1 #include <linux/bitops.h>
2 #include <linux/slab.h>
6 #include <linux/pagemap.h>
7 #include <linux/page-flags.h>
8 #include <linux/module.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 static struct kmem_cache *extent_state_cache;
21 static struct kmem_cache *extent_buffer_cache;
23 static LIST_HEAD(buffers);
24 static LIST_HEAD(states);
28 static DEFINE_SPINLOCK(leak_lock);
31 #define BUFFER_LRU_MAX 64
36 struct rb_node rb_node;
39 struct extent_page_data {
41 struct extent_io_tree *tree;
42 get_extent_t *get_extent;
44 /* tells writepage not to lock the state bits for this range
45 * it still does the unlocking
47 unsigned int extent_locked:1;
49 /* tells the submit_bio code to use a WRITE_SYNC */
50 unsigned int sync_io:1;
53 int __init extent_io_init(void)
55 extent_state_cache = kmem_cache_create("extent_state",
56 sizeof(struct extent_state), 0,
57 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
58 if (!extent_state_cache)
61 extent_buffer_cache = kmem_cache_create("extent_buffers",
62 sizeof(struct extent_buffer), 0,
63 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
64 if (!extent_buffer_cache)
65 goto free_state_cache;
69 kmem_cache_destroy(extent_state_cache);
73 void extent_io_exit(void)
75 struct extent_state *state;
76 struct extent_buffer *eb;
78 while (!list_empty(&states)) {
79 state = list_entry(states.next, struct extent_state, leak_list);
80 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
81 "state %lu in tree %p refs %d\n",
82 (unsigned long long)state->start,
83 (unsigned long long)state->end,
84 state->state, state->tree, atomic_read(&state->refs));
85 list_del(&state->leak_list);
86 kmem_cache_free(extent_state_cache, state);
90 while (!list_empty(&buffers)) {
91 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
92 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
93 "refs %d\n", (unsigned long long)eb->start,
94 eb->len, atomic_read(&eb->refs));
95 list_del(&eb->leak_list);
96 kmem_cache_free(extent_buffer_cache, eb);
98 if (extent_state_cache)
99 kmem_cache_destroy(extent_state_cache);
100 if (extent_buffer_cache)
101 kmem_cache_destroy(extent_buffer_cache);
104 void extent_io_tree_init(struct extent_io_tree *tree,
105 struct address_space *mapping, gfp_t mask)
107 tree->state = RB_ROOT;
108 tree->buffer = RB_ROOT;
110 tree->dirty_bytes = 0;
111 spin_lock_init(&tree->lock);
112 spin_lock_init(&tree->buffer_lock);
113 tree->mapping = mapping;
116 static struct extent_state *alloc_extent_state(gfp_t mask)
118 struct extent_state *state;
123 state = kmem_cache_alloc(extent_state_cache, mask);
130 spin_lock_irqsave(&leak_lock, flags);
131 list_add(&state->leak_list, &states);
132 spin_unlock_irqrestore(&leak_lock, flags);
134 atomic_set(&state->refs, 1);
135 init_waitqueue_head(&state->wq);
139 static void free_extent_state(struct extent_state *state)
143 if (atomic_dec_and_test(&state->refs)) {
147 WARN_ON(state->tree);
149 spin_lock_irqsave(&leak_lock, flags);
150 list_del(&state->leak_list);
151 spin_unlock_irqrestore(&leak_lock, flags);
153 kmem_cache_free(extent_state_cache, state);
157 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
158 struct rb_node *node)
160 struct rb_node **p = &root->rb_node;
161 struct rb_node *parent = NULL;
162 struct tree_entry *entry;
166 entry = rb_entry(parent, struct tree_entry, rb_node);
168 if (offset < entry->start)
170 else if (offset > entry->end)
176 entry = rb_entry(node, struct tree_entry, rb_node);
177 rb_link_node(node, parent, p);
178 rb_insert_color(node, root);
182 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
183 struct rb_node **prev_ret,
184 struct rb_node **next_ret)
186 struct rb_root *root = &tree->state;
187 struct rb_node *n = root->rb_node;
188 struct rb_node *prev = NULL;
189 struct rb_node *orig_prev = NULL;
190 struct tree_entry *entry;
191 struct tree_entry *prev_entry = NULL;
194 entry = rb_entry(n, struct tree_entry, rb_node);
198 if (offset < entry->start)
200 else if (offset > entry->end)
208 while (prev && offset > prev_entry->end) {
209 prev = rb_next(prev);
210 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
217 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
218 while (prev && offset < prev_entry->start) {
219 prev = rb_prev(prev);
220 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
227 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
230 struct rb_node *prev = NULL;
233 ret = __etree_search(tree, offset, &prev, NULL);
239 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
240 u64 offset, struct rb_node *node)
242 struct rb_root *root = &tree->buffer;
243 struct rb_node **p = &root->rb_node;
244 struct rb_node *parent = NULL;
245 struct extent_buffer *eb;
249 eb = rb_entry(parent, struct extent_buffer, rb_node);
251 if (offset < eb->start)
253 else if (offset > eb->start)
259 rb_link_node(node, parent, p);
260 rb_insert_color(node, root);
264 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
267 struct rb_root *root = &tree->buffer;
268 struct rb_node *n = root->rb_node;
269 struct extent_buffer *eb;
272 eb = rb_entry(n, struct extent_buffer, rb_node);
273 if (offset < eb->start)
275 else if (offset > eb->start)
283 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
284 struct extent_state *other)
286 if (tree->ops && tree->ops->merge_extent_hook)
287 tree->ops->merge_extent_hook(tree->mapping->host, new,
292 * utility function to look for merge candidates inside a given range.
293 * Any extents with matching state are merged together into a single
294 * extent in the tree. Extents with EXTENT_IO in their state field
295 * are not merged because the end_io handlers need to be able to do
296 * operations on them without sleeping (or doing allocations/splits).
298 * This should be called with the tree lock held.
300 static int merge_state(struct extent_io_tree *tree,
301 struct extent_state *state)
303 struct extent_state *other;
304 struct rb_node *other_node;
306 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
309 other_node = rb_prev(&state->rb_node);
311 other = rb_entry(other_node, struct extent_state, rb_node);
312 if (other->end == state->start - 1 &&
313 other->state == state->state) {
314 merge_cb(tree, state, other);
315 state->start = other->start;
317 rb_erase(&other->rb_node, &tree->state);
318 free_extent_state(other);
321 other_node = rb_next(&state->rb_node);
323 other = rb_entry(other_node, struct extent_state, rb_node);
324 if (other->start == state->end + 1 &&
325 other->state == state->state) {
326 merge_cb(tree, state, other);
327 other->start = state->start;
329 rb_erase(&state->rb_node, &tree->state);
330 free_extent_state(state);
338 static int set_state_cb(struct extent_io_tree *tree,
339 struct extent_state *state,
342 if (tree->ops && tree->ops->set_bit_hook) {
343 return tree->ops->set_bit_hook(tree->mapping->host,
344 state->start, state->end,
351 static void clear_state_cb(struct extent_io_tree *tree,
352 struct extent_state *state,
355 if (tree->ops && tree->ops->clear_bit_hook)
356 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
360 * insert an extent_state struct into the tree. 'bits' are set on the
361 * struct before it is inserted.
363 * This may return -EEXIST if the extent is already there, in which case the
364 * state struct is freed.
366 * The tree lock is not taken internally. This is a utility function and
367 * probably isn't what you want to call (see set/clear_extent_bit).
369 static int insert_state(struct extent_io_tree *tree,
370 struct extent_state *state, u64 start, u64 end,
373 struct rb_node *node;
377 printk(KERN_ERR "btrfs end < start %llu %llu\n",
378 (unsigned long long)end,
379 (unsigned long long)start);
382 state->start = start;
384 ret = set_state_cb(tree, state, bits);
388 if (bits & EXTENT_DIRTY)
389 tree->dirty_bytes += end - start + 1;
390 state->state |= bits;
391 node = tree_insert(&tree->state, end, &state->rb_node);
393 struct extent_state *found;
394 found = rb_entry(node, struct extent_state, rb_node);
395 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
396 "%llu %llu\n", (unsigned long long)found->start,
397 (unsigned long long)found->end,
398 (unsigned long long)start, (unsigned long long)end);
399 free_extent_state(state);
403 merge_state(tree, state);
407 static int split_cb(struct extent_io_tree *tree, struct extent_state *orig,
410 if (tree->ops && tree->ops->split_extent_hook)
411 return tree->ops->split_extent_hook(tree->mapping->host,
417 * split a given extent state struct in two, inserting the preallocated
418 * struct 'prealloc' as the newly created second half. 'split' indicates an
419 * offset inside 'orig' where it should be split.
422 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
423 * are two extent state structs in the tree:
424 * prealloc: [orig->start, split - 1]
425 * orig: [ split, orig->end ]
427 * The tree locks are not taken by this function. They need to be held
430 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
431 struct extent_state *prealloc, u64 split)
433 struct rb_node *node;
435 split_cb(tree, orig, split);
437 prealloc->start = orig->start;
438 prealloc->end = split - 1;
439 prealloc->state = orig->state;
442 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
444 free_extent_state(prealloc);
447 prealloc->tree = tree;
452 * utility function to clear some bits in an extent state struct.
453 * it will optionally wake up any one waiting on this state (wake == 1), or
454 * forcibly remove the state from the tree (delete == 1).
