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);
942 err = insert_state(tree, prealloc, start, end, &bits);
944 BUG_ON(err == -EEXIST);
947 state = rb_entry(node, struct extent_state, rb_node);
949 last_start = state->start;
950 last_end = state->end;
953 * | ---- desired range ---- |
956 * Just lock what we found and keep going
958 if (state->start == start && state->end <= end) {
959 struct rb_node *next_node;
961 set_state_bits(tree, state, &bits);
962 clear_state_bit(tree, state, &clear_bits, 0);
964 merge_state(tree, state);
965 if (last_end == (u64)-1)
968 start = last_end + 1;
969 next_node = rb_next(&state->rb_node);
970 if (next_node && start < end && prealloc && !need_resched()) {
971 state = rb_entry(next_node, struct extent_state,
973 if (state->start == start)
980 * | ---- desired range ---- |
983 * | ------------- state -------------- |
985 * We need to split the extent we found, and may flip bits on
988 * If the extent we found extends past our
989 * range, we just split and search again. It'll get split
990 * again the next time though.
992 * If the extent we found is inside our range, we set the
995 if (state->start < start) {
996 prealloc = alloc_extent_state_atomic(prealloc);
1001 err = split_state(tree, state, prealloc, start);
1002 BUG_ON(err == -EEXIST);
1006 if (state->end <= end) {
1007 set_state_bits(tree, state, &bits);
1008 clear_state_bit(tree, state, &clear_bits, 0);
1009 merge_state(tree, state);
1010 if (last_end == (u64)-1)
1012 start = last_end + 1;
1017 * | ---- desired range ---- |
1018 * | state | or | state |
1020 * There's a hole, we need to insert something in it and
1021 * ignore the extent we found.
1023 if (state->start > start) {
1025 if (end < last_start)
1028 this_end = last_start - 1;
1030 prealloc = alloc_extent_state_atomic(prealloc);
1037 * Avoid to free 'prealloc' if it can be merged with
1040 err = insert_state(tree, prealloc, start, this_end,
1042 BUG_ON(err == -EEXIST);
1044 free_extent_state(prealloc);
1049 start = this_end + 1;
1053 * | ---- desired range ---- |
1055 * We need to split the extent, and set the bit
1058 if (state->start <= end && state->end > end) {
1059 prealloc = alloc_extent_state_atomic(prealloc);
1065 err = split_state(tree, state, prealloc, end + 1);
1066 BUG_ON(err == -EEXIST);
1068 set_state_bits(tree, prealloc, &bits);
1069 clear_state_bit(tree, prealloc, &clear_bits, 0);
1071 merge_state(tree, prealloc);
1079 spin_unlock(&tree->lock);
1081 free_extent_state(prealloc);
1088 spin_unlock(&tree->lock);
1089 if (mask & __GFP_WAIT)
1094 /* wrappers around set/clear extent bit */
1095 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1098 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
1102 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1103 int bits, gfp_t mask)
1105 return set_extent_bit(tree, start, end, bits, 0, NULL,
1109 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1110 int bits, gfp_t mask)
1112 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1115 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1116 struct extent_state **cached_state, gfp_t mask)
1118 return set_extent_bit(tree, start, end,
1119 EXTENT_DELALLOC | EXTENT_UPTODATE,
1120 0, NULL, cached_state, mask);
1123 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1126 return clear_extent_bit(tree, start, end,
1127 EXTENT_DIRTY | EXTENT_DELALLOC |
1128 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1131 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1134 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
1138 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1139 struct extent_state **cached_state, gfp_t mask)
1141 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1142 NULL, cached_state, mask);
1145 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
1146 u64 end, struct extent_state **cached_state,
1149 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1150 cached_state, mask);
1154 * either insert or lock state struct between start and end use mask to tell
1155 * us if waiting is desired.
1157 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1158 int bits, struct extent_state **cached_state, gfp_t mask)
1163 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1164 EXTENT_LOCKED, &failed_start,
1165 cached_state, mask);
1166 if (err == -EEXIST && (mask & __GFP_WAIT)) {
1167 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1168 start = failed_start;
1172 WARN_ON(start > end);
1177 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1179 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1182 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1188 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1189 &failed_start, NULL, mask);
1190 if (err == -EEXIST) {
1191 if (failed_start > start)
1192 clear_extent_bit(tree, start, failed_start - 1,
1193 EXTENT_LOCKED, 1, 0, NULL, mask);
1199 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1200 struct extent_state **cached, gfp_t mask)
1202 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1206 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1208 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1212 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1214 unsigned long index = start >> PAGE_CACHE_SHIFT;
1215 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1218 while (index <= end_index) {
1219 page = find_get_page(inode->i_mapping, index);
1220 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1221 clear_page_dirty_for_io(page);
1222 page_cache_release(page);
1228 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1230 unsigned long index = start >> PAGE_CACHE_SHIFT;
1231 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1234 while (index <= end_index) {
1235 page = find_get_page(inode->i_mapping, index);
1236 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1237 account_page_redirty(page);
1238 __set_page_dirty_nobuffers(page);
1239 page_cache_release(page);
1246 * helper function to set both pages and extents in the tree writeback
1248 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1250 unsigned long index = start >> PAGE_CACHE_SHIFT;
1251 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1254 while (index <= end_index) {
1255 page = find_get_page(tree->mapping, index);
1257 set_page_writeback(page);
1258 page_cache_release(page);
1264 /* find the first state struct with 'bits' set after 'start', and
1265 * return it. tree->lock must be held. NULL will returned if
1266 * nothing was found after 'start'
1268 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1269 u64 start, int bits)
1271 struct rb_node *node;
1272 struct extent_state *state;
1275 * this search will find all the extents that end after
1278 node = tree_search(tree, start);
1283 state = rb_entry(node, struct extent_state, rb_node);
1284 if (state->end >= start && (state->state & bits))
1287 node = rb_next(node);
1296 * find the first offset in the io tree with 'bits' set. zero is
1297 * returned if we find something, and *start_ret and *end_ret are
1298 * set to reflect the state struct that was found.
1300 * If nothing was found, 1 is returned, < 0 on error
1302 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1303 u64 *start_ret, u64 *end_ret, int bits)
1305 struct extent_state *state;
1308 spin_lock(&tree->lock);
1309 state = find_first_extent_bit_state(tree, start, bits);
1311 *start_ret = state->start;
1312 *end_ret = state->end;
1315 spin_unlock(&tree->lock);
1320 * find a contiguous range of bytes in the file marked as delalloc, not
1321 * more than 'max_bytes'. start and end are used to return the range,
1323 * 1 is returned if we find something, 0 if nothing was in the tree
1325 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1326 u64 *start, u64 *end, u64 max_bytes,
1327 struct extent_state **cached_state)
1329 struct rb_node *node;
1330 struct extent_state *state;
1331 u64 cur_start = *start;
1333 u64 total_bytes = 0;
1335 spin_lock(&tree->lock);
1338 * this search will find all the extents that end after
1341 node = tree_search(tree, cur_start);
1349 state = rb_entry(node, struct extent_state, rb_node);
1350 if (found && (state->start != cur_start ||
1351 (state->state & EXTENT_BOUNDARY))) {
1354 if (!(state->state & EXTENT_DELALLOC)) {
1360 *start = state->start;
1361 *cached_state = state;
1362 atomic_inc(&state->refs);
1366 cur_start = state->end + 1;
1367 node = rb_next(node);
1370 total_bytes += state->end - state->start + 1;
1371 if (total_bytes >= max_bytes)
1375 spin_unlock(&tree->lock);
1379 static noinline int __unlock_for_delalloc(struct inode *inode,
1380 struct page *locked_page,
1384 struct page *pages[16];
1385 unsigned long index = start >> PAGE_CACHE_SHIFT;
1386 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1387 unsigned long nr_pages = end_index - index + 1;
1390 if (index == locked_page->index && end_index == index)
1393 while (nr_pages > 0) {
1394 ret = find_get_pages_contig(inode->i_mapping, index,
1395 min_t(unsigned long, nr_pages,
1396 ARRAY_SIZE(pages)), pages);
1397 for (i = 0; i < ret; i++) {
1398 if (pages[i] != locked_page)
1399 unlock_page(pages[i]);
1400 page_cache_release(pages[i]);
1409 static noinline int lock_delalloc_pages(struct inode *inode,
1410 struct page *locked_page,
1414 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1415 unsigned long start_index = index;
1416 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1417 unsigned long pages_locked = 0;
1418 struct page *pages[16];
1419 unsigned long nrpages;
1423 /* the caller is responsible for locking the start index */
1424 if (index == locked_page->index && index == end_index)
1427 /* skip the page at the start index */
1428 nrpages = end_index - index + 1;
1429 while (nrpages > 0) {
