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"
22 static struct kmem_cache *extent_state_cache;
23 static struct kmem_cache *extent_buffer_cache;
25 static LIST_HEAD(buffers);
26 static LIST_HEAD(states);
30 static DEFINE_SPINLOCK(leak_lock);
33 #define BUFFER_LRU_MAX 64
38 struct rb_node rb_node;
41 struct extent_page_data {
43 struct extent_io_tree *tree;
44 get_extent_t *get_extent;
46 /* tells writepage not to lock the state bits for this range
47 * it still does the unlocking
49 unsigned int extent_locked:1;
51 /* tells the submit_bio code to use a WRITE_SYNC */
52 unsigned int sync_io:1;
55 int __init extent_io_init(void)
57 extent_state_cache = kmem_cache_create("extent_state",
58 sizeof(struct extent_state), 0,
59 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
60 if (!extent_state_cache)
63 extent_buffer_cache = kmem_cache_create("extent_buffers",
64 sizeof(struct extent_buffer), 0,
65 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
66 if (!extent_buffer_cache)
67 goto free_state_cache;
71 kmem_cache_destroy(extent_state_cache);
75 void extent_io_exit(void)
77 struct extent_state *state;
78 struct extent_buffer *eb;
80 while (!list_empty(&states)) {
81 state = list_entry(states.next, struct extent_state, leak_list);
82 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
83 "state %lu in tree %p refs %d\n",
84 (unsigned long long)state->start,
85 (unsigned long long)state->end,
86 state->state, state->tree, atomic_read(&state->refs));
87 list_del(&state->leak_list);
88 kmem_cache_free(extent_state_cache, state);
92 while (!list_empty(&buffers)) {
93 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
94 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
95 "refs %d\n", (unsigned long long)eb->start,
96 eb->len, atomic_read(&eb->refs));
97 list_del(&eb->leak_list);
98 kmem_cache_free(extent_buffer_cache, eb);
100 if (extent_state_cache)
101 kmem_cache_destroy(extent_state_cache);
102 if (extent_buffer_cache)
103 kmem_cache_destroy(extent_buffer_cache);
106 void extent_io_tree_init(struct extent_io_tree *tree,
107 struct address_space *mapping)
109 tree->state = RB_ROOT;
110 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
112 tree->dirty_bytes = 0;
113 spin_lock_init(&tree->lock);
114 spin_lock_init(&tree->buffer_lock);
115 tree->mapping = mapping;
118 static struct extent_state *alloc_extent_state(gfp_t mask)
120 struct extent_state *state;
125 state = kmem_cache_alloc(extent_state_cache, mask);
132 spin_lock_irqsave(&leak_lock, flags);
133 list_add(&state->leak_list, &states);
134 spin_unlock_irqrestore(&leak_lock, flags);
136 atomic_set(&state->refs, 1);
137 init_waitqueue_head(&state->wq);
141 void free_extent_state(struct extent_state *state)
145 if (atomic_dec_and_test(&state->refs)) {
149 WARN_ON(state->tree);
151 spin_lock_irqsave(&leak_lock, flags);
152 list_del(&state->leak_list);
153 spin_unlock_irqrestore(&leak_lock, flags);
155 kmem_cache_free(extent_state_cache, state);
159 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
160 struct rb_node *node)
162 struct rb_node **p = &root->rb_node;
163 struct rb_node *parent = NULL;
164 struct tree_entry *entry;
168 entry = rb_entry(parent, struct tree_entry, rb_node);
170 if (offset < entry->start)
172 else if (offset > entry->end)
178 entry = rb_entry(node, struct tree_entry, rb_node);
179 rb_link_node(node, parent, p);
180 rb_insert_color(node, root);
184 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
185 struct rb_node **prev_ret,
186 struct rb_node **next_ret)
188 struct rb_root *root = &tree->state;
189 struct rb_node *n = root->rb_node;
190 struct rb_node *prev = NULL;
191 struct rb_node *orig_prev = NULL;
192 struct tree_entry *entry;
193 struct tree_entry *prev_entry = NULL;
196 entry = rb_entry(n, struct tree_entry, rb_node);
200 if (offset < entry->start)
202 else if (offset > entry->end)
210 while (prev && offset > prev_entry->end) {
211 prev = rb_next(prev);
212 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
219 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
220 while (prev && offset < prev_entry->start) {
221 prev = rb_prev(prev);
222 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
229 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
232 struct rb_node *prev = NULL;
235 ret = __etree_search(tree, offset, &prev, NULL);
241 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
242 struct extent_state *other)
244 if (tree->ops && tree->ops->merge_extent_hook)
245 tree->ops->merge_extent_hook(tree->mapping->host, new,
250 * utility function to look for merge candidates inside a given range.
251 * Any extents with matching state are merged together into a single
252 * extent in the tree. Extents with EXTENT_IO in their state field
253 * are not merged because the end_io handlers need to be able to do
254 * operations on them without sleeping (or doing allocations/splits).
256 * This should be called with the tree lock held.
258 static void merge_state(struct extent_io_tree *tree,
259 struct extent_state *state)
261 struct extent_state *other;
262 struct rb_node *other_node;
264 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
267 other_node = rb_prev(&state->rb_node);
269 other = rb_entry(other_node, struct extent_state, rb_node);
270 if (other->end == state->start - 1 &&
271 other->state == state->state) {
272 merge_cb(tree, state, other);
273 state->start = other->start;
275 rb_erase(&other->rb_node, &tree->state);
276 free_extent_state(other);
279 other_node = rb_next(&state->rb_node);
281 other = rb_entry(other_node, struct extent_state, rb_node);
282 if (other->start == state->end + 1 &&
283 other->state == state->state) {
284 merge_cb(tree, state, other);
285 state->end = other->end;
287 rb_erase(&other->rb_node, &tree->state);
288 free_extent_state(other);
293 static void set_state_cb(struct extent_io_tree *tree,
294 struct extent_state *state, int *bits)
296 if (tree->ops && tree->ops->set_bit_hook)
297 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
300 static void clear_state_cb(struct extent_io_tree *tree,
301 struct extent_state *state, int *bits)
303 if (tree->ops && tree->ops->clear_bit_hook)
304 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
307 static void set_state_bits(struct extent_io_tree *tree,
308 struct extent_state *state, int *bits);
311 * insert an extent_state struct into the tree. 'bits' are set on the
312 * struct before it is inserted.
314 * This may return -EEXIST if the extent is already there, in which case the
315 * state struct is freed.
317 * The tree lock is not taken internally. This is a utility function and
318 * probably isn't what you want to call (see set/clear_extent_bit).
320 static int insert_state(struct extent_io_tree *tree,
321 struct extent_state *state, u64 start, u64 end,
324 struct rb_node *node;
327 printk(KERN_ERR "btrfs end < start %llu %llu\n",
328 (unsigned long long)end,
329 (unsigned long long)start);
332 state->start = start;
335 set_state_bits(tree, state, bits);
337 node = tree_insert(&tree->state, end, &state->rb_node);
339 struct extent_state *found;
340 found = rb_entry(node, struct extent_state, rb_node);
341 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
342 "%llu %llu\n", (unsigned long long)found->start,
343 (unsigned long long)found->end,
344 (unsigned long long)start, (unsigned long long)end);
348 merge_state(tree, state);
352 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
355 if (tree->ops && tree->ops->split_extent_hook)
356 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
360 * split a given extent state struct in two, inserting the preallocated
361 * struct 'prealloc' as the newly created second half. 'split' indicates an
362 * offset inside 'orig' where it should be split.
365 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
366 * are two extent state structs in the tree:
367 * prealloc: [orig->start, split - 1]
368 * orig: [ split, orig->end ]
370 * The tree locks are not taken by this function. They need to be held
373 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
374 struct extent_state *prealloc, u64 split)
376 struct rb_node *node;
378 split_cb(tree, orig, split);
380 prealloc->start = orig->start;
381 prealloc->end = split - 1;
382 prealloc->state = orig->state;
385 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
387 free_extent_state(prealloc);
390 prealloc->tree = tree;
395 * utility function to clear some bits in an extent state struct.
396 * it will optionally wake up any one waiting on this state (wake == 1), or
397 * forcibly remove the state from the tree (delete == 1).
399 * If no bits are set on the state struct after clearing things, the
400 * struct is freed and removed from the tree
402 static int clear_state_bit(struct extent_io_tree *tree,
403 struct extent_state *state,
406 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
407 int ret = state->state & bits_to_clear;
409 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
410 u64 range = state->end - state->start + 1;
411 WARN_ON(range > tree->dirty_bytes);
412 tree->dirty_bytes -= range;
414 clear_state_cb(tree, state, bits);
415 state->state &= ~bits_to_clear;
418 if (state->state == 0) {
420 rb_erase(&state->rb_node, &tree->state);
422 free_extent_state(state);
427 merge_state(tree, state);
432 static struct extent_state *
433 alloc_extent_state_atomic(struct extent_state *prealloc)
436 prealloc = alloc_extent_state(GFP_ATOMIC);
442 * clear some bits on a range in the tree. This may require splitting
443 * or inserting elements in the tree, so the gfp mask is used to
444 * indicate which allocations or sleeping are allowed.
446 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
447 * the given range from the tree regardless of state (ie for truncate).
449 * the range [start, end] is inclusive.
451 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
452 * bits were already set, or zero if none of the bits were already set.
454 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
455 int bits, int wake, int delete,
456 struct extent_state **cached_state,
459 struct extent_state *state;
460 struct extent_state *cached;
461 struct extent_state *prealloc = NULL;
462 struct rb_node *next_node;
463 struct rb_node *node;
470 bits |= ~EXTENT_CTLBITS;
471 bits |= EXTENT_FIRST_DELALLOC;
473 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
476 if (!prealloc && (mask & __GFP_WAIT)) {
477 prealloc = alloc_extent_state(mask);
482 spin_lock(&tree->lock);
484 cached = *cached_state;
487 *cached_state = NULL;
491 if (cached && cached->tree && cached->start <= start &&
492 cached->end > start) {
494 atomic_dec(&cached->refs);
499 free_extent_state(cached);
502 * this search will find the extents that end after
505 node = tree_search(tree, start);
508 state = rb_entry(node, struct extent_state, rb_node);
510 if (state->start > end)
512 WARN_ON(state->end < start);
513 last_end = state->end;
516 * | ---- desired range ---- |
518 * | ------------- state -------------- |
520 * We need to split the extent we found, and may flip
521 * bits on second half.
523 * If the extent we found extends past our range, we
524 * just split and search again. It'll get split again
525 * the next time though.
527 * If the extent we found is inside our range, we clear
528 * the desired bit on it.
