1 #include <linux/bitops.h>
2 #include <linux/slab.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include "extent_io.h"
14 #include "extent_map.h"
17 #include "btrfs_inode.h"
19 static struct kmem_cache *extent_state_cache;
20 static struct kmem_cache *extent_buffer_cache;
22 static LIST_HEAD(buffers);
23 static LIST_HEAD(states);
27 static DEFINE_SPINLOCK(leak_lock);
30 #define BUFFER_LRU_MAX 64
35 struct rb_node rb_node;
38 struct extent_page_data {
40 struct extent_io_tree *tree;
41 get_extent_t *get_extent;
43 /* tells writepage not to lock the state bits for this range
44 * it still does the unlocking
46 unsigned int extent_locked:1;
48 /* tells the submit_bio code to use a WRITE_SYNC */
49 unsigned int sync_io:1;
52 int __init extent_io_init(void)
54 extent_state_cache = kmem_cache_create("extent_state",
55 sizeof(struct extent_state), 0,
56 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
57 if (!extent_state_cache)
60 extent_buffer_cache = kmem_cache_create("extent_buffers",
61 sizeof(struct extent_buffer), 0,
62 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
63 if (!extent_buffer_cache)
64 goto free_state_cache;
68 kmem_cache_destroy(extent_state_cache);
72 void extent_io_exit(void)
74 struct extent_state *state;
75 struct extent_buffer *eb;
77 while (!list_empty(&states)) {
78 state = list_entry(states.next, struct extent_state, leak_list);
79 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
80 "state %lu in tree %p refs %d\n",
81 (unsigned long long)state->start,
82 (unsigned long long)state->end,
83 state->state, state->tree, atomic_read(&state->refs));
84 list_del(&state->leak_list);
85 kmem_cache_free(extent_state_cache, state);
89 while (!list_empty(&buffers)) {
90 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
91 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
92 "refs %d\n", (unsigned long long)eb->start,
93 eb->len, atomic_read(&eb->refs));
94 list_del(&eb->leak_list);
95 kmem_cache_free(extent_buffer_cache, eb);
97 if (extent_state_cache)
98 kmem_cache_destroy(extent_state_cache);
99 if (extent_buffer_cache)
100 kmem_cache_destroy(extent_buffer_cache);
103 void extent_io_tree_init(struct extent_io_tree *tree,
104 struct address_space *mapping, gfp_t mask)
106 tree->state = RB_ROOT;
107 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
109 tree->dirty_bytes = 0;
110 spin_lock_init(&tree->lock);
111 spin_lock_init(&tree->buffer_lock);
112 tree->mapping = mapping;
115 static struct extent_state *alloc_extent_state(gfp_t mask)
117 struct extent_state *state;
122 state = kmem_cache_alloc(extent_state_cache, mask);
129 spin_lock_irqsave(&leak_lock, flags);
130 list_add(&state->leak_list, &states);
131 spin_unlock_irqrestore(&leak_lock, flags);
133 atomic_set(&state->refs, 1);
134 init_waitqueue_head(&state->wq);
138 void free_extent_state(struct extent_state *state)
142 if (atomic_dec_and_test(&state->refs)) {
146 WARN_ON(state->tree);
148 spin_lock_irqsave(&leak_lock, flags);
149 list_del(&state->leak_list);
150 spin_unlock_irqrestore(&leak_lock, flags);
152 kmem_cache_free(extent_state_cache, state);
156 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
157 struct rb_node *node)
159 struct rb_node **p = &root->rb_node;
160 struct rb_node *parent = NULL;
161 struct tree_entry *entry;
165 entry = rb_entry(parent, struct tree_entry, rb_node);
167 if (offset < entry->start)
169 else if (offset > entry->end)
175 entry = rb_entry(node, struct tree_entry, rb_node);
176 rb_link_node(node, parent, p);
177 rb_insert_color(node, root);
181 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
182 struct rb_node **prev_ret,
183 struct rb_node **next_ret)
185 struct rb_root *root = &tree->state;
186 struct rb_node *n = root->rb_node;
187 struct rb_node *prev = NULL;
188 struct rb_node *orig_prev = NULL;
189 struct tree_entry *entry;
190 struct tree_entry *prev_entry = NULL;
193 entry = rb_entry(n, struct tree_entry, rb_node);
197 if (offset < entry->start)
199 else if (offset > entry->end)
207 while (prev && offset > prev_entry->end) {
208 prev = rb_next(prev);
209 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
216 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
217 while (prev && offset < prev_entry->start) {
218 prev = rb_prev(prev);
219 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
226 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
229 struct rb_node *prev = NULL;
232 ret = __etree_search(tree, offset, &prev, NULL);
238 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
239 struct extent_state *other)
241 if (tree->ops && tree->ops->merge_extent_hook)
242 tree->ops->merge_extent_hook(tree->mapping->host, new,
247 * utility function to look for merge candidates inside a given range.
248 * Any extents with matching state are merged together into a single
249 * extent in the tree. Extents with EXTENT_IO in their state field
250 * are not merged because the end_io handlers need to be able to do
251 * operations on them without sleeping (or doing allocations/splits).
253 * This should be called with the tree lock held.
255 static int merge_state(struct extent_io_tree *tree,
256 struct extent_state *state)
258 struct extent_state *other;
259 struct rb_node *other_node;
261 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
264 other_node = rb_prev(&state->rb_node);
266 other = rb_entry(other_node, struct extent_state, rb_node);
267 if (other->end == state->start - 1 &&
268 other->state == state->state) {
269 merge_cb(tree, state, other);
270 state->start = other->start;
272 rb_erase(&other->rb_node, &tree->state);
273 free_extent_state(other);
276 other_node = rb_next(&state->rb_node);
278 other = rb_entry(other_node, struct extent_state, rb_node);
279 if (other->start == state->end + 1 &&
280 other->state == state->state) {
281 merge_cb(tree, state, other);
282 other->start = state->start;
284 rb_erase(&state->rb_node, &tree->state);
285 free_extent_state(state);
293 static int set_state_cb(struct extent_io_tree *tree,
294 struct extent_state *state, int *bits)
296 if (tree->ops && tree->ops->set_bit_hook) {
297 return tree->ops->set_bit_hook(tree->mapping->host,
304 static void clear_state_cb(struct extent_io_tree *tree,
305 struct extent_state *state, int *bits)
307 if (tree->ops && tree->ops->clear_bit_hook)
308 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
312 * insert an extent_state struct into the tree. 'bits' are set on the
313 * struct before it is inserted.
315 * This may return -EEXIST if the extent is already there, in which case the
316 * state struct is freed.
318 * The tree lock is not taken internally. This is a utility function and
319 * probably isn't what you want to call (see set/clear_extent_bit).
321 static int insert_state(struct extent_io_tree *tree,
322 struct extent_state *state, u64 start, u64 end,
325 struct rb_node *node;
326 int bits_to_set = *bits & ~EXTENT_CTLBITS;
330 printk(KERN_ERR "btrfs end < start %llu %llu\n",
331 (unsigned long long)end,
332 (unsigned long long)start);
335 state->start = start;
337 ret = set_state_cb(tree, state, bits);
341 if (bits_to_set & EXTENT_DIRTY)
342 tree->dirty_bytes += end - start + 1;
343 state->state |= bits_to_set;
344 node = tree_insert(&tree->state, end, &state->rb_node);
346 struct extent_state *found;
347 found = rb_entry(node, struct extent_state, rb_node);
348 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
349 "%llu %llu\n", (unsigned long long)found->start,
350 (unsigned long long)found->end,
351 (unsigned long long)start, (unsigned long long)end);
352 free_extent_state(state);
356 merge_state(tree, state);
360 static int split_cb(struct extent_io_tree *tree, struct extent_state *orig,
363 if (tree->ops && tree->ops->split_extent_hook)
364 return tree->ops->split_extent_hook(tree->mapping->host,
370 * split a given extent state struct in two, inserting the preallocated
371 * struct 'prealloc' as the newly created second half. 'split' indicates an
372 * offset inside 'orig' where it should be split.