1430 ret = find_get_pages_contig(inode->i_mapping, index,
1431 min_t(unsigned long,
1432 nrpages, ARRAY_SIZE(pages)), pages);
1437 /* now we have an array of pages, lock them all */
1438 for (i = 0; i < ret; i++) {
1440 * the caller is taking responsibility for
1443 if (pages[i] != locked_page) {
1444 lock_page(pages[i]);
1445 if (!PageDirty(pages[i]) ||
1446 pages[i]->mapping != inode->i_mapping) {
1448 unlock_page(pages[i]);
1449 page_cache_release(pages[i]);
1453 page_cache_release(pages[i]);
1462 if (ret && pages_locked) {
1463 __unlock_for_delalloc(inode, locked_page,
1465 ((u64)(start_index + pages_locked - 1)) <<
1472 * find a contiguous range of bytes in the file marked as delalloc, not
1473 * more than 'max_bytes'. start and end are used to return the range,
1475 * 1 is returned if we find something, 0 if nothing was in the tree
1477 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1478 struct extent_io_tree *tree,
1479 struct page *locked_page,
1480 u64 *start, u64 *end,
1486 struct extent_state *cached_state = NULL;
1491 /* step one, find a bunch of delalloc bytes starting at start */
1492 delalloc_start = *start;
1494 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1495 max_bytes, &cached_state);
1496 if (!found || delalloc_end <= *start) {
1497 *start = delalloc_start;
1498 *end = delalloc_end;
1499 free_extent_state(cached_state);
1504 * start comes from the offset of locked_page. We have to lock
1505 * pages in order, so we can't process delalloc bytes before
1508 if (delalloc_start < *start)
1509 delalloc_start = *start;
1512 * make sure to limit the number of pages we try to lock down
1515 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1516 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1518 /* step two, lock all the pages after the page that has start */
1519 ret = lock_delalloc_pages(inode, locked_page,
1520 delalloc_start, delalloc_end);
1521 if (ret == -EAGAIN) {
1522 /* some of the pages are gone, lets avoid looping by
1523 * shortening the size of the delalloc range we're searching
1525 free_extent_state(cached_state);
1526 cached_state = NULL;
1528 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1529 max_bytes = PAGE_CACHE_SIZE - offset;
1539 /* step three, lock the state bits for the whole range */
1540 lock_extent_bits(tree, delalloc_start, delalloc_end,
1541 0, &cached_state, GFP_NOFS);
1543 /* then test to make sure it is all still delalloc */
1544 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1545 EXTENT_DELALLOC, 1, cached_state);
1547 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1548 &cached_state, GFP_NOFS);
1549 __unlock_for_delalloc(inode, locked_page,
1550 delalloc_start, delalloc_end);
1554 free_extent_state(cached_state);
1555 *start = delalloc_start;
1556 *end = delalloc_end;
1561 int extent_clear_unlock_delalloc(struct inode *inode,
1562 struct extent_io_tree *tree,
1563 u64 start, u64 end, struct page *locked_page,
1567 struct page *pages[16];
1568 unsigned long index = start >> PAGE_CACHE_SHIFT;
1569 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1570 unsigned long nr_pages = end_index - index + 1;
1574 if (op & EXTENT_CLEAR_UNLOCK)
1575 clear_bits |= EXTENT_LOCKED;
1576 if (op & EXTENT_CLEAR_DIRTY)
1577 clear_bits |= EXTENT_DIRTY;
1579 if (op & EXTENT_CLEAR_DELALLOC)
1580 clear_bits |= EXTENT_DELALLOC;
1582 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1583 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1584 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1585 EXTENT_SET_PRIVATE2)))
1588 while (nr_pages > 0) {
1589 ret = find_get_pages_contig(inode->i_mapping, index,
1590 min_t(unsigned long,
1591 nr_pages, ARRAY_SIZE(pages)), pages);
1592 for (i = 0; i < ret; i++) {
1594 if (op & EXTENT_SET_PRIVATE2)
1595 SetPagePrivate2(pages[i]);
1597 if (pages[i] == locked_page) {
1598 page_cache_release(pages[i]);
1601 if (op & EXTENT_CLEAR_DIRTY)
1602 clear_page_dirty_for_io(pages[i]);
1603 if (op & EXTENT_SET_WRITEBACK)
1604 set_page_writeback(pages[i]);
1605 if (op & EXTENT_END_WRITEBACK)
1606 end_page_writeback(pages[i]);
1607 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1608 unlock_page(pages[i]);
1609 page_cache_release(pages[i]);
1619 * count the number of bytes in the tree that have a given bit(s)
1620 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1621 * cached. The total number found is returned.
1623 u64 count_range_bits(struct extent_io_tree *tree,
1624 u64 *start, u64 search_end, u64 max_bytes,
1625 unsigned long bits, int contig)
1627 struct rb_node *node;
1628 struct extent_state *state;
1629 u64 cur_start = *start;
1630 u64 total_bytes = 0;
1634 if (search_end <= cur_start) {
1639 spin_lock(&tree->lock);
1640 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1641 total_bytes = tree->dirty_bytes;
1645 * this search will find all the extents that end after
1648 node = tree_search(tree, cur_start);
1653 state = rb_entry(node, struct extent_state, rb_node);
1654 if (state->start > search_end)
1656 if (contig && found && state->start > last + 1)
1658 if (state->end >= cur_start && (state->state & bits) == bits) {
1659 total_bytes += min(search_end, state->end) + 1 -
1660 max(cur_start, state->start);
1661 if (total_bytes >= max_bytes)
1664 *start = max(cur_start, state->start);
1668 } else if (contig && found) {
1671 node = rb_next(node);
1676 spin_unlock(&tree->lock);
1681 * set the private field for a given byte offset in the tree. If there isn't
1682 * an extent_state there already, this does nothing.
1684 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1686 struct rb_node *node;
1687 struct extent_state *state;
1690 spin_lock(&tree->lock);
1692 * this search will find all the extents that end after
1695 node = tree_search(tree, start);
1700 state = rb_entry(node, struct extent_state, rb_node);
1701 if (state->start != start) {
1705 state->private = private;
1707 spin_unlock(&tree->lock);
1711 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1713 struct rb_node *node;
1714 struct extent_state *state;
1717 spin_lock(&tree->lock);
1719 * this search will find all the extents that end after
1722 node = tree_search(tree, start);
1727 state = rb_entry(node, struct extent_state, rb_node);
1728 if (state->start != start) {
1732 *private = state->private;
1734 spin_unlock(&tree->lock);
1739 * searches a range in the state tree for a given mask.
1740 * If 'filled' == 1, this returns 1 only if every extent in the tree
1741 * has the bits set. Otherwise, 1 is returned if any bit in the
1742 * range is found set.
1744 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1745 int bits, int filled, struct extent_state *cached)
1747 struct extent_state *state = NULL;
1748 struct rb_node *node;
1751 spin_lock(&tree->lock);
1752 if (cached && cached->tree && cached->start <= start &&
1753 cached->end > start)
1754 node = &cached->rb_node;
1756 node = tree_search(tree, start);
1757 while (node && start <= end) {
1758 state = rb_entry(node, struct extent_state, rb_node);
1760 if (filled && state->start > start) {
1765 if (state->start > end)
1768 if (state->state & bits) {
1772 } else if (filled) {
1777 if (state->end == (u64)-1)
1780 start = state->end + 1;
1783 node = rb_next(node);
1790 spin_unlock(&tree->lock);
1795 * helper function to set a given page up to date if all the
1796 * extents in the tree for that page are up to date
1798 static int check_page_uptodate(struct extent_io_tree *tree,
1801 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1802 u64 end = start + PAGE_CACHE_SIZE - 1;
1803 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1804 SetPageUptodate(page);
1809 * helper function to unlock a page if all the extents in the tree
1810 * for that page are unlocked
1812 static int check_page_locked(struct extent_io_tree *tree,
1815 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1816 u64 end = start + PAGE_CACHE_SIZE - 1;
1817 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1823 * helper function to end page writeback if all the extents
1824 * in the tree for that page are done with writeback
1826 static int check_page_writeback(struct extent_io_tree *tree,
1829 end_page_writeback(page);
1834 * When IO fails, either with EIO or csum verification fails, we
1835 * try other mirrors that might have a good copy of the data. This
1836 * io_failure_record is used to record state as we go through all the
1837 * mirrors. If another mirror has good data, the page is set up to date
1838 * and things continue. If a good mirror can't be found, the original
1839 * bio end_io callback is called to indicate things have failed.
1841 struct io_failure_record {
1846 unsigned long bio_flags;
1852 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1857 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1859 set_state_private(failure_tree, rec->start, 0);
1860 ret = clear_extent_bits(failure_tree, rec->start,
1861 rec->start + rec->len - 1,
1862 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1867 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1868 rec->start + rec->len - 1,
1869 EXTENT_DAMAGED, GFP_NOFS);
1878 static void repair_io_failure_callback(struct bio *bio, int err)
1880 complete(bio->bi_private);
1884 * this bypasses the standard btrfs submit functions deliberately, as
1885 * the standard behavior is to write all copies in a raid setup. here we only
1886 * want to write the one bad copy. so we do the mapping for ourselves and issue
1887 * submit_bio directly.
1888 * to avoid any synchonization issues, wait for the data after writing, which
1889 * actually prevents the read that triggered the error from finishing.
1890 * currently, there can be no more than two copies of every data bit. thus,
1891 * exactly one rewrite is required.
