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_node = NULL;
108 tree->buffer.rb_node = NULL;
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 ret = state->state & bits;
465 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
466 u64 range = state->end - state->start + 1;
467 WARN_ON(range > tree->dirty_bytes);
468 tree->dirty_bytes -= range;
470 clear_state_cb(tree, state, bits);
471 state->state &= ~bits;
474 if (delete || state->state == 0) {
476 clear_state_cb(tree, state, state->state);
477 rb_erase(&state->rb_node, &tree->state);
479 free_extent_state(state);
484 merge_state(tree, state);
490 * clear some bits on a range in the tree. This may require splitting
491 * or inserting elements in the tree, so the gfp mask is used to
492 * indicate which allocations or sleeping are allowed.
494 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
495 * the given range from the tree regardless of state (ie for truncate).
497 * the range [start, end] is inclusive.
499 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
500 * bits were already set, or zero if none of the bits were already set.
502 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
503 int bits, int wake, int delete,
504 struct extent_state **cached_state,
507 struct extent_state *state;
508 struct extent_state *cached;
509 struct extent_state *prealloc = NULL;
510 struct rb_node *next_node;
511 struct rb_node *node;
517 if (!prealloc && (mask & __GFP_WAIT)) {
518 prealloc = alloc_extent_state(mask);
523 spin_lock(&tree->lock);
525 cached = *cached_state;
526 *cached_state = NULL;
528 if (cached && cached->tree && cached->start == start) {
529 atomic_dec(&cached->refs);
533 free_extent_state(cached);
536 * this search will find the extents that end after
539 node = tree_search(tree, start);
542 state = rb_entry(node, struct extent_state, rb_node);
544 if (state->start > end)
546 WARN_ON(state->end < start);
547 last_end = state->end;
550 * | ---- desired range ---- |
552 * | ------------- state -------------- |
554 * We need to split the extent we found, and may flip
555 * bits on second half.
557 * If the extent we found extends past our range, we
558 * just split and search again. It'll get split again
559 * the next time though.
561 * If the extent we found is inside our range, we clear
562 * the desired bit on it.
565 if (state->start < start) {
567 prealloc = alloc_extent_state(GFP_ATOMIC);
568 err = split_state(tree, state, prealloc, start);
569 BUG_ON(err == -EEXIST);
573 if (state->end <= end) {
574 set |= clear_state_bit(tree, state, bits, wake,
576 if (last_end == (u64)-1)
578 start = last_end + 1;
583 * | ---- desired range ---- |
585 * We need to split the extent, and clear the bit
588 if (state->start <= end && state->end > end) {
590 prealloc = alloc_extent_state(GFP_ATOMIC);
591 err = split_state(tree, state, prealloc, end + 1);
592 BUG_ON(err == -EEXIST);
596 set |= clear_state_bit(tree, prealloc, bits, wake, delete);
602 if (state->end < end && prealloc && !need_resched())
603 next_node = rb_next(&state->rb_node);
607 set |= clear_state_bit(tree, state, bits, wake, delete);
608 if (last_end == (u64)-1)
610 start = last_end + 1;
611 if (start <= end && next_node) {
612 state = rb_entry(next_node, struct extent_state,
614 if (state->start == start)
620 spin_unlock(&tree->lock);
622 free_extent_state(prealloc);
629 spin_unlock(&tree->lock);
630 if (mask & __GFP_WAIT)
635 static int wait_on_state(struct extent_io_tree *tree,
636 struct extent_state *state)
637 __releases(tree->lock)
638 __acquires(tree->lock)
641 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
642 spin_unlock(&tree->lock);
644 spin_lock(&tree->lock);
645 finish_wait(&state->wq, &wait);
650 * waits for one or more bits to clear on a range in the state tree.
651 * The range [start, end] is inclusive.
652 * The tree lock is taken by this function
654 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
656 struct extent_state *state;
657 struct rb_node *node;
659 spin_lock(&tree->lock);
663 * this search will find all the extents that end after
666 node = tree_search(tree, start);
670 state = rb_entry(node, struct extent_state, rb_node);
672 if (state->start > end)
675 if (state->state & bits) {
676 start = state->start;
677 atomic_inc(&state->refs);
678 wait_on_state(tree, state);
679 free_extent_state(state);
682 start = state->end + 1;
687 if (need_resched()) {
688 spin_unlock(&tree->lock);
690 spin_lock(&tree->lock);
694 spin_unlock(&tree->lock);
698 static int set_state_bits(struct extent_io_tree *tree,
699 struct extent_state *state,
704 ret = set_state_cb(tree, state, bits);
708 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
709 u64 range = state->end - state->start + 1;
710 tree->dirty_bytes += range;
712 state->state |= bits;
717 static void cache_state(struct extent_state *state,
718 struct extent_state **cached_ptr)
720 if (cached_ptr && !(*cached_ptr)) {
721 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
723 atomic_inc(&state->refs);
729 * set some bits on a range in the tree. This may require allocations or
730 * sleeping, so the gfp mask is used to indicate what is allowed.
732 * If any of the exclusive bits are set, this will fail with -EEXIST if some
733 * part of the range already has the desired bits set. The start of the
734 * existing range is returned in failed_start in this case.
736 * [start, end] is inclusive This takes the tree lock.
739 static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
740 int bits, int exclusive_bits, u64 *failed_start,
741 struct extent_state **cached_state,
744 struct extent_state *state;
745 struct extent_state *prealloc = NULL;
746 struct rb_node *node;
752 if (!prealloc && (mask & __GFP_WAIT)) {
753 prealloc = alloc_extent_state(mask);
758 spin_lock(&tree->lock);
759 if (cached_state && *cached_state) {
760 state = *cached_state;
761 if (state->start == start && state->tree) {
762 node = &state->rb_node;
767 * this search will find all the extents that end after
770 node = tree_search(tree, start);
772 err = insert_state(tree, prealloc, start, end, bits);
774 BUG_ON(err == -EEXIST);
777 state = rb_entry(node, struct extent_state, rb_node);
779 last_start = state->start;
780 last_end = state->end;
783 * | ---- desired range ---- |
786 * Just lock what we found and keep going
788 if (state->start == start && state->end <= end) {
789 struct rb_node *next_node;
790 if (state->state & exclusive_bits) {
791 *failed_start = state->start;
796 err = set_state_bits(tree, state, bits);
800 cache_state(state, cached_state);
801 merge_state(tree, state);
802 if (last_end == (u64)-1)
805 start = last_end + 1;
806 if (start < end && prealloc && !need_resched()) {
807 next_node = rb_next(node);
809 state = rb_entry(next_node, struct extent_state,
811 if (state->start == start)
819 * | ---- desired range ---- |
822 * | ------------- state -------------- |
824 * We need to split the extent we found, and may flip bits on
827 * If the extent we found extends past our
828 * range, we just split and search again. It'll get split
829 * again the next time though.
831 * If the extent we found is inside our range, we set the
834 if (state->start < start) {
835 if (state->state & exclusive_bits) {
836 *failed_start = start;
840 err = split_state(tree, state, prealloc, start);
841 BUG_ON(err == -EEXIST);
845 if (state->end <= end) {
846 err = set_state_bits(tree, state, bits);
849 cache_state(state, cached_state);
850 merge_state(tree, state);
851 if (last_end == (u64)-1)
853 start = last_end + 1;
858 * | ---- desired range ---- |
859 * | state | or | state |
861 * There's a hole, we need to insert something in it and
862 * ignore the extent we found.
864 if (state->start > start) {
866 if (end < last_start)
869 this_end = last_start - 1;
870 err = insert_state(tree, prealloc, start, this_end,
872 BUG_ON(err == -EEXIST);
877 cache_state(prealloc, cached_state);
879 start = this_end + 1;
883 * | ---- desired range ---- |
885 * We need to split the extent, and set the bit
888 if (state->start <= end && state->end > end) {
889 if (state->state & exclusive_bits) {
890 *failed_start = start;
894 err = split_state(tree, state, prealloc, end + 1);
895 BUG_ON(err == -EEXIST);
897 err = set_state_bits(tree, prealloc, bits);
902 cache_state(prealloc, cached_state);
903 merge_state(tree, prealloc);
911 spin_unlock(&tree->lock);
913 free_extent_state(prealloc);
920 spin_unlock(&tree->lock);
921 if (mask & __GFP_WAIT)
926 /* wrappers around set/clear extent bit */
927 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
930 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
934 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
935 int bits, gfp_t mask)
937 return set_extent_bit(tree, start, end, bits, 0, NULL,
941 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
942 int bits, gfp_t mask)
944 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
947 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
950 return set_extent_bit(tree, start, end,
951 EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
952 0, NULL, NULL, mask);
955 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
958 return clear_extent_bit(tree, start, end,
959 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
963 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
966 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
970 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
973 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0,
977 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
980 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
984 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
987 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
991 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
993 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
997 * either insert or lock state struct between start and end use mask to tell
998 * us if waiting is desired.
