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"
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
27 static LIST_HEAD(buffers);
28 static LIST_HEAD(states);
32 static DEFINE_SPINLOCK(leak_lock);
35 #define BUFFER_LRU_MAX 64
40 struct rb_node rb_node;
43 struct extent_page_data {
45 struct extent_io_tree *tree;
46 get_extent_t *get_extent;
48 /* tells writepage not to lock the state bits for this range
49 * it still does the unlocking
51 unsigned int extent_locked:1;
53 /* tells the submit_bio code to use a WRITE_SYNC */
54 unsigned int sync_io:1;
57 static noinline void flush_write_bio(void *data);
58 static inline struct btrfs_fs_info *
59 tree_fs_info(struct extent_io_tree *tree)
61 return btrfs_sb(tree->mapping->host->i_sb);
64 int __init extent_io_init(void)
66 extent_state_cache = kmem_cache_create("extent_state",
67 sizeof(struct extent_state), 0,
68 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
69 if (!extent_state_cache)
72 extent_buffer_cache = kmem_cache_create("extent_buffers",
73 sizeof(struct extent_buffer), 0,
74 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
75 if (!extent_buffer_cache)
76 goto free_state_cache;
80 kmem_cache_destroy(extent_state_cache);
84 void extent_io_exit(void)
86 struct extent_state *state;
87 struct extent_buffer *eb;
89 while (!list_empty(&states)) {
90 state = list_entry(states.next, struct extent_state, leak_list);
91 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
92 "state %lu in tree %p refs %d\n",
93 (unsigned long long)state->start,
94 (unsigned long long)state->end,
95 state->state, state->tree, atomic_read(&state->refs));
96 list_del(&state->leak_list);
97 kmem_cache_free(extent_state_cache, state);
101 while (!list_empty(&buffers)) {
102 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
103 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
104 "refs %d\n", (unsigned long long)eb->start,
105 eb->len, atomic_read(&eb->refs));
106 list_del(&eb->leak_list);
107 kmem_cache_free(extent_buffer_cache, eb);
109 if (extent_state_cache)
110 kmem_cache_destroy(extent_state_cache);
111 if (extent_buffer_cache)
112 kmem_cache_destroy(extent_buffer_cache);
115 void extent_io_tree_init(struct extent_io_tree *tree,
116 struct address_space *mapping)
118 tree->state = RB_ROOT;
119 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
121 tree->dirty_bytes = 0;
122 spin_lock_init(&tree->lock);
123 spin_lock_init(&tree->buffer_lock);
124 tree->mapping = mapping;
127 static struct extent_state *alloc_extent_state(gfp_t mask)
129 struct extent_state *state;
134 state = kmem_cache_alloc(extent_state_cache, mask);
141 spin_lock_irqsave(&leak_lock, flags);
142 list_add(&state->leak_list, &states);
143 spin_unlock_irqrestore(&leak_lock, flags);
145 atomic_set(&state->refs, 1);
146 init_waitqueue_head(&state->wq);
147 trace_alloc_extent_state(state, mask, _RET_IP_);
151 void free_extent_state(struct extent_state *state)
155 if (atomic_dec_and_test(&state->refs)) {
159 WARN_ON(state->tree);
161 spin_lock_irqsave(&leak_lock, flags);
162 list_del(&state->leak_list);
163 spin_unlock_irqrestore(&leak_lock, flags);
165 trace_free_extent_state(state, _RET_IP_);
166 kmem_cache_free(extent_state_cache, state);
170 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
171 struct rb_node *node)
173 struct rb_node **p = &root->rb_node;
174 struct rb_node *parent = NULL;
175 struct tree_entry *entry;
179 entry = rb_entry(parent, struct tree_entry, rb_node);
181 if (offset < entry->start)
183 else if (offset > entry->end)
189 entry = rb_entry(node, struct tree_entry, rb_node);
190 rb_link_node(node, parent, p);
191 rb_insert_color(node, root);
195 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
196 struct rb_node **prev_ret,
197 struct rb_node **next_ret)
199 struct rb_root *root = &tree->state;
200 struct rb_node *n = root->rb_node;
201 struct rb_node *prev = NULL;
202 struct rb_node *orig_prev = NULL;
203 struct tree_entry *entry;
204 struct tree_entry *prev_entry = NULL;
207 entry = rb_entry(n, struct tree_entry, rb_node);
211 if (offset < entry->start)
213 else if (offset > entry->end)
221 while (prev && offset > prev_entry->end) {
222 prev = rb_next(prev);
223 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
230 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
231 while (prev && offset < prev_entry->start) {
232 prev = rb_prev(prev);
233 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
240 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
243 struct rb_node *prev = NULL;
246 ret = __etree_search(tree, offset, &prev, NULL);
252 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
253 struct extent_state *other)
255 if (tree->ops && tree->ops->merge_extent_hook)
256 tree->ops->merge_extent_hook(tree->mapping->host, new,
261 * utility function to look for merge candidates inside a given range.
262 * Any extents with matching state are merged together into a single
263 * extent in the tree. Extents with EXTENT_IO in their state field
264 * are not merged because the end_io handlers need to be able to do
265 * operations on them without sleeping (or doing allocations/splits).
267 * This should be called with the tree lock held.
269 static void merge_state(struct extent_io_tree *tree,
270 struct extent_state *state)
272 struct extent_state *other;
273 struct rb_node *other_node;
275 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
278 other_node = rb_prev(&state->rb_node);
280 other = rb_entry(other_node, struct extent_state, rb_node);
281 if (other->end == state->start - 1 &&
282 other->state == state->state) {
283 merge_cb(tree, state, other);
284 state->start = other->start;
286 rb_erase(&other->rb_node, &tree->state);
287 free_extent_state(other);
290 other_node = rb_next(&state->rb_node);
292 other = rb_entry(other_node, struct extent_state, rb_node);
293 if (other->start == state->end + 1 &&
294 other->state == state->state) {
295 merge_cb(tree, state, other);
296 state->end = other->end;
298 rb_erase(&other->rb_node, &tree->state);
299 free_extent_state(other);
304 static void set_state_cb(struct extent_io_tree *tree,
305 struct extent_state *state, int *bits)
307 if (tree->ops && tree->ops->set_bit_hook)
308 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
311 static void clear_state_cb(struct extent_io_tree *tree,
312 struct extent_state *state, int *bits)
314 if (tree->ops && tree->ops->clear_bit_hook)
315 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
318 static void set_state_bits(struct extent_io_tree *tree,
319 struct extent_state *state, int *bits);
322 * insert an extent_state struct into the tree. 'bits' are set on the
323 * struct before it is inserted.
325 * This may return -EEXIST if the extent is already there, in which case the
326 * state struct is freed.
328 * The tree lock is not taken internally. This is a utility function and
329 * probably isn't what you want to call (see set/clear_extent_bit).
331 static int insert_state(struct extent_io_tree *tree,
332 struct extent_state *state, u64 start, u64 end,
335 struct rb_node *node;
338 printk(KERN_ERR "btrfs end < start %llu %llu\n",
339 (unsigned long long)end,
340 (unsigned long long)start);
343 state->start = start;
346 set_state_bits(tree, state, bits);
348 node = tree_insert(&tree->state, end, &state->rb_node);
350 struct extent_state *found;
351 found = rb_entry(node, struct extent_state, rb_node);
352 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
353 "%llu %llu\n", (unsigned long long)found->start,
354 (unsigned long long)found->end,
355 (unsigned long long)start, (unsigned long long)end);
359 merge_state(tree, state);
363 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
366 if (tree->ops && tree->ops->split_extent_hook)
367 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
371 * split a given extent state struct in two, inserting the preallocated
372 * struct 'prealloc' as the newly created second half. 'split' indicates an
373 * offset inside 'orig' where it should be split.
376 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
377 * are two extent state structs in the tree:
378 * prealloc: [orig->start, split - 1]
379 * orig: [ split, orig->end ]
381 * The tree locks are not taken by this function. They need to be held
384 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
385 struct extent_state *prealloc, u64 split)
387 struct rb_node *node;
389 split_cb(tree, orig, split);
391 prealloc->start = orig->start;
392 prealloc->end = split - 1;
393 prealloc->state = orig->state;
396 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
398 free_extent_state(prealloc);
401 prealloc->tree = tree;
406 * utility function to clear some bits in an extent state struct.
407 * it will optionally wake up any one waiting on this state (wake == 1), or
408 * forcibly remove the state from the tree (delete == 1).
410 * If no bits are set on the state struct after clearing things, the
411 * struct is freed and removed from the tree
413 static int clear_state_bit(struct extent_io_tree *tree,
414 struct extent_state *state,
417 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
418 int ret = state->state & bits_to_clear;
420 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
421 u64 range = state->end - state->start + 1;
422 WARN_ON(range > tree->dirty_bytes);
423 tree->dirty_bytes -= range;
425 clear_state_cb(tree, state, bits);
426 state->state &= ~bits_to_clear;
429 if (state->state == 0) {
431 rb_erase(&state->rb_node, &tree->state);
433 free_extent_state(state);
438 merge_state(tree, state);
443 static struct extent_state *
444 alloc_extent_state_atomic(struct extent_state *prealloc)
447 prealloc = alloc_extent_state(GFP_ATOMIC);
452 void extent_io_tree_panic(struct extent_io_tree *tree, int err)
454 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
455 "Extent tree was modified by another "
456 "thread while locked.");
460 * clear some bits on a range in the tree. This may require splitting
461 * or inserting elements in the tree, so the gfp mask is used to
462 * indicate which allocations or sleeping are allowed.
464 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
465 * the given range from the tree regardless of state (ie for truncate).
467 * the range [start, end] is inclusive.
469 * This takes the tree lock, and returns 0 on success and < 0 on error.
471 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
472 int bits, int wake, int delete,
473 struct extent_state **cached_state,
476 struct extent_state *state;
477 struct extent_state *cached;
478 struct extent_state *prealloc = NULL;
479 struct rb_node *next_node;
480 struct rb_node *node;
486 bits |= ~EXTENT_CTLBITS;
487 bits |= EXTENT_FIRST_DELALLOC;
489 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
492 if (!prealloc && (mask & __GFP_WAIT)) {
493 prealloc = alloc_extent_state(mask);
498 spin_lock(&tree->lock);
500 cached = *cached_state;
503 *cached_state = NULL;
507 if (cached && cached->tree && cached->start <= start &&
508 cached->end > start) {
510 atomic_dec(&cached->refs);
515 free_extent_state(cached);
518 * this search will find the extents that end after
521 node = tree_search(tree, start);
524 state = rb_entry(node, struct extent_state, rb_node);
526 if (state->start > end)
528 WARN_ON(state->end < start);
529 last_end = state->end;
531 if (state->end < end && !need_resched())
532 next_node = rb_next(&state->rb_node);
536 /* the state doesn't have the wanted bits, go ahead */
537 if (!(state->state & bits))
541 * | ---- desired range ---- |
543 * | ------------- state -------------- |
545 * We need to split the extent we found, and may flip
546 * bits on second half.
548 * If the extent we found extends past our range, we
549 * just split and search again. It'll get split again
550 * the next time though.
552 * If the extent we found is inside our range, we clear
553 * the desired bit on it.
556 if (state->start < start) {
557 prealloc = alloc_extent_state_atomic(prealloc);
559 err = split_state(tree, state, prealloc, start);
561 extent_io_tree_panic(tree, err);
566 if (state->end <= end) {
567 clear_state_bit(tree, state, &bits, wake);
568 if (last_end == (u64)-1)
570 start = last_end + 1;
575 * | ---- desired range ---- |
577 * We need to split the extent, and clear the bit
580 if (state->start <= end && state->end > end) {
581 prealloc = alloc_extent_state_atomic(prealloc);
583 err = split_state(tree, state, prealloc, end + 1);
585 extent_io_tree_panic(tree, err);
590 clear_state_bit(tree, prealloc, &bits, wake);
596 clear_state_bit(tree, state, &bits, wake);
598 if (last_end == (u64)-1)
600 start = last_end + 1;
601 if (start <= end && next_node) {
602 state = rb_entry(next_node, struct extent_state,
609 spin_unlock(&tree->lock);
611 free_extent_state(prealloc);
618 spin_unlock(&tree->lock);
619 if (mask & __GFP_WAIT)
624 static void wait_on_state(struct extent_io_tree *tree,
625 struct extent_state *state)
626 __releases(tree->lock)
627 __acquires(tree->lock)
630 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
631 spin_unlock(&tree->lock);
633 spin_lock(&tree->lock);
634 finish_wait(&state->wq, &wait);
638 * waits for one or more bits to clear on a range in the state tree.