375 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
376 * are two extent state structs in the tree:
377 * prealloc: [orig->start, split - 1]
378 * orig: [ split, orig->end ]
380 * The tree locks are not taken by this function. They need to be held
383 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
384 struct extent_state *prealloc, u64 split)
386 struct rb_node *node;
388 split_cb(tree, orig, split);
390 prealloc->start = orig->start;
391 prealloc->end = split - 1;
392 prealloc->state = orig->state;
395 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
397 free_extent_state(prealloc);
400 prealloc->tree = tree;
405 * utility function to clear some bits in an extent state struct.
406 * it will optionally wake up any one waiting on this state (wake == 1), or
407 * forcibly remove the state from the tree (delete == 1).
409 * If no bits are set on the state struct after clearing things, the
410 * struct is freed and removed from the tree
412 static int clear_state_bit(struct extent_io_tree *tree,
413 struct extent_state *state,
416 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
417 int ret = state->state & bits_to_clear;
419 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
420 u64 range = state->end - state->start + 1;
421 WARN_ON(range > tree->dirty_bytes);
422 tree->dirty_bytes -= range;
424 clear_state_cb(tree, state, bits);
425 state->state &= ~bits_to_clear;
428 if (state->state == 0) {
430 rb_erase(&state->rb_node, &tree->state);
432 free_extent_state(state);
437 merge_state(tree, state);
443 * clear some bits on a range in the tree. This may require splitting
444 * or inserting elements in the tree, so the gfp mask is used to
445 * indicate which allocations or sleeping are allowed.
447 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
448 * the given range from the tree regardless of state (ie for truncate).
450 * the range [start, end] is inclusive.
452 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
453 * bits were already set, or zero if none of the bits were already set.
455 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
456 int bits, int wake, int delete,
457 struct extent_state **cached_state,
460 struct extent_state *state;
461 struct extent_state *cached;
462 struct extent_state *prealloc = NULL;
463 struct rb_node *next_node;
464 struct rb_node *node;
471 bits |= ~EXTENT_CTLBITS;
472 bits |= EXTENT_FIRST_DELALLOC;
474 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
477 if (!prealloc && (mask & __GFP_WAIT)) {
478 prealloc = alloc_extent_state(mask);
483 spin_lock(&tree->lock);
485 cached = *cached_state;
488 *cached_state = NULL;
492 if (cached && cached->tree && cached->start == start) {
494 atomic_dec(&cached->refs);
499 free_extent_state(cached);
502 * this search will find the extents that end after
505 node = tree_search(tree, start);
508 state = rb_entry(node, struct extent_state, rb_node);
510 if (state->start > end)
512 WARN_ON(state->end < start);
513 last_end = state->end;
516 * | ---- desired range ---- |
518 * | ------------- state -------------- |
520 * We need to split the extent we found, and may flip
521 * bits on second half.
523 * If the extent we found extends past our range, we
524 * just split and search again. It'll get split again
525 * the next time though.
527 * If the extent we found is inside our range, we clear
528 * the desired bit on it.
531 if (state->start < start) {
533 prealloc = alloc_extent_state(GFP_ATOMIC);
534 err = split_state(tree, state, prealloc, start);
535 BUG_ON(err == -EEXIST);
539 if (state->end <= end) {
540 set |= clear_state_bit(tree, state, &bits, wake);
541 if (last_end == (u64)-1)
543 start = last_end + 1;
548 * | ---- desired range ---- |
550 * We need to split the extent, and clear the bit
553 if (state->start <= end && state->end > end) {
555 prealloc = alloc_extent_state(GFP_ATOMIC);
556 err = split_state(tree, state, prealloc, end + 1);
557 BUG_ON(err == -EEXIST);
561 set |= clear_state_bit(tree, prealloc, &bits, wake);
567 if (state->end < end && prealloc && !need_resched())
568 next_node = rb_next(&state->rb_node);
572 set |= clear_state_bit(tree, state, &bits, wake);
573 if (last_end == (u64)-1)
575 start = last_end + 1;
576 if (start <= end && next_node) {
577 state = rb_entry(next_node, struct extent_state,
579 if (state->start == start)
585 spin_unlock(&tree->lock);
587 free_extent_state(prealloc);
594 spin_unlock(&tree->lock);
595 if (mask & __GFP_WAIT)
600 static int wait_on_state(struct extent_io_tree *tree,
601 struct extent_state *state)
602 __releases(tree->lock)
603 __acquires(tree->lock)
606 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
607 spin_unlock(&tree->lock);
609 spin_lock(&tree->lock);
610 finish_wait(&state->wq, &wait);
615 * waits for one or more bits to clear on a range in the state tree.
616 * The range [start, end] is inclusive.
617 * The tree lock is taken by this function
619 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
621 struct extent_state *state;
622 struct rb_node *node;
624 spin_lock(&tree->lock);
628 * this search will find all the extents that end after
631 node = tree_search(tree, start);
635 state = rb_entry(node, struct extent_state, rb_node);
637 if (state->start > end)
640 if (state->state & bits) {
641 start = state->start;
642 atomic_inc(&state->refs);
643 wait_on_state(tree, state);
644 free_extent_state(state);
647 start = state->end + 1;
652 if (need_resched()) {
653 spin_unlock(&tree->lock);
655 spin_lock(&tree->lock);
659 spin_unlock(&tree->lock);
663 static int set_state_bits(struct extent_io_tree *tree,
664 struct extent_state *state,
668 int bits_to_set = *bits & ~EXTENT_CTLBITS;
670 ret = set_state_cb(tree, state, bits);
673 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
674 u64 range = state->end - state->start + 1;
675 tree->dirty_bytes += range;
677 state->state |= bits_to_set;
682 static void cache_state(struct extent_state *state,
683 struct extent_state **cached_ptr)
685 if (cached_ptr && !(*cached_ptr)) {
686 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
688 atomic_inc(&state->refs);
694 * set some bits on a range in the tree. This may require allocations or
695 * sleeping, so the gfp mask is used to indicate what is allowed.
697 * If any of the exclusive bits are set, this will fail with -EEXIST if some
698 * part of the range already has the desired bits set. The start of the
699 * existing range is returned in failed_start in this case.
701 * [start, end] is inclusive This takes the tree lock.
704 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
705 int bits, int exclusive_bits, u64 *failed_start,
706 struct extent_state **cached_state, gfp_t mask)
708 struct extent_state *state;
709 struct extent_state *prealloc = NULL;
710 struct rb_node *node;
715 bits |= EXTENT_FIRST_DELALLOC;
717 if (!prealloc && (mask & __GFP_WAIT)) {
718 prealloc = alloc_extent_state(mask);
723 spin_lock(&tree->lock);
724 if (cached_state && *cached_state) {
725 state = *cached_state;
726 if (state->start == start && state->tree) {
727 node = &state->rb_node;
732 * this search will find all the extents that end after
735 node = tree_search(tree, start);
737 err = insert_state(tree, prealloc, start, end, &bits);
739 BUG_ON(err == -EEXIST);
742 state = rb_entry(node, struct extent_state, rb_node);
744 last_start = state->start;
745 last_end = state->end;
748 * | ---- desired range ---- |
751 * Just lock what we found and keep going
753 if (state->start == start && state->end <= end) {
754 struct rb_node *next_node;
755 if (state->state & exclusive_bits) {
756 *failed_start = state->start;
761 err = set_state_bits(tree, state, &bits);
765 cache_state(state, cached_state);
766 merge_state(tree, state);
767 if (last_end == (u64)-1)
770 start = last_end + 1;
771 if (start < end && prealloc && !need_resched()) {
772 next_node = rb_next(node);
774 state = rb_entry(next_node, struct extent_state,
776 if (state->start == start)
784 * | ---- desired range ---- |
787 * | ------------- state -------------- |
789 * We need to split the extent we found, and may flip bits on
792 * If the extent we found extends past our
793 * range, we just split and search again. It'll get split
794 * again the next time though.