531 if (state->start < start) {
532 prealloc = alloc_extent_state_atomic(prealloc);
534 err = split_state(tree, state, prealloc, start);
535 BUG_ON(err == -EEXIST);
539 if (state->end <= end) {
540 set |= clear_state_bit(tree, state, &bits, wake);
541 if (last_end == (u64)-1)
543 start = last_end + 1;
548 * | ---- desired range ---- |
550 * We need to split the extent, and clear the bit
553 if (state->start <= end && state->end > end) {
554 prealloc = alloc_extent_state_atomic(prealloc);
556 err = split_state(tree, state, prealloc, end + 1);
557 BUG_ON(err == -EEXIST);
561 set |= clear_state_bit(tree, prealloc, &bits, wake);
567 if (state->end < end && prealloc && !need_resched())
568 next_node = rb_next(&state->rb_node);
572 set |= clear_state_bit(tree, state, &bits, wake);
573 if (last_end == (u64)-1)
575 start = last_end + 1;
576 if (start <= end && next_node) {
577 state = rb_entry(next_node, struct extent_state,
579 if (state->start == start)
585 spin_unlock(&tree->lock);
587 free_extent_state(prealloc);
594 spin_unlock(&tree->lock);
595 if (mask & __GFP_WAIT)
600 static int wait_on_state(struct extent_io_tree *tree,
601 struct extent_state *state)
602 __releases(tree->lock)
603 __acquires(tree->lock)
606 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
607 spin_unlock(&tree->lock);
609 spin_lock(&tree->lock);
610 finish_wait(&state->wq, &wait);
615 * waits for one or more bits to clear on a range in the state tree.
616 * The range [start, end] is inclusive.
617 * The tree lock is taken by this function
619 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
621 struct extent_state *state;
622 struct rb_node *node;
624 spin_lock(&tree->lock);
628 * this search will find all the extents that end after
631 node = tree_search(tree, start);
635 state = rb_entry(node, struct extent_state, rb_node);
637 if (state->start > end)
640 if (state->state & bits) {
641 start = state->start;
642 atomic_inc(&state->refs);
643 wait_on_state(tree, state);
644 free_extent_state(state);
647 start = state->end + 1;
652 cond_resched_lock(&tree->lock);
655 spin_unlock(&tree->lock);
659 static void set_state_bits(struct extent_io_tree *tree,
660 struct extent_state *state,
663 int bits_to_set = *bits & ~EXTENT_CTLBITS;
665 set_state_cb(tree, state, bits);
666 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
667 u64 range = state->end - state->start + 1;
668 tree->dirty_bytes += range;
670 state->state |= bits_to_set;
673 static void cache_state(struct extent_state *state,
674 struct extent_state **cached_ptr)
676 if (cached_ptr && !(*cached_ptr)) {
677 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
679 atomic_inc(&state->refs);
684 static void uncache_state(struct extent_state **cached_ptr)
686 if (cached_ptr && (*cached_ptr)) {
687 struct extent_state *state = *cached_ptr;
689 free_extent_state(state);
694 * set some bits on a range in the tree. This may require allocations or
695 * sleeping, so the gfp mask is used to indicate what is allowed.
697 * If any of the exclusive bits are set, this will fail with -EEXIST if some
698 * part of the range already has the desired bits set. The start of the
699 * existing range is returned in failed_start in this case.
701 * [start, end] is inclusive This takes the tree lock.
704 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
705 int bits, int exclusive_bits, u64 *failed_start,
706 struct extent_state **cached_state, gfp_t mask)
708 struct extent_state *state;
709 struct extent_state *prealloc = NULL;
710 struct rb_node *node;
715 bits |= EXTENT_FIRST_DELALLOC;
717 if (!prealloc && (mask & __GFP_WAIT)) {
718 prealloc = alloc_extent_state(mask);
722 spin_lock(&tree->lock);
723 if (cached_state && *cached_state) {
724 state = *cached_state;
725 if (state->start <= start && state->end > start &&
727 node = &state->rb_node;
732 * this search will find all the extents that end after
735 node = tree_search(tree, start);
737 prealloc = alloc_extent_state_atomic(prealloc);
739 err = insert_state(tree, prealloc, start, end, &bits);
741 BUG_ON(err == -EEXIST);
744 state = rb_entry(node, struct extent_state, rb_node);
746 last_start = state->start;
747 last_end = state->end;
750 * | ---- desired range ---- |
753 * Just lock what we found and keep going
755 if (state->start == start && state->end <= end) {
756 struct rb_node *next_node;
757 if (state->state & exclusive_bits) {
758 *failed_start = state->start;
763 set_state_bits(tree, state, &bits);
765 cache_state(state, cached_state);
766 merge_state(tree, state);
767 if (last_end == (u64)-1)
770 start = last_end + 1;
771 next_node = rb_next(&state->rb_node);
772 if (next_node && start < end && prealloc && !need_resched()) {
773 state = rb_entry(next_node, struct extent_state,
775 if (state->start == start)
782 * | ---- desired range ---- |
785 * | ------------- state -------------- |
787 * We need to split the extent we found, and may flip bits on
790 * If the extent we found extends past our
791 * range, we just split and search again. It'll get split
792 * again the next time though.
794 * If the extent we found is inside our range, we set the
797 if (state->start < start) {
798 if (state->state & exclusive_bits) {
799 *failed_start = start;
804 prealloc = alloc_extent_state_atomic(prealloc);
806 err = split_state(tree, state, prealloc, start);
807 BUG_ON(err == -EEXIST);
811 if (state->end <= end) {
812 set_state_bits(tree, state, &bits);
813 cache_state(state, cached_state);
814 merge_state(tree, state);
815 if (last_end == (u64)-1)
817 start = last_end + 1;
822 * | ---- desired range ---- |
823 * | state | or | state |
825 * There's a hole, we need to insert something in it and
826 * ignore the extent we found.
828 if (state->start > start) {
830 if (end < last_start)
833 this_end = last_start - 1;
835 prealloc = alloc_extent_state_atomic(prealloc);
839 * Avoid to free 'prealloc' if it can be merged with
842 err = insert_state(tree, prealloc, start, this_end,
844 BUG_ON(err == -EEXIST);
846 free_extent_state(prealloc);
850 cache_state(prealloc, cached_state);
852 start = this_end + 1;
856 * | ---- desired range ---- |
858 * We need to split the extent, and set the bit
861 if (state->start <= end && state->end > end) {
862 if (state->state & exclusive_bits) {
863 *failed_start = start;
868 prealloc = alloc_extent_state_atomic(prealloc);
870 err = split_state(tree, state, prealloc, end + 1);
871 BUG_ON(err == -EEXIST);
873 set_state_bits(tree, prealloc, &bits);
874 cache_state(prealloc, cached_state);
875 merge_state(tree, prealloc);
883 spin_unlock(&tree->lock);
885 free_extent_state(prealloc);
892 spin_unlock(&tree->lock);
893 if (mask & __GFP_WAIT)
899 * convert_extent - convert all bits in a given range from one bit to another
900 * @tree: the io tree to search
901 * @start: the start offset in bytes
902 * @end: the end offset in bytes (inclusive)
903 * @bits: the bits to set in this range
904 * @clear_bits: the bits to clear in this range
905 * @mask: the allocation mask
907 * This will go through and set bits for the given range. If any states exist
908 * already in this range they are set with the given bit and cleared of the
909 * clear_bits. This is only meant to be used by things that are mergeable, ie
910 * converting from say DELALLOC to DIRTY. This is not meant to be used with
911 * boundary bits like LOCK.
913 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
914 int bits, int clear_bits, gfp_t mask)
916 struct extent_state *state;
917 struct extent_state *prealloc = NULL;
918 struct rb_node *node;
924 if (!prealloc && (mask & __GFP_WAIT)) {
925 prealloc = alloc_extent_state(mask);
930 spin_lock(&tree->lock);
932 * this search will find all the extents that end after
935 node = tree_search(tree, start);
937 prealloc = alloc_extent_state_atomic(prealloc);
940 err = insert_state(tree, prealloc, start, end, &bits);
942 BUG_ON(err == -EEXIST);
945 state = rb_entry(node, struct extent_state, rb_node);
947 last_start = state->start;
948 last_end = state->end;
951 * | ---- desired range ---- |
954 * Just lock what we found and keep going
956 if (state->start == start && state->end <= end) {
957 struct rb_node *next_node;
959 set_state_bits(tree, state, &bits);
960 clear_state_bit(tree, state, &clear_bits, 0);
962 merge_state(tree, state);
963 if (last_end == (u64)-1)
966 start = last_end + 1;
967 next_node = rb_next(&state->rb_node);
968 if (next_node && start < end && prealloc && !need_resched()) {
969 state = rb_entry(next_node, struct extent_state,
971 if (state->start == start)
978 * | ---- desired range ---- |
981 * | ------------- state -------------- |
983 * We need to split the extent we found, and may flip bits on
986 * If the extent we found extends past our
987 * range, we just split and search again. It'll get split
988 * again the next time though.
990 * If the extent we found is inside our range, we set the
993 if (state->start < start) {
994 prealloc = alloc_extent_state_atomic(prealloc);
997 err = split_state(tree, state, prealloc, start);
998 BUG_ON(err == -EEXIST);
1002 if (state->end <= end) {
1003 set_state_bits(tree, state, &bits);
1004 clear_state_bit(tree, state, &clear_bits, 0);
1005 merge_state(tree, state);
1006 if (last_end == (u64)-1)
1008 start = last_end + 1;
1013 * | ---- desired range ---- |
1014 * | state | or | state |
1016 * There's a hole, we need to insert something in it and
1017 * ignore the extent we found.
1019 if (state->start > start) {
1021 if (end < last_start)
1024 this_end = last_start - 1;
1026 prealloc = alloc_extent_state_atomic(prealloc);
1031 * Avoid to free 'prealloc' if it can be merged with
1034 err = insert_state(tree, prealloc, start, this_end,
1036 BUG_ON(err == -EEXIST);
1038 free_extent_state(prealloc);
1043 start = this_end + 1;
1047 * | ---- desired range ---- |
1049 * We need to split the extent, and set the bit
1052 if (state->start <= end && state->end > end) {
1053 prealloc = alloc_extent_state_atomic(prealloc);
1057 err = split_state(tree, state, prealloc, end + 1);
1058 BUG_ON(err == -EEXIST);
1060 set_state_bits(tree, prealloc, &bits);
1061 clear_state_bit(tree, prealloc, &clear_bits, 0);
1063 merge_state(tree, prealloc);
1071 spin_unlock(&tree->lock);
1073 free_extent_state(prealloc);
1080 spin_unlock(&tree->lock);
1081 if (mask & __GFP_WAIT)
1086 /* wrappers around set/clear extent bit */
1087 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1090 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
1094 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1095 int bits, gfp_t mask)
1097 return set_extent_bit(tree, start, end, bits, 0, NULL,
1101 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1102 int bits, gfp_t mask)
1104 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1107 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1108 struct extent_state **cached_state, gfp_t mask)
1110 return set_extent_bit(tree, start, end,
1111 EXTENT_DELALLOC | EXTENT_UPTODATE,
1112 0, NULL, cached_state, mask);
1115 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1118 return clear_extent_bit(tree, start, end,
1119 EXTENT_DIRTY | EXTENT_DELALLOC |
1120 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1123 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1126 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
1130 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1131 struct extent_state **cached_state, gfp_t mask)
1133 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1134 NULL, cached_state, mask);
1137 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
1138 u64 end, struct extent_state **cached_state,
1141 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1142 cached_state, mask);
1146 * either insert or lock state struct between start and end use mask to tell
1147 * us if waiting is desired.
