1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 static struct kmem_cache *extent_state_cache;
22 static struct kmem_cache *extent_buffer_cache;
24 static LIST_HEAD(buffers);
25 static LIST_HEAD(states);
29 static DEFINE_SPINLOCK(leak_lock);
32 #define BUFFER_LRU_MAX 64
37 struct rb_node rb_node;
40 struct extent_page_data {
42 struct extent_io_tree *tree;
43 get_extent_t *get_extent;
45 /* tells writepage not to lock the state bits for this range
46 * it still does the unlocking
48 unsigned int extent_locked:1;
50 /* tells the submit_bio code to use a WRITE_SYNC */
51 unsigned int sync_io:1;
54 int __init extent_io_init(void)
56 extent_state_cache = kmem_cache_create("extent_state",
57 sizeof(struct extent_state), 0,
58 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
59 if (!extent_state_cache)
62 extent_buffer_cache = kmem_cache_create("extent_buffers",
63 sizeof(struct extent_buffer), 0,
64 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
65 if (!extent_buffer_cache)
66 goto free_state_cache;
70 kmem_cache_destroy(extent_state_cache);
74 void extent_io_exit(void)
76 struct extent_state *state;
77 struct extent_buffer *eb;
79 while (!list_empty(&states)) {
80 state = list_entry(states.next, struct extent_state, leak_list);
81 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
82 "state %lu in tree %p refs %d\n",
83 (unsigned long long)state->start,
84 (unsigned long long)state->end,
85 state->state, state->tree, atomic_read(&state->refs));
86 list_del(&state->leak_list);
87 kmem_cache_free(extent_state_cache, state);
91 while (!list_empty(&buffers)) {
92 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
93 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
94 "refs %d\n", (unsigned long long)eb->start,
95 eb->len, atomic_read(&eb->refs));
96 list_del(&eb->leak_list);
97 kmem_cache_free(extent_buffer_cache, eb);
99 if (extent_state_cache)
100 kmem_cache_destroy(extent_state_cache);
101 if (extent_buffer_cache)
102 kmem_cache_destroy(extent_buffer_cache);
105 void extent_io_tree_init(struct extent_io_tree *tree,
106 struct address_space *mapping)
108 tree->state = RB_ROOT;
109 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
111 tree->dirty_bytes = 0;
112 spin_lock_init(&tree->lock);
113 spin_lock_init(&tree->buffer_lock);
114 tree->mapping = mapping;
117 static struct extent_state *alloc_extent_state(gfp_t mask)
119 struct extent_state *state;
124 state = kmem_cache_alloc(extent_state_cache, mask);
131 spin_lock_irqsave(&leak_lock, flags);
132 list_add(&state->leak_list, &states);
133 spin_unlock_irqrestore(&leak_lock, flags);
135 atomic_set(&state->refs, 1);
136 init_waitqueue_head(&state->wq);
140 void free_extent_state(struct extent_state *state)
144 if (atomic_dec_and_test(&state->refs)) {
148 WARN_ON(state->tree);
150 spin_lock_irqsave(&leak_lock, flags);
151 list_del(&state->leak_list);
152 spin_unlock_irqrestore(&leak_lock, flags);
154 kmem_cache_free(extent_state_cache, state);
158 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
159 struct rb_node *node)
161 struct rb_node **p = &root->rb_node;
162 struct rb_node *parent = NULL;
163 struct tree_entry *entry;
167 entry = rb_entry(parent, struct tree_entry, rb_node);
169 if (offset < entry->start)
171 else if (offset > entry->end)
177 entry = rb_entry(node, struct tree_entry, rb_node);
178 rb_link_node(node, parent, p);
179 rb_insert_color(node, root);
183 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
184 struct rb_node **prev_ret,
185 struct rb_node **next_ret)
187 struct rb_root *root = &tree->state;
188 struct rb_node *n = root->rb_node;
189 struct rb_node *prev = NULL;
190 struct rb_node *orig_prev = NULL;
191 struct tree_entry *entry;
192 struct tree_entry *prev_entry = NULL;
195 entry = rb_entry(n, struct tree_entry, rb_node);
199 if (offset < entry->start)
201 else if (offset > entry->end)
209 while (prev && offset > prev_entry->end) {
210 prev = rb_next(prev);
211 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
218 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
219 while (prev && offset < prev_entry->start) {
220 prev = rb_prev(prev);
221 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
228 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
231 struct rb_node *prev = NULL;
234 ret = __etree_search(tree, offset, &prev, NULL);
240 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
241 struct extent_state *other)
243 if (tree->ops && tree->ops->merge_extent_hook)
244 tree->ops->merge_extent_hook(tree->mapping->host, new,
249 * utility function to look for merge candidates inside a given range.
250 * Any extents with matching state are merged together into a single
251 * extent in the tree. Extents with EXTENT_IO in their state field
252 * are not merged because the end_io handlers need to be able to do
253 * operations on them without sleeping (or doing allocations/splits).
255 * This should be called with the tree lock held.
257 static int merge_state(struct extent_io_tree *tree,
258 struct extent_state *state)
260 struct extent_state *other;
261 struct rb_node *other_node;
263 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
266 other_node = rb_prev(&state->rb_node);
268 other = rb_entry(other_node, struct extent_state, rb_node);
269 if (other->end == state->start - 1 &&
270 other->state == state->state) {
271 merge_cb(tree, state, other);
272 state->start = other->start;
274 rb_erase(&other->rb_node, &tree->state);
275 free_extent_state(other);
278 other_node = rb_next(&state->rb_node);
280 other = rb_entry(other_node, struct extent_state, rb_node);
281 if (other->start == state->end + 1 &&
282 other->state == state->state) {
283 merge_cb(tree, state, other);
284 other->start = state->start;
286 rb_erase(&state->rb_node, &tree->state);
287 free_extent_state(state);
295 static int set_state_cb(struct extent_io_tree *tree,
296 struct extent_state *state, int *bits)
298 if (tree->ops && tree->ops->set_bit_hook) {
299 return tree->ops->set_bit_hook(tree->mapping->host,
306 static void clear_state_cb(struct extent_io_tree *tree,
307 struct extent_state *state, int *bits)
309 if (tree->ops && tree->ops->clear_bit_hook)
310 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
314 * insert an extent_state struct into the tree. 'bits' are set on the
315 * struct before it is inserted.
317 * This may return -EEXIST if the extent is already there, in which case the
318 * state struct is freed.
320 * The tree lock is not taken internally. This is a utility function and
321 * probably isn't what you want to call (see set/clear_extent_bit).
323 static int insert_state(struct extent_io_tree *tree,
324 struct extent_state *state, u64 start, u64 end,
327 struct rb_node *node;
328 int bits_to_set = *bits & ~EXTENT_CTLBITS;
332 printk(KERN_ERR "btrfs end < start %llu %llu\n",
333 (unsigned long long)end,
334 (unsigned long long)start);
337 state->start = start;
339 ret = set_state_cb(tree, state, bits);
343 if (bits_to_set & EXTENT_DIRTY)
344 tree->dirty_bytes += end - start + 1;
345 state->state |= bits_to_set;
346 node = tree_insert(&tree->state, end, &state->rb_node);
348 struct extent_state *found;
349 found = rb_entry(node, struct extent_state, rb_node);
350 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
351 "%llu %llu\n", (unsigned long long)found->start,
352 (unsigned long long)found->end,
353 (unsigned long long)start, (unsigned long long)end);
354 free_extent_state(state);
358 merge_state(tree, state);
362 static int split_cb(struct extent_io_tree *tree, struct extent_state *orig,
365 if (tree->ops && tree->ops->split_extent_hook)
366 return tree->ops->split_extent_hook(tree->mapping->host,
372 * split a given extent state struct in two, inserting the preallocated
373 * struct 'prealloc' as the newly created second half. 'split' indicates an
374 * offset inside 'orig' where it should be split.
377 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
378 * are two extent state structs in the tree:
379 * prealloc: [orig->start, split - 1]
380 * orig: [ split, orig->end ]
382 * The tree locks are not taken by this function. They need to be held
385 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
386 struct extent_state *prealloc, u64 split)
388 struct rb_node *node;
390 split_cb(tree, orig, split);
392 prealloc->start = orig->start;
393 prealloc->end = split - 1;
394 prealloc->state = orig->state;
397 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
399 free_extent_state(prealloc);
402 prealloc->tree = tree;
407 * utility function to clear some bits in an extent state struct.
408 * it will optionally wake up any one waiting on this state (wake == 1), or
409 * forcibly remove the state from the tree (delete == 1).
411 * If no bits are set on the state struct after clearing things, the
412 * struct is freed and removed from the tree
414 static int clear_state_bit(struct extent_io_tree *tree,
415 struct extent_state *state,
418 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
419 int ret = state->state & bits_to_clear;
421 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
422 u64 range = state->end - state->start + 1;
423 WARN_ON(range > tree->dirty_bytes);
424 tree->dirty_bytes -= range;
426 clear_state_cb(tree, state, bits);
427 state->state &= ~bits_to_clear;
430 if (state->state == 0) {
432 rb_erase(&state->rb_node, &tree->state);
434 free_extent_state(state);
439 merge_state(tree, state);
444 static struct extent_state *
445 alloc_extent_state_atomic(struct extent_state *prealloc)
448 prealloc = alloc_extent_state(GFP_ATOMIC);
454 * clear some bits on a range in the tree. This may require splitting
455 * or inserting elements in the tree, so the gfp mask is used to
456 * indicate which allocations or sleeping are allowed.
458 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
459 * the given range from the tree regardless of state (ie for truncate).
461 * the range [start, end] is inclusive.