1893 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1894 u64 length, u64 logical, struct page *page,
1898 struct btrfs_device *dev;
1899 DECLARE_COMPLETION_ONSTACK(compl);
1902 struct btrfs_bio *bbio = NULL;
1905 BUG_ON(!mirror_num);
1907 bio = bio_alloc(GFP_NOFS, 1);
1910 bio->bi_private = &compl;
1911 bio->bi_end_io = repair_io_failure_callback;
1913 map_length = length;
1915 ret = btrfs_map_block(map_tree, WRITE, logical,
1916 &map_length, &bbio, mirror_num);
1921 BUG_ON(mirror_num != bbio->mirror_num);
1922 sector = bbio->stripes[mirror_num-1].physical >> 9;
1923 bio->bi_sector = sector;
1924 dev = bbio->stripes[mirror_num-1].dev;
1926 if (!dev || !dev->bdev || !dev->writeable) {
1930 bio->bi_bdev = dev->bdev;
1931 bio_add_page(bio, page, length, start-page_offset(page));
1932 submit_bio(WRITE_SYNC, bio);
1933 wait_for_completion(&compl);
1935 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1936 /* try to remap that extent elsewhere? */
1941 printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s "
1942 "sector %llu)\n", page->mapping->host->i_ino, start,
1950 * each time an IO finishes, we do a fast check in the IO failure tree
1951 * to see if we need to process or clean up an io_failure_record
1953 static int clean_io_failure(u64 start, struct page *page)
1956 u64 private_failure;
1957 struct io_failure_record *failrec;
1958 struct btrfs_mapping_tree *map_tree;
1959 struct extent_state *state;
1963 struct inode *inode = page->mapping->host;
1966 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1967 (u64)-1, 1, EXTENT_DIRTY, 0);
1971 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1976 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1977 BUG_ON(!failrec->this_mirror);
1979 if (failrec->in_validation) {
1980 /* there was no real error, just free the record */
1981 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1987 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1988 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1991 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1993 if (state && state->start == failrec->start) {
1994 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
1995 num_copies = btrfs_num_copies(map_tree, failrec->logical,
1997 if (num_copies > 1) {
1998 ret = repair_io_failure(map_tree, start, failrec->len,
1999 failrec->logical, page,
2000 failrec->failed_mirror);
2007 ret = free_io_failure(inode, failrec, did_repair);
2013 * this is a generic handler for readpage errors (default
2014 * readpage_io_failed_hook). if other copies exist, read those and write back
2015 * good data to the failed position. does not investigate in remapping the
2016 * failed extent elsewhere, hoping the device will be smart enough to do this as
2020 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2021 u64 start, u64 end, int failed_mirror,
2022 struct extent_state *state)
2024 struct io_failure_record *failrec = NULL;
2026 struct extent_map *em;
2027 struct inode *inode = page->mapping->host;
2028 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2029 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2030 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2037 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2039 ret = get_state_private(failure_tree, start, &private);
2041 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2044 failrec->start = start;
2045 failrec->len = end - start + 1;
2046 failrec->this_mirror = 0;
2047 failrec->bio_flags = 0;
2048 failrec->in_validation = 0;
2050 read_lock(&em_tree->lock);
2051 em = lookup_extent_mapping(em_tree, start, failrec->len);
2053 read_unlock(&em_tree->lock);
2058 if (em->start > start || em->start + em->len < start) {
2059 free_extent_map(em);
2062 read_unlock(&em_tree->lock);
2064 if (!em || IS_ERR(em)) {
2068 logical = start - em->start;
2069 logical = em->block_start + logical;
2070 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2071 logical = em->block_start;
2072 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2073 extent_set_compress_type(&failrec->bio_flags,
2076 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2077 "len=%llu\n", logical, start, failrec->len);
2078 failrec->logical = logical;
2079 free_extent_map(em);
2081 /* set the bits in the private failure tree */
2082 ret = set_extent_bits(failure_tree, start, end,
2083 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2085 ret = set_state_private(failure_tree, start,
2086 (u64)(unsigned long)failrec);
2087 /* set the bits in the inode's tree */
2089 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2096 failrec = (struct io_failure_record *)(unsigned long)private;
2097 pr_debug("bio_readpage_error: (found) logical=%llu, "
2098 "start=%llu, len=%llu, validation=%d\n",
2099 failrec->logical, failrec->start, failrec->len,
2100 failrec->in_validation);
2102 * when data can be on disk more than twice, add to failrec here
2103 * (e.g. with a list for failed_mirror) to make
2104 * clean_io_failure() clean all those errors at once.
2107 num_copies = btrfs_num_copies(
2108 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2109 failrec->logical, failrec->len);
2110 if (num_copies == 1) {
2112 * we only have a single copy of the data, so don't bother with
2113 * all the retry and error correction code that follows. no
2114 * matter what the error is, it is very likely to persist.
2116 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2117 "state=%p, num_copies=%d, next_mirror %d, "
2118 "failed_mirror %d\n", state, num_copies,
2119 failrec->this_mirror, failed_mirror);
2120 free_io_failure(inode, failrec, 0);
2125 spin_lock(&tree->lock);
2126 state = find_first_extent_bit_state(tree, failrec->start,
2128 if (state && state->start != failrec->start)
2130 spin_unlock(&tree->lock);
2134 * there are two premises:
2135 * a) deliver good data to the caller
2136 * b) correct the bad sectors on disk
2138 if (failed_bio->bi_vcnt > 1) {
2140 * to fulfill b), we need to know the exact failing sectors, as
2141 * we don't want to rewrite any more than the failed ones. thus,
2142 * we need separate read requests for the failed bio
2144 * if the following BUG_ON triggers, our validation request got
2145 * merged. we need separate requests for our algorithm to work.
2147 BUG_ON(failrec->in_validation);
2148 failrec->in_validation = 1;
2149 failrec->this_mirror = failed_mirror;
2150 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2153 * we're ready to fulfill a) and b) alongside. get a good copy
2154 * of the failed sector and if we succeed, we have setup
2155 * everything for repair_io_failure to do the rest for us.
2157 if (failrec->in_validation) {
2158 BUG_ON(failrec->this_mirror != failed_mirror);
2159 failrec->in_validation = 0;
2160 failrec->this_mirror = 0;
2162 failrec->failed_mirror = failed_mirror;
2163 failrec->this_mirror++;
2164 if (failrec->this_mirror == failed_mirror)
2165 failrec->this_mirror++;
2166 read_mode = READ_SYNC;
2169 if (!state || failrec->this_mirror > num_copies) {
2170 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2171 "next_mirror %d, failed_mirror %d\n", state,
2172 num_copies, failrec->this_mirror, failed_mirror);
2173 free_io_failure(inode, failrec, 0);
2177 bio = bio_alloc(GFP_NOFS, 1);
2178 bio->bi_private = state;
2179 bio->bi_end_io = failed_bio->bi_end_io;
2180 bio->bi_sector = failrec->logical >> 9;
2181 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2184 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2186 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2187 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2188 failrec->this_mirror, num_copies, failrec->in_validation);
2190 tree->ops->submit_bio_hook(inode, read_mode, bio, failrec->this_mirror,
2191 failrec->bio_flags, 0);
2195 /* lots and lots of room for performance fixes in the end_bio funcs */
2198 * after a writepage IO is done, we need to:
2199 * clear the uptodate bits on error
2200 * clear the writeback bits in the extent tree for this IO
2201 * end_page_writeback if the page has no more pending IO
2203 * Scheduling is not allowed, so the extent state tree is expected
2204 * to have one and only one object corresponding to this IO.
2206 static void end_bio_extent_writepage(struct bio *bio, int err)
2208 int uptodate = err == 0;
2209 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2210 struct extent_io_tree *tree;
2217 struct page *page = bvec->bv_page;
2218 tree = &BTRFS_I(page->mapping->host)->io_tree;
2220 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2222 end = start + bvec->bv_len - 1;
2224 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2229 if (--bvec >= bio->bi_io_vec)
2230 prefetchw(&bvec->bv_page->flags);
2231 if (tree->ops && tree->ops->writepage_end_io_hook) {
2232 ret = tree->ops->writepage_end_io_hook(page, start,
2233 end, NULL, uptodate);
2238 if (!uptodate && tree->ops &&
2239 tree->ops->writepage_io_failed_hook) {
2240 ret = tree->ops->writepage_io_failed_hook(bio, page,
2243 uptodate = (err == 0);
2249 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
2250 ClearPageUptodate(page);
2255 end_page_writeback(page);
2257 check_page_writeback(tree, page);
2258 } while (bvec >= bio->bi_io_vec);
2264 * after a readpage IO is done, we need to:
2265 * clear the uptodate bits on error
2266 * set the uptodate bits if things worked
2267 * set the page up to date if all extents in the tree are uptodate
2268 * clear the lock bit in the extent tree
2269 * unlock the page if there are no other extents locked for it
2271 * Scheduling is not allowed, so the extent state tree is expected
2272 * to have one and only one object corresponding to this IO.