1149 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1150 int bits, struct extent_state **cached_state, gfp_t mask)
1155 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1156 EXTENT_LOCKED, &failed_start,
1157 cached_state, mask);
1158 if (err == -EEXIST && (mask & __GFP_WAIT)) {
1159 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1160 start = failed_start;
1164 WARN_ON(start > end);
1169 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1171 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1174 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1180 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1181 &failed_start, NULL, mask);
1182 if (err == -EEXIST) {
1183 if (failed_start > start)
1184 clear_extent_bit(tree, start, failed_start - 1,
1185 EXTENT_LOCKED, 1, 0, NULL, mask);
1191 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1192 struct extent_state **cached, gfp_t mask)
1194 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1198 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1200 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1205 * helper function to set both pages and extents in the tree writeback
1207 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1209 unsigned long index = start >> PAGE_CACHE_SHIFT;
1210 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1213 while (index <= end_index) {
1214 page = find_get_page(tree->mapping, index);
1216 set_page_writeback(page);
1217 page_cache_release(page);
1223 /* find the first state struct with 'bits' set after 'start', and
1224 * return it. tree->lock must be held. NULL will returned if
1225 * nothing was found after 'start'
1227 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1228 u64 start, int bits)
1230 struct rb_node *node;
1231 struct extent_state *state;
1234 * this search will find all the extents that end after
1237 node = tree_search(tree, start);
1242 state = rb_entry(node, struct extent_state, rb_node);
1243 if (state->end >= start && (state->state & bits))
1246 node = rb_next(node);
1255 * find the first offset in the io tree with 'bits' set. zero is
1256 * returned if we find something, and *start_ret and *end_ret are
1257 * set to reflect the state struct that was found.
1259 * If nothing was found, 1 is returned, < 0 on error
1261 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1262 u64 *start_ret, u64 *end_ret, int bits)
1264 struct extent_state *state;
1267 spin_lock(&tree->lock);
1268 state = find_first_extent_bit_state(tree, start, bits);
1270 *start_ret = state->start;
1271 *end_ret = state->end;
1274 spin_unlock(&tree->lock);
1279 * find a contiguous range of bytes in the file marked as delalloc, not
1280 * more than 'max_bytes'. start and end are used to return the range,
1282 * 1 is returned if we find something, 0 if nothing was in the tree
1284 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1285 u64 *start, u64 *end, u64 max_bytes,
1286 struct extent_state **cached_state)
1288 struct rb_node *node;
1289 struct extent_state *state;
1290 u64 cur_start = *start;
1292 u64 total_bytes = 0;
1294 spin_lock(&tree->lock);
1297 * this search will find all the extents that end after
1300 node = tree_search(tree, cur_start);
1308 state = rb_entry(node, struct extent_state, rb_node);
1309 if (found && (state->start != cur_start ||
1310 (state->state & EXTENT_BOUNDARY))) {
1313 if (!(state->state & EXTENT_DELALLOC)) {
1319 *start = state->start;
1320 *cached_state = state;
1321 atomic_inc(&state->refs);
1325 cur_start = state->end + 1;
1326 node = rb_next(node);
1329 total_bytes += state->end - state->start + 1;
1330 if (total_bytes >= max_bytes)
1334 spin_unlock(&tree->lock);
1338 static noinline int __unlock_for_delalloc(struct inode *inode,
1339 struct page *locked_page,
1343 struct page *pages[16];
1344 unsigned long index = start >> PAGE_CACHE_SHIFT;
1345 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1346 unsigned long nr_pages = end_index - index + 1;
1349 if (index == locked_page->index && end_index == index)
1352 while (nr_pages > 0) {
1353 ret = find_get_pages_contig(inode->i_mapping, index,
1354 min_t(unsigned long, nr_pages,
1355 ARRAY_SIZE(pages)), pages);
1356 for (i = 0; i < ret; i++) {
1357 if (pages[i] != locked_page)
1358 unlock_page(pages[i]);
1359 page_cache_release(pages[i]);
1368 static noinline int lock_delalloc_pages(struct inode *inode,
1369 struct page *locked_page,
1373 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1374 unsigned long start_index = index;
1375 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1376 unsigned long pages_locked = 0;
1377 struct page *pages[16];
1378 unsigned long nrpages;
1382 /* the caller is responsible for locking the start index */
1383 if (index == locked_page->index && index == end_index)
1386 /* skip the page at the start index */
1387 nrpages = end_index - index + 1;
1388 while (nrpages > 0) {
1389 ret = find_get_pages_contig(inode->i_mapping, index,
1390 min_t(unsigned long,
1391 nrpages, ARRAY_SIZE(pages)), pages);
1396 /* now we have an array of pages, lock them all */
1397 for (i = 0; i < ret; i++) {
1399 * the caller is taking responsibility for
1402 if (pages[i] != locked_page) {
1403 lock_page(pages[i]);
1404 if (!PageDirty(pages[i]) ||
1405 pages[i]->mapping != inode->i_mapping) {
1407 unlock_page(pages[i]);
1408 page_cache_release(pages[i]);
1412 page_cache_release(pages[i]);
1421 if (ret && pages_locked) {
1422 __unlock_for_delalloc(inode, locked_page,
1424 ((u64)(start_index + pages_locked - 1)) <<
1431 * find a contiguous range of bytes in the file marked as delalloc, not
1432 * more than 'max_bytes'. start and end are used to return the range,
1434 * 1 is returned if we find something, 0 if nothing was in the tree
1436 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1437 struct extent_io_tree *tree,
1438 struct page *locked_page,
1439 u64 *start, u64 *end,
1445 struct extent_state *cached_state = NULL;
1450 /* step one, find a bunch of delalloc bytes starting at start */
1451 delalloc_start = *start;
1453 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1454 max_bytes, &cached_state);
1455 if (!found || delalloc_end <= *start) {
1456 *start = delalloc_start;
1457 *end = delalloc_end;
1458 free_extent_state(cached_state);
1463 * start comes from the offset of locked_page. We have to lock
1464 * pages in order, so we can't process delalloc bytes before
1467 if (delalloc_start < *start)
1468 delalloc_start = *start;
1471 * make sure to limit the number of pages we try to lock down
1474 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1475 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1477 /* step two, lock all the pages after the page that has start */
1478 ret = lock_delalloc_pages(inode, locked_page,
1479 delalloc_start, delalloc_end);
1480 if (ret == -EAGAIN) {
1481 /* some of the pages are gone, lets avoid looping by
1482 * shortening the size of the delalloc range we're searching
1484 free_extent_state(cached_state);
1486 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1487 max_bytes = PAGE_CACHE_SIZE - offset;
1497 /* step three, lock the state bits for the whole range */
1498 lock_extent_bits(tree, delalloc_start, delalloc_end,
1499 0, &cached_state, GFP_NOFS);
1501 /* then test to make sure it is all still delalloc */
1502 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1503 EXTENT_DELALLOC, 1, cached_state);
1505 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1506 &cached_state, GFP_NOFS);
1507 __unlock_for_delalloc(inode, locked_page,
1508 delalloc_start, delalloc_end);
1512 free_extent_state(cached_state);
1513 *start = delalloc_start;
1514 *end = delalloc_end;
1519 int extent_clear_unlock_delalloc(struct inode *inode,
1520 struct extent_io_tree *tree,
1521 u64 start, u64 end, struct page *locked_page,
1525 struct page *pages[16];
1526 unsigned long index = start >> PAGE_CACHE_SHIFT;
1527 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1528 unsigned long nr_pages = end_index - index + 1;
1532 if (op & EXTENT_CLEAR_UNLOCK)
1533 clear_bits |= EXTENT_LOCKED;
1534 if (op & EXTENT_CLEAR_DIRTY)
1535 clear_bits |= EXTENT_DIRTY;
1537 if (op & EXTENT_CLEAR_DELALLOC)
1538 clear_bits |= EXTENT_DELALLOC;
1540 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1541 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1542 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1543 EXTENT_SET_PRIVATE2)))
1546 while (nr_pages > 0) {
1547 ret = find_get_pages_contig(inode->i_mapping, index,
1548 min_t(unsigned long,
1549 nr_pages, ARRAY_SIZE(pages)), pages);
1550 for (i = 0; i < ret; i++) {
1552 if (op & EXTENT_SET_PRIVATE2)
1553 SetPagePrivate2(pages[i]);
1555 if (pages[i] == locked_page) {
1556 page_cache_release(pages[i]);
1559 if (op & EXTENT_CLEAR_DIRTY)
1560 clear_page_dirty_for_io(pages[i]);
1561 if (op & EXTENT_SET_WRITEBACK)
1562 set_page_writeback(pages[i]);
1563 if (op & EXTENT_END_WRITEBACK)
1564 end_page_writeback(pages[i]);
1565 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1566 unlock_page(pages[i]);
1567 page_cache_release(pages[i]);
1577 * count the number of bytes in the tree that have a given bit(s)
1578 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1579 * cached. The total number found is returned.
1581 u64 count_range_bits(struct extent_io_tree *tree,
1582 u64 *start, u64 search_end, u64 max_bytes,
1583 unsigned long bits, int contig)
1585 struct rb_node *node;
1586 struct extent_state *state;
1587 u64 cur_start = *start;
1588 u64 total_bytes = 0;
1592 if (search_end <= cur_start) {
1597 spin_lock(&tree->lock);
1598 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1599 total_bytes = tree->dirty_bytes;
1603 * this search will find all the extents that end after
1606 node = tree_search(tree, cur_start);
1611 state = rb_entry(node, struct extent_state, rb_node);
1612 if (state->start > search_end)
1614 if (contig && found && state->start > last + 1)
1616 if (state->end >= cur_start && (state->state & bits) == bits) {
1617 total_bytes += min(search_end, state->end) + 1 -
1618 max(cur_start, state->start);
1619 if (total_bytes >= max_bytes)
1622 *start = max(cur_start, state->start);
1626 } else if (contig && found) {
1629 node = rb_next(node);
1634 spin_unlock(&tree->lock);
1639 * set the private field for a given byte offset in the tree. If there isn't
1640 * an extent_state there already, this does nothing.
1642 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1644 struct rb_node *node;
1645 struct extent_state *state;
1648 spin_lock(&tree->lock);
1650 * this search will find all the extents that end after
1653 node = tree_search(tree, start);
1658 state = rb_entry(node, struct extent_state, rb_node);
1659 if (state->start != start) {
1663 state->private = private;
1665 spin_unlock(&tree->lock);
1669 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1671 struct rb_node *node;
1672 struct extent_state *state;
1675 spin_lock(&tree->lock);
1677 * this search will find all the extents that end after
1680 node = tree_search(tree, start);
1685 state = rb_entry(node, struct extent_state, rb_node);
1686 if (state->start != start) {
1690 *private = state->private;
1692 spin_unlock(&tree->lock);
1697 * searches a range in the state tree for a given mask.
1698 * If 'filled' == 1, this returns 1 only if every extent in the tree
1699 * has the bits set. Otherwise, 1 is returned if any bit in the
1700 * range is found set.
