_next(&state->rb_node);
1491 state = rb_entry(n, struct extent_state,
1493 if (state->state & bits)
1497 free_extent_state(*cached_state);
1498 *cached_state = NULL;
1501 free_extent_state(*cached_state);
1502 *cached_state = NULL;
1505 state = find_first_extent_bit_state(tree, start, bits);
1508 cache_state_if_flags(state, cached_state, 0);
1509 *start_ret = state->start;
1510 *end_ret = state->end;
1514 spin_unlock(&tree->lock);
1519 * find a contiguous range of bytes in the file marked as delalloc, not
1520 * more than 'max_bytes'. start and end are used to return the range,
1522 * 1 is returned if we find something, 0 if nothing was in the tree
1524 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1525 u64 *start, u64 *end, u64 max_bytes,
1526 struct extent_state **cached_state)
1528 struct rb_node *node;
1529 struct extent_state *state;
1530 u64 cur_start = *start;
1532 u64 total_bytes = 0;
1534 spin_lock(&tree->lock);
1537 * this search will find all the extents that end after
1540 node = tree_search(tree, cur_start);
1548 state = rb_entry(node, struct extent_state, rb_node);
1549 if (found && (state->start != cur_start ||
1550 (state->state & EXTENT_BOUNDARY))) {
1553 if (!(state->state & EXTENT_DELALLOC)) {
1559 *start = state->start;
1560 *cached_state = state;
1561 atomic_inc(&state->refs);
1565 cur_start = state->end + 1;
1566 node = rb_next(node);
1567 total_bytes += state->end - state->start + 1;
1568 if (total_bytes >= max_bytes)
1574 spin_unlock(&tree->lock);
1578 static noinline void __unlock_for_delalloc(struct inode *inode,
1579 struct page *locked_page,
1583 struct page *pages[16];
1584 unsigned long index = start >> PAGE_CACHE_SHIFT;
1585 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1586 unsigned long nr_pages = end_index - index + 1;
1589 if (index == locked_page->index && end_index == index)
1592 while (nr_pages > 0) {
1593 ret = find_get_pages_contig(inode->i_mapping, index,
1594 min_t(unsigned long, nr_pages,
1595 ARRAY_SIZE(pages)), pages);
1596 for (i = 0; i < ret; i++) {
1597 if (pages[i] != locked_page)
1598 unlock_page(pages[i]);
1599 page_cache_release(pages[i]);
1607 static noinline int lock_delalloc_pages(struct inode *inode,
1608 struct page *locked_page,
1612 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1613 unsigned long start_index = index;
1614 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1615 unsigned long pages_locked = 0;
1616 struct page *pages[16];
1617 unsigned long nrpages;
1621 /* the caller is responsible for locking the start index */
1622 if (index == locked_page->index && index == end_index)
1625 /* skip the page at the start index */
1626 nrpages = end_index - index + 1;
1627 while (nrpages > 0) {
1628 ret = find_get_pages_contig(inode->i_mapping, index,
1629 min_t(unsigned long,
1630 nrpages, ARRAY_SIZE(pages)), pages);
1635 /* now we have an array of pages, lock them all */
1636 for (i = 0; i < ret; i++) {
1638 * the caller is taking responsibility for
1641 if (pages[i] != locked_page) {
1642 lock_page(pages[i]);
1643 if (!PageDirty(pages[i]) ||
1644 pages[i]->mapping != inode->i_mapping) {
1646 unlock_page(pages[i]);
1647 page_cache_release(pages[i]);
1651 page_cache_release(pages[i]);
1660 if (ret && pages_locked) {
1661 __unlock_for_delalloc(inode, locked_page,
1663 ((u64)(start_index + pages_locked - 1)) <<
1670 * find a contiguous range of bytes in the file marked as delalloc, not
1671 * more than 'max_bytes'. start and end are used to return the range,
1673 * 1 is returned if we find something, 0 if nothing was in the tree
1675 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1676 struct extent_io_tree *tree,
1677 struct page *locked_page, u64 *start,
1678 u64 *end, u64 max_bytes)
1683 struct extent_state *cached_state = NULL;
1688 /* step one, find a bunch of delalloc bytes starting at start */
1689 delalloc_start = *start;
1691 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1692 max_bytes, &cached_state);
1693 if (!found || delalloc_end <= *start) {
1694 *start = delalloc_start;
1695 *end = delalloc_end;
1696 free_extent_state(cached_state);
1701 * start comes from the offset of locked_page. We have to lock
1702 * pages in order, so we can't process delalloc bytes before
1705 if (delalloc_start < *start)
1706 delalloc_start = *start;
1709 * make sure to limit the number of pages we try to lock down
1711 if (delalloc_end + 1 - delalloc_start > max_bytes)
1712 delalloc_end = delalloc_start + max_bytes - 1;
1714 /* step two, lock all the pages after the page that has start */
1715 ret = lock_delalloc_pages(inode, locked_page,
1716 delalloc_start, delalloc_end);
1717 if (ret == -EAGAIN) {
1718 /* some of the pages are gone, lets avoid looping by
1719 * shortening the size of the delalloc range we're searching
1721 free_extent_state(cached_state);
1722 cached_state = NULL;
1724 max_bytes = PAGE_CACHE_SIZE;
1732 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1734 /* step three, lock the state bits for the whole range */
1735 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1737 /* then test to make sure it is all still delalloc */
1738 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1739 EXTENT_DELALLOC, 1, cached_state);
1741 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1742 &cached_state, GFP_NOFS);
1743 __unlock_for_delalloc(inode, locked_page,
1744 delalloc_start, delalloc_end);
1748 free_extent_state(cached_state);
1749 *start = delalloc_start;
1750 *end = delalloc_end;
1755 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1756 struct page *locked_page,
1757 unsigned clear_bits,
1758 unsigned long page_ops)
1760 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1762 struct page *pages[16];
1763 unsigned long index = start >> PAGE_CACHE_SHIFT;
1764 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1765 unsigned long nr_pages = end_index - index + 1;
1768 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1772 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1773 mapping_set_error(inode->i_mapping, -EIO);
1775 while (nr_pages > 0) {
1776 ret = find_get_pages_contig(inode->i_mapping, index,
1777 min_t(unsigned long,
1778 nr_pages, ARRAY_SIZE(pages)), pages);
1779 for (i = 0; i < ret; i++) {
1781 if (page_ops & PAGE_SET_PRIVATE2)
1782 SetPagePrivate2(pages[i]);
1784 if (pages[i] == locked_page) {
1785 page_cache_release(pages[i]);
1788 if (page_ops & PAGE_CLEAR_DIRTY)
1789 clear_page_dirty_for_io(pages[i]);
1790 if (page_ops & PAGE_SET_WRITEBACK)
1791 set_page_writeback(pages[i]);
1792 if (page_ops & PAGE_SET_ERROR)
1793 SetPageError(pages[i]);
1794 if (page_ops & PAGE_END_WRITEBACK)
1795 end_page_writeback(pages[i]);
1796 if (page_ops & PAGE_UNLOCK)
1797 unlock_page(pages[i]);
1798 page_cache_release(pages[i]);
1808 * count the number of bytes in the tree that have a given bit(s)
1809 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1810 * cached. The total number found is returned.
1812 u64 count_range_bits(struct extent_io_tree *tree,
1813 u64 *start, u64 search_end, u64 max_bytes,
1814 unsigned bits, int contig)
1816 struct rb_node *node;
1817 struct extent_state *state;
1818 u64 cur_start = *start;
1819 u64 total_bytes = 0;
1823 if (WARN_ON(search_end <= cur_start))
1826 spin_lock(&tree->lock);
1827 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1828 total_bytes = tree->dirty_bytes;
1832 * this search will find all the extents that end after
1835 node = tree_search(tree, cur_start);
1840 state = rb_entry(node, struct extent_state, rb_node);
1841 if (state->start > search_end)
1843 if (contig && found && state->start > last + 1)
1845 if (state->end >= cur_start && (state->state & bits) == bits) {
1846 total_bytes += min(search_end, state->end) + 1 -
1847 max(cur_start, state->start);
1848 if (total_bytes >= max_bytes)
1851 *start = max(cur_start, state->start);
1855 } else if (contig && found) {
1858 node = rb_next(node);
1863 spin_unlock(&tree->lock);
1868 * set the private field for a given byte offset in the tree. If there isn't
1869 * an extent_state there already, this does nothing.
1871 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1873 struct rb_node *node;
1874 struct extent_state *state;
1877 spin_lock(&tree->lock);
1879 * this search will find all the extents that end after
1882 node = tree_search(tree, start);
1887 state = rb_entry(node, struct extent_state, rb_node);
1888 if (state->start != start) {
1892 state->private = private;
1894 spin_unlock(&tree->lock);
1898 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1900 struct rb_node *node;
1901 struct extent_state *state;
1904 spin_lock(&tree->lock);
1906 * this search will find all the extents that end after
1909 node = tree_search(tree, start);
1914 state = rb_entry(node, struct extent_state, rb_node);
1915 if (state->start != start) {
1919 *private = state->private;
1921 spin_unlock(&tree->lock);
1926 * searches a range in the state tree for a given mask.
1927 * If 'filled' == 1, this returns 1 only if every extent in the tree
1928 * has the bits set. Otherwise, 1 is returned if any bit in the
1929 * range is found set.
1931 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1932 unsigned bits, int filled, struct extent_state *cached)
1934 struct extent_state *state = NULL;
1935 struct rb_node *node;
1938 spin_lock(&tree->lock);
1939 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1940 cached->end > start)
1941 node = &cached->rb_node;
1943 node = tree_search(tree, start);
1944 while (node && start <= end) {
1945 state = rb_entry(node, struct extent_state, rb_node);
1947 if (filled && state->start > start) {
1952 if (state->start > end)
1955 if (state->state & bits) {
1959 } else if (filled) {
1964 if (state->end == (u64)-1)
1967 start = state->end + 1;
1970 node = rb_next(node);
1977 spin_unlock(&tree->lock);
1982 * helper function to set a given page up to date if all the
1983 * extents in the tree for that page are up to date
1985 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1987 u64 start = page_offset(page);
1988 u64 end = start + PAGE_CACHE_SIZE - 1;
1989 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1990 SetPageUptodate(page);
1993 int free_io_failure(struct inode *inode, struct io_failure_record *rec)
1997 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1999 set_state_private(failure_tree, rec->start, 0);
2000 ret = clear_extent_bits(failure_tree, rec->start,
2001 rec->start + rec->len - 1,
2002 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2006 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
2007 rec->start + rec->len - 1,
2008 EXTENT_DAMAGED, GFP_NOFS);
2017 * this bypasses the standard btrfs submit functions deliberately, as
2018 * the standard behavior is to write all copies in a raid setup. here we only
2019 * want to write the one bad copy. so we do the mapping for ourselves and issue
2020 * submit_bio directly.