456 * If no bits are set on the state struct after clearing things, the
457 * struct is freed and removed from the tree
459 static int clear_state_bit(struct extent_io_tree *tree,
460 struct extent_state *state, int bits, int wake,
463 int bits_to_clear = bits & ~EXTENT_DO_ACCOUNTING;
464 int ret = state->state & bits_to_clear;
466 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
467 u64 range = state->end - state->start + 1;
468 WARN_ON(range > tree->dirty_bytes);
469 tree->dirty_bytes -= range;
471 clear_state_cb(tree, state, bits);
472 state->state &= ~bits_to_clear;
475 if (delete || state->state == 0) {
477 clear_state_cb(tree, state, state->state);
478 rb_erase(&state->rb_node, &tree->state);
480 free_extent_state(state);
485 merge_state(tree, state);
491 * clear some bits on a range in the tree. This may require splitting
492 * or inserting elements in the tree, so the gfp mask is used to
493 * indicate which allocations or sleeping are allowed.
495 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
496 * the given range from the tree regardless of state (ie for truncate).
498 * the range [start, end] is inclusive.
500 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
501 * bits were already set, or zero if none of the bits were already set.
503 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
504 int bits, int wake, int delete,
505 struct extent_state **cached_state,
508 struct extent_state *state;
509 struct extent_state *cached;
510 struct extent_state *prealloc = NULL;
511 struct rb_node *next_node;
512 struct rb_node *node;
518 if (!prealloc && (mask & __GFP_WAIT)) {
519 prealloc = alloc_extent_state(mask);
524 spin_lock(&tree->lock);
526 cached = *cached_state;
527 *cached_state = NULL;
529 if (cached && cached->tree && cached->start == start) {
530 atomic_dec(&cached->refs);
534 free_extent_state(cached);
537 * this search will find the extents that end after
540 node = tree_search(tree, start);
543 state = rb_entry(node, struct extent_state, rb_node);
545 if (state->start > end)
547 WARN_ON(state->end < start);
548 last_end = state->end;
551 * | ---- desired range ---- |
553 * | ------------- state -------------- |
555 * We need to split the extent we found, and may flip
556 * bits on second half.
558 * If the extent we found extends past our range, we
559 * just split and search again. It'll get split again
560 * the next time though.
562 * If the extent we found is inside our range, we clear
563 * the desired bit on it.
566 if (state->start < start) {
568 prealloc = alloc_extent_state(GFP_ATOMIC);
569 err = split_state(tree, state, prealloc, start);
570 BUG_ON(err == -EEXIST);
574 if (state->end <= end) {
575 set |= clear_state_bit(tree, state, bits, wake,
577 if (last_end == (u64)-1)
579 start = last_end + 1;
584 * | ---- desired range ---- |
586 * We need to split the extent, and clear the bit
589 if (state->start <= end && state->end > end) {
591 prealloc = alloc_extent_state(GFP_ATOMIC);
592 err = split_state(tree, state, prealloc, end + 1);
593 BUG_ON(err == -EEXIST);
597 set |= clear_state_bit(tree, prealloc, bits, wake, delete);
603 if (state->end < end && prealloc && !need_resched())
604 next_node = rb_next(&state->rb_node);
608 set |= clear_state_bit(tree, state, bits, wake, delete);
609 if (last_end == (u64)-1)
611 start = last_end + 1;
612 if (start <= end && next_node) {
613 state = rb_entry(next_node, struct extent_state,
615 if (state->start == start)
621 spin_unlock(&tree->lock);
623 free_extent_state(prealloc);
630 spin_unlock(&tree->lock);
631 if (mask & __GFP_WAIT)
636 static int wait_on_state(struct extent_io_tree *tree,
637 struct extent_state *state)
638 __releases(tree->lock)
639 __acquires(tree->lock)
642 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
643 spin_unlock(&tree->lock);
645 spin_lock(&tree->lock);
646 finish_wait(&state->wq, &wait);
651 * waits for one or more bits to clear on a range in the state tree.
652 * The range [start, end] is inclusive.
653 * The tree lock is taken by this function
655 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
657 struct extent_state *state;
658 struct rb_node *node;
660 spin_lock(&tree->lock);
664 * this search will find all the extents that end after
667 node = tree_search(tree, start);
671 state = rb_entry(node, struct extent_state, rb_node);
673 if (state->start > end)
676 if (state->state & bits) {
677 start = state->start;
678 atomic_inc(&state->refs);
679 wait_on_state(tree, state);
680 free_extent_state(state);
683 start = state->end + 1;
688 if (need_resched()) {
689 spin_unlock(&tree->lock);
691 spin_lock(&tree->lock);
695 spin_unlock(&tree->lock);
699 static int set_state_bits(struct extent_io_tree *tree,
700 struct extent_state *state,
705 ret = set_state_cb(tree, state, bits);
709 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
710 u64 range = state->end - state->start + 1;
711 tree->dirty_bytes += range;
713 state->state |= bits;
718 static void cache_state(struct extent_state *state,
719 struct extent_state **cached_ptr)
721 if (cached_ptr && !(*cached_ptr)) {
722 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
724 atomic_inc(&state->refs);
730 * set some bits on a range in the tree. This may require allocations or
731 * sleeping, so the gfp mask is used to indicate what is allowed.
733 * If any of the exclusive bits are set, this will fail with -EEXIST if some
734 * part of the range already has the desired bits set. The start of the
735 * existing range is returned in failed_start in this case.
737 * [start, end] is inclusive This takes the tree lock.
740 static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
741 int bits, int exclusive_bits, u64 *failed_start,
742 struct extent_state **cached_state,
745 struct extent_state *state;
746 struct extent_state *prealloc = NULL;
747 struct rb_node *node;
753 if (!prealloc && (mask & __GFP_WAIT)) {
754 prealloc = alloc_extent_state(mask);
759 spin_lock(&tree->lock);
760 if (cached_state && *cached_state) {
761 state = *cached_state;
762 if (state->start == start && state->tree) {
763 node = &state->rb_node;
768 * this search will find all the extents that end after
771 node = tree_search(tree, start);
773 err = insert_state(tree, prealloc, start, end, bits);
775 BUG_ON(err == -EEXIST);
778 state = rb_entry(node, struct extent_state, rb_node);
780 last_start = state->start;
781 last_end = state->end;
784 * | ---- desired range ---- |
787 * Just lock what we found and keep going
789 if (state->start == start && state->end <= end) {
790 struct rb_node *next_node;
791 if (state->state & exclusive_bits) {
792 *failed_start = state->start;
797 err = set_state_bits(tree, state, bits);
801 cache_state(state, cached_state);
802 merge_state(tree, state);
803 if (last_end == (u64)-1)
806 start = last_end + 1;
807 if (start < end && prealloc && !need_resched()) {
808 next_node = rb_next(node);
810 state = rb_entry(next_node, struct extent_state,
812 if (state->start == start)
820 * | ---- desired range ---- |
823 * | ------------- state -------------- |
825 * We need to split the extent we found, and may flip bits on
828 * If the extent we found extends past our
829 * range, we just split and search again. It'll get split
830 * again the next time though.
832 * If the extent we found is inside our range, we set the
835 if (state->start < start) {
836 if (state->state & exclusive_bits) {
837 *failed_start = start;
841 err = split_state(tree, state, prealloc, start);
842 BUG_ON(err == -EEXIST);
846 if (state->end <= end) {
847 err = set_state_bits(tree, state, bits);
850 cache_state(state, cached_state);
851 merge_state(tree, state);
852 if (last_end == (u64)-1)
854 start = last_end + 1;
859 * | ---- desired range ---- |
860 * | state | or | state |
862 * There's a hole, we need to insert something in it and
863 * ignore the extent we found.
865 if (state->start > start) {
867 if (end < last_start)
870 this_end = last_start - 1;
871 err = insert_state(tree, prealloc, start, this_end,
873 BUG_ON(err == -EEXIST);
878 cache_state(prealloc, cached_state);
880 start = this_end + 1;
884 * | ---- desired range ---- |
886 * We need to split the extent, and set the bit
889 if (state->start <= end && state->end > end) {
890 if (state->state & exclusive_bits) {
891 *failed_start = start;
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,
951 return set_extent_bit(tree, start, end,
952 EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
953 0, NULL, NULL, 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,
965 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
968 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
972 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
975 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0,
979 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
982 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
986 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
989 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
993 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
995 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
999 * either insert or lock state struct between start and end use mask to tell
1000 * us if waiting is desired.
1002 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1003 int bits, struct extent_state **cached_state, gfp_t mask)
1008 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1009 EXTENT_LOCKED, &failed_start,
1010 cached_state, mask);
1011 if (err == -EEXIST && (mask & __GFP_WAIT)) {
1012 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1013 start = failed_start;
1017 WARN_ON(start > end);
1022 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1024 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1027 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1033 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1034 &failed_start, NULL, mask);
1035 if (err == -EEXIST) {
1036 if (failed_start > start)
1037 clear_extent_bit(tree, start, failed_start - 1,
1038 EXTENT_LOCKED, 1, 0, NULL, mask);
1044 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1045 struct extent_state **cached, gfp_t mask)
1047 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1051 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1054 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1059 * helper function to set pages and extents in the tree dirty
1061 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
1063 unsigned long index = start >> PAGE_CACHE_SHIFT;
1064 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1067 while (index <= end_index) {
1068 page = find_get_page(tree->mapping, index);
1070 __set_page_dirty_nobuffers(page);
1071 page_cache_release(page);
1078 * helper function to set both pages and extents in the tree writeback
1080 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1082 unsigned long index = start >> PAGE_CACHE_SHIFT;
1083 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1086 while (index <= end_index) {
1087 page = find_get_page(tree->mapping, index);
1089 set_page_writeback(page);
1090 page_cache_release(page);
1097 * find the first offset in the io tree with 'bits' set. zero is
1098 * returned if we find something, and *start_ret and *end_ret are
1099 * set to reflect the state struct that was found.