1000 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1001 int bits, struct extent_state **cached_state, gfp_t mask)
1006 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1007 EXTENT_LOCKED, &failed_start,
1008 cached_state, mask);
1009 if (err == -EEXIST && (mask & __GFP_WAIT)) {
1010 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1011 start = failed_start;
1015 WARN_ON(start > end);
1020 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1022 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1025 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1031 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1032 &failed_start, NULL, mask);
1033 if (err == -EEXIST) {
1034 if (failed_start > start)
1035 clear_extent_bit(tree, start, failed_start - 1,
1036 EXTENT_LOCKED, 1, 0, NULL, mask);
1042 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1043 struct extent_state **cached, gfp_t mask)
1045 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1049 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1052 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1057 * helper function to set pages and extents in the tree dirty
1059 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
1061 unsigned long index = start >> PAGE_CACHE_SHIFT;
1062 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1065 while (index <= end_index) {
1066 page = find_get_page(tree->mapping, index);
1068 __set_page_dirty_nobuffers(page);
1069 page_cache_release(page);
1076 * helper function to set both pages and extents in the tree writeback
1078 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1080 unsigned long index = start >> PAGE_CACHE_SHIFT;
1081 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1084 while (index <= end_index) {
1085 page = find_get_page(tree->mapping, index);
1087 set_page_writeback(page);
1088 page_cache_release(page);
1095 * find the first offset in the io tree with 'bits' set. zero is
1096 * returned if we find something, and *start_ret and *end_ret are
1097 * set to reflect the state struct that was found.
1099 * If nothing was found, 1 is returned, < 0 on error
1101 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1102 u64 *start_ret, u64 *end_ret, int bits)
1104 struct rb_node *node;
1105 struct extent_state *state;
1108 spin_lock(&tree->lock);
1110 * this search will find all the extents that end after
1113 node = tree_search(tree, start);
1118 state = rb_entry(node, struct extent_state, rb_node);
1119 if (state->end >= start && (state->state & bits)) {
1120 *start_ret = state->start;
1121 *end_ret = state->end;
1125 node = rb_next(node);
1130 spin_unlock(&tree->lock);
1134 /* find the first state struct with 'bits' set after 'start', and
1135 * return it. tree->lock must be held. NULL will returned if
1136 * nothing was found after 'start'
1138 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1139 u64 start, int bits)
1141 struct rb_node *node;
1142 struct extent_state *state;
1145 * this search will find all the extents that end after
1148 node = tree_search(tree, start);
1153 state = rb_entry(node, struct extent_state, rb_node);
1154 if (state->end >= start && (state->state & bits))
1157 node = rb_next(node);
1166 * find a contiguous range of bytes in the file marked as delalloc, not
1167 * more than 'max_bytes'. start and end are used to return the range,
1169 * 1 is returned if we find something, 0 if nothing was in the tree
1171 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1172 u64 *start, u64 *end, u64 max_bytes)
1174 struct rb_node *node;
1175 struct extent_state *state;
1176 u64 cur_start = *start;
1178 u64 total_bytes = 0;
1180 spin_lock(&tree->lock);
1183 * this search will find all the extents that end after
1186 node = tree_search(tree, cur_start);
1194 state = rb_entry(node, struct extent_state, rb_node);
1195 if (found && (state->start != cur_start ||
1196 (state->state & EXTENT_BOUNDARY))) {
1199 if (!(state->state & EXTENT_DELALLOC)) {
1205 *start = state->start;
1208 cur_start = state->end + 1;
1209 node = rb_next(node);
1212 total_bytes += state->end - state->start + 1;
1213 if (total_bytes >= max_bytes)
1217 spin_unlock(&tree->lock);
1221 static noinline int __unlock_for_delalloc(struct inode *inode,
1222 struct page *locked_page,
1226 struct page *pages[16];
1227 unsigned long index = start >> PAGE_CACHE_SHIFT;
1228 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1229 unsigned long nr_pages = end_index - index + 1;
1232 if (index == locked_page->index && end_index == index)
1235 while (nr_pages > 0) {
1236 ret = find_get_pages_contig(inode->i_mapping, index,
1237 min_t(unsigned long, nr_pages,
1238 ARRAY_SIZE(pages)), pages);
1239 for (i = 0; i < ret; i++) {
1240 if (pages[i] != locked_page)
1241 unlock_page(pages[i]);
1242 page_cache_release(pages[i]);
1251 static noinline int lock_delalloc_pages(struct inode *inode,
1252 struct page *locked_page,
1256 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1257 unsigned long start_index = index;
1258 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1259 unsigned long pages_locked = 0;
1260 struct page *pages[16];
1261 unsigned long nrpages;
1265 /* the caller is responsible for locking the start index */
1266 if (index == locked_page->index && index == end_index)
1269 /* skip the page at the start index */
1270 nrpages = end_index - index + 1;
1271 while (nrpages > 0) {
1272 ret = find_get_pages_contig(inode->i_mapping, index,
1273 min_t(unsigned long,
1274 nrpages, ARRAY_SIZE(pages)), pages);
1279 /* now we have an array of pages, lock them all */
1280 for (i = 0; i < ret; i++) {
1282 * the caller is taking responsibility for
1285 if (pages[i] != locked_page) {
1286 lock_page(pages[i]);
1287 if (!PageDirty(pages[i]) ||
1288 pages[i]->mapping != inode->i_mapping) {
1290 unlock_page(pages[i]);
1291 page_cache_release(pages[i]);
1295 page_cache_release(pages[i]);
1304 if (ret && pages_locked) {
1305 __unlock_for_delalloc(inode, locked_page,
1307 ((u64)(start_index + pages_locked - 1)) <<
1314 * find a contiguous range of bytes in the file marked as delalloc, not
1315 * more than 'max_bytes'. start and end are used to return the range,
1317 * 1 is returned if we find something, 0 if nothing was in the tree
1319 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1320 struct extent_io_tree *tree,
1321 struct page *locked_page,
1322 u64 *start, u64 *end,
1328 struct extent_state *cached_state = NULL;
1333 /* step one, find a bunch of delalloc bytes starting at start */
1334 delalloc_start = *start;
1336 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1338 if (!found || delalloc_end <= *start) {
1339 *start = delalloc_start;
1340 *end = delalloc_end;
1345 * start comes from the offset of locked_page. We have to lock
1346 * pages in order, so we can't process delalloc bytes before
1349 if (delalloc_start < *start)
1350 delalloc_start = *start;
1353 * make sure to limit the number of pages we try to lock down
1356 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1357 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1359 /* step two, lock all the pages after the page that has start */
1360 ret = lock_delalloc_pages(inode, locked_page,
1361 delalloc_start, delalloc_end);
1362 if (ret == -EAGAIN) {
1363 /* some of the pages are gone, lets avoid looping by
1364 * shortening the size of the delalloc range we're searching
1366 free_extent_state(cached_state);
1368 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1369 max_bytes = PAGE_CACHE_SIZE - offset;
1379 /* step three, lock the state bits for the whole range */
1380 lock_extent_bits(tree, delalloc_start, delalloc_end,
1381 0, &cached_state, GFP_NOFS);
1383 /* then test to make sure it is all still delalloc */
1384 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1385 EXTENT_DELALLOC, 1, cached_state);
1387 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1388 &cached_state, GFP_NOFS);
1389 __unlock_for_delalloc(inode, locked_page,
1390 delalloc_start, delalloc_end);
1394 free_extent_state(cached_state);
1395 *start = delalloc_start;
1396 *end = delalloc_end;
1401 int extent_clear_unlock_delalloc(struct inode *inode,
1402 struct extent_io_tree *tree,
1403 u64 start, u64 end, struct page *locked_page,
1407 struct page *pages[16];
1408 unsigned long index = start >> PAGE_CACHE_SHIFT;
1409 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1410 unsigned long nr_pages = end_index - index + 1;
1414 if (op & EXTENT_CLEAR_UNLOCK)
1415 clear_bits |= EXTENT_LOCKED;
1416 if (op & EXTENT_CLEAR_DIRTY)
1417 clear_bits |= EXTENT_DIRTY;
1419 if (op & EXTENT_CLEAR_DELALLOC)
1420 clear_bits |= EXTENT_DELALLOC;
1422 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1423 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK |
1424 EXTENT_END_WRITEBACK | EXTENT_SET_PRIVATE2)))
1427 while (nr_pages > 0) {
1428 ret = find_get_pages_contig(inode->i_mapping, index,
1429 min_t(unsigned long,
1430 nr_pages, ARRAY_SIZE(pages)), pages);
1431 for (i = 0; i < ret; i++) {
1433 if (op & EXTENT_SET_PRIVATE2)
1434 SetPagePrivate2(pages[i]);
1436 if (pages[i] == locked_page) {
1437 page_cache_release(pages[i]);
1440 if (op & EXTENT_CLEAR_DIRTY)
1441 clear_page_dirty_for_io(pages[i]);
1442 if (op & EXTENT_SET_WRITEBACK)
1443 set_page_writeback(pages[i]);
1444 if (op & EXTENT_END_WRITEBACK)
1445 end_page_writeback(pages[i]);
1446 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1447 unlock_page(pages[i]);
1448 page_cache_release(pages[i]);
1458 * count the number of bytes in the tree that have a given bit(s)
1459 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1460 * cached. The total number found is returned.