2021 * to avoid any synchronization issues, wait for the data after writing, which
2022 * actually prevents the read that triggered the error from finishing.
2023 * currently, there can be no more than two copies of every data bit. thus,
2024 * exactly one rewrite is required.
2026 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
2027 struct page *page, unsigned int pg_offset, int mirror_num)
2029 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2031 struct btrfs_device *dev;
2034 struct btrfs_bio *bbio = NULL;
2035 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2038 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2039 BUG_ON(!mirror_num);
2041 /* we can't repair anything in raid56 yet */
2042 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2045 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2048 bio->bi_iter.bi_size = 0;
2049 map_length = length;
2051 ret = btrfs_map_block(fs_info, WRITE, logical,
2052 &map_length, &bbio, mirror_num);
2057 BUG_ON(mirror_num != bbio->mirror_num);
2058 sector = bbio->stripes[mirror_num-1].physical >> 9;
2059 bio->bi_iter.bi_sector = sector;
2060 dev = bbio->stripes[mirror_num-1].dev;
2062 if (!dev || !dev->bdev || !dev->writeable) {
2066 bio->bi_bdev = dev->bdev;
2067 bio_add_page(bio, page, length, pg_offset);
2069 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
2070 /* try to remap that extent elsewhere? */
2072 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2076 printk_ratelimited_in_rcu(KERN_INFO
2077 "BTRFS: read error corrected: ino %llu off %llu (dev %s sector %llu)\n",
2078 btrfs_ino(inode), start,
2079 rcu_str_deref(dev->name), sector);
2084 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2087 u64 start = eb->start;
2088 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2091 if (root->fs_info->sb->s_flags & MS_RDONLY)
2094 for (i = 0; i < num_pages; i++) {
2095 struct page *p = eb->pages[i];
2097 ret = repair_io_failure(root->fs_info->btree_inode, start,
2098 PAGE_CACHE_SIZE, start, p,
2099 start - page_offset(p), mirror_num);
2102 start += PAGE_CACHE_SIZE;
2109 * each time an IO finishes, we do a fast check in the IO failure tree
2110 * to see if we need to process or clean up an io_failure_record
2112 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
2113 unsigned int pg_offset)
2116 u64 private_failure;
2117 struct io_failure_record *failrec;
2118 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2119 struct extent_state *state;
2124 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2125 (u64)-1, 1, EXTENT_DIRTY, 0);
2129 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2134 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2135 BUG_ON(!failrec->this_mirror);
2137 if (failrec->in_validation) {
2138 /* there was no real error, just free the record */
2139 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2143 if (fs_info->sb->s_flags & MS_RDONLY)
2146 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2147 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2150 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2152 if (state && state->start <= failrec->start &&
2153 state->end >= failrec->start + failrec->len - 1) {
2154 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2156 if (num_copies > 1) {
2157 repair_io_failure(inode, start, failrec->len,
2158 failrec->logical, page,
2159 pg_offset, failrec->failed_mirror);
2164 free_io_failure(inode, failrec);
2170 * Can be called when
2171 * - hold extent lock
2172 * - under ordered extent
2173 * - the inode is freeing
2175 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end)
2177 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2178 struct io_failure_record *failrec;
2179 struct extent_state *state, *next;
2181 if (RB_EMPTY_ROOT(&failure_tree->state))
2184 spin_lock(&failure_tree->lock);
2185 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2187 if (state->start > end)
2190 ASSERT(state->end <= end);
2192 next = next_state(state);
2194 failrec = (struct io_failure_record *)state->private;
2195 free_extent_state(state);
2200 spin_unlock(&failure_tree->lock);
2203 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2204 struct io_failure_record **failrec_ret)
2206 struct io_failure_record *failrec;
2208 struct extent_map *em;
2209 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2210 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2211 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2215 ret = get_state_private(failure_tree, start, &private);
2217 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2221 failrec->start = start;
2222 failrec->len = end - start + 1;
2223 failrec->this_mirror = 0;
2224 failrec->bio_flags = 0;
2225 failrec->in_validation = 0;
2227 read_lock(&em_tree->lock);
2228 em = lookup_extent_mapping(em_tree, start, failrec->len);
2230 read_unlock(&em_tree->lock);
2235 if (em->start > start || em->start + em->len <= start) {
2236 free_extent_map(em);
2239 read_unlock(&em_tree->lock);
2245 logical = start - em->start;
2246 logical = em->block_start + logical;
2247 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2248 logical = em->block_start;
2249 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2250 extent_set_compress_type(&failrec->bio_flags,
2254 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2255 logical, start, failrec->len);
2257 failrec->logical = logical;
2258 free_extent_map(em);
2260 /* set the bits in the private failure tree */
2261 ret = set_extent_bits(failure_tree, start, end,
2262 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2264 ret = set_state_private(failure_tree, start,
2265 (u64)(unsigned long)failrec);
2266 /* set the bits in the inode's tree */
2268 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2275 failrec = (struct io_failure_record *)(unsigned long)private;
2276 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2277 failrec->logical, failrec->start, failrec->len,
2278 failrec->in_validation);
2280 * when data can be on disk more than twice, add to failrec here
2281 * (e.g. with a list for failed_mirror) to make
2282 * clean_io_failure() clean all those errors at once.
2286 *failrec_ret = failrec;
2291 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2292 struct io_failure_record *failrec, int failed_mirror)
2296 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2297 failrec->logical, failrec->len);
2298 if (num_copies == 1) {
2300 * we only have a single copy of the data, so don't bother with
2301 * all the retry and error correction code that follows. no
2302 * matter what the error is, it is very likely to persist.
2304 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2305 num_copies, failrec->this_mirror, failed_mirror);
2310 * there are two premises:
2311 * a) deliver good data to the caller
2312 * b) correct the bad sectors on disk
2314 if (failed_bio->bi_vcnt > 1) {
2316 * to fulfill b), we need to know the exact failing sectors, as
2317 * we don't want to rewrite any more than the failed ones. thus,
2318 * we need separate read requests for the failed bio
2320 * if the following BUG_ON triggers, our validation request got
2321 * merged. we need separate requests for our algorithm to work.
2323 BUG_ON(failrec->in_validation);
2324 failrec->in_validation = 1;
2325 failrec->this_mirror = failed_mirror;
2328 * we're ready to fulfill a) and b) alongside. get a good copy
2329 * of the failed sector and if we succeed, we have setup
2330 * everything for repair_io_failure to do the rest for us.
2332 if (failrec->in_validation) {
2333 BUG_ON(failrec->this_mirror != failed_mirror);
2334 failrec->in_validation = 0;
2335 failrec->this_mirror = 0;
2337 failrec->failed_mirror = failed_mirror;
2338 failrec->this_mirror++;
2339 if (failrec->this_mirror == failed_mirror)
2340 failrec->this_mirror++;
2343 if (failrec->this_mirror > num_copies) {
2344 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2345 num_copies, failrec->this_mirror, failed_mirror);
2353 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2354 struct io_failure_record *failrec,
2355 struct page *page, int pg_offset, int icsum,
2356 bio_end_io_t *endio_func, void *data)
2359 struct btrfs_io_bio *btrfs_failed_bio;
2360 struct btrfs_io_bio *btrfs_bio;
2362 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2366 bio->bi_end_io = endio_func;
2367 bio->bi_iter.bi_sector = failrec->logical >> 9;
2368 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2369 bio->bi_iter.bi_size = 0;
2370 bio->bi_private = data;
2372 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2373 if (btrfs_failed_bio->csum) {
2374 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2375 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2377 btrfs_bio = btrfs_io_bio(bio);
2378 btrfs_bio->csum = btrfs_bio->csum_inline;
2380 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2384 bio_add_page(bio, page, failrec->len, pg_offset);
2390 * this is a generic handler for readpage errors (default
2391 * readpage_io_failed_hook). if other copies exist, read those and write back
2392 * good data to the failed position. does not investigate in remapping the
2393 * failed extent elsewhere, hoping the device will be smart enough to do this as
2397 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2398 struct page *page, u64 start, u64 end,
2401 struct io_failure_record *failrec;
2402 struct inode *inode = page->mapping->host;
2403 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2408 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2410 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2414 ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror);
2416 free_io_failure(inode, failrec);
2420 if (failed_bio->bi_vcnt > 1)
2421 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2423 read_mode = READ_SYNC;
2425 phy_offset >>= inode->i_sb->s_blocksize_bits;
2426 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2427 start - page_offset(page),
2428 (int)phy_offset, failed_bio->bi_end_io,
2431 free_io_failure(inode, failrec);
2435 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2436 read_mode, failrec->this_mirror, failrec->in_validation);
2438 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2439 failrec->this_mirror,
2440 failrec->bio_flags, 0);
2442 free_io_failure(inode, failrec);
2449 /* lots and lots of room for performance fixes in the end_bio funcs */
2451 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2453 int uptodate = (err == 0);
2454 struct extent_io_tree *tree;
2457 tree = &BTRFS_I(page->mapping->host)->io_tree;
2459 if (tree->ops && tree->ops->writepage_end_io_hook) {
2460 ret = tree->ops->writepage_end_io_hook(page, start,
2461 end, NULL, uptodate);
2467 ClearPageUptodate(page);
2469 ret = ret < 0 ? ret : -EIO;
2470 mapping_set_error(page->mapping, ret);
2476 * after a writepage IO is done, we need to:
2477 * clear the uptodate bits on error
2478 * clear the writeback bits in the extent tree for this IO
2479 * end_page_writeback if the page has no more pending IO
2481 * Scheduling is not allowed, so the extent state tree is expected
2482 * to have one and only one object corresponding to this IO.
2484 static void end_bio_extent_writepage(struct bio *bio, int err)
2486 struct bio_vec *bvec;
2491 bio_for_each_segment_all(bvec, bio, i) {
2492 struct page *page = bvec->bv_page;
2494 /* We always issue full-page reads, but if some block
2495 * in a page fails to read, blk_update_request() will
2496 * advance bv_offset and adjust bv_len to compensate.