463 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
464 * bits were already set, or zero if none of the bits were already set.
466 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
467 int bits, int wake, int delete,
468 struct extent_state **cached_state,
471 struct extent_state *state;
472 struct extent_state *cached;
473 struct extent_state *prealloc = NULL;
474 struct rb_node *next_node;
475 struct rb_node *node;
482 bits |= ~EXTENT_CTLBITS;
483 bits |= EXTENT_FIRST_DELALLOC;
485 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
488 if (!prealloc && (mask & __GFP_WAIT)) {
489 prealloc = alloc_extent_state(mask);
494 spin_lock(&tree->lock);
496 cached = *cached_state;
499 *cached_state = NULL;
503 if (cached && cached->tree && cached->start == start) {
505 atomic_dec(&cached->refs);
510 free_extent_state(cached);
513 * this search will find the extents that end after
516 node = tree_search(tree, start);
519 state = rb_entry(node, struct extent_state, rb_node);
521 if (state->start > end)
523 WARN_ON(state->end < start);
524 last_end = state->end;
527 * | ---- desired range ---- |
529 * | ------------- state -------------- |
531 * We need to split the extent we found, and may flip
532 * bits on second half.
534 * If the extent we found extends past our range, we
535 * just split and search again. It'll get split again
536 * the next time though.
538 * If the extent we found is inside our range, we clear
539 * the desired bit on it.
542 if (state->start < start) {
543 prealloc = alloc_extent_state_atomic(prealloc);
545 err = split_state(tree, state, prealloc, start);
546 BUG_ON(err == -EEXIST);
550 if (state->end <= end) {
551 set |= clear_state_bit(tree, state, &bits, wake);
552 if (last_end == (u64)-1)
554 start = last_end + 1;
559 * | ---- desired range ---- |
561 * We need to split the extent, and clear the bit
564 if (state->start <= end && state->end > end) {
565 prealloc = alloc_extent_state_atomic(prealloc);
567 err = split_state(tree, state, prealloc, end + 1);
568 BUG_ON(err == -EEXIST);
572 set |= clear_state_bit(tree, prealloc, &bits, wake);
578 if (state->end < end && prealloc && !need_resched())
579 next_node = rb_next(&state->rb_node);
583 set |= clear_state_bit(tree, state, &bits, wake);
584 if (last_end == (u64)-1)
586 start = last_end + 1;
587 if (start <= end && next_node) {
588 state = rb_entry(next_node, struct extent_state,
590 if (state->start == start)
596 spin_unlock(&tree->lock);
598 free_extent_state(prealloc);
605 spin_unlock(&tree->lock);
606 if (mask & __GFP_WAIT)
611 static int wait_on_state(struct extent_io_tree *tree,
612 struct extent_state *state)
613 __releases(tree->lock)
614 __acquires(tree->lock)
617 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
618 spin_unlock(&tree->lock);
620 spin_lock(&tree->lock);
621 finish_wait(&state->wq, &wait);
626 * waits for one or more bits to clear on a range in the state tree.
627 * The range [start, end] is inclusive.
628 * The tree lock is taken by this function
630 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
632 struct extent_state *state;
633 struct rb_node *node;
635 spin_lock(&tree->lock);
639 * this search will find all the extents that end after
642 node = tree_search(tree, start);
646 state = rb_entry(node, struct extent_state, rb_node);
648 if (state->start > end)
651 if (state->state & bits) {
652 start = state->start;
653 atomic_inc(&state->refs);
654 wait_on_state(tree, state);
655 free_extent_state(state);
658 start = state->end + 1;
663 if (need_resched()) {
664 spin_unlock(&tree->lock);
666 spin_lock(&tree->lock);
670 spin_unlock(&tree->lock);
674 static int set_state_bits(struct extent_io_tree *tree,
675 struct extent_state *state,
679 int bits_to_set = *bits & ~EXTENT_CTLBITS;
681 ret = set_state_cb(tree, state, bits);
684 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
685 u64 range = state->end - state->start + 1;
686 tree->dirty_bytes += range;
688 state->state |= bits_to_set;
693 static void cache_state(struct extent_state *state,
694 struct extent_state **cached_ptr)
696 if (cached_ptr && !(*cached_ptr)) {
697 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
699 atomic_inc(&state->refs);
704 static void uncache_state(struct extent_state **cached_ptr)
706 if (cached_ptr && (*cached_ptr)) {
707 struct extent_state *state = *cached_ptr;
709 free_extent_state(state);
714 * set some bits on a range in the tree. This may require allocations or
715 * sleeping, so the gfp mask is used to indicate what is allowed.
717 * If any of the exclusive bits are set, this will fail with -EEXIST if some
718 * part of the range already has the desired bits set. The start of the
719 * existing range is returned in failed_start in this case.
721 * [start, end] is inclusive This takes the tree lock.
724 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
725 int bits, int exclusive_bits, u64 *failed_start,
726 struct extent_state **cached_state, gfp_t mask)
728 struct extent_state *state;
729 struct extent_state *prealloc = NULL;
730 struct rb_node *node;
735 bits |= EXTENT_FIRST_DELALLOC;
737 if (!prealloc && (mask & __GFP_WAIT)) {
738 prealloc = alloc_extent_state(mask);
742 spin_lock(&tree->lock);
743 if (cached_state && *cached_state) {
744 state = *cached_state;
745 if (state->start == start && state->tree) {
746 node = &state->rb_node;
751 * this search will find all the extents that end after
754 node = tree_search(tree, start);
756 prealloc = alloc_extent_state_atomic(prealloc);
758 err = insert_state(tree, prealloc, start, end, &bits);
760 BUG_ON(err == -EEXIST);
763 state = rb_entry(node, struct extent_state, rb_node);
765 last_start = state->start;
766 last_end = state->end;
769 * | ---- desired range ---- |
772 * Just lock what we found and keep going
774 if (state->start == start && state->end <= end) {
775 struct rb_node *next_node;
776 if (state->state & exclusive_bits) {
777 *failed_start = state->start;
782 err = set_state_bits(tree, state, &bits);
786 next_node = rb_next(node);
787 cache_state(state, cached_state);
788 merge_state(tree, state);
789 if (last_end == (u64)-1)
792 start = last_end + 1;
793 if (next_node && start < end && prealloc && !need_resched()) {
794 state = rb_entry(next_node, struct extent_state,
796 if (state->start == start)
803 * | ---- desired range ---- |
806 * | ------------- state -------------- |
808 * We need to split the extent we found, and may flip bits on
811 * If the extent we found extends past our
812 * range, we just split and search again. It'll get split
813 * again the next time though.
815 * If the extent we found is inside our range, we set the
818 if (state->start < start) {
819 if (state->state & exclusive_bits) {
820 *failed_start = start;
825 prealloc = alloc_extent_state_atomic(prealloc);
827 err = split_state(tree, state, prealloc, start);
828 BUG_ON(err == -EEXIST);
832 if (state->end <= end) {
833 err = set_state_bits(tree, state, &bits);
836 cache_state(state, cached_state);
837 merge_state(tree, state);
838 if (last_end == (u64)-1)
840 start = last_end + 1;
845 * | ---- desired range ---- |
846 * | state | or | state |
848 * There's a hole, we need to insert something in it and
849 * ignore the extent we found.
851 if (state->start > start) {
853 if (end < last_start)
856 this_end = last_start - 1;
858 prealloc = alloc_extent_state_atomic(prealloc);
862 * Avoid to free 'prealloc' if it can be merged with
865 atomic_inc(&prealloc->refs);
866 err = insert_state(tree, prealloc, start, this_end,
868 BUG_ON(err == -EEXIST);
870 free_extent_state(prealloc);
874 cache_state(prealloc, cached_state);
875 free_extent_state(prealloc);
877 start = this_end + 1;
881 * | ---- desired range ---- |
883 * We need to split the extent, and set the bit
886 if (state->start <= end && state->end > end) {
887 if (state->state & exclusive_bits) {
888 *failed_start = start;
893 prealloc = alloc_extent_state_atomic(prealloc);
895 err = split_state(tree, state, prealloc, end + 1);
896 BUG_ON(err == -EEXIST);
898 err = set_state_bits(tree, prealloc, &bits);
903 cache_state(prealloc, cached_state);
904 merge_state(tree, prealloc);
912 spin_unlock(&tree->lock);
914 free_extent_state(prealloc);
921 spin_unlock(&tree->lock);
922 if (mask & __GFP_WAIT)
927 /* wrappers around set/clear extent bit */
928 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
931 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
935 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
936 int bits, gfp_t mask)
938 return set_extent_bit(tree, start, end, bits, 0, NULL,
942 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
943 int bits, gfp_t mask)
945 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
948 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
949 struct extent_state **cached_state, gfp_t mask)
951 return set_extent_bit(tree, start, end,
952 EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
953 0, NULL, cached_state, mask);
956 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
959 return clear_extent_bit(tree, start, end,
960 EXTENT_DIRTY | EXTENT_DELALLOC |
961 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
964 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
967 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
971 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
972 struct extent_state **cached_state, gfp_t mask)
974 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
975 NULL, cached_state, mask);
978 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
979 u64 end, struct extent_state **cached_state,
982 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
987 * either insert or lock state struct between start and end use mask to tell
988 * us if waiting is desired.