639 * The range [start, end] is inclusive.
640 * The tree lock is taken by this function
642 void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
644 struct extent_state *state;
645 struct rb_node *node;
647 spin_lock(&tree->lock);
651 * this search will find all the extents that end after
654 node = tree_search(tree, start);
658 state = rb_entry(node, struct extent_state, rb_node);
660 if (state->start > end)
663 if (state->state & bits) {
664 start = state->start;
665 atomic_inc(&state->refs);
666 wait_on_state(tree, state);
667 free_extent_state(state);
670 start = state->end + 1;
675 cond_resched_lock(&tree->lock);
678 spin_unlock(&tree->lock);
681 static void set_state_bits(struct extent_io_tree *tree,
682 struct extent_state *state,
685 int bits_to_set = *bits & ~EXTENT_CTLBITS;
687 set_state_cb(tree, state, bits);
688 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
689 u64 range = state->end - state->start + 1;
690 tree->dirty_bytes += range;
692 state->state |= bits_to_set;
695 static void cache_state(struct extent_state *state,
696 struct extent_state **cached_ptr)
698 if (cached_ptr && !(*cached_ptr)) {
699 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
701 atomic_inc(&state->refs);
706 static void uncache_state(struct extent_state **cached_ptr)
708 if (cached_ptr && (*cached_ptr)) {
709 struct extent_state *state = *cached_ptr;
711 free_extent_state(state);
716 * set some bits on a range in the tree. This may require allocations or
717 * sleeping, so the gfp mask is used to indicate what is allowed.
719 * If any of the exclusive bits are set, this will fail with -EEXIST if some
720 * part of the range already has the desired bits set. The start of the
721 * existing range is returned in failed_start in this case.
723 * [start, end] is inclusive This takes the tree lock.
726 static int __must_check
727 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
728 int bits, int exclusive_bits, u64 *failed_start,
729 struct extent_state **cached_state, gfp_t mask)
731 struct extent_state *state;
732 struct extent_state *prealloc = NULL;
733 struct rb_node *node;
738 bits |= EXTENT_FIRST_DELALLOC;
740 if (!prealloc && (mask & __GFP_WAIT)) {
741 prealloc = alloc_extent_state(mask);
745 spin_lock(&tree->lock);
746 if (cached_state && *cached_state) {
747 state = *cached_state;
748 if (state->start <= start && state->end > start &&
750 node = &state->rb_node;
755 * this search will find all the extents that end after
758 node = tree_search(tree, start);
760 prealloc = alloc_extent_state_atomic(prealloc);
762 err = insert_state(tree, prealloc, start, end, &bits);
764 extent_io_tree_panic(tree, err);
769 state = rb_entry(node, struct extent_state, rb_node);
771 last_start = state->start;
772 last_end = state->end;
775 * | ---- desired range ---- |
778 * Just lock what we found and keep going
780 if (state->start == start && state->end <= end) {
781 struct rb_node *next_node;
782 if (state->state & exclusive_bits) {
783 *failed_start = state->start;
788 set_state_bits(tree, state, &bits);
790 cache_state(state, cached_state);
791 merge_state(tree, state);
792 if (last_end == (u64)-1)
795 start = last_end + 1;
796 next_node = rb_next(&state->rb_node);
797 if (next_node && start < end && prealloc && !need_resched()) {
798 state = rb_entry(next_node, struct extent_state,
800 if (state->start == start)
807 * | ---- desired range ---- |
810 * | ------------- state -------------- |
812 * We need to split the extent we found, and may flip bits on
815 * If the extent we found extends past our
816 * range, we just split and search again. It'll get split
817 * again the next time though.
819 * If the extent we found is inside our range, we set the
822 if (state->start < start) {
823 if (state->state & exclusive_bits) {
824 *failed_start = start;
829 prealloc = alloc_extent_state_atomic(prealloc);
831 err = split_state(tree, state, prealloc, start);
833 extent_io_tree_panic(tree, err);
838 if (state->end <= end) {
839 set_state_bits(tree, state, &bits);
840 cache_state(state, cached_state);
841 merge_state(tree, state);
842 if (last_end == (u64)-1)
844 start = last_end + 1;
849 * | ---- desired range ---- |
850 * | state | or | state |
852 * There's a hole, we need to insert something in it and
853 * ignore the extent we found.
855 if (state->start > start) {
857 if (end < last_start)
860 this_end = last_start - 1;
862 prealloc = alloc_extent_state_atomic(prealloc);
866 * Avoid to free 'prealloc' if it can be merged with
869 err = insert_state(tree, prealloc, start, this_end,
872 extent_io_tree_panic(tree, err);
874 cache_state(prealloc, cached_state);
876 start = this_end + 1;
880 * | ---- desired range ---- |
882 * We need to split the extent, and set the bit
885 if (state->start <= end && state->end > end) {
886 if (state->state & exclusive_bits) {
887 *failed_start = start;
892 prealloc = alloc_extent_state_atomic(prealloc);
894 err = split_state(tree, state, prealloc, end + 1);
896 extent_io_tree_panic(tree, err);
898 set_state_bits(tree, prealloc, &bits);
899 cache_state(prealloc, cached_state);
900 merge_state(tree, prealloc);
908 spin_unlock(&tree->lock);
910 free_extent_state(prealloc);
917 spin_unlock(&tree->lock);
918 if (mask & __GFP_WAIT)
923 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
924 u64 *failed_start, struct extent_state **cached_state,
927 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
933 * convert_extent - convert all bits in a given range from one bit to another
934 * @tree: the io tree to search
935 * @start: the start offset in bytes
936 * @end: the end offset in bytes (inclusive)
937 * @bits: the bits to set in this range
938 * @clear_bits: the bits to clear in this range
939 * @mask: the allocation mask
941 * This will go through and set bits for the given range. If any states exist
942 * already in this range they are set with the given bit and cleared of the
943 * clear_bits. This is only meant to be used by things that are mergeable, ie
944 * converting from say DELALLOC to DIRTY. This is not meant to be used with
945 * boundary bits like LOCK.
947 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
948 int bits, int clear_bits, gfp_t mask)
950 struct extent_state *state;
951 struct extent_state *prealloc = NULL;
952 struct rb_node *node;
958 if (!prealloc && (mask & __GFP_WAIT)) {
959 prealloc = alloc_extent_state(mask);
964 spin_lock(&tree->lock);
966 * this search will find all the extents that end after
969 node = tree_search(tree, start);
971 prealloc = alloc_extent_state_atomic(prealloc);
976 err = insert_state(tree, prealloc, start, end, &bits);
979 extent_io_tree_panic(tree, err);
982 state = rb_entry(node, struct extent_state, rb_node);
984 last_start = state->start;
985 last_end = state->end;
988 * | ---- desired range ---- |
991 * Just lock what we found and keep going
993 if (state->start == start && state->end <= end) {
994 struct rb_node *next_node;
996 set_state_bits(tree, state, &bits);
997 clear_state_bit(tree, state, &clear_bits, 0);
998 if (last_end == (u64)-1)
1001 start = last_end + 1;
1002 next_node = rb_next(&state->rb_node);
1003 if (next_node && start < end && prealloc && !need_resched()) {
1004 state = rb_entry(next_node, struct extent_state,
1006 if (state->start == start)
1013 * | ---- desired range ---- |
1016 * | ------------- state -------------- |
1018 * We need to split the extent we found, and may flip bits on
1021 * If the extent we found extends past our
1022 * range, we just split and search again. It'll get split
1023 * again the next time though.
1025 * If the extent we found is inside our range, we set the
1026 * desired bit on it.
1028 if (state->start < start) {
1029 prealloc = alloc_extent_state_atomic(prealloc);
1034 err = split_state(tree, state, prealloc, start);
1036 extent_io_tree_panic(tree, err);
1040 if (state->end <= end) {
1041 set_state_bits(tree, state, &bits);
1042 clear_state_bit(tree, state, &clear_bits, 0);
1043 if (last_end == (u64)-1)
1045 start = last_end + 1;
1050 * | ---- desired range ---- |
1051 * | state | or | state |
1053 * There's a hole, we need to insert something in it and
1054 * ignore the extent we found.
1056 if (state->start > start) {
1058 if (end < last_start)
1061 this_end = last_start - 1;
1063 prealloc = alloc_extent_state_atomic(prealloc);
1070 * Avoid to free 'prealloc' if it can be merged with
1073 err = insert_state(tree, prealloc, start, this_end,
1076 extent_io_tree_panic(tree, err);
1078 start = this_end + 1;
1082 * | ---- desired range ---- |
1084 * We need to split the extent, and set the bit
1087 if (state->start <= end && state->end > end) {
1088 prealloc = alloc_extent_state_atomic(prealloc);
1094 err = split_state(tree, state, prealloc, end + 1);
1096 extent_io_tree_panic(tree, err);
1098 set_state_bits(tree, prealloc, &bits);
1099 clear_state_bit(tree, prealloc, &clear_bits, 0);
1107 spin_unlock(&tree->lock);
1109 free_extent_state(prealloc);
1116 spin_unlock(&tree->lock);
1117 if (mask & __GFP_WAIT)
1122 /* wrappers around set/clear extent bit */
1123 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1126 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1130 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1131 int bits, gfp_t mask)
1133 return set_extent_bit(tree, start, end, bits, NULL,
1137 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1138 int bits, gfp_t mask)
1140 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1143 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1144 struct extent_state **cached_state, gfp_t mask)
1146 return set_extent_bit(tree, start, end,
1147 EXTENT_DELALLOC | EXTENT_UPTODATE,
1148 NULL, cached_state, mask);
1151 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1154 return clear_extent_bit(tree, start, end,
1155 EXTENT_DIRTY | EXTENT_DELALLOC |
1156 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1159 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1162 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1166 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1167 struct extent_state **cached_state, gfp_t mask)
1169 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1170 cached_state, mask);
1173 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
1174 u64 end, struct extent_state **cached_state,
1177 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1178 cached_state, mask);
1182 * either insert or lock state struct between start and end use mask to tell
1183 * us if waiting is desired.
1185 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1186 int bits, struct extent_state **cached_state)
1191 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1192 EXTENT_LOCKED, &failed_start,
1193 cached_state, GFP_NOFS);
1194 if (err == -EEXIST) {
1195 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1196 start = failed_start;
1199 WARN_ON(start > end);
1204 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1206 return lock_extent_bits(tree, start, end, 0, NULL);
1209 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1214 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1215 &failed_start, NULL, GFP_NOFS);
1216 if (err == -EEXIST) {
1217 if (failed_start > start)
1218 clear_extent_bit(tree, start, failed_start - 1,
1219 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1225 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1226 struct extent_state **cached, gfp_t mask)
1228 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1232 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1234 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1239 * helper function to set both pages and extents in the tree writeback
1241 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1243 unsigned long index = start >> PAGE_CACHE_SHIFT;
1244 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1247 while (index <= end_index) {
1248 page = find_get_page(tree->mapping, index);
1249 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1250 set_page_writeback(page);
1251 page_cache_release(page);
1257 /* find the first state struct with 'bits' set after 'start', and
1258 * return it. tree->lock must be held. NULL will returned if
1259 * nothing was found after 'start'
1261 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1262 u64 start, int bits)
1264 struct rb_node *node;
1265 struct extent_state *state;
1268 * this search will find all the extents that end after
1271 node = tree_search(tree, start);
1276 state = rb_entry(node, struct extent_state, rb_node);
1277 if (state->end >= start && (state->state & bits))
1280 node = rb_next(node);
1289 * find the first offset in the io tree with 'bits' set. zero is
1290 * returned if we find something, and *start_ret and *end_ret are
1291 * set to reflect the state struct that was found.