2497 * Print a warning for nonzero offsets, and an error
2498 * if they don't add up to a full page. */
2499 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2500 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2501 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2502 "partial page write in btrfs with offset %u and length %u",
2503 bvec->bv_offset, bvec->bv_len);
2505 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2506 "incomplete page write in btrfs with offset %u and "
2508 bvec->bv_offset, bvec->bv_len);
2511 start = page_offset(page);
2512 end = start + bvec->bv_offset + bvec->bv_len - 1;
2514 if (end_extent_writepage(page, err, start, end))
2517 end_page_writeback(page);
2524 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2527 struct extent_state *cached = NULL;
2528 u64 end = start + len - 1;
2530 if (uptodate && tree->track_uptodate)
2531 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2532 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2536 * after a readpage IO is done, we need to:
2537 * clear the uptodate bits on error
2538 * set the uptodate bits if things worked
2539 * set the page up to date if all extents in the tree are uptodate
2540 * clear the lock bit in the extent tree
2541 * unlock the page if there are no other extents locked for it
2543 * Scheduling is not allowed, so the extent state tree is expected
2544 * to have one and only one object corresponding to this IO.
2546 static void end_bio_extent_readpage(struct bio *bio, int err)
2548 struct bio_vec *bvec;
2549 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2550 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2551 struct extent_io_tree *tree;
2556 u64 extent_start = 0;
2565 bio_for_each_segment_all(bvec, bio, i) {
2566 struct page *page = bvec->bv_page;
2567 struct inode *inode = page->mapping->host;
2569 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2570 "mirror=%u\n", (u64)bio->bi_iter.bi_sector, err,
2571 io_bio->mirror_num);
2572 tree = &BTRFS_I(inode)->io_tree;
2574 /* We always issue full-page reads, but if some block
2575 * in a page fails to read, blk_update_request() will
2576 * advance bv_offset and adjust bv_len to compensate.
2577 * Print a warning for nonzero offsets, and an error
2578 * if they don't add up to a full page. */
2579 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2580 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2581 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2582 "partial page read in btrfs with offset %u and length %u",
2583 bvec->bv_offset, bvec->bv_len);
2585 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2586 "incomplete page read in btrfs with offset %u and "
2588 bvec->bv_offset, bvec->bv_len);
2591 start = page_offset(page);
2592 end = start + bvec->bv_offset + bvec->bv_len - 1;
2595 mirror = io_bio->mirror_num;
2596 if (likely(uptodate && tree->ops &&
2597 tree->ops->readpage_end_io_hook)) {
2598 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2604 clean_io_failure(inode, start, page, 0);
2607 if (likely(uptodate))
2610 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2611 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2613 test_bit(BIO_UPTODATE, &bio->bi_flags))
2617 * The generic bio_readpage_error handles errors the
2618 * following way: If possible, new read requests are
2619 * created and submitted and will end up in
2620 * end_bio_extent_readpage as well (if we're lucky, not
2621 * in the !uptodate case). In that case it returns 0 and
2622 * we just go on with the next page in our bio. If it
2623 * can't handle the error it will return -EIO and we
2624 * remain responsible for that page.
2626 ret = bio_readpage_error(bio, offset, page, start, end,
2630 test_bit(BIO_UPTODATE, &bio->bi_flags);
2638 if (likely(uptodate)) {
2639 loff_t i_size = i_size_read(inode);
2640 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2643 /* Zero out the end if this page straddles i_size */
2644 off = i_size & (PAGE_CACHE_SIZE-1);
2645 if (page->index == end_index && off)
2646 zero_user_segment(page, off, PAGE_CACHE_SIZE);
2647 SetPageUptodate(page);
2649 ClearPageUptodate(page);
2655 if (unlikely(!uptodate)) {
2657 endio_readpage_release_extent(tree,
2663 endio_readpage_release_extent(tree, start,
2664 end - start + 1, 0);
2665 } else if (!extent_len) {
2666 extent_start = start;
2667 extent_len = end + 1 - start;
2668 } else if (extent_start + extent_len == start) {
2669 extent_len += end + 1 - start;
2671 endio_readpage_release_extent(tree, extent_start,
2672 extent_len, uptodate);
2673 extent_start = start;
2674 extent_len = end + 1 - start;
2679 endio_readpage_release_extent(tree, extent_start, extent_len,
2682 io_bio->end_io(io_bio, err);
2687 * this allocates from the btrfs_bioset. We're returning a bio right now
2688 * but you can call btrfs_io_bio for the appropriate container_of magic
2691 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2694 struct btrfs_io_bio *btrfs_bio;
2697 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2699 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2700 while (!bio && (nr_vecs /= 2)) {
2701 bio = bio_alloc_bioset(gfp_flags,
2702 nr_vecs, btrfs_bioset);
2707 bio->bi_bdev = bdev;
2708 bio->bi_iter.bi_sector = first_sector;
2709 btrfs_bio = btrfs_io_bio(bio);
2710 btrfs_bio->csum = NULL;
2711 btrfs_bio->csum_allocated = NULL;
2712 btrfs_bio->end_io = NULL;
2717 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2719 struct btrfs_io_bio *btrfs_bio;
2722 new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2724 btrfs_bio = btrfs_io_bio(new);
2725 btrfs_bio->csum = NULL;
2726 btrfs_bio->csum_allocated = NULL;
2727 btrfs_bio->end_io = NULL;
2732 /* this also allocates from the btrfs_bioset */
2733 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2735 struct btrfs_io_bio *btrfs_bio;
2738 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2740 btrfs_bio = btrfs_io_bio(bio);
2741 btrfs_bio->csum = NULL;
2742 btrfs_bio->csum_allocated = NULL;
2743 btrfs_bio->end_io = NULL;
2749 static int __must_check submit_one_bio(int rw, struct bio *bio,
2750 int mirror_num, unsigned long bio_flags)
2753 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2754 struct page *page = bvec->bv_page;
2755 struct extent_io_tree *tree = bio->bi_private;
2758 start = page_offset(page) + bvec->bv_offset;
2760 bio->bi_private = NULL;
2764 if (tree->ops && tree->ops->submit_bio_hook)
2765 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2766 mirror_num, bio_flags, start);
2768 btrfsic_submit_bio(rw, bio);
2770 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2776 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2777 unsigned long offset, size_t size, struct bio *bio,
2778 unsigned long bio_flags)
2781 if (tree->ops && tree->ops->merge_bio_hook)
2782 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2789 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2790 struct page *page, sector_t sector,
2791 size_t size, unsigned long offset,
2792 struct block_device *bdev,
2793 struct bio **bio_ret,
2794 unsigned long max_pages,
2795 bio_end_io_t end_io_func,
2797 unsigned long prev_bio_flags,
2798 unsigned long bio_flags)
2804 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2805 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2806 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2808 if (bio_ret && *bio_ret) {
2811 contig = bio->bi_iter.bi_sector == sector;
2813 contig = bio_end_sector(bio) == sector;
2815 if (prev_bio_flags != bio_flags || !contig ||
2816 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2817 bio_add_page(bio, page, page_size, offset) < page_size) {
2818 ret = submit_one_bio(rw, bio, mirror_num,
2827 if (this_compressed)
2830 nr = bio_get_nr_vecs(bdev);
2832 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2836 bio_add_page(bio, page, page_size, offset);
2837 bio->bi_end_io = end_io_func;
2838 bio->bi_private = tree;
2843 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2848 static void attach_extent_buffer_page(struct extent_buffer *eb,
2851 if (!PagePrivate(page)) {
2852 SetPagePrivate(page);
2853 page_cache_get(page);
2854 set_page_private(page, (unsigned long)eb);
2856 WARN_ON(page->private != (unsigned long)eb);
2860 void set_page_extent_mapped(struct page *page)
2862 if (!PagePrivate(page)) {
2863 SetPagePrivate(page);
2864 page_cache_get(page);
2865 set_page_private(page, EXTENT_PAGE_PRIVATE);
2869 static struct extent_map *
2870 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2871 u64 start, u64 len, get_extent_t *get_extent,
2872 struct extent_map **em_cached)
2874 struct extent_map *em;
2876 if (em_cached && *em_cached) {
2878 if (extent_map_in_tree(em) && start >= em->start &&
2879 start < extent_map_end(em)) {
2880 atomic_inc(&em->refs);
2884 free_extent_map(em);
2888 em = get_extent(inode, page, pg_offset, start, len, 0);
2889 if (em_cached && !IS_ERR_OR_NULL(em)) {
2891 atomic_inc(&em->refs);
2897 * basic readpage implementation. Locked extent state structs are inserted
2898 * into the tree that are removed when the IO is done (by the end_io
2900 * XXX JDM: This needs looking at to ensure proper page locking
2902 static int __do_readpage(struct extent_io_tree *tree,
2904 get_extent_t *get_extent,
2905 struct extent_map **em_cached,
2906 struct bio **bio, int mirror_num,
2907 unsigned long *bio_flags, int rw)
2909 struct inode *inode = page->mapping->host;
2910 u64 start = page_offset(page);
2911 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2915 u64 last_byte = i_size_read(inode);
2919 struct extent_map *em;
2920 struct block_device *bdev;
2923 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2924 size_t pg_offset = 0;
2926 size_t disk_io_size;
2927 size_t blocksize = inode->i_sb->s_blocksize;
2928 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2930 set_page_extent_mapped(page);
2933 if (!PageUptodate(page)) {
2934 if (cleancache_get_page(page) == 0) {
2935 BUG_ON(blocksize != PAGE_SIZE);
2936 unlock_extent(tree, start, end);
2941 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2943 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2946 iosize = PAGE_CACHE_SIZE - zero_offset;
2947 userpage = kmap_atomic(page);
2948 memset(userpage + zero_offset, 0, iosize);
2949 flush_dcache_page(page);
2950 kunmap_atomic(userpage);
2953 while (cur <= end) {
2954 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2956 if (cur >= last_byte) {
2958 struct extent_state *cached = NULL;
2960 iosize = PAGE_CACHE_SIZE - pg_offset;
2961 userpage = kmap_atomic(page);
2962 memset(userpage + pg_offset, 0, iosize);
2963 flush_dcache_page(page);
2964 kunmap_atomic(userpage);
2965 set_extent_uptodate(tree, cur, cur + iosize - 1,
2968 unlock_extent_cached(tree, cur,
2973 em = __get_extent_map(inode, page, pg_offset, cur,
2974 end - cur + 1, get_extent, em_cached);
2975 if (IS_ERR_OR_NULL(em)) {
2978 unlock_extent(tree, cur, end);
2981 extent_offset = cur - em->start;
2982 BUG_ON(extent_map_end(em) <= cur);
2985 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2986 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2987 extent_set_compress_type(&this_bio_flag,
2991 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2992 cur_end = min(extent_map_end(em) - 1, end);
2993 iosize = ALIGN(iosize, blocksize);
2994 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2995 disk_io_size = em->block_len;
2996 sector = em->block_start >> 9;
2998 sector = (em->block_start + extent_offset) >> 9;
2999 disk_io_size = iosize;
3002 block_start = em->block_start;
3003 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3004 block_start = EXTENT_MAP_HOLE;
3005 free_extent_map(em);
3008 /* we've found a hole, just zero and go on */
3009 if (block_start == EXTENT_MAP_HOLE) {
3011 struct extent_state *cached = NULL;
3013 userpage = kmap_atomic(page);
3014 memset(userpage + pg_offset, 0, iosize);
3015 flush_dcache_page(page);
3016 kunmap_atomic(userpage);
3018 set_extent_uptodate(tree, cur, cur + iosize - 1,
3020 unlock_extent_cached(tree, cur, cur + iosize - 1,
3023 pg_offset += iosize;
3026 /* the get_extent function already copied into the page */
3027 if (test_range_bit(tree, cur, cur_end,
3028 EXTENT_UPTODATE, 1, NULL)) {
3029 check_page_uptodate(tree, page);
3031 unlock_extent(tree, cur, cur + iosize - 1);
3033 pg_offset += iosize;
3036 /* we have an inline extent but it didn't get marked up
3037 * to date. Error out