990 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
991 int bits, struct extent_state **cached_state, gfp_t mask)
996 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
997 EXTENT_LOCKED, &failed_start,
999 if (err == -EEXIST && (mask & __GFP_WAIT)) {
1000 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1001 start = failed_start;
1005 WARN_ON(start > end);
1010 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1012 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1015 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1021 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1022 &failed_start, NULL, mask);
1023 if (err == -EEXIST) {
1024 if (failed_start > start)
1025 clear_extent_bit(tree, start, failed_start - 1,
1026 EXTENT_LOCKED, 1, 0, NULL, mask);
1032 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1033 struct extent_state **cached, gfp_t mask)
1035 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1039 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1041 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1046 * helper function to set both pages and extents in the tree writeback
1048 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1050 unsigned long index = start >> PAGE_CACHE_SHIFT;
1051 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1054 while (index <= end_index) {
1055 page = find_get_page(tree->mapping, index);
1057 set_page_writeback(page);
1058 page_cache_release(page);
1065 * find the first offset in the io tree with 'bits' set. zero is
1066 * returned if we find something, and *start_ret and *end_ret are
1067 * set to reflect the state struct that was found.
1069 * If nothing was found, 1 is returned, < 0 on error
1071 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1072 u64 *start_ret, u64 *end_ret, int bits)
1074 struct rb_node *node;
1075 struct extent_state *state;
1078 spin_lock(&tree->lock);
1080 * this search will find all the extents that end after
1083 node = tree_search(tree, start);
1088 state = rb_entry(node, struct extent_state, rb_node);
1089 if (state->end >= start && (state->state & bits)) {
1090 *start_ret = state->start;
1091 *end_ret = state->end;
1095 node = rb_next(node);
1100 spin_unlock(&tree->lock);
1104 /* find the first state struct with 'bits' set after 'start', and
1105 * return it. tree->lock must be held. NULL will returned if
1106 * nothing was found after 'start'
1108 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1109 u64 start, int bits)
1111 struct rb_node *node;
1112 struct extent_state *state;
1115 * this search will find all the extents that end after
1118 node = tree_search(tree, start);
1123 state = rb_entry(node, struct extent_state, rb_node);
1124 if (state->end >= start && (state->state & bits))
1127 node = rb_next(node);
1136 * find a contiguous range of bytes in the file marked as delalloc, not
1137 * more than 'max_bytes'. start and end are used to return the range,
1139 * 1 is returned if we find something, 0 if nothing was in the tree
1141 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1142 u64 *start, u64 *end, u64 max_bytes,
1143 struct extent_state **cached_state)
1145 struct rb_node *node;
1146 struct extent_state *state;
1147 u64 cur_start = *start;
1149 u64 total_bytes = 0;
1151 spin_lock(&tree->lock);
1154 * this search will find all the extents that end after
1157 node = tree_search(tree, cur_start);
1165 state = rb_entry(node, struct extent_state, rb_node);
1166 if (found && (state->start != cur_start ||
1167 (state->state & EXTENT_BOUNDARY))) {
1170 if (!(state->state & EXTENT_DELALLOC)) {
1176 *start = state->start;
1177 *cached_state = state;
1178 atomic_inc(&state->refs);
1182 cur_start = state->end + 1;
1183 node = rb_next(node);
1186 total_bytes += state->end - state->start + 1;
1187 if (total_bytes >= max_bytes)
1191 spin_unlock(&tree->lock);
1195 static noinline int __unlock_for_delalloc(struct inode *inode,
1196 struct page *locked_page,
1200 struct page *pages[16];
1201 unsigned long index = start >> PAGE_CACHE_SHIFT;
1202 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1203 unsigned long nr_pages = end_index - index + 1;
1206 if (index == locked_page->index && end_index == index)
1209 while (nr_pages > 0) {
1210 ret = find_get_pages_contig(inode->i_mapping, index,
1211 min_t(unsigned long, nr_pages,
1212 ARRAY_SIZE(pages)), pages);
1213 for (i = 0; i < ret; i++) {
1214 if (pages[i] != locked_page)
1215 unlock_page(pages[i]);
1216 page_cache_release(pages[i]);
1225 static noinline int lock_delalloc_pages(struct inode *inode,
1226 struct page *locked_page,
1230 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1231 unsigned long start_index = index;
1232 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1233 unsigned long pages_locked = 0;
1234 struct page *pages[16];
1235 unsigned long nrpages;
1239 /* the caller is responsible for locking the start index */
1240 if (index == locked_page->index && index == end_index)
1243 /* skip the page at the start index */
1244 nrpages = end_index - index + 1;
1245 while (nrpages > 0) {
1246 ret = find_get_pages_contig(inode->i_mapping, index,
1247 min_t(unsigned long,
1248 nrpages, ARRAY_SIZE(pages)), pages);
1253 /* now we have an array of pages, lock them all */
1254 for (i = 0; i < ret; i++) {
1256 * the caller is taking responsibility for
1259 if (pages[i] != locked_page) {
1260 lock_page(pages[i]);
1261 if (!PageDirty(pages[i]) ||
1262 pages[i]->mapping != inode->i_mapping) {
1264 unlock_page(pages[i]);
1265 page_cache_release(pages[i]);
1269 page_cache_release(pages[i]);
1278 if (ret && pages_locked) {
1279 __unlock_for_delalloc(inode, locked_page,
1281 ((u64)(start_index + pages_locked - 1)) <<
1288 * find a contiguous range of bytes in the file marked as delalloc, not
1289 * more than 'max_bytes'. start and end are used to return the range,
1291 * 1 is returned if we find something, 0 if nothing was in the tree
1293 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1294 struct extent_io_tree *tree,
1295 struct page *locked_page,
1296 u64 *start, u64 *end,
1302 struct extent_state *cached_state = NULL;
1307 /* step one, find a bunch of delalloc bytes starting at start */
1308 delalloc_start = *start;
1310 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1311 max_bytes, &cached_state);
1312 if (!found || delalloc_end <= *start) {
1313 *start = delalloc_start;
1314 *end = delalloc_end;
1315 free_extent_state(cached_state);
1320 * start comes from the offset of locked_page. We have to lock
1321 * pages in order, so we can't process delalloc bytes before
1324 if (delalloc_start < *start)
1325 delalloc_start = *start;
1328 * make sure to limit the number of pages we try to lock down
1331 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1332 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1334 /* step two, lock all the pages after the page that has start */
1335 ret = lock_delalloc_pages(inode, locked_page,
1336 delalloc_start, delalloc_end);
1337 if (ret == -EAGAIN) {
1338 /* some of the pages are gone, lets avoid looping by
1339 * shortening the size of the delalloc range we're searching
1341 free_extent_state(cached_state);
1343 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1344 max_bytes = PAGE_CACHE_SIZE - offset;
1354 /* step three, lock the state bits for the whole range */
1355 lock_extent_bits(tree, delalloc_start, delalloc_end,
1356 0, &cached_state, GFP_NOFS);
1358 /* then test to make sure it is all still delalloc */
1359 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1360 EXTENT_DELALLOC, 1, cached_state);
1362 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1363 &cached_state, GFP_NOFS);
1364 __unlock_for_delalloc(inode, locked_page,
1365 delalloc_start, delalloc_end);
1369 free_extent_state(cached_state);
1370 *start = delalloc_start;
1371 *end = delalloc_end;
1376 int extent_clear_unlock_delalloc(struct inode *inode,
1377 struct extent_io_tree *tree,
1378 u64 start, u64 end, struct page *locked_page,
1382 struct page *pages[16];
1383 unsigned long index = start >> PAGE_CACHE_SHIFT;
1384 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1385 unsigned long nr_pages = end_index - index + 1;
1389 if (op & EXTENT_CLEAR_UNLOCK)
1390 clear_bits |= EXTENT_LOCKED;
1391 if (op & EXTENT_CLEAR_DIRTY)
1392 clear_bits |= EXTENT_DIRTY;
1394 if (op & EXTENT_CLEAR_DELALLOC)
1395 clear_bits |= EXTENT_DELALLOC;
1397 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1398 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1399 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1400 EXTENT_SET_PRIVATE2)))
1403 while (nr_pages > 0) {
1404 ret = find_get_pages_contig(inode->i_mapping, index,
1405 min_t(unsigned long,
1406 nr_pages, ARRAY_SIZE(pages)), pages);
1407 for (i = 0; i < ret; i++) {
1409 if (op & EXTENT_SET_PRIVATE2)
1410 SetPagePrivate2(pages[i]);
1412 if (pages[i] == locked_page) {
1413 page_cache_release(pages[i]);
1416 if (op & EXTENT_CLEAR_DIRTY)
1417 clear_page_dirty_for_io(pages[i]);
1418 if (op & EXTENT_SET_WRITEBACK)
1419 set_page_writeback(pages[i]);
1420 if (op & EXTENT_END_WRITEBACK)
1421 end_page_writeback(pages[i]);
1422 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1423 unlock_page(pages[i]);
1424 page_cache_release(pages[i]);
1434 * count the number of bytes in the tree that have a given bit(s)
1435 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1436 * cached. The total number found is returned.
1438 u64 count_range_bits(struct extent_io_tree *tree,
1439 u64 *start, u64 search_end, u64 max_bytes,
1440 unsigned long bits, int contig)
1442 struct rb_node *node;
1443 struct extent_state *state;
1444 u64 cur_start = *start;
1445 u64 total_bytes = 0;
1449 if (search_end <= cur_start) {
1454 spin_lock(&tree->lock);
1455 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1456 total_bytes = tree->dirty_bytes;
1460 * this search will find all the extents that end after
1463 node = tree_search(tree, cur_start);
1468 state = rb_entry(node, struct extent_state, rb_node);
1469 if (state->start > search_end)
1471 if (contig && found && state->start > last + 1)
1473 if (state->end >= cur_start && (state->state & bits) == bits) {
1474 total_bytes += min(search_end, state->end) + 1 -
1475 max(cur_start, state->start);
1476 if (total_bytes >= max_bytes)
1479 *start = max(cur_start, state->start);
1483 } else if (contig && found) {
1486 node = rb_next(node);
1491 spin_unlock(&tree->lock);
1496 * set the private field for a given byte offset in the tree. If there isn't
1497 * an extent_state there already, this does nothing.