796 * If the extent we found is inside our range, we set the
799 if (state->start < start) {
800 if (state->state & exclusive_bits) {
801 *failed_start = start;
805 err = split_state(tree, state, prealloc, start);
806 BUG_ON(err == -EEXIST);
810 if (state->end <= end) {
811 err = set_state_bits(tree, state, &bits);
814 cache_state(state, cached_state);
815 merge_state(tree, state);
816 if (last_end == (u64)-1)
818 start = last_end + 1;
823 * | ---- desired range ---- |
824 * | state | or | state |
826 * There's a hole, we need to insert something in it and
827 * ignore the extent we found.
829 if (state->start > start) {
831 if (end < last_start)
834 this_end = last_start - 1;
835 err = insert_state(tree, prealloc, start, this_end,
837 BUG_ON(err == -EEXIST);
842 cache_state(prealloc, cached_state);
844 start = this_end + 1;
848 * | ---- desired range ---- |
850 * We need to split the extent, and set the bit
853 if (state->start <= end && state->end > end) {
854 if (state->state & exclusive_bits) {
855 *failed_start = start;
859 err = split_state(tree, state, prealloc, end + 1);
860 BUG_ON(err == -EEXIST);
862 err = set_state_bits(tree, prealloc, &bits);
867 cache_state(prealloc, cached_state);
868 merge_state(tree, prealloc);
876 spin_unlock(&tree->lock);
878 free_extent_state(prealloc);
885 spin_unlock(&tree->lock);
886 if (mask & __GFP_WAIT)
891 /* wrappers around set/clear extent bit */
892 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
895 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
899 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
900 int bits, gfp_t mask)
902 return set_extent_bit(tree, start, end, bits, 0, NULL,
906 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
907 int bits, gfp_t mask)
909 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
912 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
913 struct extent_state **cached_state, gfp_t mask)
915 return set_extent_bit(tree, start, end,
916 EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE,
917 0, NULL, cached_state, mask);
920 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
923 return clear_extent_bit(tree, start, end,
924 EXTENT_DIRTY | EXTENT_DELALLOC |
925 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
928 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
931 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
935 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
938 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0,
942 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
945 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
949 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
950 u64 end, struct extent_state **cached_state,
953 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
957 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
959 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
963 * either insert or lock state struct between start and end use mask to tell
964 * us if waiting is desired.
966 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
967 int bits, struct extent_state **cached_state, gfp_t mask)
972 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
973 EXTENT_LOCKED, &failed_start,
975 if (err == -EEXIST && (mask & __GFP_WAIT)) {
976 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
977 start = failed_start;
981 WARN_ON(start > end);
986 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
988 return lock_extent_bits(tree, start, end, 0, NULL, mask);
991 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
997 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
998 &failed_start, NULL, mask);
999 if (err == -EEXIST) {
1000 if (failed_start > start)
1001 clear_extent_bit(tree, start, failed_start - 1,
1002 EXTENT_LOCKED, 1, 0, NULL, mask);
1008 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1009 struct extent_state **cached, gfp_t mask)
1011 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1015 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1018 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1023 * helper function to set pages and extents in the tree dirty
1025 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
1027 unsigned long index = start >> PAGE_CACHE_SHIFT;
1028 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1031 while (index <= end_index) {
1032 page = find_get_page(tree->mapping, index);
1034 __set_page_dirty_nobuffers(page);
1035 page_cache_release(page);
1042 * helper function to set both pages and extents in the tree writeback
1044 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1046 unsigned long index = start >> PAGE_CACHE_SHIFT;
1047 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1050 while (index <= end_index) {
1051 page = find_get_page(tree->mapping, index);
1053 set_page_writeback(page);
1054 page_cache_release(page);
1061 * find the first offset in the io tree with 'bits' set. zero is
1062 * returned if we find something, and *start_ret and *end_ret are
1063 * set to reflect the state struct that was found.
1065 * If nothing was found, 1 is returned, < 0 on error
1067 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1068 u64 *start_ret, u64 *end_ret, int bits)
1070 struct rb_node *node;
1071 struct extent_state *state;
1074 spin_lock(&tree->lock);
1076 * this search will find all the extents that end after
1079 node = tree_search(tree, start);
1084 state = rb_entry(node, struct extent_state, rb_node);
1085 if (state->end >= start && (state->state & bits)) {
1086 *start_ret = state->start;
1087 *end_ret = state->end;
1091 node = rb_next(node);
1096 spin_unlock(&tree->lock);
1100 /* find the first state struct with 'bits' set after 'start', and
1101 * return it. tree->lock must be held. NULL will returned if
1102 * nothing was found after 'start'
1104 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1105 u64 start, int bits)
1107 struct rb_node *node;
1108 struct extent_state *state;
1111 * this search will find all the extents that end after
1114 node = tree_search(tree, start);
1119 state = rb_entry(node, struct extent_state, rb_node);
1120 if (state->end >= start && (state->state & bits))
1123 node = rb_next(node);
1132 * find a contiguous range of bytes in the file marked as delalloc, not
1133 * more than 'max_bytes'. start and end are used to return the range,
1135 * 1 is returned if we find something, 0 if nothing was in the tree
1137 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1138 u64 *start, u64 *end, u64 max_bytes,
1139 struct extent_state **cached_state)
1141 struct rb_node *node;
1142 struct extent_state *state;
1143 u64 cur_start = *start;
1145 u64 total_bytes = 0;
1147 spin_lock(&tree->lock);
1150 * this search will find all the extents that end after
1153 node = tree_search(tree, cur_start);
1161 state = rb_entry(node, struct extent_state, rb_node);
1162 if (found && (state->start != cur_start ||
1163 (state->state & EXTENT_BOUNDARY))) {
1166 if (!(state->state & EXTENT_DELALLOC)) {
1172 *start = state->start;
1173 *cached_state = state;
1174 atomic_inc(&state->refs);
1178 cur_start = state->end + 1;
1179 node = rb_next(node);
1182 total_bytes += state->end - state->start + 1;
1183 if (total_bytes >= max_bytes)
1187 spin_unlock(&tree->lock);
1191 static noinline int __unlock_for_delalloc(struct inode *inode,
1192 struct page *locked_page,
1196 struct page *pages[16];
1197 unsigned long index = start >> PAGE_CACHE_SHIFT;
1198 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1199 unsigned long nr_pages = end_index - index + 1;
1202 if (index == locked_page->index && end_index == index)
1205 while (nr_pages > 0) {
1206 ret = find_get_pages_contig(inode->i_mapping, index,
1207 min_t(unsigned long, nr_pages,
1208 ARRAY_SIZE(pages)), pages);
1209 for (i = 0; i < ret; i++) {
1210 if (pages[i] != locked_page)
1211 unlock_page(pages[i]);
1212 page_cache_release(pages[i]);
1221 static noinline int lock_delalloc_pages(struct inode *inode,
1222 struct page *locked_page,
1226 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1227 unsigned long start_index = index;
1228 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1229 unsigned long pages_locked = 0;
1230 struct page *pages[16];
1231 unsigned long nrpages;
1235 /* the caller is responsible for locking the start index */
1236 if (index == locked_page->index && index == end_index)
1239 /* skip the page at the start index */
1240 nrpages = end_index - index + 1;
