1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/spinlock.h>
8 #include <linux/blkdev.h>
9 #include <linux/swap.h>
10 #include <linux/writeback.h>
11 #include <linux/pagevec.h>
12 #include <linux/prefetch.h>
13 #include <linux/cleancache.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 #include "check-integrity.h"
22 #include "rcu-string.h"
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
26 static struct bio_set *btrfs_bioset;
28 #ifdef CONFIG_BTRFS_DEBUG
29 static LIST_HEAD(buffers);
30 static LIST_HEAD(states);
32 static DEFINE_SPINLOCK(leak_lock);
35 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
39 spin_lock_irqsave(&leak_lock, flags);
41 spin_unlock_irqrestore(&leak_lock, flags);
45 void btrfs_leak_debug_del(struct list_head *entry)
49 spin_lock_irqsave(&leak_lock, flags);
51 spin_unlock_irqrestore(&leak_lock, flags);
55 void btrfs_leak_debug_check(void)
57 struct extent_state *state;
58 struct extent_buffer *eb;
60 while (!list_empty(&states)) {
61 state = list_entry(states.next, struct extent_state, leak_list);
62 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
63 "state %lu in tree %p refs %d\n",
64 state->start, state->end, state->state, state->tree,
65 atomic_read(&state->refs));
66 list_del(&state->leak_list);
67 kmem_cache_free(extent_state_cache, state);
70 while (!list_empty(&buffers)) {
71 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
72 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
74 eb->start, eb->len, atomic_read(&eb->refs));
75 list_del(&eb->leak_list);
76 kmem_cache_free(extent_buffer_cache, eb);
80 #define btrfs_debug_check_extent_io_range(inode, start, end) \
81 __btrfs_debug_check_extent_io_range(__func__, (inode), (start), (end))
82 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
83 struct inode *inode, u64 start, u64 end)
85 u64 isize = i_size_read(inode);
87 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
88 printk_ratelimited(KERN_DEBUG
89 "btrfs: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
90 caller, btrfs_ino(inode), isize, start, end);
94 #define btrfs_leak_debug_add(new, head) do {} while (0)
95 #define btrfs_leak_debug_del(entry) do {} while (0)
96 #define btrfs_leak_debug_check() do {} while (0)
97 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
100 #define BUFFER_LRU_MAX 64
105 struct rb_node rb_node;
108 struct extent_page_data {
110 struct extent_io_tree *tree;
111 get_extent_t *get_extent;
112 unsigned long bio_flags;
114 /* tells writepage not to lock the state bits for this range
115 * it still does the unlocking
117 unsigned int extent_locked:1;
119 /* tells the submit_bio code to use a WRITE_SYNC */
120 unsigned int sync_io:1;
123 static noinline void flush_write_bio(void *data);
124 static inline struct btrfs_fs_info *
125 tree_fs_info(struct extent_io_tree *tree)
127 return btrfs_sb(tree->mapping->host->i_sb);
130 int __init extent_io_init(void)
132 extent_state_cache = kmem_cache_create("btrfs_extent_state",
133 sizeof(struct extent_state), 0,
134 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
135 if (!extent_state_cache)
138 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
139 sizeof(struct extent_buffer), 0,
140 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
141 if (!extent_buffer_cache)
142 goto free_state_cache;
144 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
145 offsetof(struct btrfs_io_bio, bio));
147 goto free_buffer_cache;
151 kmem_cache_destroy(extent_buffer_cache);
152 extent_buffer_cache = NULL;
155 kmem_cache_destroy(extent_state_cache);
156 extent_state_cache = NULL;
160 void extent_io_exit(void)
162 btrfs_leak_debug_check();
165 * Make sure all delayed rcu free are flushed before we
169 if (extent_state_cache)
170 kmem_cache_destroy(extent_state_cache);
171 if (extent_buffer_cache)
172 kmem_cache_destroy(extent_buffer_cache);
174 bioset_free(btrfs_bioset);
177 void extent_io_tree_init(struct extent_io_tree *tree,
178 struct address_space *mapping)
180 tree->state = RB_ROOT;
181 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
183 tree->dirty_bytes = 0;
184 spin_lock_init(&tree->lock);
185 spin_lock_init(&tree->buffer_lock);
186 tree->mapping = mapping;
189 static struct extent_state *alloc_extent_state(gfp_t mask)
191 struct extent_state *state;
193 state = kmem_cache_alloc(extent_state_cache, mask);
199 btrfs_leak_debug_add(&state->leak_list, &states);
200 atomic_set(&state->refs, 1);
201 init_waitqueue_head(&state->wq);
202 trace_alloc_extent_state(state, mask, _RET_IP_);
206 void free_extent_state(struct extent_state *state)
210 if (atomic_dec_and_test(&state->refs)) {
211 WARN_ON(state->tree);
212 btrfs_leak_debug_del(&state->leak_list);
213 trace_free_extent_state(state, _RET_IP_);
214 kmem_cache_free(extent_state_cache, state);
218 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
219 struct rb_node *node)
221 struct rb_node **p = &root->rb_node;
222 struct rb_node *parent = NULL;
223 struct tree_entry *entry;
227 entry = rb_entry(parent, struct tree_entry, rb_node);
229 if (offset < entry->start)
231 else if (offset > entry->end)
237 rb_link_node(node, parent, p);
238 rb_insert_color(node, root);
242 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
243 struct rb_node **prev_ret,
244 struct rb_node **next_ret)
246 struct rb_root *root = &tree->state;
247 struct rb_node *n = root->rb_node;
248 struct rb_node *prev = NULL;
249 struct rb_node *orig_prev = NULL;
250 struct tree_entry *entry;
251 struct tree_entry *prev_entry = NULL;
254 entry = rb_entry(n, struct tree_entry, rb_node);
258 if (offset < entry->start)
260 else if (offset > entry->end)
268 while (prev && offset > prev_entry->end) {
269 prev = rb_next(prev);
270 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
277 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
278 while (prev && offset < prev_entry->start) {
279 prev = rb_prev(prev);
280 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
287 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
290 struct rb_node *prev = NULL;
293 ret = __etree_search(tree, offset, &prev, NULL);
299 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
300 struct extent_state *other)
302 if (tree->ops && tree->ops->merge_extent_hook)
303 tree->ops->merge_extent_hook(tree->mapping->host, new,
308 * utility function to look for merge candidates inside a given range.
309 * Any extents with matching state are merged together into a single
310 * extent in the tree. Extents with EXTENT_IO in their state field
311 * are not merged because the end_io handlers need to be able to do
312 * operations on them without sleeping (or doing allocations/splits).
314 * This should be called with the tree lock held.
316 static void merge_state(struct extent_io_tree *tree,
317 struct extent_state *state)
319 struct extent_state *other;
320 struct rb_node *other_node;
322 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
325 other_node = rb_prev(&state->rb_node);
327 other = rb_entry(other_node, struct extent_state, rb_node);
328 if (other->end == state->start - 1 &&
329 other->state == state->state) {
330 merge_cb(tree, state, other);
331 state->start = other->start;
333 rb_erase(&other->rb_node, &tree->state);
334 free_extent_state(other);
337 other_node = rb_next(&state->rb_node);
339 other = rb_entry(other_node, struct extent_state, rb_node);
340 if (other->start == state->end + 1 &&
341 other->state == state->state) {
342 merge_cb(tree, state, other);
343 state->end = other->end;
345 rb_erase(&other->rb_node, &tree->state);
346 free_extent_state(other);
351 static void set_state_cb(struct extent_io_tree *tree,
352 struct extent_state *state, unsigned long *bits)
354 if (tree->ops && tree->ops->set_bit_hook)
355 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
358 static void clear_state_cb(struct extent_io_tree *tree,
359 struct extent_state *state, unsigned long *bits)
361 if (tree->ops && tree->ops->clear_bit_hook)
362 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
365 static void set_state_bits(struct extent_io_tree *tree,
366 struct extent_state *state, unsigned long *bits);
369 * insert an extent_state struct into the tree. 'bits' are set on the
370 * struct before it is inserted.
372 * This may return -EEXIST if the extent is already there, in which case the
373 * state struct is freed.
375 * The tree lock is not taken internally. This is a utility function and
376 * probably isn't what you want to call (see set/clear_extent_bit).
378 static int insert_state(struct extent_io_tree *tree,
379 struct extent_state *state, u64 start, u64 end,
382 struct rb_node *node;
385 WARN(1, KERN_ERR "btrfs end < start %llu %llu\n",
387 state->start = start;
390 set_state_bits(tree, state, bits);
392 node = tree_insert(&tree->state, end, &state->rb_node);
394 struct extent_state *found;
395 found = rb_entry(node, struct extent_state, rb_node);
396 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
398 found->start, found->end, start, end);
402 merge_state(tree, state);
406 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
409 if (tree->ops && tree->ops->split_extent_hook)
410 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
414 * split a given extent state struct in two, inserting the preallocated
415 * struct 'prealloc' as the newly created second half. 'split' indicates an
416 * offset inside 'orig' where it should be split.
419 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
420 * are two extent state structs in the tree:
421 * prealloc: [orig->start, split - 1]
422 * orig: [ split, orig->end ]
424 * The tree locks are not taken by this function. They need to be held
427 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
428 struct extent_state *prealloc, u64 split)
430 struct rb_node *node;
432 split_cb(tree, orig, split);
434 prealloc->start = orig->start;
435 prealloc->end = split - 1;
436 prealloc->state = orig->state;
439 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
441 free_extent_state(prealloc);
444 prealloc->tree = tree;
448 static struct extent_state *next_state(struct extent_state *state)
450 struct rb_node *next = rb_next(&state->rb_node);
452 return rb_entry(next, struct extent_state, rb_node);
458 * utility function to clear some bits in an extent state struct.
459 * it will optionally wake up any one waiting on this state (wake == 1).
461 * If no bits are set on the state struct after clearing things, the
462 * struct is freed and removed from the tree
464 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
465 struct extent_state *state,
466 unsigned long *bits, int wake)
468 struct extent_state *next;
469 unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
471 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
472 u64 range = state->end - state->start + 1;
473 WARN_ON(range > tree->dirty_bytes);
474 tree->dirty_bytes -= range;
476 clear_state_cb(tree, state, bits);
477 state->state &= ~bits_to_clear;
480 if (state->state == 0) {
481 next = next_state(state);
483 rb_erase(&state->rb_node, &tree->state);
485 free_extent_state(state);
490 merge_state(tree, state);
491 next = next_state(state);
496 static struct extent_state *
497 alloc_extent_state_atomic(struct extent_state *prealloc)
500 prealloc = alloc_extent_state(GFP_ATOMIC);
505 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
507 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
508 "Extent tree was modified by another "
509 "thread while locked.");
513 * clear some bits on a range in the tree. This may require splitting
514 * or inserting elements in the tree, so the gfp mask is used to
515 * indicate which allocations or sleeping are allowed.
517 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
518 * the given range from the tree regardless of state (ie for truncate).
520 * the range [start, end] is inclusive.
522 * This takes the tree lock, and returns 0 on success and < 0 on error.
524 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
525 unsigned long bits, int wake, int delete,
526 struct extent_state **cached_state,
529 struct extent_state *state;
530 struct extent_state *cached;
531 struct extent_state *prealloc = NULL;
532 struct rb_node *node;
537 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
539 if (bits & EXTENT_DELALLOC)
540 bits |= EXTENT_NORESERVE;
543 bits |= ~EXTENT_CTLBITS;
544 bits |= EXTENT_FIRST_DELALLOC;
546 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
549 if (!prealloc && (mask & __GFP_WAIT)) {
550 prealloc = alloc_extent_state(mask);
555 spin_lock(&tree->lock);
557 cached = *cached_state;
560 *cached_state = NULL;
564 if (cached && cached->tree && cached->start <= start &&
565 cached->end > start) {
567 atomic_dec(&cached->refs);
572 free_extent_state(cached);
575 * this search will find the extents that end after
578 node = tree_search(tree, start);
581 state = rb_entry(node, struct extent_state, rb_node);
583 if (state->start > end)
585 WARN_ON(state->end < start);
586 last_end = state->end;
588 /* the state doesn't have the wanted bits, go ahead */
589 if (!(state->state & bits)) {
590 state = next_state(state);
595 * | ---- desired range ---- |
597 * | ------------- state -------------- |
599 * We need to split the extent we found, and may flip
600 * bits on second half.
602 * If the extent we found extends past our range, we
603 * just split and search again. It'll get split again
604 * the next time though.
606 * If the extent we found is inside our range, we clear
607 * the desired bit on it.
610 if (state->start < start) {
611 prealloc = alloc_extent_state_atomic(prealloc);
613 err = split_state(tree, state, prealloc, start);
615 extent_io_tree_panic(tree, err);
620 if (state->end <= end) {
621 state = clear_state_bit(tree, state, &bits, wake);
627 * | ---- desired range ---- |
629 * We need to split the extent, and clear the bit
632 if (state->start <= end && state->end > end) {
633 prealloc = alloc_extent_state_atomic(prealloc);
635 err = split_state(tree, state, prealloc, end + 1);
637 extent_io_tree_panic(tree, err);
642 clear_state_bit(tree, prealloc, &bits, wake);
648 state = clear_state_bit(tree, state, &bits, wake);
650 if (last_end == (u64)-1)
652 start = last_end + 1;
653 if (start <= end && state && !need_resched())
658 spin_unlock(&tree->lock);
660 free_extent_state(prealloc);
667 spin_unlock(&tree->lock);
668 if (mask & __GFP_WAIT)
673 static void wait_on_state(struct extent_io_tree *tree,
674 struct extent_state *state)
675 __releases(tree->lock)
676 __acquires(tree->lock)
679 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
680 spin_unlock(&tree->lock);
682 spin_lock(&tree->lock);
683 finish_wait(&state->wq, &wait);
687 * waits for one or more bits to clear on a range in the state tree.