1293 * If nothing was found, 1 is returned, < 0 on error
1295 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1296 u64 *start_ret, u64 *end_ret, int bits)
1298 struct extent_state *state;
1301 spin_lock(&tree->lock);
1302 state = find_first_extent_bit_state(tree, start, bits);
1304 *start_ret = state->start;
1305 *end_ret = state->end;
1308 spin_unlock(&tree->lock);
1313 * find a contiguous range of bytes in the file marked as delalloc, not
1314 * more than 'max_bytes'. start and end are used to return the range,
1316 * 1 is returned if we find something, 0 if nothing was in the tree
1318 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1319 u64 *start, u64 *end, u64 max_bytes,
1320 struct extent_state **cached_state)
1322 struct rb_node *node;
1323 struct extent_state *state;
1324 u64 cur_start = *start;
1326 u64 total_bytes = 0;
1328 spin_lock(&tree->lock);
1331 * this search will find all the extents that end after
1334 node = tree_search(tree, cur_start);
1342 state = rb_entry(node, struct extent_state, rb_node);
1343 if (found && (state->start != cur_start ||
1344 (state->state & EXTENT_BOUNDARY))) {
1347 if (!(state->state & EXTENT_DELALLOC)) {
1353 *start = state->start;
1354 *cached_state = state;
1355 atomic_inc(&state->refs);
1359 cur_start = state->end + 1;
1360 node = rb_next(node);
1363 total_bytes += state->end - state->start + 1;
1364 if (total_bytes >= max_bytes)
1368 spin_unlock(&tree->lock);
1372 static noinline void __unlock_for_delalloc(struct inode *inode,
1373 struct page *locked_page,
1377 struct page *pages[16];
1378 unsigned long index = start >> PAGE_CACHE_SHIFT;
1379 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1380 unsigned long nr_pages = end_index - index + 1;
1383 if (index == locked_page->index && end_index == index)
1386 while (nr_pages > 0) {
1387 ret = find_get_pages_contig(inode->i_mapping, index,
1388 min_t(unsigned long, nr_pages,
1389 ARRAY_SIZE(pages)), pages);
1390 for (i = 0; i < ret; i++) {
1391 if (pages[i] != locked_page)
1392 unlock_page(pages[i]);
1393 page_cache_release(pages[i]);
1401 static noinline int lock_delalloc_pages(struct inode *inode,
1402 struct page *locked_page,
1406 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1407 unsigned long start_index = index;
1408 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1409 unsigned long pages_locked = 0;
1410 struct page *pages[16];
1411 unsigned long nrpages;
1415 /* the caller is responsible for locking the start index */
1416 if (index == locked_page->index && index == end_index)
1419 /* skip the page at the start index */
1420 nrpages = end_index - index + 1;
1421 while (nrpages > 0) {
1422 ret = find_get_pages_contig(inode->i_mapping, index,
1423 min_t(unsigned long,
1424 nrpages, ARRAY_SIZE(pages)), pages);
1429 /* now we have an array of pages, lock them all */
1430 for (i = 0; i < ret; i++) {
1432 * the caller is taking responsibility for
1435 if (pages[i] != locked_page) {
1436 lock_page(pages[i]);
1437 if (!PageDirty(pages[i]) ||
1438 pages[i]->mapping != inode->i_mapping) {
1440 unlock_page(pages[i]);
1441 page_cache_release(pages[i]);
1445 page_cache_release(pages[i]);
1454 if (ret && pages_locked) {
1455 __unlock_for_delalloc(inode, locked_page,
1457 ((u64)(start_index + pages_locked - 1)) <<
1464 * find a contiguous range of bytes in the file marked as delalloc, not
1465 * more than 'max_bytes'. start and end are used to return the range,
1467 * 1 is returned if we find something, 0 if nothing was in the tree
1469 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1470 struct extent_io_tree *tree,
1471 struct page *locked_page,
1472 u64 *start, u64 *end,
1478 struct extent_state *cached_state = NULL;
1483 /* step one, find a bunch of delalloc bytes starting at start */
1484 delalloc_start = *start;
1486 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1487 max_bytes, &cached_state);
1488 if (!found || delalloc_end <= *start) {
1489 *start = delalloc_start;
1490 *end = delalloc_end;
1491 free_extent_state(cached_state);
1496 * start comes from the offset of locked_page. We have to lock
1497 * pages in order, so we can't process delalloc bytes before
1500 if (delalloc_start < *start)
1501 delalloc_start = *start;
1504 * make sure to limit the number of pages we try to lock down
1507 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1508 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1510 /* step two, lock all the pages after the page that has start */
1511 ret = lock_delalloc_pages(inode, locked_page,
1512 delalloc_start, delalloc_end);
1513 if (ret == -EAGAIN) {
1514 /* some of the pages are gone, lets avoid looping by
1515 * shortening the size of the delalloc range we're searching
1517 free_extent_state(cached_state);
1519 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1520 max_bytes = PAGE_CACHE_SIZE - offset;
1528 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1530 /* step three, lock the state bits for the whole range */
1531 lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1533 /* then test to make sure it is all still delalloc */
1534 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1535 EXTENT_DELALLOC, 1, cached_state);
1537 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1538 &cached_state, GFP_NOFS);
1539 __unlock_for_delalloc(inode, locked_page,
1540 delalloc_start, delalloc_end);
1544 free_extent_state(cached_state);
1545 *start = delalloc_start;
1546 *end = delalloc_end;
1551 int extent_clear_unlock_delalloc(struct inode *inode,
1552 struct extent_io_tree *tree,
1553 u64 start, u64 end, struct page *locked_page,
1557 struct page *pages[16];
1558 unsigned long index = start >> PAGE_CACHE_SHIFT;
1559 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1560 unsigned long nr_pages = end_index - index + 1;
1564 if (op & EXTENT_CLEAR_UNLOCK)
1565 clear_bits |= EXTENT_LOCKED;
1566 if (op & EXTENT_CLEAR_DIRTY)
1567 clear_bits |= EXTENT_DIRTY;
1569 if (op & EXTENT_CLEAR_DELALLOC)
1570 clear_bits |= EXTENT_DELALLOC;
1572 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1573 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1574 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1575 EXTENT_SET_PRIVATE2)))
1578 while (nr_pages > 0) {
1579 ret = find_get_pages_contig(inode->i_mapping, index,
1580 min_t(unsigned long,
1581 nr_pages, ARRAY_SIZE(pages)), pages);
1582 for (i = 0; i < ret; i++) {
1584 if (op & EXTENT_SET_PRIVATE2)
1585 SetPagePrivate2(pages[i]);
1587 if (pages[i] == locked_page) {
1588 page_cache_release(pages[i]);
1591 if (op & EXTENT_CLEAR_DIRTY)
1592 clear_page_dirty_for_io(pages[i]);
1593 if (op & EXTENT_SET_WRITEBACK)
1594 set_page_writeback(pages[i]);
1595 if (op & EXTENT_END_WRITEBACK)
1596 end_page_writeback(pages[i]);
1597 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1598 unlock_page(pages[i]);
1599 page_cache_release(pages[i]);
1609 * count the number of bytes in the tree that have a given bit(s)
1610 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1611 * cached. The total number found is returned.
1613 u64 count_range_bits(struct extent_io_tree *tree,
1614 u64 *start, u64 search_end, u64 max_bytes,
1615 unsigned long bits, int contig)
1617 struct rb_node *node;
1618 struct extent_state *state;
1619 u64 cur_start = *start;
1620 u64 total_bytes = 0;
1624 if (search_end <= cur_start) {
1629 spin_lock(&tree->lock);
1630 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1631 total_bytes = tree->dirty_bytes;
1635 * this search will find all the extents that end after
1638 node = tree_search(tree, cur_start);
1643 state = rb_entry(node, struct extent_state, rb_node);
1644 if (state->start > search_end)
1646 if (contig && found && state->start > last + 1)
1648 if (state->end >= cur_start && (state->state & bits) == bits) {
1649 total_bytes += min(search_end, state->end) + 1 -
1650 max(cur_start, state->start);
1651 if (total_bytes >= max_bytes)
1654 *start = max(cur_start, state->start);
1658 } else if (contig && found) {
1661 node = rb_next(node);
1666 spin_unlock(&tree->lock);
1671 * set the private field for a given byte offset in the tree. If there isn't
1672 * an extent_state there already, this does nothing.
1674 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1676 struct rb_node *node;
1677 struct extent_state *state;
1680 spin_lock(&tree->lock);
1682 * this search will find all the extents that end after
1685 node = tree_search(tree, start);
1690 state = rb_entry(node, struct extent_state, rb_node);
1691 if (state->start != start) {
1695 state->private = private;
1697 spin_unlock(&tree->lock);
1701 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1703 struct rb_node *node;
1704 struct extent_state *state;
1707 spin_lock(&tree->lock);
1709 * this search will find all the extents that end after
1712 node = tree_search(tree, start);
1717 state = rb_entry(node, struct extent_state, rb_node);
1718 if (state->start != start) {
1722 *private = state->private;
1724 spin_unlock(&tree->lock);
1729 * searches a range in the state tree for a given mask.
1730 * If 'filled' == 1, this returns 1 only if every extent in the tree
1731 * has the bits set. Otherwise, 1 is returned if any bit in the
1732 * range is found set.
1734 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1735 int bits, int filled, struct extent_state *cached)
1737 struct extent_state *state = NULL;
1738 struct rb_node *node;
1741 spin_lock(&tree->lock);
1742 if (cached && cached->tree && cached->start <= start &&
1743 cached->end > start)
1744 node = &cached->rb_node;
1746 node = tree_search(tree, start);
1747 while (node && start <= end) {
1748 state = rb_entry(node, struct extent_state, rb_node);
1750 if (filled && state->start > start) {
1755 if (state->start > end)
1758 if (state->state & bits) {
1762 } else if (filled) {
1767 if (state->end == (u64)-1)
1770 start = state->end + 1;
1773 node = rb_next(node);
1780 spin_unlock(&tree->lock);
1785 * helper function to set a given page up to date if all the
1786 * extents in the tree for that page are up to date
1788 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1790 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1791 u64 end = start + PAGE_CACHE_SIZE - 1;
1792 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1793 SetPageUptodate(page);
1797 * helper function to unlock a page if all the extents in the tree
1798 * for that page are unlocked
1800 static void check_page_locked(struct extent_io_tree *tree, struct page *page)
1802 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1803 u64 end = start + PAGE_CACHE_SIZE - 1;
1804 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1809 * helper function to end page writeback if all the extents
1810 * in the tree for that page are done with writeback
1812 static void check_page_writeback(struct extent_io_tree *tree,
1815 end_page_writeback(page);
1819 * When IO fails, either with EIO or csum verification fails, we
1820 * try other mirrors that might have a good copy of the data. This
1821 * io_failure_record is used to record state as we go through all the
1822 * mirrors. If another mirror has good data, the page is set up to date
1823 * and things continue. If a good mirror can't be found, the original
1824 * bio end_io callback is called to indicate things have failed.
1826 struct io_failure_record {
1831 unsigned long bio_flags;
1837 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1842 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1844 set_state_private(failure_tree, rec->start, 0);
1845 ret = clear_extent_bits(failure_tree, rec->start,
1846 rec->start + rec->len - 1,
1847 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1852 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1853 rec->start + rec->len - 1,
1854 EXTENT_DAMAGED, GFP_NOFS);
1863 static void repair_io_failure_callback(struct bio *bio, int err)
1865 complete(bio->bi_private);
1869 * this bypasses the standard btrfs submit functions deliberately, as
1870 * the standard behavior is to write all copies in a raid setup. here we only
1871 * want to write the one bad copy. so we do the mapping for ourselves and issue
1872 * submit_bio directly.
1873 * to avoid any synchonization issues, wait for the data after writing, which
1874 * actually prevents the read that triggered the error from finishing.
1875 * currently, there can be no more than two copies of every data bit. thus,
1876 * exactly one rewrite is required.
