1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 static struct kmem_cache *extent_state_cache;
24 static struct kmem_cache *extent_buffer_cache;
26 static LIST_HEAD(buffers);
27 static LIST_HEAD(states);
31 static DEFINE_SPINLOCK(leak_lock);
34 #define BUFFER_LRU_MAX 64
39 struct rb_node rb_node;
42 struct extent_page_data {
44 struct extent_io_tree *tree;
45 get_extent_t *get_extent;
47 /* tells writepage not to lock the state bits for this range
48 * it still does the unlocking
50 unsigned int extent_locked:1;
52 /* tells the submit_bio code to use a WRITE_SYNC */
53 unsigned int sync_io:1;
56 static inline struct btrfs_fs_info *
57 tree_fs_info(struct extent_io_tree *tree)
59 return btrfs_sb(tree->mapping->host->i_sb);
62 int __init extent_io_init(void)
64 extent_state_cache = kmem_cache_create("extent_state",
65 sizeof(struct extent_state), 0,
66 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
67 if (!extent_state_cache)
70 extent_buffer_cache = kmem_cache_create("extent_buffers",
71 sizeof(struct extent_buffer), 0,
72 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
73 if (!extent_buffer_cache)
74 goto free_state_cache;
78 kmem_cache_destroy(extent_state_cache);
82 void extent_io_exit(void)
84 struct extent_state *state;
85 struct extent_buffer *eb;
87 while (!list_empty(&states)) {
88 state = list_entry(states.next, struct extent_state, leak_list);
89 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
90 "state %lu in tree %p refs %d\n",
91 (unsigned long long)state->start,
92 (unsigned long long)state->end,
93 state->state, state->tree, atomic_read(&state->refs));
94 list_del(&state->leak_list);
95 kmem_cache_free(extent_state_cache, state);
99 while (!list_empty(&buffers)) {
100 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
101 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
102 "refs %d\n", (unsigned long long)eb->start,
103 eb->len, atomic_read(&eb->refs));
104 list_del(&eb->leak_list);
105 kmem_cache_free(extent_buffer_cache, eb);
107 if (extent_state_cache)
108 kmem_cache_destroy(extent_state_cache);
109 if (extent_buffer_cache)
110 kmem_cache_destroy(extent_buffer_cache);
113 void extent_io_tree_init(struct extent_io_tree *tree,
114 struct address_space *mapping)
116 tree->state = RB_ROOT;
117 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
119 tree->dirty_bytes = 0;
120 spin_lock_init(&tree->lock);
121 spin_lock_init(&tree->buffer_lock);
122 tree->mapping = mapping;
125 static struct extent_state *alloc_extent_state(gfp_t mask)
127 struct extent_state *state;
132 state = kmem_cache_alloc(extent_state_cache, mask);
139 spin_lock_irqsave(&leak_lock, flags);
140 list_add(&state->leak_list, &states);
141 spin_unlock_irqrestore(&leak_lock, flags);
143 atomic_set(&state->refs, 1);
144 init_waitqueue_head(&state->wq);
145 trace_alloc_extent_state(state, mask, _RET_IP_);
149 void free_extent_state(struct extent_state *state)
153 if (atomic_dec_and_test(&state->refs)) {
157 WARN_ON(state->tree);
159 spin_lock_irqsave(&leak_lock, flags);
160 list_del(&state->leak_list);
161 spin_unlock_irqrestore(&leak_lock, flags);
163 trace_free_extent_state(state, _RET_IP_);
164 kmem_cache_free(extent_state_cache, state);
168 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
169 struct rb_node *node)
171 struct rb_node **p = &root->rb_node;
172 struct rb_node *parent = NULL;
173 struct tree_entry *entry;
177 entry = rb_entry(parent, struct tree_entry, rb_node);
179 if (offset < entry->start)
181 else if (offset > entry->end)
187 entry = rb_entry(node, struct tree_entry, rb_node);
188 rb_link_node(node, parent, p);
189 rb_insert_color(node, root);
193 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
194 struct rb_node **prev_ret,
195 struct rb_node **next_ret)
197 struct rb_root *root = &tree->state;
198 struct rb_node *n = root->rb_node;
199 struct rb_node *prev = NULL;
200 struct rb_node *orig_prev = NULL;
201 struct tree_entry *entry;
202 struct tree_entry *prev_entry = NULL;
205 entry = rb_entry(n, struct tree_entry, rb_node);
209 if (offset < entry->start)
211 else if (offset > entry->end)
219 while (prev && offset > prev_entry->end) {
220 prev = rb_next(prev);
221 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
228 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
229 while (prev && offset < prev_entry->start) {
230 prev = rb_prev(prev);
231 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
238 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
241 struct rb_node *prev = NULL;
244 ret = __etree_search(tree, offset, &prev, NULL);
250 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
251 struct extent_state *other)
253 if (tree->ops && tree->ops->merge_extent_hook)
254 tree->ops->merge_extent_hook(tree->mapping->host, new,
259 * utility function to look for merge candidates inside a given range.
260 * Any extents with matching state are merged together into a single
261 * extent in the tree. Extents with EXTENT_IO in their state field
262 * are not merged because the end_io handlers need to be able to do
263 * operations on them without sleeping (or doing allocations/splits).
265 * This should be called with the tree lock held.
267 static void merge_state(struct extent_io_tree *tree,
268 struct extent_state *state)
270 struct extent_state *other;
271 struct rb_node *other_node;
273 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
276 other_node = rb_prev(&state->rb_node);
278 other = rb_entry(other_node, struct extent_state, rb_node);
279 if (other->end == state->start - 1 &&
280 other->state == state->state) {
281 merge_cb(tree, state, other);
282 state->start = other->start;
284 rb_erase(&other->rb_node, &tree->state);
285 free_extent_state(other);
288 other_node = rb_next(&state->rb_node);
290 other = rb_entry(other_node, struct extent_state, rb_node);
291 if (other->start == state->end + 1 &&
292 other->state == state->state) {
293 merge_cb(tree, state, other);
294 state->end = other->end;
296 rb_erase(&other->rb_node, &tree->state);
297 free_extent_state(other);
302 static void set_state_cb(struct extent_io_tree *tree,
303 struct extent_state *state, int *bits)
305 if (tree->ops && tree->ops->set_bit_hook)
306 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
309 static void clear_state_cb(struct extent_io_tree *tree,
310 struct extent_state *state, int *bits)
312 if (tree->ops && tree->ops->clear_bit_hook)
313 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
316 static void set_state_bits(struct extent_io_tree *tree,
317 struct extent_state *state, int *bits);
320 * insert an extent_state struct into the tree. 'bits' are set on the
321 * struct before it is inserted.
323 * This may return -EEXIST if the extent is already there, in which case the
324 * state struct is freed.
326 * The tree lock is not taken internally. This is a utility function and
327 * probably isn't what you want to call (see set/clear_extent_bit).
329 static int insert_state(struct extent_io_tree *tree,
330 struct extent_state *state, u64 start, u64 end,
333 struct rb_node *node;
336 printk(KERN_ERR "btrfs end < start %llu %llu\n",
337 (unsigned long long)end,
338 (unsigned long long)start);
341 state->start = start;
344 set_state_bits(tree, state, bits);
346 node = tree_insert(&tree->state, end, &state->rb_node);
348 struct extent_state *found;
349 found = rb_entry(node, struct extent_state, rb_node);
350 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
351 "%llu %llu\n", (unsigned long long)found->start,
352 (unsigned long long)found->end,
353 (unsigned long long)start, (unsigned long long)end);
357 merge_state(tree, state);
361 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
364 if (tree->ops && tree->ops->split_extent_hook)
365 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
369 * split a given extent state struct in two, inserting the preallocated
370 * struct 'prealloc' as the newly created second half. 'split' indicates an
371 * offset inside 'orig' where it should be split.
374 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
375 * are two extent state structs in the tree:
376 * prealloc: [orig->start, split - 1]
377 * orig: [ split, orig->end ]
379 * The tree locks are not taken by this function. They need to be held
382 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
383 struct extent_state *prealloc, u64 split)
385 struct rb_node *node;
387 split_cb(tree, orig, split);
389 prealloc->start = orig->start;
390 prealloc->end = split - 1;
391 prealloc->state = orig->state;
394 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
396 free_extent_state(prealloc);
399 prealloc->tree = tree;
404 * utility function to clear some bits in an extent state struct.
405 * it will optionally wake up any one waiting on this state (wake == 1), or
406 * forcibly remove the state from the tree (delete == 1).
408 * If no bits are set on the state struct after clearing things, the
409 * struct is freed and removed from the tree
411 static int clear_state_bit(struct extent_io_tree *tree,
412 struct extent_state *state,
415 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
416 int ret = state->state & bits_to_clear;
418 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
419 u64 range = state->end - state->start + 1;
420 WARN_ON(range > tree->dirty_bytes);
421 tree->dirty_bytes -= range;
423 clear_state_cb(tree, state, bits);
424 state->state &= ~bits_to_clear;
427 if (state->state == 0) {
429 rb_erase(&state->rb_node, &tree->state);
431 free_extent_state(state);
436 merge_state(tree, state);
441 static struct extent_state *
442 alloc_extent_state_atomic(struct extent_state *prealloc)
445 prealloc = alloc_extent_state(GFP_ATOMIC);
450 void extent_io_tree_panic(struct extent_io_tree *tree, int err)
452 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
453 "Extent tree was modified by another "
454 "thread while locked.");
458 * clear some bits on a range in the tree. This may require splitting
459 * or inserting elements in the tree, so the gfp mask is used to
460 * indicate which allocations or sleeping are allowed.
462 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
463 * the given range from the tree regardless of state (ie for truncate).
465 * the range [start, end] is inclusive.
467 * This takes the tree lock, and returns 0 on success and < 0 on error.
469 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
470 int bits, int wake, int delete,
471 struct extent_state **cached_state,
474 struct extent_state *state;
475 struct extent_state *cached;
476 struct extent_state *prealloc = NULL;
477 struct rb_node *next_node;
478 struct rb_node *node;
484 bits |= ~EXTENT_CTLBITS;
485 bits |= EXTENT_FIRST_DELALLOC;
487 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
490 if (!prealloc && (mask & __GFP_WAIT)) {
491 prealloc = alloc_extent_state(mask);
496 spin_lock(&tree->lock);
498 cached = *cached_state;
501 *cached_state = NULL;
505 if (cached && cached->tree && cached->start <= start &&
506 cached->end > start) {
508 atomic_dec(&cached->refs);
513 free_extent_state(cached);
516 * this search will find the extents that end after
519 node = tree_search(tree, start);
522 state = rb_entry(node, struct extent_state, rb_node);
524 if (state->start > end)
526 WARN_ON(state->end < start);
527 last_end = state->end;
529 if (state->end < end && !need_resched())
530 next_node = rb_next(&state->rb_node);
534 /* the state doesn't have the wanted bits, go ahead */
535 if (!(state->state & bits))
539 * | ---- desired range ---- |
541 * | ------------- state -------------- |
543 * We need to split the extent we found, and may flip
544 * bits on second half.
546 * If the extent we found extends past our range, we
547 * just split and search again. It'll get split again
548 * the next time though.
550 * If the extent we found is inside our range, we clear
551 * the desired bit on it.