1241 while (nrpages > 0) {
1242 ret = find_get_pages_contig(inode->i_mapping, index,
1243 min_t(unsigned long,
1244 nrpages, ARRAY_SIZE(pages)), pages);
1249 /* now we have an array of pages, lock them all */
1250 for (i = 0; i < ret; i++) {
1252 * the caller is taking responsibility for
1255 if (pages[i] != locked_page) {
1256 lock_page(pages[i]);
1257 if (!PageDirty(pages[i]) ||
1258 pages[i]->mapping != inode->i_mapping) {
1260 unlock_page(pages[i]);
1261 page_cache_release(pages[i]);
1265 page_cache_release(pages[i]);
1274 if (ret && pages_locked) {
1275 __unlock_for_delalloc(inode, locked_page,
1277 ((u64)(start_index + pages_locked - 1)) <<
1284 * find a contiguous range of bytes in the file marked as delalloc, not
1285 * more than 'max_bytes'. start and end are used to return the range,
1287 * 1 is returned if we find something, 0 if nothing was in the tree
1289 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1290 struct extent_io_tree *tree,
1291 struct page *locked_page,
1292 u64 *start, u64 *end,
1298 struct extent_state *cached_state = NULL;
1303 /* step one, find a bunch of delalloc bytes starting at start */
1304 delalloc_start = *start;
1306 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1307 max_bytes, &cached_state);
1308 if (!found || delalloc_end <= *start) {
1309 *start = delalloc_start;
1310 *end = delalloc_end;
1311 free_extent_state(cached_state);
1316 * start comes from the offset of locked_page. We have to lock
1317 * pages in order, so we can't process delalloc bytes before
1320 if (delalloc_start < *start)
1321 delalloc_start = *start;
1324 * make sure to limit the number of pages we try to lock down
1327 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1328 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1330 /* step two, lock all the pages after the page that has start */
1331 ret = lock_delalloc_pages(inode, locked_page,
1332 delalloc_start, delalloc_end);
1333 if (ret == -EAGAIN) {
1334 /* some of the pages are gone, lets avoid looping by
1335 * shortening the size of the delalloc range we're searching
1337 free_extent_state(cached_state);
1339 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1340 max_bytes = PAGE_CACHE_SIZE - offset;
1350 /* step three, lock the state bits for the whole range */
1351 lock_extent_bits(tree, delalloc_start, delalloc_end,
1352 0, &cached_state, GFP_NOFS);
1354 /* then test to make sure it is all still delalloc */
1355 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1356 EXTENT_DELALLOC, 1, cached_state);
1358 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1359 &cached_state, GFP_NOFS);
1360 __unlock_for_delalloc(inode, locked_page,
1361 delalloc_start, delalloc_end);
1365 free_extent_state(cached_state);
1366 *start = delalloc_start;
1367 *end = delalloc_end;
1372 int extent_clear_unlock_delalloc(struct inode *inode,
1373 struct extent_io_tree *tree,
1374 u64 start, u64 end, struct page *locked_page,
1378 struct page *pages[16];
1379 unsigned long index = start >> PAGE_CACHE_SHIFT;
1380 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1381 unsigned long nr_pages = end_index - index + 1;
1385 if (op & EXTENT_CLEAR_UNLOCK)
1386 clear_bits |= EXTENT_LOCKED;
1387 if (op & EXTENT_CLEAR_DIRTY)
1388 clear_bits |= EXTENT_DIRTY;
1390 if (op & EXTENT_CLEAR_DELALLOC)
1391 clear_bits |= EXTENT_DELALLOC;
1393 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1394 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1395 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1396 EXTENT_SET_PRIVATE2)))
1399 while (nr_pages > 0) {
1400 ret = find_get_pages_contig(inode->i_mapping, index,
1401 min_t(unsigned long,
1402 nr_pages, ARRAY_SIZE(pages)), pages);
1403 for (i = 0; i < ret; i++) {
1405 if (op & EXTENT_SET_PRIVATE2)
1406 SetPagePrivate2(pages[i]);
1408 if (pages[i] == locked_page) {
1409 page_cache_release(pages[i]);
1412 if (op & EXTENT_CLEAR_DIRTY)
1413 clear_page_dirty_for_io(pages[i]);
1414 if (op & EXTENT_SET_WRITEBACK)
1415 set_page_writeback(pages[i]);
1416 if (op & EXTENT_END_WRITEBACK)
1417 end_page_writeback(pages[i]);
1418 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1419 unlock_page(pages[i]);
1420 page_cache_release(pages[i]);
1430 * count the number of bytes in the tree that have a given bit(s)
1431 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1432 * cached. The total number found is returned.
1434 u64 count_range_bits(struct extent_io_tree *tree,
1435 u64 *start, u64 search_end, u64 max_bytes,
1438 struct rb_node *node;
1439 struct extent_state *state;
1440 u64 cur_start = *start;
1441 u64 total_bytes = 0;
1444 if (search_end <= cur_start) {
1449 spin_lock(&tree->lock);
1450 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1451 total_bytes = tree->dirty_bytes;
1455 * this search will find all the extents that end after
1458 node = tree_search(tree, cur_start);
1463 state = rb_entry(node, struct extent_state, rb_node);
1464 if (state->start > search_end)
1466 if (state->end >= cur_start && (state->state & bits)) {
1467 total_bytes += min(search_end, state->end) + 1 -
1468 max(cur_start, state->start);
1469 if (total_bytes >= max_bytes)
1472 *start = state->start;
1476 node = rb_next(node);
1481 spin_unlock(&tree->lock);
1486 * set the private field for a given byte offset in the tree. If there isn't
1487 * an extent_state there already, this does nothing.
1489 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1491 struct rb_node *node;
1492 struct extent_state *state;
1495 spin_lock(&tree->lock);
1497 * this search will find all the extents that end after
1500 node = tree_search(tree, start);
1505 state = rb_entry(node, struct extent_state, rb_node);
1506 if (state->start != start) {
1510 state->private = private;
1512 spin_unlock(&tree->lock);
1516 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1518 struct rb_node *node;
1519 struct extent_state *state;
1522 spin_lock(&tree->lock);
1524 * this search will find all the extents that end after
1527 node = tree_search(tree, start);
1532 state = rb_entry(node, struct extent_state, rb_node);
1533 if (state->start != start) {
1537 *private = state->private;
1539 spin_unlock(&tree->lock);
1544 * searches a range in the state tree for a given mask.
1545 * If 'filled' == 1, this returns 1 only if every extent in the tree
1546 * has the bits set. Otherwise, 1 is returned if any bit in the
1547 * range is found set.
1549 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1550 int bits, int filled, struct extent_state *cached)
1552 struct extent_state *state = NULL;
1553 struct rb_node *node;
1556 spin_lock(&tree->lock);
1557 if (cached && cached->tree && cached->start == start)
1558 node = &cached->rb_node;
1560 node = tree_search(tree, start);
1561 while (node && start <= end) {
1562 state = rb_entry(node, struct extent_state, rb_node);
1564 if (filled && state->start > start) {
1569 if (state->start > end)
1572 if (state->state & bits) {
1576 } else if (filled) {
1581 if (state->end == (u64)-1)
1584 start = state->end + 1;
1587 node = rb_next(node);
1594 spin_unlock(&tree->lock);
1599 * helper function to set a given page up to date if all the
1600 * extents in the tree for that page are up to date
1602 static int check_page_uptodate(struct extent_io_tree *tree,
1605 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1606 u64 end = start + PAGE_CACHE_SIZE - 1;
1607 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1608 SetPageUptodate(page);
1613 * helper function to unlock a page if all the extents in the tree
1614 * for that page are unlocked
1616 static int check_page_locked(struct extent_io_tree *tree,
1619 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1620 u64 end = start + PAGE_CACHE_SIZE - 1;
1621 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1627 * helper function to end page writeback if all the extents
1628 * in the tree for that page are done with writeback
1630 static int check_page_writeback(struct extent_io_tree *tree,
1633 end_page_writeback(page);
1637 /* lots and lots of room for performance fixes in the end_bio funcs */
1640 * after a writepage IO is done, we need to:
1641 * clear the uptodate bits on error
1642 * clear the writeback bits in the extent tree for this IO
1643 * end_page_writeback if the page has no more pending IO
1645 * Scheduling is not allowed, so the extent state tree is expected
1646 * to have one and only one object corresponding to this IO.