1462 u64 count_range_bits(struct extent_io_tree *tree,
1463 u64 *start, u64 search_end, u64 max_bytes,
1466 struct rb_node *node;
1467 struct extent_state *state;
1468 u64 cur_start = *start;
1469 u64 total_bytes = 0;
1472 if (search_end <= cur_start) {
1477 spin_lock(&tree->lock);
1478 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1479 total_bytes = tree->dirty_bytes;
1483 * this search will find all the extents that end after
1486 node = tree_search(tree, cur_start);
1491 state = rb_entry(node, struct extent_state, rb_node);
1492 if (state->start > search_end)
1494 if (state->end >= cur_start && (state->state & bits)) {
1495 total_bytes += min(search_end, state->end) + 1 -
1496 max(cur_start, state->start);
1497 if (total_bytes >= max_bytes)
1500 *start = state->start;
1504 node = rb_next(node);
1509 spin_unlock(&tree->lock);
1514 * set the private field for a given byte offset in the tree. If there isn't
1515 * an extent_state there already, this does nothing.
1517 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1519 struct rb_node *node;
1520 struct extent_state *state;
1523 spin_lock(&tree->lock);
1525 * this search will find all the extents that end after
1528 node = tree_search(tree, start);
1533 state = rb_entry(node, struct extent_state, rb_node);
1534 if (state->start != start) {
1538 state->private = private;
1540 spin_unlock(&tree->lock);
1544 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1546 struct rb_node *node;
1547 struct extent_state *state;
1550 spin_lock(&tree->lock);
1552 * this search will find all the extents that end after
1555 node = tree_search(tree, start);
1560 state = rb_entry(node, struct extent_state, rb_node);
1561 if (state->start != start) {
1565 *private = state->private;
1567 spin_unlock(&tree->lock);
1572 * searches a range in the state tree for a given mask.
1573 * If 'filled' == 1, this returns 1 only if every extent in the tree
1574 * has the bits set. Otherwise, 1 is returned if any bit in the
1575 * range is found set.
1577 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1578 int bits, int filled, struct extent_state *cached)
1580 struct extent_state *state = NULL;
1581 struct rb_node *node;
1584 spin_lock(&tree->lock);
1585 if (cached && cached->tree && cached->start == start)
1586 node = &cached->rb_node;
1588 node = tree_search(tree, start);
1589 while (node && start <= end) {
1590 state = rb_entry(node, struct extent_state, rb_node);
1592 if (filled && state->start > start) {
1597 if (state->start > end)
1600 if (state->state & bits) {
1604 } else if (filled) {
1609 if (state->end == (u64)-1)
1612 start = state->end + 1;
1615 node = rb_next(node);
1622 spin_unlock(&tree->lock);
1627 * helper function to set a given page up to date if all the
1628 * extents in the tree for that page are up to date
1630 static int check_page_uptodate(struct extent_io_tree *tree,
1633 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1634 u64 end = start + PAGE_CACHE_SIZE - 1;
1635 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1636 SetPageUptodate(page);
1641 * helper function to unlock a page if all the extents in the tree
1642 * for that page are unlocked
1644 static int check_page_locked(struct extent_io_tree *tree,
1647 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1648 u64 end = start + PAGE_CACHE_SIZE - 1;
1649 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1655 * helper function to end page writeback if all the extents
1656 * in the tree for that page are done with writeback
1658 static int check_page_writeback(struct extent_io_tree *tree,
1661 end_page_writeback(page);
1665 /* lots and lots of room for performance fixes in the end_bio funcs */
1668 * after a writepage IO is done, we need to:
1669 * clear the uptodate bits on error
1670 * clear the writeback bits in the extent tree for this IO
1671 * end_page_writeback if the page has no more pending IO
1673 * Scheduling is not allowed, so the extent state tree is expected
1674 * to have one and only one object corresponding to this IO.
1676 static void end_bio_extent_writepage(struct bio *bio, int err)
1678 int uptodate = err == 0;
1679 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1680 struct extent_io_tree *tree;
1687 struct page *page = bvec->bv_page;
1688 tree = &BTRFS_I(page->mapping->host)->io_tree;
1690 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1692 end = start + bvec->bv_len - 1;
1694 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1699 if (--bvec >= bio->bi_io_vec)
1700 prefetchw(&bvec->bv_page->flags);
1701 if (tree->ops && tree->ops->writepage_end_io_hook) {
1702 ret = tree->ops->writepage_end_io_hook(page, start,
1703 end, NULL, uptodate);
1708 if (!uptodate && tree->ops &&
1709 tree->ops->writepage_io_failed_hook) {
1710 ret = tree->ops->writepage_io_failed_hook(bio, page,
1713 uptodate = (err == 0);
1719 clear_extent_uptodate(tree, start, end, GFP_NOFS);
1720 ClearPageUptodate(page);
1725 end_page_writeback(page);
1727 check_page_writeback(tree, page);
1728 } while (bvec >= bio->bi_io_vec);
1734 * after a readpage IO is done, we need to:
1735 * clear the uptodate bits on error
1736 * set the uptodate bits if things worked
1737 * set the page up to date if all extents in the tree are uptodate
1738 * clear the lock bit in the extent tree
1739 * unlock the page if there are no other extents locked for it
1741 * Scheduling is not allowed, so the extent state tree is expected
1742 * to have one and only one object corresponding to this IO.