554 if (state->start < start) {
555 prealloc = alloc_extent_state_atomic(prealloc);
557 err = split_state(tree, state, prealloc, start);
559 extent_io_tree_panic(tree, err);
564 if (state->end <= end) {
565 clear_state_bit(tree, state, &bits, wake);
566 if (last_end == (u64)-1)
568 start = last_end + 1;
573 * | ---- desired range ---- |
575 * We need to split the extent, and clear the bit
578 if (state->start <= end && state->end > end) {
579 prealloc = alloc_extent_state_atomic(prealloc);
581 err = split_state(tree, state, prealloc, end + 1);
583 extent_io_tree_panic(tree, err);
588 clear_state_bit(tree, prealloc, &bits, wake);
594 clear_state_bit(tree, state, &bits, wake);
596 if (last_end == (u64)-1)
598 start = last_end + 1;
599 if (start <= end && next_node) {
600 state = rb_entry(next_node, struct extent_state,
607 spin_unlock(&tree->lock);
609 free_extent_state(prealloc);
616 spin_unlock(&tree->lock);
617 if (mask & __GFP_WAIT)
622 static void wait_on_state(struct extent_io_tree *tree,
623 struct extent_state *state)
624 __releases(tree->lock)
625 __acquires(tree->lock)
628 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
629 spin_unlock(&tree->lock);
631 spin_lock(&tree->lock);
632 finish_wait(&state->wq, &wait);
636 * waits for one or more bits to clear on a range in the state tree.
637 * The range [start, end] is inclusive.
638 * The tree lock is taken by this function
640 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
642 struct extent_state *state;
643 struct rb_node *node;
645 spin_lock(&tree->lock);
649 * this search will find all the extents that end after
652 node = tree_search(tree, start);
656 state = rb_entry(node, struct extent_state, rb_node);
658 if (state->start > end)
661 if (state->state & bits) {
662 start = state->start;
663 atomic_inc(&state->refs);
664 wait_on_state(tree, state);
665 free_extent_state(state);
668 start = state->end + 1;
673 cond_resched_lock(&tree->lock);
676 spin_unlock(&tree->lock);
679 static void set_state_bits(struct extent_io_tree *tree,
680 struct extent_state *state,
683 int bits_to_set = *bits & ~EXTENT_CTLBITS;
685 set_state_cb(tree, state, bits);
686 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
687 u64 range = state->end - state->start + 1;
688 tree->dirty_bytes += range;
690 state->state |= bits_to_set;
693 static void cache_state(struct extent_state *state,
694 struct extent_state **cached_ptr)
696 if (cached_ptr && !(*cached_ptr)) {
697 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
699 atomic_inc(&state->refs);
704 static void uncache_state(struct extent_state **cached_ptr)
706 if (cached_ptr && (*cached_ptr)) {
707 struct extent_state *state = *cached_ptr;
709 free_extent_state(state);
714 * set some bits on a range in the tree. This may require allocations or
715 * sleeping, so the gfp mask is used to indicate what is allowed.
717 * If any of the exclusive bits are set, this will fail with -EEXIST if some
718 * part of the range already has the desired bits set. The start of the
719 * existing range is returned in failed_start in this case.
721 * [start, end] is inclusive This takes the tree lock.
724 static int __must_check
725 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
726 int bits, int exclusive_bits, u64 *failed_start,
727 struct extent_state **cached_state, gfp_t mask)
729 struct extent_state *state;
730 struct extent_state *prealloc = NULL;
731 struct rb_node *node;
736 bits |= EXTENT_FIRST_DELALLOC;
738 if (!prealloc && (mask & __GFP_WAIT)) {
739 prealloc = alloc_extent_state(mask);
743 spin_lock(&tree->lock);
744 if (cached_state && *cached_state) {
745 state = *cached_state;
746 if (state->start <= start && state->end > start &&
748 node = &state->rb_node;
753 * this search will find all the extents that end after
756 node = tree_search(tree, start);
758 prealloc = alloc_extent_state_atomic(prealloc);
760 err = insert_state(tree, prealloc, start, end, &bits);
762 extent_io_tree_panic(tree, err);
767 state = rb_entry(node, struct extent_state, rb_node);
769 last_start = state->start;
770 last_end = state->end;
773 * | ---- desired range ---- |
776 * Just lock what we found and keep going
778 if (state->start == start && state->end <= end) {
779 struct rb_node *next_node;
780 if (state->state & exclusive_bits) {
781 *failed_start = state->start;
786 set_state_bits(tree, state, &bits);
788 cache_state(state, cached_state);
789 merge_state(tree, state);
790 if (last_end == (u64)-1)
793 start = last_end + 1;
794 next_node = rb_next(&state->rb_node);
795 if (next_node && start < end && prealloc && !need_resched()) {
796 state = rb_entry(next_node, struct extent_state,
798 if (state->start == start)
805 * | ---- desired range ---- |
808 * | ------------- state -------------- |
810 * We need to split the extent we found, and may flip bits on
813 * If the extent we found extends past our
814 * range, we just split and search again. It'll get split
815 * again the next time though.
817 * If the extent we found is inside our range, we set the
820 if (state->start < start) {
821 if (state->state & exclusive_bits) {
822 *failed_start = start;
827 prealloc = alloc_extent_state_atomic(prealloc);
829 err = split_state(tree, state, prealloc, start);
831 extent_io_tree_panic(tree, err);
836 if (state->end <= end) {
837 set_state_bits(tree, state, &bits);
838 cache_state(state, cached_state);
839 merge_state(tree, state);
840 if (last_end == (u64)-1)
842 start = last_end + 1;
847 * | ---- desired range ---- |
848 * | state | or | state |
850 * There's a hole, we need to insert something in it and
851 * ignore the extent we found.
853 if (state->start > start) {
855 if (end < last_start)
858 this_end = last_start - 1;
860 prealloc = alloc_extent_state_atomic(prealloc);
864 * Avoid to free 'prealloc' if it can be merged with
867 err = insert_state(tree, prealloc, start, this_end,
870 extent_io_tree_panic(tree, err);
872 cache_state(prealloc, cached_state);
874 start = this_end + 1;
878 * | ---- desired range ---- |
880 * We need to split the extent, and set the bit
883 if (state->start <= end && state->end > end) {
884 if (state->state & exclusive_bits) {
885 *failed_start = start;
890 prealloc = alloc_extent_state_atomic(prealloc);
892 err = split_state(tree, state, prealloc, end + 1);
894 extent_io_tree_panic(tree, err);
896 set_state_bits(tree, prealloc, &bits);
897 cache_state(prealloc, cached_state);
898 merge_state(tree, prealloc);
906 spin_unlock(&tree->lock);
908 free_extent_state(prealloc);
915 spin_unlock(&tree->lock);
916 if (mask & __GFP_WAIT)
921 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
922 u64 *failed_start, struct extent_state **cached_state,
925 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
931 * convert_extent - convert all bits in a given range from one bit to another
932 * @tree: the io tree to search
933 * @start: the start offset in bytes
934 * @end: the end offset in bytes (inclusive)
935 * @bits: the bits to set in this range
936 * @clear_bits: the bits to clear in this range
937 * @mask: the allocation mask
939 * This will go through and set bits for the given range. If any states exist
940 * already in this range they are set with the given bit and cleared of the
941 * clear_bits. This is only meant to be used by things that are mergeable, ie
942 * converting from say DELALLOC to DIRTY. This is not meant to be used with
943 * boundary bits like LOCK.
945 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
946 int bits, int clear_bits, gfp_t mask)
948 struct extent_state *state;
949 struct extent_state *prealloc = NULL;
950 struct rb_node *node;
956 if (!prealloc && (mask & __GFP_WAIT)) {
957 prealloc = alloc_extent_state(mask);
962 spin_lock(&tree->lock);
964 * this search will find all the extents that end after
967 node = tree_search(tree, start);
969 prealloc = alloc_extent_state_atomic(prealloc);
974 err = insert_state(tree, prealloc, start, end, &bits);
977 extent_io_tree_panic(tree, err);
980 state = rb_entry(node, struct extent_state, rb_node);
982 last_start = state->start;
983 last_end = state->end;
986 * | ---- desired range ---- |
989 * Just lock what we found and keep going
991 if (state->start == start && state->end <= end) {
992 struct rb_node *next_node;
994 set_state_bits(tree, state, &bits);
995 clear_state_bit(tree, state, &clear_bits, 0);
996 if (last_end == (u64)-1)
999 start = last_end + 1;
1000 next_node = rb_next(&state->rb_node);
1001 if (next_node && start < end && prealloc && !need_resched()) {
1002 state = rb_entry(next_node, struct extent_state,
1004 if (state->start == start)
1011 * | ---- desired range ---- |
1014 * | ------------- state -------------- |
1016 * We need to split the extent we found, and may flip bits on
1019 * If the extent we found extends past our
1020 * range, we just split and search again. It'll get split
1021 * again the next time though.
1023 * If the extent we found is inside our range, we set the
1024 * desired bit on it.
1026 if (state->start < start) {
1027 prealloc = alloc_extent_state_atomic(prealloc);
1032 err = split_state(tree, state, prealloc, start);
1034 extent_io_tree_panic(tree, err);
1038 if (state->end <= end) {
1039 set_state_bits(tree, state, &bits);
1040 clear_state_bit(tree, state, &clear_bits, 0);
1041 if (last_end == (u64)-1)
1043 start = last_end + 1;
1048 * | ---- desired range ---- |
1049 * | state | or | state |
1051 * There's a hole, we need to insert something in it and
1052 * ignore the extent we found.
1054 if (state->start > start) {
1056 if (end < last_start)
1059 this_end = last_start - 1;
1061 prealloc = alloc_extent_state_atomic(prealloc);
1068 * Avoid to free 'prealloc' if it can be merged with
1071 err = insert_state(tree, prealloc, start, this_end,
1074 extent_io_tree_panic(tree, err);
1076 start = this_end + 1;
1080 * | ---- desired range ---- |
1082 * We need to split the extent, and set the bit
1085 if (state->start <= end && state->end > end) {
1086 prealloc = alloc_extent_state_atomic(prealloc);
1092 err = split_state(tree, state, prealloc, end + 1);
1094 extent_io_tree_panic(tree, err);
1096 set_state_bits(tree, prealloc, &bits);
1097 clear_state_bit(tree, prealloc, &clear_bits, 0);
1105 spin_unlock(&tree->lock);
1107 free_extent_state(prealloc);
1114 spin_unlock(&tree->lock);
1115 if (mask & __GFP_WAIT)
1120 /* wrappers around set/clear extent bit */
1121 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1124 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1128 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1129 int bits, gfp_t mask)
1131 return set_extent_bit(tree, start, end, bits, NULL,
1135 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1136 int bits, gfp_t mask)
1138 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1141 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1142 struct extent_state **cached_state, gfp_t mask)
1144 return set_extent_bit(tree, start, end,
1145 EXTENT_DELALLOC | EXTENT_UPTODATE,
1146 NULL, cached_state, mask);
1149 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1152 return clear_extent_bit(tree, start, end,
1153 EXTENT_DIRTY | EXTENT_DELALLOC |
1154 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1157 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1160 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1164 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1165 struct extent_state **cached_state, gfp_t mask)
1167 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1168 cached_state, mask);
1171 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
1172 u64 end, struct extent_state **cached_state,
1175 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1176 cached_state, mask);
1180 * either insert or lock state struct between start and end use mask to tell
1181 * us if waiting is desired.