1101 * If nothing was found, 1 is returned, < 0 on error
1103 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1104 u64 *start_ret, u64 *end_ret, int bits)
1106 struct rb_node *node;
1107 struct extent_state *state;
1110 spin_lock(&tree->lock);
1112 * this search will find all the extents that end after
1115 node = tree_search(tree, start);
1120 state = rb_entry(node, struct extent_state, rb_node);
1121 if (state->end >= start && (state->state & bits)) {
1122 *start_ret = state->start;
1123 *end_ret = state->end;
1127 node = rb_next(node);
1132 spin_unlock(&tree->lock);
1136 /* find the first state struct with 'bits' set after 'start', and
1137 * return it. tree->lock must be held. NULL will returned if
1138 * nothing was found after 'start'
1140 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1141 u64 start, int bits)
1143 struct rb_node *node;
1144 struct extent_state *state;
1147 * this search will find all the extents that end after
1150 node = tree_search(tree, start);
1155 state = rb_entry(node, struct extent_state, rb_node);
1156 if (state->end >= start && (state->state & bits))
1159 node = rb_next(node);
1168 * find a contiguous range of bytes in the file marked as delalloc, not
1169 * more than 'max_bytes'. start and end are used to return the range,
1171 * 1 is returned if we find something, 0 if nothing was in the tree
1173 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1174 u64 *start, u64 *end, u64 max_bytes,
1175 struct extent_state **cached_state)
1177 struct rb_node *node;
1178 struct extent_state *state;
1179 u64 cur_start = *start;
1181 u64 total_bytes = 0;
1183 spin_lock(&tree->lock);
1186 * this search will find all the extents that end after
1189 node = tree_search(tree, cur_start);
1197 state = rb_entry(node, struct extent_state, rb_node);
1198 if (found && (state->start != cur_start ||
1199 (state->state & EXTENT_BOUNDARY))) {
1202 if (!(state->state & EXTENT_DELALLOC)) {
1208 *start = state->start;
1209 *cached_state = state;
1210 atomic_inc(&state->refs);
1214 cur_start = state->end + 1;
1215 node = rb_next(node);
1218 total_bytes += state->end - state->start + 1;
1219 if (total_bytes >= max_bytes)
1223 spin_unlock(&tree->lock);
1227 static noinline int __unlock_for_delalloc(struct inode *inode,
1228 struct page *locked_page,
1232 struct page *pages[16];
1233 unsigned long index = start >> PAGE_CACHE_SHIFT;
1234 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1235 unsigned long nr_pages = end_index - index + 1;
1238 if (index == locked_page->index && end_index == index)
1241 while (nr_pages > 0) {
1242 ret = find_get_pages_contig(inode->i_mapping, index,
1243 min_t(unsigned long, nr_pages,
1244 ARRAY_SIZE(pages)), pages);
1245 for (i = 0; i < ret; i++) {
1246 if (pages[i] != locked_page)
1247 unlock_page(pages[i]);
1248 page_cache_release(pages[i]);
1257 static noinline int lock_delalloc_pages(struct inode *inode,
1258 struct page *locked_page,
1262 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1263 unsigned long start_index = index;
1264 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1265 unsigned long pages_locked = 0;
1266 struct page *pages[16];
1267 unsigned long nrpages;
1271 /* the caller is responsible for locking the start index */
1272 if (index == locked_page->index && index == end_index)
1275 /* skip the page at the start index */
1276 nrpages = end_index - index + 1;
1277 while (nrpages > 0) {
1278 ret = find_get_pages_contig(inode->i_mapping, index,
1279 min_t(unsigned long,
1280 nrpages, ARRAY_SIZE(pages)), pages);
1285 /* now we have an array of pages, lock them all */
1286 for (i = 0; i < ret; i++) {
1288 * the caller is taking responsibility for
1291 if (pages[i] != locked_page) {
1292 lock_page(pages[i]);
1293 if (!PageDirty(pages[i]) ||
1294 pages[i]->mapping != inode->i_mapping) {
1296 unlock_page(pages[i]);
1297 page_cache_release(pages[i]);
1301 page_cache_release(pages[i]);
1310 if (ret && pages_locked) {
1311 __unlock_for_delalloc(inode, locked_page,
1313 ((u64)(start_index + pages_locked - 1)) <<
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_lock_delalloc_range(struct inode *inode,
1326 struct extent_io_tree *tree,
1327 struct page *locked_page,
1328 u64 *start, u64 *end,
1334 struct extent_state *cached_state = NULL;
1339 /* step one, find a bunch of delalloc bytes starting at start */
1340 delalloc_start = *start;
1342 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1343 max_bytes, &cached_state);
1344 if (!found || delalloc_end <= *start) {
1345 *start = delalloc_start;
1346 *end = delalloc_end;
1347 free_extent_state(cached_state);
1352 * start comes from the offset of locked_page. We have to lock
1353 * pages in order, so we can't process delalloc bytes before
1356 if (delalloc_start < *start)
1357 delalloc_start = *start;
1360 * make sure to limit the number of pages we try to lock down
1363 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1364 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1366 /* step two, lock all the pages after the page that has start */
1367 ret = lock_delalloc_pages(inode, locked_page,
1368 delalloc_start, delalloc_end);
1369 if (ret == -EAGAIN) {
1370 /* some of the pages are gone, lets avoid looping by
1371 * shortening the size of the delalloc range we're searching
1373 free_extent_state(cached_state);
1375 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1376 max_bytes = PAGE_CACHE_SIZE - offset;
1386 /* step three, lock the state bits for the whole range */
1387 lock_extent_bits(tree, delalloc_start, delalloc_end,
1388 0, &cached_state, GFP_NOFS);
1390 /* then test to make sure it is all still delalloc */
1391 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1392 EXTENT_DELALLOC, 1, cached_state);
1394 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1395 &cached_state, GFP_NOFS);
1396 __unlock_for_delalloc(inode, locked_page,
1397 delalloc_start, delalloc_end);
1401 free_extent_state(cached_state);
1402 *start = delalloc_start;
1403 *end = delalloc_end;
1408 int extent_clear_unlock_delalloc(struct inode *inode,
1409 struct extent_io_tree *tree,
1410 u64 start, u64 end, struct page *locked_page,
1414 struct page *pages[16];
1415 unsigned long index = start >> PAGE_CACHE_SHIFT;
1416 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1417 unsigned long nr_pages = end_index - index + 1;
1421 if (op & EXTENT_CLEAR_UNLOCK)
1422 clear_bits |= EXTENT_LOCKED;
1423 if (op & EXTENT_CLEAR_DIRTY)
1424 clear_bits |= EXTENT_DIRTY;
1426 if (op & EXTENT_CLEAR_DELALLOC)
1427 clear_bits |= EXTENT_DELALLOC;
1429 if (op & EXTENT_CLEAR_ACCOUNTING)
1430 clear_bits |= EXTENT_DO_ACCOUNTING;
1432 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1433 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1434 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1435 EXTENT_SET_PRIVATE2)))
1438 while (nr_pages > 0) {
1439 ret = find_get_pages_contig(inode->i_mapping, index,
1440 min_t(unsigned long,
1441 nr_pages, ARRAY_SIZE(pages)), pages);
1442 for (i = 0; i < ret; i++) {
1444 if (op & EXTENT_SET_PRIVATE2)
1445 SetPagePrivate2(pages[i]);
1447 if (pages[i] == locked_page) {
1448 page_cache_release(pages[i]);
1451 if (op & EXTENT_CLEAR_DIRTY)
1452 clear_page_dirty_for_io(pages[i]);
1453 if (op & EXTENT_SET_WRITEBACK)
1454 set_page_writeback(pages[i]);
1455 if (op & EXTENT_END_WRITEBACK)
1456 end_page_writeback(pages[i]);
1457 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1458 unlock_page(pages[i]);
1459 page_cache_release(pages[i]);
1469 * count the number of bytes in the tree that have a given bit(s)
1470 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1471 * cached. The total number found is returned.
1473 u64 count_range_bits(struct extent_io_tree *tree,
1474 u64 *start, u64 search_end, u64 max_bytes,
1477 struct rb_node *node;
1478 struct extent_state *state;
1479 u64 cur_start = *start;
1480 u64 total_bytes = 0;
1483 if (search_end <= cur_start) {
1488 spin_lock(&tree->lock);
1489 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1490 total_bytes = tree->dirty_bytes;
1494 * this search will find all the extents that end after
1497 node = tree_search(tree, cur_start);
1502 state = rb_entry(node, struct extent_state, rb_node);
1503 if (state->start > search_end)
1505 if (state->end >= cur_start && (state->state & bits)) {
1506 total_bytes += min(search_end, state->end) + 1 -
1507 max(cur_start, state->start);
1508 if (total_bytes >= max_bytes)
1511 *start = state->start;
1515 node = rb_next(node);
1520 spin_unlock(&tree->lock);
1525 * set the private field for a given byte offset in the tree. If there isn't
1526 * an extent_state there already, this does nothing.