1702 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1703 int bits, int filled, struct extent_state *cached)
1705 struct extent_state *state = NULL;
1706 struct rb_node *node;
1709 spin_lock(&tree->lock);
1710 if (cached && cached->tree && cached->start <= start &&
1711 cached->end > start)
1712 node = &cached->rb_node;
1714 node = tree_search(tree, start);
1715 while (node && start <= end) {
1716 state = rb_entry(node, struct extent_state, rb_node);
1718 if (filled && state->start > start) {
1723 if (state->start > end)
1726 if (state->state & bits) {
1730 } else if (filled) {
1735 if (state->end == (u64)-1)
1738 start = state->end + 1;
1741 node = rb_next(node);
1748 spin_unlock(&tree->lock);
1753 * helper function to set a given page up to date if all the
1754 * extents in the tree for that page are up to date
1756 static int check_page_uptodate(struct extent_io_tree *tree,
1759 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1760 u64 end = start + PAGE_CACHE_SIZE - 1;
1761 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1762 SetPageUptodate(page);
1767 * helper function to unlock a page if all the extents in the tree
1768 * for that page are unlocked
1770 static int check_page_locked(struct extent_io_tree *tree,
1773 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1774 u64 end = start + PAGE_CACHE_SIZE - 1;
1775 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1781 * helper function to end page writeback if all the extents
1782 * in the tree for that page are done with writeback
1784 static int check_page_writeback(struct extent_io_tree *tree,
1787 end_page_writeback(page);
1792 * When IO fails, either with EIO or csum verification fails, we
1793 * try other mirrors that might have a good copy of the data. This
1794 * io_failure_record is used to record state as we go through all the
1795 * mirrors. If another mirror has good data, the page is set up to date
1796 * and things continue. If a good mirror can't be found, the original
1797 * bio end_io callback is called to indicate things have failed.
1799 struct io_failure_record {
1804 unsigned long bio_flags;
1810 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1815 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1817 set_state_private(failure_tree, rec->start, 0);
1818 ret = clear_extent_bits(failure_tree, rec->start,
1819 rec->start + rec->len - 1,
1820 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1825 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1826 rec->start + rec->len - 1,
1827 EXTENT_DAMAGED, GFP_NOFS);
1836 static void repair_io_failure_callback(struct bio *bio, int err)
1838 complete(bio->bi_private);
1842 * this bypasses the standard btrfs submit functions deliberately, as
1843 * the standard behavior is to write all copies in a raid setup. here we only
1844 * want to write the one bad copy. so we do the mapping for ourselves and issue
1845 * submit_bio directly.
1846 * to avoid any synchonization issues, wait for the data after writing, which
1847 * actually prevents the read that triggered the error from finishing.
1848 * currently, there can be no more than two copies of every data bit. thus,
1849 * exactly one rewrite is required.
1851 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1852 u64 length, u64 logical, struct page *page,
1856 struct btrfs_device *dev;
1857 DECLARE_COMPLETION_ONSTACK(compl);
1860 struct btrfs_bio *bbio = NULL;
1863 BUG_ON(!mirror_num);
1865 bio = bio_alloc(GFP_NOFS, 1);
1868 bio->bi_private = &compl;
1869 bio->bi_end_io = repair_io_failure_callback;
1871 map_length = length;
1873 ret = btrfs_map_block(map_tree, WRITE, logical,
1874 &map_length, &bbio, mirror_num);
1879 BUG_ON(mirror_num != bbio->mirror_num);
1880 sector = bbio->stripes[mirror_num-1].physical >> 9;
1881 bio->bi_sector = sector;
1882 dev = bbio->stripes[mirror_num-1].dev;
1884 if (!dev || !dev->bdev || !dev->writeable) {
1888 bio->bi_bdev = dev->bdev;
1889 bio_add_page(bio, page, length, start-page_offset(page));
1890 submit_bio(WRITE_SYNC, bio);
1891 wait_for_completion(&compl);
1893 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1894 /* try to remap that extent elsewhere? */
1899 printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s "
1900 "sector %llu)\n", page->mapping->host->i_ino, start,
1908 * each time an IO finishes, we do a fast check in the IO failure tree
1909 * to see if we need to process or clean up an io_failure_record
1911 static int clean_io_failure(u64 start, struct page *page)
1914 u64 private_failure;
1915 struct io_failure_record *failrec;
1916 struct btrfs_mapping_tree *map_tree;
1917 struct extent_state *state;
1921 struct inode *inode = page->mapping->host;
1924 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1925 (u64)-1, 1, EXTENT_DIRTY, 0);
1929 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1934 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1935 BUG_ON(!failrec->this_mirror);
1937 if (failrec->in_validation) {
1938 /* there was no real error, just free the record */
1939 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1945 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1946 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1949 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1951 if (state && state->start == failrec->start) {
1952 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
1953 num_copies = btrfs_num_copies(map_tree, failrec->logical,
1955 if (num_copies > 1) {
1956 ret = repair_io_failure(map_tree, start, failrec->len,
1957 failrec->logical, page,
1958 failrec->failed_mirror);
1965 ret = free_io_failure(inode, failrec, did_repair);
1971 * this is a generic handler for readpage errors (default
1972 * readpage_io_failed_hook). if other copies exist, read those and write back
1973 * good data to the failed position. does not investigate in remapping the
1974 * failed extent elsewhere, hoping the device will be smart enough to do this as
1978 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
1979 u64 start, u64 end, int failed_mirror,
1980 struct extent_state *state)
1982 struct io_failure_record *failrec = NULL;
1984 struct extent_map *em;
1985 struct inode *inode = page->mapping->host;
1986 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1987 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1988 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1995 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
1997 ret = get_state_private(failure_tree, start, &private);
1999 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2002 failrec->start = start;
2003 failrec->len = end - start + 1;
2004 failrec->this_mirror = 0;
2005 failrec->bio_flags = 0;
2006 failrec->in_validation = 0;
2008 read_lock(&em_tree->lock);
2009 em = lookup_extent_mapping(em_tree, start, failrec->len);
2011 read_unlock(&em_tree->lock);
2016 if (em->start > start || em->start + em->len < start) {
2017 free_extent_map(em);
2020 read_unlock(&em_tree->lock);
2022 if (!em || IS_ERR(em)) {
2026 logical = start - em->start;
2027 logical = em->block_start + logical;
2028 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2029 logical = em->block_start;
2030 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2031 extent_set_compress_type(&failrec->bio_flags,
2034 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2035 "len=%llu\n", logical, start, failrec->len);
2036 failrec->logical = logical;
2037 free_extent_map(em);
2039 /* set the bits in the private failure tree */
2040 ret = set_extent_bits(failure_tree, start, end,
2041 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2043 ret = set_state_private(failure_tree, start,
2044 (u64)(unsigned long)failrec);
2045 /* set the bits in the inode's tree */
2047 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2054 failrec = (struct io_failure_record *)(unsigned long)private;
2055 pr_debug("bio_readpage_error: (found) logical=%llu, "
2056 "start=%llu, len=%llu, validation=%d\n",
2057 failrec->logical, failrec->start, failrec->len,
2058 failrec->in_validation);
2060 * when data can be on disk more than twice, add to failrec here
2061 * (e.g. with a list for failed_mirror) to make
2062 * clean_io_failure() clean all those errors at once.
2065 num_copies = btrfs_num_copies(
2066 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2067 failrec->logical, failrec->len);
2068 if (num_copies == 1) {
2070 * we only have a single copy of the data, so don't bother with
2071 * all the retry and error correction code that follows. no
2072 * matter what the error is, it is very likely to persist.
2074 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2075 "state=%p, num_copies=%d, next_mirror %d, "
2076 "failed_mirror %d\n", state, num_copies,
2077 failrec->this_mirror, failed_mirror);
2078 free_io_failure(inode, failrec, 0);
2083 spin_lock(&tree->lock);
2084 state = find_first_extent_bit_state(tree, failrec->start,
2086 if (state && state->start != failrec->start)
2088 spin_unlock(&tree->lock);
2092 * there are two premises:
2093 * a) deliver good data to the caller
2094 * b) correct the bad sectors on disk
2096 if (failed_bio->bi_vcnt > 1) {
2098 * to fulfill b), we need to know the exact failing sectors, as
2099 * we don't want to rewrite any more than the failed ones. thus,
2100 * we need separate read requests for the failed bio
2102 * if the following BUG_ON triggers, our validation request got
2103 * merged. we need separate requests for our algorithm to work.
2105 BUG_ON(failrec->in_validation);
2106 failrec->in_validation = 1;
2107 failrec->this_mirror = failed_mirror;
2108 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2111 * we're ready to fulfill a) and b) alongside. get a good copy
2112 * of the failed sector and if we succeed, we have setup
2113 * everything for repair_io_failure to do the rest for us.
2115 if (failrec->in_validation) {
2116 BUG_ON(failrec->this_mirror != failed_mirror);
2117 failrec->in_validation = 0;
2118 failrec->this_mirror = 0;
2120 failrec->failed_mirror = failed_mirror;
2121 failrec->this_mirror++;
2122 if (failrec->this_mirror == failed_mirror)
2123 failrec->this_mirror++;
2124 read_mode = READ_SYNC;
2127 if (!state || failrec->this_mirror > num_copies) {
2128 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2129 "next_mirror %d, failed_mirror %d\n", state,
2130 num_copies, failrec->this_mirror, failed_mirror);
2131 free_io_failure(inode, failrec, 0);
2135 bio = bio_alloc(GFP_NOFS, 1);
2136 bio->bi_private = state;
2137 bio->bi_end_io = failed_bio->bi_end_io;
2138 bio->bi_sector = failrec->logical >> 9;
2139 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2142 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2144 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2145 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2146 failrec->this_mirror, num_copies, failrec->in_validation);
2148 tree->ops->submit_bio_hook(inode, read_mode, bio, failrec->this_mirror,
2149 failrec->bio_flags, 0);
2153 /* lots and lots of room for performance fixes in the end_bio funcs */
2156 * after a writepage IO is done, we need to:
2157 * clear the uptodate bits on error
2158 * clear the writeback bits in the extent tree for this IO
2159 * end_page_writeback if the page has no more pending IO
2161 * Scheduling is not allowed, so the extent state tree is expected
2162 * to have one and only one object corresponding to this IO.
2164 static void end_bio_extent_writepage(struct bio *bio, int err)
2166 int uptodate = err == 0;
2167 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2168 struct extent_io_tree *tree;
2175 struct page *page = bvec->bv_page;
2176 tree = &BTRFS_I(page->mapping->host)->io_tree;
2178 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2180 end = start + bvec->bv_len - 1;
2182 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2187 if (--bvec >= bio->bi_io_vec)
2188 prefetchw(&bvec->bv_page->flags);
2189 if (tree->ops && tree->ops->writepage_end_io_hook) {
2190 ret = tree->ops->writepage_end_io_hook(page, start,
2191 end, NULL, uptodate);
2196 if (!uptodate && tree->ops &&
2197 tree->ops->writepage_io_failed_hook) {
2198 ret = tree->ops->writepage_io_failed_hook(bio, page,
2201 uptodate = (err == 0);
2207 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
2208 ClearPageUptodate(page);
2213 end_page_writeback(page);
2215 check_page_writeback(tree, page);
2216 } while (bvec >= bio->bi_io_vec);
2222 * after a readpage IO is done, we need to:
2223 * clear the uptodate bits on error
2224 * set the uptodate bits if things worked
2225 * set the page up to date if all extents in the tree are uptodate
2226 * clear the lock bit in the extent tree
2227 * unlock the page if there are no other extents locked for it
2229 * Scheduling is not allowed, so the extent state tree is expected
2230 * to have one and only one object corresponding to this IO.