688 * The range [start, end] is inclusive.
689 * The tree lock is taken by this function
691 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
694 struct extent_state *state;
695 struct rb_node *node;
697 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
699 spin_lock(&tree->lock);
703 * this search will find all the extents that end after
706 node = tree_search(tree, start);
710 state = rb_entry(node, struct extent_state, rb_node);
712 if (state->start > end)
715 if (state->state & bits) {
716 start = state->start;
717 atomic_inc(&state->refs);
718 wait_on_state(tree, state);
719 free_extent_state(state);
722 start = state->end + 1;
727 cond_resched_lock(&tree->lock);
730 spin_unlock(&tree->lock);
733 static void set_state_bits(struct extent_io_tree *tree,
734 struct extent_state *state,
737 unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
739 set_state_cb(tree, state, bits);
740 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
741 u64 range = state->end - state->start + 1;
742 tree->dirty_bytes += range;
744 state->state |= bits_to_set;
747 static void cache_state(struct extent_state *state,
748 struct extent_state **cached_ptr)
750 if (cached_ptr && !(*cached_ptr)) {
751 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
753 atomic_inc(&state->refs);
759 * set some bits on a range in the tree. This may require allocations or
760 * sleeping, so the gfp mask is used to indicate what is allowed.
762 * If any of the exclusive bits are set, this will fail with -EEXIST if some
763 * part of the range already has the desired bits set. The start of the
764 * existing range is returned in failed_start in this case.
766 * [start, end] is inclusive This takes the tree lock.
769 static int __must_check
770 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
771 unsigned long bits, unsigned long exclusive_bits,
772 u64 *failed_start, struct extent_state **cached_state,
775 struct extent_state *state;
776 struct extent_state *prealloc = NULL;
777 struct rb_node *node;
782 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
784 bits |= EXTENT_FIRST_DELALLOC;
786 if (!prealloc && (mask & __GFP_WAIT)) {
787 prealloc = alloc_extent_state(mask);
791 spin_lock(&tree->lock);
792 if (cached_state && *cached_state) {
793 state = *cached_state;
794 if (state->start <= start && state->end > start &&
796 node = &state->rb_node;
801 * this search will find all the extents that end after
804 node = tree_search(tree, start);
806 prealloc = alloc_extent_state_atomic(prealloc);
808 err = insert_state(tree, prealloc, start, end, &bits);
810 extent_io_tree_panic(tree, err);
815 state = rb_entry(node, struct extent_state, rb_node);
817 last_start = state->start;
818 last_end = state->end;
821 * | ---- desired range ---- |
824 * Just lock what we found and keep going
826 if (state->start == start && state->end <= end) {
827 if (state->state & exclusive_bits) {
828 *failed_start = state->start;
833 set_state_bits(tree, state, &bits);
834 cache_state(state, cached_state);
835 merge_state(tree, state);
836 if (last_end == (u64)-1)
838 start = last_end + 1;
839 state = next_state(state);
840 if (start < end && state && state->start == start &&
847 * | ---- desired range ---- |
850 * | ------------- state -------------- |
852 * We need to split the extent we found, and may flip bits on
855 * If the extent we found extends past our
856 * range, we just split and search again. It'll get split
857 * again the next time though.
859 * If the extent we found is inside our range, we set the
862 if (state->start < start) {
863 if (state->state & exclusive_bits) {
864 *failed_start = start;
869 prealloc = alloc_extent_state_atomic(prealloc);
871 err = split_state(tree, state, prealloc, start);
873 extent_io_tree_panic(tree, err);
878 if (state->end <= end) {
879 set_state_bits(tree, state, &bits);
880 cache_state(state, cached_state);
881 merge_state(tree, state);
882 if (last_end == (u64)-1)
884 start = last_end + 1;
885 state = next_state(state);
886 if (start < end && state && state->start == start &&
893 * | ---- desired range ---- |
894 * | state | or | state |
896 * There's a hole, we need to insert something in it and
897 * ignore the extent we found.
899 if (state->start > start) {
901 if (end < last_start)
904 this_end = last_start - 1;
906 prealloc = alloc_extent_state_atomic(prealloc);
910 * Avoid to free 'prealloc' if it can be merged with
913 err = insert_state(tree, prealloc, start, this_end,
916 extent_io_tree_panic(tree, err);
918 cache_state(prealloc, cached_state);
920 start = this_end + 1;
924 * | ---- desired range ---- |
926 * We need to split the extent, and set the bit
929 if (state->start <= end && state->end > end) {
930 if (state->state & exclusive_bits) {
931 *failed_start = start;
936 prealloc = alloc_extent_state_atomic(prealloc);
938 err = split_state(tree, state, prealloc, end + 1);
940 extent_io_tree_panic(tree, err);
942 set_state_bits(tree, prealloc, &bits);
943 cache_state(prealloc, cached_state);
944 merge_state(tree, prealloc);
952 spin_unlock(&tree->lock);
954 free_extent_state(prealloc);
961 spin_unlock(&tree->lock);
962 if (mask & __GFP_WAIT)
967 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
968 unsigned long bits, u64 * failed_start,
969 struct extent_state **cached_state, gfp_t mask)
971 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
977 * convert_extent_bit - convert all bits in a given range from one bit to
979 * @tree: the io tree to search
980 * @start: the start offset in bytes
981 * @end: the end offset in bytes (inclusive)
982 * @bits: the bits to set in this range
983 * @clear_bits: the bits to clear in this range
984 * @cached_state: state that we're going to cache
985 * @mask: the allocation mask
987 * This will go through and set bits for the given range. If any states exist
988 * already in this range they are set with the given bit and cleared of the
989 * clear_bits. This is only meant to be used by things that are mergeable, ie
990 * converting from say DELALLOC to DIRTY. This is not meant to be used with
991 * boundary bits like LOCK.
993 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
994 unsigned long bits, unsigned long clear_bits,
995 struct extent_state **cached_state, gfp_t mask)
997 struct extent_state *state;
998 struct extent_state *prealloc = NULL;
999 struct rb_node *node;
1004 btrfs_debug_check_extent_io_range(tree->mapping->host, start, end);
1007 if (!prealloc && (mask & __GFP_WAIT)) {
1008 prealloc = alloc_extent_state(mask);
1013 spin_lock(&tree->lock);
1014 if (cached_state && *cached_state) {
1015 state = *cached_state;
1016 if (state->start <= start && state->end > start &&
1018 node = &state->rb_node;
1024 * this search will find all the extents that end after
1027 node = tree_search(tree, start);
1029 prealloc = alloc_extent_state_atomic(prealloc);
1034 err = insert_state(tree, prealloc, start, end, &bits);
1037 extent_io_tree_panic(tree, err);
1040 state = rb_entry(node, struct extent_state, rb_node);
1042 last_start = state->start;
1043 last_end = state->end;
1046 * | ---- desired range ---- |
1049 * Just lock what we found and keep going
1051 if (state->start == start && state->end <= end) {
1052 set_state_bits(tree, state, &bits);
1053 cache_state(state, cached_state);
1054 state = clear_state_bit(tree, state, &clear_bits, 0);
1055 if (last_end == (u64)-1)
1057 start = last_end + 1;
1058 if (start < end && state && state->start == start &&
1065 * | ---- desired range ---- |
1068 * | ------------- state -------------- |
1070 * We need to split the extent we found, and may flip bits on
1073 * If the extent we found extends past our
1074 * range, we just split and search again. It'll get split
1075 * again the next time though.
1077 * If the extent we found is inside our range, we set the
1078 * desired bit on it.
1080 if (state->start < start) {
1081 prealloc = alloc_extent_state_atomic(prealloc);
1086 err = split_state(tree, state, prealloc, start);
1088 extent_io_tree_panic(tree, err);
1092 if (state->end <= end) {
1093 set_state_bits(tree, state, &bits);
1094 cache_state(state, cached_state);
1095 state = clear_state_bit(tree, state, &clear_bits, 0);
1096 if (last_end == (u64)-1)
1098 start = last_end + 1;
1099 if (start < end && state && state->start == start &&
1106 * | ---- desired range ---- |
1107 * | state | or | state |
1109 * There's a hole, we need to insert something in it and
1110 * ignore the extent we found.
1112 if (state->start > start) {
1114 if (end < last_start)
1117 this_end = last_start - 1;
1119 prealloc = alloc_extent_state_atomic(prealloc);
1126 * Avoid to free 'prealloc' if it can be merged with
1129 err = insert_state(tree, prealloc, start, this_end,
1132 extent_io_tree_panic(tree, err);
1133 cache_state(prealloc, cached_state);
1135 start = this_end + 1;
1139 * | ---- desired range ---- |
1141 * We need to split the extent, and set the bit
1144 if (state->start <= end && state->end > end) {
1145 prealloc = alloc_extent_state_atomic(prealloc);
1151 err = split_state(tree, state, prealloc, end + 1);
1153 extent_io_tree_panic(tree, err);
1155 set_state_bits(tree, prealloc, &bits);
1156 cache_state(prealloc, cached_state);
1157 clear_state_bit(tree, prealloc, &clear_bits, 0);
1165 spin_unlock(&tree->lock);
1167 free_extent_state(prealloc);
1174 spin_unlock(&tree->lock);
1175 if (mask & __GFP_WAIT)
1180 /* wrappers around set/clear extent bit */
1181 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1184 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1188 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1189 unsigned long bits, gfp_t mask)
1191 return set_extent_bit(tree, start, end, bits, NULL,
1195 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1196 unsigned long bits, gfp_t mask)
1198 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1201 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1202 struct extent_state **cached_state, gfp_t mask)
1204 return set_extent_bit(tree, start, end,
1205 EXTENT_DELALLOC | EXTENT_UPTODATE,
1206 NULL, cached_state, mask);
1209 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1210 struct extent_state **cached_state, gfp_t mask)
1212 return set_extent_bit(tree, start, end,
1213 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1214 NULL, cached_state, mask);
1217 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1220 return clear_extent_bit(tree, start, end,
1221 EXTENT_DIRTY | EXTENT_DELALLOC |
1222 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1225 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1228 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1232 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1233 struct extent_state **cached_state, gfp_t mask)
1235 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1236 cached_state, mask);
1239 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1240 struct extent_state **cached_state, gfp_t mask)
1242 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1243 cached_state, mask);
1247 * either insert or lock state struct between start and end use mask to tell
1248 * us if waiting is desired.
1250 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1251 unsigned long bits, struct extent_state **cached_state)
1256 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1257 EXTENT_LOCKED, &failed_start,
1258 cached_state, GFP_NOFS);
1259 if (err == -EEXIST) {
1260 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1261 start = failed_start;
1264 WARN_ON(start > end);
1269 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1271 return lock_extent_bits(tree, start, end, 0, NULL);
1274 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1279 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1280 &failed_start, NULL, GFP_NOFS);
1281 if (err == -EEXIST) {
1282 if (failed_start > start)
1283 clear_extent_bit(tree, start, failed_start - 1,
1284 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1290 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1291 struct extent_state **cached, gfp_t mask)
1293 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1297 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1299 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1303 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1305 unsigned long index = start >> PAGE_CACHE_SHIFT;
1306 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1309 while (index <= end_index) {
1310 page = find_get_page(inode->i_mapping, index);
1311 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1312 clear_page_dirty_for_io(page);
1313 page_cache_release(page);
1319 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1321 unsigned long index = start >> PAGE_CACHE_SHIFT;
1322 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1325 while (index <= end_index) {
1326 page = find_get_page(inode->i_mapping, index);
1327 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1328 account_page_redirty(page);
1329 __set_page_dirty_nobuffers(page);
1330 page_cache_release(page);
1337 * helper function to set both pages and extents in the tree writeback
1339 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1341 unsigned long index = start >> PAGE_CACHE_SHIFT;
1342 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1345 while (index <= end_index) {
1346 page = find_get_page(tree->mapping, index);
1347 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1348 set_page_writeback(page);
1349 page_cache_release(page);
1355 /* find the first state struct with 'bits' set after 'start', and
1356 * return it. tree->lock must be held. NULL will returned if
1357 * nothing was found after 'start'
1359 static struct extent_state *
1360 find_first_extent_bit_state(struct extent_io_tree *tree,
1361 u64 start, unsigned long bits)
1363 struct rb_node *node;
1364 struct extent_state *state;
1367 * this search will find all the extents that end after
1370 node = tree_search(tree, start);
1375 state = rb_entry(node, struct extent_state, rb_node);
1376 if (state->end >= start && (state->state & bits))
1379 node = rb_next(node);
1388 * find the first offset in the io tree with 'bits' set. zero is
1389 * returned if we find something, and *start_ret and *end_ret are
1390 * set to reflect the state struct that was found.
1392 * If nothing was found, 1 is returned. If found something, return 0.