1183 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1184 int bits, struct extent_state **cached_state)
1189 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1190 EXTENT_LOCKED, &failed_start,
1191 cached_state, GFP_NOFS);
1192 if (err == -EEXIST) {
1193 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1194 start = failed_start;
1197 WARN_ON(start > end);
1202 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1204 return lock_extent_bits(tree, start, end, 0, NULL);
1207 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1212 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1213 &failed_start, NULL, GFP_NOFS);
1214 if (err == -EEXIST) {
1215 if (failed_start > start)
1216 clear_extent_bit(tree, start, failed_start - 1,
1217 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1223 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1224 struct extent_state **cached, gfp_t mask)
1226 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1230 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1232 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1237 * helper function to set both pages and extents in the tree writeback
1239 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1241 unsigned long index = start >> PAGE_CACHE_SHIFT;
1242 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1245 while (index <= end_index) {
1246 page = find_get_page(tree->mapping, index);
1247 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1248 set_page_writeback(page);
1249 page_cache_release(page);
1255 /* find the first state struct with 'bits' set after 'start', and
1256 * return it. tree->lock must be held. NULL will returned if
1257 * nothing was found after 'start'
1259 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1260 u64 start, int bits)
1262 struct rb_node *node;
1263 struct extent_state *state;
1266 * this search will find all the extents that end after
1269 node = tree_search(tree, start);
1274 state = rb_entry(node, struct extent_state, rb_node);
1275 if (state->end >= start && (state->state & bits))
1278 node = rb_next(node);
1287 * find the first offset in the io tree with 'bits' set. zero is
1288 * returned if we find something, and *start_ret and *end_ret are
1289 * set to reflect the state struct that was found.
1291 * If nothing was found, 1 is returned, < 0 on error
1293 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1294 u64 *start_ret, u64 *end_ret, int bits)
1296 struct extent_state *state;
1299 spin_lock(&tree->lock);
1300 state = find_first_extent_bit_state(tree, start, bits);
1302 *start_ret = state->start;
1303 *end_ret = state->end;
1306 spin_unlock(&tree->lock);
1311 * find a contiguous range of bytes in the file marked as delalloc, not
1312 * more than 'max_bytes'. start and end are used to return the range,
1314 * 1 is returned if we find something, 0 if nothing was in the tree
1316 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1317 u64 *start, u64 *end, u64 max_bytes,
1318 struct extent_state **cached_state)
1320 struct rb_node *node;
1321 struct extent_state *state;
1322 u64 cur_start = *start;
1324 u64 total_bytes = 0;
1326 spin_lock(&tree->lock);
1329 * this search will find all the extents that end after
1332 node = tree_search(tree, cur_start);
1340 state = rb_entry(node, struct extent_state, rb_node);
1341 if (found && (state->start != cur_start ||
1342 (state->state & EXTENT_BOUNDARY))) {
1345 if (!(state->state & EXTENT_DELALLOC)) {
1351 *start = state->start;
1352 *cached_state = state;
1353 atomic_inc(&state->refs);
1357 cur_start = state->end + 1;
1358 node = rb_next(node);
1361 total_bytes += state->end - state->start + 1;
1362 if (total_bytes >= max_bytes)
1366 spin_unlock(&tree->lock);
1370 static noinline void __unlock_for_delalloc(struct inode *inode,
1371 struct page *locked_page,
1375 struct page *pages[16];
1376 unsigned long index = start >> PAGE_CACHE_SHIFT;
1377 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1378 unsigned long nr_pages = end_index - index + 1;
1381 if (index == locked_page->index && end_index == index)
1384 while (nr_pages > 0) {
1385 ret = find_get_pages_contig(inode->i_mapping, index,
1386 min_t(unsigned long, nr_pages,
1387 ARRAY_SIZE(pages)), pages);
1388 for (i = 0; i < ret; i++) {
1389 if (pages[i] != locked_page)
1390 unlock_page(pages[i]);
1391 page_cache_release(pages[i]);
1399 static noinline int lock_delalloc_pages(struct inode *inode,
1400 struct page *locked_page,
1404 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1405 unsigned long start_index = index;
1406 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1407 unsigned long pages_locked = 0;
1408 struct page *pages[16];
1409 unsigned long nrpages;
1413 /* the caller is responsible for locking the start index */
1414 if (index == locked_page->index && index == end_index)
1417 /* skip the page at the start index */
1418 nrpages = end_index - index + 1;
1419 while (nrpages > 0) {
1420 ret = find_get_pages_contig(inode->i_mapping, index,
1421 min_t(unsigned long,
1422 nrpages, ARRAY_SIZE(pages)), pages);
1427 /* now we have an array of pages, lock them all */
1428 for (i = 0; i < ret; i++) {
1430 * the caller is taking responsibility for
1433 if (pages[i] != locked_page) {
1434 lock_page(pages[i]);
1435 if (!PageDirty(pages[i]) ||
1436 pages[i]->mapping != inode->i_mapping) {
1438 unlock_page(pages[i]);
1439 page_cache_release(pages[i]);
1443 page_cache_release(pages[i]);
1452 if (ret && pages_locked) {
1453 __unlock_for_delalloc(inode, locked_page,
1455 ((u64)(start_index + pages_locked - 1)) <<
1462 * find a contiguous range of bytes in the file marked as delalloc, not
1463 * more than 'max_bytes'. start and end are used to return the range,
1465 * 1 is returned if we find something, 0 if nothing was in the tree
1467 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1468 struct extent_io_tree *tree,
1469 struct page *locked_page,
1470 u64 *start, u64 *end,
1476 struct extent_state *cached_state = NULL;
1481 /* step one, find a bunch of delalloc bytes starting at start */
1482 delalloc_start = *start;
1484 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1485 max_bytes, &cached_state);
1486 if (!found || delalloc_end <= *start) {
1487 *start = delalloc_start;
1488 *end = delalloc_end;
1489 free_extent_state(cached_state);
1494 * start comes from the offset of locked_page. We have to lock
1495 * pages in order, so we can't process delalloc bytes before
1498 if (delalloc_start < *start)
1499 delalloc_start = *start;
1502 * make sure to limit the number of pages we try to lock down
1505 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1506 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1508 /* step two, lock all the pages after the page that has start */
1509 ret = lock_delalloc_pages(inode, locked_page,
1510 delalloc_start, delalloc_end);
1511 if (ret == -EAGAIN) {
1512 /* some of the pages are gone, lets avoid looping by
1513 * shortening the size of the delalloc range we're searching
1515 free_extent_state(cached_state);
1517 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1518 max_bytes = PAGE_CACHE_SIZE - offset;
1526 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1528 /* step three, lock the state bits for the whole range */
1529 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1531 /* then test to make sure it is all still delalloc */
1532 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1533 EXTENT_DELALLOC, 1, cached_state);
1535 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1536 &cached_state, GFP_NOFS);
1537 __unlock_for_delalloc(inode, locked_page,
1538 delalloc_start, delalloc_end);
1542 free_extent_state(cached_state);
1543 *start = delalloc_start;
1544 *end = delalloc_end;
1549 int extent_clear_unlock_delalloc(struct inode *inode,
1550 struct extent_io_tree *tree,
1551 u64 start, u64 end, struct page *locked_page,
1555 struct page *pages[16];
1556 unsigned long index = start >> PAGE_CACHE_SHIFT;
1557 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1558 unsigned long nr_pages = end_index - index + 1;
1562 if (op & EXTENT_CLEAR_UNLOCK)
1563 clear_bits |= EXTENT_LOCKED;
1564 if (op & EXTENT_CLEAR_DIRTY)
1565 clear_bits |= EXTENT_DIRTY;
1567 if (op & EXTENT_CLEAR_DELALLOC)
1568 clear_bits |= EXTENT_DELALLOC;
1570 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1571 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1572 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1573 EXTENT_SET_PRIVATE2)))
1576 while (nr_pages > 0) {
1577 ret = find_get_pages_contig(inode->i_mapping, index,
1578 min_t(unsigned long,
1579 nr_pages, ARRAY_SIZE(pages)), pages);
1580 for (i = 0; i < ret; i++) {
1582 if (op & EXTENT_SET_PRIVATE2)
1583 SetPagePrivate2(pages[i]);
1585 if (pages[i] == locked_page) {
1586 page_cache_release(pages[i]);
1589 if (op & EXTENT_CLEAR_DIRTY)
1590 clear_page_dirty_for_io(pages[i]);
1591 if (op & EXTENT_SET_WRITEBACK)
1592 set_page_writeback(pages[i]);
1593 if (op & EXTENT_END_WRITEBACK)
1594 end_page_writeback(pages[i]);
1595 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1596 unlock_page(pages[i]);
1597 page_cache_release(pages[i]);
1607 * count the number of bytes in the tree that have a given bit(s)
1608 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1609 * cached. The total number found is returned.
1611 u64 count_range_bits(struct extent_io_tree *tree,
1612 u64 *start, u64 search_end, u64 max_bytes,
1613 unsigned long bits, int contig)
1615 struct rb_node *node;
1616 struct extent_state *state;
1617 u64 cur_start = *start;
1618 u64 total_bytes = 0;
1622 if (search_end <= cur_start) {
1627 spin_lock(&tree->lock);
1628 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1629 total_bytes = tree->dirty_bytes;
1633 * this search will find all the extents that end after
1636 node = tree_search(tree, cur_start);
1641 state = rb_entry(node, struct extent_state, rb_node);
1642 if (state->start > search_end)
1644 if (contig && found && state->start > last + 1)
1646 if (state->end >= cur_start && (state->state & bits) == bits) {
1647 total_bytes += min(search_end, state->end) + 1 -
1648 max(cur_start, state->start);
1649 if (total_bytes >= max_bytes)
1652 *start = max(cur_start, state->start);
1656 } else if (contig && found) {
1659 node = rb_next(node);
1664 spin_unlock(&tree->lock);
1669 * set the private field for a given byte offset in the tree. If there isn't
1670 * an extent_state there already, this does nothing.
1672 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1674 struct rb_node *node;
1675 struct extent_state *state;
1678 spin_lock(&tree->lock);
1680 * this search will find all the extents that end after
1683 node = tree_search(tree, start);
1688 state = rb_entry(node, struct extent_state, rb_node);
1689 if (state->start != start) {
1693 state->private = private;
1695 spin_unlock(&tree->lock);
1699 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1701 struct rb_node *node;
1702 struct extent_state *state;
1705 spin_lock(&tree->lock);
1707 * this search will find all the extents that end after
1710 node = tree_search(tree, start);
1715 state = rb_entry(node, struct extent_state, rb_node);
1716 if (state->start != start) {
1720 *private = state->private;
1722 spin_unlock(&tree->lock);
1727 * searches a range in the state tree for a given mask.
1728 * If 'filled' == 1, this returns 1 only if every extent in the tree
1729 * has the bits set. Otherwise, 1 is returned if any bit in the
1730 * range is found set.