2274 static void end_bio_extent_readpage(struct bio *bio, int err)
2276 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2277 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2278 struct bio_vec *bvec = bio->bi_io_vec;
2279 struct extent_io_tree *tree;
2289 struct page *page = bvec->bv_page;
2290 struct extent_state *cached = NULL;
2291 struct extent_state *state;
2293 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2294 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2295 (long int)bio->bi_bdev);
2296 tree = &BTRFS_I(page->mapping->host)->io_tree;
2298 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2300 end = start + bvec->bv_len - 1;
2302 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2307 if (++bvec <= bvec_end)
2308 prefetchw(&bvec->bv_page->flags);
2310 spin_lock(&tree->lock);
2311 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2312 if (state && state->start == start) {
2314 * take a reference on the state, unlock will drop
2317 cache_state(state, &cached);
2319 spin_unlock(&tree->lock);
2321 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2322 ret = tree->ops->readpage_end_io_hook(page, start, end,
2327 clean_io_failure(start, page);
2331 failed_mirror = (int)(unsigned long)bio->bi_bdev;
2333 * The generic bio_readpage_error handles errors the
2334 * following way: If possible, new read requests are
2335 * created and submitted and will end up in
2336 * end_bio_extent_readpage as well (if we're lucky, not
2337 * in the !uptodate case). In that case it returns 0 and
2338 * we just go on with the next page in our bio. If it
2339 * can't handle the error it will return -EIO and we
2340 * remain responsible for that page.
2342 ret = bio_readpage_error(bio, page, start, end,
2343 failed_mirror, NULL);
2347 test_bit(BIO_UPTODATE, &bio->bi_flags);
2350 uncache_state(&cached);
2353 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2354 ret = tree->ops->readpage_io_failed_hook(
2355 bio, page, start, end,
2356 failed_mirror, state);
2363 set_extent_uptodate(tree, start, end, &cached,
2366 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2370 SetPageUptodate(page);
2372 ClearPageUptodate(page);
2378 check_page_uptodate(tree, page);
2380 ClearPageUptodate(page);
2383 check_page_locked(tree, page);
2385 } while (bvec <= bvec_end);
2391 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2396 bio = bio_alloc(gfp_flags, nr_vecs);
2398 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2399 while (!bio && (nr_vecs /= 2))
2400 bio = bio_alloc(gfp_flags, nr_vecs);
2405 bio->bi_bdev = bdev;
2406 bio->bi_sector = first_sector;
2411 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
2412 unsigned long bio_flags)
2415 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2416 struct page *page = bvec->bv_page;
2417 struct extent_io_tree *tree = bio->bi_private;
2420 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2422 bio->bi_private = NULL;
2426 if (tree->ops && tree->ops->submit_bio_hook)
2427 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2428 mirror_num, bio_flags, start);
2430 submit_bio(rw, bio);
2432 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2438 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2439 struct page *page, sector_t sector,
2440 size_t size, unsigned long offset,
2441 struct block_device *bdev,
2442 struct bio **bio_ret,
2443 unsigned long max_pages,
2444 bio_end_io_t end_io_func,
2446 unsigned long prev_bio_flags,
2447 unsigned long bio_flags,
2448 bool force_bio_submit)
2454 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2455 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2456 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2458 if (bio_ret && *bio_ret) {
2461 contig = bio->bi_sector == sector;
2463 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2466 if (prev_bio_flags != bio_flags || !contig ||
2468 (tree->ops && tree->ops->merge_bio_hook &&
2469 tree->ops->merge_bio_hook(page, offset, page_size, bio,
2471 bio_add_page(bio, page, page_size, offset) < page_size) {
2472 ret = submit_one_bio(rw, bio, mirror_num,
2479 if (this_compressed)
2482 nr = bio_get_nr_vecs(bdev);
2484 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2488 bio_add_page(bio, page, page_size, offset);
2489 bio->bi_end_io = end_io_func;
2490 bio->bi_private = tree;
2495 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2500 void set_page_extent_mapped(struct page *page)
2502 if (!PagePrivate(page)) {
2503 SetPagePrivate(page);
2504 page_cache_get(page);
2505 set_page_private(page, EXTENT_PAGE_PRIVATE);
2509 static void set_page_extent_head(struct page *page, unsigned long len)
2511 WARN_ON(!PagePrivate(page));
2512 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
2516 * basic readpage implementation. Locked extent state structs are inserted
2517 * into the tree that are removed when the IO is done (by the end_io
2520 static int __extent_read_full_page(struct extent_io_tree *tree,
2522 get_extent_t *get_extent,
2523 struct bio **bio, int mirror_num,
2524 unsigned long *bio_flags,
2527 struct inode *inode = page->mapping->host;
2528 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2529 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2533 u64 last_byte = i_size_read(inode);
2537 struct extent_map *em;
2538 struct block_device *bdev;
2539 struct btrfs_ordered_extent *ordered;
2542 size_t pg_offset = 0;
2544 size_t disk_io_size;
2545 size_t blocksize = inode->i_sb->s_blocksize;
2546 unsigned long this_bio_flag = 0;
2548 set_page_extent_mapped(page);
2550 if (!PageUptodate(page)) {
2551 if (cleancache_get_page(page) == 0) {
2552 BUG_ON(blocksize != PAGE_SIZE);
2559 lock_extent(tree, start, end, GFP_NOFS);
2560 ordered = btrfs_lookup_ordered_extent(inode, start);
2563 unlock_extent(tree, start, end, GFP_NOFS);
2564 btrfs_start_ordered_extent(inode, ordered, 1);
2565 btrfs_put_ordered_extent(ordered);
2568 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2570 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2573 iosize = PAGE_CACHE_SIZE - zero_offset;
2574 userpage = kmap_atomic(page, KM_USER0);
2575 memset(userpage + zero_offset, 0, iosize);
2576 flush_dcache_page(page);
2577 kunmap_atomic(userpage, KM_USER0);
2580 while (cur <= end) {
2581 bool force_bio_submit = false;
2583 if (cur >= last_byte) {
2585 struct extent_state *cached = NULL;
2587 iosize = PAGE_CACHE_SIZE - pg_offset;
2588 userpage = kmap_atomic(page, KM_USER0);
2589 memset(userpage + pg_offset, 0, iosize);
2590 flush_dcache_page(page);
2591 kunmap_atomic(userpage, KM_USER0);
2592 set_extent_uptodate(tree, cur, cur + iosize - 1,
2594 unlock_extent_cached(tree, cur, cur + iosize - 1,
2598 em = get_extent(inode, page, pg_offset, cur,
2600 if (IS_ERR_OR_NULL(em)) {
2602 unlock_extent(tree, cur, end, GFP_NOFS);
2605 extent_offset = cur - em->start;
2606 BUG_ON(extent_map_end(em) <= cur);
2609 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2610 this_bio_flag = EXTENT_BIO_COMPRESSED;
2611 extent_set_compress_type(&this_bio_flag,
2615 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2616 cur_end = min(extent_map_end(em) - 1, end);
2617 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2618 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2619 disk_io_size = em->block_len;
2620 sector = em->block_start >> 9;
2622 sector = (em->block_start + extent_offset) >> 9;
2623 disk_io_size = iosize;
2626 block_start = em->block_start;
2627 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2628 block_start = EXTENT_MAP_HOLE;
2631 * If we have a file range that points to a compressed extent
2632 * and it's followed by a consecutive file range that points to
2633 * to the same compressed extent (possibly with a different
2634 * offset and/or length, so it either points to the whole extent
2635 * or only part of it), we must make sure we do not submit a
2636 * single bio to populate the pages for the 2 ranges because
2637 * this makes the compressed extent read zero out the pages
2638 * belonging to the 2nd range. Imagine the following scenario:
2641 * [0 - 8K] [8K - 24K]
2644 * points to extent X, points to extent X,
2645 * offset 4K, length of 8K offset 0, length 16K
2647 * [extent X, compressed length = 4K uncompressed length = 16K]
2649 * If the bio to read the compressed extent covers both ranges,
2650 * it will decompress extent X into the pages belonging to the
2651 * first range and then it will stop, zeroing out the remaining
2652 * pages that belong to the other range that points to extent X.
2653 * So here we make sure we submit 2 bios, one for the first
2654 * range and another one for the third range. Both will target
2655 * the same physical extent from disk, but we can't currently
2656 * make the compressed bio endio callback populate the pages
2657 * for both ranges because each compressed bio is tightly
2658 * coupled with a single extent map, and each range can have
2659 * an extent map with a different offset value relative to the
2660 * uncompressed data of our extent and different lengths. This
2661 * is a corner case so we prioritize correctness over
2662 * non-optimal behavior (submitting 2 bios for the same extent).