2232 static void end_bio_extent_readpage(struct bio *bio, int err)
2234 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2235 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2236 struct bio_vec *bvec = bio->bi_io_vec;
2237 struct extent_io_tree *tree;
2247 struct page *page = bvec->bv_page;
2248 struct extent_state *cached = NULL;
2249 struct extent_state *state;
2251 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2252 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2253 (long int)bio->bi_bdev);
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 <= bvec_end)
2266 prefetchw(&bvec->bv_page->flags);
2268 spin_lock(&tree->lock);
2269 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2270 if (state && state->start == start) {
2272 * take a reference on the state, unlock will drop
2275 cache_state(state, &cached);
2277 spin_unlock(&tree->lock);
2279 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2280 ret = tree->ops->readpage_end_io_hook(page, start, end,
2285 clean_io_failure(start, page);
2289 failed_mirror = (int)(unsigned long)bio->bi_bdev;
2291 * The generic bio_readpage_error handles errors the
2292 * following way: If possible, new read requests are
2293 * created and submitted and will end up in
2294 * end_bio_extent_readpage as well (if we're lucky, not
2295 * in the !uptodate case). In that case it returns 0 and
2296 * we just go on with the next page in our bio. If it
2297 * can't handle the error it will return -EIO and we
2298 * remain responsible for that page.
2300 ret = bio_readpage_error(bio, page, start, end,
2301 failed_mirror, NULL);
2305 test_bit(BIO_UPTODATE, &bio->bi_flags);
2308 uncache_state(&cached);
2311 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2312 ret = tree->ops->readpage_io_failed_hook(
2313 bio, page, start, end,
2314 failed_mirror, state);
2321 set_extent_uptodate(tree, start, end, &cached,
2324 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2328 SetPageUptodate(page);
2330 ClearPageUptodate(page);
2336 check_page_uptodate(tree, page);
2338 ClearPageUptodate(page);
2341 check_page_locked(tree, page);
2343 } while (bvec <= bvec_end);
2349 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2354 bio = bio_alloc(gfp_flags, nr_vecs);
2356 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2357 while (!bio && (nr_vecs /= 2))
2358 bio = bio_alloc(gfp_flags, nr_vecs);
2363 bio->bi_bdev = bdev;
2364 bio->bi_sector = first_sector;
2369 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
2370 unsigned long bio_flags)
2373 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2374 struct page *page = bvec->bv_page;
2375 struct extent_io_tree *tree = bio->bi_private;
2378 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2380 bio->bi_private = NULL;
2384 if (tree->ops && tree->ops->submit_bio_hook)
2385 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2386 mirror_num, bio_flags, start);
2388 submit_bio(rw, bio);
2390 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2396 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2397 struct page *page, sector_t sector,
2398 size_t size, unsigned long offset,
2399 struct block_device *bdev,
2400 struct bio **bio_ret,
2401 unsigned long max_pages,
2402 bio_end_io_t end_io_func,
2404 unsigned long prev_bio_flags,
2405 unsigned long bio_flags)
2411 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2412 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2413 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2415 if (bio_ret && *bio_ret) {
2418 contig = bio->bi_sector == sector;
2420 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2423 if (prev_bio_flags != bio_flags || !contig ||
2424 (tree->ops && tree->ops->merge_bio_hook &&
2425 tree->ops->merge_bio_hook(page, offset, page_size, bio,
2427 bio_add_page(bio, page, page_size, offset) < page_size) {
2428 ret = submit_one_bio(rw, bio, mirror_num,
2435 if (this_compressed)
2438 nr = bio_get_nr_vecs(bdev);
2440 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2444 bio_add_page(bio, page, page_size, offset);
2445 bio->bi_end_io = end_io_func;
2446 bio->bi_private = tree;
2451 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2456 void set_page_extent_mapped(struct page *page)
2458 if (!PagePrivate(page)) {
2459 SetPagePrivate(page);
2460 page_cache_get(page);
2461 set_page_private(page, EXTENT_PAGE_PRIVATE);
2465 static void set_page_extent_head(struct page *page, unsigned long len)
2467 WARN_ON(!PagePrivate(page));
2468 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
2472 * basic readpage implementation. Locked extent state structs are inserted
2473 * into the tree that are removed when the IO is done (by the end_io
2476 static int __extent_read_full_page(struct extent_io_tree *tree,
2478 get_extent_t *get_extent,
2479 struct bio **bio, int mirror_num,
2480 unsigned long *bio_flags)
2482 struct inode *inode = page->mapping->host;
2483 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2484 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2488 u64 last_byte = i_size_read(inode);
2492 struct extent_map *em;
2493 struct block_device *bdev;
2494 struct btrfs_ordered_extent *ordered;
2497 size_t pg_offset = 0;
2499 size_t disk_io_size;
2500 size_t blocksize = inode->i_sb->s_blocksize;
2501 unsigned long this_bio_flag = 0;
2503 set_page_extent_mapped(page);
2505 if (!PageUptodate(page)) {
2506 if (cleancache_get_page(page) == 0) {
2507 BUG_ON(blocksize != PAGE_SIZE);
2514 lock_extent(tree, start, end, GFP_NOFS);
2515 ordered = btrfs_lookup_ordered_extent(inode, start);
2518 unlock_extent(tree, start, end, GFP_NOFS);
2519 btrfs_start_ordered_extent(inode, ordered, 1);
2520 btrfs_put_ordered_extent(ordered);
2523 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2525 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2528 iosize = PAGE_CACHE_SIZE - zero_offset;
2529 userpage = kmap_atomic(page, KM_USER0);
2530 memset(userpage + zero_offset, 0, iosize);
2531 flush_dcache_page(page);
2532 kunmap_atomic(userpage, KM_USER0);
2535 while (cur <= end) {
2536 if (cur >= last_byte) {
2538 struct extent_state *cached = NULL;
2540 iosize = PAGE_CACHE_SIZE - pg_offset;
2541 userpage = kmap_atomic(page, KM_USER0);
2542 memset(userpage + pg_offset, 0, iosize);
2543 flush_dcache_page(page);
2544 kunmap_atomic(userpage, KM_USER0);
2545 set_extent_uptodate(tree, cur, cur + iosize - 1,
2547 unlock_extent_cached(tree, cur, cur + iosize - 1,
2551 em = get_extent(inode, page, pg_offset, cur,
2553 if (IS_ERR_OR_NULL(em)) {
2555 unlock_extent(tree, cur, end, GFP_NOFS);
2558 extent_offset = cur - em->start;
2559 BUG_ON(extent_map_end(em) <= cur);
2562 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2563 this_bio_flag = EXTENT_BIO_COMPRESSED;
2564 extent_set_compress_type(&this_bio_flag,
2568 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2569 cur_end = min(extent_map_end(em) - 1, end);
2570 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2571 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2572 disk_io_size = em->block_len;
2573 sector = em->block_start >> 9;
2575 sector = (em->block_start + extent_offset) >> 9;
2576 disk_io_size = iosize;
2579 block_start = em->block_start;
2580 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2581 block_start = EXTENT_MAP_HOLE;
2582 free_extent_map(em);
2585 /* we've found a hole, just zero and go on */
2586 if (block_start == EXTENT_MAP_HOLE) {
2588 struct extent_state *cached = NULL;
2590 userpage = kmap_atomic(page, KM_USER0);
2591 memset(userpage + pg_offset, 0, iosize);
2592 flush_dcache_page(page);
2593 kunmap_atomic(userpage, KM_USER0);
2595 set_extent_uptodate(tree, cur, cur + iosize - 1,
2597 unlock_extent_cached(tree, cur, cur + iosize - 1,
2600 pg_offset += iosize;
2603 /* the get_extent function already copied into the page */
2604 if (test_range_bit(tree, cur, cur_end,
2605 EXTENT_UPTODATE, 1, NULL)) {
2606 check_page_uptodate(tree, page);
2607 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2609 pg_offset += iosize;
2612 /* we have an inline extent but it didn't get marked up
2613 * to date. Error out
2615 if (block_start == EXTENT_MAP_INLINE) {
2617 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2619 pg_offset += iosize;
2624 if (tree->ops && tree->ops->readpage_io_hook) {
2625 ret = tree->ops->readpage_io_hook(page, cur,
2629 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2631 ret = submit_extent_page(READ, tree, page,
2632 sector, disk_io_size, pg_offset,
2634 end_bio_extent_readpage, mirror_num,
2638 *bio_flags = this_bio_flag;
2643 pg_offset += iosize;
2647 if (!PageError(page))
2648 SetPageUptodate(page);
2654 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2655 get_extent_t *get_extent, int mirror_num)
2657 struct bio *bio = NULL;
2658 unsigned long bio_flags = 0;
2661 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2664 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2668 static noinline void update_nr_written(struct page *page,
2669 struct writeback_control *wbc,
2670 unsigned long nr_written)
2672 wbc->nr_to_write -= nr_written;
2673 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2674 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2675 page->mapping->writeback_index = page->index + nr_written;
2679 * the writepage semantics are similar to regular writepage. extent
2680 * records are inserted to lock ranges in the tree, and as dirty areas
2681 * are found, they are marked writeback. Then the lock bits are removed
2682 * and the end_io handler clears the writeback ranges
2684 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2687 struct inode *inode = page->mapping->host;
2688 struct extent_page_data *epd = data;
2689 struct extent_io_tree *tree = epd->tree;
2690 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2692 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2696 u64 last_byte = i_size_read(inode);
2700 struct extent_state *cached_state = NULL;
2701 struct extent_map *em;
2702 struct block_device *bdev;
2705 size_t pg_offset = 0;
2707 loff_t i_size = i_size_read(inode);
2708 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2714 unsigned long nr_written = 0;
2715 bool fill_delalloc = true;
2717 if (wbc->sync_mode == WB_SYNC_ALL)
2718 write_flags = WRITE_SYNC;
2720 write_flags = WRITE;
2722 trace___extent_writepage(page, inode, wbc);
2724 WARN_ON(!PageLocked(page));
2726 ClearPageError(page);
2728 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2729 if (page->index > end_index ||
2730 (page->index == end_index && !pg_offset)) {
2731 page->mapping->a_ops->invalidatepage(page, 0);
2736 if (page->index == end_index) {
2739 userpage = kmap_atomic(page, KM_USER0);
2740 memset(userpage + pg_offset, 0,
2741 PAGE_CACHE_SIZE - pg_offset);
2742 kunmap_atomic(userpage, KM_USER0);
2743 flush_dcache_page(page);