1732 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1733 int bits, int filled, struct extent_state *cached)
1735 struct extent_state *state = NULL;
1736 struct rb_node *node;
1739 spin_lock(&tree->lock);
1740 if (cached && cached->tree && cached->start <= start &&
1741 cached->end > start)
1742 node = &cached->rb_node;
1744 node = tree_search(tree, start);
1745 while (node && start <= end) {
1746 state = rb_entry(node, struct extent_state, rb_node);
1748 if (filled && state->start > start) {
1753 if (state->start > end)
1756 if (state->state & bits) {
1760 } else if (filled) {
1765 if (state->end == (u64)-1)
1768 start = state->end + 1;
1771 node = rb_next(node);
1778 spin_unlock(&tree->lock);
1783 * helper function to set a given page up to date if all the
1784 * extents in the tree for that page are up to date
1786 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1788 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1789 u64 end = start + PAGE_CACHE_SIZE - 1;
1790 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1791 SetPageUptodate(page);
1795 * helper function to unlock a page if all the extents in the tree
1796 * for that page are unlocked
1798 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1800 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1801 u64 end = start + PAGE_CACHE_SIZE - 1;
1802 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1807 * helper function to end page writeback if all the extents
1808 * in the tree for that page are done with writeback
1810 static void check_page_writeback(struct extent_io_tree *tree,
1813 end_page_writeback(page);
1817 * When IO fails, either with EIO or csum verification fails, we
1818 * try other mirrors that might have a good copy of the data. This
1819 * io_failure_record is used to record state as we go through all the
1820 * mirrors. If another mirror has good data, the page is set up to date
1821 * and things continue. If a good mirror can't be found, the original
1822 * bio end_io callback is called to indicate things have failed.
1824 struct io_failure_record {
1829 unsigned long bio_flags;
1835 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1840 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1842 set_state_private(failure_tree, rec->start, 0);
1843 ret = clear_extent_bits(failure_tree, rec->start,
1844 rec->start + rec->len - 1,
1845 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1850 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1851 rec->start + rec->len - 1,
1852 EXTENT_DAMAGED, GFP_NOFS);
1861 static void repair_io_failure_callback(struct bio *bio, int err)
1863 complete(bio->bi_private);
1867 * this bypasses the standard btrfs submit functions deliberately, as
1868 * the standard behavior is to write all copies in a raid setup. here we only
1869 * want to write the one bad copy. so we do the mapping for ourselves and issue
1870 * submit_bio directly.
1871 * to avoid any synchonization issues, wait for the data after writing, which
1872 * actually prevents the read that triggered the error from finishing.
1873 * currently, there can be no more than two copies of every data bit. thus,
1874 * exactly one rewrite is required.
1876 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1877 u64 length, u64 logical, struct page *page,
1881 struct btrfs_device *dev;
1882 DECLARE_COMPLETION_ONSTACK(compl);
1885 struct btrfs_bio *bbio = NULL;
1888 BUG_ON(!mirror_num);
1890 bio = bio_alloc(GFP_NOFS, 1);
1893 bio->bi_private = &compl;
1894 bio->bi_end_io = repair_io_failure_callback;
1896 map_length = length;
1898 ret = btrfs_map_block(map_tree, WRITE, logical,
1899 &map_length, &bbio, mirror_num);
1904 BUG_ON(mirror_num != bbio->mirror_num);
1905 sector = bbio->stripes[mirror_num-1].physical >> 9;
1906 bio->bi_sector = sector;
1907 dev = bbio->stripes[mirror_num-1].dev;
1909 if (!dev || !dev->bdev || !dev->writeable) {
1913 bio->bi_bdev = dev->bdev;
1914 bio_add_page(bio, page, length, start-page_offset(page));
1915 btrfsic_submit_bio(WRITE_SYNC, bio);
1916 wait_for_completion(&compl);
1918 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1919 /* try to remap that extent elsewhere? */
1924 printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s "
1925 "sector %llu)\n", page->mapping->host->i_ino, start,
1933 * each time an IO finishes, we do a fast check in the IO failure tree
1934 * to see if we need to process or clean up an io_failure_record
1936 static int clean_io_failure(u64 start, struct page *page)
1939 u64 private_failure;
1940 struct io_failure_record *failrec;
1941 struct btrfs_mapping_tree *map_tree;
1942 struct extent_state *state;
1946 struct inode *inode = page->mapping->host;
1949 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1950 (u64)-1, 1, EXTENT_DIRTY, 0);
1954 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1959 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1960 BUG_ON(!failrec->this_mirror);
1962 if (failrec->in_validation) {
1963 /* there was no real error, just free the record */
1964 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1970 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1971 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1974 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1976 if (state && state->start == failrec->start) {
1977 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
1978 num_copies = btrfs_num_copies(map_tree, failrec->logical,
1980 if (num_copies > 1) {
1981 ret = repair_io_failure(map_tree, start, failrec->len,
1982 failrec->logical, page,
1983 failrec->failed_mirror);
1990 ret = free_io_failure(inode, failrec, did_repair);
1996 * this is a generic handler for readpage errors (default
1997 * readpage_io_failed_hook). if other copies exist, read those and write back
1998 * good data to the failed position. does not investigate in remapping the
1999 * failed extent elsewhere, hoping the device will be smart enough to do this as
2003 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2004 u64 start, u64 end, int failed_mirror,
2005 struct extent_state *state)
2007 struct io_failure_record *failrec = NULL;
2009 struct extent_map *em;
2010 struct inode *inode = page->mapping->host;
2011 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2012 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2013 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2020 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2022 ret = get_state_private(failure_tree, start, &private);
2024 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2027 failrec->start = start;
2028 failrec->len = end - start + 1;
2029 failrec->this_mirror = 0;
2030 failrec->bio_flags = 0;
2031 failrec->in_validation = 0;
2033 read_lock(&em_tree->lock);
2034 em = lookup_extent_mapping(em_tree, start, failrec->len);
2036 read_unlock(&em_tree->lock);
2041 if (em->start > start || em->start + em->len < start) {
2042 free_extent_map(em);
2045 read_unlock(&em_tree->lock);
2047 if (!em || IS_ERR(em)) {
2051 logical = start - em->start;
2052 logical = em->block_start + logical;
2053 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2054 logical = em->block_start;
2055 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2056 extent_set_compress_type(&failrec->bio_flags,
2059 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2060 "len=%llu\n", logical, start, failrec->len);
2061 failrec->logical = logical;
2062 free_extent_map(em);
2064 /* set the bits in the private failure tree */
2065 ret = set_extent_bits(failure_tree, start, end,
2066 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2068 ret = set_state_private(failure_tree, start,
2069 (u64)(unsigned long)failrec);
2070 /* set the bits in the inode's tree */
2072 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2079 failrec = (struct io_failure_record *)(unsigned long)private;
2080 pr_debug("bio_readpage_error: (found) logical=%llu, "
2081 "start=%llu, len=%llu, validation=%d\n",
2082 failrec->logical, failrec->start, failrec->len,
2083 failrec->in_validation);
2085 * when data can be on disk more than twice, add to failrec here
2086 * (e.g. with a list for failed_mirror) to make
2087 * clean_io_failure() clean all those errors at once.
2090 num_copies = btrfs_num_copies(
2091 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2092 failrec->logical, failrec->len);
2093 if (num_copies == 1) {
2095 * we only have a single copy of the data, so don't bother with
2096 * all the retry and error correction code that follows. no
2097 * matter what the error is, it is very likely to persist.
2099 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2100 "state=%p, num_copies=%d, next_mirror %d, "
2101 "failed_mirror %d\n", state, num_copies,
2102 failrec->this_mirror, failed_mirror);
2103 free_io_failure(inode, failrec, 0);
2108 spin_lock(&tree->lock);
2109 state = find_first_extent_bit_state(tree, failrec->start,
2111 if (state && state->start != failrec->start)
2113 spin_unlock(&tree->lock);
2117 * there are two premises:
2118 * a) deliver good data to the caller
2119 * b) correct the bad sectors on disk
2121 if (failed_bio->bi_vcnt > 1) {
2123 * to fulfill b), we need to know the exact failing sectors, as
2124 * we don't want to rewrite any more than the failed ones. thus,
2125 * we need separate read requests for the failed bio
2127 * if the following BUG_ON triggers, our validation request got
2128 * merged. we need separate requests for our algorithm to work.
2130 BUG_ON(failrec->in_validation);
2131 failrec->in_validation = 1;
2132 failrec->this_mirror = failed_mirror;
2133 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2136 * we're ready to fulfill a) and b) alongside. get a good copy
2137 * of the failed sector and if we succeed, we have setup
2138 * everything for repair_io_failure to do the rest for us.
2140 if (failrec->in_validation) {
2141 BUG_ON(failrec->this_mirror != failed_mirror);
2142 failrec->in_validation = 0;
2143 failrec->this_mirror = 0;
2145 failrec->failed_mirror = failed_mirror;
2146 failrec->this_mirror++;
2147 if (failrec->this_mirror == failed_mirror)
2148 failrec->this_mirror++;
2149 read_mode = READ_SYNC;
2152 if (!state || failrec->this_mirror > num_copies) {
2153 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2154 "next_mirror %d, failed_mirror %d\n", state,
2155 num_copies, failrec->this_mirror, failed_mirror);
2156 free_io_failure(inode, failrec, 0);
2160 bio = bio_alloc(GFP_NOFS, 1);
2161 bio->bi_private = state;
2162 bio->bi_end_io = failed_bio->bi_end_io;
2163 bio->bi_sector = failrec->logical >> 9;
2164 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2167 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2169 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2170 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2171 failrec->this_mirror, num_copies, failrec->in_validation);
2173 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2174 failrec->this_mirror,
2175 failrec->bio_flags, 0);
2179 /* lots and lots of room for performance fixes in the end_bio funcs */
2181 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2183 int uptodate = (err == 0);
2184 struct extent_io_tree *tree;
2187 tree = &BTRFS_I(page->mapping->host)->io_tree;
2189 if (tree->ops && tree->ops->writepage_end_io_hook) {
2190 ret = tree->ops->writepage_end_io_hook(page, start,
2191 end, NULL, uptodate);
2196 if (!uptodate && tree->ops &&
2197 tree->ops->writepage_io_failed_hook) {
2198 ret = tree->ops->writepage_io_failed_hook(NULL, page,
2200 /* Writeback already completed */
2206 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
2207 ClearPageUptodate(page);
2214 * after a writepage IO is done, we need to:
2215 * clear the uptodate bits on error
2216 * clear the writeback bits in the extent tree for this IO
2217 * end_page_writeback if the page has no more pending IO
2219 * Scheduling is not allowed, so the extent state tree is expected
2220 * to have one and only one object corresponding to this IO.
2222 static void end_bio_extent_writepage(struct bio *bio, int err)
2224 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2225 struct extent_io_tree *tree;
2231 struct page *page = bvec->bv_page;
2232 tree = &BTRFS_I(page->mapping->host)->io_tree;
2234 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2236 end = start + bvec->bv_len - 1;
2238 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2243 if (--bvec >= bio->bi_io_vec)
2244 prefetchw(&bvec->bv_page->flags);
2246 if (end_extent_writepage(page, err, start, end))
2250 end_page_writeback(page);
2252 check_page_writeback(tree, page);
2253 } while (bvec >= bio->bi_io_vec);
2259 * after a readpage IO is done, we need to:
2260 * clear the uptodate bits on error
2261 * set the uptodate bits if things worked
2262 * set the page up to date if all extents in the tree are uptodate
2263 * clear the lock bit in the extent tree
2264 * unlock the page if there are no other extents locked for it
2266 * Scheduling is not allowed, so the extent state tree is expected
2267 * to have one and only one object corresponding to this IO.