1878 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1879 u64 length, u64 logical, struct page *page,
1883 struct btrfs_device *dev;
1884 DECLARE_COMPLETION_ONSTACK(compl);
1887 struct btrfs_bio *bbio = NULL;
1890 BUG_ON(!mirror_num);
1892 bio = bio_alloc(GFP_NOFS, 1);
1895 bio->bi_private = &compl;
1896 bio->bi_end_io = repair_io_failure_callback;
1898 map_length = length;
1900 ret = btrfs_map_block(map_tree, WRITE, logical,
1901 &map_length, &bbio, mirror_num);
1906 BUG_ON(mirror_num != bbio->mirror_num);
1907 sector = bbio->stripes[mirror_num-1].physical >> 9;
1908 bio->bi_sector = sector;
1909 dev = bbio->stripes[mirror_num-1].dev;
1911 if (!dev || !dev->bdev || !dev->writeable) {
1915 bio->bi_bdev = dev->bdev;
1916 bio_add_page(bio, page, length, start-page_offset(page));
1917 btrfsic_submit_bio(WRITE_SYNC, bio);
1918 wait_for_completion(&compl);
1920 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1921 /* try to remap that extent elsewhere? */
1926 printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s "
1927 "sector %llu)\n", page->mapping->host->i_ino, start,
1934 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
1937 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1938 u64 start = eb->start;
1939 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
1942 for (i = 0; i < num_pages; i++) {
1943 struct page *p = extent_buffer_page(eb, i);
1944 ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE,
1945 start, p, mirror_num);
1948 start += PAGE_CACHE_SIZE;
1955 * each time an IO finishes, we do a fast check in the IO failure tree
1956 * to see if we need to process or clean up an io_failure_record
1958 static int clean_io_failure(u64 start, struct page *page)
1961 u64 private_failure;
1962 struct io_failure_record *failrec;
1963 struct btrfs_mapping_tree *map_tree;
1964 struct extent_state *state;
1968 struct inode *inode = page->mapping->host;
1971 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1972 (u64)-1, 1, EXTENT_DIRTY, 0);
1976 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1981 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1982 BUG_ON(!failrec->this_mirror);
1984 if (failrec->in_validation) {
1985 /* there was no real error, just free the record */
1986 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1992 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1993 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1996 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1998 if (state && state->start == failrec->start) {
1999 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
2000 num_copies = btrfs_num_copies(map_tree, failrec->logical,
2002 if (num_copies > 1) {
2003 ret = repair_io_failure(map_tree, start, failrec->len,
2004 failrec->logical, page,
2005 failrec->failed_mirror);
2012 ret = free_io_failure(inode, failrec, did_repair);
2018 * this is a generic handler for readpage errors (default
2019 * readpage_io_failed_hook). if other copies exist, read those and write back
2020 * good data to the failed position. does not investigate in remapping the
2021 * failed extent elsewhere, hoping the device will be smart enough to do this as
2025 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2026 u64 start, u64 end, int failed_mirror,
2027 struct extent_state *state)
2029 struct io_failure_record *failrec = NULL;
2031 struct extent_map *em;
2032 struct inode *inode = page->mapping->host;
2033 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2034 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2035 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2042 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2044 ret = get_state_private(failure_tree, start, &private);
2046 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2049 failrec->start = start;
2050 failrec->len = end - start + 1;
2051 failrec->this_mirror = 0;
2052 failrec->bio_flags = 0;
2053 failrec->in_validation = 0;
2055 read_lock(&em_tree->lock);
2056 em = lookup_extent_mapping(em_tree, start, failrec->len);
2058 read_unlock(&em_tree->lock);
2063 if (em->start > start || em->start + em->len < start) {
2064 free_extent_map(em);
2067 read_unlock(&em_tree->lock);
2069 if (!em || IS_ERR(em)) {
2073 logical = start - em->start;
2074 logical = em->block_start + logical;
2075 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2076 logical = em->block_start;
2077 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2078 extent_set_compress_type(&failrec->bio_flags,
2081 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2082 "len=%llu\n", logical, start, failrec->len);
2083 failrec->logical = logical;
2084 free_extent_map(em);
2086 /* set the bits in the private failure tree */
2087 ret = set_extent_bits(failure_tree, start, end,
2088 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2090 ret = set_state_private(failure_tree, start,
2091 (u64)(unsigned long)failrec);
2092 /* set the bits in the inode's tree */
2094 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2101 failrec = (struct io_failure_record *)(unsigned long)private;
2102 pr_debug("bio_readpage_error: (found) logical=%llu, "
2103 "start=%llu, len=%llu, validation=%d\n",
2104 failrec->logical, failrec->start, failrec->len,
2105 failrec->in_validation);
2107 * when data can be on disk more than twice, add to failrec here
2108 * (e.g. with a list for failed_mirror) to make
2109 * clean_io_failure() clean all those errors at once.
2112 num_copies = btrfs_num_copies(
2113 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2114 failrec->logical, failrec->len);
2115 if (num_copies == 1) {
2117 * we only have a single copy of the data, so don't bother with
2118 * all the retry and error correction code that follows. no
2119 * matter what the error is, it is very likely to persist.
2121 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2122 "state=%p, num_copies=%d, next_mirror %d, "
2123 "failed_mirror %d\n", state, num_copies,
2124 failrec->this_mirror, failed_mirror);
2125 free_io_failure(inode, failrec, 0);
2130 spin_lock(&tree->lock);
2131 state = find_first_extent_bit_state(tree, failrec->start,
2133 if (state && state->start != failrec->start)
2135 spin_unlock(&tree->lock);
2139 * there are two premises:
2140 * a) deliver good data to the caller
2141 * b) correct the bad sectors on disk
2143 if (failed_bio->bi_vcnt > 1) {
2145 * to fulfill b), we need to know the exact failing sectors, as
2146 * we don't want to rewrite any more than the failed ones. thus,
2147 * we need separate read requests for the failed bio
2149 * if the following BUG_ON triggers, our validation request got
2150 * merged. we need separate requests for our algorithm to work.
2152 BUG_ON(failrec->in_validation);
2153 failrec->in_validation = 1;
2154 failrec->this_mirror = failed_mirror;
2155 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2158 * we're ready to fulfill a) and b) alongside. get a good copy
2159 * of the failed sector and if we succeed, we have setup
2160 * everything for repair_io_failure to do the rest for us.
2162 if (failrec->in_validation) {
2163 BUG_ON(failrec->this_mirror != failed_mirror);
2164 failrec->in_validation = 0;
2165 failrec->this_mirror = 0;
2167 failrec->failed_mirror = failed_mirror;
2168 failrec->this_mirror++;
2169 if (failrec->this_mirror == failed_mirror)
2170 failrec->this_mirror++;
2171 read_mode = READ_SYNC;
2174 if (!state || failrec->this_mirror > num_copies) {
2175 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2176 "next_mirror %d, failed_mirror %d\n", state,
2177 num_copies, failrec->this_mirror, failed_mirror);
2178 free_io_failure(inode, failrec, 0);
2182 bio = bio_alloc(GFP_NOFS, 1);
2183 bio->bi_private = state;
2184 bio->bi_end_io = failed_bio->bi_end_io;
2185 bio->bi_sector = failrec->logical >> 9;
2186 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2189 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2191 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2192 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2193 failrec->this_mirror, num_copies, failrec->in_validation);
2195 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2196 failrec->this_mirror,
2197 failrec->bio_flags, 0);
2201 /* lots and lots of room for performance fixes in the end_bio funcs */
2203 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2205 int uptodate = (err == 0);
2206 struct extent_io_tree *tree;
2209 tree = &BTRFS_I(page->mapping->host)->io_tree;
2211 if (tree->ops && tree->ops->writepage_end_io_hook) {
2212 ret = tree->ops->writepage_end_io_hook(page, start,
2213 end, NULL, uptodate);
2218 if (!uptodate && tree->ops &&
2219 tree->ops->writepage_io_failed_hook) {
2220 ret = tree->ops->writepage_io_failed_hook(NULL, page,
2222 /* Writeback already completed */
2228 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
2229 ClearPageUptodate(page);
2236 * after a writepage IO is done, we need to:
2237 * clear the uptodate bits on error
2238 * clear the writeback bits in the extent tree for this IO
2239 * end_page_writeback if the page has no more pending IO
2241 * Scheduling is not allowed, so the extent state tree is expected
2242 * to have one and only one object corresponding to this IO.
2244 static void end_bio_extent_writepage(struct bio *bio, int err)
2246 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2247 struct extent_io_tree *tree;
2253 struct page *page = bvec->bv_page;
2254 tree = &BTRFS_I(page->mapping->host)->io_tree;
2256 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2258 end = start + bvec->bv_len - 1;
2260 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2265 if (--bvec >= bio->bi_io_vec)
2266 prefetchw(&bvec->bv_page->flags);
2268 if (end_extent_writepage(page, err, start, end))
2272 end_page_writeback(page);
2274 check_page_writeback(tree, page);
2275 } while (bvec >= bio->bi_io_vec);
2281 * after a readpage IO is done, we need to:
2282 * clear the uptodate bits on error
2283 * set the uptodate bits if things worked
2284 * set the page up to date if all extents in the tree are uptodate
2285 * clear the lock bit in the extent tree
2286 * unlock the page if there are no other extents locked for it
2288 * Scheduling is not allowed, so the extent state tree is expected
2289 * to have one and only one object corresponding to this IO.
2291 static void end_bio_extent_readpage(struct bio *bio, int err)
2293 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2294 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2295 struct bio_vec *bvec = bio->bi_io_vec;
2296 struct extent_io_tree *tree;
2307 struct page *page = bvec->bv_page;
2308 struct extent_state *cached = NULL;
2309 struct extent_state *state;
2311 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2312 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2313 (long int)bio->bi_bdev);
2314 tree = &BTRFS_I(page->mapping->host)->io_tree;
2316 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2318 end = start + bvec->bv_len - 1;
2320 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2325 if (++bvec <= bvec_end)
2326 prefetchw(&bvec->bv_page->flags);
2328 spin_lock(&tree->lock);
2329 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2330 if (state && state->start == start) {
2332 * take a reference on the state, unlock will drop
2335 cache_state(state, &cached);
2337 spin_unlock(&tree->lock);
2339 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2340 ret = tree->ops->readpage_end_io_hook(page, start, end,
2345 clean_io_failure(start, page);
2349 failed_mirror = (int)(unsigned long)bio->bi_bdev;
2351 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2352 ret = tree->ops->readpage_io_failed_hook(page, failed_mirror);
2354 test_bit(BIO_UPTODATE, &bio->bi_flags))
2356 } else if (!uptodate) {
2358 * The generic bio_readpage_error handles errors the
2359 * following way: If possible, new read requests are
2360 * created and submitted and will end up in
2361 * end_bio_extent_readpage as well (if we're lucky, not
2362 * in the !uptodate case). In that case it returns 0 and
2363 * we just go on with the next page in our bio. If it
2364 * can't handle the error it will return -EIO and we
2365 * remain responsible for that page.
2367 ret = bio_readpage_error(bio, page, start, end,
2368 failed_mirror, NULL);
2371 test_bit(BIO_UPTODATE, &bio->bi_flags);
2374 uncache_state(&cached);
2379 if (uptodate && tree->track_uptodate) {
2380 set_extent_uptodate(tree, start, end, &cached,
2383 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2387 SetPageUptodate(page);
2389 ClearPageUptodate(page);
2395 check_page_uptodate(tree, page);
2397 ClearPageUptodate(page);
2400 check_page_locked(tree, page);
2402 } while (bvec <= bvec_end);
2408 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2413 bio = bio_alloc(gfp_flags, nr_vecs);
2415 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2416 while (!bio && (nr_vecs /= 2))
2417 bio = bio_alloc(gfp_flags, nr_vecs);
2422 bio->bi_bdev = bdev;
2423 bio->bi_sector = first_sector;
2429 * Since writes are async, they will only return -ENOMEM.
2430 * Reads can return the full range of I/O error conditions.