1528 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1530 struct rb_node *node;
1531 struct extent_state *state;
1534 spin_lock(&tree->lock);
1536 * this search will find all the extents that end after
1539 node = tree_search(tree, start);
1544 state = rb_entry(node, struct extent_state, rb_node);
1545 if (state->start != start) {
1549 state->private = private;
1551 spin_unlock(&tree->lock);
1555 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1557 struct rb_node *node;
1558 struct extent_state *state;
1561 spin_lock(&tree->lock);
1563 * this search will find all the extents that end after
1566 node = tree_search(tree, start);
1571 state = rb_entry(node, struct extent_state, rb_node);
1572 if (state->start != start) {
1576 *private = state->private;
1578 spin_unlock(&tree->lock);
1583 * searches a range in the state tree for a given mask.
1584 * If 'filled' == 1, this returns 1 only if every extent in the tree
1585 * has the bits set. Otherwise, 1 is returned if any bit in the
1586 * range is found set.
1588 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1589 int bits, int filled, struct extent_state *cached)
1591 struct extent_state *state = NULL;
1592 struct rb_node *node;
1595 spin_lock(&tree->lock);
1596 if (cached && cached->tree && cached->start == start)
1597 node = &cached->rb_node;
1599 node = tree_search(tree, start);
1600 while (node && start <= end) {
1601 state = rb_entry(node, struct extent_state, rb_node);
1603 if (filled && state->start > start) {
1608 if (state->start > end)
1611 if (state->state & bits) {
1615 } else if (filled) {
1620 if (state->end == (u64)-1)
1623 start = state->end + 1;
1626 node = rb_next(node);
1633 spin_unlock(&tree->lock);
1638 * helper function to set a given page up to date if all the
1639 * extents in the tree for that page are up to date
1641 static int check_page_uptodate(struct extent_io_tree *tree,
1644 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1645 u64 end = start + PAGE_CACHE_SIZE - 1;
1646 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1647 SetPageUptodate(page);
1652 * helper function to unlock a page if all the extents in the tree
1653 * for that page are unlocked
1655 static int check_page_locked(struct extent_io_tree *tree,
1658 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1659 u64 end = start + PAGE_CACHE_SIZE - 1;
1660 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1666 * helper function to end page writeback if all the extents
1667 * in the tree for that page are done with writeback
1669 static int check_page_writeback(struct extent_io_tree *tree,
1672 end_page_writeback(page);
1676 /* lots and lots of room for performance fixes in the end_bio funcs */
1679 * after a writepage IO is done, we need to:
1680 * clear the uptodate bits on error
1681 * clear the writeback bits in the extent tree for this IO
1682 * end_page_writeback if the page has no more pending IO
1684 * Scheduling is not allowed, so the extent state tree is expected
1685 * to have one and only one object corresponding to this IO.
1687 static void end_bio_extent_writepage(struct bio *bio, int err)
1689 int uptodate = err == 0;
1690 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1691 struct extent_io_tree *tree;
1698 struct page *page = bvec->bv_page;
1699 tree = &BTRFS_I(page->mapping->host)->io_tree;
1701 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1703 end = start + bvec->bv_len - 1;
1705 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1710 if (--bvec >= bio->bi_io_vec)
1711 prefetchw(&bvec->bv_page->flags);
1712 if (tree->ops && tree->ops->writepage_end_io_hook) {
1713 ret = tree->ops->writepage_end_io_hook(page, start,
1714 end, NULL, uptodate);
1719 if (!uptodate && tree->ops &&
1720 tree->ops->writepage_io_failed_hook) {
1721 ret = tree->ops->writepage_io_failed_hook(bio, page,
1724 uptodate = (err == 0);
1730 clear_extent_uptodate(tree, start, end, GFP_NOFS);
1731 ClearPageUptodate(page);
1736 end_page_writeback(page);
1738 check_page_writeback(tree, page);
1739 } while (bvec >= bio->bi_io_vec);
1745 * after a readpage IO is done, we need to:
1746 * clear the uptodate bits on error
1747 * set the uptodate bits if things worked
1748 * set the page up to date if all extents in the tree are uptodate
1749 * clear the lock bit in the extent tree
1750 * unlock the page if there are no other extents locked for it
1752 * Scheduling is not allowed, so the extent state tree is expected
1753 * to have one and only one object corresponding to this IO.
1755 static void end_bio_extent_readpage(struct bio *bio, int err)
1757 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1758 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
1759 struct bio_vec *bvec = bio->bi_io_vec;
1760 struct extent_io_tree *tree;
1770 struct page *page = bvec->bv_page;
1771 tree = &BTRFS_I(page->mapping->host)->io_tree;
1773 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1775 end = start + bvec->bv_len - 1;
1777 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1782 if (++bvec <= bvec_end)
1783 prefetchw(&bvec->bv_page->flags);
1785 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1786 ret = tree->ops->readpage_end_io_hook(page, start, end,
1791 if (!uptodate && tree->ops &&
1792 tree->ops->readpage_io_failed_hook) {
1793 ret = tree->ops->readpage_io_failed_hook(bio, page,
1797 test_bit(BIO_UPTODATE, &bio->bi_flags);
1805 set_extent_uptodate(tree, start, end,
1808 unlock_extent(tree, start, end, GFP_ATOMIC);
1812 SetPageUptodate(page);
1814 ClearPageUptodate(page);
1820 check_page_uptodate(tree, page);
1822 ClearPageUptodate(page);
1825 check_page_locked(tree, page);
1827 } while (bvec <= bvec_end);
1833 * IO done from prepare_write is pretty simple, we just unlock
1834 * the structs in the extent tree when done, and set the uptodate bits
1837 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1839 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1840 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1841 struct extent_io_tree *tree;
1846 struct page *page = bvec->bv_page;
1847 tree = &BTRFS_I(page->mapping->host)->io_tree;
1849 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1851 end = start + bvec->bv_len - 1;
1853 if (--bvec >= bio->bi_io_vec)
1854 prefetchw(&bvec->bv_page->flags);
1857 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1859 ClearPageUptodate(page);
1863 unlock_extent(tree, start, end, GFP_ATOMIC);
1865 } while (bvec >= bio->bi_io_vec);
1871 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1876 bio = bio_alloc(gfp_flags, nr_vecs);
1878 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1879 while (!bio && (nr_vecs /= 2))
1880 bio = bio_alloc(gfp_flags, nr_vecs);
1885 bio->bi_bdev = bdev;
1886 bio->bi_sector = first_sector;
1891 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1892 unsigned long bio_flags)
1895 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1896 struct page *page = bvec->bv_page;
1897 struct extent_io_tree *tree = bio->bi_private;
1901 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1902 end = start + bvec->bv_len - 1;
1904 bio->bi_private = NULL;
1908 if (tree->ops && tree->ops->submit_bio_hook)
1909 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1910 mirror_num, bio_flags);
1912 submit_bio(rw, bio);
1913 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1919 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1920 struct page *page, sector_t sector,
1921 size_t size, unsigned long offset,
1922 struct block_device *bdev,
1923 struct bio **bio_ret,
1924 unsigned long max_pages,
1925 bio_end_io_t end_io_func,
1927 unsigned long prev_bio_flags,
1928 unsigned long bio_flags)
1934 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1935 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1936 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1938 if (bio_ret && *bio_ret) {
1941 contig = bio->bi_sector == sector;
1943 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1946 if (prev_bio_flags != bio_flags || !contig ||
1947 (tree->ops && tree->ops->merge_bio_hook &&
1948 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1950 bio_add_page(bio, page, page_size, offset) < page_size) {
1951 ret = submit_one_bio(rw, bio, mirror_num,
1958 if (this_compressed)
1961 nr = bio_get_nr_vecs(bdev);
1963 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1965 bio_add_page(bio, page, page_size, offset);
1966 bio->bi_end_io = end_io_func;
1967 bio->bi_private = tree;
1972 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1977 void set_page_extent_mapped(struct page *page)
1979 if (!PagePrivate(page)) {
1980 SetPagePrivate(page);
1981 page_cache_get(page);
1982 set_page_private(page, EXTENT_PAGE_PRIVATE);
1986 static void set_page_extent_head(struct page *page, unsigned long len)
1988 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1992 * basic readpage implementation. Locked extent state structs are inserted
1993 * into the tree that are removed when the IO is done (by the end_io
1996 static int __extent_read_full_page(struct extent_io_tree *tree,
1998 get_extent_t *get_extent,
1999 struct bio **bio, int mirror_num,
2000 unsigned long *bio_flags)
2002 struct inode *inode = page->mapping->host;
2003 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2004 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2008 u64 last_byte = i_size_read(inode);
2012 struct extent_map *em;
2013 struct block_device *bdev;
2016 size_t page_offset = 0;
2018 size_t disk_io_size;
2019 size_t blocksize = inode->i_sb->s_blocksize;
2020 unsigned long this_bio_flag = 0;
2022 set_page_extent_mapped(page);
2025 lock_extent(tree, start, end, GFP_NOFS);
2027 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2029 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2032 iosize = PAGE_CACHE_SIZE - zero_offset;
2033 userpage = kmap_atomic(page, KM_USER0);
2034 memset(userpage + zero_offset, 0, iosize);
2035 flush_dcache_page(page);
2036 kunmap_atomic(userpage, KM_USER0);
2039 while (cur <= end) {
2040 if (cur >= last_byte) {
2042 iosize = PAGE_CACHE_SIZE - page_offset;
2043 userpage = kmap_atomic(page, KM_USER0);
2044 memset(userpage + page_offset, 0, iosize);
2045 flush_dcache_page(page);
2046 kunmap_atomic(userpage, KM_USER0);
2047 set_extent_uptodate(tree, cur, cur + iosize - 1,
2049 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2052 em = get_extent(inode, page, page_offset, cur,