2269 static void end_bio_extent_readpage(struct bio *bio, int err)
2271 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2272 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2273 struct bio_vec *bvec = bio->bi_io_vec;
2274 struct extent_io_tree *tree;
2284 struct page *page = bvec->bv_page;
2285 struct extent_state *cached = NULL;
2286 struct extent_state *state;
2288 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2289 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2290 (long int)bio->bi_bdev);
2291 tree = &BTRFS_I(page->mapping->host)->io_tree;
2293 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2295 end = start + bvec->bv_len - 1;
2297 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2302 if (++bvec <= bvec_end)
2303 prefetchw(&bvec->bv_page->flags);
2305 spin_lock(&tree->lock);
2306 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2307 if (state && state->start == start) {
2309 * take a reference on the state, unlock will drop
2312 cache_state(state, &cached);
2314 spin_unlock(&tree->lock);
2316 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2317 ret = tree->ops->readpage_end_io_hook(page, start, end,
2322 clean_io_failure(start, page);
2326 failed_mirror = (int)(unsigned long)bio->bi_bdev;
2328 * The generic bio_readpage_error handles errors the
2329 * following way: If possible, new read requests are
2330 * created and submitted and will end up in
2331 * end_bio_extent_readpage as well (if we're lucky, not
2332 * in the !uptodate case). In that case it returns 0 and
2333 * we just go on with the next page in our bio. If it
2334 * can't handle the error it will return -EIO and we
2335 * remain responsible for that page.
2337 ret = bio_readpage_error(bio, page, start, end,
2338 failed_mirror, NULL);
2342 test_bit(BIO_UPTODATE, &bio->bi_flags);
2345 uncache_state(&cached);
2348 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2349 ret = tree->ops->readpage_io_failed_hook(
2350 bio, page, start, end,
2351 failed_mirror, state);
2358 set_extent_uptodate(tree, start, end, &cached,
2361 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2365 SetPageUptodate(page);
2367 ClearPageUptodate(page);
2373 check_page_uptodate(tree, page);
2375 ClearPageUptodate(page);
2378 check_page_locked(tree, page);
2380 } while (bvec <= bvec_end);
2386 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2391 bio = bio_alloc(gfp_flags, nr_vecs);
2393 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2394 while (!bio && (nr_vecs /= 2))
2395 bio = bio_alloc(gfp_flags, nr_vecs);
2400 bio->bi_bdev = bdev;
2401 bio->bi_sector = first_sector;
2407 * Since writes are async, they will only return -ENOMEM.
2408 * Reads can return the full range of I/O error conditions.
2410 static int __must_check submit_one_bio(int rw, struct bio *bio,
2411 int mirror_num, unsigned long bio_flags)
2414 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2415 struct page *page = bvec->bv_page;
2416 struct extent_io_tree *tree = bio->bi_private;
2419 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2421 bio->bi_private = NULL;
2425 if (tree->ops && tree->ops->submit_bio_hook)
2426 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2427 mirror_num, bio_flags, start);
2429 btrfsic_submit_bio(rw, bio);
2431 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2437 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2438 unsigned long offset, size_t size, struct bio *bio,
2439 unsigned long bio_flags)
2442 if (tree->ops && tree->ops->merge_bio_hook)
2443 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2450 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2451 struct page *page, sector_t sector,
2452 size_t size, unsigned long offset,
2453 struct block_device *bdev,
2454 struct bio **bio_ret,
2455 unsigned long max_pages,
2456 bio_end_io_t end_io_func,
2458 unsigned long prev_bio_flags,
2459 unsigned long bio_flags)
2465 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2466 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2467 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2469 if (bio_ret && *bio_ret) {
2472 contig = bio->bi_sector == sector;
2474 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2477 if (prev_bio_flags != bio_flags || !contig ||
2478 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2479 bio_add_page(bio, page, page_size, offset) < page_size) {
2480 ret = submit_one_bio(rw, bio, mirror_num,
2489 if (this_compressed)
2492 nr = bio_get_nr_vecs(bdev);
2494 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2498 bio_add_page(bio, page, page_size, offset);
2499 bio->bi_end_io = end_io_func;
2500 bio->bi_private = tree;
2505 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2510 void set_page_extent_mapped(struct page *page)
2512 if (!PagePrivate(page)) {
2513 SetPagePrivate(page);
2514 page_cache_get(page);
2515 set_page_private(page, EXTENT_PAGE_PRIVATE);
2519 static void set_page_extent_head(struct page *page, unsigned long len)
2521 WARN_ON(!PagePrivate(page));
2522 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
2526 * basic readpage implementation. Locked extent state structs are inserted
2527 * into the tree that are removed when the IO is done (by the end_io
2529 * XXX JDM: This needs looking at to ensure proper page locking
2531 static int __extent_read_full_page(struct extent_io_tree *tree,
2533 get_extent_t *get_extent,
2534 struct bio **bio, int mirror_num,
2535 unsigned long *bio_flags)
2537 struct inode *inode = page->mapping->host;
2538 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2539 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2543 u64 last_byte = i_size_read(inode);
2547 struct extent_map *em;
2548 struct block_device *bdev;
2549 struct btrfs_ordered_extent *ordered;
2552 size_t pg_offset = 0;
2554 size_t disk_io_size;
2555 size_t blocksize = inode->i_sb->s_blocksize;
2556 unsigned long this_bio_flag = 0;
2558 set_page_extent_mapped(page);
2560 if (!PageUptodate(page)) {
2561 if (cleancache_get_page(page) == 0) {
2562 BUG_ON(blocksize != PAGE_SIZE);
2569 lock_extent(tree, start, end);
2570 ordered = btrfs_lookup_ordered_extent(inode, start);
2573 unlock_extent(tree, start, end);
2574 btrfs_start_ordered_extent(inode, ordered, 1);
2575 btrfs_put_ordered_extent(ordered);
2578 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2580 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2583 iosize = PAGE_CACHE_SIZE - zero_offset;
2584 userpage = kmap_atomic(page, KM_USER0);
2585 memset(userpage + zero_offset, 0, iosize);
2586 flush_dcache_page(page);
2587 kunmap_atomic(userpage, KM_USER0);
2590 while (cur <= end) {
2591 if (cur >= last_byte) {
2593 struct extent_state *cached = NULL;
2595 iosize = PAGE_CACHE_SIZE - pg_offset;
2596 userpage = kmap_atomic(page, KM_USER0);
2597 memset(userpage + pg_offset, 0, iosize);
2598 flush_dcache_page(page);
2599 kunmap_atomic(userpage, KM_USER0);
2600 set_extent_uptodate(tree, cur, cur + iosize - 1,
2602 unlock_extent_cached(tree, cur, cur + iosize - 1,
2606 em = get_extent(inode, page, pg_offset, cur,
2608 if (IS_ERR_OR_NULL(em)) {
2610 unlock_extent(tree, cur, end);
2613 extent_offset = cur - em->start;
2614 BUG_ON(extent_map_end(em) <= cur);
2617 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2618 this_bio_flag = EXTENT_BIO_COMPRESSED;
2619 extent_set_compress_type(&this_bio_flag,
2623 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2624 cur_end = min(extent_map_end(em) - 1, end);
2625 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2626 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2627 disk_io_size = em->block_len;
2628 sector = em->block_start >> 9;
2630 sector = (em->block_start + extent_offset) >> 9;
2631 disk_io_size = iosize;
2634 block_start = em->block_start;
2635 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2636 block_start = EXTENT_MAP_HOLE;
2637 free_extent_map(em);
2640 /* we've found a hole, just zero and go on */
2641 if (block_start == EXTENT_MAP_HOLE) {
2643 struct extent_state *cached = NULL;
2645 userpage = kmap_atomic(page, KM_USER0);
2646 memset(userpage + pg_offset, 0, iosize);
2647 flush_dcache_page(page);
2648 kunmap_atomic(userpage, KM_USER0);
2650 set_extent_uptodate(tree, cur, cur + iosize - 1,
2652 unlock_extent_cached(tree, cur, cur + iosize - 1,
2655 pg_offset += iosize;
2658 /* the get_extent function already copied into the page */
2659 if (test_range_bit(tree, cur, cur_end,
2660 EXTENT_UPTODATE, 1, NULL)) {
2661 check_page_uptodate(tree, page);
2662 unlock_extent(tree, cur, cur + iosize - 1);
2664 pg_offset += iosize;
2667 /* we have an inline extent but it didn't get marked up
2668 * to date. Error out
2670 if (block_start == EXTENT_MAP_INLINE) {
2672 unlock_extent(tree, cur, cur + iosize - 1);
2674 pg_offset += iosize;
2679 if (tree->ops && tree->ops->readpage_io_hook) {
2680 ret = tree->ops->readpage_io_hook(page, cur,
2684 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2686 ret = submit_extent_page(READ, tree, page,
2687 sector, disk_io_size, pg_offset,
2689 end_bio_extent_readpage, mirror_num,
2692 BUG_ON(ret == -ENOMEM);
2694 *bio_flags = this_bio_flag;
2699 pg_offset += iosize;
2703 if (!PageError(page))
2704 SetPageUptodate(page);
2710 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2711 get_extent_t *get_extent, int mirror_num)
2713 struct bio *bio = NULL;
2714 unsigned long bio_flags = 0;
2717 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2720 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2724 static noinline void update_nr_written(struct page *page,
2725 struct writeback_control *wbc,
2726 unsigned long nr_written)
2728 wbc->nr_to_write -= nr_written;
2729 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2730 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2731 page->mapping->writeback_index = page->index + nr_written;
2735 * the writepage semantics are similar to regular writepage. extent
2736 * records are inserted to lock ranges in the tree, and as dirty areas
2737 * are found, they are marked writeback. Then the lock bits are removed
2738 * and the end_io handler clears the writeback ranges
2740 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2743 struct inode *inode = page->mapping->host;
2744 struct extent_page_data *epd = data;
2745 struct extent_io_tree *tree = epd->tree;
2746 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2748 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2752 u64 last_byte = i_size_read(inode);
2756 struct extent_state *cached_state = NULL;
2757 struct extent_map *em;
2758 struct block_device *bdev;
2761 size_t pg_offset = 0;
2763 loff_t i_size = i_size_read(inode);
2764 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2770 unsigned long nr_written = 0;
2771 bool fill_delalloc = true;
2773 if (wbc->sync_mode == WB_SYNC_ALL)
2774 write_flags = WRITE_SYNC;
2776 write_flags = WRITE;
2778 trace___extent_writepage(page, inode, wbc);
2780 WARN_ON(!PageLocked(page));
2782 ClearPageError(page);
2784 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2785 if (page->index > end_index ||
2786 (page->index == end_index && !pg_offset)) {