3039 if (block_start == EXTENT_MAP_INLINE) {
3042 unlock_extent(tree, cur, cur + iosize - 1);
3044 pg_offset += iosize;
3049 ret = submit_extent_page(rw, tree, page,
3050 sector, disk_io_size, pg_offset,
3052 end_bio_extent_readpage, mirror_num,
3057 *bio_flags = this_bio_flag;
3061 unlock_extent(tree, cur, cur + iosize - 1);
3064 pg_offset += iosize;
3068 if (!PageError(page))
3069 SetPageUptodate(page);
3075 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3076 struct page *pages[], int nr_pages,
3078 get_extent_t *get_extent,
3079 struct extent_map **em_cached,
3080 struct bio **bio, int mirror_num,
3081 unsigned long *bio_flags, int rw)
3083 struct inode *inode;
3084 struct btrfs_ordered_extent *ordered;
3087 inode = pages[0]->mapping->host;
3089 lock_extent(tree, start, end);
3090 ordered = btrfs_lookup_ordered_range(inode, start,
3094 unlock_extent(tree, start, end);
3095 btrfs_start_ordered_extent(inode, ordered, 1);
3096 btrfs_put_ordered_extent(ordered);
3099 for (index = 0; index < nr_pages; index++) {
3100 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3101 mirror_num, bio_flags, rw);
3102 page_cache_release(pages[index]);
3106 static void __extent_readpages(struct extent_io_tree *tree,
3107 struct page *pages[],
3108 int nr_pages, get_extent_t *get_extent,
3109 struct extent_map **em_cached,
3110 struct bio **bio, int mirror_num,
3111 unsigned long *bio_flags, int rw)
3117 int first_index = 0;
3119 for (index = 0; index < nr_pages; index++) {
3120 page_start = page_offset(pages[index]);
3123 end = start + PAGE_CACHE_SIZE - 1;
3124 first_index = index;
3125 } else if (end + 1 == page_start) {
3126 end += PAGE_CACHE_SIZE;
3128 __do_contiguous_readpages(tree, &pages[first_index],
3129 index - first_index, start,
3130 end, get_extent, em_cached,
3131 bio, mirror_num, bio_flags,
3134 end = start + PAGE_CACHE_SIZE - 1;
3135 first_index = index;
3140 __do_contiguous_readpages(tree, &pages[first_index],
3141 index - first_index, start,
3142 end, get_extent, em_cached, bio,
3143 mirror_num, bio_flags, rw);
3146 static int __extent_read_full_page(struct extent_io_tree *tree,
3148 get_extent_t *get_extent,
3149 struct bio **bio, int mirror_num,
3150 unsigned long *bio_flags, int rw)
3152 struct inode *inode = page->mapping->host;
3153 struct btrfs_ordered_extent *ordered;
3154 u64 start = page_offset(page);
3155 u64 end = start + PAGE_CACHE_SIZE - 1;
3159 lock_extent(tree, start, end);
3160 ordered = btrfs_lookup_ordered_extent(inode, start);
3163 unlock_extent(tree, start, end);
3164 btrfs_start_ordered_extent(inode, ordered, 1);
3165 btrfs_put_ordered_extent(ordered);
3168 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3173 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3174 get_extent_t *get_extent, int mirror_num)
3176 struct bio *bio = NULL;
3177 unsigned long bio_flags = 0;
3180 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3183 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3187 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3188 get_extent_t *get_extent, int mirror_num)
3190 struct bio *bio = NULL;
3191 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3194 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3197 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3201 static noinline void update_nr_written(struct page *page,
3202 struct writeback_control *wbc,
3203 unsigned long nr_written)
3205 wbc->nr_to_write -= nr_written;
3206 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3207 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3208 page->mapping->writeback_index = page->index + nr_written;
3212 * helper for __extent_writepage, doing all of the delayed allocation setup.
3214 * This returns 1 if our fill_delalloc function did all the work required
3215 * to write the page (copy into inline extent). In this case the IO has
3216 * been started and the page is already unlocked.
3218 * This returns 0 if all went well (page still locked)
3219 * This returns < 0 if there were errors (page still locked)
3221 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3222 struct page *page, struct writeback_control *wbc,
3223 struct extent_page_data *epd,
3225 unsigned long *nr_written)
3227 struct extent_io_tree *tree = epd->tree;
3228 u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1;
3230 u64 delalloc_to_write = 0;
3231 u64 delalloc_end = 0;
3233 int page_started = 0;
3235 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3238 while (delalloc_end < page_end) {
3239 nr_delalloc = find_lock_delalloc_range(inode, tree,
3244 if (nr_delalloc == 0) {
3245 delalloc_start = delalloc_end + 1;
3248 ret = tree->ops->fill_delalloc(inode, page,
3253 /* File system has been set read-only */
3256 /* fill_delalloc should be return < 0 for error
3257 * but just in case, we use > 0 here meaning the
3258 * IO is started, so we don't want to return > 0
3259 * unless things are going well.
3261 ret = ret < 0 ? ret : -EIO;
3265 * delalloc_end is already one less than the total
3266 * length, so we don't subtract one from
3269 delalloc_to_write += (delalloc_end - delalloc_start +
3272 delalloc_start = delalloc_end + 1;
3274 if (wbc->nr_to_write < delalloc_to_write) {
3277 if (delalloc_to_write < thresh * 2)
3278 thresh = delalloc_to_write;
3279 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3283 /* did the fill delalloc function already unlock and start
3288 * we've unlocked the page, so we can't update
3289 * the mapping's writeback index, just update
3292 wbc->nr_to_write -= *nr_written;
3303 * helper for __extent_writepage. This calls the writepage start hooks,
3304 * and does the loop to map the page into extents and bios.
3306 * We return 1 if the IO is started and the page is unlocked,
3307 * 0 if all went well (page still locked)
3308 * < 0 if there were errors (page still locked)
3310 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3312 struct writeback_control *wbc,
3313 struct extent_page_data *epd,
3315 unsigned long nr_written,
3316 int write_flags, int *nr_ret)
3318 struct extent_io_tree *tree = epd->tree;
3319 u64 start = page_offset(page);
3320 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3327 struct extent_state *cached_state = NULL;
3328 struct extent_map *em;
3329 struct block_device *bdev;
3330 size_t pg_offset = 0;
3336 if (tree->ops && tree->ops->writepage_start_hook) {
3337 ret = tree->ops->writepage_start_hook(page, start,
3340 /* Fixup worker will requeue */
3342 wbc->pages_skipped++;
3344 redirty_page_for_writepage(wbc, page);
3346 update_nr_written(page, wbc, nr_written);
3354 * we don't want to touch the inode after unlocking the page,
3355 * so we update the mapping writeback index now
3357 update_nr_written(page, wbc, nr_written + 1);
3360 if (i_size <= start) {
3361 if (tree->ops && tree->ops->writepage_end_io_hook)
3362 tree->ops->writepage_end_io_hook(page, start,
3367 blocksize = inode->i_sb->s_blocksize;
3369 while (cur <= end) {
3371 if (cur >= i_size) {
3372 if (tree->ops && tree->ops->writepage_end_io_hook)
3373 tree->ops->writepage_end_io_hook(page, cur,
3377 em = epd->get_extent(inode, page, pg_offset, cur,
3379 if (IS_ERR_OR_NULL(em)) {
3381 ret = PTR_ERR_OR_ZERO(em);
3385 extent_offset = cur - em->start;
3386 em_end = extent_map_end(em);
3387 BUG_ON(em_end <= cur);
3389 iosize = min(em_end - cur, end - cur + 1);
3390 iosize = ALIGN(iosize, blocksize);
3391 sector = (em->block_start + extent_offset) >> 9;
3393 block_start = em->block_start;
3394 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3395 free_extent_map(em);
3399 * compressed and inline extents are written through other
3402 if (compressed || block_start == EXTENT_MAP_HOLE ||
3403 block_start == EXTENT_MAP_INLINE) {
3405 * end_io notification does not happen here for
3406 * compressed extents
3408 if (!compressed && tree->ops &&
3409 tree->ops->writepage_end_io_hook)
3410 tree->ops->writepage_end_io_hook(page, cur,
3413 else if (compressed) {
3414 /* we don't want to end_page_writeback on
3415 * a compressed extent. this happens
3422 pg_offset += iosize;
3426 if (tree->ops && tree->ops->writepage_io_hook) {
3427 ret = tree->ops->writepage_io_hook(page, cur,
3435 unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1;
3437 set_range_writeback(tree, cur, cur + iosize - 1);
3438 if (!PageWriteback(page)) {
3439 btrfs_err(BTRFS_I(inode)->root->fs_info,
3440 "page %lu not writeback, cur %llu end %llu",
3441 page->index, cur, end);
3444 ret = submit_extent_page(write_flags, tree, page,
3445 sector, iosize, pg_offset,
3446 bdev, &epd->bio, max_nr,
3447 end_bio_extent_writepage,
3453 pg_offset += iosize;
3461 /* drop our reference on any cached states */
3462 free_extent_state(cached_state);
3467 * the writepage semantics are similar to regular writepage. extent
3468 * records are inserted to lock ranges in the tree, and as dirty areas
3469 * are found, they are marked writeback. Then the lock bits are removed
3470 * and the end_io handler clears the writeback ranges
3472 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3475 struct inode *inode = page->mapping->host;
3476 struct extent_page_data *epd = data;
3477 u64 start = page_offset(page);
3478 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3481 size_t pg_offset = 0;
3482 loff_t i_size = i_size_read(inode);
3483 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3485 unsigned long nr_written = 0;
3487 if (wbc->sync_mode == WB_SYNC_ALL)
3488 write_flags = WRITE_SYNC;
3490 write_flags = WRITE;
3492 trace___extent_writepage(page, inode, wbc);
3494 WARN_ON(!PageLocked(page));
3496 ClearPageError(page);
3498 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3499 if (page->index > end_index ||
3500 (page->index == end_index && !pg_offset)) {
3501 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3506 if (page->index == end_index) {
3509 userpage = kmap_atomic(page);
3510 memset(userpage + pg_offset, 0,
3511 PAGE_CACHE_SIZE - pg_offset);
3512 kunmap_atomic(userpage);
3513 flush_dcache_page(page);
3518 set_page_extent_mapped(page);
3520 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3526 ret = __extent_writepage_io(inode, page, wbc, epd,
3527 i_size, nr_written, write_flags, &nr);
3533 /* make sure the mapping tag for page dirty gets cleared */
3534 set_page_writeback(page);
3535 end_page_writeback(page);
3537 if (PageError(page)) {
3538 ret = ret < 0 ? ret : -EIO;
3539 end_extent_writepage(page, ret, start, page_end);
3548 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3550 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3551 TASK_UNINTERRUPTIBLE);
3554 static noinline_for_stack int
3555 lock_extent_buffer_for_io(struct extent_buffer *eb,
3556 struct btrfs_fs_info *fs_info,
3557 struct extent_page_data *epd)
3559 unsigned long i, num_pages;
3563 if (!btrfs_try_tree_write_lock(eb)) {
3565 flush_write_bio(epd);
3566 btrfs_tree_lock(eb);
3569 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3570 btrfs_tree_unlock(eb);
3574 flush_write_bio(epd);
3578 wait_on_extent_buffer_writeback(eb);
3579 btrfs_tree_lock(eb);
3580 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3582 btrfs_tree_unlock(eb);
3587 * We need to do this to prevent races in people who check if the eb is
3588 * under IO since we can end up having no IO bits set for a short period
3591 spin_lock(&eb->refs_lock);
3592 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3593 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3594 spin_unlock(&eb->refs_lock);
3595 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3596 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3598 fs_info->dirty_metadata_batch);
3601 spin_unlock(&eb->refs_lock);
3604 btrfs_tree_unlock(eb);
3609 num_pages = num_extent_pages(eb->start, eb->len);
3610 for (i = 0; i < num_pages; i++) {
3611 struct page *p = eb->pages[i];
3613 if (!trylock_page(p)) {
3615 flush_write_bio(epd);
3625 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3627 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3628 smp_mb__after_atomic();
3629 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3632 static void set_btree_ioerr(struct page *page)
3634 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3635 struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode);
3638 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3642 * If writeback for a btree extent that doesn't belong to a log tree
3643 * failed, increment the counter transaction->eb_write_errors.