1744 static void end_bio_extent_readpage(struct bio *bio, int err)
1746 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1747 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1748 struct extent_io_tree *tree;
1758 struct page *page = bvec->bv_page;
1759 tree = &BTRFS_I(page->mapping->host)->io_tree;
1761 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1763 end = start + bvec->bv_len - 1;
1765 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1770 if (--bvec >= bio->bi_io_vec)
1771 prefetchw(&bvec->bv_page->flags);
1773 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1774 ret = tree->ops->readpage_end_io_hook(page, start, end,
1779 if (!uptodate && tree->ops &&
1780 tree->ops->readpage_io_failed_hook) {
1781 ret = tree->ops->readpage_io_failed_hook(bio, page,
1785 test_bit(BIO_UPTODATE, &bio->bi_flags);
1793 set_extent_uptodate(tree, start, end,
1796 unlock_extent(tree, start, end, GFP_ATOMIC);
1800 SetPageUptodate(page);
1802 ClearPageUptodate(page);
1808 check_page_uptodate(tree, page);
1810 ClearPageUptodate(page);
1813 check_page_locked(tree, page);
1815 } while (bvec >= bio->bi_io_vec);
1821 * IO done from prepare_write is pretty simple, we just unlock
1822 * the structs in the extent tree when done, and set the uptodate bits
1825 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1827 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1828 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1829 struct extent_io_tree *tree;
1834 struct page *page = bvec->bv_page;
1835 tree = &BTRFS_I(page->mapping->host)->io_tree;
1837 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1839 end = start + bvec->bv_len - 1;
1841 if (--bvec >= bio->bi_io_vec)
1842 prefetchw(&bvec->bv_page->flags);
1845 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1847 ClearPageUptodate(page);
1851 unlock_extent(tree, start, end, GFP_ATOMIC);
1853 } while (bvec >= bio->bi_io_vec);
1859 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1864 bio = bio_alloc(gfp_flags, nr_vecs);
1866 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1867 while (!bio && (nr_vecs /= 2))
1868 bio = bio_alloc(gfp_flags, nr_vecs);
1873 bio->bi_bdev = bdev;
1874 bio->bi_sector = first_sector;
1879 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1880 unsigned long bio_flags)
1883 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1884 struct page *page = bvec->bv_page;
1885 struct extent_io_tree *tree = bio->bi_private;
1889 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1890 end = start + bvec->bv_len - 1;
1892 bio->bi_private = NULL;
1896 if (tree->ops && tree->ops->submit_bio_hook)
1897 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1898 mirror_num, bio_flags);
1900 submit_bio(rw, bio);
1901 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1907 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1908 struct page *page, sector_t sector,
1909 size_t size, unsigned long offset,
1910 struct block_device *bdev,
1911 struct bio **bio_ret,
1912 unsigned long max_pages,
1913 bio_end_io_t end_io_func,
1915 unsigned long prev_bio_flags,
1916 unsigned long bio_flags)
1922 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1923 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1924 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1926 if (bio_ret && *bio_ret) {
1929 contig = bio->bi_sector == sector;
1931 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1934 if (prev_bio_flags != bio_flags || !contig ||
1935 (tree->ops && tree->ops->merge_bio_hook &&
1936 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1938 bio_add_page(bio, page, page_size, offset) < page_size) {
1939 ret = submit_one_bio(rw, bio, mirror_num,
1946 if (this_compressed)
1949 nr = bio_get_nr_vecs(bdev);
1951 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1953 bio_add_page(bio, page, page_size, offset);
1954 bio->bi_end_io = end_io_func;
1955 bio->bi_private = tree;
1960 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1965 void set_page_extent_mapped(struct page *page)
1967 if (!PagePrivate(page)) {
1968 SetPagePrivate(page);
1969 page_cache_get(page);
1970 set_page_private(page, EXTENT_PAGE_PRIVATE);
1974 static void set_page_extent_head(struct page *page, unsigned long len)
1976 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1980 * basic readpage implementation. Locked extent state structs are inserted
1981 * into the tree that are removed when the IO is done (by the end_io
1984 static int __extent_read_full_page(struct extent_io_tree *tree,
1986 get_extent_t *get_extent,
1987 struct bio **bio, int mirror_num,
1988 unsigned long *bio_flags)
1990 struct inode *inode = page->mapping->host;
1991 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1992 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1996 u64 last_byte = i_size_read(inode);
2000 struct extent_map *em;
2001 struct block_device *bdev;
2004 size_t page_offset = 0;
2006 size_t disk_io_size;
2007 size_t blocksize = inode->i_sb->s_blocksize;
2008 unsigned long this_bio_flag = 0;
2010 set_page_extent_mapped(page);
2013 lock_extent(tree, start, end, GFP_NOFS);
2015 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2017 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2020 iosize = PAGE_CACHE_SIZE - zero_offset;
2021 userpage = kmap_atomic(page, KM_USER0);
2022 memset(userpage + zero_offset, 0, iosize);
2023 flush_dcache_page(page);
2024 kunmap_atomic(userpage, KM_USER0);
2027 while (cur <= end) {
2028 if (cur >= last_byte) {
2030 iosize = PAGE_CACHE_SIZE - page_offset;
2031 userpage = kmap_atomic(page, KM_USER0);
2032 memset(userpage + page_offset, 0, iosize);
2033 flush_dcache_page(page);
2034 kunmap_atomic(userpage, KM_USER0);
2035 set_extent_uptodate(tree, cur, cur + iosize - 1,
2037 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2040 em = get_extent(inode, page, page_offset, cur,
2042 if (IS_ERR(em) || !em) {
2044 unlock_extent(tree, cur, end, GFP_NOFS);
2047 extent_offset = cur - em->start;
2048 BUG_ON(extent_map_end(em) <= cur);
2051 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2052 this_bio_flag = EXTENT_BIO_COMPRESSED;
2054 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2055 cur_end = min(extent_map_end(em) - 1, end);
2056 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2057 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2058 disk_io_size = em->block_len;
2059 sector = em->block_start >> 9;
2061 sector = (em->block_start + extent_offset) >> 9;
2062 disk_io_size = iosize;
2065 block_start = em->block_start;
2066 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2067 block_start = EXTENT_MAP_HOLE;
2068 free_extent_map(em);
2071 /* we've found a hole, just zero and go on */
2072 if (block_start == EXTENT_MAP_HOLE) {
2074 userpage = kmap_atomic(page, KM_USER0);
2075 memset(userpage + page_offset, 0, iosize);
2076 flush_dcache_page(page);
2077 kunmap_atomic(userpage, KM_USER0);
2079 set_extent_uptodate(tree, cur, cur + iosize - 1,
2081 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2083 page_offset += iosize;
2086 /* the get_extent function already copied into the page */
2087 if (test_range_bit(tree, cur, cur_end,
2088 EXTENT_UPTODATE, 1, NULL)) {
2089 check_page_uptodate(tree, page);
2090 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2092 page_offset += iosize;
2095 /* we have an inline extent but it didn't get marked up
2096 * to date. Error out
2098 if (block_start == EXTENT_MAP_INLINE) {
2100 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2102 page_offset += iosize;
2107 if (tree->ops && tree->ops->readpage_io_hook) {
2108 ret = tree->ops->readpage_io_hook(page, cur,
2112 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2114 ret = submit_extent_page(READ, tree, page,
2115 sector, disk_io_size, page_offset,
2117 end_bio_extent_readpage, mirror_num,
2121 *bio_flags = this_bio_flag;
2126 page_offset += iosize;
2129 if (!PageError(page))
2130 SetPageUptodate(page);
2136 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2137 get_extent_t *get_extent)
2139 struct bio *bio = NULL;
2140 unsigned long bio_flags = 0;
2143 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2146 submit_one_bio(READ, bio, 0, bio_flags);
2150 static noinline void update_nr_written(struct page *page,
2151 struct writeback_control *wbc,
2152 unsigned long nr_written)
2154 wbc->nr_to_write -= nr_written;
2155 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2156 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2157 page->mapping->writeback_index = page->index + nr_written;
2161 * the writepage semantics are similar to regular writepage. extent
2162 * records are inserted to lock ranges in the tree, and as dirty areas
2163 * are found, they are marked writeback. Then the lock bits are removed
2164 * and the end_io handler clears the writeback ranges
2166 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2169 struct inode *inode = page->mapping->host;
2170 struct extent_page_data *epd = data;
2171 struct extent_io_tree *tree = epd->tree;
2172 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2174 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2178 u64 last_byte = i_size_read(inode);
2183 struct extent_state *cached_state = NULL;
2184 struct extent_map *em;
2185 struct block_device *bdev;
2188 size_t pg_offset = 0;
2190 loff_t i_size = i_size_read(inode);