2432 static int __must_check submit_one_bio(int rw, struct bio *bio,
2433 int mirror_num, unsigned long bio_flags)
2436 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2437 struct page *page = bvec->bv_page;
2438 struct extent_io_tree *tree = bio->bi_private;
2441 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2443 bio->bi_private = NULL;
2447 if (tree->ops && tree->ops->submit_bio_hook)
2448 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2449 mirror_num, bio_flags, start);
2451 btrfsic_submit_bio(rw, bio);
2453 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2459 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2460 unsigned long offset, size_t size, struct bio *bio,
2461 unsigned long bio_flags)
2464 if (tree->ops && tree->ops->merge_bio_hook)
2465 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2472 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2473 struct page *page, sector_t sector,
2474 size_t size, unsigned long offset,
2475 struct block_device *bdev,
2476 struct bio **bio_ret,
2477 unsigned long max_pages,
2478 bio_end_io_t end_io_func,
2480 unsigned long prev_bio_flags,
2481 unsigned long bio_flags)
2487 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2488 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2489 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2491 if (bio_ret && *bio_ret) {
2494 contig = bio->bi_sector == sector;
2496 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2499 if (prev_bio_flags != bio_flags || !contig ||
2500 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2501 bio_add_page(bio, page, page_size, offset) < page_size) {
2502 ret = submit_one_bio(rw, bio, mirror_num,
2511 if (this_compressed)
2514 nr = bio_get_nr_vecs(bdev);
2516 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2520 bio_add_page(bio, page, page_size, offset);
2521 bio->bi_end_io = end_io_func;
2522 bio->bi_private = tree;
2527 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2532 void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
2534 if (!PagePrivate(page)) {
2535 SetPagePrivate(page);
2536 page_cache_get(page);
2537 set_page_private(page, (unsigned long)eb);
2539 WARN_ON(page->private != (unsigned long)eb);
2543 void set_page_extent_mapped(struct page *page)
2545 if (!PagePrivate(page)) {
2546 SetPagePrivate(page);
2547 page_cache_get(page);
2548 set_page_private(page, EXTENT_PAGE_PRIVATE);
2553 * basic readpage implementation. Locked extent state structs are inserted
2554 * into the tree that are removed when the IO is done (by the end_io
2556 * XXX JDM: This needs looking at to ensure proper page locking
2558 static int __extent_read_full_page(struct extent_io_tree *tree,
2560 get_extent_t *get_extent,
2561 struct bio **bio, int mirror_num,
2562 unsigned long *bio_flags)
2564 struct inode *inode = page->mapping->host;
2565 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2566 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2570 u64 last_byte = i_size_read(inode);
2574 struct extent_map *em;
2575 struct block_device *bdev;
2576 struct btrfs_ordered_extent *ordered;
2579 size_t pg_offset = 0;
2581 size_t disk_io_size;
2582 size_t blocksize = inode->i_sb->s_blocksize;
2583 unsigned long this_bio_flag = 0;
2585 set_page_extent_mapped(page);
2587 if (!PageUptodate(page)) {
2588 if (cleancache_get_page(page) == 0) {
2589 BUG_ON(blocksize != PAGE_SIZE);
2596 lock_extent(tree, start, end);
2597 ordered = btrfs_lookup_ordered_extent(inode, start);
2600 unlock_extent(tree, start, end);
2601 btrfs_start_ordered_extent(inode, ordered, 1);
2602 btrfs_put_ordered_extent(ordered);
2605 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2607 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2610 iosize = PAGE_CACHE_SIZE - zero_offset;
2611 userpage = kmap_atomic(page);
2612 memset(userpage + zero_offset, 0, iosize);
2613 flush_dcache_page(page);
2614 kunmap_atomic(userpage);
2617 while (cur <= end) {
2618 if (cur >= last_byte) {
2620 struct extent_state *cached = NULL;
2622 iosize = PAGE_CACHE_SIZE - pg_offset;
2623 userpage = kmap_atomic(page);
2624 memset(userpage + pg_offset, 0, iosize);
2625 flush_dcache_page(page);
2626 kunmap_atomic(userpage);
2627 set_extent_uptodate(tree, cur, cur + iosize - 1,
2629 unlock_extent_cached(tree, cur, cur + iosize - 1,
2633 em = get_extent(inode, page, pg_offset, cur,
2635 if (IS_ERR_OR_NULL(em)) {
2637 unlock_extent(tree, cur, end);
2640 extent_offset = cur - em->start;
2641 BUG_ON(extent_map_end(em) <= cur);
2644 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2645 this_bio_flag = EXTENT_BIO_COMPRESSED;
2646 extent_set_compress_type(&this_bio_flag,
2650 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2651 cur_end = min(extent_map_end(em) - 1, end);
2652 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2653 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2654 disk_io_size = em->block_len;
2655 sector = em->block_start >> 9;
2657 sector = (em->block_start + extent_offset) >> 9;
2658 disk_io_size = iosize;
2661 block_start = em->block_start;
2662 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2663 block_start = EXTENT_MAP_HOLE;
2664 free_extent_map(em);
2667 /* we've found a hole, just zero and go on */
2668 if (block_start == EXTENT_MAP_HOLE) {
2670 struct extent_state *cached = NULL;
2672 userpage = kmap_atomic(page);
2673 memset(userpage + pg_offset, 0, iosize);
2674 flush_dcache_page(page);
2675 kunmap_atomic(userpage);
2677 set_extent_uptodate(tree, cur, cur + iosize - 1,
2679 unlock_extent_cached(tree, cur, cur + iosize - 1,
2682 pg_offset += iosize;
2685 /* the get_extent function already copied into the page */
2686 if (test_range_bit(tree, cur, cur_end,
2687 EXTENT_UPTODATE, 1, NULL)) {
2688 check_page_uptodate(tree, page);
2689 unlock_extent(tree, cur, cur + iosize - 1);
2691 pg_offset += iosize;
2694 /* we have an inline extent but it didn't get marked up
2695 * to date. Error out
2697 if (block_start == EXTENT_MAP_INLINE) {
2699 unlock_extent(tree, cur, cur + iosize - 1);
2701 pg_offset += iosize;
2706 if (tree->ops && tree->ops->readpage_io_hook) {
2707 ret = tree->ops->readpage_io_hook(page, cur,
2711 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2713 ret = submit_extent_page(READ, tree, page,
2714 sector, disk_io_size, pg_offset,
2716 end_bio_extent_readpage, mirror_num,
2719 BUG_ON(ret == -ENOMEM);
2721 *bio_flags = this_bio_flag;
2726 pg_offset += iosize;
2730 if (!PageError(page))
2731 SetPageUptodate(page);
2737 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2738 get_extent_t *get_extent, int mirror_num)
2740 struct bio *bio = NULL;
2741 unsigned long bio_flags = 0;
2744 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2747 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2751 static noinline void update_nr_written(struct page *page,
2752 struct writeback_control *wbc,
2753 unsigned long nr_written)
2755 wbc->nr_to_write -= nr_written;
2756 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2757 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2758 page->mapping->writeback_index = page->index + nr_written;
2762 * the writepage semantics are similar to regular writepage. extent
2763 * records are inserted to lock ranges in the tree, and as dirty areas
2764 * are found, they are marked writeback. Then the lock bits are removed
2765 * and the end_io handler clears the writeback ranges
2767 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2770 struct inode *inode = page->mapping->host;
2771 struct extent_page_data *epd = data;
2772 struct extent_io_tree *tree = epd->tree;
2773 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2775 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2779 u64 last_byte = i_size_read(inode);
2783 struct extent_state *cached_state = NULL;
2784 struct extent_map *em;
2785 struct block_device *bdev;
2788 size_t pg_offset = 0;
2790 loff_t i_size = i_size_read(inode);
2791 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2797 unsigned long nr_written = 0;
2798 bool fill_delalloc = true;
2800 if (wbc->sync_mode == WB_SYNC_ALL)
2801 write_flags = WRITE_SYNC;
2803 write_flags = WRITE;
2805 trace___extent_writepage(page, inode, wbc);
2807 WARN_ON(!PageLocked(page));
2809 ClearPageError(page);
2811 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2812 if (page->index > end_index ||
2813 (page->index == end_index && !pg_offset)) {
2814 page->mapping->a_ops->invalidatepage(page, 0);
2819 if (page->index == end_index) {
2822 userpage = kmap_atomic(page);
2823 memset(userpage + pg_offset, 0,
2824 PAGE_CACHE_SIZE - pg_offset);
2825 kunmap_atomic(userpage);
2826 flush_dcache_page(page);
2830 set_page_extent_mapped(page);
2832 if (!tree->ops || !tree->ops->fill_delalloc)
2833 fill_delalloc = false;
2835 delalloc_start = start;
2838 if (!epd->extent_locked && fill_delalloc) {
2839 u64 delalloc_to_write = 0;
2841 * make sure the wbc mapping index is at least updated
2844 update_nr_written(page, wbc, 0);
2846 while (delalloc_end < page_end) {
2847 nr_delalloc = find_lock_delalloc_range(inode, tree,
2852 if (nr_delalloc == 0) {
2853 delalloc_start = delalloc_end + 1;
2856 ret = tree->ops->fill_delalloc(inode, page,
2861 /* File system has been set read-only */
2867 * delalloc_end is already one less than the total
2868 * length, so we don't subtract one from
2871 delalloc_to_write += (delalloc_end - delalloc_start +
2874 delalloc_start = delalloc_end + 1;
2876 if (wbc->nr_to_write < delalloc_to_write) {
2879 if (delalloc_to_write < thresh * 2)
2880 thresh = delalloc_to_write;
2881 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2885 /* did the fill delalloc function already unlock and start
2891 * we've unlocked the page, so we can't update
2892 * the mapping's writeback index, just update
2895 wbc->nr_to_write -= nr_written;
2899 if (tree->ops && tree->ops->writepage_start_hook) {
2900 ret = tree->ops->writepage_start_hook(page, start,
2903 /* Fixup worker will requeue */
2905 wbc->pages_skipped++;
2907 redirty_page_for_writepage(wbc, page);
2908 update_nr_written(page, wbc, nr_written);
2916 * we don't want to touch the inode after unlocking the page,
2917 * so we update the mapping writeback index now
2919 update_nr_written(page, wbc, nr_written + 1);
2922 if (last_byte <= start) {
2923 if (tree->ops && tree->ops->writepage_end_io_hook)
2924 tree->ops->writepage_end_io_hook(page, start,
2929 blocksize = inode->i_sb->s_blocksize;
2931 while (cur <= end) {
2932 if (cur >= last_byte) {
2933 if (tree->ops && tree->ops->writepage_end_io_hook)
2934 tree->ops->writepage_end_io_hook(page, cur,
2938 em = epd->get_extent(inode, page, pg_offset, cur,
2940 if (IS_ERR_OR_NULL(em)) {
2945 extent_offset = cur - em->start;
2946 BUG_ON(extent_map_end(em) <= cur);
2948 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2949 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2950 sector = (em->block_start + extent_offset) >> 9;
2952 block_start = em->block_start;
2953 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2954 free_extent_map(em);
2958 * compressed and inline extents are written through other
2961 if (compressed || block_start == EXTENT_MAP_HOLE ||
2962 block_start == EXTENT_MAP_INLINE) {
2964 * end_io notification does not happen here for
2965 * compressed extents
2967 if (!compressed && tree->ops &&
2968 tree->ops->writepage_end_io_hook)
2969 tree->ops->writepage_end_io_hook(page, cur,
2972 else if (compressed) {
2973 /* we don't want to end_page_writeback on
2974 * a compressed extent. this happens
2981 pg_offset += iosize;
2984 /* leave this out until we have a page_mkwrite call */
2985 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2986 EXTENT_DIRTY, 0, NULL)) {
2988 pg_offset += iosize;
2992 if (tree->ops && tree->ops->writepage_io_hook) {
2993 ret = tree->ops->writepage_io_hook(page, cur,
3001 unsigned long max_nr = end_index + 1;
3003 set_range_writeback(tree, cur, cur + iosize - 1);
3004 if (!PageWriteback(page)) {
3005 printk(KERN_ERR "btrfs warning page %lu not "
3006 "writeback, cur %llu end %llu\n",
3007 page->index, (unsigned long long)cur,
3008 (unsigned long long)end);
3011 ret = submit_extent_page(write_flags, tree, page,
3012 sector, iosize, pg_offset,
3013 bdev, &epd->bio, max_nr,
3014 end_bio_extent_writepage,
3020 pg_offset += iosize;
3025 /* make sure the mapping tag for page dirty gets cleared */
3026 set_page_writeback(page);
3027 end_page_writeback(page);
3033 /* drop our reference on any cached states */
3034 free_extent_state(cached_state);
3038 static int eb_wait(void *word)
3044 static void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3046 wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3047 TASK_UNINTERRUPTIBLE);
3050 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3051 struct btrfs_fs_info *fs_info,
3052 struct extent_page_data *epd)
3054 unsigned long i, num_pages;
3058 if (!btrfs_try_tree_write_lock(eb)) {
3060 flush_write_bio(epd);
3061 btrfs_tree_lock(eb);
3064 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3065 btrfs_tree_unlock(eb);
3069 flush_write_bio(epd);
3073 wait_on_extent_buffer_writeback(eb);
3074 btrfs_tree_lock(eb);
3075 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3077 btrfs_tree_unlock(eb);
3081 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3082 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3083 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3084 spin_lock(&fs_info->delalloc_lock);
3085 if (fs_info->dirty_metadata_bytes >= eb->len)
3086 fs_info->dirty_metadata_bytes -= eb->len;
3089 spin_unlock(&fs_info->delalloc_lock);
3093 btrfs_tree_unlock(eb);
3098 num_pages = num_extent_pages(eb->start, eb->len);
3099 for (i = 0; i < num_pages; i++) {
3100 struct page *p = extent_buffer_page(eb, i);
3102 if (!trylock_page(p)) {
3104 flush_write_bio(epd);
3114 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3116 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3117 smp_mb__after_clear_bit();
3118 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3121 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3123 int uptodate = err == 0;
3124 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3125 struct extent_buffer *eb;
3129 struct page *page = bvec->bv_page;
3132 eb = (struct extent_buffer *)page->private;
3134 done = atomic_dec_and_test(&eb->io_pages);
3136 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3137 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3138 ClearPageUptodate(page);
3142 end_page_writeback(page);
3147 end_extent_buffer_writeback(eb);
3148 } while (bvec >= bio->bi_io_vec);
3154 static int write_one_eb(struct extent_buffer *eb,
3155 struct btrfs_fs_info *fs_info,
3156 struct writeback_control *wbc,
3157 struct extent_page_data *epd)
3159 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3160 u64 offset = eb->start;
3161 unsigned long i, num_pages;
3162 int rw = (epd->sync_io ? WRITE_SYNC : WRITE);
3165 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3166 num_pages = num_extent_pages(eb->start, eb->len);
3167 atomic_set(&eb->io_pages, num_pages);
3168 for (i = 0; i < num_pages; i++) {
3169 struct page *p = extent_buffer_page(eb, i);
3171 clear_page_dirty_for_io(p);
3172 set_page_writeback(p);
3173 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3174 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3175 -1, end_bio_extent_buffer_writepage,
3178 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3180 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3181 end_extent_buffer_writeback(eb);
3185 offset += PAGE_CACHE_SIZE;
3186 update_nr_written(p, wbc, 1);
3190 if (unlikely(ret)) {
3191 for (; i < num_pages; i++) {
3192 struct page *p = extent_buffer_page(eb, i);
3200 int btree_write_cache_pages(struct address_space *mapping,
3201 struct writeback_control *wbc)
3203 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3204 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3205 struct extent_buffer *eb, *prev_eb = NULL;
3206 struct extent_page_data epd = {
3210 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3214 int nr_to_write_done = 0;
3215 struct pagevec pvec;
3218 pgoff_t end; /* Inclusive */
3222 pagevec_init(&pvec, 0);
3223 if (wbc->range_cyclic) {
3224 index = mapping->writeback_index; /* Start from prev offset */
3227 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3228 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3231 if (wbc->sync_mode == WB_SYNC_ALL)
3232 tag = PAGECACHE_TAG_TOWRITE;
3234 tag = PAGECACHE_TAG_DIRTY;
3236 if (wbc->sync_mode == WB_SYNC_ALL)
3237 tag_pages_for_writeback(mapping, index, end);
3238 while (!done && !nr_to_write_done && (index <= end) &&
3239 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3240 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3244 for (i = 0; i < nr_pages; i++) {
3245 struct page *page = pvec.pages[i];
3247 if (!PagePrivate(page))
3250 if (!wbc->range_cyclic && page->index > end) {
3255 eb = (struct extent_buffer *)page->private;
3264 if (!atomic_inc_not_zero(&eb->refs)) {
3270 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3272 free_extent_buffer(eb);
3276 ret = write_one_eb(eb, fs_info, wbc, &epd);
3279 free_extent_buffer(eb);
3282 free_extent_buffer(eb);
3285 * the filesystem may choose to bump up nr_to_write.