1499 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1501 struct rb_node *node;
1502 struct extent_state *state;
1505 spin_lock(&tree->lock);
1507 * this search will find all the extents that end after
1510 node = tree_search(tree, start);
1515 state = rb_entry(node, struct extent_state, rb_node);
1516 if (state->start != start) {
1520 state->private = private;
1522 spin_unlock(&tree->lock);
1526 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1528 struct rb_node *node;
1529 struct extent_state *state;
1532 spin_lock(&tree->lock);
1534 * this search will find all the extents that end after
1537 node = tree_search(tree, start);
1542 state = rb_entry(node, struct extent_state, rb_node);
1543 if (state->start != start) {
1547 *private = state->private;
1549 spin_unlock(&tree->lock);
1554 * searches a range in the state tree for a given mask.
1555 * If 'filled' == 1, this returns 1 only if every extent in the tree
1556 * has the bits set. Otherwise, 1 is returned if any bit in the
1557 * range is found set.
1559 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1560 int bits, int filled, struct extent_state *cached)
1562 struct extent_state *state = NULL;
1563 struct rb_node *node;
1566 spin_lock(&tree->lock);
1567 if (cached && cached->tree && cached->start == start)
1568 node = &cached->rb_node;
1570 node = tree_search(tree, start);
1571 while (node && start <= end) {
1572 state = rb_entry(node, struct extent_state, rb_node);
1574 if (filled && state->start > start) {
1579 if (state->start > end)
1582 if (state->state & bits) {
1586 } else if (filled) {
1591 if (state->end == (u64)-1)
1594 start = state->end + 1;
1597 node = rb_next(node);
1604 spin_unlock(&tree->lock);
1609 * helper function to set a given page up to date if all the
1610 * extents in the tree for that page are up to date
1612 static int check_page_uptodate(struct extent_io_tree *tree,
1615 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1616 u64 end = start + PAGE_CACHE_SIZE - 1;
1617 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1618 SetPageUptodate(page);
1623 * helper function to unlock a page if all the extents in the tree
1624 * for that page are unlocked
1626 static int check_page_locked(struct extent_io_tree *tree,
1629 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1630 u64 end = start + PAGE_CACHE_SIZE - 1;
1631 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1637 * helper function to end page writeback if all the extents
1638 * in the tree for that page are done with writeback
1640 static int check_page_writeback(struct extent_io_tree *tree,
1643 end_page_writeback(page);
1647 /* lots and lots of room for performance fixes in the end_bio funcs */
1650 * after a writepage IO is done, we need to:
1651 * clear the uptodate bits on error
1652 * clear the writeback bits in the extent tree for this IO
1653 * end_page_writeback if the page has no more pending IO
1655 * Scheduling is not allowed, so the extent state tree is expected
1656 * to have one and only one object corresponding to this IO.
1658 static void end_bio_extent_writepage(struct bio *bio, int err)
1660 int uptodate = err == 0;
1661 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1662 struct extent_io_tree *tree;
1669 struct page *page = bvec->bv_page;
1670 tree = &BTRFS_I(page->mapping->host)->io_tree;
1672 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1674 end = start + bvec->bv_len - 1;
1676 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1681 if (--bvec >= bio->bi_io_vec)
1682 prefetchw(&bvec->bv_page->flags);
1683 if (tree->ops && tree->ops->writepage_end_io_hook) {
1684 ret = tree->ops->writepage_end_io_hook(page, start,
1685 end, NULL, uptodate);
1690 if (!uptodate && tree->ops &&
1691 tree->ops->writepage_io_failed_hook) {
1692 ret = tree->ops->writepage_io_failed_hook(bio, page,
1695 uptodate = (err == 0);
1701 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
1702 ClearPageUptodate(page);
1707 end_page_writeback(page);
1709 check_page_writeback(tree, page);
1710 } while (bvec >= bio->bi_io_vec);
1716 * after a readpage IO is done, we need to:
1717 * clear the uptodate bits on error
1718 * set the uptodate bits if things worked
1719 * set the page up to date if all extents in the tree are uptodate
1720 * clear the lock bit in the extent tree
1721 * unlock the page if there are no other extents locked for it
1723 * Scheduling is not allowed, so the extent state tree is expected
1724 * to have one and only one object corresponding to this IO.
1726 static void end_bio_extent_readpage(struct bio *bio, int err)
1728 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1729 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
1730 struct bio_vec *bvec = bio->bi_io_vec;
1731 struct extent_io_tree *tree;
1741 struct page *page = bvec->bv_page;
1742 struct extent_state *cached = NULL;
1743 struct extent_state *state;
1745 tree = &BTRFS_I(page->mapping->host)->io_tree;
1747 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1749 end = start + bvec->bv_len - 1;
1751 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1756 if (++bvec <= bvec_end)
1757 prefetchw(&bvec->bv_page->flags);
1759 spin_lock(&tree->lock);
1760 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
1761 if (state && state->start == start) {
1763 * take a reference on the state, unlock will drop
1766 cache_state(state, &cached);
1768 spin_unlock(&tree->lock);
1770 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1771 ret = tree->ops->readpage_end_io_hook(page, start, end,
1776 if (!uptodate && tree->ops &&
1777 tree->ops->readpage_io_failed_hook) {
1778 ret = tree->ops->readpage_io_failed_hook(bio, page,
1782 test_bit(BIO_UPTODATE, &bio->bi_flags);
1785 uncache_state(&cached);
1791 set_extent_uptodate(tree, start, end, &cached,
1794 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
1798 SetPageUptodate(page);
1800 ClearPageUptodate(page);
1806 check_page_uptodate(tree, page);
1808 ClearPageUptodate(page);
1811 check_page_locked(tree, page);
1813 } while (bvec <= bvec_end);
1819 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1824 bio = bio_alloc(gfp_flags, nr_vecs);
1826 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1827 while (!bio && (nr_vecs /= 2))
1828 bio = bio_alloc(gfp_flags, nr_vecs);
1833 bio->bi_bdev = bdev;
1834 bio->bi_sector = first_sector;
1839 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1840 unsigned long bio_flags)
1843 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1844 struct page *page = bvec->bv_page;
1845 struct extent_io_tree *tree = bio->bi_private;
1848 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1850 bio->bi_private = NULL;
1854 if (tree->ops && tree->ops->submit_bio_hook)
1855 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1856 mirror_num, bio_flags, start);
1858 submit_bio(rw, bio);
1859 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1865 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1866 struct page *page, sector_t sector,
1867 size_t size, unsigned long offset,
1868 struct block_device *bdev,
1869 struct bio **bio_ret,
1870 unsigned long max_pages,
1871 bio_end_io_t end_io_func,
1873 unsigned long prev_bio_flags,
1874 unsigned long bio_flags)
1880 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1881 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1882 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1884 if (bio_ret && *bio_ret) {
1887 contig = bio->bi_sector == sector;
1889 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1892 if (prev_bio_flags != bio_flags || !contig ||
1893 (tree->ops && tree->ops->merge_bio_hook &&
1894 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1896 bio_add_page(bio, page, page_size, offset) < page_size) {
1897 ret = submit_one_bio(rw, bio, mirror_num,
1904 if (this_compressed)
1907 nr = bio_get_nr_vecs(bdev);
1909 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1913 bio_add_page(bio, page, page_size, offset);
1914 bio->bi_end_io = end_io_func;
1915 bio->bi_private = tree;
1920 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1925 void set_page_extent_mapped(struct page *page)
1927 if (!PagePrivate(page)) {
1928 SetPagePrivate(page);
1929 page_cache_get(page);
1930 set_page_private(page, EXTENT_PAGE_PRIVATE);
1934 static void set_page_extent_head(struct page *page, unsigned long len)
1936 WARN_ON(!PagePrivate(page));
1937 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1941 * basic readpage implementation. Locked extent state structs are inserted
1942 * into the tree that are removed when the IO is done (by the end_io
1945 static int __extent_read_full_page(struct extent_io_tree *tree,
1947 get_extent_t *get_extent,
1948 struct bio **bio, int mirror_num,
1949 unsigned long *bio_flags)
1951 struct inode *inode = page->mapping->host;
1952 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1953 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1957 u64 last_byte = i_size_read(inode);
1961 struct extent_map *em;
1962 struct block_device *bdev;
1963 struct btrfs_ordered_extent *ordered;
1966 size_t pg_offset = 0;
1968 size_t disk_io_size;
1969 size_t blocksize = inode->i_sb->s_blocksize;
1970 unsigned long this_bio_flag = 0;
1972 set_page_extent_mapped(page);
1974 if (!PageUptodate(page)) {
1975 if (cleancache_get_page(page) == 0) {
1976 BUG_ON(blocksize != PAGE_SIZE);
1983 lock_extent(tree, start, end, GFP_NOFS);
1984 ordered = btrfs_lookup_ordered_extent(inode, start);
1987 unlock_extent(tree, start, end, GFP_NOFS);
1988 btrfs_start_ordered_extent(inode, ordered, 1);