2191 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2197 unsigned long nr_written = 0;
2199 if (wbc->sync_mode == WB_SYNC_ALL)
2200 write_flags = WRITE_SYNC_PLUG;
2202 write_flags = WRITE;
2204 WARN_ON(!PageLocked(page));
2205 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2206 if (page->index > end_index ||
2207 (page->index == end_index && !pg_offset)) {
2208 page->mapping->a_ops->invalidatepage(page, 0);
2213 if (page->index == end_index) {
2216 userpage = kmap_atomic(page, KM_USER0);
2217 memset(userpage + pg_offset, 0,
2218 PAGE_CACHE_SIZE - pg_offset);
2219 kunmap_atomic(userpage, KM_USER0);
2220 flush_dcache_page(page);
2224 set_page_extent_mapped(page);
2226 delalloc_start = start;
2229 if (!epd->extent_locked) {
2230 u64 delalloc_to_write = 0;
2232 * make sure the wbc mapping index is at least updated
2235 update_nr_written(page, wbc, 0);
2237 while (delalloc_end < page_end) {
2238 nr_delalloc = find_lock_delalloc_range(inode, tree,
2243 if (nr_delalloc == 0) {
2244 delalloc_start = delalloc_end + 1;
2247 tree->ops->fill_delalloc(inode, page, delalloc_start,
2248 delalloc_end, &page_started,
2251 * delalloc_end is already one less than the total
2252 * length, so we don't subtract one from
2255 delalloc_to_write += (delalloc_end - delalloc_start +
2258 delalloc_start = delalloc_end + 1;
2260 if (wbc->nr_to_write < delalloc_to_write) {
2263 if (delalloc_to_write < thresh * 2)
2264 thresh = delalloc_to_write;
2265 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2269 /* did the fill delalloc function already unlock and start
2275 * we've unlocked the page, so we can't update
2276 * the mapping's writeback index, just update
2279 wbc->nr_to_write -= nr_written;
2283 if (tree->ops && tree->ops->writepage_start_hook) {
2284 ret = tree->ops->writepage_start_hook(page, start,
2286 if (ret == -EAGAIN) {
2287 redirty_page_for_writepage(wbc, page);
2288 update_nr_written(page, wbc, nr_written);
2296 * we don't want to touch the inode after unlocking the page,
2297 * so we update the mapping writeback index now
2299 update_nr_written(page, wbc, nr_written + 1);
2302 if (last_byte <= start) {
2303 if (tree->ops && tree->ops->writepage_end_io_hook)
2304 tree->ops->writepage_end_io_hook(page, start,
2306 unlock_start = page_end + 1;
2310 blocksize = inode->i_sb->s_blocksize;
2312 while (cur <= end) {
2313 if (cur >= last_byte) {
2314 if (tree->ops && tree->ops->writepage_end_io_hook)
2315 tree->ops->writepage_end_io_hook(page, cur,
2317 unlock_start = page_end + 1;
2320 em = epd->get_extent(inode, page, pg_offset, cur,
2322 if (IS_ERR(em) || !em) {
2327 extent_offset = cur - em->start;
2328 BUG_ON(extent_map_end(em) <= cur);
2330 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2331 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2332 sector = (em->block_start + extent_offset) >> 9;
2334 block_start = em->block_start;
2335 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2336 free_extent_map(em);
2340 * compressed and inline extents are written through other
2343 if (compressed || block_start == EXTENT_MAP_HOLE ||
2344 block_start == EXTENT_MAP_INLINE) {
2346 * end_io notification does not happen here for
2347 * compressed extents
2349 if (!compressed && tree->ops &&
2350 tree->ops->writepage_end_io_hook)
2351 tree->ops->writepage_end_io_hook(page, cur,
2354 else if (compressed) {
2355 /* we don't want to end_page_writeback on
2356 * a compressed extent. this happens
2363 pg_offset += iosize;
2367 /* leave this out until we have a page_mkwrite call */
2368 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2369 EXTENT_DIRTY, 0, NULL)) {
2371 pg_offset += iosize;
2375 if (tree->ops && tree->ops->writepage_io_hook) {
2376 ret = tree->ops->writepage_io_hook(page, cur,
2384 unsigned long max_nr = end_index + 1;
2386 set_range_writeback(tree, cur, cur + iosize - 1);
2387 if (!PageWriteback(page)) {
2388 printk(KERN_ERR "btrfs warning page %lu not "
2389 "writeback, cur %llu end %llu\n",
2390 page->index, (unsigned long long)cur,
2391 (unsigned long long)end);
2394 ret = submit_extent_page(write_flags, tree, page,
2395 sector, iosize, pg_offset,
2396 bdev, &epd->bio, max_nr,
2397 end_bio_extent_writepage,
2403 pg_offset += iosize;
2408 /* make sure the mapping tag for page dirty gets cleared */
2409 set_page_writeback(page);
2410 end_page_writeback(page);
2416 /* drop our reference on any cached states */
2417 free_extent_state(cached_state);
2422 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2423 * @mapping: address space structure to write
2424 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2425 * @writepage: function called for each page
2426 * @data: data passed to writepage function
2428 * If a page is already under I/O, write_cache_pages() skips it, even
2429 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2430 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2431 * and msync() need to guarantee that all the data which was dirty at the time
2432 * the call was made get new I/O started against them. If wbc->sync_mode is
2433 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2434 * existing IO to complete.
2436 static int extent_write_cache_pages(struct extent_io_tree *tree,
2437 struct address_space *mapping,
2438 struct writeback_control *wbc,
2439 writepage_t writepage, void *data,
2440 void (*flush_fn)(void *))
2444 int nr_to_write_done = 0;
2445 struct pagevec pvec;
2448 pgoff_t end; /* Inclusive */
2450 int range_whole = 0;
2452 pagevec_init(&pvec, 0);
2453 if (wbc->range_cyclic) {
2454 index = mapping->writeback_index; /* Start from prev offset */
2457 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2458 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2459 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2464 while (!done && !nr_to_write_done && (index <= end) &&
2465 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2466 PAGECACHE_TAG_DIRTY, min(end - index,
2467 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2471 for (i = 0; i < nr_pages; i++) {
2472 struct page *page = pvec.pages[i];
2475 * At this point we hold neither mapping->tree_lock nor
2476 * lock on the page itself: the page may be truncated or
2477 * invalidated (changing page->mapping to NULL), or even
2478 * swizzled back from swapper_space to tmpfs file
2481 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2482 tree->ops->write_cache_pages_lock_hook(page);
2486 if (unlikely(page->mapping != mapping)) {
2491 if (!wbc->range_cyclic && page->index > end) {
2497 if (wbc->sync_mode != WB_SYNC_NONE) {
2498 if (PageWriteback(page))
2500 wait_on_page_writeback(page);
2503 if (PageWriteback(page) ||
2504 !clear_page_dirty_for_io(page)) {
2509 ret = (*writepage)(page, wbc, data);
2511 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2519 * the filesystem may choose to bump up nr_to_write.
2520 * We have to make sure to honor the new nr_to_write
2523 nr_to_write_done = wbc->nr_to_write <= 0;
2525 pagevec_release(&pvec);
2528 if (!scanned && !done) {
2530 * We hit the last page and there is more work to be done: wrap
2531 * back to the start of the file
2540 static void flush_epd_write_bio(struct extent_page_data *epd)
2544 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2546 submit_one_bio(WRITE, epd->bio, 0, 0);
2551 static noinline void flush_write_bio(void *data)
2553 struct extent_page_data *epd = data;
2554 flush_epd_write_bio(epd);
2557 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2558 get_extent_t *get_extent,
2559 struct writeback_control *wbc)
2562 struct address_space *mapping = page->mapping;
2563 struct extent_page_data epd = {
2566 .get_extent = get_extent,
2568 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2570 struct writeback_control wbc_writepages = {
2572 .sync_mode = wbc->sync_mode,
2573 .older_than_this = NULL,
2575 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2576 .range_end = (loff_t)-1,
2579 ret = __extent_writepage(page, wbc, &epd);
2581 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2582 __extent_writepage, &epd, flush_write_bio);
2583 flush_epd_write_bio(&epd);
2587 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2588 u64 start, u64 end, get_extent_t *get_extent,
2592 struct address_space *mapping = inode->i_mapping;
2594 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2597 struct extent_page_data epd = {
2600 .get_extent = get_extent,
2602 .sync_io = mode == WB_SYNC_ALL,
2604 struct writeback_control wbc_writepages = {
2605 .bdi = inode->i_mapping->backing_dev_info,
2607 .older_than_this = NULL,
2608 .nr_to_write = nr_pages * 2,
2609 .range_start = start,
2610 .range_end = end + 1,
2613 while (start <= end) {
2614 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2615 if (clear_page_dirty_for_io(page))
2616 ret = __extent_writepage(page, &wbc_writepages, &epd);
2618 if (tree->ops && tree->ops->writepage_end_io_hook)
2619 tree->ops->writepage_end_io_hook(page, start,
2620 start + PAGE_CACHE_SIZE - 1,
2624 page_cache_release(page);
2625 start += PAGE_CACHE_SIZE;
2628 flush_epd_write_bio(&epd);
2632 int extent_writepages(struct extent_io_tree *tree,
2633 struct address_space *mapping,
2634 get_extent_t *get_extent,
2635 struct writeback_control *wbc)
2638 struct extent_page_data epd = {
2641 .get_extent = get_extent,
2643 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2646 ret = extent_write_cache_pages(tree, mapping, wbc,