2664 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
2665 prev_em_start && *prev_em_start != (u64)-1 &&
2666 *prev_em_start != em->orig_start)
2667 force_bio_submit = true;
2670 *prev_em_start = em->orig_start;
2672 free_extent_map(em);
2675 /* we've found a hole, just zero and go on */
2676 if (block_start == EXTENT_MAP_HOLE) {
2678 struct extent_state *cached = NULL;
2680 userpage = kmap_atomic(page, KM_USER0);
2681 memset(userpage + pg_offset, 0, iosize);
2682 flush_dcache_page(page);
2683 kunmap_atomic(userpage, KM_USER0);
2685 set_extent_uptodate(tree, cur, cur + iosize - 1,
2687 unlock_extent_cached(tree, cur, cur + iosize - 1,
2690 pg_offset += iosize;
2693 /* the get_extent function already copied into the page */
2694 if (test_range_bit(tree, cur, cur_end,
2695 EXTENT_UPTODATE, 1, NULL)) {
2696 check_page_uptodate(tree, page);
2697 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2699 pg_offset += iosize;
2702 /* we have an inline extent but it didn't get marked up
2703 * to date. Error out
2705 if (block_start == EXTENT_MAP_INLINE) {
2707 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2709 pg_offset += iosize;
2714 if (tree->ops && tree->ops->readpage_io_hook) {
2715 ret = tree->ops->readpage_io_hook(page, cur,
2719 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2721 ret = submit_extent_page(READ, tree, page,
2722 sector, disk_io_size, pg_offset,
2724 end_bio_extent_readpage, mirror_num,
2729 *bio_flags = this_bio_flag;
2734 pg_offset += iosize;
2738 if (!PageError(page))
2739 SetPageUptodate(page);
2745 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2746 get_extent_t *get_extent, int mirror_num)
2748 struct bio *bio = NULL;
2749 unsigned long bio_flags = 0;
2752 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2755 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2759 static noinline void update_nr_written(struct page *page,
2760 struct writeback_control *wbc,
2761 unsigned long nr_written)
2763 wbc->nr_to_write -= nr_written;
2764 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2765 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2766 page->mapping->writeback_index = page->index + nr_written;
2770 * the writepage semantics are similar to regular writepage. extent
2771 * records are inserted to lock ranges in the tree, and as dirty areas
2772 * are found, they are marked writeback. Then the lock bits are removed
2773 * and the end_io handler clears the writeback ranges
2775 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2778 struct inode *inode = page->mapping->host;
2779 struct extent_page_data *epd = data;
2780 struct extent_io_tree *tree = epd->tree;
2781 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2783 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2787 u64 last_byte = i_size_read(inode);
2791 struct extent_state *cached_state = NULL;
2792 struct extent_map *em;
2793 struct block_device *bdev;
2796 size_t pg_offset = 0;
2798 loff_t i_size = i_size_read(inode);
2799 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2805 unsigned long nr_written = 0;
2806 bool fill_delalloc = true;
2808 if (wbc->sync_mode == WB_SYNC_ALL)
2809 write_flags = WRITE_SYNC;
2811 write_flags = WRITE;
2813 trace___extent_writepage(page, inode, wbc);
2815 WARN_ON(!PageLocked(page));
2817 ClearPageError(page);
2819 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2820 if (page->index > end_index ||
2821 (page->index == end_index && !pg_offset)) {
2822 page->mapping->a_ops->invalidatepage(page, 0);
2827 if (page->index == end_index) {
2830 userpage = kmap_atomic(page, KM_USER0);
2831 memset(userpage + pg_offset, 0,
2832 PAGE_CACHE_SIZE - pg_offset);
2833 kunmap_atomic(userpage, KM_USER0);
2834 flush_dcache_page(page);
2838 set_page_extent_mapped(page);
2840 if (!tree->ops || !tree->ops->fill_delalloc)
2841 fill_delalloc = false;
2843 delalloc_start = start;
2846 if (!epd->extent_locked && fill_delalloc) {
2847 u64 delalloc_to_write = 0;
2849 * make sure the wbc mapping index is at least updated
2852 update_nr_written(page, wbc, 0);
2854 while (delalloc_end < page_end) {
2855 nr_delalloc = find_lock_delalloc_range(inode, tree,
2860 if (nr_delalloc == 0) {
2861 delalloc_start = delalloc_end + 1;
2864 tree->ops->fill_delalloc(inode, page, delalloc_start,
2865 delalloc_end, &page_started,
2868 * delalloc_end is already one less than the total
2869 * length, so we don't subtract one from
2872 delalloc_to_write += (delalloc_end - delalloc_start +
2875 delalloc_start = delalloc_end + 1;
2877 if (wbc->nr_to_write < delalloc_to_write) {
2880 if (delalloc_to_write < thresh * 2)
2881 thresh = delalloc_to_write;
2882 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2886 /* did the fill delalloc function already unlock and start
2892 * we've unlocked the page, so we can't update
2893 * the mapping's writeback index, just update
2896 wbc->nr_to_write -= nr_written;
2900 if (tree->ops && tree->ops->writepage_start_hook) {
2901 ret = tree->ops->writepage_start_hook(page, start,
2903 if (ret == -EAGAIN) {
2904 redirty_page_for_writepage(wbc, page);
2905 update_nr_written(page, wbc, nr_written);
2913 * we don't want to touch the inode after unlocking the page,
2914 * so we update the mapping writeback index now
2916 update_nr_written(page, wbc, nr_written + 1);
2919 if (last_byte <= start) {
2920 if (tree->ops && tree->ops->writepage_end_io_hook)
2921 tree->ops->writepage_end_io_hook(page, start,
2926 blocksize = inode->i_sb->s_blocksize;
2928 while (cur <= end) {
2929 if (cur >= last_byte) {
2930 if (tree->ops && tree->ops->writepage_end_io_hook)
2931 tree->ops->writepage_end_io_hook(page, cur,
2935 em = epd->get_extent(inode, page, pg_offset, cur,
2937 if (IS_ERR_OR_NULL(em)) {
2942 extent_offset = cur - em->start;
2943 BUG_ON(extent_map_end(em) <= cur);
2945 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2946 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2947 sector = (em->block_start + extent_offset) >> 9;
2949 block_start = em->block_start;
2950 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2951 free_extent_map(em);
2955 * compressed and inline extents are written through other
2958 if (compressed || block_start == EXTENT_MAP_HOLE ||
2959 block_start == EXTENT_MAP_INLINE) {
2961 * end_io notification does not happen here for
2962 * compressed extents
2964 if (!compressed && tree->ops &&
2965 tree->ops->writepage_end_io_hook)
2966 tree->ops->writepage_end_io_hook(page, cur,
2969 else if (compressed) {
2970 /* we don't want to end_page_writeback on
2971 * a compressed extent. this happens
2978 pg_offset += iosize;
2981 /* leave this out until we have a page_mkwrite call */
2982 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2983 EXTENT_DIRTY, 0, NULL)) {
2985 pg_offset += iosize;
2989 if (tree->ops && tree->ops->writepage_io_hook) {
2990 ret = tree->ops->writepage_io_hook(page, cur,
2998 unsigned long max_nr = end_index + 1;
3000 set_range_writeback(tree, cur, cur + iosize - 1);
3001 if (!PageWriteback(page)) {
3002 printk(KERN_ERR "btrfs warning page %lu not "
3003 "writeback, cur %llu end %llu\n",
3004 page->index, (unsigned long long)cur,
3005 (unsigned long long)end);
3008 ret = submit_extent_page(write_flags, tree, page,
3009 sector, iosize, pg_offset,
3010 bdev, &epd->bio, max_nr,
3011 end_bio_extent_writepage,
3017 pg_offset += iosize;
3022 /* make sure the mapping tag for page dirty gets cleared */
3023 set_page_writeback(page);
3024 end_page_writeback(page);
3030 /* drop our reference on any cached states */
3031 free_extent_state(cached_state);
3036 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3037 * @mapping: address space structure to write
3038 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3039 * @writepage: function called for each page
3040 * @data: data passed to writepage function
3042 * If a page is already under I/O, write_cache_pages() skips it, even
3043 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3044 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3045 * and msync() need to guarantee that all the data which was dirty at the time
3046 * the call was made get new I/O started against them. If wbc->sync_mode is
3047 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3048 * existing IO to complete.
3050 static int extent_write_cache_pages(struct extent_io_tree *tree,
3051 struct address_space *mapping,
3052 struct writeback_control *wbc,
3053 writepage_t writepage, void *data,
3054 void (*flush_fn)(void *))
3058 int nr_to_write_done = 0;
3059 struct pagevec pvec;
3062 pgoff_t end; /* Inclusive */
3066 pagevec_init(&pvec, 0);
3067 if (wbc->range_cyclic) {
3068 index = mapping->writeback_index; /* Start from prev offset */
3071 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3072 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3075 if (wbc->sync_mode == WB_SYNC_ALL)
3076 tag = PAGECACHE_TAG_TOWRITE;
3078 tag = PAGECACHE_TAG_DIRTY;
3080 if (wbc->sync_mode == WB_SYNC_ALL)
3081 tag_pages_for_writeback(mapping, index, end);
3082 while (!done && !nr_to_write_done && (index <= end) &&
3083 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3084 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3088 for (i = 0; i < nr_pages; i++) {
3089 struct page *page = pvec.pages[i];
3092 * At this point we hold neither mapping->tree_lock nor
3093 * lock on the page itself: the page may be truncated or
3094 * invalidated (changing page->mapping to NULL), or even
3095 * swizzled back from swapper_space to tmpfs file
3099 tree->ops->write_cache_pages_lock_hook) {
3100 tree->ops->write_cache_pages_lock_hook(page,
3103 if (!trylock_page(page)) {
3109 if (unlikely(page->mapping != mapping)) {
3114 if (!wbc->range_cyclic && page->index > end) {
3120 if (wbc->sync_mode != WB_SYNC_NONE) {
3121 if (PageWriteback(page))
3123 wait_on_page_writeback(page);
3126 if (PageWriteback(page) ||
3127 !clear_page_dirty_for_io(page)) {
3132 ret = (*writepage)(page, wbc, data);
3134 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3142 * the filesystem may choose to bump up nr_to_write.