1989 btrfs_put_ordered_extent(ordered);
1992 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1994 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
1997 iosize = PAGE_CACHE_SIZE - zero_offset;
1998 userpage = kmap_atomic(page, KM_USER0);
1999 memset(userpage + zero_offset, 0, iosize);
2000 flush_dcache_page(page);
2001 kunmap_atomic(userpage, KM_USER0);
2004 while (cur <= end) {
2005 if (cur >= last_byte) {
2007 struct extent_state *cached = NULL;
2009 iosize = PAGE_CACHE_SIZE - pg_offset;
2010 userpage = kmap_atomic(page, KM_USER0);
2011 memset(userpage + pg_offset, 0, iosize);
2012 flush_dcache_page(page);
2013 kunmap_atomic(userpage, KM_USER0);
2014 set_extent_uptodate(tree, cur, cur + iosize - 1,
2016 unlock_extent_cached(tree, cur, cur + iosize - 1,
2020 em = get_extent(inode, page, pg_offset, cur,
2022 if (IS_ERR_OR_NULL(em)) {
2024 unlock_extent(tree, cur, end, GFP_NOFS);
2027 extent_offset = cur - em->start;
2028 BUG_ON(extent_map_end(em) <= cur);
2031 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2032 this_bio_flag = EXTENT_BIO_COMPRESSED;
2033 extent_set_compress_type(&this_bio_flag,
2037 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2038 cur_end = min(extent_map_end(em) - 1, end);
2039 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2040 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2041 disk_io_size = em->block_len;
2042 sector = em->block_start >> 9;
2044 sector = (em->block_start + extent_offset) >> 9;
2045 disk_io_size = iosize;
2048 block_start = em->block_start;
2049 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2050 block_start = EXTENT_MAP_HOLE;
2051 free_extent_map(em);
2054 /* we've found a hole, just zero and go on */
2055 if (block_start == EXTENT_MAP_HOLE) {
2057 struct extent_state *cached = NULL;
2059 userpage = kmap_atomic(page, KM_USER0);
2060 memset(userpage + pg_offset, 0, iosize);
2061 flush_dcache_page(page);
2062 kunmap_atomic(userpage, KM_USER0);
2064 set_extent_uptodate(tree, cur, cur + iosize - 1,
2066 unlock_extent_cached(tree, cur, cur + iosize - 1,
2069 pg_offset += iosize;
2072 /* the get_extent function already copied into the page */
2073 if (test_range_bit(tree, cur, cur_end,
2074 EXTENT_UPTODATE, 1, NULL)) {
2075 check_page_uptodate(tree, page);
2076 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2078 pg_offset += iosize;
2081 /* we have an inline extent but it didn't get marked up
2082 * to date. Error out
2084 if (block_start == EXTENT_MAP_INLINE) {
2086 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2088 pg_offset += iosize;
2093 if (tree->ops && tree->ops->readpage_io_hook) {
2094 ret = tree->ops->readpage_io_hook(page, cur,
2098 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2100 ret = submit_extent_page(READ, tree, page,
2101 sector, disk_io_size, pg_offset,
2103 end_bio_extent_readpage, mirror_num,
2107 *bio_flags = this_bio_flag;
2112 pg_offset += iosize;
2116 if (!PageError(page))
2117 SetPageUptodate(page);
2123 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2124 get_extent_t *get_extent)
2126 struct bio *bio = NULL;
2127 unsigned long bio_flags = 0;
2130 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2133 ret = submit_one_bio(READ, bio, 0, bio_flags);
2137 static noinline void update_nr_written(struct page *page,
2138 struct writeback_control *wbc,
2139 unsigned long nr_written)
2141 wbc->nr_to_write -= nr_written;
2142 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2143 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2144 page->mapping->writeback_index = page->index + nr_written;
2148 * the writepage semantics are similar to regular writepage. extent
2149 * records are inserted to lock ranges in the tree, and as dirty areas
2150 * are found, they are marked writeback. Then the lock bits are removed
2151 * and the end_io handler clears the writeback ranges
2153 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2156 struct inode *inode = page->mapping->host;
2157 struct extent_page_data *epd = data;
2158 struct extent_io_tree *tree = epd->tree;
2159 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2161 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2165 u64 last_byte = i_size_read(inode);
2169 struct extent_state *cached_state = NULL;
2170 struct extent_map *em;
2171 struct block_device *bdev;
2174 size_t pg_offset = 0;
2176 loff_t i_size = i_size_read(inode);
2177 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2183 unsigned long nr_written = 0;
2185 if (wbc->sync_mode == WB_SYNC_ALL)
2186 write_flags = WRITE_SYNC;
2188 write_flags = WRITE;
2190 trace___extent_writepage(page, inode, wbc);
2192 WARN_ON(!PageLocked(page));
2193 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2194 if (page->index > end_index ||
2195 (page->index == end_index && !pg_offset)) {
2196 page->mapping->a_ops->invalidatepage(page, 0);
2201 if (page->index == end_index) {
2204 userpage = kmap_atomic(page, KM_USER0);
2205 memset(userpage + pg_offset, 0,
2206 PAGE_CACHE_SIZE - pg_offset);
2207 kunmap_atomic(userpage, KM_USER0);
2208 flush_dcache_page(page);
2212 set_page_extent_mapped(page);
2214 delalloc_start = start;
2217 if (!epd->extent_locked) {
2218 u64 delalloc_to_write = 0;
2220 * make sure the wbc mapping index is at least updated
2223 update_nr_written(page, wbc, 0);
2225 while (delalloc_end < page_end) {
2226 nr_delalloc = find_lock_delalloc_range(inode, tree,
2231 if (nr_delalloc == 0) {
2232 delalloc_start = delalloc_end + 1;
2235 tree->ops->fill_delalloc(inode, page, delalloc_start,
2236 delalloc_end, &page_started,
2239 * delalloc_end is already one less than the total
2240 * length, so we don't subtract one from
2243 delalloc_to_write += (delalloc_end - delalloc_start +
2246 delalloc_start = delalloc_end + 1;
2248 if (wbc->nr_to_write < delalloc_to_write) {
2251 if (delalloc_to_write < thresh * 2)
2252 thresh = delalloc_to_write;
2253 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2257 /* did the fill delalloc function already unlock and start
2263 * we've unlocked the page, so we can't update
2264 * the mapping's writeback index, just update
2267 wbc->nr_to_write -= nr_written;
2271 if (tree->ops && tree->ops->writepage_start_hook) {
2272 ret = tree->ops->writepage_start_hook(page, start,
2274 if (ret == -EAGAIN) {
2275 redirty_page_for_writepage(wbc, page);
2276 update_nr_written(page, wbc, nr_written);
2284 * we don't want to touch the inode after unlocking the page,
2285 * so we update the mapping writeback index now
2287 update_nr_written(page, wbc, nr_written + 1);
2290 if (last_byte <= start) {
2291 if (tree->ops && tree->ops->writepage_end_io_hook)
2292 tree->ops->writepage_end_io_hook(page, start,
2297 blocksize = inode->i_sb->s_blocksize;
2299 while (cur <= end) {
2300 if (cur >= last_byte) {
2301 if (tree->ops && tree->ops->writepage_end_io_hook)
2302 tree->ops->writepage_end_io_hook(page, cur,
2306 em = epd->get_extent(inode, page, pg_offset, cur,
2308 if (IS_ERR_OR_NULL(em)) {
2313 extent_offset = cur - em->start;
2314 BUG_ON(extent_map_end(em) <= cur);
2316 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2317 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2318 sector = (em->block_start + extent_offset) >> 9;
2320 block_start = em->block_start;
2321 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2322 free_extent_map(em);
2326 * compressed and inline extents are written through other
2329 if (compressed || block_start == EXTENT_MAP_HOLE ||
2330 block_start == EXTENT_MAP_INLINE) {
2332 * end_io notification does not happen here for
2333 * compressed extents
2335 if (!compressed && tree->ops &&
2336 tree->ops->writepage_end_io_hook)
2337 tree->ops->writepage_end_io_hook(page, cur,
2340 else if (compressed) {
2341 /* we don't want to end_page_writeback on
2342 * a compressed extent. this happens
2349 pg_offset += iosize;
2352 /* leave this out until we have a page_mkwrite call */
2353 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2354 EXTENT_DIRTY, 0, NULL)) {
2356 pg_offset += iosize;
2360 if (tree->ops && tree->ops->writepage_io_hook) {
2361 ret = tree->ops->writepage_io_hook(page, cur,
2369 unsigned long max_nr = end_index + 1;
2371 set_range_writeback(tree, cur, cur + iosize - 1);
2372 if (!PageWriteback(page)) {
2373 printk(KERN_ERR "btrfs warning page %lu not "
2374 "writeback, cur %llu end %llu\n",
2375 page->index, (unsigned long long)cur,
2376 (unsigned long long)end);
2379 ret = submit_extent_page(write_flags, tree, page,
2380 sector, iosize, pg_offset,
2381 bdev, &epd->bio, max_nr,
2382 end_bio_extent_writepage,
2388 pg_offset += iosize;
2393 /* make sure the mapping tag for page dirty gets cleared */
2394 set_page_writeback(page);
2395 end_page_writeback(page);
2401 /* drop our reference on any cached states */
2402 free_extent_state(cached_state);
2407 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2408 * @mapping: address space structure to write
2409 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2410 * @writepage: function called for each page
2411 * @data: data passed to writepage function
2413 * If a page is already under I/O, write_cache_pages() skips it, even
2414 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2415 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2416 * and msync() need to guarantee that all the data which was dirty at the time
2417 * the call was made get new I/O started against them. If wbc->sync_mode is
2418 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2419 * existing IO to complete.