2647 __extent_writepage, &epd,
2649 flush_epd_write_bio(&epd);
2653 int extent_readpages(struct extent_io_tree *tree,
2654 struct address_space *mapping,
2655 struct list_head *pages, unsigned nr_pages,
2656 get_extent_t get_extent)
2658 struct bio *bio = NULL;
2660 struct pagevec pvec;
2661 unsigned long bio_flags = 0;
2663 pagevec_init(&pvec, 0);
2664 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2665 struct page *page = list_entry(pages->prev, struct page, lru);
2667 prefetchw(&page->flags);
2668 list_del(&page->lru);
2670 * what we want to do here is call add_to_page_cache_lru,
2671 * but that isn't exported, so we reproduce it here
2673 if (!add_to_page_cache(page, mapping,
2674 page->index, GFP_KERNEL)) {
2676 /* open coding of lru_cache_add, also not exported */
2677 page_cache_get(page);
2678 if (!pagevec_add(&pvec, page))
2679 __pagevec_lru_add_file(&pvec);
2680 __extent_read_full_page(tree, page, get_extent,
2681 &bio, 0, &bio_flags);
2683 page_cache_release(page);
2685 if (pagevec_count(&pvec))
2686 __pagevec_lru_add_file(&pvec);
2687 BUG_ON(!list_empty(pages));
2689 submit_one_bio(READ, bio, 0, bio_flags);
2694 * basic invalidatepage code, this waits on any locked or writeback
2695 * ranges corresponding to the page, and then deletes any extent state
2696 * records from the tree
2698 int extent_invalidatepage(struct extent_io_tree *tree,
2699 struct page *page, unsigned long offset)
2701 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2702 u64 end = start + PAGE_CACHE_SIZE - 1;
2703 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2705 start += (offset + blocksize - 1) & ~(blocksize - 1);
2709 lock_extent(tree, start, end, GFP_NOFS);
2710 wait_on_page_writeback(page);
2711 clear_extent_bit(tree, start, end,
2712 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2713 1, 1, NULL, GFP_NOFS);
2718 * simple commit_write call, set_range_dirty is used to mark both
2719 * the pages and the extent records as dirty
2721 int extent_commit_write(struct extent_io_tree *tree,
2722 struct inode *inode, struct page *page,
2723 unsigned from, unsigned to)
2725 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2727 set_page_extent_mapped(page);
2728 set_page_dirty(page);
2730 if (pos > inode->i_size) {
2731 i_size_write(inode, pos);
2732 mark_inode_dirty(inode);
2737 int extent_prepare_write(struct extent_io_tree *tree,
2738 struct inode *inode, struct page *page,
2739 unsigned from, unsigned to, get_extent_t *get_extent)
2741 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2742 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2744 u64 orig_block_start;
2747 struct extent_map *em;
2748 unsigned blocksize = 1 << inode->i_blkbits;
2749 size_t page_offset = 0;
2750 size_t block_off_start;
2751 size_t block_off_end;
2757 set_page_extent_mapped(page);
2759 block_start = (page_start + from) & ~((u64)blocksize - 1);
2760 block_end = (page_start + to - 1) | (blocksize - 1);
2761 orig_block_start = block_start;
2763 lock_extent(tree, page_start, page_end, GFP_NOFS);
2764 while (block_start <= block_end) {
2765 em = get_extent(inode, page, page_offset, block_start,
2766 block_end - block_start + 1, 1);
2767 if (IS_ERR(em) || !em)
2770 cur_end = min(block_end, extent_map_end(em) - 1);
2771 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2772 block_off_end = block_off_start + blocksize;
2773 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2775 if (!PageUptodate(page) && isnew &&
2776 (block_off_end > to || block_off_start < from)) {
2779 kaddr = kmap_atomic(page, KM_USER0);
2780 if (block_off_end > to)
2781 memset(kaddr + to, 0, block_off_end - to);
2782 if (block_off_start < from)
2783 memset(kaddr + block_off_start, 0,
2784 from - block_off_start);
2785 flush_dcache_page(page);
2786 kunmap_atomic(kaddr, KM_USER0);
2788 if ((em->block_start != EXTENT_MAP_HOLE &&
2789 em->block_start != EXTENT_MAP_INLINE) &&
2790 !isnew && !PageUptodate(page) &&
2791 (block_off_end > to || block_off_start < from) &&
2792 !test_range_bit(tree, block_start, cur_end,
2793 EXTENT_UPTODATE, 1, NULL)) {
2795 u64 extent_offset = block_start - em->start;
2797 sector = (em->block_start + extent_offset) >> 9;
2798 iosize = (cur_end - block_start + blocksize) &
2799 ~((u64)blocksize - 1);
2801 * we've already got the extent locked, but we
2802 * need to split the state such that our end_bio
2803 * handler can clear the lock.
2805 set_extent_bit(tree, block_start,
2806 block_start + iosize - 1,
2807 EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS);
2808 ret = submit_extent_page(READ, tree, page,
2809 sector, iosize, page_offset, em->bdev,
2811 end_bio_extent_preparewrite, 0,
2814 block_start = block_start + iosize;
2816 set_extent_uptodate(tree, block_start, cur_end,
2818 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2819 block_start = cur_end + 1;
2821 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2822 free_extent_map(em);
2825 wait_extent_bit(tree, orig_block_start,
2826 block_end, EXTENT_LOCKED);
2828 check_page_uptodate(tree, page);
2830 /* FIXME, zero out newly allocated blocks on error */
2835 * a helper for releasepage, this tests for areas of the page that
2836 * are locked or under IO and drops the related state bits if it is safe
2839 int try_release_extent_state(struct extent_map_tree *map,
2840 struct extent_io_tree *tree, struct page *page,
2843 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2844 u64 end = start + PAGE_CACHE_SIZE - 1;
2847 if (test_range_bit(tree, start, end,
2848 EXTENT_IOBITS, 0, NULL))
2851 if ((mask & GFP_NOFS) == GFP_NOFS)
2854 * at this point we can safely clear everything except the
2855 * locked bit and the nodatasum bit
2857 clear_extent_bit(tree, start, end,
2858 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
2865 * a helper for releasepage. As long as there are no locked extents
2866 * in the range corresponding to the page, both state records and extent
2867 * map records are removed
2869 int try_release_extent_mapping(struct extent_map_tree *map,
2870 struct extent_io_tree *tree, struct page *page,
2873 struct extent_map *em;
2874 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2875 u64 end = start + PAGE_CACHE_SIZE - 1;
2877 if ((mask & __GFP_WAIT) &&
2878 page->mapping->host->i_size > 16 * 1024 * 1024) {
2880 while (start <= end) {
2881 len = end - start + 1;
2882 write_lock(&map->lock);
2883 em = lookup_extent_mapping(map, start, len);
2884 if (!em || IS_ERR(em)) {
2885 write_unlock(&map->lock);
2888 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2889 em->start != start) {
2890 write_unlock(&map->lock);
2891 free_extent_map(em);
2894 if (!test_range_bit(tree, em->start,
2895 extent_map_end(em) - 1,
2896 EXTENT_LOCKED | EXTENT_WRITEBACK,
2898 remove_extent_mapping(map, em);
2899 /* once for the rb tree */
2900 free_extent_map(em);
2902 start = extent_map_end(em);
2903 write_unlock(&map->lock);
2906 free_extent_map(em);
2909 return try_release_extent_state(map, tree, page, mask);
2912 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2913 get_extent_t *get_extent)
2915 struct inode *inode = mapping->host;
2916 u64 start = iblock << inode->i_blkbits;
2917 sector_t sector = 0;
2918 size_t blksize = (1 << inode->i_blkbits);
2919 struct extent_map *em;
2921 lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2923 em = get_extent(inode, NULL, 0, start, blksize, 0);
2924 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2926 if (!em || IS_ERR(em))
2929 if (em->block_start > EXTENT_MAP_LAST_BYTE)
2932 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2934 free_extent_map(em);
2938 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2939 __u64 start, __u64 len, get_extent_t *get_extent)
2943 u64 max = start + len;
2946 struct extent_map *em = NULL;
2948 u64 em_start = 0, em_len = 0;
2949 unsigned long emflags;
2955 lock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2957 em = get_extent(inode, NULL, 0, off, max - off, 0);
2965 off = em->start + em->len;
2969 em_start = em->start;
2975 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
2977 flags |= FIEMAP_EXTENT_LAST;
2978 } else if (em->block_start == EXTENT_MAP_HOLE) {
2979 flags |= FIEMAP_EXTENT_UNWRITTEN;
2980 } else if (em->block_start == EXTENT_MAP_INLINE) {
2981 flags |= (FIEMAP_EXTENT_DATA_INLINE |
2982 FIEMAP_EXTENT_NOT_ALIGNED);
2983 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
2984 flags |= (FIEMAP_EXTENT_DELALLOC |
2985 FIEMAP_EXTENT_UNKNOWN);
2987 disko = em->block_start;
2989 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2990 flags |= FIEMAP_EXTENT_ENCODED;
2992 emflags = em->flags;
2993 free_extent_map(em);
2997 em = get_extent(inode, NULL, 0, off, max - off, 0);
3004 emflags = em->flags;
3006 if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) {
3007 flags |= FIEMAP_EXTENT_LAST;
3011 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3017 free_extent_map(em);
3019 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
3024 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
3028 struct address_space *mapping;
3031 return eb->first_page;
3032 i += eb->start >> PAGE_CACHE_SHIFT;
3033 mapping = eb->first_page->mapping;
3038 * extent_buffer_page is only called after pinning the page
3039 * by increasing the reference count. So we know the page must
3040 * be in the radix tree.