2787 page->mapping->a_ops->invalidatepage(page, 0);
2792 if (page->index == end_index) {
2795 userpage = kmap_atomic(page, KM_USER0);
2796 memset(userpage + pg_offset, 0,
2797 PAGE_CACHE_SIZE - pg_offset);
2798 kunmap_atomic(userpage, KM_USER0);
2799 flush_dcache_page(page);
2803 set_page_extent_mapped(page);
2805 if (!tree->ops || !tree->ops->fill_delalloc)
2806 fill_delalloc = false;
2808 delalloc_start = start;
2811 if (!epd->extent_locked && fill_delalloc) {
2812 u64 delalloc_to_write = 0;
2814 * make sure the wbc mapping index is at least updated
2817 update_nr_written(page, wbc, 0);
2819 while (delalloc_end < page_end) {
2820 nr_delalloc = find_lock_delalloc_range(inode, tree,
2825 if (nr_delalloc == 0) {
2826 delalloc_start = delalloc_end + 1;
2829 ret = tree->ops->fill_delalloc(inode, page,
2834 /* File system has been set read-only */
2840 * delalloc_end is already one less than the total
2841 * length, so we don't subtract one from
2844 delalloc_to_write += (delalloc_end - delalloc_start +
2847 delalloc_start = delalloc_end + 1;
2849 if (wbc->nr_to_write < delalloc_to_write) {
2852 if (delalloc_to_write < thresh * 2)
2853 thresh = delalloc_to_write;
2854 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2858 /* did the fill delalloc function already unlock and start
2864 * we've unlocked the page, so we can't update
2865 * the mapping's writeback index, just update
2868 wbc->nr_to_write -= nr_written;
2872 if (tree->ops && tree->ops->writepage_start_hook) {
2873 ret = tree->ops->writepage_start_hook(page, start,
2876 /* Fixup worker will requeue */
2878 wbc->pages_skipped++;
2880 redirty_page_for_writepage(wbc, page);
2881 update_nr_written(page, wbc, nr_written);
2889 * we don't want to touch the inode after unlocking the page,
2890 * so we update the mapping writeback index now
2892 update_nr_written(page, wbc, nr_written + 1);
2895 if (last_byte <= start) {
2896 if (tree->ops && tree->ops->writepage_end_io_hook)
2897 tree->ops->writepage_end_io_hook(page, start,
2902 blocksize = inode->i_sb->s_blocksize;
2904 while (cur <= end) {
2905 if (cur >= last_byte) {
2906 if (tree->ops && tree->ops->writepage_end_io_hook)
2907 tree->ops->writepage_end_io_hook(page, cur,
2911 em = epd->get_extent(inode, page, pg_offset, cur,
2913 if (IS_ERR_OR_NULL(em)) {
2918 extent_offset = cur - em->start;
2919 BUG_ON(extent_map_end(em) <= cur);
2921 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2922 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2923 sector = (em->block_start + extent_offset) >> 9;
2925 block_start = em->block_start;
2926 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2927 free_extent_map(em);
2931 * compressed and inline extents are written through other
2934 if (compressed || block_start == EXTENT_MAP_HOLE ||
2935 block_start == EXTENT_MAP_INLINE) {
2937 * end_io notification does not happen here for
2938 * compressed extents
2940 if (!compressed && tree->ops &&
2941 tree->ops->writepage_end_io_hook)
2942 tree->ops->writepage_end_io_hook(page, cur,
2945 else if (compressed) {
2946 /* we don't want to end_page_writeback on
2947 * a compressed extent. this happens
2954 pg_offset += iosize;
2957 /* leave this out until we have a page_mkwrite call */
2958 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2959 EXTENT_DIRTY, 0, NULL)) {
2961 pg_offset += iosize;
2965 if (tree->ops && tree->ops->writepage_io_hook) {
2966 ret = tree->ops->writepage_io_hook(page, cur,
2974 unsigned long max_nr = end_index + 1;
2976 set_range_writeback(tree, cur, cur + iosize - 1);
2977 if (!PageWriteback(page)) {
2978 printk(KERN_ERR "btrfs warning page %lu not "
2979 "writeback, cur %llu end %llu\n",
2980 page->index, (unsigned long long)cur,
2981 (unsigned long long)end);
2984 ret = submit_extent_page(write_flags, tree, page,
2985 sector, iosize, pg_offset,
2986 bdev, &epd->bio, max_nr,
2987 end_bio_extent_writepage,
2993 pg_offset += iosize;
2998 /* make sure the mapping tag for page dirty gets cleared */
2999 set_page_writeback(page);
3000 end_page_writeback(page);
3006 /* drop our reference on any cached states */
3007 free_extent_state(cached_state);
3012 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3013 * @mapping: address space structure to write
3014 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3015 * @writepage: function called for each page
3016 * @data: data passed to writepage function
3018 * If a page is already under I/O, write_cache_pages() skips it, even
3019 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3020 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3021 * and msync() need to guarantee that all the data which was dirty at the time
3022 * the call was made get new I/O started against them. If wbc->sync_mode is
3023 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3024 * existing IO to complete.
3026 static int extent_write_cache_pages(struct extent_io_tree *tree,
3027 struct address_space *mapping,
3028 struct writeback_control *wbc,
3029 writepage_t writepage, void *data,
3030 void (*flush_fn)(void *))
3034 int nr_to_write_done = 0;
3035 struct pagevec pvec;
3038 pgoff_t end; /* Inclusive */
3042 pagevec_init(&pvec, 0);
3043 if (wbc->range_cyclic) {
3044 index = mapping->writeback_index; /* Start from prev offset */
3047 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3048 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3051 if (wbc->sync_mode == WB_SYNC_ALL)
3052 tag = PAGECACHE_TAG_TOWRITE;
3054 tag = PAGECACHE_TAG_DIRTY;
3056 if (wbc->sync_mode == WB_SYNC_ALL)
3057 tag_pages_for_writeback(mapping, index, end);
3058 while (!done && !nr_to_write_done && (index <= end) &&
3059 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3060 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3064 for (i = 0; i < nr_pages; i++) {
3065 struct page *page = pvec.pages[i];
3068 * At this point we hold neither mapping->tree_lock nor
3069 * lock on the page itself: the page may be truncated or
3070 * invalidated (changing page->mapping to NULL), or even
3071 * swizzled back from swapper_space to tmpfs file
3075 tree->ops->write_cache_pages_lock_hook) {
3076 tree->ops->write_cache_pages_lock_hook(page,
3079 if (!trylock_page(page)) {
3085 if (unlikely(page->mapping != mapping)) {
3090 if (!wbc->range_cyclic && page->index > end) {
3096 if (wbc->sync_mode != WB_SYNC_NONE) {
3097 if (PageWriteback(page))
3099 wait_on_page_writeback(page);
3102 if (PageWriteback(page) ||
3103 !clear_page_dirty_for_io(page)) {
3108 ret = (*writepage)(page, wbc, data);
3110 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3118 * the filesystem may choose to bump up nr_to_write.
3119 * We have to make sure to honor the new nr_to_write
3122 nr_to_write_done = wbc->nr_to_write <= 0;
3124 pagevec_release(&pvec);
3127 if (!scanned && !done) {
3129 * We hit the last page and there is more work to be done: wrap
3130 * back to the start of the file
3139 static void flush_epd_write_bio(struct extent_page_data *epd)
3148 ret = submit_one_bio(rw, epd->bio, 0, 0);
3149 BUG_ON(ret < 0); /* -ENOMEM */
3154 static noinline void flush_write_bio(void *data)
3156 struct extent_page_data *epd = data;
3157 flush_epd_write_bio(epd);
3160 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3161 get_extent_t *get_extent,
3162 struct writeback_control *wbc)
3165 struct extent_page_data epd = {
3168 .get_extent = get_extent,
3170 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3173 ret = __extent_writepage(page, wbc, &epd);
3175 flush_epd_write_bio(&epd);
3179 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3180 u64 start, u64 end, get_extent_t *get_extent,
3184 struct address_space *mapping = inode->i_mapping;
3186 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3189 struct extent_page_data epd = {
3192 .get_extent = get_extent,
3194 .sync_io = mode == WB_SYNC_ALL,
3196 struct writeback_control wbc_writepages = {
3198 .nr_to_write = nr_pages * 2,
3199 .range_start = start,
3200 .range_end = end + 1,
3203 while (start <= end) {
3204 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3205 if (clear_page_dirty_for_io(page))
3206 ret = __extent_writepage(page, &wbc_writepages, &epd);
3208 if (tree->ops && tree->ops->writepage_end_io_hook)
3209 tree->ops->writepage_end_io_hook(page, start,
3210 start + PAGE_CACHE_SIZE - 1,
3214 page_cache_release(page);
3215 start += PAGE_CACHE_SIZE;
3218 flush_epd_write_bio(&epd);
3222 int extent_writepages(struct extent_io_tree *tree,
3223 struct address_space *mapping,
3224 get_extent_t *get_extent,
3225 struct writeback_control *wbc)
3228 struct extent_page_data epd = {
3231 .get_extent = get_extent,
3233 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3236 ret = extent_write_cache_pages(tree, mapping, wbc,
3237 __extent_writepage, &epd,
3239 flush_epd_write_bio(&epd);
3243 int extent_readpages(struct extent_io_tree *tree,
3244 struct address_space *mapping,
3245 struct list_head *pages, unsigned nr_pages,
3246 get_extent_t get_extent)
3248 struct bio *bio = NULL;
3250 unsigned long bio_flags = 0;
3252 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3253 struct page *page = list_entry(pages->prev, struct page, lru);
3255 prefetchw(&page->flags);
3256 list_del(&page->lru);
3257 if (!add_to_page_cache_lru(page, mapping,
3258 page->index, GFP_NOFS)) {
3259 __extent_read_full_page(tree, page, get_extent,
3260 &bio, 0, &bio_flags);
3262 page_cache_release(page);
3264 BUG_ON(!list_empty(pages));
3266 return submit_one_bio(READ, bio, 0, bio_flags);
3271 * basic invalidatepage code, this waits on any locked or writeback
3272 * ranges corresponding to the page, and then deletes any extent state
3273 * records from the tree
3275 int extent_invalidatepage(struct extent_io_tree *tree,
3276 struct page *page, unsigned long offset)
3278 struct extent_state *cached_state = NULL;
3279 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3280 u64 end = start + PAGE_CACHE_SIZE - 1;
3281 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3283 start += (offset + blocksize - 1) & ~(blocksize - 1);
3287 lock_extent_bits(tree, start, end, 0, &cached_state);
3288 wait_on_page_writeback(page);
3289 clear_extent_bit(tree, start, end,
3290 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3291 EXTENT_DO_ACCOUNTING,
3292 1, 1, &cached_state, GFP_NOFS);
3297 * a helper for releasepage, this tests for areas of the page that
3298 * are locked or under IO and drops the related state bits if it is safe
3301 int try_release_extent_state(struct extent_map_tree *map,
3302 struct extent_io_tree *tree, struct page *page,
3305 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3306 u64 end = start + PAGE_CACHE_SIZE - 1;
3309 if (test_range_bit(tree, start, end,
3310 EXTENT_IOBITS, 0, NULL))
3313 if ((mask & GFP_NOFS) == GFP_NOFS)
3316 * at this point we can safely clear everything except the
3317 * locked bit and the nodatasum bit
3319 ret = clear_extent_bit(tree, start, end,
3320 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3323 /* if clear_extent_bit failed for enomem reasons,
3324 * we can't allow the release to continue.