2421 static int extent_write_cache_pages(struct extent_io_tree *tree,
2422 struct address_space *mapping,
2423 struct writeback_control *wbc,
2424 writepage_t writepage, void *data,
2425 void (*flush_fn)(void *))
2429 int nr_to_write_done = 0;
2430 struct pagevec pvec;
2433 pgoff_t end; /* Inclusive */
2436 pagevec_init(&pvec, 0);
2437 if (wbc->range_cyclic) {
2438 index = mapping->writeback_index; /* Start from prev offset */
2441 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2442 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2446 while (!done && !nr_to_write_done && (index <= end) &&
2447 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2448 PAGECACHE_TAG_DIRTY, min(end - index,
2449 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2453 for (i = 0; i < nr_pages; i++) {
2454 struct page *page = pvec.pages[i];
2457 * At this point we hold neither mapping->tree_lock nor
2458 * lock on the page itself: the page may be truncated or
2459 * invalidated (changing page->mapping to NULL), or even
2460 * swizzled back from swapper_space to tmpfs file
2463 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2464 tree->ops->write_cache_pages_lock_hook(page);
2468 if (unlikely(page->mapping != mapping)) {
2473 if (!wbc->range_cyclic && page->index > end) {
2479 if (wbc->sync_mode != WB_SYNC_NONE) {
2480 if (PageWriteback(page))
2482 wait_on_page_writeback(page);
2485 if (PageWriteback(page) ||
2486 !clear_page_dirty_for_io(page)) {
2491 ret = (*writepage)(page, wbc, data);
2493 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2501 * the filesystem may choose to bump up nr_to_write.
2502 * We have to make sure to honor the new nr_to_write
2505 nr_to_write_done = wbc->nr_to_write <= 0;
2507 pagevec_release(&pvec);
2510 if (!scanned && !done) {
2512 * We hit the last page and there is more work to be done: wrap
2513 * back to the start of the file
2522 static void flush_epd_write_bio(struct extent_page_data *epd)
2526 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2528 submit_one_bio(WRITE, epd->bio, 0, 0);
2533 static noinline void flush_write_bio(void *data)
2535 struct extent_page_data *epd = data;
2536 flush_epd_write_bio(epd);
2539 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2540 get_extent_t *get_extent,
2541 struct writeback_control *wbc)
2544 struct address_space *mapping = page->mapping;
2545 struct extent_page_data epd = {
2548 .get_extent = get_extent,
2550 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2552 struct writeback_control wbc_writepages = {
2553 .sync_mode = wbc->sync_mode,
2554 .older_than_this = NULL,
2556 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2557 .range_end = (loff_t)-1,
2560 ret = __extent_writepage(page, wbc, &epd);
2562 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2563 __extent_writepage, &epd, flush_write_bio);
2564 flush_epd_write_bio(&epd);
2568 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2569 u64 start, u64 end, get_extent_t *get_extent,
2573 struct address_space *mapping = inode->i_mapping;
2575 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2578 struct extent_page_data epd = {
2581 .get_extent = get_extent,
2583 .sync_io = mode == WB_SYNC_ALL,
2585 struct writeback_control wbc_writepages = {
2587 .older_than_this = NULL,
2588 .nr_to_write = nr_pages * 2,
2589 .range_start = start,
2590 .range_end = end + 1,
2593 while (start <= end) {
2594 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2595 if (clear_page_dirty_for_io(page))
2596 ret = __extent_writepage(page, &wbc_writepages, &epd);
2598 if (tree->ops && tree->ops->writepage_end_io_hook)
2599 tree->ops->writepage_end_io_hook(page, start,
2600 start + PAGE_CACHE_SIZE - 1,
2604 page_cache_release(page);
2605 start += PAGE_CACHE_SIZE;
2608 flush_epd_write_bio(&epd);
2612 int extent_writepages(struct extent_io_tree *tree,
2613 struct address_space *mapping,
2614 get_extent_t *get_extent,
2615 struct writeback_control *wbc)
2618 struct extent_page_data epd = {
2621 .get_extent = get_extent,
2623 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2626 ret = extent_write_cache_pages(tree, mapping, wbc,
2627 __extent_writepage, &epd,
2629 flush_epd_write_bio(&epd);
2633 int extent_readpages(struct extent_io_tree *tree,
2634 struct address_space *mapping,
2635 struct list_head *pages, unsigned nr_pages,
2636 get_extent_t get_extent)
2638 struct bio *bio = NULL;
2640 unsigned long bio_flags = 0;
2642 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2643 struct page *page = list_entry(pages->prev, struct page, lru);
2645 prefetchw(&page->flags);
2646 list_del(&page->lru);
2647 if (!add_to_page_cache_lru(page, mapping,
2648 page->index, GFP_NOFS)) {
2649 __extent_read_full_page(tree, page, get_extent,
2650 &bio, 0, &bio_flags);
2652 page_cache_release(page);
2654 BUG_ON(!list_empty(pages));
2656 submit_one_bio(READ, bio, 0, bio_flags);
2661 * basic invalidatepage code, this waits on any locked or writeback
2662 * ranges corresponding to the page, and then deletes any extent state
2663 * records from the tree
2665 int extent_invalidatepage(struct extent_io_tree *tree,
2666 struct page *page, unsigned long offset)
2668 struct extent_state *cached_state = NULL;
2669 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2670 u64 end = start + PAGE_CACHE_SIZE - 1;
2671 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2673 start += (offset + blocksize - 1) & ~(blocksize - 1);
2677 lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS);
2678 wait_on_page_writeback(page);
2679 clear_extent_bit(tree, start, end,
2680 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2681 EXTENT_DO_ACCOUNTING,
2682 1, 1, &cached_state, GFP_NOFS);
2687 * a helper for releasepage, this tests for areas of the page that
2688 * are locked or under IO and drops the related state bits if it is safe
2691 int try_release_extent_state(struct extent_map_tree *map,
2692 struct extent_io_tree *tree, struct page *page,
2695 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2696 u64 end = start + PAGE_CACHE_SIZE - 1;
2699 if (test_range_bit(tree, start, end,
2700 EXTENT_IOBITS, 0, NULL))
2703 if ((mask & GFP_NOFS) == GFP_NOFS)
2706 * at this point we can safely clear everything except the
2707 * locked bit and the nodatasum bit
2709 ret = clear_extent_bit(tree, start, end,
2710 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
2713 /* if clear_extent_bit failed for enomem reasons,
2714 * we can't allow the release to continue.
2725 * a helper for releasepage. As long as there are no locked extents
2726 * in the range corresponding to the page, both state records and extent
2727 * map records are removed
2729 int try_release_extent_mapping(struct extent_map_tree *map,
2730 struct extent_io_tree *tree, struct page *page,
2733 struct extent_map *em;
2734 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2735 u64 end = start + PAGE_CACHE_SIZE - 1;
2737 if ((mask & __GFP_WAIT) &&
2738 page->mapping->host->i_size > 16 * 1024 * 1024) {
2740 while (start <= end) {
2741 len = end - start + 1;
2742 write_lock(&map->lock);
2743 em = lookup_extent_mapping(map, start, len);
2744 if (IS_ERR_OR_NULL(em)) {
2745 write_unlock(&map->lock);
2748 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2749 em->start != start) {
2750 write_unlock(&map->lock);
2751 free_extent_map(em);
2754 if (!test_range_bit(tree, em->start,
2755 extent_map_end(em) - 1,
2756 EXTENT_LOCKED | EXTENT_WRITEBACK,
2758 remove_extent_mapping(map, em);
2759 /* once for the rb tree */
2760 free_extent_map(em);
2762 start = extent_map_end(em);
2763 write_unlock(&map->lock);
2766 free_extent_map(em);
2769 return try_release_extent_state(map, tree, page, mask);
2773 * helper function for fiemap, which doesn't want to see any holes.