3043 p = radix_tree_lookup(&mapping->page_tree, i);
3049 static inline unsigned long num_extent_pages(u64 start, u64 len)
3051 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3052 (start >> PAGE_CACHE_SHIFT);
3055 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3060 struct extent_buffer *eb = NULL;
3062 unsigned long flags;
3065 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3068 spin_lock_init(&eb->lock);
3069 init_waitqueue_head(&eb->lock_wq);
3072 spin_lock_irqsave(&leak_lock, flags);
3073 list_add(&eb->leak_list, &buffers);
3074 spin_unlock_irqrestore(&leak_lock, flags);
3076 atomic_set(&eb->refs, 1);
3081 static void __free_extent_buffer(struct extent_buffer *eb)
3084 unsigned long flags;
3085 spin_lock_irqsave(&leak_lock, flags);
3086 list_del(&eb->leak_list);
3087 spin_unlock_irqrestore(&leak_lock, flags);
3089 kmem_cache_free(extent_buffer_cache, eb);
3092 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3093 u64 start, unsigned long len,
3097 unsigned long num_pages = num_extent_pages(start, len);
3099 unsigned long index = start >> PAGE_CACHE_SHIFT;
3100 struct extent_buffer *eb;
3101 struct extent_buffer *exists = NULL;
3103 struct address_space *mapping = tree->mapping;
3106 spin_lock(&tree->buffer_lock);
3107 eb = buffer_search(tree, start);
3109 atomic_inc(&eb->refs);
3110 spin_unlock(&tree->buffer_lock);
3111 mark_page_accessed(eb->first_page);
3114 spin_unlock(&tree->buffer_lock);
3116 eb = __alloc_extent_buffer(tree, start, len, mask);
3121 eb->first_page = page0;
3124 page_cache_get(page0);
3125 mark_page_accessed(page0);
3126 set_page_extent_mapped(page0);
3127 set_page_extent_head(page0, len);
3128 uptodate = PageUptodate(page0);
3132 for (; i < num_pages; i++, index++) {
3133 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
3138 set_page_extent_mapped(p);
3139 mark_page_accessed(p);
3142 set_page_extent_head(p, len);
3144 set_page_private(p, EXTENT_PAGE_PRIVATE);
3146 if (!PageUptodate(p))
3151 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3153 spin_lock(&tree->buffer_lock);
3154 exists = buffer_tree_insert(tree, start, &eb->rb_node);
3156 /* add one reference for the caller */
3157 atomic_inc(&exists->refs);
3158 spin_unlock(&tree->buffer_lock);
3161 spin_unlock(&tree->buffer_lock);
3163 /* add one reference for the tree */
3164 atomic_inc(&eb->refs);
3168 if (!atomic_dec_and_test(&eb->refs))
3170 for (index = 1; index < i; index++)
3171 page_cache_release(extent_buffer_page(eb, index));
3172 page_cache_release(extent_buffer_page(eb, 0));
3173 __free_extent_buffer(eb);
3177 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3178 u64 start, unsigned long len,
3181 struct extent_buffer *eb;
3183 spin_lock(&tree->buffer_lock);
3184 eb = buffer_search(tree, start);
3186 atomic_inc(&eb->refs);
3187 spin_unlock(&tree->buffer_lock);
3190 mark_page_accessed(eb->first_page);
3195 void free_extent_buffer(struct extent_buffer *eb)
3200 if (!atomic_dec_and_test(&eb->refs))
3206 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3207 struct extent_buffer *eb)
3210 unsigned long num_pages;
3213 num_pages = num_extent_pages(eb->start, eb->len);
3215 for (i = 0; i < num_pages; i++) {
3216 page = extent_buffer_page(eb, i);
3217 if (!PageDirty(page))
3222 set_page_extent_head(page, eb->len);
3224 set_page_private(page, EXTENT_PAGE_PRIVATE);
3226 clear_page_dirty_for_io(page);
3227 spin_lock_irq(&page->mapping->tree_lock);
3228 if (!PageDirty(page)) {
3229 radix_tree_tag_clear(&page->mapping->page_tree,
3231 PAGECACHE_TAG_DIRTY);
3233 spin_unlock_irq(&page->mapping->tree_lock);
3239 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3240 struct extent_buffer *eb)
3242 return wait_on_extent_writeback(tree, eb->start,
3243 eb->start + eb->len - 1);
3246 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3247 struct extent_buffer *eb)
3250 unsigned long num_pages;
3253 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3254 num_pages = num_extent_pages(eb->start, eb->len);
3255 for (i = 0; i < num_pages; i++)
3256 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3260 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3261 struct extent_buffer *eb)
3265 unsigned long num_pages;
3267 num_pages = num_extent_pages(eb->start, eb->len);
3268 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3270 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3272 for (i = 0; i < num_pages; i++) {
3273 page = extent_buffer_page(eb, i);
3275 ClearPageUptodate(page);
3280 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3281 struct extent_buffer *eb)
3285 unsigned long num_pages;
3287 num_pages = num_extent_pages(eb->start, eb->len);
3289 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3291 for (i = 0; i < num_pages; i++) {
3292 page = extent_buffer_page(eb, i);
3293 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3294 ((i == num_pages - 1) &&
3295 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3296 check_page_uptodate(tree, page);
3299 SetPageUptodate(page);
3304 int extent_range_uptodate(struct extent_io_tree *tree,
3309 int pg_uptodate = 1;
3311 unsigned long index;
3313 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL);
3316 while (start <= end) {
3317 index = start >> PAGE_CACHE_SHIFT;
3318 page = find_get_page(tree->mapping, index);
3319 uptodate = PageUptodate(page);
3320 page_cache_release(page);
3325 start += PAGE_CACHE_SIZE;
3330 int extent_buffer_uptodate(struct extent_io_tree *tree,
3331 struct extent_buffer *eb)
3334 unsigned long num_pages;
3337 int pg_uptodate = 1;
3339 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3342 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3343 EXTENT_UPTODATE, 1, NULL);
3347 num_pages = num_extent_pages(eb->start, eb->len);
3348 for (i = 0; i < num_pages; i++) {
3349 page = extent_buffer_page(eb, i);
3350 if (!PageUptodate(page)) {
3358 int read_extent_buffer_pages(struct extent_io_tree *tree,
3359 struct extent_buffer *eb,
3360 u64 start, int wait,
3361 get_extent_t *get_extent, int mirror_num)
3364 unsigned long start_i;
3368 int locked_pages = 0;
3369 int all_uptodate = 1;
3370 int inc_all_pages = 0;
3371 unsigned long num_pages;
3372 struct bio *bio = NULL;
3373 unsigned long bio_flags = 0;
3375 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3378 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3379 EXTENT_UPTODATE, 1, NULL)) {
3384 WARN_ON(start < eb->start);
3385 start_i = (start >> PAGE_CACHE_SHIFT) -
3386 (eb->start >> PAGE_CACHE_SHIFT);
3391 num_pages = num_extent_pages(eb->start, eb->len);
3392 for (i = start_i; i < num_pages; i++) {
3393 page = extent_buffer_page(eb, i);
3395 if (!trylock_page(page))
3401 if (!PageUptodate(page))
3406 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3410 for (i = start_i; i < num_pages; i++) {
3411 page = extent_buffer_page(eb, i);
3413 page_cache_get(page);
3414 if (!PageUptodate(page)) {
3417 ClearPageError(page);
3418 err = __extent_read_full_page(tree, page,
3420 mirror_num, &bio_flags);
3429 submit_one_bio(READ, bio, mirror_num, bio_flags);
3434 for (i = start_i; i < num_pages; i++) {
3435 page = extent_buffer_page(eb, i);
3436 wait_on_page_locked(page);
3437 if (!PageUptodate(page))
3442 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3447 while (locked_pages > 0) {
3448 page = extent_buffer_page(eb, i);
3456 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3457 unsigned long start,
3464 char *dst = (char *)dstv;
3465 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3466 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3468 WARN_ON(start > eb->len);
3469 WARN_ON(start + len > eb->start + eb->len);
3471 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3474 page = extent_buffer_page(eb, i);
3476 cur = min(len, (PAGE_CACHE_SIZE - offset));
3477 kaddr = kmap_atomic(page, KM_USER1);
3478 memcpy(dst, kaddr + offset, cur);
3479 kunmap_atomic(kaddr, KM_USER1);