1394 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1395 u64 *start_ret, u64 *end_ret, unsigned long bits,
1396 struct extent_state **cached_state)
1398 struct extent_state *state;
1402 spin_lock(&tree->lock);
1403 if (cached_state && *cached_state) {
1404 state = *cached_state;
1405 if (state->end == start - 1 && state->tree) {
1406 n = rb_next(&state->rb_node);
1408 state = rb_entry(n, struct extent_state,
1410 if (state->state & bits)
1414 free_extent_state(*cached_state);
1415 *cached_state = NULL;
1418 free_extent_state(*cached_state);
1419 *cached_state = NULL;
1422 state = find_first_extent_bit_state(tree, start, bits);
1425 cache_state(state, cached_state);
1426 *start_ret = state->start;
1427 *end_ret = state->end;
1431 spin_unlock(&tree->lock);
1436 * find a contiguous range of bytes in the file marked as delalloc, not
1437 * more than 'max_bytes'. start and end are used to return the range,
1439 * 1 is returned if we find something, 0 if nothing was in the tree
1441 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1442 u64 *start, u64 *end, u64 max_bytes,
1443 struct extent_state **cached_state)
1445 struct rb_node *node;
1446 struct extent_state *state;
1447 u64 cur_start = *start;
1449 u64 total_bytes = 0;
1451 spin_lock(&tree->lock);
1454 * this search will find all the extents that end after
1457 node = tree_search(tree, cur_start);
1465 state = rb_entry(node, struct extent_state, rb_node);
1466 if (found && (state->start != cur_start ||
1467 (state->state & EXTENT_BOUNDARY))) {
1470 if (!(state->state & EXTENT_DELALLOC)) {
1476 *start = state->start;
1477 *cached_state = state;
1478 atomic_inc(&state->refs);
1482 cur_start = state->end + 1;
1483 node = rb_next(node);
1484 total_bytes += state->end - state->start + 1;
1485 if (total_bytes >= max_bytes) {
1486 *end = *start + max_bytes - 1;
1493 spin_unlock(&tree->lock);
1497 static noinline void __unlock_for_delalloc(struct inode *inode,
1498 struct page *locked_page,
1502 struct page *pages[16];
1503 unsigned long index = start >> PAGE_CACHE_SHIFT;
1504 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1505 unsigned long nr_pages = end_index - index + 1;
1508 if (index == locked_page->index && end_index == index)
1511 while (nr_pages > 0) {
1512 ret = find_get_pages_contig(inode->i_mapping, index,
1513 min_t(unsigned long, nr_pages,
1514 ARRAY_SIZE(pages)), pages);
1515 for (i = 0; i < ret; i++) {
1516 if (pages[i] != locked_page)
1517 unlock_page(pages[i]);
1518 page_cache_release(pages[i]);
1526 static noinline int lock_delalloc_pages(struct inode *inode,
1527 struct page *locked_page,
1531 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1532 unsigned long start_index = index;
1533 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1534 unsigned long pages_locked = 0;
1535 struct page *pages[16];
1536 unsigned long nrpages;
1540 /* the caller is responsible for locking the start index */
1541 if (index == locked_page->index && index == end_index)
1544 /* skip the page at the start index */
1545 nrpages = end_index - index + 1;
1546 while (nrpages > 0) {
1547 ret = find_get_pages_contig(inode->i_mapping, index,
1548 min_t(unsigned long,
1549 nrpages, ARRAY_SIZE(pages)), pages);
1554 /* now we have an array of pages, lock them all */
1555 for (i = 0; i < ret; i++) {
1557 * the caller is taking responsibility for
1560 if (pages[i] != locked_page) {
1561 lock_page(pages[i]);
1562 if (!PageDirty(pages[i]) ||
1563 pages[i]->mapping != inode->i_mapping) {
1565 unlock_page(pages[i]);
1566 page_cache_release(pages[i]);
1570 page_cache_release(pages[i]);
1579 if (ret && pages_locked) {
1580 __unlock_for_delalloc(inode, locked_page,
1582 ((u64)(start_index + pages_locked - 1)) <<
1589 * find a contiguous range of bytes in the file marked as delalloc, not
1590 * more than 'max_bytes'. start and end are used to return the range,
1592 * 1 is returned if we find something, 0 if nothing was in the tree
1594 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1595 struct extent_io_tree *tree,
1596 struct page *locked_page,
1597 u64 *start, u64 *end,
1603 struct extent_state *cached_state = NULL;
1608 /* step one, find a bunch of delalloc bytes starting at start */
1609 delalloc_start = *start;
1611 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1612 max_bytes, &cached_state);
1613 if (!found || delalloc_end <= *start) {
1614 *start = delalloc_start;
1615 *end = delalloc_end;
1616 free_extent_state(cached_state);
1621 * start comes from the offset of locked_page. We have to lock
1622 * pages in order, so we can't process delalloc bytes before
1625 if (delalloc_start < *start)
1626 delalloc_start = *start;
1629 * make sure to limit the number of pages we try to lock down
1632 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1633 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1635 /* step two, lock all the pages after the page that has start */
1636 ret = lock_delalloc_pages(inode, locked_page,
1637 delalloc_start, delalloc_end);
1638 if (ret == -EAGAIN) {
1639 /* some of the pages are gone, lets avoid looping by
1640 * shortening the size of the delalloc range we're searching
1642 free_extent_state(cached_state);
1644 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1645 max_bytes = PAGE_CACHE_SIZE - offset;
1653 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1655 /* step three, lock the state bits for the whole range */
1656 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1658 /* then test to make sure it is all still delalloc */
1659 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1660 EXTENT_DELALLOC, 1, cached_state);
1662 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1663 &cached_state, GFP_NOFS);
1664 __unlock_for_delalloc(inode, locked_page,
1665 delalloc_start, delalloc_end);
1669 free_extent_state(cached_state);
1670 *start = delalloc_start;
1671 *end = delalloc_end;
1676 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1677 struct page *locked_page,
1678 unsigned long clear_bits,
1679 unsigned long page_ops)
1681 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1683 struct page *pages[16];
1684 unsigned long index = start >> PAGE_CACHE_SHIFT;
1685 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1686 unsigned long nr_pages = end_index - index + 1;
1689 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1693 while (nr_pages > 0) {
1694 ret = find_get_pages_contig(inode->i_mapping, index,
1695 min_t(unsigned long,
1696 nr_pages, ARRAY_SIZE(pages)), pages);
1697 for (i = 0; i < ret; i++) {
1699 if (page_ops & PAGE_SET_PRIVATE2)
1700 SetPagePrivate2(pages[i]);
1702 if (pages[i] == locked_page) {
1703 page_cache_release(pages[i]);
1706 if (page_ops & PAGE_CLEAR_DIRTY)
1707 clear_page_dirty_for_io(pages[i]);
1708 if (page_ops & PAGE_SET_WRITEBACK)
1709 set_page_writeback(pages[i]);
1710 if (page_ops & PAGE_END_WRITEBACK)
1711 end_page_writeback(pages[i]);
1712 if (page_ops & PAGE_UNLOCK)
1713 unlock_page(pages[i]);
1714 page_cache_release(pages[i]);
1724 * count the number of bytes in the tree that have a given bit(s)
1725 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1726 * cached. The total number found is returned.
1728 u64 count_range_bits(struct extent_io_tree *tree,
1729 u64 *start, u64 search_end, u64 max_bytes,
1730 unsigned long bits, int contig)
1732 struct rb_node *node;
1733 struct extent_state *state;
1734 u64 cur_start = *start;
1735 u64 total_bytes = 0;
1739 if (search_end <= cur_start) {
1744 spin_lock(&tree->lock);
1745 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1746 total_bytes = tree->dirty_bytes;
1750 * this search will find all the extents that end after
1753 node = tree_search(tree, cur_start);
1758 state = rb_entry(node, struct extent_state, rb_node);
1759 if (state->start > search_end)
1761 if (contig && found && state->start > last + 1)
1763 if (state->end >= cur_start && (state->state & bits) == bits) {
1764 total_bytes += min(search_end, state->end) + 1 -
1765 max(cur_start, state->start);
1766 if (total_bytes >= max_bytes)
1769 *start = max(cur_start, state->start);
1773 } else if (contig && found) {
1776 node = rb_next(node);
1781 spin_unlock(&tree->lock);
1786 * set the private field for a given byte offset in the tree. If there isn't
1787 * an extent_state there already, this does nothing.
1789 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1791 struct rb_node *node;
1792 struct extent_state *state;
1795 spin_lock(&tree->lock);
1797 * this search will find all the extents that end after
1800 node = tree_search(tree, start);
1805 state = rb_entry(node, struct extent_state, rb_node);
1806 if (state->start != start) {
1810 state->private = private;
1812 spin_unlock(&tree->lock);
1816 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1818 struct rb_node *node;
1819 struct extent_state *state;
1822 spin_lock(&tree->lock);
1824 * this search will find all the extents that end after
1827 node = tree_search(tree, start);
1832 state = rb_entry(node, struct extent_state, rb_node);
1833 if (state->start != start) {
1837 *private = state->private;
1839 spin_unlock(&tree->lock);
1844 * searches a range in the state tree for a given mask.
1845 * If 'filled' == 1, this returns 1 only if every extent in the tree
1846 * has the bits set. Otherwise, 1 is returned if any bit in the
1847 * range is found set.
1849 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1850 unsigned long bits, int filled, struct extent_state *cached)
1852 struct extent_state *state = NULL;
1853 struct rb_node *node;
1856 spin_lock(&tree->lock);
1857 if (cached && cached->tree && cached->start <= start &&
1858 cached->end > start)
1859 node = &cached->rb_node;
1861 node = tree_search(tree, start);
1862 while (node && start <= end) {
1863 state = rb_entry(node, struct extent_state, rb_node);
1865 if (filled && state->start > start) {
1870 if (state->start > end)
1873 if (state->state & bits) {
1877 } else if (filled) {
1882 if (state->end == (u64)-1)
1885 start = state->end + 1;
1888 node = rb_next(node);
1895 spin_unlock(&tree->lock);
1900 * helper function to set a given page up to date if all the
1901 * extents in the tree for that page are up to date
1903 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1905 u64 start = page_offset(page);
1906 u64 end = start + PAGE_CACHE_SIZE - 1;
1907 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1908 SetPageUptodate(page);
1912 * When IO fails, either with EIO or csum verification fails, we
1913 * try other mirrors that might have a good copy of the data. This
1914 * io_failure_record is used to record state as we go through all the
1915 * mirrors. If another mirror has good data, the page is set up to date
1916 * and things continue. If a good mirror can't be found, the original
1917 * bio end_io callback is called to indicate things have failed.
1919 struct io_failure_record {
1924 unsigned long bio_flags;
1930 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1935 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1937 set_state_private(failure_tree, rec->start, 0);
1938 ret = clear_extent_bits(failure_tree, rec->start,
1939 rec->start + rec->len - 1,
1940 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1944 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1945 rec->start + rec->len - 1,
1946 EXTENT_DAMAGED, GFP_NOFS);
1954 static void repair_io_failure_callback(struct bio *bio, int err)
1956 complete(bio->bi_private);
1960 * this bypasses the standard btrfs submit functions deliberately, as
1961 * the standard behavior is to write all copies in a raid setup. here we only
1962 * want to write the one bad copy. so we do the mapping for ourselves and issue
1963 * submit_bio directly.
1964 * to avoid any synchronization issues, wait for the data after writing, which
1965 * actually prevents the read that triggered the error from finishing.
1966 * currently, there can be no more than two copies of every data bit. thus,
1967 * exactly one rewrite is required.