2747 set_page_extent_mapped(page);
2749 if (!tree->ops || !tree->ops->fill_delalloc)
2750 fill_delalloc = false;
2752 delalloc_start = start;
2755 if (!epd->extent_locked && fill_delalloc) {
2756 u64 delalloc_to_write = 0;
2758 * make sure the wbc mapping index is at least updated
2761 update_nr_written(page, wbc, 0);
2763 while (delalloc_end < page_end) {
2764 nr_delalloc = find_lock_delalloc_range(inode, tree,
2769 if (nr_delalloc == 0) {
2770 delalloc_start = delalloc_end + 1;
2773 tree->ops->fill_delalloc(inode, page, delalloc_start,
2774 delalloc_end, &page_started,
2777 * delalloc_end is already one less than the total
2778 * length, so we don't subtract one from
2781 delalloc_to_write += (delalloc_end - delalloc_start +
2784 delalloc_start = delalloc_end + 1;
2786 if (wbc->nr_to_write < delalloc_to_write) {
2789 if (delalloc_to_write < thresh * 2)
2790 thresh = delalloc_to_write;
2791 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2795 /* did the fill delalloc function already unlock and start
2801 * we've unlocked the page, so we can't update
2802 * the mapping's writeback index, just update
2805 wbc->nr_to_write -= nr_written;
2809 if (tree->ops && tree->ops->writepage_start_hook) {
2810 ret = tree->ops->writepage_start_hook(page, start,
2812 if (ret == -EAGAIN) {
2813 redirty_page_for_writepage(wbc, page);
2814 update_nr_written(page, wbc, nr_written);
2822 * we don't want to touch the inode after unlocking the page,
2823 * so we update the mapping writeback index now
2825 update_nr_written(page, wbc, nr_written + 1);
2828 if (last_byte <= start) {
2829 if (tree->ops && tree->ops->writepage_end_io_hook)
2830 tree->ops->writepage_end_io_hook(page, start,
2835 blocksize = inode->i_sb->s_blocksize;
2837 while (cur <= end) {
2838 if (cur >= last_byte) {
2839 if (tree->ops && tree->ops->writepage_end_io_hook)
2840 tree->ops->writepage_end_io_hook(page, cur,
2844 em = epd->get_extent(inode, page, pg_offset, cur,
2846 if (IS_ERR_OR_NULL(em)) {
2851 extent_offset = cur - em->start;
2852 BUG_ON(extent_map_end(em) <= cur);
2854 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2855 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2856 sector = (em->block_start + extent_offset) >> 9;
2858 block_start = em->block_start;
2859 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2860 free_extent_map(em);
2864 * compressed and inline extents are written through other
2867 if (compressed || block_start == EXTENT_MAP_HOLE ||
2868 block_start == EXTENT_MAP_INLINE) {
2870 * end_io notification does not happen here for
2871 * compressed extents
2873 if (!compressed && tree->ops &&
2874 tree->ops->writepage_end_io_hook)
2875 tree->ops->writepage_end_io_hook(page, cur,
2878 else if (compressed) {
2879 /* we don't want to end_page_writeback on
2880 * a compressed extent. this happens
2887 pg_offset += iosize;
2890 /* leave this out until we have a page_mkwrite call */
2891 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2892 EXTENT_DIRTY, 0, NULL)) {
2894 pg_offset += iosize;
2898 if (tree->ops && tree->ops->writepage_io_hook) {
2899 ret = tree->ops->writepage_io_hook(page, cur,
2907 unsigned long max_nr = end_index + 1;
2909 set_range_writeback(tree, cur, cur + iosize - 1);
2910 if (!PageWriteback(page)) {
2911 printk(KERN_ERR "btrfs warning page %lu not "
2912 "writeback, cur %llu end %llu\n",
2913 page->index, (unsigned long long)cur,
2914 (unsigned long long)end);
2917 ret = submit_extent_page(write_flags, tree, page,
2918 sector, iosize, pg_offset,
2919 bdev, &epd->bio, max_nr,
2920 end_bio_extent_writepage,
2926 pg_offset += iosize;
2931 /* make sure the mapping tag for page dirty gets cleared */
2932 set_page_writeback(page);
2933 end_page_writeback(page);
2939 /* drop our reference on any cached states */
2940 free_extent_state(cached_state);
2945 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2946 * @mapping: address space structure to write
2947 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2948 * @writepage: function called for each page
2949 * @data: data passed to writepage function
2951 * If a page is already under I/O, write_cache_pages() skips it, even
2952 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2953 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2954 * and msync() need to guarantee that all the data which was dirty at the time
2955 * the call was made get new I/O started against them. If wbc->sync_mode is
2956 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2957 * existing IO to complete.
2959 static int extent_write_cache_pages(struct extent_io_tree *tree,
2960 struct address_space *mapping,
2961 struct writeback_control *wbc,
2962 writepage_t writepage, void *data,
2963 void (*flush_fn)(void *))
2967 int nr_to_write_done = 0;
2968 struct pagevec pvec;
2971 pgoff_t end; /* Inclusive */
2975 pagevec_init(&pvec, 0);
2976 if (wbc->range_cyclic) {
2977 index = mapping->writeback_index; /* Start from prev offset */
2980 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2981 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2984 if (wbc->sync_mode == WB_SYNC_ALL)
2985 tag = PAGECACHE_TAG_TOWRITE;
2987 tag = PAGECACHE_TAG_DIRTY;
2989 if (wbc->sync_mode == WB_SYNC_ALL)
2990 tag_pages_for_writeback(mapping, index, end);
2991 while (!done && !nr_to_write_done && (index <= end) &&
2992 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2993 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2997 for (i = 0; i < nr_pages; i++) {
2998 struct page *page = pvec.pages[i];
3001 * At this point we hold neither mapping->tree_lock nor
3002 * lock on the page itself: the page may be truncated or
3003 * invalidated (changing page->mapping to NULL), or even
3004 * swizzled back from swapper_space to tmpfs file
3008 tree->ops->write_cache_pages_lock_hook) {
3009 tree->ops->write_cache_pages_lock_hook(page,
3012 if (!trylock_page(page)) {
3018 if (unlikely(page->mapping != mapping)) {
3023 if (!wbc->range_cyclic && page->index > end) {
3029 if (wbc->sync_mode != WB_SYNC_NONE) {
3030 if (PageWriteback(page))
3032 wait_on_page_writeback(page);
3035 if (PageWriteback(page) ||
3036 !clear_page_dirty_for_io(page)) {
3041 ret = (*writepage)(page, wbc, data);
3043 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3051 * the filesystem may choose to bump up nr_to_write.
3052 * We have to make sure to honor the new nr_to_write
3055 nr_to_write_done = wbc->nr_to_write <= 0;
3057 pagevec_release(&pvec);
3060 if (!scanned && !done) {
3062 * We hit the last page and there is more work to be done: wrap
3063 * back to the start of the file
3072 static void flush_epd_write_bio(struct extent_page_data *epd)
3076 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
3078 submit_one_bio(WRITE, epd->bio, 0, 0);
3083 static noinline void flush_write_bio(void *data)
3085 struct extent_page_data *epd = data;
3086 flush_epd_write_bio(epd);
3089 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3090 get_extent_t *get_extent,
3091 struct writeback_control *wbc)
3094 struct extent_page_data epd = {
3097 .get_extent = get_extent,
3099 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3102 ret = __extent_writepage(page, wbc, &epd);
3104 flush_epd_write_bio(&epd);
3108 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3109 u64 start, u64 end, get_extent_t *get_extent,
3113 struct address_space *mapping = inode->i_mapping;
3115 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3118 struct extent_page_data epd = {
3121 .get_extent = get_extent,
3123 .sync_io = mode == WB_SYNC_ALL,
3125 struct writeback_control wbc_writepages = {
3127 .nr_to_write = nr_pages * 2,
3128 .range_start = start,
3129 .range_end = end + 1,
3132 while (start <= end) {
3133 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3134 if (clear_page_dirty_for_io(page))
3135 ret = __extent_writepage(page, &wbc_writepages, &epd);
3137 if (tree->ops && tree->ops->writepage_end_io_hook)
3138 tree->ops->writepage_end_io_hook(page, start,
3139 start + PAGE_CACHE_SIZE - 1,
3143 page_cache_release(page);
3144 start += PAGE_CACHE_SIZE;
3147 flush_epd_write_bio(&epd);
3151 int extent_writepages(struct extent_io_tree *tree,
3152 struct address_space *mapping,
3153 get_extent_t *get_extent,
3154 struct writeback_control *wbc)
3157 struct extent_page_data epd = {
3160 .get_extent = get_extent,
3162 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3165 ret = extent_write_cache_pages(tree, mapping, wbc,
3166 __extent_writepage, &epd,
3168 flush_epd_write_bio(&epd);
3172 int extent_readpages(struct extent_io_tree *tree,
3173 struct address_space *mapping,
3174 struct list_head *pages, unsigned nr_pages,
3175 get_extent_t get_extent)
3177 struct bio *bio = NULL;
3179 unsigned long bio_flags = 0;
3181 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3182 struct page *page = list_entry(pages->prev, struct page, lru);
3184 prefetchw(&page->flags);
3185 list_del(&page->lru);
3186 if (!add_to_page_cache_lru(page, mapping,
3187 page->index, GFP_NOFS)) {
3188 __extent_read_full_page(tree, page, get_extent,
3189 &bio, 0, &bio_flags);
3191 page_cache_release(page);
3193 BUG_ON(!list_empty(pages));
3195 submit_one_bio(READ, bio, 0, bio_flags);
3200 * basic invalidatepage code, this waits on any locked or writeback
3201 * ranges corresponding to the page, and then deletes any extent state
3202 * records from the tree
3204 int extent_invalidatepage(struct extent_io_tree *tree,
3205 struct page *page, unsigned long offset)
3207 struct extent_state *cached_state = NULL;
3208 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3209 u64 end = start + PAGE_CACHE_SIZE - 1;
3210 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3212 start += (offset + blocksize - 1) & ~(blocksize - 1);
3216 lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS);
3217 wait_on_page_writeback(page);
3218 clear_extent_bit(tree, start, end,
3219 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3220 EXTENT_DO_ACCOUNTING,
3221 1, 1, &cached_state, GFP_NOFS);
3226 * a helper for releasepage, this tests for areas of the page that
3227 * are locked or under IO and drops the related state bits if it is safe
3230 int try_release_extent_state(struct extent_map_tree *map,
3231 struct extent_io_tree *tree, struct page *page,
3234 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3235 u64 end = start + PAGE_CACHE_SIZE - 1;
3238 if (test_range_bit(tree, start, end,
3239 EXTENT_IOBITS, 0, NULL))
3242 if ((mask & GFP_NOFS) == GFP_NOFS)
3245 * at this point we can safely clear everything except the
3246 * locked bit and the nodatasum bit
3248 ret = clear_extent_bit(tree, start, end,
3249 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3252 /* if clear_extent_bit failed for enomem reasons,
3253 * we can't allow the release to continue.