3143 * We have to make sure to honor the new nr_to_write
3146 nr_to_write_done = wbc->nr_to_write <= 0;
3148 pagevec_release(&pvec);
3151 if (!scanned && !done) {
3153 * We hit the last page and there is more work to be done: wrap
3154 * back to the start of the file
3163 static void flush_epd_write_bio(struct extent_page_data *epd)
3167 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
3169 submit_one_bio(WRITE, epd->bio, 0, 0);
3174 static noinline void flush_write_bio(void *data)
3176 struct extent_page_data *epd = data;
3177 flush_epd_write_bio(epd);
3180 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3181 get_extent_t *get_extent,
3182 struct writeback_control *wbc)
3185 struct extent_page_data epd = {
3188 .get_extent = get_extent,
3190 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3193 ret = __extent_writepage(page, wbc, &epd);
3195 flush_epd_write_bio(&epd);
3199 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3200 u64 start, u64 end, get_extent_t *get_extent,
3204 struct address_space *mapping = inode->i_mapping;
3206 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3209 struct extent_page_data epd = {
3212 .get_extent = get_extent,
3214 .sync_io = mode == WB_SYNC_ALL,
3216 struct writeback_control wbc_writepages = {
3218 .nr_to_write = nr_pages * 2,
3219 .range_start = start,
3220 .range_end = end + 1,
3223 while (start <= end) {
3224 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3225 if (clear_page_dirty_for_io(page))
3226 ret = __extent_writepage(page, &wbc_writepages, &epd);
3228 if (tree->ops && tree->ops->writepage_end_io_hook)
3229 tree->ops->writepage_end_io_hook(page, start,
3230 start + PAGE_CACHE_SIZE - 1,
3234 page_cache_release(page);
3235 start += PAGE_CACHE_SIZE;
3238 flush_epd_write_bio(&epd);
3242 int extent_writepages(struct extent_io_tree *tree,
3243 struct address_space *mapping,
3244 get_extent_t *get_extent,
3245 struct writeback_control *wbc)
3248 struct extent_page_data epd = {
3251 .get_extent = get_extent,
3253 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3256 ret = extent_write_cache_pages(tree, mapping, wbc,
3257 __extent_writepage, &epd,
3259 flush_epd_write_bio(&epd);
3263 int extent_readpages(struct extent_io_tree *tree,
3264 struct address_space *mapping,
3265 struct list_head *pages, unsigned nr_pages,
3266 get_extent_t get_extent)
3268 struct bio *bio = NULL;
3270 unsigned long bio_flags = 0;
3271 u64 prev_em_start = (u64)-1;
3273 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3274 struct page *page = list_entry(pages->prev, struct page, lru);
3276 prefetchw(&page->flags);
3277 list_del(&page->lru);
3278 if (!add_to_page_cache_lru(page, mapping,
3279 page->index, GFP_NOFS)) {
3280 __extent_read_full_page(tree, page, get_extent,
3281 &bio, 0, &bio_flags,
3284 page_cache_release(page);
3286 BUG_ON(!list_empty(pages));
3288 submit_one_bio(READ, bio, 0, bio_flags);
3293 * basic invalidatepage code, this waits on any locked or writeback
3294 * ranges corresponding to the page, and then deletes any extent state
3295 * records from the tree
3297 int extent_invalidatepage(struct extent_io_tree *tree,
3298 struct page *page, unsigned long offset)
3300 struct extent_state *cached_state = NULL;
3301 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3302 u64 end = start + PAGE_CACHE_SIZE - 1;
3303 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3305 start += (offset + blocksize - 1) & ~(blocksize - 1);
3309 lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS);
3310 wait_on_page_writeback(page);
3311 clear_extent_bit(tree, start, end,
3312 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3313 EXTENT_DO_ACCOUNTING,
3314 1, 1, &cached_state, GFP_NOFS);
3319 * a helper for releasepage, this tests for areas of the page that
3320 * are locked or under IO and drops the related state bits if it is safe
3323 int try_release_extent_state(struct extent_map_tree *map,
3324 struct extent_io_tree *tree, struct page *page,
3327 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3328 u64 end = start + PAGE_CACHE_SIZE - 1;
3331 if (test_range_bit(tree, start, end,
3332 EXTENT_IOBITS, 0, NULL))
3335 if ((mask & GFP_NOFS) == GFP_NOFS)
3338 * at this point we can safely clear everything except the
3339 * locked bit and the nodatasum bit
3341 ret = clear_extent_bit(tree, start, end,
3342 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3345 /* if clear_extent_bit failed for enomem reasons,
3346 * we can't allow the release to continue.
3357 * a helper for releasepage. As long as there are no locked extents
3358 * in the range corresponding to the page, both state records and extent
3359 * map records are removed
3361 int try_release_extent_mapping(struct extent_map_tree *map,
3362 struct extent_io_tree *tree, struct page *page,
3365 struct extent_map *em;
3366 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3367 u64 end = start + PAGE_CACHE_SIZE - 1;
3369 if ((mask & __GFP_WAIT) &&
3370 page->mapping->host->i_size > 16 * 1024 * 1024) {
3372 while (start <= end) {
3373 len = end - start + 1;
3374 write_lock(&map->lock);
3375 em = lookup_extent_mapping(map, start, len);
3376 if (IS_ERR_OR_NULL(em)) {
3377 write_unlock(&map->lock);
3380 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3381 em->start != start) {
3382 write_unlock(&map->lock);
3383 free_extent_map(em);
3386 if (!test_range_bit(tree, em->start,
3387 extent_map_end(em) - 1,
3388 EXTENT_LOCKED | EXTENT_WRITEBACK,
3390 remove_extent_mapping(map, em);
3391 /* once for the rb tree */
3392 free_extent_map(em);
3394 start = extent_map_end(em);
3395 write_unlock(&map->lock);
3398 free_extent_map(em);
3401 return try_release_extent_state(map, tree, page, mask);
3405 * helper function for fiemap, which doesn't want to see any holes.
3406 * This maps until we find something past 'last'
3408 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3411 get_extent_t *get_extent)
3413 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3414 struct extent_map *em;
3421 len = last - offset;
3424 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3425 em = get_extent(inode, NULL, 0, offset, len, 0);
3426 if (IS_ERR_OR_NULL(em))
3429 /* if this isn't a hole return it */
3430 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3431 em->block_start != EXTENT_MAP_HOLE) {
3435 /* this is a hole, advance to the next extent */
3436 offset = extent_map_end(em);
3437 free_extent_map(em);
3444 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3445 __u64 start, __u64 len, get_extent_t *get_extent)
3449 u64 max = start + len;
3453 u64 last_for_get_extent = 0;
3455 u64 isize = i_size_read(inode);
3456 struct btrfs_key found_key;
3457 struct extent_map *em = NULL;
3458 struct extent_state *cached_state = NULL;
3459 struct btrfs_path *path;
3460 struct btrfs_file_extent_item *item;
3465 unsigned long emflags;
3470 path = btrfs_alloc_path();
3473 path->leave_spinning = 1;
3475 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3476 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3479 * lookup the last file extent. We're not using i_size here
3480 * because there might be preallocation past i_size
3482 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3483 path, btrfs_ino(inode), -1, 0);
3485 btrfs_free_path(path);
3490 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3491 struct btrfs_file_extent_item);
3492 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3493 found_type = btrfs_key_type(&found_key);
3495 /* No extents, but there might be delalloc bits */
3496 if (found_key.objectid != btrfs_ino(inode) ||
3497 found_type != BTRFS_EXTENT_DATA_KEY) {
3498 /* have to trust i_size as the end */
3500 last_for_get_extent = isize;
3503 * remember the start of the last extent. There are a
3504 * bunch of different factors that go into the length of the
3505 * extent, so its much less complex to remember where it started
3507 last = found_key.offset;
3508 last_for_get_extent = last + 1;
3510 btrfs_free_path(path);
3513 * we might have some extents allocated but more delalloc past those
3514 * extents. so, we trust isize unless the start of the last extent is
3519 last_for_get_extent = isize;
3522 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3523 &cached_state, GFP_NOFS);
3525 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3535 u64 offset_in_extent;
3537 /* break if the extent we found is outside the range */
3538 if (em->start >= max || extent_map_end(em) < off)
3542 * get_extent may return an extent that starts before our
3543 * requested range. We have to make sure the ranges
3544 * we return to fiemap always move forward and don't
3545 * overlap, so adjust the offsets here
3547 em_start = max(em->start, off);
3550 * record the offset from the start of the extent
3551 * for adjusting the disk offset below
3553 offset_in_extent = em_start - em->start;
3554 em_end = extent_map_end(em);
3555 em_len = em_end - em_start;
3556 emflags = em->flags;
3561 * bump off for our next call to get_extent
3563 off = extent_map_end(em);
3567 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3569 flags |= FIEMAP_EXTENT_LAST;
3570 } else if (em->block_start == EXTENT_MAP_INLINE) {
3571 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3572 FIEMAP_EXTENT_NOT_ALIGNED);
3573 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3574 flags |= (FIEMAP_EXTENT_DELALLOC |
3575 FIEMAP_EXTENT_UNKNOWN);
3577 disko = em->block_start + offset_in_extent;
3579 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3580 flags |= FIEMAP_EXTENT_ENCODED;
3582 free_extent_map(em);
3584 if ((em_start >= last) || em_len == (u64)-1 ||
3585 (last == (u64)-1 && isize <= em_end)) {
3586 flags |= FIEMAP_EXTENT_LAST;
3590 /* now scan forward to see if this is really the last extent. */
3591 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3598 flags |= FIEMAP_EXTENT_LAST;
3601 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3607 free_extent_map(em);
3609 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3610 &cached_state, GFP_NOFS);
3614 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3618 struct address_space *mapping;
3621 return eb->first_page;
3622 i += eb->start >> PAGE_CACHE_SHIFT;
3623 mapping = eb->first_page->mapping;
3628 * extent_buffer_page is only called after pinning the page
3629 * by increasing the reference count. So we know the page must
3630 * be in the radix tree.