3286 * We have to make sure to honor the new nr_to_write
3289 nr_to_write_done = wbc->nr_to_write <= 0;
3291 pagevec_release(&pvec);
3294 if (!scanned && !done) {
3296 * We hit the last page and there is more work to be done: wrap
3297 * back to the start of the file
3303 flush_write_bio(&epd);
3308 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3309 * @mapping: address space structure to write
3310 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3311 * @writepage: function called for each page
3312 * @data: data passed to writepage function
3314 * If a page is already under I/O, write_cache_pages() skips it, even
3315 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3316 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3317 * and msync() need to guarantee that all the data which was dirty at the time
3318 * the call was made get new I/O started against them. If wbc->sync_mode is
3319 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3320 * existing IO to complete.
3322 static int extent_write_cache_pages(struct extent_io_tree *tree,
3323 struct address_space *mapping,
3324 struct writeback_control *wbc,
3325 writepage_t writepage, void *data,
3326 void (*flush_fn)(void *))
3330 int nr_to_write_done = 0;
3331 struct pagevec pvec;
3334 pgoff_t end; /* Inclusive */
3338 pagevec_init(&pvec, 0);
3339 if (wbc->range_cyclic) {
3340 index = mapping->writeback_index; /* Start from prev offset */
3343 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3344 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3347 if (wbc->sync_mode == WB_SYNC_ALL)
3348 tag = PAGECACHE_TAG_TOWRITE;
3350 tag = PAGECACHE_TAG_DIRTY;
3352 if (wbc->sync_mode == WB_SYNC_ALL)
3353 tag_pages_for_writeback(mapping, index, end);
3354 while (!done && !nr_to_write_done && (index <= end) &&
3355 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3356 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3360 for (i = 0; i < nr_pages; i++) {
3361 struct page *page = pvec.pages[i];
3364 * At this point we hold neither mapping->tree_lock nor
3365 * lock on the page itself: the page may be truncated or
3366 * invalidated (changing page->mapping to NULL), or even
3367 * swizzled back from swapper_space to tmpfs file
3371 tree->ops->write_cache_pages_lock_hook) {
3372 tree->ops->write_cache_pages_lock_hook(page,
3375 if (!trylock_page(page)) {
3381 if (unlikely(page->mapping != mapping)) {
3386 if (!wbc->range_cyclic && page->index > end) {
3392 if (wbc->sync_mode != WB_SYNC_NONE) {
3393 if (PageWriteback(page))
3395 wait_on_page_writeback(page);
3398 if (PageWriteback(page) ||
3399 !clear_page_dirty_for_io(page)) {
3404 ret = (*writepage)(page, wbc, data);
3406 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3414 * the filesystem may choose to bump up nr_to_write.
3415 * We have to make sure to honor the new nr_to_write
3418 nr_to_write_done = wbc->nr_to_write <= 0;
3420 pagevec_release(&pvec);
3423 if (!scanned && !done) {
3425 * We hit the last page and there is more work to be done: wrap
3426 * back to the start of the file
3435 static void flush_epd_write_bio(struct extent_page_data *epd)
3444 ret = submit_one_bio(rw, epd->bio, 0, 0);
3445 BUG_ON(ret < 0); /* -ENOMEM */
3450 static noinline void flush_write_bio(void *data)
3452 struct extent_page_data *epd = data;
3453 flush_epd_write_bio(epd);
3456 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3457 get_extent_t *get_extent,
3458 struct writeback_control *wbc)
3461 struct extent_page_data epd = {
3464 .get_extent = get_extent,
3466 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3469 ret = __extent_writepage(page, wbc, &epd);
3471 flush_epd_write_bio(&epd);
3475 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3476 u64 start, u64 end, get_extent_t *get_extent,
3480 struct address_space *mapping = inode->i_mapping;
3482 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3485 struct extent_page_data epd = {
3488 .get_extent = get_extent,
3490 .sync_io = mode == WB_SYNC_ALL,
3492 struct writeback_control wbc_writepages = {
3494 .nr_to_write = nr_pages * 2,
3495 .range_start = start,
3496 .range_end = end + 1,
3499 while (start <= end) {
3500 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3501 if (clear_page_dirty_for_io(page))
3502 ret = __extent_writepage(page, &wbc_writepages, &epd);
3504 if (tree->ops && tree->ops->writepage_end_io_hook)
3505 tree->ops->writepage_end_io_hook(page, start,
3506 start + PAGE_CACHE_SIZE - 1,
3510 page_cache_release(page);
3511 start += PAGE_CACHE_SIZE;
3514 flush_epd_write_bio(&epd);
3518 int extent_writepages(struct extent_io_tree *tree,
3519 struct address_space *mapping,
3520 get_extent_t *get_extent,
3521 struct writeback_control *wbc)
3524 struct extent_page_data epd = {
3527 .get_extent = get_extent,
3529 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3532 ret = extent_write_cache_pages(tree, mapping, wbc,
3533 __extent_writepage, &epd,
3535 flush_epd_write_bio(&epd);
3539 int extent_readpages(struct extent_io_tree *tree,
3540 struct address_space *mapping,
3541 struct list_head *pages, unsigned nr_pages,
3542 get_extent_t get_extent)
3544 struct bio *bio = NULL;
3546 unsigned long bio_flags = 0;
3548 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3549 struct page *page = list_entry(pages->prev, struct page, lru);
3551 prefetchw(&page->flags);
3552 list_del(&page->lru);
3553 if (!add_to_page_cache_lru(page, mapping,
3554 page->index, GFP_NOFS)) {
3555 __extent_read_full_page(tree, page, get_extent,
3556 &bio, 0, &bio_flags);
3558 page_cache_release(page);
3560 BUG_ON(!list_empty(pages));
3562 return submit_one_bio(READ, bio, 0, bio_flags);
3567 * basic invalidatepage code, this waits on any locked or writeback
3568 * ranges corresponding to the page, and then deletes any extent state
3569 * records from the tree
3571 int extent_invalidatepage(struct extent_io_tree *tree,
3572 struct page *page, unsigned long offset)
3574 struct extent_state *cached_state = NULL;
3575 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3576 u64 end = start + PAGE_CACHE_SIZE - 1;
3577 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3579 start += (offset + blocksize - 1) & ~(blocksize - 1);
3583 lock_extent_bits(tree, start, end, 0, &cached_state);
3584 wait_on_page_writeback(page);
3585 clear_extent_bit(tree, start, end,
3586 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3587 EXTENT_DO_ACCOUNTING,
3588 1, 1, &cached_state, GFP_NOFS);
3593 * a helper for releasepage, this tests for areas of the page that
3594 * are locked or under IO and drops the related state bits if it is safe
3597 int try_release_extent_state(struct extent_map_tree *map,
3598 struct extent_io_tree *tree, struct page *page,
3601 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3602 u64 end = start + PAGE_CACHE_SIZE - 1;
3605 if (test_range_bit(tree, start, end,
3606 EXTENT_IOBITS, 0, NULL))
3609 if ((mask & GFP_NOFS) == GFP_NOFS)
3612 * at this point we can safely clear everything except the
3613 * locked bit and the nodatasum bit
3615 ret = clear_extent_bit(tree, start, end,
3616 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3619 /* if clear_extent_bit failed for enomem reasons,
3620 * we can't allow the release to continue.
3631 * a helper for releasepage. As long as there are no locked extents
3632 * in the range corresponding to the page, both state records and extent
3633 * map records are removed
3635 int try_release_extent_mapping(struct extent_map_tree *map,
3636 struct extent_io_tree *tree, struct page *page,
3639 struct extent_map *em;
3640 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3641 u64 end = start + PAGE_CACHE_SIZE - 1;
3643 if ((mask & __GFP_WAIT) &&
3644 page->mapping->host->i_size > 16 * 1024 * 1024) {
3646 while (start <= end) {
3647 len = end - start + 1;
3648 write_lock(&map->lock);
3649 em = lookup_extent_mapping(map, start, len);
3651 write_unlock(&map->lock);
3654 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3655 em->start != start) {
3656 write_unlock(&map->lock);
3657 free_extent_map(em);
3660 if (!test_range_bit(tree, em->start,
3661 extent_map_end(em) - 1,
3662 EXTENT_LOCKED | EXTENT_WRITEBACK,
3664 remove_extent_mapping(map, em);
3665 /* once for the rb tree */
3666 free_extent_map(em);
3668 start = extent_map_end(em);
3669 write_unlock(&map->lock);
3672 free_extent_map(em);
3675 return try_release_extent_state(map, tree, page, mask);
3679 * helper function for fiemap, which doesn't want to see any holes.