1969 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
1970 u64 length, u64 logical, struct page *page,
1974 struct btrfs_device *dev;
1975 DECLARE_COMPLETION_ONSTACK(compl);
1978 struct btrfs_bio *bbio = NULL;
1979 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1982 BUG_ON(!mirror_num);
1984 /* we can't repair anything in raid56 yet */
1985 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
1988 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1991 bio->bi_private = &compl;
1992 bio->bi_end_io = repair_io_failure_callback;
1994 map_length = length;
1996 ret = btrfs_map_block(fs_info, WRITE, logical,
1997 &map_length, &bbio, mirror_num);
2002 BUG_ON(mirror_num != bbio->mirror_num);
2003 sector = bbio->stripes[mirror_num-1].physical >> 9;
2004 bio->bi_sector = sector;
2005 dev = bbio->stripes[mirror_num-1].dev;
2007 if (!dev || !dev->bdev || !dev->writeable) {
2011 bio->bi_bdev = dev->bdev;
2012 bio_add_page(bio, page, length, start - page_offset(page));
2013 btrfsic_submit_bio(WRITE_SYNC, bio);
2014 wait_for_completion(&compl);
2016 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2017 /* try to remap that extent elsewhere? */
2019 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2023 printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
2024 "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2025 start, rcu_str_deref(dev->name), sector);
2031 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2034 u64 start = eb->start;
2035 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2038 for (i = 0; i < num_pages; i++) {
2039 struct page *p = extent_buffer_page(eb, i);
2040 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2041 start, p, mirror_num);
2044 start += PAGE_CACHE_SIZE;
2051 * each time an IO finishes, we do a fast check in the IO failure tree
2052 * to see if we need to process or clean up an io_failure_record
2054 static int clean_io_failure(u64 start, struct page *page)
2057 u64 private_failure;
2058 struct io_failure_record *failrec;
2059 struct btrfs_fs_info *fs_info;
2060 struct extent_state *state;
2064 struct inode *inode = page->mapping->host;
2067 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2068 (u64)-1, 1, EXTENT_DIRTY, 0);
2072 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2077 failrec = (struct io_failure_record *)(unsigned long) private_failure;
2078 BUG_ON(!failrec->this_mirror);
2080 if (failrec->in_validation) {
2081 /* there was no real error, just free the record */
2082 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2088 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2089 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2092 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2094 if (state && state->start <= failrec->start &&
2095 state->end >= failrec->start + failrec->len - 1) {
2096 fs_info = BTRFS_I(inode)->root->fs_info;
2097 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2099 if (num_copies > 1) {
2100 ret = repair_io_failure(fs_info, start, failrec->len,
2101 failrec->logical, page,
2102 failrec->failed_mirror);
2110 ret = free_io_failure(inode, failrec, did_repair);
2116 * this is a generic handler for readpage errors (default
2117 * readpage_io_failed_hook). if other copies exist, read those and write back
2118 * good data to the failed position. does not investigate in remapping the
2119 * failed extent elsewhere, hoping the device will be smart enough to do this as
2123 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2124 struct page *page, u64 start, u64 end,
2127 struct io_failure_record *failrec = NULL;
2129 struct extent_map *em;
2130 struct inode *inode = page->mapping->host;
2131 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2132 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2133 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2135 struct btrfs_io_bio *btrfs_failed_bio;
2136 struct btrfs_io_bio *btrfs_bio;
2142 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2144 ret = get_state_private(failure_tree, start, &private);
2146 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2149 failrec->start = start;
2150 failrec->len = end - start + 1;
2151 failrec->this_mirror = 0;
2152 failrec->bio_flags = 0;
2153 failrec->in_validation = 0;
2155 read_lock(&em_tree->lock);
2156 em = lookup_extent_mapping(em_tree, start, failrec->len);
2158 read_unlock(&em_tree->lock);
2163 if (em->start > start || em->start + em->len < start) {
2164 free_extent_map(em);
2167 read_unlock(&em_tree->lock);
2173 logical = start - em->start;
2174 logical = em->block_start + logical;
2175 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2176 logical = em->block_start;
2177 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2178 extent_set_compress_type(&failrec->bio_flags,
2181 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2182 "len=%llu\n", logical, start, failrec->len);
2183 failrec->logical = logical;
2184 free_extent_map(em);
2186 /* set the bits in the private failure tree */
2187 ret = set_extent_bits(failure_tree, start, end,
2188 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2190 ret = set_state_private(failure_tree, start,
2191 (u64)(unsigned long)failrec);
2192 /* set the bits in the inode's tree */
2194 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2201 failrec = (struct io_failure_record *)(unsigned long)private;
2202 pr_debug("bio_readpage_error: (found) logical=%llu, "
2203 "start=%llu, len=%llu, validation=%d\n",
2204 failrec->logical, failrec->start, failrec->len,
2205 failrec->in_validation);
2207 * when data can be on disk more than twice, add to failrec here
2208 * (e.g. with a list for failed_mirror) to make
2209 * clean_io_failure() clean all those errors at once.
2212 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2213 failrec->logical, failrec->len);
2214 if (num_copies == 1) {
2216 * we only have a single copy of the data, so don't bother with
2217 * all the retry and error correction code that follows. no
2218 * matter what the error is, it is very likely to persist.
2220 pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2221 num_copies, failrec->this_mirror, failed_mirror);
2222 free_io_failure(inode, failrec, 0);
2227 * there are two premises:
2228 * a) deliver good data to the caller
2229 * b) correct the bad sectors on disk
2231 if (failed_bio->bi_vcnt > 1) {
2233 * to fulfill b), we need to know the exact failing sectors, as
2234 * we don't want to rewrite any more than the failed ones. thus,
2235 * we need separate read requests for the failed bio
2237 * if the following BUG_ON triggers, our validation request got
2238 * merged. we need separate requests for our algorithm to work.
2240 BUG_ON(failrec->in_validation);
2241 failrec->in_validation = 1;
2242 failrec->this_mirror = failed_mirror;
2243 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2246 * we're ready to fulfill a) and b) alongside. get a good copy
2247 * of the failed sector and if we succeed, we have setup
2248 * everything for repair_io_failure to do the rest for us.
2250 if (failrec->in_validation) {
2251 BUG_ON(failrec->this_mirror != failed_mirror);
2252 failrec->in_validation = 0;
2253 failrec->this_mirror = 0;
2255 failrec->failed_mirror = failed_mirror;
2256 failrec->this_mirror++;
2257 if (failrec->this_mirror == failed_mirror)
2258 failrec->this_mirror++;
2259 read_mode = READ_SYNC;
2262 if (failrec->this_mirror > num_copies) {
2263 pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2264 num_copies, failrec->this_mirror, failed_mirror);
2265 free_io_failure(inode, failrec, 0);
2269 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2271 free_io_failure(inode, failrec, 0);
2274 bio->bi_end_io = failed_bio->bi_end_io;
2275 bio->bi_sector = failrec->logical >> 9;
2276 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2279 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2280 if (btrfs_failed_bio->csum) {
2281 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2282 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2284 btrfs_bio = btrfs_io_bio(bio);
2285 btrfs_bio->csum = btrfs_bio->csum_inline;
2286 phy_offset >>= inode->i_sb->s_blocksize_bits;
2287 phy_offset *= csum_size;
2288 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset,
2292 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2294 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2295 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2296 failrec->this_mirror, num_copies, failrec->in_validation);
2298 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2299 failrec->this_mirror,
2300 failrec->bio_flags, 0);
2304 /* lots and lots of room for performance fixes in the end_bio funcs */
2306 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2308 int uptodate = (err == 0);
2309 struct extent_io_tree *tree;
2312 tree = &BTRFS_I(page->mapping->host)->io_tree;
2314 if (tree->ops && tree->ops->writepage_end_io_hook) {
2315 ret = tree->ops->writepage_end_io_hook(page, start,
2316 end, NULL, uptodate);
2322 ClearPageUptodate(page);
2329 * after a writepage IO is done, we need to:
2330 * clear the uptodate bits on error
2331 * clear the writeback bits in the extent tree for this IO
2332 * end_page_writeback if the page has no more pending IO
2334 * Scheduling is not allowed, so the extent state tree is expected
2335 * to have one and only one object corresponding to this IO.
2337 static void end_bio_extent_writepage(struct bio *bio, int err)
2339 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2340 struct extent_io_tree *tree;
2345 struct page *page = bvec->bv_page;
2346 tree = &BTRFS_I(page->mapping->host)->io_tree;
2348 /* We always issue full-page reads, but if some block
2349 * in a page fails to read, blk_update_request() will
2350 * advance bv_offset and adjust bv_len to compensate.
2351 * Print a warning for nonzero offsets, and an error
2352 * if they don't add up to a full page. */
2353 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2354 printk("%s page write in btrfs with offset %u and length %u\n",
2355 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2356 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2357 bvec->bv_offset, bvec->bv_len);
2359 start = page_offset(page);
2360 end = start + bvec->bv_offset + bvec->bv_len - 1;
2362 if (--bvec >= bio->bi_io_vec)
2363 prefetchw(&bvec->bv_page->flags);
2365 if (end_extent_writepage(page, err, start, end))
2368 end_page_writeback(page);
2369 } while (bvec >= bio->bi_io_vec);
2375 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2378 struct extent_state *cached = NULL;
2379 u64 end = start + len - 1;
2381 if (uptodate && tree->track_uptodate)
2382 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2383 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2387 * after a readpage IO is done, we need to:
2388 * clear the uptodate bits on error
2389 * set the uptodate bits if things worked
2390 * set the page up to date if all extents in the tree are uptodate
2391 * clear the lock bit in the extent tree
2392 * unlock the page if there are no other extents locked for it
2394 * Scheduling is not allowed, so the extent state tree is expected
2395 * to have one and only one object corresponding to this IO.
2397 static void end_bio_extent_readpage(struct bio *bio, int err)
2399 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2400 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2401 struct bio_vec *bvec = bio->bi_io_vec;
2402 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2403 struct extent_io_tree *tree;
2408 u64 extent_start = 0;
2417 struct page *page = bvec->bv_page;
2418 struct inode *inode = page->mapping->host;
2420 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2421 "mirror=%lu\n", (u64)bio->bi_sector, err,
2422 io_bio->mirror_num);
2423 tree = &BTRFS_I(inode)->io_tree;
2425 /* We always issue full-page reads, but if some block
2426 * in a page fails to read, blk_update_request() will
2427 * advance bv_offset and adjust bv_len to compensate.
2428 * Print a warning for nonzero offsets, and an error
2429 * if they don't add up to a full page. */
2430 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2431 printk("%s page read in btrfs with offset %u and length %u\n",
2432 bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2433 ? KERN_ERR "partial" : KERN_INFO "incomplete",
2434 bvec->bv_offset, bvec->bv_len);
2436 start = page_offset(page);
2437 end = start + bvec->bv_offset + bvec->bv_len - 1;
2440 if (++bvec <= bvec_end)
2441 prefetchw(&bvec->bv_page->flags);
2443 mirror = io_bio->mirror_num;
2444 if (likely(uptodate && tree->ops &&
2445 tree->ops->readpage_end_io_hook)) {
2446 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2452 clean_io_failure(start, page);
2455 if (likely(uptodate))
2458 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2459 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2461 test_bit(BIO_UPTODATE, &bio->bi_flags))
2465 * The generic bio_readpage_error handles errors the
2466 * following way: If possible, new read requests are
2467 * created and submitted and will end up in
2468 * end_bio_extent_readpage as well (if we're lucky, not
2469 * in the !uptodate case). In that case it returns 0 and
2470 * we just go on with the next page in our bio. If it
2471 * can't handle the error it will return -EIO and we
2472 * remain responsible for that page.