2054 if (IS_ERR(em) || !em) {
2056 unlock_extent(tree, cur, end, GFP_NOFS);
2059 extent_offset = cur - em->start;
2060 BUG_ON(extent_map_end(em) <= cur);
2063 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2064 this_bio_flag = EXTENT_BIO_COMPRESSED;
2066 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2067 cur_end = min(extent_map_end(em) - 1, end);
2068 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2069 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2070 disk_io_size = em->block_len;
2071 sector = em->block_start >> 9;
2073 sector = (em->block_start + extent_offset) >> 9;
2074 disk_io_size = iosize;
2077 block_start = em->block_start;
2078 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2079 block_start = EXTENT_MAP_HOLE;
2080 free_extent_map(em);
2083 /* we've found a hole, just zero and go on */
2084 if (block_start == EXTENT_MAP_HOLE) {
2086 userpage = kmap_atomic(page, KM_USER0);
2087 memset(userpage + page_offset, 0, iosize);
2088 flush_dcache_page(page);
2089 kunmap_atomic(userpage, KM_USER0);
2091 set_extent_uptodate(tree, cur, cur + iosize - 1,
2093 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2095 page_offset += iosize;
2098 /* the get_extent function already copied into the page */
2099 if (test_range_bit(tree, cur, cur_end,
2100 EXTENT_UPTODATE, 1, NULL)) {
2101 check_page_uptodate(tree, page);
2102 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2104 page_offset += iosize;
2107 /* we have an inline extent but it didn't get marked up
2108 * to date. Error out
2110 if (block_start == EXTENT_MAP_INLINE) {
2112 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2114 page_offset += iosize;
2119 if (tree->ops && tree->ops->readpage_io_hook) {
2120 ret = tree->ops->readpage_io_hook(page, cur,
2124 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2126 ret = submit_extent_page(READ, tree, page,
2127 sector, disk_io_size, page_offset,
2129 end_bio_extent_readpage, mirror_num,
2133 *bio_flags = this_bio_flag;
2138 page_offset += iosize;
2141 if (!PageError(page))
2142 SetPageUptodate(page);
2148 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2149 get_extent_t *get_extent)
2151 struct bio *bio = NULL;
2152 unsigned long bio_flags = 0;
2155 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2158 submit_one_bio(READ, bio, 0, bio_flags);
2162 static noinline void update_nr_written(struct page *page,
2163 struct writeback_control *wbc,
2164 unsigned long nr_written)
2166 wbc->nr_to_write -= nr_written;
2167 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2168 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2169 page->mapping->writeback_index = page->index + nr_written;
2173 * the writepage semantics are similar to regular writepage. extent
2174 * records are inserted to lock ranges in the tree, and as dirty areas
2175 * are found, they are marked writeback. Then the lock bits are removed
2176 * and the end_io handler clears the writeback ranges
2178 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2181 struct inode *inode = page->mapping->host;
2182 struct extent_page_data *epd = data;
2183 struct extent_io_tree *tree = epd->tree;
2184 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2186 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2190 u64 last_byte = i_size_read(inode);
2195 struct extent_state *cached_state = NULL;
2196 struct extent_map *em;
2197 struct block_device *bdev;
2200 size_t pg_offset = 0;
2202 loff_t i_size = i_size_read(inode);
2203 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2209 unsigned long nr_written = 0;
2211 if (wbc->sync_mode == WB_SYNC_ALL)
2212 write_flags = WRITE_SYNC_PLUG;
2214 write_flags = WRITE;
2216 WARN_ON(!PageLocked(page));
2217 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2218 if (page->index > end_index ||
2219 (page->index == end_index && !pg_offset)) {
2220 page->mapping->a_ops->invalidatepage(page, 0);
2225 if (page->index == end_index) {
2228 userpage = kmap_atomic(page, KM_USER0);
2229 memset(userpage + pg_offset, 0,
2230 PAGE_CACHE_SIZE - pg_offset);
2231 kunmap_atomic(userpage, KM_USER0);
2232 flush_dcache_page(page);
2236 set_page_extent_mapped(page);
2238 delalloc_start = start;
2241 if (!epd->extent_locked) {
2242 u64 delalloc_to_write = 0;
2244 * make sure the wbc mapping index is at least updated
2247 update_nr_written(page, wbc, 0);
2249 while (delalloc_end < page_end) {
2250 nr_delalloc = find_lock_delalloc_range(inode, tree,
2255 if (nr_delalloc == 0) {
2256 delalloc_start = delalloc_end + 1;
2259 tree->ops->fill_delalloc(inode, page, delalloc_start,
2260 delalloc_end, &page_started,
2263 * delalloc_end is already one less than the total
2264 * length, so we don't subtract one from
2267 delalloc_to_write += (delalloc_end - delalloc_start +
2270 delalloc_start = delalloc_end + 1;
2272 if (wbc->nr_to_write < delalloc_to_write) {
2275 if (delalloc_to_write < thresh * 2)
2276 thresh = delalloc_to_write;
2277 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2281 /* did the fill delalloc function already unlock and start
2287 * we've unlocked the page, so we can't update
2288 * the mapping's writeback index, just update
2291 wbc->nr_to_write -= nr_written;
2295 if (tree->ops && tree->ops->writepage_start_hook) {
2296 ret = tree->ops->writepage_start_hook(page, start,
2298 if (ret == -EAGAIN) {
2299 redirty_page_for_writepage(wbc, page);
2300 update_nr_written(page, wbc, nr_written);
2308 * we don't want to touch the inode after unlocking the page,
2309 * so we update the mapping writeback index now
2311 update_nr_written(page, wbc, nr_written + 1);
2314 if (last_byte <= start) {
2315 if (tree->ops && tree->ops->writepage_end_io_hook)
2316 tree->ops->writepage_end_io_hook(page, start,
2318 unlock_start = page_end + 1;
2322 blocksize = inode->i_sb->s_blocksize;
2324 while (cur <= end) {
2325 if (cur >= last_byte) {
2326 if (tree->ops && tree->ops->writepage_end_io_hook)
2327 tree->ops->writepage_end_io_hook(page, cur,
2329 unlock_start = page_end + 1;
2332 em = epd->get_extent(inode, page, pg_offset, cur,
2334 if (IS_ERR(em) || !em) {
2339 extent_offset = cur - em->start;
2340 BUG_ON(extent_map_end(em) <= cur);
2342 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2343 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2344 sector = (em->block_start + extent_offset) >> 9;
2346 block_start = em->block_start;
2347 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2348 free_extent_map(em);
2352 * compressed and inline extents are written through other
2355 if (compressed || block_start == EXTENT_MAP_HOLE ||
2356 block_start == EXTENT_MAP_INLINE) {
2358 * end_io notification does not happen here for
2359 * compressed extents
2361 if (!compressed && tree->ops &&
2362 tree->ops->writepage_end_io_hook)
2363 tree->ops->writepage_end_io_hook(page, cur,
2366 else if (compressed) {
2367 /* we don't want to end_page_writeback on
2368 * a compressed extent. this happens
2375 pg_offset += iosize;
2379 /* leave this out until we have a page_mkwrite call */
2380 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2381 EXTENT_DIRTY, 0, NULL)) {
2383 pg_offset += iosize;
2387 if (tree->ops && tree->ops->writepage_io_hook) {
2388 ret = tree->ops->writepage_io_hook(page, cur,
2396 unsigned long max_nr = end_index + 1;
2398 set_range_writeback(tree, cur, cur + iosize - 1);
2399 if (!PageWriteback(page)) {
2400 printk(KERN_ERR "btrfs warning page %lu not "
2401 "writeback, cur %llu end %llu\n",
2402 page->index, (unsigned long long)cur,
2403 (unsigned long long)end);
2406 ret = submit_extent_page(write_flags, tree, page,
2407 sector, iosize, pg_offset,
2408 bdev, &epd->bio, max_nr,
2409 end_bio_extent_writepage,
2415 pg_offset += iosize;
2420 /* make sure the mapping tag for page dirty gets cleared */
2421 set_page_writeback(page);
2422 end_page_writeback(page);
2428 /* drop our reference on any cached states */
2429 free_extent_state(cached_state);
2434 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2435 * @mapping: address space structure to write
2436 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2437 * @writepage: function called for each page
2438 * @data: data passed to writepage function
2440 * If a page is already under I/O, write_cache_pages() skips it, even
2441 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2442 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2443 * and msync() need to guarantee that all the data which was dirty at the time
2444 * the call was made get new I/O started against them. If wbc->sync_mode is
2445 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2446 * existing IO to complete.
2448 static int extent_write_cache_pages(struct extent_io_tree *tree,
2449 struct address_space *mapping,
2450 struct writeback_control *wbc,
2451 writepage_t writepage, void *data,
2452 void (*flush_fn)(void *))
2456 int nr_to_write_done = 0;
2457 struct pagevec pvec;
2460 pgoff_t end; /* Inclusive */
2462 int range_whole = 0;
2464 pagevec_init(&pvec, 0);
2465 if (wbc->range_cyclic) {
2466 index = mapping->writeback_index; /* Start from prev offset */
2469 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2470 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2471 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2476 while (!done && !nr_to_write_done && (index <= end) &&
2477 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2478 PAGECACHE_TAG_DIRTY, min(end - index,
2479 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2483 for (i = 0; i < nr_pages; i++) {
2484 struct page *page = pvec.pages[i];
2487 * At this point we hold neither mapping->tree_lock nor
2488 * lock on the page itself: the page may be truncated or
2489 * invalidated (changing page->mapping to NULL), or even
2490 * swizzled back from swapper_space to tmpfs file
2493 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2494 tree->ops->write_cache_pages_lock_hook(page);
2498 if (unlikely(page->mapping != mapping)) {
2503 if (!wbc->range_cyclic && page->index > end) {
2509 if (wbc->sync_mode != WB_SYNC_NONE) {
2510 if (PageWriteback(page))
2512 wait_on_page_writeback(page);
2515 if (PageWriteback(page) ||
2516 !clear_page_dirty_for_io(page)) {
2521 ret = (*writepage)(page, wbc, data);
2523 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2531 * the filesystem may choose to bump up nr_to_write.