3335 * a helper for releasepage. As long as there are no locked extents
3336 * in the range corresponding to the page, both state records and extent
3337 * map records are removed
3339 int try_release_extent_mapping(struct extent_map_tree *map,
3340 struct extent_io_tree *tree, struct page *page,
3343 struct extent_map *em;
3344 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3345 u64 end = start + PAGE_CACHE_SIZE - 1;
3347 if ((mask & __GFP_WAIT) &&
3348 page->mapping->host->i_size > 16 * 1024 * 1024) {
3350 while (start <= end) {
3351 len = end - start + 1;
3352 write_lock(&map->lock);
3353 em = lookup_extent_mapping(map, start, len);
3355 write_unlock(&map->lock);
3358 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3359 em->start != start) {
3360 write_unlock(&map->lock);
3361 free_extent_map(em);
3364 if (!test_range_bit(tree, em->start,
3365 extent_map_end(em) - 1,
3366 EXTENT_LOCKED | EXTENT_WRITEBACK,
3368 remove_extent_mapping(map, em);
3369 /* once for the rb tree */
3370 free_extent_map(em);
3372 start = extent_map_end(em);
3373 write_unlock(&map->lock);
3376 free_extent_map(em);
3379 return try_release_extent_state(map, tree, page, mask);
3383 * helper function for fiemap, which doesn't want to see any holes.
3384 * This maps until we find something past 'last'
3386 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3389 get_extent_t *get_extent)
3391 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3392 struct extent_map *em;
3399 len = last - offset;
3402 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3403 em = get_extent(inode, NULL, 0, offset, len, 0);
3404 if (IS_ERR_OR_NULL(em))
3407 /* if this isn't a hole return it */
3408 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3409 em->block_start != EXTENT_MAP_HOLE) {
3413 /* this is a hole, advance to the next extent */
3414 offset = extent_map_end(em);
3415 free_extent_map(em);
3422 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3423 __u64 start, __u64 len, get_extent_t *get_extent)
3427 u64 max = start + len;
3431 u64 last_for_get_extent = 0;
3433 u64 isize = i_size_read(inode);
3434 struct btrfs_key found_key;
3435 struct extent_map *em = NULL;
3436 struct extent_state *cached_state = NULL;
3437 struct btrfs_path *path;
3438 struct btrfs_file_extent_item *item;
3443 unsigned long emflags;
3448 path = btrfs_alloc_path();
3451 path->leave_spinning = 1;
3453 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3454 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3457 * lookup the last file extent. We're not using i_size here
3458 * because there might be preallocation past i_size
3460 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3461 path, btrfs_ino(inode), -1, 0);
3463 btrfs_free_path(path);
3468 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3469 struct btrfs_file_extent_item);
3470 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3471 found_type = btrfs_key_type(&found_key);
3473 /* No extents, but there might be delalloc bits */
3474 if (found_key.objectid != btrfs_ino(inode) ||
3475 found_type != BTRFS_EXTENT_DATA_KEY) {
3476 /* have to trust i_size as the end */
3478 last_for_get_extent = isize;
3481 * remember the start of the last extent. There are a
3482 * bunch of different factors that go into the length of the
3483 * extent, so its much less complex to remember where it started
3485 last = found_key.offset;
3486 last_for_get_extent = last + 1;
3488 btrfs_free_path(path);
3491 * we might have some extents allocated but more delalloc past those
3492 * extents. so, we trust isize unless the start of the last extent is
3497 last_for_get_extent = isize;
3500 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3503 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3513 u64 offset_in_extent;
3515 /* break if the extent we found is outside the range */
3516 if (em->start >= max || extent_map_end(em) < off)
3520 * get_extent may return an extent that starts before our
3521 * requested range. We have to make sure the ranges
3522 * we return to fiemap always move forward and don't
3523 * overlap, so adjust the offsets here
3525 em_start = max(em->start, off);
3528 * record the offset from the start of the extent
3529 * for adjusting the disk offset below
3531 offset_in_extent = em_start - em->start;
3532 em_end = extent_map_end(em);
3533 em_len = em_end - em_start;
3534 emflags = em->flags;
3539 * bump off for our next call to get_extent
3541 off = extent_map_end(em);
3545 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3547 flags |= FIEMAP_EXTENT_LAST;
3548 } else if (em->block_start == EXTENT_MAP_INLINE) {
3549 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3550 FIEMAP_EXTENT_NOT_ALIGNED);
3551 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3552 flags |= (FIEMAP_EXTENT_DELALLOC |
3553 FIEMAP_EXTENT_UNKNOWN);
3555 disko = em->block_start + offset_in_extent;
3557 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3558 flags |= FIEMAP_EXTENT_ENCODED;
3560 free_extent_map(em);
3562 if ((em_start >= last) || em_len == (u64)-1 ||
3563 (last == (u64)-1 && isize <= em_end)) {
3564 flags |= FIEMAP_EXTENT_LAST;
3568 /* now scan forward to see if this is really the last extent. */
3569 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3576 flags |= FIEMAP_EXTENT_LAST;
3579 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3585 free_extent_map(em);
3587 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3588 &cached_state, GFP_NOFS);
3592 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3596 struct address_space *mapping;
3599 return eb->first_page;
3600 i += eb->start >> PAGE_CACHE_SHIFT;
3601 mapping = eb->first_page->mapping;
3606 * extent_buffer_page is only called after pinning the page
3607 * by increasing the reference count. So we know the page must
3608 * be in the radix tree.
3611 p = radix_tree_lookup(&mapping->page_tree, i);
3617 inline unsigned long num_extent_pages(u64 start, u64 len)
3619 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3620 (start >> PAGE_CACHE_SHIFT);
3623 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3628 struct extent_buffer *eb = NULL;
3630 unsigned long flags;
3633 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3638 rwlock_init(&eb->lock);
3639 atomic_set(&eb->write_locks, 0);
3640 atomic_set(&eb->read_locks, 0);
3641 atomic_set(&eb->blocking_readers, 0);
3642 atomic_set(&eb->blocking_writers, 0);
3643 atomic_set(&eb->spinning_readers, 0);
3644 atomic_set(&eb->spinning_writers, 0);
3645 eb->lock_nested = 0;
3646 init_waitqueue_head(&eb->write_lock_wq);
3647 init_waitqueue_head(&eb->read_lock_wq);
3650 spin_lock_irqsave(&leak_lock, flags);
3651 list_add(&eb->leak_list, &buffers);
3652 spin_unlock_irqrestore(&leak_lock, flags);
3654 atomic_set(&eb->refs, 1);
3659 static void __free_extent_buffer(struct extent_buffer *eb)
3662 unsigned long flags;
3663 spin_lock_irqsave(&leak_lock, flags);
3664 list_del(&eb->leak_list);
3665 spin_unlock_irqrestore(&leak_lock, flags);
3667 kmem_cache_free(extent_buffer_cache, eb);
3671 * Helper for releasing extent buffer page.
3673 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3674 unsigned long start_idx)
3676 unsigned long index;
3679 if (!eb->first_page)
3682 index = num_extent_pages(eb->start, eb->len);
3683 if (start_idx >= index)
3688 page = extent_buffer_page(eb, index);
3690 page_cache_release(page);
3691 } while (index != start_idx);
3695 * Helper for releasing the extent buffer.
3697 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3699 btrfs_release_extent_buffer_page(eb, 0);
3700 __free_extent_buffer(eb);
3703 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3704 u64 start, unsigned long len,
3707 unsigned long num_pages = num_extent_pages(start, len);
3709 unsigned long index = start >> PAGE_CACHE_SHIFT;
3710 struct extent_buffer *eb;
3711 struct extent_buffer *exists = NULL;
3713 struct address_space *mapping = tree->mapping;
3718 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3719 if (eb && atomic_inc_not_zero(&eb->refs)) {
3721 mark_page_accessed(eb->first_page);
3726 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
3731 eb->first_page = page0;
3734 page_cache_get(page0);
3735 mark_page_accessed(page0);
3736 set_page_extent_mapped(page0);
3737 set_page_extent_head(page0, len);
3738 uptodate = PageUptodate(page0);
3742 for (; i < num_pages; i++, index++) {
3743 p = find_or_create_page(mapping, index, GFP_NOFS);
3748 set_page_extent_mapped(p);
3749 mark_page_accessed(p);
3752 set_page_extent_head(p, len);
3754 set_page_private(p, EXTENT_PAGE_PRIVATE);
3756 if (!PageUptodate(p))
3760 * see below about how we avoid a nasty race with release page
3761 * and why we unlock later
3767 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3769 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3773 spin_lock(&tree->buffer_lock);
3774 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3775 if (ret == -EEXIST) {
3776 exists = radix_tree_lookup(&tree->buffer,
3777 start >> PAGE_CACHE_SHIFT);
3778 /* add one reference for the caller */
3779 atomic_inc(&exists->refs);
3780 spin_unlock(&tree->buffer_lock);
3781 radix_tree_preload_end();
3784 /* add one reference for the tree */
3785 atomic_inc(&eb->refs);
3786 spin_unlock(&tree->buffer_lock);
3787 radix_tree_preload_end();
3790 * there is a race where release page may have
3791 * tried to find this extent buffer in the radix
3792 * but failed. It will tell the VM it is safe to
3793 * reclaim the, and it will clear the page private bit.