2474 ret = bio_readpage_error(bio, offset, page, start, end,
2478 test_bit(BIO_UPTODATE, &bio->bi_flags);
2485 if (likely(uptodate)) {
2486 loff_t i_size = i_size_read(inode);
2487 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2490 /* Zero out the end if this page straddles i_size */
2491 offset = i_size & (PAGE_CACHE_SIZE-1);
2492 if (page->index == end_index && offset)
2493 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2494 SetPageUptodate(page);
2496 ClearPageUptodate(page);
2502 if (unlikely(!uptodate)) {
2504 endio_readpage_release_extent(tree,
2510 endio_readpage_release_extent(tree, start,
2511 end - start + 1, 0);
2512 } else if (!extent_len) {
2513 extent_start = start;
2514 extent_len = end + 1 - start;
2515 } else if (extent_start + extent_len == start) {
2516 extent_len += end + 1 - start;
2518 endio_readpage_release_extent(tree, extent_start,
2519 extent_len, uptodate);
2520 extent_start = start;
2521 extent_len = end + 1 - start;
2523 } while (bvec <= bvec_end);
2526 endio_readpage_release_extent(tree, extent_start, extent_len,
2529 io_bio->end_io(io_bio, err);
2534 * this allocates from the btrfs_bioset. We're returning a bio right now
2535 * but you can call btrfs_io_bio for the appropriate container_of magic
2538 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2541 struct btrfs_io_bio *btrfs_bio;
2544 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2546 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2547 while (!bio && (nr_vecs /= 2)) {
2548 bio = bio_alloc_bioset(gfp_flags,
2549 nr_vecs, btrfs_bioset);
2555 bio->bi_bdev = bdev;
2556 bio->bi_sector = first_sector;
2557 btrfs_bio = btrfs_io_bio(bio);
2558 btrfs_bio->csum = NULL;
2559 btrfs_bio->csum_allocated = NULL;
2560 btrfs_bio->end_io = NULL;
2565 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2567 return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2571 /* this also allocates from the btrfs_bioset */
2572 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2574 struct btrfs_io_bio *btrfs_bio;
2577 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2579 btrfs_bio = btrfs_io_bio(bio);
2580 btrfs_bio->csum = NULL;
2581 btrfs_bio->csum_allocated = NULL;
2582 btrfs_bio->end_io = NULL;
2588 static int __must_check submit_one_bio(int rw, struct bio *bio,
2589 int mirror_num, unsigned long bio_flags)
2592 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2593 struct page *page = bvec->bv_page;
2594 struct extent_io_tree *tree = bio->bi_private;
2597 start = page_offset(page) + bvec->bv_offset;
2599 bio->bi_private = NULL;
2603 if (tree->ops && tree->ops->submit_bio_hook)
2604 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2605 mirror_num, bio_flags, start);
2607 btrfsic_submit_bio(rw, bio);
2609 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2615 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2616 unsigned long offset, size_t size, struct bio *bio,
2617 unsigned long bio_flags)
2620 if (tree->ops && tree->ops->merge_bio_hook)
2621 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2628 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2629 struct page *page, sector_t sector,
2630 size_t size, unsigned long offset,
2631 struct block_device *bdev,
2632 struct bio **bio_ret,
2633 unsigned long max_pages,
2634 bio_end_io_t end_io_func,
2636 unsigned long prev_bio_flags,
2637 unsigned long bio_flags)
2643 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2644 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2645 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2647 if (bio_ret && *bio_ret) {
2650 contig = bio->bi_sector == sector;
2652 contig = bio_end_sector(bio) == sector;
2654 if (prev_bio_flags != bio_flags || !contig ||
2655 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2656 bio_add_page(bio, page, page_size, offset) < page_size) {
2657 ret = submit_one_bio(rw, bio, mirror_num,
2666 if (this_compressed)
2669 nr = bio_get_nr_vecs(bdev);
2671 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2675 bio_add_page(bio, page, page_size, offset);
2676 bio->bi_end_io = end_io_func;
2677 bio->bi_private = tree;
2682 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2687 static void attach_extent_buffer_page(struct extent_buffer *eb,
2690 if (!PagePrivate(page)) {
2691 SetPagePrivate(page);
2692 page_cache_get(page);
2693 set_page_private(page, (unsigned long)eb);
2695 WARN_ON(page->private != (unsigned long)eb);
2699 void set_page_extent_mapped(struct page *page)
2701 if (!PagePrivate(page)) {
2702 SetPagePrivate(page);
2703 page_cache_get(page);
2704 set_page_private(page, EXTENT_PAGE_PRIVATE);
2708 static struct extent_map *
2709 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2710 u64 start, u64 len, get_extent_t *get_extent,
2711 struct extent_map **em_cached)
2713 struct extent_map *em;
2715 if (em_cached && *em_cached) {
2717 if (em->in_tree && start >= em->start &&
2718 start < extent_map_end(em)) {
2719 atomic_inc(&em->refs);
2723 free_extent_map(em);
2727 em = get_extent(inode, page, pg_offset, start, len, 0);
2728 if (em_cached && !IS_ERR_OR_NULL(em)) {
2730 atomic_inc(&em->refs);
2736 * basic readpage implementation. Locked extent state structs are inserted
2737 * into the tree that are removed when the IO is done (by the end_io
2739 * XXX JDM: This needs looking at to ensure proper page locking
2741 static int __do_readpage(struct extent_io_tree *tree,
2743 get_extent_t *get_extent,
2744 struct extent_map **em_cached,
2745 struct bio **bio, int mirror_num,
2746 unsigned long *bio_flags, int rw)
2748 struct inode *inode = page->mapping->host;
2749 u64 start = page_offset(page);
2750 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2754 u64 last_byte = i_size_read(inode);
2758 struct extent_map *em;
2759 struct block_device *bdev;
2762 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2763 size_t pg_offset = 0;
2765 size_t disk_io_size;
2766 size_t blocksize = inode->i_sb->s_blocksize;
2767 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2769 set_page_extent_mapped(page);
2772 if (!PageUptodate(page)) {
2773 if (cleancache_get_page(page) == 0) {
2774 BUG_ON(blocksize != PAGE_SIZE);
2775 unlock_extent(tree, start, end);
2780 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2782 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2785 iosize = PAGE_CACHE_SIZE - zero_offset;
2786 userpage = kmap_atomic(page);
2787 memset(userpage + zero_offset, 0, iosize);
2788 flush_dcache_page(page);
2789 kunmap_atomic(userpage);
2792 while (cur <= end) {
2793 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2795 if (cur >= last_byte) {
2797 struct extent_state *cached = NULL;
2799 iosize = PAGE_CACHE_SIZE - pg_offset;
2800 userpage = kmap_atomic(page);
2801 memset(userpage + pg_offset, 0, iosize);
2802 flush_dcache_page(page);
2803 kunmap_atomic(userpage);
2804 set_extent_uptodate(tree, cur, cur + iosize - 1,
2807 unlock_extent_cached(tree, cur,
2812 em = __get_extent_map(inode, page, pg_offset, cur,
2813 end - cur + 1, get_extent, em_cached);
2814 if (IS_ERR_OR_NULL(em)) {
2817 unlock_extent(tree, cur, end);
2820 extent_offset = cur - em->start;
2821 BUG_ON(extent_map_end(em) <= cur);
2824 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2825 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2826 extent_set_compress_type(&this_bio_flag,
2830 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2831 cur_end = min(extent_map_end(em) - 1, end);
2832 iosize = ALIGN(iosize, blocksize);
2833 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2834 disk_io_size = em->block_len;
2835 sector = em->block_start >> 9;
2837 sector = (em->block_start + extent_offset) >> 9;
2838 disk_io_size = iosize;
2841 block_start = em->block_start;
2842 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2843 block_start = EXTENT_MAP_HOLE;
2844 free_extent_map(em);
2847 /* we've found a hole, just zero and go on */
2848 if (block_start == EXTENT_MAP_HOLE) {
2850 struct extent_state *cached = NULL;
2852 userpage = kmap_atomic(page);
2853 memset(userpage + pg_offset, 0, iosize);
2854 flush_dcache_page(page);
2855 kunmap_atomic(userpage);
2857 set_extent_uptodate(tree, cur, cur + iosize - 1,
2859 unlock_extent_cached(tree, cur, cur + iosize - 1,
2862 pg_offset += iosize;
2865 /* the get_extent function already copied into the page */
2866 if (test_range_bit(tree, cur, cur_end,
2867 EXTENT_UPTODATE, 1, NULL)) {
2868 check_page_uptodate(tree, page);
2870 unlock_extent(tree, cur, cur + iosize - 1);
2872 pg_offset += iosize;
2875 /* we have an inline extent but it didn't get marked up
2876 * to date. Error out
2878 if (block_start == EXTENT_MAP_INLINE) {
2881 unlock_extent(tree, cur, cur + iosize - 1);
2883 pg_offset += iosize;
2888 ret = submit_extent_page(rw, tree, page,
2889 sector, disk_io_size, pg_offset,
2891 end_bio_extent_readpage, mirror_num,
2896 *bio_flags = this_bio_flag;
2900 unlock_extent(tree, cur, cur + iosize - 1);
2903 pg_offset += iosize;
2907 if (!PageError(page))
2908 SetPageUptodate(page);
2914 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
2915 struct page *pages[], int nr_pages,
2917 get_extent_t *get_extent,
2918 struct extent_map **em_cached,
2919 struct bio **bio, int mirror_num,
2920 unsigned long *bio_flags, int rw)
2922 struct inode *inode;
2923 struct btrfs_ordered_extent *ordered;
2926 inode = pages[0]->mapping->host;
2928 lock_extent(tree, start, end);
2929 ordered = btrfs_lookup_ordered_range(inode, start,
2933 unlock_extent(tree, start, end);
2934 btrfs_start_ordered_extent(inode, ordered, 1);
2935 btrfs_put_ordered_extent(ordered);
2938 for (index = 0; index < nr_pages; index++) {
2939 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
2940 mirror_num, bio_flags, rw);
2941 page_cache_release(pages[index]);
2945 static void __extent_readpages(struct extent_io_tree *tree,
2946 struct page *pages[],
2947 int nr_pages, get_extent_t *get_extent,
2948 struct extent_map **em_cached,
2949 struct bio **bio, int mirror_num,
2950 unsigned long *bio_flags, int rw)
2956 int first_index = 0;
2958 for (index = 0; index < nr_pages; index++) {
2959 page_start = page_offset(pages[index]);
2962 end = start + PAGE_CACHE_SIZE - 1;
2963 first_index = index;
2964 } else if (end + 1 == page_start) {
2965 end += PAGE_CACHE_SIZE;
2967 __do_contiguous_readpages(tree, &pages[first_index],
2968 index - first_index, start,
2969 end, get_extent, em_cached,
2970 bio, mirror_num, bio_flags,
2973 end = start + PAGE_CACHE_SIZE - 1;
2974 first_index = index;
2979 __do_contiguous_readpages(tree, &pages[first_index],
2980 index - first_index, start,
2981 end, get_extent, em_cached, bio,
2982 mirror_num, bio_flags, rw);
2985 static int __extent_read_full_page(struct extent_io_tree *tree,
2987 get_extent_t *get_extent,
2988 struct bio **bio, int mirror_num,
2989 unsigned long *bio_flags, int rw)
2991 struct inode *inode = page->mapping->host;
2992 struct btrfs_ordered_extent *ordered;
2993 u64 start = page_offset(page);
2994 u64 end = start + PAGE_CACHE_SIZE - 1;
2998 lock_extent(tree, start, end);
2999 ordered = btrfs_lookup_ordered_extent(inode, start);
3002 unlock_extent(tree, start, end);
3003 btrfs_start_ordered_extent(inode, ordered, 1);
3004 btrfs_put_ordered_extent(ordered);
3007 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3012 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3013 get_extent_t *get_extent, int mirror_num)
3015 struct bio *bio = NULL;
3016 unsigned long bio_flags = 0;
3019 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3022 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3026 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3027 get_extent_t *get_extent, int mirror_num)
3029 struct bio *bio = NULL;
3030 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3033 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3036 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3040 static noinline void update_nr_written(struct page *page,
3041 struct writeback_control *wbc,
3042 unsigned long nr_written)
3044 wbc->nr_to_write -= nr_written;
3045 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3046 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3047 page->mapping->writeback_index = page->index + nr_written;
3051 * the writepage semantics are similar to regular writepage. extent
3052 * records are inserted to lock ranges in the tree, and as dirty areas
3053 * are found, they are marked writeback. Then the lock bits are removed
3054 * and the end_io handler clears the writeback ranges
3056 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3059 struct inode *inode = page->mapping->host;
3060 struct extent_page_data *epd = data;
3061 struct extent_io_tree *tree = epd->tree;
3062 u64 start = page_offset(page);
3064 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3068 u64 last_byte = i_size_read(inode);
3072 struct extent_state *cached_state = NULL;
3073 struct extent_map *em;
3074 struct block_device *bdev;
3077 size_t pg_offset = 0;
3079 loff_t i_size = i_size_read(inode);
3080 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3086 unsigned long nr_written = 0;
3087 bool fill_delalloc = true;
3089 if (wbc->sync_mode == WB_SYNC_ALL)
3090 write_flags = WRITE_SYNC;
3092 write_flags = WRITE;
3094 trace___extent_writepage(page, inode, wbc);
3096 WARN_ON(!PageLocked(page));
3098 ClearPageError(page);
3100 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3101 if (page->index > end_index ||
3102 (page->index == end_index && !pg_offset)) {
3103 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3108 if (page->index == end_index) {
3111 userpage = kmap_atomic(page);
3112 memset(userpage + pg_offset, 0,
3113 PAGE_CACHE_SIZE - pg_offset);
3114 kunmap_atomic(userpage);
3115 flush_dcache_page(page);
3119 set_page_extent_mapped(page);
3121 if (!tree->ops || !tree->ops->fill_delalloc)
3122 fill_delalloc = false;
3124 delalloc_start = start;
3127 if (!epd->extent_locked && fill_delalloc) {
3128 u64 delalloc_to_write = 0;
3130 * make sure the wbc mapping index is at least updated
3133 update_nr_written(page, wbc, 0);
3135 while (delalloc_end < page_end) {
3136 nr_delalloc = find_lock_delalloc_range(inode, tree,
3141 if (nr_delalloc == 0) {
3142 delalloc_start = delalloc_end + 1;
3145 ret = tree->ops->fill_delalloc(inode, page,
3150 /* File system has been set read-only */
3156 * delalloc_end is already one less than the total
3157 * length, so we don't subtract one from
3160 delalloc_to_write += (delalloc_end - delalloc_start +
3163 delalloc_start = delalloc_end + 1;
3165 if (wbc->nr_to_write < delalloc_to_write) {
3168 if (delalloc_to_write < thresh * 2)
3169 thresh = delalloc_to_write;
3170 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3174 /* did the fill delalloc function already unlock and start
3180 * we've unlocked the page, so we can't update
3181 * the mapping's writeback index, just update
3184 wbc->nr_to_write -= nr_written;
3188 if (tree->ops && tree->ops->writepage_start_hook) {
3189 ret = tree->ops->writepage_start_hook(page, start,
3192 /* Fixup worker will requeue */
3194 wbc->pages_skipped++;
3196 redirty_page_for_writepage(wbc, page);
3197 update_nr_written(page, wbc, nr_written);
3205 * we don't want to touch the inode after unlocking the page,
3206 * so we update the mapping writeback index now
3208 update_nr_written(page, wbc, nr_written + 1);
3211 if (last_byte <= start) {
3212 if (tree->ops && tree->ops->writepage_end_io_hook)
3213 tree->ops->writepage_end_io_hook(page, start,
3218 blocksize = inode->i_sb->s_blocksize;