3633 p = radix_tree_lookup(&mapping->page_tree, i);
3639 inline unsigned long num_extent_pages(u64 start, u64 len)
3641 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3642 (start >> PAGE_CACHE_SHIFT);
3645 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3650 struct extent_buffer *eb = NULL;
3652 unsigned long flags;
3655 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3660 rwlock_init(&eb->lock);
3661 atomic_set(&eb->write_locks, 0);
3662 atomic_set(&eb->read_locks, 0);
3663 atomic_set(&eb->blocking_readers, 0);
3664 atomic_set(&eb->blocking_writers, 0);
3665 atomic_set(&eb->spinning_readers, 0);
3666 atomic_set(&eb->spinning_writers, 0);
3667 init_waitqueue_head(&eb->write_lock_wq);
3668 init_waitqueue_head(&eb->read_lock_wq);
3671 spin_lock_irqsave(&leak_lock, flags);
3672 list_add(&eb->leak_list, &buffers);
3673 spin_unlock_irqrestore(&leak_lock, flags);
3675 atomic_set(&eb->refs, 1);
3680 static void __free_extent_buffer(struct extent_buffer *eb)
3683 unsigned long flags;
3684 spin_lock_irqsave(&leak_lock, flags);
3685 list_del(&eb->leak_list);
3686 spin_unlock_irqrestore(&leak_lock, flags);
3688 kmem_cache_free(extent_buffer_cache, eb);
3692 * Helper for releasing extent buffer page.
3694 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3695 unsigned long start_idx)
3697 unsigned long index;
3700 if (!eb->first_page)
3703 index = num_extent_pages(eb->start, eb->len);
3704 if (start_idx >= index)
3709 page = extent_buffer_page(eb, index);
3711 page_cache_release(page);
3712 } while (index != start_idx);
3716 * Helper for releasing the extent buffer.
3718 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3720 btrfs_release_extent_buffer_page(eb, 0);
3721 __free_extent_buffer(eb);
3724 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3725 u64 start, unsigned long len,
3728 unsigned long num_pages = num_extent_pages(start, len);
3730 unsigned long index = start >> PAGE_CACHE_SHIFT;
3731 struct extent_buffer *eb;
3732 struct extent_buffer *exists = NULL;
3734 struct address_space *mapping = tree->mapping;
3739 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3740 if (eb && atomic_inc_not_zero(&eb->refs)) {
3742 mark_page_accessed(eb->first_page);
3747 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
3752 eb->first_page = page0;
3755 page_cache_get(page0);
3756 mark_page_accessed(page0);
3757 set_page_extent_mapped(page0);
3758 set_page_extent_head(page0, len);
3759 uptodate = PageUptodate(page0);
3763 for (; i < num_pages; i++, index++) {
3764 p = find_or_create_page(mapping, index, GFP_NOFS);
3769 set_page_extent_mapped(p);
3770 mark_page_accessed(p);
3773 set_page_extent_head(p, len);
3775 set_page_private(p, EXTENT_PAGE_PRIVATE);
3777 if (!PageUptodate(p))
3781 * see below about how we avoid a nasty race with release page
3782 * and why we unlock later
3788 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3790 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3794 spin_lock(&tree->buffer_lock);
3795 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3796 if (ret == -EEXIST) {
3797 exists = radix_tree_lookup(&tree->buffer,
3798 start >> PAGE_CACHE_SHIFT);
3799 /* add one reference for the caller */
3800 atomic_inc(&exists->refs);
3801 spin_unlock(&tree->buffer_lock);
3802 radix_tree_preload_end();
3805 /* add one reference for the tree */
3806 atomic_inc(&eb->refs);
3807 spin_unlock(&tree->buffer_lock);
3808 radix_tree_preload_end();
3811 * there is a race where release page may have
3812 * tried to find this extent buffer in the radix
3813 * but failed. It will tell the VM it is safe to
3814 * reclaim the, and it will clear the page private bit.
3815 * We must make sure to set the page private bit properly
3816 * after the extent buffer is in the radix tree so
3817 * it doesn't get lost
3819 set_page_extent_mapped(eb->first_page);
3820 set_page_extent_head(eb->first_page, eb->len);
3822 unlock_page(eb->first_page);
3826 if (eb->first_page && !page0)
3827 unlock_page(eb->first_page);
3829 if (!atomic_dec_and_test(&eb->refs))
3831 btrfs_release_extent_buffer(eb);
3835 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3836 u64 start, unsigned long len)
3838 struct extent_buffer *eb;
3841 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3842 if (eb && atomic_inc_not_zero(&eb->refs)) {
3844 mark_page_accessed(eb->first_page);
3852 void free_extent_buffer(struct extent_buffer *eb)
3857 if (!atomic_dec_and_test(&eb->refs))
3863 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3864 struct extent_buffer *eb)
3867 unsigned long num_pages;
3870 num_pages = num_extent_pages(eb->start, eb->len);
3872 for (i = 0; i < num_pages; i++) {
3873 page = extent_buffer_page(eb, i);
3874 if (!PageDirty(page))
3878 WARN_ON(!PagePrivate(page));
3880 set_page_extent_mapped(page);
3882 set_page_extent_head(page, eb->len);
3884 clear_page_dirty_for_io(page);
3885 spin_lock_irq(&page->mapping->tree_lock);
3886 if (!PageDirty(page)) {
3887 radix_tree_tag_clear(&page->mapping->page_tree,
3889 PAGECACHE_TAG_DIRTY);
3891 spin_unlock_irq(&page->mapping->tree_lock);
3892 ClearPageError(page);
3898 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3899 struct extent_buffer *eb)
3902 unsigned long num_pages;
3905 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3906 num_pages = num_extent_pages(eb->start, eb->len);
3907 for (i = 0; i < num_pages; i++)
3908 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3912 static int __eb_straddles_pages(u64 start, u64 len)
3914 if (len < PAGE_CACHE_SIZE)
3916 if (start & (PAGE_CACHE_SIZE - 1))
3918 if ((start + len) & (PAGE_CACHE_SIZE - 1))
3923 static int eb_straddles_pages(struct extent_buffer *eb)
3925 return __eb_straddles_pages(eb->start, eb->len);
3928 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3929 struct extent_buffer *eb,
3930 struct extent_state **cached_state)
3934 unsigned long num_pages;
3936 num_pages = num_extent_pages(eb->start, eb->len);
3937 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3939 if (eb_straddles_pages(eb)) {
3940 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3941 cached_state, GFP_NOFS);
3943 for (i = 0; i < num_pages; i++) {
3944 page = extent_buffer_page(eb, i);
3946 ClearPageUptodate(page);
3951 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3952 struct extent_buffer *eb)
3956 unsigned long num_pages;
3958 num_pages = num_extent_pages(eb->start, eb->len);
3960 if (eb_straddles_pages(eb)) {
3961 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3964 for (i = 0; i < num_pages; i++) {
3965 page = extent_buffer_page(eb, i);
3966 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3967 ((i == num_pages - 1) &&
3968 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3969 check_page_uptodate(tree, page);
3972 SetPageUptodate(page);
3977 int extent_range_uptodate(struct extent_io_tree *tree,
3982 int pg_uptodate = 1;
3984 unsigned long index;
3986 if (__eb_straddles_pages(start, end - start + 1)) {
3987 ret = test_range_bit(tree, start, end,
3988 EXTENT_UPTODATE, 1, NULL);
3992 while (start <= end) {
3993 index = start >> PAGE_CACHE_SHIFT;
3994 page = find_get_page(tree->mapping, index);
3995 uptodate = PageUptodate(page);
3996 page_cache_release(page);
4001 start += PAGE_CACHE_SIZE;
4006 int extent_buffer_uptodate(struct extent_io_tree *tree,
4007 struct extent_buffer *eb,
4008 struct extent_state *cached_state)
4011 unsigned long num_pages;
4014 int pg_uptodate = 1;
4016 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4019 if (eb_straddles_pages(eb)) {
4020 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
4021 EXTENT_UPTODATE, 1, cached_state);
4026 num_pages = num_extent_pages(eb->start, eb->len);
4027 for (i = 0; i < num_pages; i++) {
4028 page = extent_buffer_page(eb, i);
4029 if (!PageUptodate(page)) {
4037 int read_extent_buffer_pages(struct extent_io_tree *tree,
4038 struct extent_buffer *eb, u64 start, int wait,
4039 get_extent_t *get_extent, int mirror_num)
4042 unsigned long start_i;
4046 int locked_pages = 0;
4047 int all_uptodate = 1;
4048 int inc_all_pages = 0;
4049 unsigned long num_pages;
4050 struct bio *bio = NULL;
4051 unsigned long bio_flags = 0;
4052 u64 prev_em_start = (u64)-1;
4054 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4057 if (eb_straddles_pages(eb)) {
4058 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
4059 EXTENT_UPTODATE, 1, NULL)) {
4065 WARN_ON(start < eb->start);
4066 start_i = (start >> PAGE_CACHE_SHIFT) -
4067 (eb->start >> PAGE_CACHE_SHIFT);
4072 num_pages = num_extent_pages(eb->start, eb->len);
4073 for (i = start_i; i < num_pages; i++) {
4074 page = extent_buffer_page(eb, i);
4075 if (wait == WAIT_NONE) {
4076 if (!trylock_page(page))
4082 if (!PageUptodate(page))