3264 * a helper for releasepage. As long as there are no locked extents
3265 * in the range corresponding to the page, both state records and extent
3266 * map records are removed
3268 int try_release_extent_mapping(struct extent_map_tree *map,
3269 struct extent_io_tree *tree, struct page *page,
3272 struct extent_map *em;
3273 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3274 u64 end = start + PAGE_CACHE_SIZE - 1;
3276 if ((mask & __GFP_WAIT) &&
3277 page->mapping->host->i_size > 16 * 1024 * 1024) {
3279 while (start <= end) {
3280 len = end - start + 1;
3281 write_lock(&map->lock);
3282 em = lookup_extent_mapping(map, start, len);
3283 if (IS_ERR_OR_NULL(em)) {
3284 write_unlock(&map->lock);
3287 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3288 em->start != start) {
3289 write_unlock(&map->lock);
3290 free_extent_map(em);
3293 if (!test_range_bit(tree, em->start,
3294 extent_map_end(em) - 1,
3295 EXTENT_LOCKED | EXTENT_WRITEBACK,
3297 remove_extent_mapping(map, em);
3298 /* once for the rb tree */
3299 free_extent_map(em);
3301 start = extent_map_end(em);
3302 write_unlock(&map->lock);
3305 free_extent_map(em);
3308 return try_release_extent_state(map, tree, page, mask);
3312 * helper function for fiemap, which doesn't want to see any holes.
3313 * This maps until we find something past 'last'
3315 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3318 get_extent_t *get_extent)
3320 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3321 struct extent_map *em;
3328 len = last - offset;
3331 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3332 em = get_extent(inode, NULL, 0, offset, len, 0);
3333 if (IS_ERR_OR_NULL(em))
3336 /* if this isn't a hole return it */
3337 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3338 em->block_start != EXTENT_MAP_HOLE) {
3342 /* this is a hole, advance to the next extent */
3343 offset = extent_map_end(em);
3344 free_extent_map(em);
3351 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3352 __u64 start, __u64 len, get_extent_t *get_extent)
3356 u64 max = start + len;
3360 u64 last_for_get_extent = 0;
3362 u64 isize = i_size_read(inode);
3363 struct btrfs_key found_key;
3364 struct extent_map *em = NULL;
3365 struct extent_state *cached_state = NULL;
3366 struct btrfs_path *path;
3367 struct btrfs_file_extent_item *item;
3372 unsigned long emflags;
3377 path = btrfs_alloc_path();
3380 path->leave_spinning = 1;
3382 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3383 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3386 * lookup the last file extent. We're not using i_size here
3387 * because there might be preallocation past i_size
3389 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3390 path, btrfs_ino(inode), -1, 0);
3392 btrfs_free_path(path);
3397 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3398 struct btrfs_file_extent_item);
3399 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3400 found_type = btrfs_key_type(&found_key);
3402 /* No extents, but there might be delalloc bits */
3403 if (found_key.objectid != btrfs_ino(inode) ||
3404 found_type != BTRFS_EXTENT_DATA_KEY) {
3405 /* have to trust i_size as the end */
3407 last_for_get_extent = isize;
3410 * remember the start of the last extent. There are a
3411 * bunch of different factors that go into the length of the
3412 * extent, so its much less complex to remember where it started
3414 last = found_key.offset;
3415 last_for_get_extent = last + 1;
3417 btrfs_free_path(path);
3420 * we might have some extents allocated but more delalloc past those
3421 * extents. so, we trust isize unless the start of the last extent is
3426 last_for_get_extent = isize;
3429 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3430 &cached_state, GFP_NOFS);
3432 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3442 u64 offset_in_extent;
3444 /* break if the extent we found is outside the range */
3445 if (em->start >= max || extent_map_end(em) < off)
3449 * get_extent may return an extent that starts before our
3450 * requested range. We have to make sure the ranges
3451 * we return to fiemap always move forward and don't
3452 * overlap, so adjust the offsets here
3454 em_start = max(em->start, off);
3457 * record the offset from the start of the extent
3458 * for adjusting the disk offset below
3460 offset_in_extent = em_start - em->start;
3461 em_end = extent_map_end(em);
3462 em_len = em_end - em_start;
3463 emflags = em->flags;
3468 * bump off for our next call to get_extent
3470 off = extent_map_end(em);
3474 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3476 flags |= FIEMAP_EXTENT_LAST;
3477 } else if (em->block_start == EXTENT_MAP_INLINE) {
3478 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3479 FIEMAP_EXTENT_NOT_ALIGNED);
3480 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3481 flags |= (FIEMAP_EXTENT_DELALLOC |
3482 FIEMAP_EXTENT_UNKNOWN);
3484 disko = em->block_start + offset_in_extent;
3486 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3487 flags |= FIEMAP_EXTENT_ENCODED;
3489 free_extent_map(em);
3491 if ((em_start >= last) || em_len == (u64)-1 ||
3492 (last == (u64)-1 && isize <= em_end)) {
3493 flags |= FIEMAP_EXTENT_LAST;
3497 /* now scan forward to see if this is really the last extent. */
3498 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3505 flags |= FIEMAP_EXTENT_LAST;
3508 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3514 free_extent_map(em);
3516 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3517 &cached_state, GFP_NOFS);
3521 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3525 struct address_space *mapping;
3528 return eb->first_page;
3529 i += eb->start >> PAGE_CACHE_SHIFT;
3530 mapping = eb->first_page->mapping;
3535 * extent_buffer_page is only called after pinning the page
3536 * by increasing the reference count. So we know the page must
3537 * be in the radix tree.
3540 p = radix_tree_lookup(&mapping->page_tree, i);
3546 inline unsigned long num_extent_pages(u64 start, u64 len)
3548 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3549 (start >> PAGE_CACHE_SHIFT);
3552 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3557 struct extent_buffer *eb = NULL;
3559 unsigned long flags;
3562 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3567 rwlock_init(&eb->lock);
3568 atomic_set(&eb->write_locks, 0);
3569 atomic_set(&eb->read_locks, 0);
3570 atomic_set(&eb->blocking_readers, 0);
3571 atomic_set(&eb->blocking_writers, 0);
3572 atomic_set(&eb->spinning_readers, 0);
3573 atomic_set(&eb->spinning_writers, 0);
3574 init_waitqueue_head(&eb->write_lock_wq);
3575 init_waitqueue_head(&eb->read_lock_wq);
3578 spin_lock_irqsave(&leak_lock, flags);
3579 list_add(&eb->leak_list, &buffers);
3580 spin_unlock_irqrestore(&leak_lock, flags);
3582 atomic_set(&eb->refs, 1);
3587 static void __free_extent_buffer(struct extent_buffer *eb)
3590 unsigned long flags;
3591 spin_lock_irqsave(&leak_lock, flags);
3592 list_del(&eb->leak_list);
3593 spin_unlock_irqrestore(&leak_lock, flags);
3595 kmem_cache_free(extent_buffer_cache, eb);
3599 * Helper for releasing extent buffer page.
3601 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3602 unsigned long start_idx)
3604 unsigned long index;
3607 if (!eb->first_page)
3610 index = num_extent_pages(eb->start, eb->len);
3611 if (start_idx >= index)
3616 page = extent_buffer_page(eb, index);
3618 page_cache_release(page);
3619 } while (index != start_idx);
3623 * Helper for releasing the extent buffer.
3625 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3627 btrfs_release_extent_buffer_page(eb, 0);
3628 __free_extent_buffer(eb);
3631 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3632 u64 start, unsigned long len,
3635 unsigned long num_pages = num_extent_pages(start, len);
3637 unsigned long index = start >> PAGE_CACHE_SHIFT;
3638 struct extent_buffer *eb;
3639 struct extent_buffer *exists = NULL;
3641 struct address_space *mapping = tree->mapping;
3646 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3647 if (eb && atomic_inc_not_zero(&eb->refs)) {
3649 mark_page_accessed(eb->first_page);
3654 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
3659 eb->first_page = page0;
3662 page_cache_get(page0);
3663 mark_page_accessed(page0);
3664 set_page_extent_mapped(page0);
3665 set_page_extent_head(page0, len);
3666 uptodate = PageUptodate(page0);
3670 for (; i < num_pages; i++, index++) {
3671 p = find_or_create_page(mapping, index, GFP_NOFS);
3676 set_page_extent_mapped(p);
3677 mark_page_accessed(p);
3680 set_page_extent_head(p, len);
3682 set_page_private(p, EXTENT_PAGE_PRIVATE);
3684 if (!PageUptodate(p))
3688 * see below about how we avoid a nasty race with release page
3689 * and why we unlock later
3695 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3697 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3701 spin_lock(&tree->buffer_lock);
3702 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3703 if (ret == -EEXIST) {
3704 exists = radix_tree_lookup(&tree->buffer,
3705 start >> PAGE_CACHE_SHIFT);
3706 /* add one reference for the caller */
3707 atomic_inc(&exists->refs);
3708 spin_unlock(&tree->buffer_lock);
3709 radix_tree_preload_end();
3712 /* add one reference for the tree */
3713 atomic_inc(&eb->refs);
3714 spin_unlock(&tree->buffer_lock);
3715 radix_tree_preload_end();
3718 * there is a race where release page may have
3719 * tried to find this extent buffer in the radix
3720 * but failed. It will tell the VM it is safe to
3721 * reclaim the, and it will clear the page private bit.