3644 * We do this because while the transaction is running and before it's
3645 * committing (when we call filemap_fdata[write|wait]_range against
3646 * the btree inode), we might have
3647 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3648 * returns an error or an error happens during writeback, when we're
3649 * committing the transaction we wouldn't know about it, since the pages
3650 * can be no longer dirty nor marked anymore for writeback (if a
3651 * subsequent modification to the extent buffer didn't happen before the
3652 * transaction commit), which makes filemap_fdata[write|wait]_range not
3653 * able to find the pages tagged with SetPageError at transaction
3654 * commit time. So if this happens we must abort the transaction,
3655 * otherwise we commit a super block with btree roots that point to
3656 * btree nodes/leafs whose content on disk is invalid - either garbage
3657 * or the content of some node/leaf from a past generation that got
3658 * cowed or deleted and is no longer valid.
3660 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3661 * not be enough - we need to distinguish between log tree extents vs
3662 * non-log tree extents, and the next filemap_fdatawait_range() call
3663 * will catch and clear such errors in the mapping - and that call might
3664 * be from a log sync and not from a transaction commit. Also, checking
3665 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3666 * not done and would not be reliable - the eb might have been released
3667 * from memory and reading it back again means that flag would not be
3668 * set (since it's a runtime flag, not persisted on disk).
3670 * Using the flags below in the btree inode also makes us achieve the
3671 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3672 * writeback for all dirty pages and before filemap_fdatawait_range()
3673 * is called, the writeback for all dirty pages had already finished
3674 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3675 * filemap_fdatawait_range() would return success, as it could not know
3676 * that writeback errors happened (the pages were no longer tagged for
3679 switch (eb->log_index) {
3681 set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags);
3684 set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
3687 set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
3690 BUG(); /* unexpected, logic error */
3694 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3696 struct bio_vec *bvec;
3697 struct extent_buffer *eb;
3700 bio_for_each_segment_all(bvec, bio, i) {
3701 struct page *page = bvec->bv_page;
3703 eb = (struct extent_buffer *)page->private;
3705 done = atomic_dec_and_test(&eb->io_pages);
3707 if (err || test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3708 ClearPageUptodate(page);
3709 set_btree_ioerr(page);
3712 end_page_writeback(page);
3717 end_extent_buffer_writeback(eb);
3723 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3724 struct btrfs_fs_info *fs_info,
3725 struct writeback_control *wbc,
3726 struct extent_page_data *epd)
3728 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3729 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3730 u64 offset = eb->start;
3731 unsigned long i, num_pages;
3732 unsigned long bio_flags = 0;
3733 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3736 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3737 num_pages = num_extent_pages(eb->start, eb->len);
3738 atomic_set(&eb->io_pages, num_pages);
3739 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3740 bio_flags = EXTENT_BIO_TREE_LOG;
3742 for (i = 0; i < num_pages; i++) {
3743 struct page *p = eb->pages[i];
3745 clear_page_dirty_for_io(p);
3746 set_page_writeback(p);
3747 ret = submit_extent_page(rw, tree, p, offset >> 9,
3748 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3749 -1, end_bio_extent_buffer_writepage,
3750 0, epd->bio_flags, bio_flags);
3751 epd->bio_flags = bio_flags;
3754 end_page_writeback(p);
3755 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3756 end_extent_buffer_writeback(eb);
3760 offset += PAGE_CACHE_SIZE;
3761 update_nr_written(p, wbc, 1);
3765 if (unlikely(ret)) {
3766 for (; i < num_pages; i++) {
3767 struct page *p = eb->pages[i];
3768 clear_page_dirty_for_io(p);
3776 int btree_write_cache_pages(struct address_space *mapping,
3777 struct writeback_control *wbc)
3779 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3780 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3781 struct extent_buffer *eb, *prev_eb = NULL;
3782 struct extent_page_data epd = {
3786 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3791 int nr_to_write_done = 0;
3792 struct pagevec pvec;
3795 pgoff_t end; /* Inclusive */
3799 pagevec_init(&pvec, 0);
3800 if (wbc->range_cyclic) {
3801 index = mapping->writeback_index; /* Start from prev offset */
3804 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3805 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3808 if (wbc->sync_mode == WB_SYNC_ALL)
3809 tag = PAGECACHE_TAG_TOWRITE;
3811 tag = PAGECACHE_TAG_DIRTY;
3813 if (wbc->sync_mode == WB_SYNC_ALL)
3814 tag_pages_for_writeback(mapping, index, end);
3815 while (!done && !nr_to_write_done && (index <= end) &&
3816 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3817 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3821 for (i = 0; i < nr_pages; i++) {
3822 struct page *page = pvec.pages[i];
3824 if (!PagePrivate(page))
3827 if (!wbc->range_cyclic && page->index > end) {
3832 spin_lock(&mapping->private_lock);
3833 if (!PagePrivate(page)) {
3834 spin_unlock(&mapping->private_lock);
3838 eb = (struct extent_buffer *)page->private;
3841 * Shouldn't happen and normally this would be a BUG_ON
3842 * but no sense in crashing the users box for something
3843 * we can survive anyway.
3846 spin_unlock(&mapping->private_lock);
3850 if (eb == prev_eb) {
3851 spin_unlock(&mapping->private_lock);
3855 ret = atomic_inc_not_zero(&eb->refs);
3856 spin_unlock(&mapping->private_lock);
3861 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3863 free_extent_buffer(eb);
3867 ret = write_one_eb(eb, fs_info, wbc, &epd);
3870 free_extent_buffer(eb);
3873 free_extent_buffer(eb);
3876 * the filesystem may choose to bump up nr_to_write.
3877 * We have to make sure to honor the new nr_to_write
3880 nr_to_write_done = wbc->nr_to_write <= 0;
3882 pagevec_release(&pvec);
3885 if (!scanned && !done) {
3887 * We hit the last page and there is more work to be done: wrap
3888 * back to the start of the file
3894 flush_write_bio(&epd);
3899 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3900 * @mapping: address space structure to write
3901 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3902 * @writepage: function called for each page
3903 * @data: data passed to writepage function
3905 * If a page is already under I/O, write_cache_pages() skips it, even
3906 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3907 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3908 * and msync() need to guarantee that all the data which was dirty at the time
3909 * the call was made get new I/O started against them. If wbc->sync_mode is
3910 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3911 * existing IO to complete.
3913 static int extent_write_cache_pages(struct extent_io_tree *tree,
3914 struct address_space *mapping,
3915 struct writeback_control *wbc,
3916 writepage_t writepage, void *data,
3917 void (*flush_fn)(void *))
3919 struct inode *inode = mapping->host;
3923 int nr_to_write_done = 0;
3924 struct pagevec pvec;
3927 pgoff_t end; /* Inclusive */
3932 * We have to hold onto the inode so that ordered extents can do their
3933 * work when the IO finishes. The alternative to this is failing to add
3934 * an ordered extent if the igrab() fails there and that is a huge pain
3935 * to deal with, so instead just hold onto the inode throughout the
3936 * writepages operation. If it fails here we are freeing up the inode
3937 * anyway and we'd rather not waste our time writing out stuff that is
3938 * going to be truncated anyway.
3943 pagevec_init(&pvec, 0);
3944 if (wbc->range_cyclic) {
3945 index = mapping->writeback_index; /* Start from prev offset */
3948 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3949 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3952 if (wbc->sync_mode == WB_SYNC_ALL)
3953 tag = PAGECACHE_TAG_TOWRITE;
3955 tag = PAGECACHE_TAG_DIRTY;
3957 if (wbc->sync_mode == WB_SYNC_ALL)
3958 tag_pages_for_writeback(mapping, index, end);
3959 while (!done && !nr_to_write_done && (index <= end) &&
3960 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3961 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3965 for (i = 0; i < nr_pages; i++) {
3966 struct page *page = pvec.pages[i];
3969 * At this point we hold neither mapping->tree_lock nor
3970 * lock on the page itself: the page may be truncated or
3971 * invalidated (changing page->mapping to NULL), or even
3972 * swizzled back from swapper_space to tmpfs file
3975 if (!trylock_page(page)) {
3980 if (unlikely(page->mapping != mapping)) {
3985 if (!wbc->range_cyclic && page->index > end) {
3991 if (wbc->sync_mode != WB_SYNC_NONE) {
3992 if (PageWriteback(page))
3994 wait_on_page_writeback(page);
3997 if (PageWriteback(page) ||
3998 !clear_page_dirty_for_io(page)) {
4003 ret = (*writepage)(page, wbc, data);
4005 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4009 if (!err && ret < 0)
4013 * the filesystem may choose to bump up nr_to_write.