1648 static void end_bio_extent_writepage(struct bio *bio, int err)
1650 int uptodate = err == 0;
1651 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1652 struct extent_io_tree *tree;
1659 struct page *page = bvec->bv_page;
1660 tree = &BTRFS_I(page->mapping->host)->io_tree;
1662 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1664 end = start + bvec->bv_len - 1;
1666 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1671 if (--bvec >= bio->bi_io_vec)
1672 prefetchw(&bvec->bv_page->flags);
1673 if (tree->ops && tree->ops->writepage_end_io_hook) {
1674 ret = tree->ops->writepage_end_io_hook(page, start,
1675 end, NULL, uptodate);
1680 if (!uptodate && tree->ops &&
1681 tree->ops->writepage_io_failed_hook) {
1682 ret = tree->ops->writepage_io_failed_hook(bio, page,
1685 uptodate = (err == 0);
1691 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
1692 ClearPageUptodate(page);
1697 end_page_writeback(page);
1699 check_page_writeback(tree, page);
1700 } while (bvec >= bio->bi_io_vec);
1706 * after a readpage IO is done, we need to:
1707 * clear the uptodate bits on error
1708 * set the uptodate bits if things worked
1709 * set the page up to date if all extents in the tree are uptodate
1710 * clear the lock bit in the extent tree
1711 * unlock the page if there are no other extents locked for it
1713 * Scheduling is not allowed, so the extent state tree is expected
1714 * to have one and only one object corresponding to this IO.
1716 static void end_bio_extent_readpage(struct bio *bio, int err)
1718 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1719 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
1720 struct bio_vec *bvec = bio->bi_io_vec;
1721 struct extent_io_tree *tree;
1731 struct page *page = bvec->bv_page;
1732 tree = &BTRFS_I(page->mapping->host)->io_tree;
1734 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1736 end = start + bvec->bv_len - 1;
1738 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1743 if (++bvec <= bvec_end)
1744 prefetchw(&bvec->bv_page->flags);
1746 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1747 ret = tree->ops->readpage_end_io_hook(page, start, end,
1752 if (!uptodate && tree->ops &&
1753 tree->ops->readpage_io_failed_hook) {
1754 ret = tree->ops->readpage_io_failed_hook(bio, page,
1758 test_bit(BIO_UPTODATE, &bio->bi_flags);
1766 set_extent_uptodate(tree, start, end,
1769 unlock_extent(tree, start, end, GFP_ATOMIC);
1773 SetPageUptodate(page);
1775 ClearPageUptodate(page);
1781 check_page_uptodate(tree, page);
1783 ClearPageUptodate(page);
1786 check_page_locked(tree, page);
1788 } while (bvec <= bvec_end);
1794 * IO done from prepare_write is pretty simple, we just unlock
1795 * the structs in the extent tree when done, and set the uptodate bits
1798 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1800 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1801 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1802 struct extent_io_tree *tree;
1807 struct page *page = bvec->bv_page;
1808 tree = &BTRFS_I(page->mapping->host)->io_tree;
1810 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1812 end = start + bvec->bv_len - 1;
1814 if (--bvec >= bio->bi_io_vec)
1815 prefetchw(&bvec->bv_page->flags);
1818 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1820 ClearPageUptodate(page);
1824 unlock_extent(tree, start, end, GFP_ATOMIC);
1826 } while (bvec >= bio->bi_io_vec);
1832 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1837 bio = bio_alloc(gfp_flags, nr_vecs);
1839 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1840 while (!bio && (nr_vecs /= 2))
1841 bio = bio_alloc(gfp_flags, nr_vecs);
1846 bio->bi_bdev = bdev;
1847 bio->bi_sector = first_sector;
1852 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1853 unsigned long bio_flags)
1856 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1857 struct page *page = bvec->bv_page;
1858 struct extent_io_tree *tree = bio->bi_private;
1861 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1863 bio->bi_private = NULL;
1867 if (tree->ops && tree->ops->submit_bio_hook)
1868 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1869 mirror_num, bio_flags, start);
1871 submit_bio(rw, bio);
1872 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1878 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1879 struct page *page, sector_t sector,
1880 size_t size, unsigned long offset,
1881 struct block_device *bdev,
1882 struct bio **bio_ret,
1883 unsigned long max_pages,
1884 bio_end_io_t end_io_func,
1886 unsigned long prev_bio_flags,
1887 unsigned long bio_flags)
1893 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1894 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1895 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1897 if (bio_ret && *bio_ret) {
1900 contig = bio->bi_sector == sector;
1902 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1905 if (prev_bio_flags != bio_flags || !contig ||
1906 (tree->ops && tree->ops->merge_bio_hook &&
1907 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1909 bio_add_page(bio, page, page_size, offset) < page_size) {
1910 ret = submit_one_bio(rw, bio, mirror_num,
1917 if (this_compressed)
1920 nr = bio_get_nr_vecs(bdev);
1922 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1924 bio_add_page(bio, page, page_size, offset);
1925 bio->bi_end_io = end_io_func;
1926 bio->bi_private = tree;
1931 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1936 void set_page_extent_mapped(struct page *page)
1938 if (!PagePrivate(page)) {
1939 SetPagePrivate(page);
1940 page_cache_get(page);
1941 set_page_private(page, EXTENT_PAGE_PRIVATE);
1945 static void set_page_extent_head(struct page *page, unsigned long len)
1947 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1951 * basic readpage implementation. Locked extent state structs are inserted
1952 * into the tree that are removed when the IO is done (by the end_io
1955 static int __extent_read_full_page(struct extent_io_tree *tree,
1957 get_extent_t *get_extent,
1958 struct bio **bio, int mirror_num,
1959 unsigned long *bio_flags)
1961 struct inode *inode = page->mapping->host;
1962 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1963 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1967 u64 last_byte = i_size_read(inode);
1971 struct extent_map *em;
1972 struct block_device *bdev;
1973 struct btrfs_ordered_extent *ordered;
1976 size_t page_offset = 0;
1978 size_t disk_io_size;
1979 size_t blocksize = inode->i_sb->s_blocksize;
1980 unsigned long this_bio_flag = 0;
1982 set_page_extent_mapped(page);
1986 lock_extent(tree, start, end, GFP_NOFS);
1987 ordered = btrfs_lookup_ordered_extent(inode, start);
1990 unlock_extent(tree, start, end, GFP_NOFS);
1991 btrfs_start_ordered_extent(inode, ordered, 1);
1992 btrfs_put_ordered_extent(ordered);
1995 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1997 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2000 iosize = PAGE_CACHE_SIZE - zero_offset;
2001 userpage = kmap_atomic(page, KM_USER0);
2002 memset(userpage + zero_offset, 0, iosize);
2003 flush_dcache_page(page);
2004 kunmap_atomic(userpage, KM_USER0);
2007 while (cur <= end) {
2008 if (cur >= last_byte) {
2010 iosize = PAGE_CACHE_SIZE - page_offset;
2011 userpage = kmap_atomic(page, KM_USER0);
2012 memset(userpage + page_offset, 0, iosize);
2013 flush_dcache_page(page);
2014 kunmap_atomic(userpage, KM_USER0);
2015 set_extent_uptodate(tree, cur, cur + iosize - 1,
2017 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2020 em = get_extent(inode, page, page_offset, cur,
2022 if (IS_ERR(em) || !em) {
2024 unlock_extent(tree, cur, end, GFP_NOFS);
2027 extent_offset = cur - em->start;
2028 BUG_ON(extent_map_end(em) <= cur);
2031 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2032 this_bio_flag = EXTENT_BIO_COMPRESSED;
2034 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2035 cur_end = min(extent_map_end(em) - 1, end);
2036 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2037 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2038 disk_io_size = em->block_len;