2774 * This maps until we find something past 'last'
2776 static struct extent_map *get_extent_skip_holes(struct inode *inode,
2779 get_extent_t *get_extent)
2781 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
2782 struct extent_map *em;
2789 len = last - offset;
2792 len = (len + sectorsize - 1) & ~(sectorsize - 1);
2793 em = get_extent(inode, NULL, 0, offset, len, 0);
2794 if (IS_ERR_OR_NULL(em))
2797 /* if this isn't a hole return it */
2798 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
2799 em->block_start != EXTENT_MAP_HOLE) {
2803 /* this is a hole, advance to the next extent */
2804 offset = extent_map_end(em);
2805 free_extent_map(em);
2812 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2813 __u64 start, __u64 len, get_extent_t *get_extent)
2817 u64 max = start + len;
2821 u64 last_for_get_extent = 0;
2823 u64 isize = i_size_read(inode);
2824 struct btrfs_key found_key;
2825 struct extent_map *em = NULL;
2826 struct extent_state *cached_state = NULL;
2827 struct btrfs_path *path;
2828 struct btrfs_file_extent_item *item;
2833 unsigned long emflags;
2838 path = btrfs_alloc_path();
2841 path->leave_spinning = 1;
2844 * lookup the last file extent. We're not using i_size here
2845 * because there might be preallocation past i_size
2847 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
2848 path, btrfs_ino(inode), -1, 0);
2850 btrfs_free_path(path);
2855 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2856 struct btrfs_file_extent_item);
2857 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
2858 found_type = btrfs_key_type(&found_key);
2860 /* No extents, but there might be delalloc bits */
2861 if (found_key.objectid != btrfs_ino(inode) ||
2862 found_type != BTRFS_EXTENT_DATA_KEY) {
2863 /* have to trust i_size as the end */
2865 last_for_get_extent = isize;
2868 * remember the start of the last extent. There are a
2869 * bunch of different factors that go into the length of the
2870 * extent, so its much less complex to remember where it started
2872 last = found_key.offset;
2873 last_for_get_extent = last + 1;
2875 btrfs_free_path(path);
2878 * we might have some extents allocated but more delalloc past those
2879 * extents. so, we trust isize unless the start of the last extent is
2884 last_for_get_extent = isize;
2887 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
2888 &cached_state, GFP_NOFS);
2890 em = get_extent_skip_holes(inode, off, last_for_get_extent,
2900 u64 offset_in_extent;
2902 /* break if the extent we found is outside the range */
2903 if (em->start >= max || extent_map_end(em) < off)
2907 * get_extent may return an extent that starts before our
2908 * requested range. We have to make sure the ranges
2909 * we return to fiemap always move forward and don't
2910 * overlap, so adjust the offsets here
2912 em_start = max(em->start, off);
2915 * record the offset from the start of the extent
2916 * for adjusting the disk offset below
2918 offset_in_extent = em_start - em->start;
2919 em_end = extent_map_end(em);
2920 em_len = em_end - em_start;
2921 emflags = em->flags;
2926 * bump off for our next call to get_extent
2928 off = extent_map_end(em);
2932 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
2934 flags |= FIEMAP_EXTENT_LAST;
2935 } else if (em->block_start == EXTENT_MAP_INLINE) {
2936 flags |= (FIEMAP_EXTENT_DATA_INLINE |
2937 FIEMAP_EXTENT_NOT_ALIGNED);
2938 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
2939 flags |= (FIEMAP_EXTENT_DELALLOC |
2940 FIEMAP_EXTENT_UNKNOWN);
2942 disko = em->block_start + offset_in_extent;
2944 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2945 flags |= FIEMAP_EXTENT_ENCODED;
2947 free_extent_map(em);
2949 if ((em_start >= last) || em_len == (u64)-1 ||
2950 (last == (u64)-1 && isize <= em_end)) {
2951 flags |= FIEMAP_EXTENT_LAST;
2955 /* now scan forward to see if this is really the last extent. */
2956 em = get_extent_skip_holes(inode, off, last_for_get_extent,
2963 flags |= FIEMAP_EXTENT_LAST;
2966 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
2972 free_extent_map(em);
2974 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
2975 &cached_state, GFP_NOFS);
2979 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2983 struct address_space *mapping;
2986 return eb->first_page;
2987 i += eb->start >> PAGE_CACHE_SHIFT;
2988 mapping = eb->first_page->mapping;
2993 * extent_buffer_page is only called after pinning the page
2994 * by increasing the reference count. So we know the page must
2995 * be in the radix tree.
2998 p = radix_tree_lookup(&mapping->page_tree, i);
3004 static inline unsigned long num_extent_pages(u64 start, u64 len)
3006 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3007 (start >> PAGE_CACHE_SHIFT);
3010 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3015 struct extent_buffer *eb = NULL;
3017 unsigned long flags;
3020 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3025 spin_lock_init(&eb->lock);
3026 init_waitqueue_head(&eb->lock_wq);
3029 spin_lock_irqsave(&leak_lock, flags);
3030 list_add(&eb->leak_list, &buffers);
3031 spin_unlock_irqrestore(&leak_lock, flags);
3033 atomic_set(&eb->refs, 1);
3038 static void __free_extent_buffer(struct extent_buffer *eb)
3041 unsigned long flags;
3042 spin_lock_irqsave(&leak_lock, flags);
3043 list_del(&eb->leak_list);
3044 spin_unlock_irqrestore(&leak_lock, flags);
3046 kmem_cache_free(extent_buffer_cache, eb);
3050 * Helper for releasing extent buffer page.
3052 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3053 unsigned long start_idx)
3055 unsigned long index;
3058 if (!eb->first_page)
3061 index = num_extent_pages(eb->start, eb->len);
3062 if (start_idx >= index)
3067 page = extent_buffer_page(eb, index);
3069 page_cache_release(page);
3070 } while (index != start_idx);
3074 * Helper for releasing the extent buffer.
3076 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3078 btrfs_release_extent_buffer_page(eb, 0);
3079 __free_extent_buffer(eb);
3082 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3083 u64 start, unsigned long len,
3086 unsigned long num_pages = num_extent_pages(start, len);
3088 unsigned long index = start >> PAGE_CACHE_SHIFT;
3089 struct extent_buffer *eb;
3090 struct extent_buffer *exists = NULL;
3092 struct address_space *mapping = tree->mapping;
3097 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3098 if (eb && atomic_inc_not_zero(&eb->refs)) {
3100 mark_page_accessed(eb->first_page);
3105 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
3110 eb->first_page = page0;
3113 page_cache_get(page0);
3114 mark_page_accessed(page0);
3115 set_page_extent_mapped(page0);
3116 set_page_extent_head(page0, len);
3117 uptodate = PageUptodate(page0);
3121 for (; i < num_pages; i++, index++) {
3122 p = find_or_create_page(mapping, index, GFP_NOFS | __GFP_HIGHMEM);
3127 set_page_extent_mapped(p);
3128 mark_page_accessed(p);
3131 set_page_extent_head(p, len);
3133 set_page_private(p, EXTENT_PAGE_PRIVATE);
3135 if (!PageUptodate(p))
3139 * see below about how we avoid a nasty race with release page
3140 * and why we unlock later
3146 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3148 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3152 spin_lock(&tree->buffer_lock);
3153 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3154 if (ret == -EEXIST) {
3155 exists = radix_tree_lookup(&tree->buffer,
3156 start >> PAGE_CACHE_SHIFT);
3157 /* add one reference for the caller */
3158 atomic_inc(&exists->refs);
3159 spin_unlock(&tree->buffer_lock);
3160 radix_tree_preload_end();
3163 /* add one reference for the tree */
3164 atomic_inc(&eb->refs);
3165 spin_unlock(&tree->buffer_lock);
3166 radix_tree_preload_end();
3169 * there is a race where release page may have
3170 * tried to find this extent buffer in the radix
3171 * but failed. It will tell the VM it is safe to
3172 * reclaim the, and it will clear the page private bit.
3173 * We must make sure to set the page private bit properly
3174 * after the extent buffer is in the radix tree so
3175 * it doesn't get lost
3177 set_page_extent_mapped(eb->first_page);
3178 set_page_extent_head(eb->first_page, eb->len);
3180 unlock_page(eb->first_page);
3184 if (eb->first_page && !page0)
3185 unlock_page(eb->first_page);
3187 if (!atomic_dec_and_test(&eb->refs))
3189 btrfs_release_extent_buffer(eb);
3193 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3194 u64 start, unsigned long len)
3196 struct extent_buffer *eb;
3199 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3200 if (eb && atomic_inc_not_zero(&eb->refs)) {
3202 mark_page_accessed(eb->first_page);
3210 void free_extent_buffer(struct extent_buffer *eb)
3215 if (!atomic_dec_and_test(&eb->refs))
3221 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3222 struct extent_buffer *eb)
3225 unsigned long num_pages;
3228 num_pages = num_extent_pages(eb->start, eb->len);
3230 for (i = 0; i < num_pages; i++) {
3231 page = extent_buffer_page(eb, i);
3232 if (!PageDirty(page))
3236 WARN_ON(!PagePrivate(page));
3238 set_page_extent_mapped(page);
3240 set_page_extent_head(page, eb->len);
3242 clear_page_dirty_for_io(page);
3243 spin_lock_irq(&page->mapping->tree_lock);
3244 if (!PageDirty(page)) {
3245 radix_tree_tag_clear(&page->mapping->page_tree,
3247 PAGECACHE_TAG_DIRTY);
3249 spin_unlock_irq(&page->mapping->tree_lock);
3255 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3256 struct extent_buffer *eb)
3259 unsigned long num_pages;
3262 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3263 num_pages = num_extent_pages(eb->start, eb->len);
3264 for (i = 0; i < num_pages; i++)
3265 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3269 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3270 struct extent_buffer *eb,
3271 struct extent_state **cached_state)
3275 unsigned long num_pages;
3277 num_pages = num_extent_pages(eb->start, eb->len);
3278 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3280 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3281 cached_state, GFP_NOFS);
3282 for (i = 0; i < num_pages; i++) {
3283 page = extent_buffer_page(eb, i);
3285 ClearPageUptodate(page);
3290 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3291 struct extent_buffer *eb)
3295 unsigned long num_pages;
3297 num_pages = num_extent_pages(eb->start, eb->len);
3299 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3301 for (i = 0; i < num_pages; i++) {
3302 page = extent_buffer_page(eb, i);
3303 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3304 ((i == num_pages - 1) &&
3305 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3306 check_page_uptodate(tree, page);
3309 SetPageUptodate(page);
3314 int extent_range_uptodate(struct extent_io_tree *tree,
3319 int pg_uptodate = 1;
3321 unsigned long index;
3323 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL);
3326 while (start <= end) {
3327 index = start >> PAGE_CACHE_SHIFT;
3328 page = find_get_page(tree->mapping, index);
3329 uptodate = PageUptodate(page);
3330 page_cache_release(page);
3335 start += PAGE_CACHE_SIZE;
3340 int extent_buffer_uptodate(struct extent_io_tree *tree,