4087 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4091 for (i = start_i; i < num_pages; i++) {
4092 page = extent_buffer_page(eb, i);
4094 WARN_ON(!PagePrivate(page));
4096 set_page_extent_mapped(page);
4098 set_page_extent_head(page, eb->len);
4101 page_cache_get(page);
4102 if (!PageUptodate(page)) {
4105 ClearPageError(page);
4106 err = __extent_read_full_page(tree, page,
4108 mirror_num, &bio_flags,
4118 submit_one_bio(READ, bio, mirror_num, bio_flags);
4120 if (ret || wait != WAIT_COMPLETE)
4123 for (i = start_i; i < num_pages; i++) {
4124 page = extent_buffer_page(eb, i);
4125 wait_on_page_locked(page);
4126 if (!PageUptodate(page))
4131 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4136 while (locked_pages > 0) {
4137 page = extent_buffer_page(eb, i);
4145 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4146 unsigned long start,
4153 char *dst = (char *)dstv;
4154 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4155 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4157 WARN_ON(start > eb->len);
4158 WARN_ON(start + len > eb->start + eb->len);
4160 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4163 page = extent_buffer_page(eb, i);
4165 cur = min(len, (PAGE_CACHE_SIZE - offset));
4166 kaddr = page_address(page);
4167 memcpy(dst, kaddr + offset, cur);
4176 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4177 unsigned long min_len, char **map,
4178 unsigned long *map_start,
4179 unsigned long *map_len)
4181 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4184 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4185 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4186 unsigned long end_i = (start_offset + start + min_len - 1) >>
4193 offset = start_offset;
4197 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4200 if (start + min_len > eb->len) {
4201 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4202 "wanted %lu %lu\n", (unsigned long long)eb->start,
4203 eb->len, start, min_len);
4208 p = extent_buffer_page(eb, i);
4209 kaddr = page_address(p);
4210 *map = kaddr + offset;
4211 *map_len = PAGE_CACHE_SIZE - offset;
4215 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4216 unsigned long start,
4223 char *ptr = (char *)ptrv;
4224 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4225 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4228 WARN_ON(start > eb->len);
4229 WARN_ON(start + len > eb->start + eb->len);
4231 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4234 page = extent_buffer_page(eb, i);
4236 cur = min(len, (PAGE_CACHE_SIZE - offset));
4238 kaddr = page_address(page);
4239 ret = memcmp(ptr, kaddr + offset, cur);
4251 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4252 unsigned long start, unsigned long len)
4258 char *src = (char *)srcv;
4259 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4260 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4262 WARN_ON(start > eb->len);
4263 WARN_ON(start + len > eb->start + eb->len);
4265 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4268 page = extent_buffer_page(eb, i);
4269 WARN_ON(!PageUptodate(page));
4271 cur = min(len, PAGE_CACHE_SIZE - offset);
4272 kaddr = page_address(page);
4273 memcpy(kaddr + offset, src, cur);
4282 void memset_extent_buffer(struct extent_buffer *eb, char c,
4283 unsigned long start, unsigned long len)
4289 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4290 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4292 WARN_ON(start > eb->len);
4293 WARN_ON(start + len > eb->start + eb->len);
4295 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4298 page = extent_buffer_page(eb, i);
4299 WARN_ON(!PageUptodate(page));
4301 cur = min(len, PAGE_CACHE_SIZE - offset);
4302 kaddr = page_address(page);
4303 memset(kaddr + offset, c, cur);
4311 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4312 unsigned long dst_offset, unsigned long src_offset,
4315 u64 dst_len = dst->len;
4320 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4321 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4323 WARN_ON(src->len != dst_len);
4325 offset = (start_offset + dst_offset) &
4326 ((unsigned long)PAGE_CACHE_SIZE - 1);
4329 page = extent_buffer_page(dst, i);
4330 WARN_ON(!PageUptodate(page));
4332 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4334 kaddr = page_address(page);
4335 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4344 static void move_pages(struct page *dst_page, struct page *src_page,
4345 unsigned long dst_off, unsigned long src_off,
4348 char *dst_kaddr = page_address(dst_page);
4349 if (dst_page == src_page) {
4350 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4352 char *src_kaddr = page_address(src_page);
4353 char *p = dst_kaddr + dst_off + len;
4354 char *s = src_kaddr + src_off + len;
4361 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4363 unsigned long distance = (src > dst) ? src - dst : dst - src;
4364 return distance < len;
4367 static void copy_pages(struct page *dst_page, struct page *src_page,
4368 unsigned long dst_off, unsigned long src_off,
4371 char *dst_kaddr = page_address(dst_page);
4374 if (dst_page != src_page) {
4375 src_kaddr = page_address(src_page);
4377 src_kaddr = dst_kaddr;
4378 BUG_ON(areas_overlap(src_off, dst_off, len));
4381 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4384 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4385 unsigned long src_offset, unsigned long len)
4388 size_t dst_off_in_page;
4389 size_t src_off_in_page;
4390 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4391 unsigned long dst_i;
4392 unsigned long src_i;
4394 if (src_offset + len > dst->len) {
4395 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4396 "len %lu dst len %lu\n", src_offset, len, dst->len);
4399 if (dst_offset + len > dst->len) {
4400 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4401 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4406 dst_off_in_page = (start_offset + dst_offset) &
4407 ((unsigned long)PAGE_CACHE_SIZE - 1);
4408 src_off_in_page = (start_offset + src_offset) &
4409 ((unsigned long)PAGE_CACHE_SIZE - 1);
4411 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4412 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4414 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4416 cur = min_t(unsigned long, cur,
4417 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4419 copy_pages(extent_buffer_page(dst, dst_i),
4420 extent_buffer_page(dst, src_i),
4421 dst_off_in_page, src_off_in_page, cur);
4429 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4430 unsigned long src_offset, unsigned long len)
4433 size_t dst_off_in_page;
4434 size_t src_off_in_page;
4435 unsigned long dst_end = dst_offset + len - 1;
4436 unsigned long src_end = src_offset + len - 1;
4437 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4438 unsigned long dst_i;
4439 unsigned long src_i;
4441 if (src_offset + len > dst->len) {
4442 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4443 "len %lu len %lu\n", src_offset, len, dst->len);
4446 if (dst_offset + len > dst->len) {
4447 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4448 "len %lu len %lu\n", dst_offset, len, dst->len);
4451 if (!areas_overlap(src_offset, dst_offset, len)) {
4452 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4456 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4457 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4459 dst_off_in_page = (start_offset + dst_end) &
4460 ((unsigned long)PAGE_CACHE_SIZE - 1);
4461 src_off_in_page = (start_offset + src_end) &
4462 ((unsigned long)PAGE_CACHE_SIZE - 1);
4464 cur = min_t(unsigned long, len, src_off_in_page + 1);
4465 cur = min(cur, dst_off_in_page + 1);
4466 move_pages(extent_buffer_page(dst, dst_i),
4467 extent_buffer_page(dst, src_i),
4468 dst_off_in_page - cur + 1,
4469 src_off_in_page - cur + 1, cur);
4477 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4479 struct extent_buffer *eb =
4480 container_of(head, struct extent_buffer, rcu_head);
4482 btrfs_release_extent_buffer(eb);
4485 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
4487 u64 start = page_offset(page);
4488 struct extent_buffer *eb;
4491 spin_lock(&tree->buffer_lock);
4492 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4494 spin_unlock(&tree->buffer_lock);
4498 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4504 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4507 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
4512 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4514 spin_unlock(&tree->buffer_lock);
4516 /* at this point we can safely release the extent buffer */
4517 if (atomic_read(&eb->refs) == 0)
4518 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);