2039 sector = em->block_start >> 9;
2041 sector = (em->block_start + extent_offset) >> 9;
2042 disk_io_size = iosize;
2045 block_start = em->block_start;
2046 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2047 block_start = EXTENT_MAP_HOLE;
2048 free_extent_map(em);
2051 /* we've found a hole, just zero and go on */
2052 if (block_start == EXTENT_MAP_HOLE) {
2054 userpage = kmap_atomic(page, KM_USER0);
2055 memset(userpage + page_offset, 0, iosize);
2056 flush_dcache_page(page);
2057 kunmap_atomic(userpage, KM_USER0);
2059 set_extent_uptodate(tree, cur, cur + iosize - 1,
2061 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2063 page_offset += iosize;
2066 /* the get_extent function already copied into the page */
2067 if (test_range_bit(tree, cur, cur_end,
2068 EXTENT_UPTODATE, 1, NULL)) {
2069 check_page_uptodate(tree, page);
2070 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2072 page_offset += iosize;
2075 /* we have an inline extent but it didn't get marked up
2076 * to date. Error out
2078 if (block_start == EXTENT_MAP_INLINE) {
2080 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2082 page_offset += iosize;
2087 if (tree->ops && tree->ops->readpage_io_hook) {
2088 ret = tree->ops->readpage_io_hook(page, cur,
2092 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2094 ret = submit_extent_page(READ, tree, page,
2095 sector, disk_io_size, page_offset,
2097 end_bio_extent_readpage, mirror_num,
2101 *bio_flags = this_bio_flag;
2106 page_offset += iosize;
2109 if (!PageError(page))
2110 SetPageUptodate(page);
2116 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2117 get_extent_t *get_extent)
2119 struct bio *bio = NULL;
2120 unsigned long bio_flags = 0;
2123 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2126 submit_one_bio(READ, bio, 0, bio_flags);
2130 static noinline void update_nr_written(struct page *page,
2131 struct writeback_control *wbc,
2132 unsigned long nr_written)
2134 wbc->nr_to_write -= nr_written;
2135 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2136 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2137 page->mapping->writeback_index = page->index + nr_written;
2141 * the writepage semantics are similar to regular writepage. extent
2142 * records are inserted to lock ranges in the tree, and as dirty areas
2143 * are found, they are marked writeback. Then the lock bits are removed
2144 * and the end_io handler clears the writeback ranges
2146 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2149 struct inode *inode = page->mapping->host;
2150 struct extent_page_data *epd = data;
2151 struct extent_io_tree *tree = epd->tree;
2152 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2154 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2158 u64 last_byte = i_size_read(inode);
2162 struct extent_state *cached_state = NULL;
2163 struct extent_map *em;
2164 struct block_device *bdev;
2167 size_t pg_offset = 0;
2169 loff_t i_size = i_size_read(inode);
2170 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2176 unsigned long nr_written = 0;
2178 if (wbc->sync_mode == WB_SYNC_ALL)
2179 write_flags = WRITE_SYNC_PLUG;
2181 write_flags = WRITE;
2183 WARN_ON(!PageLocked(page));
2184 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2185 if (page->index > end_index ||
2186 (page->index == end_index && !pg_offset)) {
2187 page->mapping->a_ops->invalidatepage(page, 0);
2192 if (page->index == end_index) {
2195 userpage = kmap_atomic(page, KM_USER0);
2196 memset(userpage + pg_offset, 0,
2197 PAGE_CACHE_SIZE - pg_offset);
2198 kunmap_atomic(userpage, KM_USER0);
2199 flush_dcache_page(page);
2203 set_page_extent_mapped(page);
2205 delalloc_start = start;
2208 if (!epd->extent_locked) {
2209 u64 delalloc_to_write = 0;
2211 * make sure the wbc mapping index is at least updated
2214 update_nr_written(page, wbc, 0);
2216 while (delalloc_end < page_end) {
2217 nr_delalloc = find_lock_delalloc_range(inode, tree,
2222 if (nr_delalloc == 0) {
2223 delalloc_start = delalloc_end + 1;
2226 tree->ops->fill_delalloc(inode, page, delalloc_start,
2227 delalloc_end, &page_started,
2230 * delalloc_end is already one less than the total
2231 * length, so we don't subtract one from
2234 delalloc_to_write += (delalloc_end - delalloc_start +
2237 delalloc_start = delalloc_end + 1;
2239 if (wbc->nr_to_write < delalloc_to_write) {
2242 if (delalloc_to_write < thresh * 2)
2243 thresh = delalloc_to_write;
2244 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2248 /* did the fill delalloc function already unlock and start
2254 * we've unlocked the page, so we can't update
2255 * the mapping's writeback index, just update
2258 wbc->nr_to_write -= nr_written;
2262 if (tree->ops && tree->ops->writepage_start_hook) {
2263 ret = tree->ops->writepage_start_hook(page, start,
2265 if (ret == -EAGAIN) {
2266 redirty_page_for_writepage(wbc, page);
2267 update_nr_written(page, wbc, nr_written);
2275 * we don't want to touch the inode after unlocking the page,
2276 * so we update the mapping writeback index now
2278 update_nr_written(page, wbc, nr_written + 1);
2281 if (last_byte <= start) {
2282 if (tree->ops && tree->ops->writepage_end_io_hook)
2283 tree->ops->writepage_end_io_hook(page, start,
2288 blocksize = inode->i_sb->s_blocksize;
2290 while (cur <= end) {
2291 if (cur >= last_byte) {
2292 if (tree->ops && tree->ops->writepage_end_io_hook)
2293 tree->ops->writepage_end_io_hook(page, cur,
2297 em = epd->get_extent(inode, page, pg_offset, cur,
2299 if (IS_ERR(em) || !em) {
2304 extent_offset = cur - em->start;
2305 BUG_ON(extent_map_end(em) <= cur);
2307 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2308 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2309 sector = (em->block_start + extent_offset) >> 9;
2311 block_start = em->block_start;
2312 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2313 free_extent_map(em);
2317 * compressed and inline extents are written through other
2320 if (compressed || block_start == EXTENT_MAP_HOLE ||
2321 block_start == EXTENT_MAP_INLINE) {
2323 * end_io notification does not happen here for
2324 * compressed extents
2326 if (!compressed && tree->ops &&
2327 tree->ops->writepage_end_io_hook)
2328 tree->ops->writepage_end_io_hook(page, cur,
2331 else if (compressed) {
2332 /* we don't want to end_page_writeback on
2333 * a compressed extent. this happens
2340 pg_offset += iosize;
2343 /* leave this out until we have a page_mkwrite call */
2344 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2345 EXTENT_DIRTY, 0, NULL)) {
2347 pg_offset += iosize;
2351 if (tree->ops && tree->ops->writepage_io_hook) {
2352 ret = tree->ops->writepage_io_hook(page, cur,
2360 unsigned long max_nr = end_index + 1;
2362 set_range_writeback(tree, cur, cur + iosize - 1);
2363 if (!PageWriteback(page)) {
2364 printk(KERN_ERR "btrfs warning page %lu not "
2365 "writeback, cur %llu end %llu\n",
2366 page->index, (unsigned long long)cur,
2367 (unsigned long long)end);
2370 ret = submit_extent_page(write_flags, tree, page,
2371 sector, iosize, pg_offset,
2372 bdev, &epd->bio, max_nr,
2373 end_bio_extent_writepage,
2379 pg_offset += iosize;
2384 /* make sure the mapping tag for page dirty gets cleared */
2385 set_page_writeback(page);
2386 end_page_writeback(page);
2392 /* drop our reference on any cached states */
2393 free_extent_state(cached_state);
2398 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2399 * @mapping: address space structure to write
2400 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2401 * @writepage: function called for each page
2402 * @data: data passed to writepage function
2404 * If a page is already under I/O, write_cache_pages() skips it, even
2405 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2406 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2407 * and msync() need to guarantee that all the data which was dirty at the time
2408 * the call was made get new I/O started against them. If wbc->sync_mode is
2409 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2410 * existing IO to complete.