3794 * We must make sure to set the page private bit properly
3795 * after the extent buffer is in the radix tree so
3796 * it doesn't get lost
3798 set_page_extent_mapped(eb->first_page);
3799 set_page_extent_head(eb->first_page, eb->len);
3801 unlock_page(eb->first_page);
3805 if (eb->first_page && !page0)
3806 unlock_page(eb->first_page);
3808 if (!atomic_dec_and_test(&eb->refs))
3810 btrfs_release_extent_buffer(eb);
3814 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3815 u64 start, unsigned long len)
3817 struct extent_buffer *eb;
3820 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3821 if (eb && atomic_inc_not_zero(&eb->refs)) {
3823 mark_page_accessed(eb->first_page);
3831 void free_extent_buffer(struct extent_buffer *eb)
3836 if (!atomic_dec_and_test(&eb->refs))
3842 void clear_extent_buffer_dirty(struct extent_io_tree *tree,
3843 struct extent_buffer *eb)
3846 unsigned long num_pages;
3849 num_pages = num_extent_pages(eb->start, eb->len);
3851 for (i = 0; i < num_pages; i++) {
3852 page = extent_buffer_page(eb, i);
3853 if (!PageDirty(page))
3857 WARN_ON(!PagePrivate(page));
3859 set_page_extent_mapped(page);
3861 set_page_extent_head(page, eb->len);
3863 clear_page_dirty_for_io(page);
3864 spin_lock_irq(&page->mapping->tree_lock);
3865 if (!PageDirty(page)) {
3866 radix_tree_tag_clear(&page->mapping->page_tree,
3868 PAGECACHE_TAG_DIRTY);
3870 spin_unlock_irq(&page->mapping->tree_lock);
3871 ClearPageError(page);
3876 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3877 struct extent_buffer *eb)
3880 unsigned long num_pages;
3883 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3884 num_pages = num_extent_pages(eb->start, eb->len);
3885 for (i = 0; i < num_pages; i++)
3886 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3890 static int __eb_straddles_pages(u64 start, u64 len)
3892 if (len < PAGE_CACHE_SIZE)
3894 if (start & (PAGE_CACHE_SIZE - 1))
3896 if ((start + len) & (PAGE_CACHE_SIZE - 1))
3901 static int eb_straddles_pages(struct extent_buffer *eb)
3903 return __eb_straddles_pages(eb->start, eb->len);
3906 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3907 struct extent_buffer *eb,
3908 struct extent_state **cached_state)
3912 unsigned long num_pages;
3914 num_pages = num_extent_pages(eb->start, eb->len);
3915 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3917 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3918 cached_state, GFP_NOFS);
3920 for (i = 0; i < num_pages; i++) {
3921 page = extent_buffer_page(eb, i);
3923 ClearPageUptodate(page);
3928 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3929 struct extent_buffer *eb)
3933 unsigned long num_pages;
3935 num_pages = num_extent_pages(eb->start, eb->len);
3937 if (eb_straddles_pages(eb)) {
3938 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3941 for (i = 0; i < num_pages; i++) {
3942 page = extent_buffer_page(eb, i);
3943 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3944 ((i == num_pages - 1) &&
3945 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3946 check_page_uptodate(tree, page);
3949 SetPageUptodate(page);
3954 int extent_range_uptodate(struct extent_io_tree *tree,
3959 int pg_uptodate = 1;
3961 unsigned long index;
3963 if (__eb_straddles_pages(start, end - start + 1)) {
3964 ret = test_range_bit(tree, start, end,
3965 EXTENT_UPTODATE, 1, NULL);
3969 while (start <= end) {
3970 index = start >> PAGE_CACHE_SHIFT;
3971 page = find_get_page(tree->mapping, index);
3974 uptodate = PageUptodate(page);
3975 page_cache_release(page);
3980 start += PAGE_CACHE_SIZE;
3985 int extent_buffer_uptodate(struct extent_io_tree *tree,
3986 struct extent_buffer *eb,
3987 struct extent_state *cached_state)
3990 unsigned long num_pages;
3993 int pg_uptodate = 1;
3995 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3998 if (eb_straddles_pages(eb)) {
3999 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
4000 EXTENT_UPTODATE, 1, cached_state);
4005 num_pages = num_extent_pages(eb->start, eb->len);
4006 for (i = 0; i < num_pages; i++) {
4007 page = extent_buffer_page(eb, i);
4008 if (!PageUptodate(page)) {
4016 int read_extent_buffer_pages(struct extent_io_tree *tree,
4017 struct extent_buffer *eb, u64 start, int wait,
4018 get_extent_t *get_extent, int mirror_num)
4021 unsigned long start_i;
4025 int locked_pages = 0;
4026 int all_uptodate = 1;
4027 int inc_all_pages = 0;
4028 unsigned long num_pages;
4029 struct bio *bio = NULL;
4030 unsigned long bio_flags = 0;
4032 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4035 if (eb_straddles_pages(eb)) {
4036 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
4037 EXTENT_UPTODATE, 1, NULL)) {
4043 WARN_ON(start < eb->start);
4044 start_i = (start >> PAGE_CACHE_SHIFT) -
4045 (eb->start >> PAGE_CACHE_SHIFT);
4050 num_pages = num_extent_pages(eb->start, eb->len);
4051 for (i = start_i; i < num_pages; i++) {
4052 page = extent_buffer_page(eb, i);
4053 if (wait == WAIT_NONE) {
4054 if (!trylock_page(page))
4060 if (!PageUptodate(page))
4065 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4069 for (i = start_i; i < num_pages; i++) {
4070 page = extent_buffer_page(eb, i);
4072 WARN_ON(!PagePrivate(page));
4074 set_page_extent_mapped(page);
4076 set_page_extent_head(page, eb->len);
4079 page_cache_get(page);
4080 if (!PageUptodate(page)) {
4083 ClearPageError(page);
4084 err = __extent_read_full_page(tree, page,
4086 mirror_num, &bio_flags);
4095 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4100 if (ret || wait != WAIT_COMPLETE)
4103 for (i = start_i; i < num_pages; i++) {
4104 page = extent_buffer_page(eb, i);
4105 wait_on_page_locked(page);
4106 if (!PageUptodate(page))
4111 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4116 while (locked_pages > 0) {
4117 page = extent_buffer_page(eb, i);
4125 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4126 unsigned long start,
4133 char *dst = (char *)dstv;
4134 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4135 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4137 WARN_ON(start > eb->len);
4138 WARN_ON(start + len > eb->start + eb->len);
4140 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4143 page = extent_buffer_page(eb, i);
4145 cur = min(len, (PAGE_CACHE_SIZE - offset));
4146 kaddr = page_address(page);
4147 memcpy(dst, kaddr + offset, cur);
4156 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4157 unsigned long min_len, char **map,
4158 unsigned long *map_start,
4159 unsigned long *map_len)
4161 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4164 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4165 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4166 unsigned long end_i = (start_offset + start + min_len - 1) >>
4173 offset = start_offset;
4177 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4180 if (start + min_len > eb->len) {
4181 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4182 "wanted %lu %lu\n", (unsigned long long)eb->start,
4183 eb->len, start, min_len);
4188 p = extent_buffer_page(eb, i);
4189 kaddr = page_address(p);
4190 *map = kaddr + offset;
4191 *map_len = PAGE_CACHE_SIZE - offset;
4195 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4196 unsigned long start,
4203 char *ptr = (char *)ptrv;
4204 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4205 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4208 WARN_ON(start > eb->len);
4209 WARN_ON(start + len > eb->start + eb->len);
4211 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4214 page = extent_buffer_page(eb, i);
4216 cur = min(len, (PAGE_CACHE_SIZE - offset));
4218 kaddr = page_address(page);
4219 ret = memcmp(ptr, kaddr + offset, cur);
4231 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4232 unsigned long start, unsigned long len)
4238 char *src = (char *)srcv;
4239 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4240 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4242 WARN_ON(start > eb->len);
4243 WARN_ON(start + len > eb->start + eb->len);
4245 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4248 page = extent_buffer_page(eb, i);
4249 WARN_ON(!PageUptodate(page));
4251 cur = min(len, PAGE_CACHE_SIZE - offset);
4252 kaddr = page_address(page);
4253 memcpy(kaddr + offset, src, cur);
4262 void memset_extent_buffer(struct extent_buffer *eb, char c,
4263 unsigned long start, unsigned long len)
4269 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4270 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4272 WARN_ON(start > eb->len);
4273 WARN_ON(start + len > eb->start + eb->len);
4275 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4278 page = extent_buffer_page(eb, i);
4279 WARN_ON(!PageUptodate(page));
4281 cur = min(len, PAGE_CACHE_SIZE - offset);
4282 kaddr = page_address(page);
4283 memset(kaddr + offset, c, cur);
4291 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4292 unsigned long dst_offset, unsigned long src_offset,
4295 u64 dst_len = dst->len;
4300 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4301 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4303 WARN_ON(src->len != dst_len);
4305 offset = (start_offset + dst_offset) &
4306 ((unsigned long)PAGE_CACHE_SIZE - 1);
4309 page = extent_buffer_page(dst, i);
4310 WARN_ON(!PageUptodate(page));
4312 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4314 kaddr = page_address(page);
4315 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4324 static void move_pages(struct page *dst_page, struct page *src_page,
4325 unsigned long dst_off, unsigned long src_off,
4328 char *dst_kaddr = page_address(dst_page);
4329 if (dst_page == src_page) {
4330 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4332 char *src_kaddr = page_address(src_page);
4333 char *p = dst_kaddr + dst_off + len;
4334 char *s = src_kaddr + src_off + len;
4341 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4343 unsigned long distance = (src > dst) ? src - dst : dst - src;
4344 return distance < len;
4347 static void copy_pages(struct page *dst_page, struct page *src_page,
4348 unsigned long dst_off, unsigned long src_off,
4351 char *dst_kaddr = page_address(dst_page);
4354 if (dst_page != src_page) {
4355 src_kaddr = page_address(src_page);
4357 src_kaddr = dst_kaddr;
4358 BUG_ON(areas_overlap(src_off, dst_off, len));
4361 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4364 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4365 unsigned long src_offset, unsigned long len)
4368 size_t dst_off_in_page;
4369 size_t src_off_in_page;
4370 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4371 unsigned long dst_i;
4372 unsigned long src_i;
4374 if (src_offset + len > dst->len) {
4375 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4376 "len %lu dst len %lu\n", src_offset, len, dst->len);
4379 if (dst_offset + len > dst->len) {
4380 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4381 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4386 dst_off_in_page = (start_offset + dst_offset) &
4387 ((unsigned long)PAGE_CACHE_SIZE - 1);
4388 src_off_in_page = (start_offset + src_offset) &
4389 ((unsigned long)PAGE_CACHE_SIZE - 1);
4391 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4392 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4394 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4396 cur = min_t(unsigned long, cur,
4397 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4399 copy_pages(extent_buffer_page(dst, dst_i),
4400 extent_buffer_page(dst, src_i),
4401 dst_off_in_page, src_off_in_page, cur);
4409 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4410 unsigned long src_offset, unsigned long len)
4413 size_t dst_off_in_page;
4414 size_t src_off_in_page;
4415 unsigned long dst_end = dst_offset + len - 1;
4416 unsigned long src_end = src_offset + len - 1;
4417 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4418 unsigned long dst_i;
4419 unsigned long src_i;
4421 if (src_offset + len > dst->len) {
4422 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4423 "len %lu len %lu\n", src_offset, len, dst->len);
4426 if (dst_offset + len > dst->len) {
4427 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4428 "len %lu len %lu\n", dst_offset, len, dst->len);
4431 if (!areas_overlap(src_offset, dst_offset, len)) {
4432 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4436 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4437 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4439 dst_off_in_page = (start_offset + dst_end) &
4440 ((unsigned long)PAGE_CACHE_SIZE - 1);
4441 src_off_in_page = (start_offset + src_end) &
4442 ((unsigned long)PAGE_CACHE_SIZE - 1);
4444 cur = min_t(unsigned long, len, src_off_in_page + 1);
4445 cur = min(cur, dst_off_in_page + 1);
4446 move_pages(extent_buffer_page(dst, dst_i),
4447 extent_buffer_page(dst, src_i),
4448 dst_off_in_page - cur + 1,
4449 src_off_in_page - cur + 1, cur);
4457 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4459 struct extent_buffer *eb =
4460 container_of(head, struct extent_buffer, rcu_head);
4462 btrfs_release_extent_buffer(eb);
4465 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
4467 u64 start = page_offset(page);
4468 struct extent_buffer *eb;
4471 spin_lock(&tree->buffer_lock);
4472 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4474 spin_unlock(&tree->buffer_lock);
4478 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4484 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4487 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
4492 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4494 spin_unlock(&tree->buffer_lock);
4496 /* at this point we can safely release the extent buffer */
4497 if (atomic_read(&eb->refs) == 0)
4498 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);