4014 * We have to make sure to honor the new nr_to_write
4017 nr_to_write_done = wbc->nr_to_write <= 0;
4019 pagevec_release(&pvec);
4022 if (!scanned && !done && !err) {
4024 * We hit the last page and there is more work to be done: wrap
4025 * back to the start of the file
4031 btrfs_add_delayed_iput(inode);
4035 static void flush_epd_write_bio(struct extent_page_data *epd)
4044 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
4045 BUG_ON(ret < 0); /* -ENOMEM */
4050 static noinline void flush_write_bio(void *data)
4052 struct extent_page_data *epd = data;
4053 flush_epd_write_bio(epd);
4056 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4057 get_extent_t *get_extent,
4058 struct writeback_control *wbc)
4061 struct extent_page_data epd = {
4064 .get_extent = get_extent,
4066 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4070 ret = __extent_writepage(page, wbc, &epd);
4072 flush_epd_write_bio(&epd);
4076 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4077 u64 start, u64 end, get_extent_t *get_extent,
4081 struct address_space *mapping = inode->i_mapping;
4083 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
4086 struct extent_page_data epd = {
4089 .get_extent = get_extent,
4091 .sync_io = mode == WB_SYNC_ALL,
4094 struct writeback_control wbc_writepages = {
4096 .nr_to_write = nr_pages * 2,
4097 .range_start = start,
4098 .range_end = end + 1,
4101 while (start <= end) {
4102 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
4103 if (clear_page_dirty_for_io(page))
4104 ret = __extent_writepage(page, &wbc_writepages, &epd);
4106 if (tree->ops && tree->ops->writepage_end_io_hook)
4107 tree->ops->writepage_end_io_hook(page, start,
4108 start + PAGE_CACHE_SIZE - 1,
4112 page_cache_release(page);
4113 start += PAGE_CACHE_SIZE;
4116 flush_epd_write_bio(&epd);
4120 int extent_writepages(struct extent_io_tree *tree,
4121 struct address_space *mapping,
4122 get_extent_t *get_extent,
4123 struct writeback_control *wbc)
4126 struct extent_page_data epd = {
4129 .get_extent = get_extent,
4131 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4135 ret = extent_write_cache_pages(tree, mapping, wbc,
4136 __extent_writepage, &epd,
4138 flush_epd_write_bio(&epd);
4142 int extent_readpages(struct extent_io_tree *tree,
4143 struct address_space *mapping,
4144 struct list_head *pages, unsigned nr_pages,
4145 get_extent_t get_extent)
4147 struct bio *bio = NULL;
4149 unsigned long bio_flags = 0;
4150 struct page *pagepool[16];
4152 struct extent_map *em_cached = NULL;
4155 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4156 page = list_entry(pages->prev, struct page, lru);
4158 prefetchw(&page->flags);
4159 list_del(&page->lru);
4160 if (add_to_page_cache_lru(page, mapping,
4161 page->index, GFP_NOFS)) {
4162 page_cache_release(page);
4166 pagepool[nr++] = page;
4167 if (nr < ARRAY_SIZE(pagepool))
4169 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4170 &bio, 0, &bio_flags, READ);
4174 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4175 &bio, 0, &bio_flags, READ);
4178 free_extent_map(em_cached);
4180 BUG_ON(!list_empty(pages));
4182 return submit_one_bio(READ, bio, 0, bio_flags);
4187 * basic invalidatepage code, this waits on any locked or writeback
4188 * ranges corresponding to the page, and then deletes any extent state
4189 * records from the tree
4191 int extent_invalidatepage(struct extent_io_tree *tree,
4192 struct page *page, unsigned long offset)
4194 struct extent_state *cached_state = NULL;
4195 u64 start = page_offset(page);
4196 u64 end = start + PAGE_CACHE_SIZE - 1;
4197 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4199 start += ALIGN(offset, blocksize);
4203 lock_extent_bits(tree, start, end, 0, &cached_state);
4204 wait_on_page_writeback(page);
4205 clear_extent_bit(tree, start, end,
4206 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4207 EXTENT_DO_ACCOUNTING,
4208 1, 1, &cached_state, GFP_NOFS);
4213 * a helper for releasepage, this tests for areas of the page that
4214 * are locked or under IO and drops the related state bits if it is safe
4217 static int try_release_extent_state(struct extent_map_tree *map,
4218 struct extent_io_tree *tree,
4219 struct page *page, gfp_t mask)
4221 u64 start = page_offset(page);
4222 u64 end = start + PAGE_CACHE_SIZE - 1;
4225 if (test_range_bit(tree, start, end,
4226 EXTENT_IOBITS, 0, NULL))
4229 if ((mask & GFP_NOFS) == GFP_NOFS)
4232 * at this point we can safely clear everything except the
4233 * locked bit and the nodatasum bit
4235 ret = clear_extent_bit(tree, start, end,
4236 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4239 /* if clear_extent_bit failed for enomem reasons,
4240 * we can't allow the release to continue.
4251 * a helper for releasepage. As long as there are no locked extents
4252 * in the range corresponding to the page, both state records and extent
4253 * map records are removed
4255 int try_release_extent_mapping(struct extent_map_tree *map,
4256 struct extent_io_tree *tree, struct page *page,
4259 struct extent_map *em;
4260 u64 start = page_offset(page);
4261 u64 end = start + PAGE_CACHE_SIZE - 1;
4263 if ((mask & __GFP_WAIT) &&
4264 page->mapping->host->i_size > 16 * 1024 * 1024) {
4266 while (start <= end) {
4267 len = end - start + 1;
4268 write_lock(&map->lock);
4269 em = lookup_extent_mapping(map, start, len);
4271 write_unlock(&map->lock);
4274 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4275 em->start != start) {
4276 write_unlock(&map->lock);
4277 free_extent_map(em);
4280 if (!test_range_bit(tree, em->start,
4281 extent_map_end(em) - 1,
4282 EXTENT_LOCKED | EXTENT_WRITEBACK,
4284 remove_extent_mapping(map, em);
4285 /* once for the rb tree */
4286 free_extent_map(em);
4288 start = extent_map_end(em);
4289 write_unlock(&map->lock);
4292 free_extent_map(em);
4295 return try_release_extent_state(map, tree, page, mask);
4299 * helper function for fiemap, which doesn't want to see any holes.
4300 * This maps until we find something past 'last'
4302 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4305 get_extent_t *get_extent)
4307 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4308 struct extent_map *em;
4315 len = last - offset;
4318 len = ALIGN(len, sectorsize);
4319 em = get_extent(inode, NULL, 0, offset, len, 0);
4320 if (IS_ERR_OR_NULL(em))
4323 /* if this isn't a hole return it */
4324 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4325 em->block_start != EXTENT_MAP_HOLE) {
4329 /* this is a hole, advance to the next extent */
4330 offset = extent_map_end(em);
4331 free_extent_map(em);
4338 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4339 __u64 start, __u64 len, get_extent_t *get_extent)
4343 u64 max = start + len;
4347 u64 last_for_get_extent = 0;
4349 u64 isize = i_size_read(inode);
4350 struct btrfs_key found_key;
4351 struct extent_map *em = NULL;
4352 struct extent_state *cached_state = NULL;
4353 struct btrfs_path *path;
4354 struct btrfs_root *root = BTRFS_I(inode)->root;
4363 path = btrfs_alloc_path();
4366 path->leave_spinning = 1;
4368 start = round_down(start, BTRFS_I(inode)->root->sectorsize);
4369 len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start;
4372 * lookup the last file extent. We're not using i_size here
4373 * because there might be preallocation past i_size
4375 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4378 btrfs_free_path(path);
4383 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4384 found_type = found_key.type;
4386 /* No extents, but there might be delalloc bits */
4387 if (found_key.objectid != btrfs_ino(inode) ||
4388 found_type != BTRFS_EXTENT_DATA_KEY) {
4389 /* have to trust i_size as the end */
4391 last_for_get_extent = isize;
4394 * remember the start of the last extent. There are a
4395 * bunch of different factors that go into the length of the
4396 * extent, so its much less complex to remember where it started
4398 last = found_key.offset;
4399 last_for_get_extent = last + 1;
4401 btrfs_release_path(path);
4404 * we might have some extents allocated but more delalloc past those
4405 * extents. so, we trust isize unless the start of the last extent is
4410 last_for_get_extent = isize;
4413 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4416 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4426 u64 offset_in_extent = 0;
4428 /* break if the extent we found is outside the range */
4429 if (em->start >= max || extent_map_end(em) < off)
4433 * get_extent may return an extent that starts before our
4434 * requested range. We have to make sure the ranges
4435 * we return to fiemap always move forward and don't
4436 * overlap, so adjust the offsets here
4438 em_start = max(em->start, off);
4441 * record the offset from the start of the extent
4442 * for adjusting the disk offset below. Only do this if the
4443 * extent isn't compressed since our in ram offset may be past
4444 * what we have actually allocated on disk.
4446 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4447 offset_in_extent = em_start - em->start;
4448 em_end = extent_map_end(em);
4449 em_len = em_end - em_start;
4454 * bump off for our next call to get_extent
4456 off = extent_map_end(em);
4460 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4462 flags |= FIEMAP_EXTENT_LAST;
4463 } else if (em->block_start == EXTENT_MAP_INLINE) {
4464 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4465 FIEMAP_EXTENT_NOT_ALIGNED);
4466 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4467 flags |= (FIEMAP_EXTENT_DELALLOC |
4468 FIEMAP_EXTENT_UNKNOWN);
4469 } else if (fieinfo->fi_extents_max) {
4470 u64 bytenr = em->block_start -
4471 (em->start - em->orig_start);
4473 disko = em->block_start + offset_in_extent;
4476 * As btrfs supports shared space, this information
4477 * can be exported to userspace tools via
4478 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4479 * then we're just getting a count and we can skip the
4482 ret = btrfs_check_shared(NULL, root->fs_info,
4484 btrfs_ino(inode), bytenr);
4488 flags |= FIEMAP_EXTENT_SHARED;
4491 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4492 flags |= FIEMAP_EXTENT_ENCODED;
4494 free_extent_map(em);
4496 if ((em_start >= last) || em_len == (u64)-1 ||
4497 (last == (u64)-1 && isize <= em_end)) {
4498 flags |= FIEMAP_EXTENT_LAST;
4502 /* now scan forward to see if this is really the last extent. */
4503 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4510 flags |= FIEMAP_EXTENT_LAST;
4513 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4519 free_extent_map(em);
4521 btrfs_free_path(path);
4522 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4523 &cached_state, GFP_NOFS);
4527 static void __free_extent_buffer(struct extent_buffer *eb)
4529 btrfs_leak_debug_del(&eb->leak_list);
4530 kmem_cache_free(extent_buffer_cache, eb);
4533 int extent_buffer_under_io(struct extent_buffer *eb)
4535 return (atomic_read(&eb->io_pages) ||
4536 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4537 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4541 * Helper for releasing extent buffer page.