3680 * This maps until we find something past 'last'
3682 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3685 get_extent_t *get_extent)
3687 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3688 struct extent_map *em;
3695 len = last - offset;
3698 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3699 em = get_extent(inode, NULL, 0, offset, len, 0);
3700 if (IS_ERR_OR_NULL(em))
3703 /* if this isn't a hole return it */
3704 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3705 em->block_start != EXTENT_MAP_HOLE) {
3709 /* this is a hole, advance to the next extent */
3710 offset = extent_map_end(em);
3711 free_extent_map(em);
3718 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3719 __u64 start, __u64 len, get_extent_t *get_extent)
3723 u64 max = start + len;
3727 u64 last_for_get_extent = 0;
3729 u64 isize = i_size_read(inode);
3730 struct btrfs_key found_key;
3731 struct extent_map *em = NULL;
3732 struct extent_state *cached_state = NULL;
3733 struct btrfs_path *path;
3734 struct btrfs_file_extent_item *item;
3739 unsigned long emflags;
3744 path = btrfs_alloc_path();
3747 path->leave_spinning = 1;
3749 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3750 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3753 * lookup the last file extent. We're not using i_size here
3754 * because there might be preallocation past i_size
3756 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3757 path, btrfs_ino(inode), -1, 0);
3759 btrfs_free_path(path);
3764 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3765 struct btrfs_file_extent_item);
3766 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3767 found_type = btrfs_key_type(&found_key);
3769 /* No extents, but there might be delalloc bits */
3770 if (found_key.objectid != btrfs_ino(inode) ||
3771 found_type != BTRFS_EXTENT_DATA_KEY) {
3772 /* have to trust i_size as the end */
3774 last_for_get_extent = isize;
3777 * remember the start of the last extent. There are a
3778 * bunch of different factors that go into the length of the
3779 * extent, so its much less complex to remember where it started
3781 last = found_key.offset;
3782 last_for_get_extent = last + 1;
3784 btrfs_free_path(path);
3787 * we might have some extents allocated but more delalloc past those
3788 * extents. so, we trust isize unless the start of the last extent is
3793 last_for_get_extent = isize;
3796 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3799 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3809 u64 offset_in_extent;
3811 /* break if the extent we found is outside the range */
3812 if (em->start >= max || extent_map_end(em) < off)
3816 * get_extent may return an extent that starts before our
3817 * requested range. We have to make sure the ranges
3818 * we return to fiemap always move forward and don't
3819 * overlap, so adjust the offsets here
3821 em_start = max(em->start, off);
3824 * record the offset from the start of the extent
3825 * for adjusting the disk offset below
3827 offset_in_extent = em_start - em->start;
3828 em_end = extent_map_end(em);
3829 em_len = em_end - em_start;
3830 emflags = em->flags;
3835 * bump off for our next call to get_extent
3837 off = extent_map_end(em);
3841 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3843 flags |= FIEMAP_EXTENT_LAST;
3844 } else if (em->block_start == EXTENT_MAP_INLINE) {
3845 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3846 FIEMAP_EXTENT_NOT_ALIGNED);
3847 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3848 flags |= (FIEMAP_EXTENT_DELALLOC |
3849 FIEMAP_EXTENT_UNKNOWN);
3851 disko = em->block_start + offset_in_extent;
3853 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3854 flags |= FIEMAP_EXTENT_ENCODED;
3856 free_extent_map(em);
3858 if ((em_start >= last) || em_len == (u64)-1 ||
3859 (last == (u64)-1 && isize <= em_end)) {
3860 flags |= FIEMAP_EXTENT_LAST;
3864 /* now scan forward to see if this is really the last extent. */
3865 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3872 flags |= FIEMAP_EXTENT_LAST;
3875 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3881 free_extent_map(em);
3883 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3884 &cached_state, GFP_NOFS);
3888 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3891 return eb->pages[i];
3894 inline unsigned long num_extent_pages(u64 start, u64 len)
3896 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3897 (start >> PAGE_CACHE_SHIFT);
3900 static void __free_extent_buffer(struct extent_buffer *eb)
3903 unsigned long flags;
3904 spin_lock_irqsave(&leak_lock, flags);
3905 list_del(&eb->leak_list);
3906 spin_unlock_irqrestore(&leak_lock, flags);
3908 if (eb->pages && eb->pages != eb->inline_pages)
3910 kmem_cache_free(extent_buffer_cache, eb);
3913 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3918 struct extent_buffer *eb = NULL;
3920 unsigned long flags;
3923 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3929 rwlock_init(&eb->lock);
3930 atomic_set(&eb->write_locks, 0);
3931 atomic_set(&eb->read_locks, 0);
3932 atomic_set(&eb->blocking_readers, 0);
3933 atomic_set(&eb->blocking_writers, 0);
3934 atomic_set(&eb->spinning_readers, 0);
3935 atomic_set(&eb->spinning_writers, 0);
3936 eb->lock_nested = 0;
3937 init_waitqueue_head(&eb->write_lock_wq);
3938 init_waitqueue_head(&eb->read_lock_wq);
3941 spin_lock_irqsave(&leak_lock, flags);
3942 list_add(&eb->leak_list, &buffers);
3943 spin_unlock_irqrestore(&leak_lock, flags);
3945 spin_lock_init(&eb->refs_lock);
3946 atomic_set(&eb->refs, 1);
3947 atomic_set(&eb->io_pages, 0);
3949 if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) {
3950 struct page **pages;
3951 int num_pages = (len + PAGE_CACHE_SIZE - 1) >>
3953 pages = kzalloc(num_pages, mask);
3955 __free_extent_buffer(eb);
3960 eb->pages = eb->inline_pages;
3966 static int extent_buffer_under_io(struct extent_buffer *eb)
3968 return (atomic_read(&eb->io_pages) ||
3969 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3970 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3974 * Helper for releasing extent buffer page.
3976 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3977 unsigned long start_idx)
3979 unsigned long index;
3982 BUG_ON(extent_buffer_under_io(eb));
3984 index = num_extent_pages(eb->start, eb->len);
3985 if (start_idx >= index)
3990 page = extent_buffer_page(eb, index);
3992 spin_lock(&page->mapping->private_lock);
3994 * We do this since we'll remove the pages after we've
3995 * removed the eb from the radix tree, so we could race
3996 * and have this page now attached to the new eb. So
3997 * only clear page_private if it's still connected to
4000 if (PagePrivate(page) &&
4001 page->private == (unsigned long)eb) {
4002 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4003 BUG_ON(PageDirty(page));
4004 BUG_ON(PageWriteback(page));
4006 * We need to make sure we haven't be attached
4009 ClearPagePrivate(page);
4010 set_page_private(page, 0);
4011 /* One for the page private */
4012 page_cache_release(page);
4014 spin_unlock(&page->mapping->private_lock);
4016 /* One for when we alloced the page */
4017 page_cache_release(page);
4019 } while (index != start_idx);
4023 * Helper for releasing the extent buffer.
4025 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4027 btrfs_release_extent_buffer_page(eb, 0);
4028 __free_extent_buffer(eb);
4031 static void check_buffer_tree_ref(struct extent_buffer *eb)
4033 /* the ref bit is tricky. We have to make sure it is set
4034 * if we have the buffer dirty. Otherwise the
4035 * code to free a buffer can end up dropping a dirty
4038 * Once the ref bit is set, it won't go away while the
4039 * buffer is dirty or in writeback, and it also won't
4040 * go away while we have the reference count on the
4043 * We can't just set the ref bit without bumping the
4044 * ref on the eb because free_extent_buffer might
4045 * see the ref bit and try to clear it. If this happens
4046 * free_extent_buffer might end up dropping our original
4047 * ref by mistake and freeing the page before we are able
4048 * to add one more ref.
4050 * So bump the ref count first, then set the bit. If someone
4051 * beat us to it, drop the ref we added.
4053 if (!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4054 atomic_inc(&eb->refs);
4055 if (test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4056 atomic_dec(&eb->refs);
4060 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4062 unsigned long num_pages, i;
4064 check_buffer_tree_ref(eb);
4066 num_pages = num_extent_pages(eb->start, eb->len);
4067 for (i = 0; i < num_pages; i++) {
4068 struct page *p = extent_buffer_page(eb, i);
4069 mark_page_accessed(p);
4073 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4074 u64 start, unsigned long len)
4076 unsigned long num_pages = num_extent_pages(start, len);
4078 unsigned long index = start >> PAGE_CACHE_SHIFT;
4079 struct extent_buffer *eb;
4080 struct extent_buffer *exists = NULL;
4082 struct address_space *mapping = tree->mapping;
4087 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4088 if (eb && atomic_inc_not_zero(&eb->refs)) {
4090 mark_extent_buffer_accessed(eb);
4095 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4099 for (i = 0; i < num_pages; i++, index++) {
4100 p = find_or_create_page(mapping, index, GFP_NOFS);
4106 spin_lock(&mapping->private_lock);
4107 if (PagePrivate(p)) {
4109 * We could have already allocated an eb for this page
4110 * and attached one so lets see if we can get a ref on
4111 * the existing eb, and if we can we know it's good and
4112 * we can just return that one, else we know we can just
4113 * overwrite page->private.
4115 exists = (struct extent_buffer *)p->private;
4116 if (atomic_inc_not_zero(&exists->refs)) {
4117 spin_unlock(&mapping->private_lock);
4119 mark_extent_buffer_accessed(exists);
4124 * Do this so attach doesn't complain and we need to
4125 * drop the ref the old guy had.
4127 ClearPagePrivate(p);
4128 WARN_ON(PageDirty(p));
4129 page_cache_release(p);
4131 attach_extent_buffer_page(eb, p);
4132 spin_unlock(&mapping->private_lock);
4133 WARN_ON(PageDirty(p));
4134 mark_page_accessed(p);
4136 if (!PageUptodate(p))
4140 * see below about how we avoid a nasty race with release page
4141 * and why we unlock later
4145 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4147 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4151 spin_lock(&tree->buffer_lock);
4152 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4153 if (ret == -EEXIST) {
4154 exists = radix_tree_lookup(&tree->buffer,
4155 start >> PAGE_CACHE_SHIFT);
4156 if (!atomic_inc_not_zero(&exists->refs)) {
4157 spin_unlock(&tree->buffer_lock);
4158 radix_tree_preload_end();
4162 spin_unlock(&tree->buffer_lock);
4163 radix_tree_preload_end();
4164 mark_extent_buffer_accessed(exists);
4167 /* add one reference for the tree */
4168 spin_lock(&eb->refs_lock);
4169 check_buffer_tree_ref(eb);
4170 spin_unlock(&eb->refs_lock);
4171 spin_unlock(&tree->buffer_lock);
4172 radix_tree_preload_end();
4175 * there is a race where release page may have
4176 * tried to find this extent buffer in the radix
4177 * but failed. It will tell the VM it is safe to
4178 * reclaim the, and it will clear the page private bit.
4179 * We must make sure to set the page private bit properly
4180 * after the extent buffer is in the radix tree so
4181 * it doesn't get lost
4183 SetPageChecked(eb->pages[0]);
4184 for (i = 1; i < num_pages; i++) {
4185 p = extent_buffer_page(eb, i);
4186 ClearPageChecked(p);
4189 unlock_page(eb->pages[0]);
4193 for (i = 0; i < num_pages; i++) {
4195 unlock_page(eb->pages[i]);
4198 if (!atomic_dec_and_test(&eb->refs))
4200 btrfs_release_extent_buffer(eb);
4204 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4205 u64 start, unsigned long len)
4207 struct extent_buffer *eb;
4210 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4211 if (eb && atomic_inc_not_zero(&eb->refs)) {
4213 mark_extent_buffer_accessed(eb);
4221 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4223 struct extent_buffer *eb =
4224 container_of(head, struct extent_buffer, rcu_head);
4226 __free_extent_buffer(eb);
4229 /* Expects to have eb->eb_lock already held */
4230 static void release_extent_buffer(struct extent_buffer *eb, gfp_t mask)
4232 WARN_ON(atomic_read(&eb->refs) == 0);
4233 if (atomic_dec_and_test(&eb->refs)) {
4234 struct extent_io_tree *tree = eb->tree;
4236 spin_unlock(&eb->refs_lock);
4238 spin_lock(&tree->buffer_lock);
4239 radix_tree_delete(&tree->buffer,
4240 eb->start >> PAGE_CACHE_SHIFT);
4241 spin_unlock(&tree->buffer_lock);
4243 /* Should be safe to release our pages at this point */
4244 btrfs_release_extent_buffer_page(eb, 0);
4246 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4249 spin_unlock(&eb->refs_lock);
4252 void free_extent_buffer(struct extent_buffer *eb)
4257 spin_lock(&eb->refs_lock);
4258 if (atomic_read(&eb->refs) == 2 &&
4259 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4260 !extent_buffer_under_io(eb) &&
4261 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4262 atomic_dec(&eb->refs);
4265 * I know this is terrible, but it's temporary until we stop tracking
4266 * the uptodate bits and such for the extent buffers.