2532 * We have to make sure to honor the new nr_to_write
2535 nr_to_write_done = wbc->nr_to_write <= 0;
2537 pagevec_release(&pvec);
2540 if (!scanned && !done) {
2542 * We hit the last page and there is more work to be done: wrap
2543 * back to the start of the file
2552 static void flush_epd_write_bio(struct extent_page_data *epd)
2556 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2558 submit_one_bio(WRITE, epd->bio, 0, 0);
2563 static noinline void flush_write_bio(void *data)
2565 struct extent_page_data *epd = data;
2566 flush_epd_write_bio(epd);
2569 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2570 get_extent_t *get_extent,
2571 struct writeback_control *wbc)
2574 struct address_space *mapping = page->mapping;
2575 struct extent_page_data epd = {
2578 .get_extent = get_extent,
2580 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2582 struct writeback_control wbc_writepages = {
2584 .sync_mode = wbc->sync_mode,
2585 .older_than_this = NULL,
2587 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2588 .range_end = (loff_t)-1,
2591 ret = __extent_writepage(page, wbc, &epd);
2593 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2594 __extent_writepage, &epd, flush_write_bio);
2595 flush_epd_write_bio(&epd);
2599 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2600 u64 start, u64 end, get_extent_t *get_extent,
2604 struct address_space *mapping = inode->i_mapping;
2606 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2609 struct extent_page_data epd = {
2612 .get_extent = get_extent,
2614 .sync_io = mode == WB_SYNC_ALL,
2616 struct writeback_control wbc_writepages = {
2617 .bdi = inode->i_mapping->backing_dev_info,
2619 .older_than_this = NULL,
2620 .nr_to_write = nr_pages * 2,
2621 .range_start = start,
2622 .range_end = end + 1,
2625 while (start <= end) {
2626 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2627 if (clear_page_dirty_for_io(page))
2628 ret = __extent_writepage(page, &wbc_writepages, &epd);
2630 if (tree->ops && tree->ops->writepage_end_io_hook)
2631 tree->ops->writepage_end_io_hook(page, start,
2632 start + PAGE_CACHE_SIZE - 1,
2636 page_cache_release(page);
2637 start += PAGE_CACHE_SIZE;
2640 flush_epd_write_bio(&epd);
2644 int extent_writepages(struct extent_io_tree *tree,
2645 struct address_space *mapping,
2646 get_extent_t *get_extent,
2647 struct writeback_control *wbc)
2650 struct extent_page_data epd = {
2653 .get_extent = get_extent,
2655 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2658 ret = extent_write_cache_pages(tree, mapping, wbc,
2659 __extent_writepage, &epd,
2661 flush_epd_write_bio(&epd);
2665 int extent_readpages(struct extent_io_tree *tree,
2666 struct address_space *mapping,
2667 struct list_head *pages, unsigned nr_pages,
2668 get_extent_t get_extent)
2670 struct bio *bio = NULL;
2672 struct pagevec pvec;
2673 unsigned long bio_flags = 0;
2675 pagevec_init(&pvec, 0);
2676 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2677 struct page *page = list_entry(pages->prev, struct page, lru);
2679 prefetchw(&page->flags);
2680 list_del(&page->lru);
2682 * what we want to do here is call add_to_page_cache_lru,
2683 * but that isn't exported, so we reproduce it here
2685 if (!add_to_page_cache(page, mapping,
2686 page->index, GFP_KERNEL)) {
2688 /* open coding of lru_cache_add, also not exported */
2689 page_cache_get(page);
2690 if (!pagevec_add(&pvec, page))
2691 __pagevec_lru_add_file(&pvec);
2692 __extent_read_full_page(tree, page, get_extent,
2693 &bio, 0, &bio_flags);
2695 page_cache_release(page);
2697 if (pagevec_count(&pvec))
2698 __pagevec_lru_add_file(&pvec);
2699 BUG_ON(!list_empty(pages));
2701 submit_one_bio(READ, bio, 0, bio_flags);
2706 * basic invalidatepage code, this waits on any locked or writeback
2707 * ranges corresponding to the page, and then deletes any extent state
2708 * records from the tree
2710 int extent_invalidatepage(struct extent_io_tree *tree,
2711 struct page *page, unsigned long offset)
2713 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2714 u64 end = start + PAGE_CACHE_SIZE - 1;
2715 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2717 start += (offset + blocksize - 1) & ~(blocksize - 1);
2721 lock_extent(tree, start, end, GFP_NOFS);
2722 wait_on_page_writeback(page);
2723 clear_extent_bit(tree, start, end,
2724 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2725 EXTENT_DO_ACCOUNTING,
2726 1, 1, NULL, GFP_NOFS);
2731 * simple commit_write call, set_range_dirty is used to mark both
2732 * the pages and the extent records as dirty
2734 int extent_commit_write(struct extent_io_tree *tree,
2735 struct inode *inode, struct page *page,
2736 unsigned from, unsigned to)
2738 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2740 set_page_extent_mapped(page);
2741 set_page_dirty(page);
2743 if (pos > inode->i_size) {
2744 i_size_write(inode, pos);
2745 mark_inode_dirty(inode);
2750 int extent_prepare_write(struct extent_io_tree *tree,
2751 struct inode *inode, struct page *page,
2752 unsigned from, unsigned to, get_extent_t *get_extent)
2754 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2755 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2757 u64 orig_block_start;
2760 struct extent_map *em;
2761 unsigned blocksize = 1 << inode->i_blkbits;
2762 size_t page_offset = 0;
2763 size_t block_off_start;
2764 size_t block_off_end;
2770 set_page_extent_mapped(page);
2772 block_start = (page_start + from) & ~((u64)blocksize - 1);
2773 block_end = (page_start + to - 1) | (blocksize - 1);
2774 orig_block_start = block_start;
2776 lock_extent(tree, page_start, page_end, GFP_NOFS);
2777 while (block_start <= block_end) {
2778 em = get_extent(inode, page, page_offset, block_start,
2779 block_end - block_start + 1, 1);
2780 if (IS_ERR(em) || !em)
2783 cur_end = min(block_end, extent_map_end(em) - 1);
2784 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2785 block_off_end = block_off_start + blocksize;
2786 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2788 if (!PageUptodate(page) && isnew &&
2789 (block_off_end > to || block_off_start < from)) {
2792 kaddr = kmap_atomic(page, KM_USER0);
2793 if (block_off_end > to)
2794 memset(kaddr + to, 0, block_off_end - to);
2795 if (block_off_start < from)
2796 memset(kaddr + block_off_start, 0,
2797 from - block_off_start);
2798 flush_dcache_page(page);
2799 kunmap_atomic(kaddr, KM_USER0);
2801 if ((em->block_start != EXTENT_MAP_HOLE &&
2802 em->block_start != EXTENT_MAP_INLINE) &&
2803 !isnew && !PageUptodate(page) &&
2804 (block_off_end > to || block_off_start < from) &&
2805 !test_range_bit(tree, block_start, cur_end,
2806 EXTENT_UPTODATE, 1, NULL)) {
2808 u64 extent_offset = block_start - em->start;
2810 sector = (em->block_start + extent_offset) >> 9;
2811 iosize = (cur_end - block_start + blocksize) &
2812 ~((u64)blocksize - 1);
2814 * we've already got the extent locked, but we
2815 * need to split the state such that our end_bio
2816 * handler can clear the lock.