4543 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4545 unsigned long index;
4547 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4549 BUG_ON(extent_buffer_under_io(eb));
4551 index = num_extent_pages(eb->start, eb->len);
4557 page = eb->pages[index];
4558 if (page && mapped) {
4559 spin_lock(&page->mapping->private_lock);
4561 * We do this since we'll remove the pages after we've
4562 * removed the eb from the radix tree, so we could race
4563 * and have this page now attached to the new eb. So
4564 * only clear page_private if it's still connected to
4567 if (PagePrivate(page) &&
4568 page->private == (unsigned long)eb) {
4569 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4570 BUG_ON(PageDirty(page));
4571 BUG_ON(PageWriteback(page));
4573 * We need to make sure we haven't be attached
4576 ClearPagePrivate(page);
4577 set_page_private(page, 0);
4578 /* One for the page private */
4579 page_cache_release(page);
4581 spin_unlock(&page->mapping->private_lock);
4585 /* One for when we alloced the page */
4586 page_cache_release(page);
4588 } while (index != 0);
4592 * Helper for releasing the extent buffer.
4594 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4596 btrfs_release_extent_buffer_page(eb);
4597 __free_extent_buffer(eb);
4600 static struct extent_buffer *
4601 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4604 struct extent_buffer *eb = NULL;
4606 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS);
4611 eb->fs_info = fs_info;
4613 rwlock_init(&eb->lock);
4614 atomic_set(&eb->write_locks, 0);
4615 atomic_set(&eb->read_locks, 0);
4616 atomic_set(&eb->blocking_readers, 0);
4617 atomic_set(&eb->blocking_writers, 0);
4618 atomic_set(&eb->spinning_readers, 0);
4619 atomic_set(&eb->spinning_writers, 0);
4620 eb->lock_nested = 0;
4621 init_waitqueue_head(&eb->write_lock_wq);
4622 init_waitqueue_head(&eb->read_lock_wq);
4624 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4626 spin_lock_init(&eb->refs_lock);
4627 atomic_set(&eb->refs, 1);
4628 atomic_set(&eb->io_pages, 0);
4631 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4633 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4634 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4635 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4640 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4644 struct extent_buffer *new;
4645 unsigned long num_pages = num_extent_pages(src->start, src->len);
4647 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4651 for (i = 0; i < num_pages; i++) {
4652 p = alloc_page(GFP_NOFS);
4654 btrfs_release_extent_buffer(new);
4657 attach_extent_buffer_page(new, p);
4658 WARN_ON(PageDirty(p));
4663 copy_extent_buffer(new, src, 0, 0, src->len);
4664 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4665 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4670 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4673 struct extent_buffer *eb;
4675 unsigned long num_pages;
4680 * Called only from tests that don't always have a fs_info
4681 * available, but we know that nodesize is 4096
4685 len = fs_info->tree_root->nodesize;
4687 num_pages = num_extent_pages(0, len);
4689 eb = __alloc_extent_buffer(fs_info, start, len);
4693 for (i = 0; i < num_pages; i++) {
4694 eb->pages[i] = alloc_page(GFP_NOFS);
4698 set_extent_buffer_uptodate(eb);
4699 btrfs_set_header_nritems(eb, 0);
4700 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4705 __free_page(eb->pages[i - 1]);
4706 __free_extent_buffer(eb);
4710 static void check_buffer_tree_ref(struct extent_buffer *eb)
4713 /* the ref bit is tricky. We have to make sure it is set
4714 * if we have the buffer dirty. Otherwise the
4715 * code to free a buffer can end up dropping a dirty
4718 * Once the ref bit is set, it won't go away while the
4719 * buffer is dirty or in writeback, and it also won't
4720 * go away while we have the reference count on the
4723 * We can't just set the ref bit without bumping the
4724 * ref on the eb because free_extent_buffer might
4725 * see the ref bit and try to clear it. If this happens
4726 * free_extent_buffer might end up dropping our original
4727 * ref by mistake and freeing the page before we are able
4728 * to add one more ref.
4730 * So bump the ref count first, then set the bit. If someone
4731 * beat us to it, drop the ref we added.
4733 refs = atomic_read(&eb->refs);
4734 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4737 spin_lock(&eb->refs_lock);
4738 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4739 atomic_inc(&eb->refs);
4740 spin_unlock(&eb->refs_lock);
4743 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4744 struct page *accessed)
4746 unsigned long num_pages, i;
4748 check_buffer_tree_ref(eb);
4750 num_pages = num_extent_pages(eb->start, eb->len);
4751 for (i = 0; i < num_pages; i++) {
4752 struct page *p = eb->pages[i];
4755 mark_page_accessed(p);
4759 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4762 struct extent_buffer *eb;
4765 eb = radix_tree_lookup(&fs_info->buffer_radix,
4766 start >> PAGE_CACHE_SHIFT);
4767 if (eb && atomic_inc_not_zero(&eb->refs)) {
4769 mark_extent_buffer_accessed(eb, NULL);
4777 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4778 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4781 struct extent_buffer *eb, *exists = NULL;
4784 eb = find_extent_buffer(fs_info, start);
4787 eb = alloc_dummy_extent_buffer(fs_info, start);
4790 eb->fs_info = fs_info;
4792 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4795 spin_lock(&fs_info->buffer_lock);
4796 ret = radix_tree_insert(&fs_info->buffer_radix,
4797 start >> PAGE_CACHE_SHIFT, eb);
4798 spin_unlock(&fs_info->buffer_lock);
4799 radix_tree_preload_end();
4800 if (ret == -EEXIST) {
4801 exists = find_extent_buffer(fs_info, start);
4807 check_buffer_tree_ref(eb);
4808 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4811 * We will free dummy extent buffer's if they come into
4812 * free_extent_buffer with a ref count of 2, but if we are using this we
4813 * want the buffers to stay in memory until we're done with them, so
4814 * bump the ref count again.
4816 atomic_inc(&eb->refs);
4819 btrfs_release_extent_buffer(eb);
4824 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4827 unsigned long len = fs_info->tree_root->nodesize;
4828 unsigned long num_pages = num_extent_pages(start, len);
4830 unsigned long index = start >> PAGE_CACHE_SHIFT;
4831 struct extent_buffer *eb;
4832 struct extent_buffer *exists = NULL;
4834 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4838 eb = find_extent_buffer(fs_info, start);
4842 eb = __alloc_extent_buffer(fs_info, start, len);
4846 for (i = 0; i < num_pages; i++, index++) {
4847 p = find_or_create_page(mapping, index, GFP_NOFS);
4851 spin_lock(&mapping->private_lock);
4852 if (PagePrivate(p)) {
4854 * We could have already allocated an eb for this page
4855 * and attached one so lets see if we can get a ref on
4856 * the existing eb, and if we can we know it's good and
4857 * we can just return that one, else we know we can just
4858 * overwrite page->private.
4860 exists = (struct extent_buffer *)p->private;
4861 if (atomic_inc_not_zero(&exists->refs)) {
4862 spin_unlock(&mapping->private_lock);
4864 page_cache_release(p);
4865 mark_extent_buffer_accessed(exists, p);
4870 * Do this so attach doesn't complain and we need to
4871 * drop the ref the old guy had.
4873 ClearPagePrivate(p);
4874 WARN_ON(PageDirty(p));
4875 page_cache_release(p);
4877 attach_extent_buffer_page(eb, p);
4878 spin_unlock(&mapping->private_lock);
4879 WARN_ON(PageDirty(p));
4881 if (!PageUptodate(p))
4885 * see below about how we avoid a nasty race with release page
4886 * and why we unlock later
4890 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4892 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4896 spin_lock(&fs_info->buffer_lock);
4897 ret = radix_tree_insert(&fs_info->buffer_radix,
4898 start >> PAGE_CACHE_SHIFT, eb);
4899 spin_unlock(&fs_info->buffer_lock);
4900 radix_tree_preload_end();
4901 if (ret == -EEXIST) {
4902 exists = find_extent_buffer(fs_info, start);
4908 /* add one reference for the tree */
4909 check_buffer_tree_ref(eb);
4910 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4913 * there is a race where release page may have
4914 * tried to find this extent buffer in the radix
4915 * but failed. It will tell the VM it is safe to
4916 * reclaim the, and it will clear the page private bit.
4917 * We must make sure to set the page private bit properly
4918 * after the extent buffer is in the radix tree so
4919 * it doesn't get lost
4921 SetPageChecked(eb->pages[0]);
4922 for (i = 1; i < num_pages; i++) {
4924 ClearPageChecked(p);
4927 unlock_page(eb->pages[0]);
4931 for (i = 0; i < num_pages; i++) {
4933 unlock_page(eb->pages[i]);
4936 WARN_ON(!atomic_dec_and_test(&eb->refs));
4937 btrfs_release_extent_buffer(eb);
4941 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4943 struct extent_buffer *eb =
4944 container_of(head, struct extent_buffer, rcu_head);
4946 __free_extent_buffer(eb);
4949 /* Expects to have eb->eb_lock already held */
4950 static int release_extent_buffer(struct extent_buffer *eb)
4952 WARN_ON(atomic_read(&eb->refs) == 0);
4953 if (atomic_dec_and_test(&eb->refs)) {
4954 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4955 struct btrfs_fs_info *fs_info = eb->fs_info;
4957 spin_unlock(&eb->refs_lock);
4959 spin_lock(&fs_info->buffer_lock);
4960 radix_tree_delete(&fs_info->buffer_radix,
4961 eb->start >> PAGE_CACHE_SHIFT);
4962 spin_unlock(&fs_info->buffer_lock);
4964 spin_unlock(&eb->refs_lock);
4967 /* Should be safe to release our pages at this point */
4968 btrfs_release_extent_buffer_page(eb);
4969 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4972 spin_unlock(&eb->refs_lock);
4977 void free_extent_buffer(struct extent_buffer *eb)
4985 refs = atomic_read(&eb->refs);
4988 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4993 spin_lock(&eb->refs_lock);
4994 if (atomic_read(&eb->refs) == 2 &&
4995 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4996 atomic_dec(&eb->refs);
4998 if (atomic_read(&eb->refs) == 2 &&
4999 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5000 !extent_buffer_under_io(eb) &&
5001 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5002 atomic_dec(&eb->refs);
5005 * I know this is terrible, but it's temporary until we stop tracking
5006 * the uptodate bits and such for the extent buffers.