2412 static int extent_write_cache_pages(struct extent_io_tree *tree,
2413 struct address_space *mapping,
2414 struct writeback_control *wbc,
2415 writepage_t writepage, void *data,
2416 void (*flush_fn)(void *))
2420 int nr_to_write_done = 0;
2421 struct pagevec pvec;
2424 pgoff_t end; /* Inclusive */
2427 pagevec_init(&pvec, 0);
2428 if (wbc->range_cyclic) {
2429 index = mapping->writeback_index; /* Start from prev offset */
2432 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2433 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2437 while (!done && !nr_to_write_done && (index <= end) &&
2438 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2439 PAGECACHE_TAG_DIRTY, min(end - index,
2440 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2444 for (i = 0; i < nr_pages; i++) {
2445 struct page *page = pvec.pages[i];
2448 * At this point we hold neither mapping->tree_lock nor
2449 * lock on the page itself: the page may be truncated or
2450 * invalidated (changing page->mapping to NULL), or even
2451 * swizzled back from swapper_space to tmpfs file
2454 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2455 tree->ops->write_cache_pages_lock_hook(page);
2459 if (unlikely(page->mapping != mapping)) {
2464 if (!wbc->range_cyclic && page->index > end) {
2470 if (wbc->sync_mode != WB_SYNC_NONE) {
2471 if (PageWriteback(page))
2473 wait_on_page_writeback(page);
2476 if (PageWriteback(page) ||
2477 !clear_page_dirty_for_io(page)) {
2482 ret = (*writepage)(page, wbc, data);
2484 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2492 * the filesystem may choose to bump up nr_to_write.
2493 * We have to make sure to honor the new nr_to_write
2496 nr_to_write_done = wbc->nr_to_write <= 0;
2498 pagevec_release(&pvec);
2501 if (!scanned && !done) {
2503 * We hit the last page and there is more work to be done: wrap
2504 * back to the start of the file
2513 static void flush_epd_write_bio(struct extent_page_data *epd)
2517 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2519 submit_one_bio(WRITE, epd->bio, 0, 0);
2524 static noinline void flush_write_bio(void *data)
2526 struct extent_page_data *epd = data;
2527 flush_epd_write_bio(epd);
2530 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2531 get_extent_t *get_extent,
2532 struct writeback_control *wbc)
2535 struct address_space *mapping = page->mapping;
2536 struct extent_page_data epd = {
2539 .get_extent = get_extent,
2541 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2543 struct writeback_control wbc_writepages = {
2544 .sync_mode = wbc->sync_mode,
2545 .older_than_this = NULL,
2547 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2548 .range_end = (loff_t)-1,
2551 ret = __extent_writepage(page, wbc, &epd);
2553 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2554 __extent_writepage, &epd, flush_write_bio);
2555 flush_epd_write_bio(&epd);
2559 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2560 u64 start, u64 end, get_extent_t *get_extent,
2564 struct address_space *mapping = inode->i_mapping;
2566 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2569 struct extent_page_data epd = {
2572 .get_extent = get_extent,
2574 .sync_io = mode == WB_SYNC_ALL,
2576 struct writeback_control wbc_writepages = {
2578 .older_than_this = NULL,
2579 .nr_to_write = nr_pages * 2,
2580 .range_start = start,
2581 .range_end = end + 1,
2584 while (start <= end) {
2585 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2586 if (clear_page_dirty_for_io(page))
2587 ret = __extent_writepage(page, &wbc_writepages, &epd);
2589 if (tree->ops && tree->ops->writepage_end_io_hook)
2590 tree->ops->writepage_end_io_hook(page, start,
2591 start + PAGE_CACHE_SIZE - 1,
2595 page_cache_release(page);
2596 start += PAGE_CACHE_SIZE;
2599 flush_epd_write_bio(&epd);
2603 int extent_writepages(struct extent_io_tree *tree,
2604 struct address_space *mapping,
2605 get_extent_t *get_extent,
2606 struct writeback_control *wbc)
2609 struct extent_page_data epd = {
2612 .get_extent = get_extent,
2614 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2617 ret = extent_write_cache_pages(tree, mapping, wbc,
2618 __extent_writepage, &epd,
2620 flush_epd_write_bio(&epd);
2624 int extent_readpages(struct extent_io_tree *tree,
2625 struct address_space *mapping,
2626 struct list_head *pages, unsigned nr_pages,
2627 get_extent_t get_extent)
2629 struct bio *bio = NULL;
2631 unsigned long bio_flags = 0;
2633 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2634 struct page *page = list_entry(pages->prev, struct page, lru);
2636 prefetchw(&page->flags);
2637 list_del(&page->lru);
2638 if (!add_to_page_cache_lru(page, mapping,
2639 page->index, GFP_KERNEL)) {
2640 __extent_read_full_page(tree, page, get_extent,
2641 &bio, 0, &bio_flags);
2643 page_cache_release(page);
2645 BUG_ON(!list_empty(pages));
2647 submit_one_bio(READ, bio, 0, bio_flags);
2652 * basic invalidatepage code, this waits on any locked or writeback
2653 * ranges corresponding to the page, and then deletes any extent state
2654 * records from the tree
2656 int extent_invalidatepage(struct extent_io_tree *tree,
2657 struct page *page, unsigned long offset)
2659 struct extent_state *cached_state = NULL;
2660 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2661 u64 end = start + PAGE_CACHE_SIZE - 1;
2662 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2664 start += (offset + blocksize - 1) & ~(blocksize - 1);
2668 lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS);
2669 wait_on_page_writeback(page);
2670 clear_extent_bit(tree, start, end,
2671 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2672 EXTENT_DO_ACCOUNTING,
2673 1, 1, &cached_state, GFP_NOFS);
2678 * simple commit_write call, set_range_dirty is used to mark both
2679 * the pages and the extent records as dirty
2681 int extent_commit_write(struct extent_io_tree *tree,
2682 struct inode *inode, struct page *page,
2683 unsigned from, unsigned to)
2685 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2687 set_page_extent_mapped(page);
2688 set_page_dirty(page);
2690 if (pos > inode->i_size) {
2691 i_size_write(inode, pos);
2692 mark_inode_dirty(inode);
2697 int extent_prepare_write(struct extent_io_tree *tree,
2698 struct inode *inode, struct page *page,
2699 unsigned from, unsigned to, get_extent_t *get_extent)
2701 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2702 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2704 u64 orig_block_start;
2707 struct extent_map *em;
2708 unsigned blocksize = 1 << inode->i_blkbits;
2709 size_t page_offset = 0;
2710 size_t block_off_start;
2711 size_t block_off_end;
2717 set_page_extent_mapped(page);
2719 block_start = (page_start + from) & ~((u64)blocksize - 1);
2720 block_end = (page_start + to - 1) | (blocksize - 1);
2721 orig_block_start = block_start;
2723 lock_extent(tree, page_start, page_end, GFP_NOFS);
2724 while (block_start <= block_end) {
2725 em = get_extent(inode, page, page_offset, block_start,
2726 block_end - block_start + 1, 1);
2727 if (IS_ERR(em) || !em)
2730 cur_end = min(block_end, extent_map_end(em) - 1);
2731 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2732 block_off_end = block_off_start + blocksize;
2733 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2735 if (!PageUptodate(page) && isnew &&
2736 (block_off_end > to || block_off_start < from)) {
2739 kaddr = kmap_atomic(page, KM_USER0);
2740 if (block_off_end > to)
2741 memset(kaddr + to, 0, block_off_end - to);
2742 if (block_off_start < from)
2743 memset(kaddr + block_off_start, 0,
2744 from - block_off_start);
2745 flush_dcache_page(page);
2746 kunmap_atomic(kaddr, KM_USER0);
2748 if ((em->block_start != EXTENT_MAP_HOLE &&
2749 em->block_start != EXTENT_MAP_INLINE) &&
2750 !isnew && !PageUptodate(page) &&
2751 (block_off_end > to || block_off_start < from) &&
2752 !test_range_bit(tree, block_start, cur_end,
2753 EXTENT_UPTODATE, 1, NULL)) {
2755 u64 extent_offset = block_start - em->start;
2757 sector = (em->block_start + extent_offset) >> 9;
2758 iosize = (cur_end - block_start + blocksize) &
2759 ~((u64)blocksize - 1);
2761 * we've already got the extent locked, but we
2762 * need to split the state such that our end_bio
2763 * handler can clear the lock.