3341 struct extent_buffer *eb,
3342 struct extent_state *cached_state)
3345 unsigned long num_pages;
3348 int pg_uptodate = 1;
3350 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3353 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3354 EXTENT_UPTODATE, 1, cached_state);
3358 num_pages = num_extent_pages(eb->start, eb->len);
3359 for (i = 0; i < num_pages; i++) {
3360 page = extent_buffer_page(eb, i);
3361 if (!PageUptodate(page)) {
3369 int read_extent_buffer_pages(struct extent_io_tree *tree,
3370 struct extent_buffer *eb,
3371 u64 start, int wait,
3372 get_extent_t *get_extent, int mirror_num)
3375 unsigned long start_i;
3379 int locked_pages = 0;
3380 int all_uptodate = 1;
3381 int inc_all_pages = 0;
3382 unsigned long num_pages;
3383 struct bio *bio = NULL;
3384 unsigned long bio_flags = 0;
3386 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3389 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3390 EXTENT_UPTODATE, 1, NULL)) {
3395 WARN_ON(start < eb->start);
3396 start_i = (start >> PAGE_CACHE_SHIFT) -
3397 (eb->start >> PAGE_CACHE_SHIFT);
3402 num_pages = num_extent_pages(eb->start, eb->len);
3403 for (i = start_i; i < num_pages; i++) {
3404 page = extent_buffer_page(eb, i);
3406 if (!trylock_page(page))
3412 if (!PageUptodate(page))
3417 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3421 for (i = start_i; i < num_pages; i++) {
3422 page = extent_buffer_page(eb, i);
3424 WARN_ON(!PagePrivate(page));
3426 set_page_extent_mapped(page);
3428 set_page_extent_head(page, eb->len);
3431 page_cache_get(page);
3432 if (!PageUptodate(page)) {
3435 ClearPageError(page);
3436 err = __extent_read_full_page(tree, page,
3438 mirror_num, &bio_flags);
3447 submit_one_bio(READ, bio, mirror_num, bio_flags);
3452 for (i = start_i; i < num_pages; i++) {
3453 page = extent_buffer_page(eb, i);
3454 wait_on_page_locked(page);
3455 if (!PageUptodate(page))
3460 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3465 while (locked_pages > 0) {
3466 page = extent_buffer_page(eb, i);
3474 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3475 unsigned long start,
3482 char *dst = (char *)dstv;
3483 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3484 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3486 WARN_ON(start > eb->len);
3487 WARN_ON(start + len > eb->start + eb->len);
3489 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3492 page = extent_buffer_page(eb, i);
3494 cur = min(len, (PAGE_CACHE_SIZE - offset));
3495 kaddr = kmap_atomic(page, KM_USER1);
3496 memcpy(dst, kaddr + offset, cur);
3497 kunmap_atomic(kaddr, KM_USER1);
3506 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3507 unsigned long min_len, char **token, char **map,
3508 unsigned long *map_start,
3509 unsigned long *map_len, int km)
3511 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3514 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3515 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3516 unsigned long end_i = (start_offset + start + min_len - 1) >>
3523 offset = start_offset;
3527 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3530 if (start + min_len > eb->len) {
3531 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3532 "wanted %lu %lu\n", (unsigned long long)eb->start,
3533 eb->len, start, min_len);
3538 p = extent_buffer_page(eb, i);
3539 kaddr = kmap_atomic(p, km);
3541 *map = kaddr + offset;
3542 *map_len = PAGE_CACHE_SIZE - offset;
3546 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3547 unsigned long min_len,
3548 char **token, char **map,
3549 unsigned long *map_start,
3550 unsigned long *map_len, int km)
3554 if (eb->map_token) {
3555 unmap_extent_buffer(eb, eb->map_token, km);
3556 eb->map_token = NULL;
3559 err = map_private_extent_buffer(eb, start, min_len, token, map,
3560 map_start, map_len, km);
3562 eb->map_token = *token;
3564 eb->map_start = *map_start;
3565 eb->map_len = *map_len;
3570 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3572 kunmap_atomic(token, km);
3575 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3576 unsigned long start,
3583 char *ptr = (char *)ptrv;
3584 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3585 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3588 WARN_ON(start > eb->len);
3589 WARN_ON(start + len > eb->start + eb->len);
3591 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3594 page = extent_buffer_page(eb, i);
3596 cur = min(len, (PAGE_CACHE_SIZE - offset));
3598 kaddr = kmap_atomic(page, KM_USER0);
3599 ret = memcmp(ptr, kaddr + offset, cur);
3600 kunmap_atomic(kaddr, KM_USER0);
3612 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3613 unsigned long start, unsigned long len)
3619 char *src = (char *)srcv;
3620 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3621 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3623 WARN_ON(start > eb->len);
3624 WARN_ON(start + len > eb->start + eb->len);
3626 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3629 page = extent_buffer_page(eb, i);
3630 WARN_ON(!PageUptodate(page));
3632 cur = min(len, PAGE_CACHE_SIZE - offset);
3633 kaddr = kmap_atomic(page, KM_USER1);
3634 memcpy(kaddr + offset, src, cur);
3635 kunmap_atomic(kaddr, KM_USER1);
3644 void memset_extent_buffer(struct extent_buffer *eb, char c,
3645 unsigned long start, unsigned long len)
3651 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3652 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3654 WARN_ON(start > eb->len);
3655 WARN_ON(start + len > eb->start + eb->len);
3657 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3660 page = extent_buffer_page(eb, i);
3661 WARN_ON(!PageUptodate(page));
3663 cur = min(len, PAGE_CACHE_SIZE - offset);
3664 kaddr = kmap_atomic(page, KM_USER0);
3665 memset(kaddr + offset, c, cur);
3666 kunmap_atomic(kaddr, KM_USER0);
3674 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3675 unsigned long dst_offset, unsigned long src_offset,
3678 u64 dst_len = dst->len;
3683 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3684 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3686 WARN_ON(src->len != dst_len);
3688 offset = (start_offset + dst_offset) &
3689 ((unsigned long)PAGE_CACHE_SIZE - 1);
3692 page = extent_buffer_page(dst, i);
3693 WARN_ON(!PageUptodate(page));
3695 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3697 kaddr = kmap_atomic(page, KM_USER0);
3698 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3699 kunmap_atomic(kaddr, KM_USER0);
3708 static void move_pages(struct page *dst_page, struct page *src_page,
3709 unsigned long dst_off, unsigned long src_off,
3712 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3713 if (dst_page == src_page) {
3714 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3716 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3717 char *p = dst_kaddr + dst_off + len;
3718 char *s = src_kaddr + src_off + len;
3723 kunmap_atomic(src_kaddr, KM_USER1);
3725 kunmap_atomic(dst_kaddr, KM_USER0);
3728 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
3730 unsigned long distance = (src > dst) ? src - dst : dst - src;
3731 return distance < len;
3734 static void copy_pages(struct page *dst_page, struct page *src_page,
3735 unsigned long dst_off, unsigned long src_off,
3738 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3741 if (dst_page != src_page) {
3742 src_kaddr = kmap_atomic(src_page, KM_USER1);
3744 src_kaddr = dst_kaddr;
3745 BUG_ON(areas_overlap(src_off, dst_off, len));
3748 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3749 kunmap_atomic(dst_kaddr, KM_USER0);
3750 if (dst_page != src_page)
3751 kunmap_atomic(src_kaddr, KM_USER1);
3754 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3755 unsigned long src_offset, unsigned long len)
3758 size_t dst_off_in_page;
3759 size_t src_off_in_page;
3760 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3761 unsigned long dst_i;
3762 unsigned long src_i;
3764 if (src_offset + len > dst->len) {
3765 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3766 "len %lu dst len %lu\n", src_offset, len, dst->len);
3769 if (dst_offset + len > dst->len) {
3770 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3771 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3776 dst_off_in_page = (start_offset + dst_offset) &
3777 ((unsigned long)PAGE_CACHE_SIZE - 1);
3778 src_off_in_page = (start_offset + src_offset) &
3779 ((unsigned long)PAGE_CACHE_SIZE - 1);
3781 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3782 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3784 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3786 cur = min_t(unsigned long, cur,
3787 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3789 copy_pages(extent_buffer_page(dst, dst_i),
3790 extent_buffer_page(dst, src_i),
3791 dst_off_in_page, src_off_in_page, cur);
3799 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3800 unsigned long src_offset, unsigned long len)
3803 size_t dst_off_in_page;
3804 size_t src_off_in_page;
3805 unsigned long dst_end = dst_offset + len - 1;
3806 unsigned long src_end = src_offset + len - 1;
3807 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3808 unsigned long dst_i;
3809 unsigned long src_i;
3811 if (src_offset + len > dst->len) {
3812 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3813 "len %lu len %lu\n", src_offset, len, dst->len);
3816 if (dst_offset + len > dst->len) {
3817 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3818 "len %lu len %lu\n", dst_offset, len, dst->len);
3821 if (!areas_overlap(src_offset, dst_offset, len)) {
3822 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3826 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3827 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3829 dst_off_in_page = (start_offset + dst_end) &
3830 ((unsigned long)PAGE_CACHE_SIZE - 1);
3831 src_off_in_page = (start_offset + src_end) &
3832 ((unsigned long)PAGE_CACHE_SIZE - 1);
3834 cur = min_t(unsigned long, len, src_off_in_page + 1);
3835 cur = min(cur, dst_off_in_page + 1);
3836 move_pages(extent_buffer_page(dst, dst_i),
3837 extent_buffer_page(dst, src_i),
3838 dst_off_in_page - cur + 1,
3839 src_off_in_page - cur + 1, cur);
3847 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3849 struct extent_buffer *eb =
3850 container_of(head, struct extent_buffer, rcu_head);
3852 btrfs_release_extent_buffer(eb);
3855 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3857 u64 start = page_offset(page);
3858 struct extent_buffer *eb;
3861 spin_lock(&tree->buffer_lock);
3862 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3864 spin_unlock(&tree->buffer_lock);
3868 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3874 * set @eb->refs to 0 if it is already 1, and then release the @eb.
3877 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
3882 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3884 spin_unlock(&tree->buffer_lock);
3886 /* at this point we can safely release the extent buffer */
3887 if (atomic_read(&eb->refs) == 0)
3888 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);