3722 * We must make sure to set the page private bit properly
3723 * after the extent buffer is in the radix tree so
3724 * it doesn't get lost
3726 set_page_extent_mapped(eb->first_page);
3727 set_page_extent_head(eb->first_page, eb->len);
3729 unlock_page(eb->first_page);
3733 if (eb->first_page && !page0)
3734 unlock_page(eb->first_page);
3736 if (!atomic_dec_and_test(&eb->refs))
3738 btrfs_release_extent_buffer(eb);
3742 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3743 u64 start, unsigned long len)
3745 struct extent_buffer *eb;
3748 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3749 if (eb && atomic_inc_not_zero(&eb->refs)) {
3751 mark_page_accessed(eb->first_page);
3759 void free_extent_buffer(struct extent_buffer *eb)
3764 if (!atomic_dec_and_test(&eb->refs))
3770 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3771 struct extent_buffer *eb)
3774 unsigned long num_pages;
3777 num_pages = num_extent_pages(eb->start, eb->len);
3779 for (i = 0; i < num_pages; i++) {
3780 page = extent_buffer_page(eb, i);
3781 if (!PageDirty(page))
3785 WARN_ON(!PagePrivate(page));
3787 set_page_extent_mapped(page);
3789 set_page_extent_head(page, eb->len);
3791 clear_page_dirty_for_io(page);
3792 spin_lock_irq(&page->mapping->tree_lock);
3793 if (!PageDirty(page)) {
3794 radix_tree_tag_clear(&page->mapping->page_tree,
3796 PAGECACHE_TAG_DIRTY);
3798 spin_unlock_irq(&page->mapping->tree_lock);
3799 ClearPageError(page);
3805 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3806 struct extent_buffer *eb)
3809 unsigned long num_pages;
3812 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3813 num_pages = num_extent_pages(eb->start, eb->len);
3814 for (i = 0; i < num_pages; i++)
3815 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3819 static int __eb_straddles_pages(u64 start, u64 len)
3821 if (len < PAGE_CACHE_SIZE)
3823 if (start & (PAGE_CACHE_SIZE - 1))
3825 if ((start + len) & (PAGE_CACHE_SIZE - 1))
3830 static int eb_straddles_pages(struct extent_buffer *eb)
3832 return __eb_straddles_pages(eb->start, eb->len);
3835 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3836 struct extent_buffer *eb,
3837 struct extent_state **cached_state)
3841 unsigned long num_pages;
3843 num_pages = num_extent_pages(eb->start, eb->len);
3844 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3846 if (eb_straddles_pages(eb)) {
3847 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3848 cached_state, GFP_NOFS);
3850 for (i = 0; i < num_pages; i++) {
3851 page = extent_buffer_page(eb, i);
3853 ClearPageUptodate(page);
3858 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3859 struct extent_buffer *eb)
3863 unsigned long num_pages;
3865 num_pages = num_extent_pages(eb->start, eb->len);
3867 if (eb_straddles_pages(eb)) {
3868 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3871 for (i = 0; i < num_pages; i++) {
3872 page = extent_buffer_page(eb, i);
3873 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3874 ((i == num_pages - 1) &&
3875 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3876 check_page_uptodate(tree, page);
3879 SetPageUptodate(page);
3884 int extent_range_uptodate(struct extent_io_tree *tree,
3889 int pg_uptodate = 1;
3891 unsigned long index;
3893 if (__eb_straddles_pages(start, end - start + 1)) {
3894 ret = test_range_bit(tree, start, end,
3895 EXTENT_UPTODATE, 1, NULL);
3899 while (start <= end) {
3900 index = start >> PAGE_CACHE_SHIFT;
3901 page = find_get_page(tree->mapping, index);
3902 uptodate = PageUptodate(page);
3903 page_cache_release(page);
3908 start += PAGE_CACHE_SIZE;
3913 int extent_buffer_uptodate(struct extent_io_tree *tree,
3914 struct extent_buffer *eb,
3915 struct extent_state *cached_state)
3918 unsigned long num_pages;
3921 int pg_uptodate = 1;
3923 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3926 if (eb_straddles_pages(eb)) {
3927 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3928 EXTENT_UPTODATE, 1, cached_state);
3933 num_pages = num_extent_pages(eb->start, eb->len);
3934 for (i = 0; i < num_pages; i++) {
3935 page = extent_buffer_page(eb, i);
3936 if (!PageUptodate(page)) {
3944 int read_extent_buffer_pages(struct extent_io_tree *tree,
3945 struct extent_buffer *eb, u64 start, int wait,
3946 get_extent_t *get_extent, int mirror_num)
3949 unsigned long start_i;
3953 int locked_pages = 0;
3954 int all_uptodate = 1;
3955 int inc_all_pages = 0;
3956 unsigned long num_pages;
3957 struct bio *bio = NULL;
3958 unsigned long bio_flags = 0;
3960 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3963 if (eb_straddles_pages(eb)) {
3964 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3965 EXTENT_UPTODATE, 1, NULL)) {
3971 WARN_ON(start < eb->start);
3972 start_i = (start >> PAGE_CACHE_SHIFT) -
3973 (eb->start >> PAGE_CACHE_SHIFT);
3978 num_pages = num_extent_pages(eb->start, eb->len);
3979 for (i = start_i; i < num_pages; i++) {
3980 page = extent_buffer_page(eb, i);
3981 if (wait == WAIT_NONE) {
3982 if (!trylock_page(page))
3988 if (!PageUptodate(page))
3993 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3997 for (i = start_i; i < num_pages; i++) {
3998 page = extent_buffer_page(eb, i);
4000 WARN_ON(!PagePrivate(page));
4002 set_page_extent_mapped(page);
4004 set_page_extent_head(page, eb->len);
4007 page_cache_get(page);
4008 if (!PageUptodate(page)) {
4011 ClearPageError(page);
4012 err = __extent_read_full_page(tree, page,
4014 mirror_num, &bio_flags);
4023 submit_one_bio(READ, bio, mirror_num, bio_flags);
4025 if (ret || wait != WAIT_COMPLETE)
4028 for (i = start_i; i < num_pages; i++) {
4029 page = extent_buffer_page(eb, i);
4030 wait_on_page_locked(page);
4031 if (!PageUptodate(page))
4036 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4041 while (locked_pages > 0) {
4042 page = extent_buffer_page(eb, i);
4050 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4051 unsigned long start,
4058 char *dst = (char *)dstv;
4059 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4060 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4062 WARN_ON(start > eb->len);
4063 WARN_ON(start + len > eb->start + eb->len);
4065 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4068 page = extent_buffer_page(eb, i);
4070 cur = min(len, (PAGE_CACHE_SIZE - offset));
4071 kaddr = page_address(page);
4072 memcpy(dst, kaddr + offset, cur);
4081 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4082 unsigned long min_len, char **map,
4083 unsigned long *map_start,
4084 unsigned long *map_len)
4086 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4089 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4090 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4091 unsigned long end_i = (start_offset + start + min_len - 1) >>
4098 offset = start_offset;
4102 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4105 if (start + min_len > eb->len) {
4106 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4107 "wanted %lu %lu\n", (unsigned long long)eb->start,
4108 eb->len, start, min_len);
4113 p = extent_buffer_page(eb, i);
4114 kaddr = page_address(p);
4115 *map = kaddr + offset;
4116 *map_len = PAGE_CACHE_SIZE - offset;
4120 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4121 unsigned long start,
4128 char *ptr = (char *)ptrv;
4129 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4130 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4133 WARN_ON(start > eb->len);
4134 WARN_ON(start + len > eb->start + eb->len);
4136 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4139 page = extent_buffer_page(eb, i);
4141 cur = min(len, (PAGE_CACHE_SIZE - offset));
4143 kaddr = page_address(page);
4144 ret = memcmp(ptr, kaddr + offset, cur);
4156 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4157 unsigned long start, unsigned long len)
4163 char *src = (char *)srcv;
4164 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4165 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4167 WARN_ON(start > eb->len);
4168 WARN_ON(start + len > eb->start + eb->len);
4170 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4173 page = extent_buffer_page(eb, i);
4174 WARN_ON(!PageUptodate(page));
4176 cur = min(len, PAGE_CACHE_SIZE - offset);
4177 kaddr = page_address(page);
4178 memcpy(kaddr + offset, src, cur);
4187 void memset_extent_buffer(struct extent_buffer *eb, char c,
4188 unsigned long start, unsigned long len)
4194 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4195 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4197 WARN_ON(start > eb->len);
4198 WARN_ON(start + len > eb->start + eb->len);
4200 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4203 page = extent_buffer_page(eb, i);
4204 WARN_ON(!PageUptodate(page));
4206 cur = min(len, PAGE_CACHE_SIZE - offset);
4207 kaddr = page_address(page);
4208 memset(kaddr + offset, c, cur);
4216 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4217 unsigned long dst_offset, unsigned long src_offset,
4220 u64 dst_len = dst->len;
4225 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4226 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4228 WARN_ON(src->len != dst_len);
4230 offset = (start_offset + dst_offset) &
4231 ((unsigned long)PAGE_CACHE_SIZE - 1);
4234 page = extent_buffer_page(dst, i);
4235 WARN_ON(!PageUptodate(page));
4237 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4239 kaddr = page_address(page);
4240 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4249 static void move_pages(struct page *dst_page, struct page *src_page,
4250 unsigned long dst_off, unsigned long src_off,
4253 char *dst_kaddr = page_address(dst_page);
4254 if (dst_page == src_page) {
4255 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4257 char *src_kaddr = page_address(src_page);
4258 char *p = dst_kaddr + dst_off + len;
4259 char *s = src_kaddr + src_off + len;
4266 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4268 unsigned long distance = (src > dst) ? src - dst : dst - src;
4269 return distance < len;
4272 static void copy_pages(struct page *dst_page, struct page *src_page,
4273 unsigned long dst_off, unsigned long src_off,
4276 char *dst_kaddr = page_address(dst_page);
4279 if (dst_page != src_page) {
4280 src_kaddr = page_address(src_page);
4282 src_kaddr = dst_kaddr;
4283 BUG_ON(areas_overlap(src_off, dst_off, len));
4286 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4289 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4290 unsigned long src_offset, unsigned long len)
4293 size_t dst_off_in_page;
4294 size_t src_off_in_page;
4295 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4296 unsigned long dst_i;
4297 unsigned long src_i;
4299 if (src_offset + len > dst->len) {
4300 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4301 "len %lu dst len %lu\n", src_offset, len, dst->len);
4304 if (dst_offset + len > dst->len) {
4305 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4306 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4311 dst_off_in_page = (start_offset + dst_offset) &
4312 ((unsigned long)PAGE_CACHE_SIZE - 1);
4313 src_off_in_page = (start_offset + src_offset) &
4314 ((unsigned long)PAGE_CACHE_SIZE - 1);
4316 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4317 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4319 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4321 cur = min_t(unsigned long, cur,
4322 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4324 copy_pages(extent_buffer_page(dst, dst_i),
4325 extent_buffer_page(dst, src_i),
4326 dst_off_in_page, src_off_in_page, cur);
4334 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4335 unsigned long src_offset, unsigned long len)
4338 size_t dst_off_in_page;
4339 size_t src_off_in_page;
4340 unsigned long dst_end = dst_offset + len - 1;
4341 unsigned long src_end = src_offset + len - 1;
4342 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4343 unsigned long dst_i;
4344 unsigned long src_i;
4346 if (src_offset + len > dst->len) {
4347 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4348 "len %lu len %lu\n", src_offset, len, dst->len);
4351 if (dst_offset + len > dst->len) {
4352 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4353 "len %lu len %lu\n", dst_offset, len, dst->len);
4356 if (!areas_overlap(src_offset, dst_offset, len)) {
4357 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4361 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4362 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4364 dst_off_in_page = (start_offset + dst_end) &
4365 ((unsigned long)PAGE_CACHE_SIZE - 1);
4366 src_off_in_page = (start_offset + src_end) &
4367 ((unsigned long)PAGE_CACHE_SIZE - 1);
4369 cur = min_t(unsigned long, len, src_off_in_page + 1);
4370 cur = min(cur, dst_off_in_page + 1);
4371 move_pages(extent_buffer_page(dst, dst_i),
4372 extent_buffer_page(dst, src_i),
4373 dst_off_in_page - cur + 1,
4374 src_off_in_page - cur + 1, cur);
4382 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4384 struct extent_buffer *eb =
4385 container_of(head, struct extent_buffer, rcu_head);
4387 btrfs_release_extent_buffer(eb);
4390 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
4392 u64 start = page_offset(page);
4393 struct extent_buffer *eb;
4396 spin_lock(&tree->buffer_lock);
4397 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4399 spin_unlock(&tree->buffer_lock);
4403 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4409 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4412 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
4417 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4419 spin_unlock(&tree->buffer_lock);
4421 /* at this point we can safely release the extent buffer */
4422 if (atomic_read(&eb->refs) == 0)
4423 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);