3220 while (cur <= end) {
3221 if (cur >= last_byte) {
3222 if (tree->ops && tree->ops->writepage_end_io_hook)
3223 tree->ops->writepage_end_io_hook(page, cur,
3227 em = epd->get_extent(inode, page, pg_offset, cur,
3229 if (IS_ERR_OR_NULL(em)) {
3234 extent_offset = cur - em->start;
3235 BUG_ON(extent_map_end(em) <= cur);
3237 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3238 iosize = ALIGN(iosize, blocksize);
3239 sector = (em->block_start + extent_offset) >> 9;
3241 block_start = em->block_start;
3242 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3243 free_extent_map(em);
3247 * compressed and inline extents are written through other
3250 if (compressed || block_start == EXTENT_MAP_HOLE ||
3251 block_start == EXTENT_MAP_INLINE) {
3253 * end_io notification does not happen here for
3254 * compressed extents
3256 if (!compressed && tree->ops &&
3257 tree->ops->writepage_end_io_hook)
3258 tree->ops->writepage_end_io_hook(page, cur,
3261 else if (compressed) {
3262 /* we don't want to end_page_writeback on
3263 * a compressed extent. this happens
3270 pg_offset += iosize;
3273 /* leave this out until we have a page_mkwrite call */
3274 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3275 EXTENT_DIRTY, 0, NULL)) {
3277 pg_offset += iosize;
3281 if (tree->ops && tree->ops->writepage_io_hook) {
3282 ret = tree->ops->writepage_io_hook(page, cur,
3290 unsigned long max_nr = end_index + 1;
3292 set_range_writeback(tree, cur, cur + iosize - 1);
3293 if (!PageWriteback(page)) {
3294 printk(KERN_ERR "btrfs warning page %lu not "
3295 "writeback, cur %llu end %llu\n",
3296 page->index, cur, end);
3299 ret = submit_extent_page(write_flags, tree, page,
3300 sector, iosize, pg_offset,
3301 bdev, &epd->bio, max_nr,
3302 end_bio_extent_writepage,
3308 pg_offset += iosize;
3313 /* make sure the mapping tag for page dirty gets cleared */
3314 set_page_writeback(page);
3315 end_page_writeback(page);
3321 /* drop our reference on any cached states */
3322 free_extent_state(cached_state);
3326 static int eb_wait(void *word)
3332 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3334 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3335 TASK_UNINTERRUPTIBLE);
3338 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3339 struct btrfs_fs_info *fs_info,
3340 struct extent_page_data *epd)
3342 unsigned long i, num_pages;
3346 if (!btrfs_try_tree_write_lock(eb)) {
3348 flush_write_bio(epd);
3349 btrfs_tree_lock(eb);
3352 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3353 btrfs_tree_unlock(eb);
3357 flush_write_bio(epd);
3361 wait_on_extent_buffer_writeback(eb);
3362 btrfs_tree_lock(eb);
3363 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3365 btrfs_tree_unlock(eb);
3370 * We need to do this to prevent races in people who check if the eb is
3371 * under IO since we can end up having no IO bits set for a short period
3374 spin_lock(&eb->refs_lock);
3375 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3376 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3377 spin_unlock(&eb->refs_lock);
3378 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3379 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3381 fs_info->dirty_metadata_batch);
3384 spin_unlock(&eb->refs_lock);
3387 btrfs_tree_unlock(eb);
3392 num_pages = num_extent_pages(eb->start, eb->len);
3393 for (i = 0; i < num_pages; i++) {
3394 struct page *p = extent_buffer_page(eb, i);
3396 if (!trylock_page(p)) {
3398 flush_write_bio(epd);
3408 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3410 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3411 smp_mb__after_clear_bit();
3412 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3415 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3417 int uptodate = err == 0;
3418 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3419 struct extent_buffer *eb;
3423 struct page *page = bvec->bv_page;
3426 eb = (struct extent_buffer *)page->private;
3428 done = atomic_dec_and_test(&eb->io_pages);
3430 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3431 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3432 ClearPageUptodate(page);
3436 end_page_writeback(page);
3441 end_extent_buffer_writeback(eb);
3442 } while (bvec >= bio->bi_io_vec);
3448 static int write_one_eb(struct extent_buffer *eb,
3449 struct btrfs_fs_info *fs_info,
3450 struct writeback_control *wbc,
3451 struct extent_page_data *epd)
3453 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3454 u64 offset = eb->start;
3455 unsigned long i, num_pages;
3456 unsigned long bio_flags = 0;
3457 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3460 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3461 num_pages = num_extent_pages(eb->start, eb->len);
3462 atomic_set(&eb->io_pages, num_pages);
3463 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3464 bio_flags = EXTENT_BIO_TREE_LOG;
3466 for (i = 0; i < num_pages; i++) {
3467 struct page *p = extent_buffer_page(eb, i);
3469 clear_page_dirty_for_io(p);
3470 set_page_writeback(p);
3471 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3472 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3473 -1, end_bio_extent_buffer_writepage,
3474 0, epd->bio_flags, bio_flags);
3475 epd->bio_flags = bio_flags;
3477 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3479 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3480 end_extent_buffer_writeback(eb);
3484 offset += PAGE_CACHE_SIZE;
3485 update_nr_written(p, wbc, 1);
3489 if (unlikely(ret)) {
3490 for (; i < num_pages; i++) {
3491 struct page *p = extent_buffer_page(eb, i);
3499 int btree_write_cache_pages(struct address_space *mapping,
3500 struct writeback_control *wbc)
3502 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3503 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3504 struct extent_buffer *eb, *prev_eb = NULL;
3505 struct extent_page_data epd = {
3509 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3514 int nr_to_write_done = 0;
3515 struct pagevec pvec;
3518 pgoff_t end; /* Inclusive */
3522 pagevec_init(&pvec, 0);
3523 if (wbc->range_cyclic) {
3524 index = mapping->writeback_index; /* Start from prev offset */
3527 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3528 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3531 if (wbc->sync_mode == WB_SYNC_ALL)
3532 tag = PAGECACHE_TAG_TOWRITE;
3534 tag = PAGECACHE_TAG_DIRTY;
3536 if (wbc->sync_mode == WB_SYNC_ALL)
3537 tag_pages_for_writeback(mapping, index, end);
3538 while (!done && !nr_to_write_done && (index <= end) &&
3539 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3540 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3544 for (i = 0; i < nr_pages; i++) {
3545 struct page *page = pvec.pages[i];
3547 if (!PagePrivate(page))
3550 if (!wbc->range_cyclic && page->index > end) {
3555 spin_lock(&mapping->private_lock);
3556 if (!PagePrivate(page)) {
3557 spin_unlock(&mapping->private_lock);
3561 eb = (struct extent_buffer *)page->private;
3564 * Shouldn't happen and normally this would be a BUG_ON
3565 * but no sense in crashing the users box for something
3566 * we can survive anyway.
3569 spin_unlock(&mapping->private_lock);
3574 if (eb == prev_eb) {
3575 spin_unlock(&mapping->private_lock);
3579 ret = atomic_inc_not_zero(&eb->refs);
3580 spin_unlock(&mapping->private_lock);
3585 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3587 free_extent_buffer(eb);
3591 ret = write_one_eb(eb, fs_info, wbc, &epd);
3594 free_extent_buffer(eb);
3597 free_extent_buffer(eb);
3600 * the filesystem may choose to bump up nr_to_write.
3601 * We have to make sure to honor the new nr_to_write
3604 nr_to_write_done = wbc->nr_to_write <= 0;
3606 pagevec_release(&pvec);
3609 if (!scanned && !done) {
3611 * We hit the last page and there is more work to be done: wrap
3612 * back to the start of the file
3618 flush_write_bio(&epd);
3623 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3624 * @mapping: address space structure to write
3625 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3626 * @writepage: function called for each page
3627 * @data: data passed to writepage function
3629 * If a page is already under I/O, write_cache_pages() skips it, even
3630 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3631 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3632 * and msync() need to guarantee that all the data which was dirty at the time
3633 * the call was made get new I/O started against them. If wbc->sync_mode is
3634 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3635 * existing IO to complete.
3637 static int extent_write_cache_pages(struct extent_io_tree *tree,
3638 struct address_space *mapping,
3639 struct writeback_control *wbc,
3640 writepage_t writepage, void *data,
3641 void (*flush_fn)(void *))
3643 struct inode *inode = mapping->host;
3646 int nr_to_write_done = 0;
3647 struct pagevec pvec;
3650 pgoff_t end; /* Inclusive */
3655 * We have to hold onto the inode so that ordered extents can do their
3656 * work when the IO finishes. The alternative to this is failing to add
3657 * an ordered extent if the igrab() fails there and that is a huge pain
3658 * to deal with, so instead just hold onto the inode throughout the
3659 * writepages operation. If it fails here we are freeing up the inode
3660 * anyway and we'd rather not waste our time writing out stuff that is
3661 * going to be truncated anyway.
3666 pagevec_init(&pvec, 0);
3667 if (wbc->range_cyclic) {
3668 index = mapping->writeback_index; /* Start from prev offset */
3671 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3672 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3675 if (wbc->sync_mode == WB_SYNC_ALL)
3676 tag = PAGECACHE_TAG_TOWRITE;
3678 tag = PAGECACHE_TAG_DIRTY;
3680 if (wbc->sync_mode == WB_SYNC_ALL)
3681 tag_pages_for_writeback(mapping, index, end);
3682 while (!done && !nr_to_write_done && (index <= end) &&
3683 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3684 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3688 for (i = 0; i < nr_pages; i++) {
3689 struct page *page = pvec.pages[i];
3692 * At this point we hold neither mapping->tree_lock nor
3693 * lock on the page itself: the page may be truncated or
3694 * invalidated (changing page->mapping to NULL), or even
3695 * swizzled back from swapper_space to tmpfs file
3698 if (!trylock_page(page)) {
3703 if (unlikely(page->mapping != mapping)) {
3708 if (!wbc->range_cyclic && page->index > end) {
3714 if (wbc->sync_mode != WB_SYNC_NONE) {
3715 if (PageWriteback(page))
3717 wait_on_page_writeback(page);
3720 if (PageWriteback(page) ||
3721 !clear_page_dirty_for_io(page)) {
3726 ret = (*writepage)(page, wbc, data);
3728 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3736 * the filesystem may choose to bump up nr_to_write.
3737 * We have to make sure to honor the new nr_to_write
3740 nr_to_write_done = wbc->nr_to_write <= 0;
3742 pagevec_release(&pvec);
3745 if (!scanned && !done) {
3747 * We hit the last page and there is more work to be done: wrap
3748 * back to the start of the file
3754 btrfs_add_delayed_iput(inode);
3758 static void flush_epd_write_bio(struct extent_page_data *epd)
3767 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3768 BUG_ON(ret < 0); /* -ENOMEM */
3773 static noinline void flush_write_bio(void *data)
3775 struct extent_page_data *epd = data;
3776 flush_epd_write_bio(epd);
3779 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3780 get_extent_t *get_extent,
3781 struct writeback_control *wbc)
3784 struct extent_page_data epd = {
3787 .get_extent = get_extent,
3789 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3793 ret = __extent_writepage(page, wbc, &epd);
3795 flush_epd_write_bio(&epd);
3799 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3800 u64 start, u64 end, get_extent_t *get_extent,
3804 struct address_space *mapping = inode->i_mapping;
3806 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3809 struct extent_page_data epd = {
3812 .get_extent = get_extent,
3814 .sync_io = mode == WB_SYNC_ALL,
3817 struct writeback_control wbc_writepages = {
3819 .nr_to_write = nr_pages * 2,
3820 .range_start = start,
3821 .range_end = end + 1,
3824 while (start <= end) {
3825 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3826 if (clear_page_dirty_for_io(page))
3827 ret = __extent_writepage(page, &wbc_writepages, &epd);
3829 if (tree->ops && tree->ops->writepage_end_io_hook)
3830 tree->ops->writepage_end_io_hook(page, start,
3831 start + PAGE_CACHE_SIZE - 1,
3835 page_cache_release(page);
3836 start += PAGE_CACHE_SIZE;
3839 flush_epd_write_bio(&epd);
3843 int extent_writepages(struct extent_io_tree *tree,
3844 struct address_space *mapping,
3845 get_extent_t *get_extent,
3846 struct writeback_control *wbc)
3849 struct extent_page_data epd = {
3852 .get_extent = get_extent,
3854 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3858 ret = extent_write_cache_pages(tree, mapping, wbc,
3859 __extent_writepage, &epd,
3861 flush_epd_write_bio(&epd);
3865 int extent_readpages(struct extent_io_tree *tree,
3866 struct address_space *mapping,
3867 struct list_head *pages, unsigned nr_pages,
3868 get_extent_t get_extent)
3870 struct bio *bio = NULL;
3872 unsigned long bio_flags = 0;
3873 struct page *pagepool[16];
3875 struct extent_map *em_cached = NULL;
3878 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3879 page = list_entry(pages->prev, struct page, lru);
3881 prefetchw(&page->flags);
3882 list_del(&page->lru);
3883 if (add_to_page_cache_lru(page, mapping,
3884 page->index, GFP_NOFS)) {
3885 page_cache_release(page);
3889 pagepool[nr++] = page;
3890 if (nr < ARRAY_SIZE(pagepool))
3892 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3893 &bio, 0, &bio_flags, READ);
3897 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
3898 &bio, 0, &bio_flags, READ);
3901 free_extent_map(em_cached);
3903 BUG_ON(!list_empty(pages));
3905 return submit_one_bio(READ, bio, 0, bio_flags);
3910 * basic invalidatepage code, this waits on any locked or writeback
3911 * ranges corresponding to the page, and then deletes any extent state
3912 * records from the tree
3914 int extent_invalidatepage(struct extent_io_tree *tree,
3915 struct page *page, unsigned long offset)
3917 struct extent_state *cached_state = NULL;
3918 u64 start = page_offset(page);
3919 u64 end = start + PAGE_CACHE_SIZE - 1;
3920 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3922 start += ALIGN(offset, blocksize);
3926 lock_extent_bits(tree, start, end, 0, &cached_state);
3927 wait_on_page_writeback(page);
3928 clear_extent_bit(tree, start, end,
3929 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3930 EXTENT_DO_ACCOUNTING,
3931 1, 1, &cached_state, GFP_NOFS);
3936 * a helper for releasepage, this tests for areas of the page that
3937 * are locked or under IO and drops the related state bits if it is safe
3940 static int try_release_extent_state(struct extent_map_tree *map,
3941 struct extent_io_tree *tree,
3942 struct page *page, gfp_t mask)
3944 u64 start = page_offset(page);
3945 u64 end = start + PAGE_CACHE_SIZE - 1;
3948 if (test_range_bit(tree, start, end,
3949 EXTENT_IOBITS, 0, NULL))
3952 if ((mask & GFP_NOFS) == GFP_NOFS)
3955 * at this point we can safely clear everything except the
3956 * locked bit and the nodatasum bit
3958 ret = clear_extent_bit(tree, start, end,
3959 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3962 /* if clear_extent_bit failed for enomem reasons,
3963 * we can't allow the release to continue.