5008 release_extent_buffer(eb);
5011 void free_extent_buffer_stale(struct extent_buffer *eb)
5016 spin_lock(&eb->refs_lock);
5017 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5019 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5020 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5021 atomic_dec(&eb->refs);
5022 release_extent_buffer(eb);
5025 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5028 unsigned long num_pages;
5031 num_pages = num_extent_pages(eb->start, eb->len);
5033 for (i = 0; i < num_pages; i++) {
5034 page = eb->pages[i];
5035 if (!PageDirty(page))
5039 WARN_ON(!PagePrivate(page));
5041 clear_page_dirty_for_io(page);
5042 spin_lock_irq(&page->mapping->tree_lock);
5043 if (!PageDirty(page)) {
5044 radix_tree_tag_clear(&page->mapping->page_tree,
5046 PAGECACHE_TAG_DIRTY);
5048 spin_unlock_irq(&page->mapping->tree_lock);
5049 ClearPageError(page);
5052 WARN_ON(atomic_read(&eb->refs) == 0);
5055 int set_extent_buffer_dirty(struct extent_buffer *eb)
5058 unsigned long num_pages;
5061 check_buffer_tree_ref(eb);
5063 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5065 num_pages = num_extent_pages(eb->start, eb->len);
5066 WARN_ON(atomic_read(&eb->refs) == 0);
5067 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5069 for (i = 0; i < num_pages; i++)
5070 set_page_dirty(eb->pages[i]);
5074 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
5078 unsigned long num_pages;
5080 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5081 num_pages = num_extent_pages(eb->start, eb->len);
5082 for (i = 0; i < num_pages; i++) {
5083 page = eb->pages[i];
5085 ClearPageUptodate(page);
5090 int set_extent_buffer_uptodate(struct extent_buffer *eb)
5094 unsigned long num_pages;
5096 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5097 num_pages = num_extent_pages(eb->start, eb->len);
5098 for (i = 0; i < num_pages; i++) {
5099 page = eb->pages[i];
5100 SetPageUptodate(page);
5105 int extent_buffer_uptodate(struct extent_buffer *eb)
5107 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5110 int read_extent_buffer_pages(struct extent_io_tree *tree,
5111 struct extent_buffer *eb, u64 start, int wait,
5112 get_extent_t *get_extent, int mirror_num)
5115 unsigned long start_i;
5119 int locked_pages = 0;
5120 int all_uptodate = 1;
5121 unsigned long num_pages;
5122 unsigned long num_reads = 0;
5123 struct bio *bio = NULL;
5124 unsigned long bio_flags = 0;
5126 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5130 WARN_ON(start < eb->start);
5131 start_i = (start >> PAGE_CACHE_SHIFT) -
5132 (eb->start >> PAGE_CACHE_SHIFT);
5137 num_pages = num_extent_pages(eb->start, eb->len);
5138 for (i = start_i; i < num_pages; i++) {
5139 page = eb->pages[i];
5140 if (wait == WAIT_NONE) {
5141 if (!trylock_page(page))
5147 if (!PageUptodate(page)) {
5154 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5158 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5159 eb->read_mirror = 0;
5160 atomic_set(&eb->io_pages, num_reads);
5161 for (i = start_i; i < num_pages; i++) {
5162 page = eb->pages[i];
5163 if (!PageUptodate(page)) {
5164 ClearPageError(page);
5165 err = __extent_read_full_page(tree, page,
5167 mirror_num, &bio_flags,
5177 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
5183 if (ret || wait != WAIT_COMPLETE)
5186 for (i = start_i; i < num_pages; i++) {
5187 page = eb->pages[i];
5188 wait_on_page_locked(page);
5189 if (!PageUptodate(page))
5197 while (locked_pages > 0) {
5198 page = eb->pages[i];
5206 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
5207 unsigned long start,
5214 char *dst = (char *)dstv;
5215 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5216 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5218 WARN_ON(start > eb->len);
5219 WARN_ON(start + len > eb->start + eb->len);
5221 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5224 page = eb->pages[i];
5226 cur = min(len, (PAGE_CACHE_SIZE - offset));
5227 kaddr = page_address(page);
5228 memcpy(dst, kaddr + offset, cur);
5237 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv,
5238 unsigned long start,
5245 char __user *dst = (char __user *)dstv;
5246 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5247 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5250 WARN_ON(start > eb->len);
5251 WARN_ON(start + len > eb->start + eb->len);
5253 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5256 page = eb->pages[i];
5258 cur = min(len, (PAGE_CACHE_SIZE - offset));
5259 kaddr = page_address(page);
5260 if (copy_to_user(dst, kaddr + offset, cur)) {
5274 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
5275 unsigned long min_len, char **map,
5276 unsigned long *map_start,
5277 unsigned long *map_len)
5279 size_t offset = start & (PAGE_CACHE_SIZE - 1);
5282 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5283 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5284 unsigned long end_i = (start_offset + start + min_len - 1) >>
5291 offset = start_offset;
5295 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
5298 if (start + min_len > eb->len) {
5299 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
5301 eb->start, eb->len, start, min_len);
5306 kaddr = page_address(p);
5307 *map = kaddr + offset;
5308 *map_len = PAGE_CACHE_SIZE - offset;
5312 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5313 unsigned long start,
5320 char *ptr = (char *)ptrv;
5321 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5322 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5325 WARN_ON(start > eb->len);
5326 WARN_ON(start + len > eb->start + eb->len);
5328 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5331 page = eb->pages[i];
5333 cur = min(len, (PAGE_CACHE_SIZE - offset));
5335 kaddr = page_address(page);
5336 ret = memcmp(ptr, kaddr + offset, cur);
5348 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5349 unsigned long start, unsigned long len)
5355 char *src = (char *)srcv;
5356 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5357 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5359 WARN_ON(start > eb->len);
5360 WARN_ON(start + len > eb->start + eb->len);
5362 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5365 page = eb->pages[i];
5366 WARN_ON(!PageUptodate(page));
5368 cur = min(len, PAGE_CACHE_SIZE - offset);
5369 kaddr = page_address(page);
5370 memcpy(kaddr + offset, src, cur);
5379 void memset_extent_buffer(struct extent_buffer *eb, char c,
5380 unsigned long start, unsigned long len)
5386 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5387 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5389 WARN_ON(start > eb->len);
5390 WARN_ON(start + len > eb->start + eb->len);
5392 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5395 page = eb->pages[i];
5396 WARN_ON(!PageUptodate(page));
5398 cur = min(len, PAGE_CACHE_SIZE - offset);
5399 kaddr = page_address(page);
5400 memset(kaddr + offset, c, cur);
5408 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5409 unsigned long dst_offset, unsigned long src_offset,
5412 u64 dst_len = dst->len;
5417 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5418 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5420 WARN_ON(src->len != dst_len);
5422 offset = (start_offset + dst_offset) &
5423 (PAGE_CACHE_SIZE - 1);
5426 page = dst->pages[i];
5427 WARN_ON(!PageUptodate(page));
5429 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5431 kaddr = page_address(page);
5432 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5441 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5443 unsigned long distance = (src > dst) ? src - dst : dst - src;
5444 return distance < len;
5447 static void copy_pages(struct page *dst_page, struct page *src_page,
5448 unsigned long dst_off, unsigned long src_off,
5451 char *dst_kaddr = page_address(dst_page);
5453 int must_memmove = 0;
5455 if (dst_page != src_page) {
5456 src_kaddr = page_address(src_page);
5458 src_kaddr = dst_kaddr;
5459 if (areas_overlap(src_off, dst_off, len))
5464 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5466 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5469 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5470 unsigned long src_offset, unsigned long len)
5473 size_t dst_off_in_page;
5474 size_t src_off_in_page;
5475 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5476 unsigned long dst_i;
5477 unsigned long src_i;
5479 if (src_offset + len > dst->len) {
5480 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5481 "len %lu dst len %lu\n", src_offset, len, dst->len);
5484 if (dst_offset + len > dst->len) {
5485 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5486 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5491 dst_off_in_page = (start_offset + dst_offset) &
5492 (PAGE_CACHE_SIZE - 1);
5493 src_off_in_page = (start_offset + src_offset) &
5494 (PAGE_CACHE_SIZE - 1);
5496 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5497 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5499 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5501 cur = min_t(unsigned long, cur,
5502 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5504 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5505 dst_off_in_page, src_off_in_page, cur);
5513 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5514 unsigned long src_offset, unsigned long len)
5517 size_t dst_off_in_page;
5518 size_t src_off_in_page;
5519 unsigned long dst_end = dst_offset + len - 1;
5520 unsigned long src_end = src_offset + len - 1;
5521 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5522 unsigned long dst_i;
5523 unsigned long src_i;
5525 if (src_offset + len > dst->len) {
5526 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5527 "len %lu len %lu\n", src_offset, len, dst->len);
5530 if (dst_offset + len > dst->len) {
5531 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5532 "len %lu len %lu\n", dst_offset, len, dst->len);
5535 if (dst_offset < src_offset) {
5536 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5540 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5541 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5543 dst_off_in_page = (start_offset + dst_end) &
5544 (PAGE_CACHE_SIZE - 1);
5545 src_off_in_page = (start_offset + src_end) &
5546 (PAGE_CACHE_SIZE - 1);
5548 cur = min_t(unsigned long, len, src_off_in_page + 1);
5549 cur = min(cur, dst_off_in_page + 1);
5550 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5551 dst_off_in_page - cur + 1,
5552 src_off_in_page - cur + 1, cur);
5560 int try_release_extent_buffer(struct page *page)
5562 struct extent_buffer *eb;
5565 * We need to make sure noboody is attaching this page to an eb right
5568 spin_lock(&page->mapping->private_lock);
5569 if (!PagePrivate(page)) {
5570 spin_unlock(&page->mapping->private_lock);
5574 eb = (struct extent_buffer *)page->private;
5578 * This is a little awful but should be ok, we need to make sure that
5579 * the eb doesn't disappear out from under us while we're looking at
5582 spin_lock(&eb->refs_lock);
5583 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5584 spin_unlock(&eb->refs_lock);
5585 spin_unlock(&page->mapping->private_lock);
5588 spin_unlock(&page->mapping->private_lock);
5591 * If tree ref isn't set then we know the ref on this eb is a real ref,
5592 * so just return, this page will likely be freed soon anyway.
5594 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5595 spin_unlock(&eb->refs_lock);
5599 return release_extent_buffer(eb);