3488 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3489 unsigned long min_len, char **token, char **map,
3490 unsigned long *map_start,
3491 unsigned long *map_len, int km)
3493 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3496 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3497 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3498 unsigned long end_i = (start_offset + start + min_len - 1) >>
3505 offset = start_offset;
3509 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3512 if (start + min_len > eb->len) {
3513 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3514 "wanted %lu %lu\n", (unsigned long long)eb->start,
3515 eb->len, start, min_len);
3519 p = extent_buffer_page(eb, i);
3520 kaddr = kmap_atomic(p, km);
3522 *map = kaddr + offset;
3523 *map_len = PAGE_CACHE_SIZE - offset;
3527 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3528 unsigned long min_len,
3529 char **token, char **map,
3530 unsigned long *map_start,
3531 unsigned long *map_len, int km)
3535 if (eb->map_token) {
3536 unmap_extent_buffer(eb, eb->map_token, km);
3537 eb->map_token = NULL;
3540 err = map_private_extent_buffer(eb, start, min_len, token, map,
3541 map_start, map_len, km);
3543 eb->map_token = *token;
3545 eb->map_start = *map_start;
3546 eb->map_len = *map_len;
3551 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3553 kunmap_atomic(token, km);
3556 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3557 unsigned long start,
3564 char *ptr = (char *)ptrv;
3565 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3566 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3569 WARN_ON(start > eb->len);
3570 WARN_ON(start + len > eb->start + eb->len);
3572 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3575 page = extent_buffer_page(eb, i);
3577 cur = min(len, (PAGE_CACHE_SIZE - offset));
3579 kaddr = kmap_atomic(page, KM_USER0);
3580 ret = memcmp(ptr, kaddr + offset, cur);
3581 kunmap_atomic(kaddr, KM_USER0);
3593 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3594 unsigned long start, unsigned long len)
3600 char *src = (char *)srcv;
3601 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3602 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3604 WARN_ON(start > eb->len);
3605 WARN_ON(start + len > eb->start + eb->len);
3607 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3610 page = extent_buffer_page(eb, i);
3611 WARN_ON(!PageUptodate(page));
3613 cur = min(len, PAGE_CACHE_SIZE - offset);
3614 kaddr = kmap_atomic(page, KM_USER1);
3615 memcpy(kaddr + offset, src, cur);
3616 kunmap_atomic(kaddr, KM_USER1);
3625 void memset_extent_buffer(struct extent_buffer *eb, char c,
3626 unsigned long start, unsigned long len)
3632 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3633 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3635 WARN_ON(start > eb->len);
3636 WARN_ON(start + len > eb->start + eb->len);
3638 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3641 page = extent_buffer_page(eb, i);
3642 WARN_ON(!PageUptodate(page));
3644 cur = min(len, PAGE_CACHE_SIZE - offset);
3645 kaddr = kmap_atomic(page, KM_USER0);
3646 memset(kaddr + offset, c, cur);
3647 kunmap_atomic(kaddr, KM_USER0);
3655 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3656 unsigned long dst_offset, unsigned long src_offset,
3659 u64 dst_len = dst->len;
3664 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3665 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3667 WARN_ON(src->len != dst_len);
3669 offset = (start_offset + dst_offset) &
3670 ((unsigned long)PAGE_CACHE_SIZE - 1);
3673 page = extent_buffer_page(dst, i);
3674 WARN_ON(!PageUptodate(page));
3676 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3678 kaddr = kmap_atomic(page, KM_USER0);
3679 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3680 kunmap_atomic(kaddr, KM_USER0);
3689 static void move_pages(struct page *dst_page, struct page *src_page,
3690 unsigned long dst_off, unsigned long src_off,
3693 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3694 if (dst_page == src_page) {
3695 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3697 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3698 char *p = dst_kaddr + dst_off + len;
3699 char *s = src_kaddr + src_off + len;
3704 kunmap_atomic(src_kaddr, KM_USER1);
3706 kunmap_atomic(dst_kaddr, KM_USER0);
3709 static void copy_pages(struct page *dst_page, struct page *src_page,
3710 unsigned long dst_off, unsigned long src_off,
3713 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3716 if (dst_page != src_page)
3717 src_kaddr = kmap_atomic(src_page, KM_USER1);
3719 src_kaddr = dst_kaddr;
3721 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3722 kunmap_atomic(dst_kaddr, KM_USER0);
3723 if (dst_page != src_page)
3724 kunmap_atomic(src_kaddr, KM_USER1);
3727 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3728 unsigned long src_offset, unsigned long len)
3731 size_t dst_off_in_page;
3732 size_t src_off_in_page;
3733 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3734 unsigned long dst_i;
3735 unsigned long src_i;
3737 if (src_offset + len > dst->len) {
3738 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3739 "len %lu dst len %lu\n", src_offset, len, dst->len);
3742 if (dst_offset + len > dst->len) {
3743 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3744 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3749 dst_off_in_page = (start_offset + dst_offset) &
3750 ((unsigned long)PAGE_CACHE_SIZE - 1);
3751 src_off_in_page = (start_offset + src_offset) &
3752 ((unsigned long)PAGE_CACHE_SIZE - 1);
3754 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3755 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3757 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3759 cur = min_t(unsigned long, cur,
3760 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3762 copy_pages(extent_buffer_page(dst, dst_i),
3763 extent_buffer_page(dst, src_i),
3764 dst_off_in_page, src_off_in_page, cur);
3772 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3773 unsigned long src_offset, unsigned long len)
3776 size_t dst_off_in_page;
3777 size_t src_off_in_page;
3778 unsigned long dst_end = dst_offset + len - 1;
3779 unsigned long src_end = src_offset + len - 1;
3780 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3781 unsigned long dst_i;
3782 unsigned long src_i;
3784 if (src_offset + len > dst->len) {
3785 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3786 "len %lu len %lu\n", src_offset, len, dst->len);
3789 if (dst_offset + len > dst->len) {
3790 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3791 "len %lu len %lu\n", dst_offset, len, dst->len);
3794 if (dst_offset < src_offset) {
3795 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3799 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3800 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3802 dst_off_in_page = (start_offset + dst_end) &
3803 ((unsigned long)PAGE_CACHE_SIZE - 1);
3804 src_off_in_page = (start_offset + src_end) &
3805 ((unsigned long)PAGE_CACHE_SIZE - 1);
3807 cur = min_t(unsigned long, len, src_off_in_page + 1);
3808 cur = min(cur, dst_off_in_page + 1);
3809 move_pages(extent_buffer_page(dst, dst_i),
3810 extent_buffer_page(dst, src_i),
3811 dst_off_in_page - cur + 1,
3812 src_off_in_page - cur + 1, cur);
3820 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3822 u64 start = page_offset(page);
3823 struct extent_buffer *eb;
3826 unsigned long num_pages;
3828 spin_lock(&tree->buffer_lock);
3829 eb = buffer_search(tree, start);
3833 if (atomic_read(&eb->refs) > 1) {
3837 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3841 /* at this point we can safely release the extent buffer */
3842 num_pages = num_extent_pages(eb->start, eb->len);
3843 for (i = 0; i < num_pages; i++)
3844 page_cache_release(extent_buffer_page(eb, i));
3845 rb_erase(&eb->rb_node, &tree->buffer);
3846 __free_extent_buffer(eb);
3848 spin_unlock(&tree->buffer_lock);