2818 set_extent_bit(tree, block_start,
2819 block_start + iosize - 1,
2820 EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS);
2821 ret = submit_extent_page(READ, tree, page,
2822 sector, iosize, page_offset, em->bdev,
2824 end_bio_extent_preparewrite, 0,
2827 block_start = block_start + iosize;
2829 set_extent_uptodate(tree, block_start, cur_end,
2831 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2832 block_start = cur_end + 1;
2834 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2835 free_extent_map(em);
2838 wait_extent_bit(tree, orig_block_start,
2839 block_end, EXTENT_LOCKED);
2841 check_page_uptodate(tree, page);
2843 /* FIXME, zero out newly allocated blocks on error */
2848 * a helper for releasepage, this tests for areas of the page that
2849 * are locked or under IO and drops the related state bits if it is safe
2852 int try_release_extent_state(struct extent_map_tree *map,
2853 struct extent_io_tree *tree, struct page *page,
2856 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2857 u64 end = start + PAGE_CACHE_SIZE - 1;
2860 if (test_range_bit(tree, start, end,
2861 EXTENT_IOBITS, 0, NULL))
2864 if ((mask & GFP_NOFS) == GFP_NOFS)
2867 * at this point we can safely clear everything except the
2868 * locked bit and the nodatasum bit
2870 clear_extent_bit(tree, start, end,
2871 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
2878 * a helper for releasepage. As long as there are no locked extents
2879 * in the range corresponding to the page, both state records and extent
2880 * map records are removed
2882 int try_release_extent_mapping(struct extent_map_tree *map,
2883 struct extent_io_tree *tree, struct page *page,
2886 struct extent_map *em;
2887 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2888 u64 end = start + PAGE_CACHE_SIZE - 1;
2890 if ((mask & __GFP_WAIT) &&
2891 page->mapping->host->i_size > 16 * 1024 * 1024) {
2893 while (start <= end) {
2894 len = end - start + 1;
2895 write_lock(&map->lock);
2896 em = lookup_extent_mapping(map, start, len);
2897 if (!em || IS_ERR(em)) {
2898 write_unlock(&map->lock);
2901 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2902 em->start != start) {
2903 write_unlock(&map->lock);
2904 free_extent_map(em);
2907 if (!test_range_bit(tree, em->start,
2908 extent_map_end(em) - 1,
2909 EXTENT_LOCKED | EXTENT_WRITEBACK,
2911 remove_extent_mapping(map, em);
2912 /* once for the rb tree */
2913 free_extent_map(em);
2915 start = extent_map_end(em);
2916 write_unlock(&map->lock);
2919 free_extent_map(em);
2922 return try_release_extent_state(map, tree, page, mask);
2925 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2926 get_extent_t *get_extent)
2928 struct inode *inode = mapping->host;
2929 u64 start = iblock << inode->i_blkbits;
2930 sector_t sector = 0;
2931 size_t blksize = (1 << inode->i_blkbits);
2932 struct extent_map *em;
2934 lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2936 em = get_extent(inode, NULL, 0, start, blksize, 0);
2937 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2939 if (!em || IS_ERR(em))
2942 if (em->block_start > EXTENT_MAP_LAST_BYTE)
2945 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2947 free_extent_map(em);
2951 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2952 __u64 start, __u64 len, get_extent_t *get_extent)
2956 u64 max = start + len;
2959 struct extent_map *em = NULL;
2961 u64 em_start = 0, em_len = 0;
2962 unsigned long emflags;
2968 lock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2970 em = get_extent(inode, NULL, 0, off, max - off, 0);
2978 off = em->start + em->len;
2982 em_start = em->start;
2988 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
2990 flags |= FIEMAP_EXTENT_LAST;
2991 } else if (em->block_start == EXTENT_MAP_HOLE) {
2992 flags |= FIEMAP_EXTENT_UNWRITTEN;
2993 } else if (em->block_start == EXTENT_MAP_INLINE) {
2994 flags |= (FIEMAP_EXTENT_DATA_INLINE |
2995 FIEMAP_EXTENT_NOT_ALIGNED);
2996 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
2997 flags |= (FIEMAP_EXTENT_DELALLOC |
2998 FIEMAP_EXTENT_UNKNOWN);
3000 disko = em->block_start;
3002 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3003 flags |= FIEMAP_EXTENT_ENCODED;
3005 emflags = em->flags;
3006 free_extent_map(em);
3010 em = get_extent(inode, NULL, 0, off, max - off, 0);
3017 emflags = em->flags;
3019 if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) {
3020 flags |= FIEMAP_EXTENT_LAST;
3024 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3030 free_extent_map(em);
3032 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
3037 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
3041 struct address_space *mapping;
3044 return eb->first_page;
3045 i += eb->start >> PAGE_CACHE_SHIFT;
3046 mapping = eb->first_page->mapping;
3051 * extent_buffer_page is only called after pinning the page
3052 * by increasing the reference count. So we know the page must
3053 * be in the radix tree.
3056 p = radix_tree_lookup(&mapping->page_tree, i);
3062 static inline unsigned long num_extent_pages(u64 start, u64 len)
3064 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3065 (start >> PAGE_CACHE_SHIFT);
3068 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3073 struct extent_buffer *eb = NULL;
3075 unsigned long flags;
3078 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3081 spin_lock_init(&eb->lock);
3082 init_waitqueue_head(&eb->lock_wq);
3085 spin_lock_irqsave(&leak_lock, flags);
3086 list_add(&eb->leak_list, &buffers);
3087 spin_unlock_irqrestore(&leak_lock, flags);
3089 atomic_set(&eb->refs, 1);
3094 static void __free_extent_buffer(struct extent_buffer *eb)
3097 unsigned long flags;
3098 spin_lock_irqsave(&leak_lock, flags);
3099 list_del(&eb->leak_list);
3100 spin_unlock_irqrestore(&leak_lock, flags);
3102 kmem_cache_free(extent_buffer_cache, eb);
3105 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3106 u64 start, unsigned long len,
3110 unsigned long num_pages = num_extent_pages(start, len);
3112 unsigned long index = start >> PAGE_CACHE_SHIFT;
3113 struct extent_buffer *eb;
3114 struct extent_buffer *exists = NULL;
3116 struct address_space *mapping = tree->mapping;
3119 spin_lock(&tree->buffer_lock);
3120 eb = buffer_search(tree, start);
3122 atomic_inc(&eb->refs);
3123 spin_unlock(&tree->buffer_lock);
3124 mark_page_accessed(eb->first_page);
3127 spin_unlock(&tree->buffer_lock);
3129 eb = __alloc_extent_buffer(tree, start, len, mask);
3134 eb->first_page = page0;
3137 page_cache_get(page0);
3138 mark_page_accessed(page0);
3139 set_page_extent_mapped(page0);
3140 set_page_extent_head(page0, len);
3141 uptodate = PageUptodate(page0);
3145 for (; i < num_pages; i++, index++) {
3146 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
3151 set_page_extent_mapped(p);
3152 mark_page_accessed(p);
3155 set_page_extent_head(p, len);
3157 set_page_private(p, EXTENT_PAGE_PRIVATE);
3159 if (!PageUptodate(p))
3164 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3166 spin_lock(&tree->buffer_lock);
3167 exists = buffer_tree_insert(tree, start, &eb->rb_node);
3169 /* add one reference for the caller */
3170 atomic_inc(&exists->refs);
3171 spin_unlock(&tree->buffer_lock);
3174 /* add one reference for the tree */
3175 atomic_inc(&eb->refs);
3176 spin_unlock(&tree->buffer_lock);
3180 if (!atomic_dec_and_test(&eb->refs))
3182 for (index = 1; index < i; index++)
3183 page_cache_release(extent_buffer_page(eb, index));
3184 page_cache_release(extent_buffer_page(eb, 0));
3185 __free_extent_buffer(eb);
3189 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3190 u64 start, unsigned long len,
3193 struct extent_buffer *eb;
3195 spin_lock(&tree->buffer_lock);
3196 eb = buffer_search(tree, start);
3198 atomic_inc(&eb->refs);
3199 spin_unlock(&tree->buffer_lock);
3202 mark_page_accessed(eb->first_page);
3207 void free_extent_buffer(struct extent_buffer *eb)
3212 if (!atomic_dec_and_test(&eb->refs))
3218 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3219 struct extent_buffer *eb)
3222 unsigned long num_pages;
3225 num_pages = num_extent_pages(eb->start, eb->len);
3227 for (i = 0; i < num_pages; i++) {
3228 page = extent_buffer_page(eb, i);
3229 if (!PageDirty(page))
3234 set_page_extent_head(page, eb->len);
3236 set_page_private(page, EXTENT_PAGE_PRIVATE);
3238 clear_page_dirty_for_io(page);
3239 spin_lock_irq(&page->mapping->tree_lock);
3240 if (!PageDirty(page)) {
3241 radix_tree_tag_clear(&page->mapping->page_tree,
3243 PAGECACHE_TAG_DIRTY);
3245 spin_unlock_irq(&page->mapping->tree_lock);
3251 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3252 struct extent_buffer *eb)
3254 return wait_on_extent_writeback(tree, eb->start,
3255 eb->start + eb->len - 1);
3258 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3259 struct extent_buffer *eb)
3262 unsigned long num_pages;
3265 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3266 num_pages = num_extent_pages(eb->start, eb->len);
3267 for (i = 0; i < num_pages; i++)
3268 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3272 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3273 struct extent_buffer *eb)
3277 unsigned long num_pages;
3279 num_pages = num_extent_pages(eb->start, eb->len);
3280 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3282 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3284 for (i = 0; i < num_pages; i++) {
3285 page = extent_buffer_page(eb, i);
3287 ClearPageUptodate(page);
3292 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3293 struct extent_buffer *eb)
3297 unsigned long num_pages;
3299 num_pages = num_extent_pages(eb->start, eb->len);
3301 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3303 for (i = 0; i < num_pages; i++) {
3304 page = extent_buffer_page(eb, i);
3305 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3306 ((i == num_pages - 1) &&
3307 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3308 check_page_uptodate(tree, page);
3311 SetPageUptodate(page);
3316 int extent_range_uptodate(struct extent_io_tree *tree,
3321 int pg_uptodate = 1;
3323 unsigned long index;
3325 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL);
3328 while (start <= end) {
3329 index = start >> PAGE_CACHE_SHIFT;
3330 page = find_get_page(tree->mapping, index);
3331 uptodate = PageUptodate(page);
3332 page_cache_release(page);
3337 start += PAGE_CACHE_SIZE;
3342 int extent_buffer_uptodate(struct extent_io_tree *tree,
3343 struct extent_buffer *eb)