4268 release_extent_buffer(eb, GFP_ATOMIC);
4271 void free_extent_buffer_stale(struct extent_buffer *eb)
4276 spin_lock(&eb->refs_lock);
4277 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4279 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4280 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4281 atomic_dec(&eb->refs);
4282 release_extent_buffer(eb, GFP_NOFS);
4285 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4288 unsigned long num_pages;
4291 num_pages = num_extent_pages(eb->start, eb->len);
4293 for (i = 0; i < num_pages; i++) {
4294 page = extent_buffer_page(eb, i);
4295 if (!PageDirty(page))
4299 WARN_ON(!PagePrivate(page));
4301 clear_page_dirty_for_io(page);
4302 spin_lock_irq(&page->mapping->tree_lock);
4303 if (!PageDirty(page)) {
4304 radix_tree_tag_clear(&page->mapping->page_tree,
4306 PAGECACHE_TAG_DIRTY);
4308 spin_unlock_irq(&page->mapping->tree_lock);
4309 ClearPageError(page);
4312 WARN_ON(atomic_read(&eb->refs) == 0);
4315 int set_extent_buffer_dirty(struct extent_buffer *eb)
4318 unsigned long num_pages;
4321 check_buffer_tree_ref(eb);
4323 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4325 num_pages = num_extent_pages(eb->start, eb->len);
4326 WARN_ON(atomic_read(&eb->refs) == 0);
4327 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4329 for (i = 0; i < num_pages; i++)
4330 set_page_dirty(extent_buffer_page(eb, i));
4334 static int range_straddles_pages(u64 start, u64 len)
4336 if (len < PAGE_CACHE_SIZE)
4338 if (start & (PAGE_CACHE_SIZE - 1))
4340 if ((start + len) & (PAGE_CACHE_SIZE - 1))
4345 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4349 unsigned long num_pages;
4351 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4352 num_pages = num_extent_pages(eb->start, eb->len);
4353 for (i = 0; i < num_pages; i++) {
4354 page = extent_buffer_page(eb, i);
4356 ClearPageUptodate(page);
4361 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4365 unsigned long num_pages;
4367 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4368 num_pages = num_extent_pages(eb->start, eb->len);
4369 for (i = 0; i < num_pages; i++) {
4370 page = extent_buffer_page(eb, i);
4371 SetPageUptodate(page);
4376 int extent_range_uptodate(struct extent_io_tree *tree,
4381 int pg_uptodate = 1;
4383 unsigned long index;
4385 if (range_straddles_pages(start, end - start + 1)) {
4386 ret = test_range_bit(tree, start, end,
4387 EXTENT_UPTODATE, 1, NULL);
4391 while (start <= end) {
4392 index = start >> PAGE_CACHE_SHIFT;
4393 page = find_get_page(tree->mapping, index);
4396 uptodate = PageUptodate(page);
4397 page_cache_release(page);
4402 start += PAGE_CACHE_SIZE;
4407 int extent_buffer_uptodate(struct extent_buffer *eb)
4409 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4412 int read_extent_buffer_pages(struct extent_io_tree *tree,
4413 struct extent_buffer *eb, u64 start, int wait,
4414 get_extent_t *get_extent, int mirror_num)
4417 unsigned long start_i;
4421 int locked_pages = 0;
4422 int all_uptodate = 1;
4423 unsigned long num_pages;
4424 unsigned long num_reads = 0;
4425 struct bio *bio = NULL;
4426 unsigned long bio_flags = 0;
4428 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4432 WARN_ON(start < eb->start);
4433 start_i = (start >> PAGE_CACHE_SHIFT) -
4434 (eb->start >> PAGE_CACHE_SHIFT);
4439 num_pages = num_extent_pages(eb->start, eb->len);
4440 for (i = start_i; i < num_pages; i++) {
4441 page = extent_buffer_page(eb, i);
4442 if (wait == WAIT_NONE) {
4443 if (!trylock_page(page))
4449 if (!PageUptodate(page)) {
4456 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4460 clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4461 eb->failed_mirror = 0;
4462 atomic_set(&eb->io_pages, num_reads);
4463 for (i = start_i; i < num_pages; i++) {
4464 page = extent_buffer_page(eb, i);
4465 if (!PageUptodate(page)) {
4466 ClearPageError(page);
4467 err = __extent_read_full_page(tree, page,
4469 mirror_num, &bio_flags);
4478 err = submit_one_bio(READ, bio, mirror_num, bio_flags);
4483 if (ret || wait != WAIT_COMPLETE)
4486 for (i = start_i; i < num_pages; i++) {
4487 page = extent_buffer_page(eb, i);
4488 wait_on_page_locked(page);
4489 if (!PageUptodate(page))
4497 while (locked_pages > 0) {
4498 page = extent_buffer_page(eb, i);
4506 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4507 unsigned long start,
4514 char *dst = (char *)dstv;
4515 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4516 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4518 WARN_ON(start > eb->len);
4519 WARN_ON(start + len > eb->start + eb->len);
4521 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4524 page = extent_buffer_page(eb, i);
4526 cur = min(len, (PAGE_CACHE_SIZE - offset));
4527 kaddr = page_address(page);
4528 memcpy(dst, kaddr + offset, cur);
4537 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4538 unsigned long min_len, char **map,
4539 unsigned long *map_start,
4540 unsigned long *map_len)
4542 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4545 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4546 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4547 unsigned long end_i = (start_offset + start + min_len - 1) >>
4554 offset = start_offset;
4558 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4561 if (start + min_len > eb->len) {
4562 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4563 "wanted %lu %lu\n", (unsigned long long)eb->start,
4564 eb->len, start, min_len);
4569 p = extent_buffer_page(eb, i);
4570 kaddr = page_address(p);
4571 *map = kaddr + offset;
4572 *map_len = PAGE_CACHE_SIZE - offset;
4576 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4577 unsigned long start,
4584 char *ptr = (char *)ptrv;
4585 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4586 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4589 WARN_ON(start > eb->len);
4590 WARN_ON(start + len > eb->start + eb->len);
4592 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4595 page = extent_buffer_page(eb, i);
4597 cur = min(len, (PAGE_CACHE_SIZE - offset));
4599 kaddr = page_address(page);
4600 ret = memcmp(ptr, kaddr + offset, cur);
4612 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4613 unsigned long start, unsigned long len)
4619 char *src = (char *)srcv;
4620 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4621 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4623 WARN_ON(start > eb->len);
4624 WARN_ON(start + len > eb->start + eb->len);
4626 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4629 page = extent_buffer_page(eb, i);
4630 WARN_ON(!PageUptodate(page));
4632 cur = min(len, PAGE_CACHE_SIZE - offset);
4633 kaddr = page_address(page);
4634 memcpy(kaddr + offset, src, cur);
4643 void memset_extent_buffer(struct extent_buffer *eb, char c,
4644 unsigned long start, unsigned long len)
4650 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4651 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4653 WARN_ON(start > eb->len);
4654 WARN_ON(start + len > eb->start + eb->len);
4656 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4659 page = extent_buffer_page(eb, i);
4660 WARN_ON(!PageUptodate(page));
4662 cur = min(len, PAGE_CACHE_SIZE - offset);
4663 kaddr = page_address(page);
4664 memset(kaddr + offset, c, cur);
4672 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4673 unsigned long dst_offset, unsigned long src_offset,
4676 u64 dst_len = dst->len;
4681 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4682 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4684 WARN_ON(src->len != dst_len);
4686 offset = (start_offset + dst_offset) &
4687 ((unsigned long)PAGE_CACHE_SIZE - 1);
4690 page = extent_buffer_page(dst, i);
4691 WARN_ON(!PageUptodate(page));
4693 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4695 kaddr = page_address(page);
4696 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4705 static void move_pages(struct page *dst_page, struct page *src_page,
4706 unsigned long dst_off, unsigned long src_off,
4709 char *dst_kaddr = page_address(dst_page);
4710 if (dst_page == src_page) {
4711 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4713 char *src_kaddr = page_address(src_page);
4714 char *p = dst_kaddr + dst_off + len;
4715 char *s = src_kaddr + src_off + len;
4722 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4724 unsigned long distance = (src > dst) ? src - dst : dst - src;
4725 return distance < len;
4728 static void copy_pages(struct page *dst_page, struct page *src_page,
4729 unsigned long dst_off, unsigned long src_off,
4732 char *dst_kaddr = page_address(dst_page);
4734 int must_memmove = 0;
4736 if (dst_page != src_page) {
4737 src_kaddr = page_address(src_page);
4739 src_kaddr = dst_kaddr;
4740 if (areas_overlap(src_off, dst_off, len))
4745 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4747 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4750 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4751 unsigned long src_offset, unsigned long len)
4754 size_t dst_off_in_page;
4755 size_t src_off_in_page;
4756 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4757 unsigned long dst_i;
4758 unsigned long src_i;
4760 if (src_offset + len > dst->len) {
4761 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4762 "len %lu dst len %lu\n", src_offset, len, dst->len);
4765 if (dst_offset + len > dst->len) {
4766 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4767 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4772 dst_off_in_page = (start_offset + dst_offset) &
4773 ((unsigned long)PAGE_CACHE_SIZE - 1);
4774 src_off_in_page = (start_offset + src_offset) &
4775 ((unsigned long)PAGE_CACHE_SIZE - 1);
4777 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4778 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4780 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4782 cur = min_t(unsigned long, cur,
4783 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4785 copy_pages(extent_buffer_page(dst, dst_i),
4786 extent_buffer_page(dst, src_i),
4787 dst_off_in_page, src_off_in_page, cur);
4795 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4796 unsigned long src_offset, unsigned long len)
4799 size_t dst_off_in_page;
4800 size_t src_off_in_page;
4801 unsigned long dst_end = dst_offset + len - 1;
4802 unsigned long src_end = src_offset + len - 1;
4803 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4804 unsigned long dst_i;
4805 unsigned long src_i;
4807 if (src_offset + len > dst->len) {
4808 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4809 "len %lu len %lu\n", src_offset, len, dst->len);
4812 if (dst_offset + len > dst->len) {
4813 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4814 "len %lu len %lu\n", dst_offset, len, dst->len);
4817 if (dst_offset < src_offset) {
4818 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4822 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4823 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4825 dst_off_in_page = (start_offset + dst_end) &
4826 ((unsigned long)PAGE_CACHE_SIZE - 1);
4827 src_off_in_page = (start_offset + src_end) &
4828 ((unsigned long)PAGE_CACHE_SIZE - 1);
4830 cur = min_t(unsigned long, len, src_off_in_page + 1);
4831 cur = min(cur, dst_off_in_page + 1);
4832 move_pages(extent_buffer_page(dst, dst_i),
4833 extent_buffer_page(dst, src_i),
4834 dst_off_in_page - cur + 1,
4835 src_off_in_page - cur + 1, cur);
4843 int try_release_extent_buffer(struct page *page, gfp_t mask)
4845 struct extent_buffer *eb;
4848 * We need to make sure noboody is attaching this page to an eb right
4851 spin_lock(&page->mapping->private_lock);
4852 if (!PagePrivate(page)) {
4853 spin_unlock(&page->mapping->private_lock);
4857 eb = (struct extent_buffer *)page->private;
4861 * This is a little awful but should be ok, we need to make sure that
4862 * the eb doesn't disappear out from under us while we're looking at
4865 spin_lock(&eb->refs_lock);
4866 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4867 spin_unlock(&eb->refs_lock);
4868 spin_unlock(&page->mapping->private_lock);
4871 spin_unlock(&page->mapping->private_lock);
4873 if ((mask & GFP_NOFS) == GFP_NOFS)
4877 * If tree ref isn't set then we know the ref on this eb is a real ref,
4878 * so just return, this page will likely be freed soon anyway.
4880 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4881 spin_unlock(&eb->refs_lock);
4884 release_extent_buffer(eb, mask);