3974 * a helper for releasepage. As long as there are no locked extents
3975 * in the range corresponding to the page, both state records and extent
3976 * map records are removed
3978 int try_release_extent_mapping(struct extent_map_tree *map,
3979 struct extent_io_tree *tree, struct page *page,
3982 struct extent_map *em;
3983 u64 start = page_offset(page);
3984 u64 end = start + PAGE_CACHE_SIZE - 1;
3986 if ((mask & __GFP_WAIT) &&
3987 page->mapping->host->i_size > 16 * 1024 * 1024) {
3989 while (start <= end) {
3990 len = end - start + 1;
3991 write_lock(&map->lock);
3992 em = lookup_extent_mapping(map, start, len);
3994 write_unlock(&map->lock);
3997 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3998 em->start != start) {
3999 write_unlock(&map->lock);
4000 free_extent_map(em);
4003 if (!test_range_bit(tree, em->start,
4004 extent_map_end(em) - 1,
4005 EXTENT_LOCKED | EXTENT_WRITEBACK,
4007 remove_extent_mapping(map, em);
4008 /* once for the rb tree */
4009 free_extent_map(em);
4011 start = extent_map_end(em);
4012 write_unlock(&map->lock);
4015 free_extent_map(em);
4018 return try_release_extent_state(map, tree, page, mask);
4022 * helper function for fiemap, which doesn't want to see any holes.
4023 * This maps until we find something past 'last'
4025 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4028 get_extent_t *get_extent)
4030 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4031 struct extent_map *em;
4038 len = last - offset;
4041 len = ALIGN(len, sectorsize);
4042 em = get_extent(inode, NULL, 0, offset, len, 0);
4043 if (IS_ERR_OR_NULL(em))
4046 /* if this isn't a hole return it */
4047 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4048 em->block_start != EXTENT_MAP_HOLE) {
4052 /* this is a hole, advance to the next extent */
4053 offset = extent_map_end(em);
4054 free_extent_map(em);
4061 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4062 __u64 start, __u64 len, get_extent_t *get_extent)
4066 u64 max = start + len;
4070 u64 last_for_get_extent = 0;
4072 u64 isize = i_size_read(inode);
4073 struct btrfs_key found_key;
4074 struct extent_map *em = NULL;
4075 struct extent_state *cached_state = NULL;
4076 struct btrfs_path *path;
4077 struct btrfs_file_extent_item *item;
4082 unsigned long emflags;
4087 path = btrfs_alloc_path();
4090 path->leave_spinning = 1;
4092 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
4093 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
4096 * lookup the last file extent. We're not using i_size here
4097 * because there might be preallocation past i_size
4099 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
4100 path, btrfs_ino(inode), -1, 0);
4102 btrfs_free_path(path);
4107 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4108 struct btrfs_file_extent_item);
4109 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4110 found_type = btrfs_key_type(&found_key);
4112 /* No extents, but there might be delalloc bits */
4113 if (found_key.objectid != btrfs_ino(inode) ||
4114 found_type != BTRFS_EXTENT_DATA_KEY) {
4115 /* have to trust i_size as the end */
4117 last_for_get_extent = isize;
4120 * remember the start of the last extent. There are a
4121 * bunch of different factors that go into the length of the
4122 * extent, so its much less complex to remember where it started
4124 last = found_key.offset;
4125 last_for_get_extent = last + 1;
4127 btrfs_free_path(path);
4130 * we might have some extents allocated but more delalloc past those
4131 * extents. so, we trust isize unless the start of the last extent is
4136 last_for_get_extent = isize;
4139 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4142 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4152 u64 offset_in_extent = 0;
4154 /* break if the extent we found is outside the range */
4155 if (em->start >= max || extent_map_end(em) < off)
4159 * get_extent may return an extent that starts before our
4160 * requested range. We have to make sure the ranges
4161 * we return to fiemap always move forward and don't
4162 * overlap, so adjust the offsets here
4164 em_start = max(em->start, off);
4167 * record the offset from the start of the extent
4168 * for adjusting the disk offset below. Only do this if the
4169 * extent isn't compressed since our in ram offset may be past
4170 * what we have actually allocated on disk.
4172 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4173 offset_in_extent = em_start - em->start;
4174 em_end = extent_map_end(em);
4175 em_len = em_end - em_start;
4176 emflags = em->flags;
4181 * bump off for our next call to get_extent
4183 off = extent_map_end(em);
4187 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4189 flags |= FIEMAP_EXTENT_LAST;
4190 } else if (em->block_start == EXTENT_MAP_INLINE) {
4191 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4192 FIEMAP_EXTENT_NOT_ALIGNED);
4193 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4194 flags |= (FIEMAP_EXTENT_DELALLOC |
4195 FIEMAP_EXTENT_UNKNOWN);
4197 disko = em->block_start + offset_in_extent;
4199 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4200 flags |= FIEMAP_EXTENT_ENCODED;
4202 free_extent_map(em);
4204 if ((em_start >= last) || em_len == (u64)-1 ||
4205 (last == (u64)-1 && isize <= em_end)) {
4206 flags |= FIEMAP_EXTENT_LAST;
4210 /* now scan forward to see if this is really the last extent. */
4211 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4218 flags |= FIEMAP_EXTENT_LAST;
4221 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4227 free_extent_map(em);
4229 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4230 &cached_state, GFP_NOFS);
4234 static void __free_extent_buffer(struct extent_buffer *eb)
4236 btrfs_leak_debug_del(&eb->leak_list);
4237 kmem_cache_free(extent_buffer_cache, eb);
4240 static int extent_buffer_under_io(struct extent_buffer *eb)
4242 return (atomic_read(&eb->io_pages) ||
4243 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4244 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4248 * Helper for releasing extent buffer page.
4250 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4251 unsigned long start_idx)
4253 unsigned long index;
4254 unsigned long num_pages;
4256 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4258 BUG_ON(extent_buffer_under_io(eb));
4260 num_pages = num_extent_pages(eb->start, eb->len);
4261 index = start_idx + num_pages;
4262 if (start_idx >= index)
4267 page = extent_buffer_page(eb, index);
4268 if (page && mapped) {
4269 spin_lock(&page->mapping->private_lock);
4271 * We do this since we'll remove the pages after we've
4272 * removed the eb from the radix tree, so we could race
4273 * and have this page now attached to the new eb. So
4274 * only clear page_private if it's still connected to
4277 if (PagePrivate(page) &&
4278 page->private == (unsigned long)eb) {
4279 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4280 BUG_ON(PageDirty(page));
4281 BUG_ON(PageWriteback(page));
4283 * We need to make sure we haven't be attached
4286 ClearPagePrivate(page);
4287 set_page_private(page, 0);
4288 /* One for the page private */
4289 page_cache_release(page);
4291 spin_unlock(&page->mapping->private_lock);
4295 /* One for when we alloced the page */
4296 page_cache_release(page);
4298 } while (index != start_idx);
4302 * Helper for releasing the extent buffer.
4304 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4306 btrfs_release_extent_buffer_page(eb, 0);
4307 __free_extent_buffer(eb);
4310 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
4315 struct extent_buffer *eb = NULL;
4317 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4324 rwlock_init(&eb->lock);
4325 atomic_set(&eb->write_locks, 0);
4326 atomic_set(&eb->read_locks, 0);
4327 atomic_set(&eb->blocking_readers, 0);
4328 atomic_set(&eb->blocking_writers, 0);
4329 atomic_set(&eb->spinning_readers, 0);
4330 atomic_set(&eb->spinning_writers, 0);
4331 eb->lock_nested = 0;
4332 init_waitqueue_head(&eb->write_lock_wq);
4333 init_waitqueue_head(&eb->read_lock_wq);
4335 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4337 spin_lock_init(&eb->refs_lock);
4338 atomic_set(&eb->refs, 1);
4339 atomic_set(&eb->io_pages, 0);
4342 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4344 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4345 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4346 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4351 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4355 struct extent_buffer *new;
4356 unsigned long num_pages = num_extent_pages(src->start, src->len);
4358 new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS);
4362 for (i = 0; i < num_pages; i++) {
4363 p = alloc_page(GFP_NOFS);
4365 btrfs_release_extent_buffer(new);
4368 attach_extent_buffer_page(new, p);
4369 WARN_ON(PageDirty(p));
4374 copy_extent_buffer(new, src, 0, 0, src->len);
4375 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4376 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4381 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4383 struct extent_buffer *eb;
4384 unsigned long num_pages = num_extent_pages(0, len);
4387 eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS);
4391 for (i = 0; i < num_pages; i++) {
4392 eb->pages[i] = alloc_page(GFP_NOFS);
4396 set_extent_buffer_uptodate(eb);
4397 btrfs_set_header_nritems(eb, 0);
4398 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4403 __free_page(eb->pages[i - 1]);
4404 __free_extent_buffer(eb);
4408 static void check_buffer_tree_ref(struct extent_buffer *eb)
4411 /* the ref bit is tricky. We have to make sure it is set
4412 * if we have the buffer dirty. Otherwise the
4413 * code to free a buffer can end up dropping a dirty
4416 * Once the ref bit is set, it won't go away while the
4417 * buffer is dirty or in writeback, and it also won't
4418 * go away while we have the reference count on the
4421 * We can't just set the ref bit without bumping the
4422 * ref on the eb because free_extent_buffer might
4423 * see the ref bit and try to clear it. If this happens
4424 * free_extent_buffer might end up dropping our original
4425 * ref by mistake and freeing the page before we are able
4426 * to add one more ref.
4428 * So bump the ref count first, then set the bit. If someone
4429 * beat us to it, drop the ref we added.
4431 refs = atomic_read(&eb->refs);
4432 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4435 spin_lock(&eb->refs_lock);
4436 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4437 atomic_inc(&eb->refs);
4438 spin_unlock(&eb->refs_lock);
4441 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4443 unsigned long num_pages, i;
4445 check_buffer_tree_ref(eb);
4447 num_pages = num_extent_pages(eb->start, eb->len);
4448 for (i = 0; i < num_pages; i++) {
4449 struct page *p = extent_buffer_page(eb, i);
4450 mark_page_accessed(p);
4454 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4455 u64 start, unsigned long len)
4457 unsigned long num_pages = num_extent_pages(start, len);
4459 unsigned long index = start >> PAGE_CACHE_SHIFT;
4460 struct extent_buffer *eb;
4461 struct extent_buffer *exists = NULL;
4463 struct address_space *mapping = tree->mapping;
4468 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4469 if (eb && atomic_inc_not_zero(&eb->refs)) {
4471 mark_extent_buffer_accessed(eb);
4476 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4480 for (i = 0; i < num_pages; i++, index++) {
4481 p = find_or_create_page(mapping, index, GFP_NOFS);
4485 spin_lock(&mapping->private_lock);
4486 if (PagePrivate(p)) {
4488 * We could have already allocated an eb for this page
4489 * and attached one so lets see if we can get a ref on
4490 * the existing eb, and if we can we know it's good and
4491 * we can just return that one, else we know we can just
4492 * overwrite page->private.
4494 exists = (struct extent_buffer *)p->private;
4495 if (atomic_inc_not_zero(&exists->refs)) {
4496 spin_unlock(&mapping->private_lock);
4498 page_cache_release(p);
4499 mark_extent_buffer_accessed(exists);
4504 * Do this so attach doesn't complain and we need to
4505 * drop the ref the old guy had.
4507 ClearPagePrivate(p);
4508 WARN_ON(PageDirty(p));
4509 page_cache_release(p);
4511 attach_extent_buffer_page(eb, p);
4512 spin_unlock(&mapping->private_lock);
4513 WARN_ON(PageDirty(p));
4514 mark_page_accessed(p);
4516 if (!PageUptodate(p))
4520 * see below about how we avoid a nasty race with release page
4521 * and why we unlock later
4525 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4527 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4531 spin_lock(&tree->buffer_lock);
4532 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4533 if (ret == -EEXIST) {
4534 exists = radix_tree_lookup(&tree->buffer,
4535 start >> PAGE_CACHE_SHIFT);
4536 if (!atomic_inc_not_zero(&exists->refs)) {
4537 spin_unlock(&tree->buffer_lock);
4538 radix_tree_preload_end();
4542 spin_unlock(&tree->buffer_lock);
4543 radix_tree_preload_end();
4544 mark_extent_buffer_accessed(exists);
4547 /* add one reference for the tree */
4548 check_buffer_tree_ref(eb);
4549 spin_unlock(&tree->buffer_lock);
4550 radix_tree_preload_end();
4553 * there is a race where release page may have
4554 * tried to find this extent buffer in the radix
4555 * but failed. It will tell the VM it is safe to
4556 * reclaim the, and it will clear the page private bit.