3346 unsigned long num_pages;
3349 int pg_uptodate = 1;
3351 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3354 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3355 EXTENT_UPTODATE, 1, NULL);
3359 num_pages = num_extent_pages(eb->start, eb->len);
3360 for (i = 0; i < num_pages; i++) {
3361 page = extent_buffer_page(eb, i);
3362 if (!PageUptodate(page)) {
3370 int read_extent_buffer_pages(struct extent_io_tree *tree,
3371 struct extent_buffer *eb,
3372 u64 start, int wait,
3373 get_extent_t *get_extent, int mirror_num)
3376 unsigned long start_i;
3380 int locked_pages = 0;
3381 int all_uptodate = 1;
3382 int inc_all_pages = 0;
3383 unsigned long num_pages;
3384 struct bio *bio = NULL;
3385 unsigned long bio_flags = 0;
3387 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3390 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3391 EXTENT_UPTODATE, 1, NULL)) {
3396 WARN_ON(start < eb->start);
3397 start_i = (start >> PAGE_CACHE_SHIFT) -
3398 (eb->start >> PAGE_CACHE_SHIFT);
3403 num_pages = num_extent_pages(eb->start, eb->len);
3404 for (i = start_i; i < num_pages; i++) {
3405 page = extent_buffer_page(eb, i);
3407 if (!trylock_page(page))
3413 if (!PageUptodate(page))
3418 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3422 for (i = start_i; i < num_pages; i++) {
3423 page = extent_buffer_page(eb, i);
3425 page_cache_get(page);
3426 if (!PageUptodate(page)) {
3429 ClearPageError(page);
3430 err = __extent_read_full_page(tree, page,
3432 mirror_num, &bio_flags);
3441 submit_one_bio(READ, bio, mirror_num, bio_flags);
3446 for (i = start_i; i < num_pages; i++) {
3447 page = extent_buffer_page(eb, i);
3448 wait_on_page_locked(page);
3449 if (!PageUptodate(page))
3454 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3459 while (locked_pages > 0) {
3460 page = extent_buffer_page(eb, i);
3468 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3469 unsigned long start,
3476 char *dst = (char *)dstv;
3477 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3478 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3480 WARN_ON(start > eb->len);
3481 WARN_ON(start + len > eb->start + eb->len);
3483 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3486 page = extent_buffer_page(eb, i);
3488 cur = min(len, (PAGE_CACHE_SIZE - offset));
3489 kaddr = kmap_atomic(page, KM_USER1);
3490 memcpy(dst, kaddr + offset, cur);
3491 kunmap_atomic(kaddr, KM_USER1);
3500 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3501 unsigned long min_len, char **token, char **map,
3502 unsigned long *map_start,
3503 unsigned long *map_len, int km)
3505 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3508 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3509 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3510 unsigned long end_i = (start_offset + start + min_len - 1) >>
3517 offset = start_offset;
3521 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3524 if (start + min_len > eb->len) {
3525 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3526 "wanted %lu %lu\n", (unsigned long long)eb->start,
3527 eb->len, start, min_len);
3531 p = extent_buffer_page(eb, i);
3532 kaddr = kmap_atomic(p, km);
3534 *map = kaddr + offset;
3535 *map_len = PAGE_CACHE_SIZE - offset;
3539 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3540 unsigned long min_len,
3541 char **token, char **map,
3542 unsigned long *map_start,
3543 unsigned long *map_len, int km)
3547 if (eb->map_token) {
3548 unmap_extent_buffer(eb, eb->map_token, km);
3549 eb->map_token = NULL;
3552 err = map_private_extent_buffer(eb, start, min_len, token, map,
3553 map_start, map_len, km);
3555 eb->map_token = *token;
3557 eb->map_start = *map_start;
3558 eb->map_len = *map_len;
3563 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3565 kunmap_atomic(token, km);
3568 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3569 unsigned long start,
3576 char *ptr = (char *)ptrv;
3577 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3578 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3581 WARN_ON(start > eb->len);
3582 WARN_ON(start + len > eb->start + eb->len);
3584 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3587 page = extent_buffer_page(eb, i);
3589 cur = min(len, (PAGE_CACHE_SIZE - offset));
3591 kaddr = kmap_atomic(page, KM_USER0);
3592 ret = memcmp(ptr, kaddr + offset, cur);
3593 kunmap_atomic(kaddr, KM_USER0);
3605 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3606 unsigned long start, unsigned long len)
3612 char *src = (char *)srcv;
3613 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3614 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3616 WARN_ON(start > eb->len);
3617 WARN_ON(start + len > eb->start + eb->len);
3619 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3622 page = extent_buffer_page(eb, i);
3623 WARN_ON(!PageUptodate(page));
3625 cur = min(len, PAGE_CACHE_SIZE - offset);
3626 kaddr = kmap_atomic(page, KM_USER1);
3627 memcpy(kaddr + offset, src, cur);
3628 kunmap_atomic(kaddr, KM_USER1);
3637 void memset_extent_buffer(struct extent_buffer *eb, char c,
3638 unsigned long start, unsigned long len)
3644 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3645 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3647 WARN_ON(start > eb->len);
3648 WARN_ON(start + len > eb->start + eb->len);
3650 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3653 page = extent_buffer_page(eb, i);
3654 WARN_ON(!PageUptodate(page));
3656 cur = min(len, PAGE_CACHE_SIZE - offset);
3657 kaddr = kmap_atomic(page, KM_USER0);
3658 memset(kaddr + offset, c, cur);
3659 kunmap_atomic(kaddr, KM_USER0);
3667 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3668 unsigned long dst_offset, unsigned long src_offset,
3671 u64 dst_len = dst->len;
3676 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3677 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3679 WARN_ON(src->len != dst_len);
3681 offset = (start_offset + dst_offset) &
3682 ((unsigned long)PAGE_CACHE_SIZE - 1);
3685 page = extent_buffer_page(dst, i);
3686 WARN_ON(!PageUptodate(page));
3688 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3690 kaddr = kmap_atomic(page, KM_USER0);
3691 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3692 kunmap_atomic(kaddr, KM_USER0);
3701 static void move_pages(struct page *dst_page, struct page *src_page,
3702 unsigned long dst_off, unsigned long src_off,
3705 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3706 if (dst_page == src_page) {
3707 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3709 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3710 char *p = dst_kaddr + dst_off + len;
3711 char *s = src_kaddr + src_off + len;
3716 kunmap_atomic(src_kaddr, KM_USER1);
3718 kunmap_atomic(dst_kaddr, KM_USER0);
3721 static void copy_pages(struct page *dst_page, struct page *src_page,
3722 unsigned long dst_off, unsigned long src_off,
3725 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3728 if (dst_page != src_page)
3729 src_kaddr = kmap_atomic(src_page, KM_USER1);
3731 src_kaddr = dst_kaddr;
3733 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3734 kunmap_atomic(dst_kaddr, KM_USER0);
3735 if (dst_page != src_page)
3736 kunmap_atomic(src_kaddr, KM_USER1);
3739 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3740 unsigned long src_offset, unsigned long len)
3743 size_t dst_off_in_page;
3744 size_t src_off_in_page;
3745 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3746 unsigned long dst_i;
3747 unsigned long src_i;
3749 if (src_offset + len > dst->len) {
3750 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3751 "len %lu dst len %lu\n", src_offset, len, dst->len);
3754 if (dst_offset + len > dst->len) {
3755 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3756 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3761 dst_off_in_page = (start_offset + dst_offset) &
3762 ((unsigned long)PAGE_CACHE_SIZE - 1);
3763 src_off_in_page = (start_offset + src_offset) &
3764 ((unsigned long)PAGE_CACHE_SIZE - 1);
3766 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3767 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3769 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3771 cur = min_t(unsigned long, cur,
3772 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3774 copy_pages(extent_buffer_page(dst, dst_i),
3775 extent_buffer_page(dst, src_i),
3776 dst_off_in_page, src_off_in_page, cur);
3784 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3785 unsigned long src_offset, unsigned long len)
3788 size_t dst_off_in_page;
3789 size_t src_off_in_page;
3790 unsigned long dst_end = dst_offset + len - 1;
3791 unsigned long src_end = src_offset + len - 1;
3792 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3793 unsigned long dst_i;
3794 unsigned long src_i;
3796 if (src_offset + len > dst->len) {
3797 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3798 "len %lu len %lu\n", src_offset, len, dst->len);
3801 if (dst_offset + len > dst->len) {
3802 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3803 "len %lu len %lu\n", dst_offset, len, dst->len);
3806 if (dst_offset < src_offset) {
3807 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3811 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3812 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3814 dst_off_in_page = (start_offset + dst_end) &
3815 ((unsigned long)PAGE_CACHE_SIZE - 1);
3816 src_off_in_page = (start_offset + src_end) &
3817 ((unsigned long)PAGE_CACHE_SIZE - 1);
3819 cur = min_t(unsigned long, len, src_off_in_page + 1);
3820 cur = min(cur, dst_off_in_page + 1);
3821 move_pages(extent_buffer_page(dst, dst_i),
3822 extent_buffer_page(dst, src_i),
3823 dst_off_in_page - cur + 1,
3824 src_off_in_page - cur + 1, cur);
3832 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3834 u64 start = page_offset(page);
3835 struct extent_buffer *eb;
3838 unsigned long num_pages;
3840 spin_lock(&tree->buffer_lock);
3841 eb = buffer_search(tree, start);
3845 if (atomic_read(&eb->refs) > 1) {
3849 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3853 /* at this point we can safely release the extent buffer */
3854 num_pages = num_extent_pages(eb->start, eb->len);
3855 for (i = 0; i < num_pages; i++)
3856 page_cache_release(extent_buffer_page(eb, i));
3857 rb_erase(&eb->rb_node, &tree->buffer);
3858 __free_extent_buffer(eb);
3860 spin_unlock(&tree->buffer_lock);