4557 * We must make sure to set the page private bit properly
4558 * after the extent buffer is in the radix tree so
4559 * it doesn't get lost
4561 SetPageChecked(eb->pages[0]);
4562 for (i = 1; i < num_pages; i++) {
4563 p = extent_buffer_page(eb, i);
4564 ClearPageChecked(p);
4567 unlock_page(eb->pages[0]);
4571 for (i = 0; i < num_pages; i++) {
4573 unlock_page(eb->pages[i]);
4576 WARN_ON(!atomic_dec_and_test(&eb->refs));
4577 btrfs_release_extent_buffer(eb);
4581 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4582 u64 start, unsigned long len)
4584 struct extent_buffer *eb;
4587 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4588 if (eb && atomic_inc_not_zero(&eb->refs)) {
4590 mark_extent_buffer_accessed(eb);
4598 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4600 struct extent_buffer *eb =
4601 container_of(head, struct extent_buffer, rcu_head);
4603 __free_extent_buffer(eb);
4606 /* Expects to have eb->eb_lock already held */
4607 static int release_extent_buffer(struct extent_buffer *eb)
4609 WARN_ON(atomic_read(&eb->refs) == 0);
4610 if (atomic_dec_and_test(&eb->refs)) {
4611 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4612 spin_unlock(&eb->refs_lock);
4614 struct extent_io_tree *tree = eb->tree;
4616 spin_unlock(&eb->refs_lock);
4618 spin_lock(&tree->buffer_lock);
4619 radix_tree_delete(&tree->buffer,
4620 eb->start >> PAGE_CACHE_SHIFT);
4621 spin_unlock(&tree->buffer_lock);
4624 /* Should be safe to release our pages at this point */
4625 btrfs_release_extent_buffer_page(eb, 0);
4626 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4629 spin_unlock(&eb->refs_lock);
4634 void free_extent_buffer(struct extent_buffer *eb)
4642 refs = atomic_read(&eb->refs);
4645 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4650 spin_lock(&eb->refs_lock);
4651 if (atomic_read(&eb->refs) == 2 &&
4652 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4653 atomic_dec(&eb->refs);
4655 if (atomic_read(&eb->refs) == 2 &&
4656 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4657 !extent_buffer_under_io(eb) &&
4658 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4659 atomic_dec(&eb->refs);
4662 * I know this is terrible, but it's temporary until we stop tracking
4663 * the uptodate bits and such for the extent buffers.
4665 release_extent_buffer(eb);
4668 void free_extent_buffer_stale(struct extent_buffer *eb)
4673 spin_lock(&eb->refs_lock);
4674 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4676 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4677 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4678 atomic_dec(&eb->refs);
4679 release_extent_buffer(eb);
4682 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4685 unsigned long num_pages;
4688 num_pages = num_extent_pages(eb->start, eb->len);
4690 for (i = 0; i < num_pages; i++) {
4691 page = extent_buffer_page(eb, i);
4692 if (!PageDirty(page))
4696 WARN_ON(!PagePrivate(page));
4698 clear_page_dirty_for_io(page);
4699 spin_lock_irq(&page->mapping->tree_lock);
4700 if (!PageDirty(page)) {
4701 radix_tree_tag_clear(&page->mapping->page_tree,
4703 PAGECACHE_TAG_DIRTY);
4705 spin_unlock_irq(&page->mapping->tree_lock);
4706 ClearPageError(page);
4709 WARN_ON(atomic_read(&eb->refs) == 0);
4712 int set_extent_buffer_dirty(struct extent_buffer *eb)
4715 unsigned long num_pages;
4718 check_buffer_tree_ref(eb);
4720 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4722 num_pages = num_extent_pages(eb->start, eb->len);
4723 WARN_ON(atomic_read(&eb->refs) == 0);
4724 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4726 for (i = 0; i < num_pages; i++)
4727 set_page_dirty(extent_buffer_page(eb, i));
4731 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4735 unsigned long num_pages;
4737 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4738 num_pages = num_extent_pages(eb->start, eb->len);
4739 for (i = 0; i < num_pages; i++) {
4740 page = extent_buffer_page(eb, i);
4742 ClearPageUptodate(page);
4747 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4751 unsigned long num_pages;
4753 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4754 num_pages = num_extent_pages(eb->start, eb->len);
4755 for (i = 0; i < num_pages; i++) {
4756 page = extent_buffer_page(eb, i);
4757 SetPageUptodate(page);
4762 int extent_buffer_uptodate(struct extent_buffer *eb)
4764 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4767 int read_extent_buffer_pages(struct extent_io_tree *tree,
4768 struct extent_buffer *eb, u64 start, int wait,
4769 get_extent_t *get_extent, int mirror_num)
4772 unsigned long start_i;
4776 int locked_pages = 0;
4777 int all_uptodate = 1;
4778 unsigned long num_pages;
4779 unsigned long num_reads = 0;
4780 struct bio *bio = NULL;
4781 unsigned long bio_flags = 0;
4783 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4787 WARN_ON(start < eb->start);
4788 start_i = (start >> PAGE_CACHE_SHIFT) -
4789 (eb->start >> PAGE_CACHE_SHIFT);
4794 num_pages = num_extent_pages(eb->start, eb->len);
4795 for (i = start_i; i < num_pages; i++) {
4796 page = extent_buffer_page(eb, i);
4797 if (wait == WAIT_NONE) {
4798 if (!trylock_page(page))
4804 if (!PageUptodate(page)) {
4811 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4815 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4816 eb->read_mirror = 0;
4817 atomic_set(&eb->io_pages, num_reads);
4818 for (i = start_i; i < num_pages; i++) {
4819 page = extent_buffer_page(eb, i);
4820 if (!PageUptodate(page)) {
4821 ClearPageError(page);
4822 err = __extent_read_full_page(tree, page,
4824 mirror_num, &bio_flags,
4834 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4840 if (ret || wait != WAIT_COMPLETE)
4843 for (i = start_i; i < num_pages; i++) {
4844 page = extent_buffer_page(eb, i);
4845 wait_on_page_locked(page);
4846 if (!PageUptodate(page))
4854 while (locked_pages > 0) {
4855 page = extent_buffer_page(eb, i);
4863 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4864 unsigned long start,
4871 char *dst = (char *)dstv;
4872 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4873 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4875 WARN_ON(start > eb->len);
4876 WARN_ON(start + len > eb->start + eb->len);
4878 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4881 page = extent_buffer_page(eb, i);
4883 cur = min(len, (PAGE_CACHE_SIZE - offset));
4884 kaddr = page_address(page);
4885 memcpy(dst, kaddr + offset, cur);
4894 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4895 unsigned long min_len, char **map,
4896 unsigned long *map_start,
4897 unsigned long *map_len)
4899 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4902 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4903 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4904 unsigned long end_i = (start_offset + start + min_len - 1) >>
4911 offset = start_offset;
4915 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4918 if (start + min_len > eb->len) {
4919 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4921 eb->start, eb->len, start, min_len);
4925 p = extent_buffer_page(eb, i);
4926 kaddr = page_address(p);
4927 *map = kaddr + offset;
4928 *map_len = PAGE_CACHE_SIZE - offset;
4932 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4933 unsigned long start,
4940 char *ptr = (char *)ptrv;
4941 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4942 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4945 WARN_ON(start > eb->len);
4946 WARN_ON(start + len > eb->start + eb->len);
4948 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4951 page = extent_buffer_page(eb, i);
4953 cur = min(len, (PAGE_CACHE_SIZE - offset));
4955 kaddr = page_address(page);
4956 ret = memcmp(ptr, kaddr + offset, cur);
4968 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4969 unsigned long start, unsigned long len)
4975 char *src = (char *)srcv;
4976 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4977 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4979 WARN_ON(start > eb->len);
4980 WARN_ON(start + len > eb->start + eb->len);
4982 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
4985 page = extent_buffer_page(eb, i);
4986 WARN_ON(!PageUptodate(page));
4988 cur = min(len, PAGE_CACHE_SIZE - offset);
4989 kaddr = page_address(page);
4990 memcpy(kaddr + offset, src, cur);
4999 void memset_extent_buffer(struct extent_buffer *eb, char c,
5000 unsigned long start, unsigned long len)
5006 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5007 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5009 WARN_ON(start > eb->len);
5010 WARN_ON(start + len > eb->start + eb->len);
5012 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5015 page = extent_buffer_page(eb, i);
5016 WARN_ON(!PageUptodate(page));
5018 cur = min(len, PAGE_CACHE_SIZE - offset);
5019 kaddr = page_address(page);
5020 memset(kaddr + offset, c, cur);
5028 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5029 unsigned long dst_offset, unsigned long src_offset,
5032 u64 dst_len = dst->len;
5037 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5038 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5040 WARN_ON(src->len != dst_len);
5042 offset = (start_offset + dst_offset) &
5043 (PAGE_CACHE_SIZE - 1);
5046 page = extent_buffer_page(dst, i);
5047 WARN_ON(!PageUptodate(page));
5049 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5051 kaddr = page_address(page);
5052 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5061 static void move_pages(struct page *dst_page, struct page *src_page,
5062 unsigned long dst_off, unsigned long src_off,
5065 char *dst_kaddr = page_address(dst_page);
5066 if (dst_page == src_page) {
5067 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
5069 char *src_kaddr = page_address(src_page);
5070 char *p = dst_kaddr + dst_off + len;
5071 char *s = src_kaddr + src_off + len;
5078 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5080 unsigned long distance = (src > dst) ? src - dst : dst - src;
5081 return distance < len;
5084 static void copy_pages(struct page *dst_page, struct page *src_page,
5085 unsigned long dst_off, unsigned long src_off,
5088 char *dst_kaddr = page_address(dst_page);
5090 int must_memmove = 0;
5092 if (dst_page != src_page) {
5093 src_kaddr = page_address(src_page);
5095 src_kaddr = dst_kaddr;
5096 if (areas_overlap(src_off, dst_off, len))
5101 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5103 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5106 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5107 unsigned long src_offset, unsigned long len)
5110 size_t dst_off_in_page;
5111 size_t src_off_in_page;
5112 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5113 unsigned long dst_i;
5114 unsigned long src_i;
5116 if (src_offset + len > dst->len) {
5117 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5118 "len %lu dst len %lu\n", src_offset, len, dst->len);
5121 if (dst_offset + len > dst->len) {
5122 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5123 "len %lu dst len %lu\n", dst_offset, len, dst->len);
5128 dst_off_in_page = (start_offset + dst_offset) &
5129 (PAGE_CACHE_SIZE - 1);
5130 src_off_in_page = (start_offset + src_offset) &
5131 (PAGE_CACHE_SIZE - 1);
5133 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5134 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5136 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5138 cur = min_t(unsigned long, cur,
5139 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5141 copy_pages(extent_buffer_page(dst, dst_i),
5142 extent_buffer_page(dst, src_i),
5143 dst_off_in_page, src_off_in_page, cur);
5151 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5152 unsigned long src_offset, unsigned long len)
5155 size_t dst_off_in_page;
5156 size_t src_off_in_page;
5157 unsigned long dst_end = dst_offset + len - 1;
5158 unsigned long src_end = src_offset + len - 1;
5159 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5160 unsigned long dst_i;
5161 unsigned long src_i;
5163 if (src_offset + len > dst->len) {
5164 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5165 "len %lu len %lu\n", src_offset, len, dst->len);
5168 if (dst_offset + len > dst->len) {
5169 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5170 "len %lu len %lu\n", dst_offset, len, dst->len);
5173 if (dst_offset < src_offset) {
5174 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5178 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5179 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5181 dst_off_in_page = (start_offset + dst_end) &
5182 (PAGE_CACHE_SIZE - 1);
5183 src_off_in_page = (start_offset + src_end) &
5184 (PAGE_CACHE_SIZE - 1);
5186 cur = min_t(unsigned long, len, src_off_in_page + 1);
5187 cur = min(cur, dst_off_in_page + 1);
5188 move_pages(extent_buffer_page(dst, dst_i),
5189 extent_buffer_page(dst, src_i),
5190 dst_off_in_page - cur + 1,
5191 src_off_in_page - cur + 1, cur);
5199 int try_release_extent_buffer(struct page *page)
5201 struct extent_buffer *eb;
5204 * We need to make sure noboody is attaching this page to an eb right
5207 spin_lock(&page->mapping->private_lock);
5208 if (!PagePrivate(page)) {
5209 spin_unlock(&page->mapping->private_lock);
5213 eb = (struct extent_buffer *)page->private;
5217 * This is a little awful but should be ok, we need to make sure that
5218 * the eb doesn't disappear out from under us while we're looking at
5221 spin_lock(&eb->refs_lock);
5222 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5223 spin_unlock(&eb->refs_lock);
5224 spin_unlock(&page->mapping->private_lock);
5227 spin_unlock(&page->mapping->private_lock);
5230 * If tree ref isn't set then we know the ref on this eb is a real ref,
5231 * so just return, this page will likely be freed soon anyway.
5233 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5234 spin_unlock(&eb->refs_lock);
5238 return release_extent_buffer(eb);