2765 set_extent_bit(tree, block_start,
2766 block_start + iosize - 1,
2767 EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS);
2768 ret = submit_extent_page(READ, tree, page,
2769 sector, iosize, page_offset, em->bdev,
2771 end_bio_extent_preparewrite, 0,
2776 block_start = block_start + iosize;
2778 set_extent_uptodate(tree, block_start, cur_end,
2780 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2781 block_start = cur_end + 1;
2783 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2784 free_extent_map(em);
2787 wait_extent_bit(tree, orig_block_start,
2788 block_end, EXTENT_LOCKED);
2790 check_page_uptodate(tree, page);
2792 /* FIXME, zero out newly allocated blocks on error */
2797 * a helper for releasepage, this tests for areas of the page that
2798 * are locked or under IO and drops the related state bits if it is safe
2801 int try_release_extent_state(struct extent_map_tree *map,
2802 struct extent_io_tree *tree, struct page *page,
2805 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2806 u64 end = start + PAGE_CACHE_SIZE - 1;
2809 if (test_range_bit(tree, start, end,
2810 EXTENT_IOBITS, 0, NULL))
2813 if ((mask & GFP_NOFS) == GFP_NOFS)
2816 * at this point we can safely clear everything except the
2817 * locked bit and the nodatasum bit
2819 clear_extent_bit(tree, start, end,
2820 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
2827 * a helper for releasepage. As long as there are no locked extents
2828 * in the range corresponding to the page, both state records and extent
2829 * map records are removed
2831 int try_release_extent_mapping(struct extent_map_tree *map,
2832 struct extent_io_tree *tree, struct page *page,
2835 struct extent_map *em;
2836 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2837 u64 end = start + PAGE_CACHE_SIZE - 1;
2839 if ((mask & __GFP_WAIT) &&
2840 page->mapping->host->i_size > 16 * 1024 * 1024) {
2842 while (start <= end) {
2843 len = end - start + 1;
2844 write_lock(&map->lock);
2845 em = lookup_extent_mapping(map, start, len);
2846 if (!em || IS_ERR(em)) {
2847 write_unlock(&map->lock);
2850 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2851 em->start != start) {
2852 write_unlock(&map->lock);
2853 free_extent_map(em);
2856 if (!test_range_bit(tree, em->start,
2857 extent_map_end(em) - 1,
2858 EXTENT_LOCKED | EXTENT_WRITEBACK,
2860 remove_extent_mapping(map, em);
2861 /* once for the rb tree */
2862 free_extent_map(em);
2864 start = extent_map_end(em);
2865 write_unlock(&map->lock);
2868 free_extent_map(em);
2871 return try_release_extent_state(map, tree, page, mask);
2874 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2875 get_extent_t *get_extent)
2877 struct inode *inode = mapping->host;
2878 struct extent_state *cached_state = NULL;
2879 u64 start = iblock << inode->i_blkbits;
2880 sector_t sector = 0;
2881 size_t blksize = (1 << inode->i_blkbits);
2882 struct extent_map *em;
2884 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2885 0, &cached_state, GFP_NOFS);
2886 em = get_extent(inode, NULL, 0, start, blksize, 0);
2887 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start,
2888 start + blksize - 1, &cached_state, GFP_NOFS);
2889 if (!em || IS_ERR(em))
2892 if (em->block_start > EXTENT_MAP_LAST_BYTE)
2895 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2897 free_extent_map(em);
2901 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2902 __u64 start, __u64 len, get_extent_t *get_extent)
2906 u64 max = start + len;
2911 struct btrfs_key found_key;
2912 struct extent_map *em = NULL;
2913 struct extent_state *cached_state = NULL;
2914 struct btrfs_path *path;
2915 struct btrfs_file_extent_item *item;
2917 u64 em_start = 0, em_len = 0;
2918 unsigned long emflags;
2924 path = btrfs_alloc_path();
2927 path->leave_spinning = 1;
2929 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
2930 path, inode->i_ino, -1, 0);
2932 btrfs_free_path(path);
2937 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2938 struct btrfs_file_extent_item);
2939 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
2940 found_type = btrfs_key_type(&found_key);
2942 /* No extents, just return */
2943 if (found_key.objectid != inode->i_ino ||
2944 found_type != BTRFS_EXTENT_DATA_KEY) {
2945 btrfs_free_path(path);
2948 last = found_key.offset;
2949 btrfs_free_path(path);
2951 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
2952 &cached_state, GFP_NOFS);
2953 em = get_extent(inode, NULL, 0, off, max - off, 0);
2963 off = em->start + em->len;
2967 if (em->block_start == EXTENT_MAP_HOLE) {
2972 em_start = em->start;
2978 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
2980 flags |= FIEMAP_EXTENT_LAST;
2981 } else if (em->block_start == EXTENT_MAP_INLINE) {
2982 flags |= (FIEMAP_EXTENT_DATA_INLINE |
2983 FIEMAP_EXTENT_NOT_ALIGNED);
2984 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
2985 flags |= (FIEMAP_EXTENT_DELALLOC |
2986 FIEMAP_EXTENT_UNKNOWN);
2988 disko = em->block_start;
2990 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2991 flags |= FIEMAP_EXTENT_ENCODED;
2994 emflags = em->flags;
2995 free_extent_map(em);
2998 em = get_extent(inode, NULL, 0, off, max - off, 0);
3005 emflags = em->flags;
3008 if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) {
3009 flags |= FIEMAP_EXTENT_LAST;
3013 if (em_start == last) {
3014 flags |= FIEMAP_EXTENT_LAST;
3019 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3026 free_extent_map(em);
3028 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3029 &cached_state, GFP_NOFS);
3033 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
3037 struct address_space *mapping;
3040 return eb->first_page;
3041 i += eb->start >> PAGE_CACHE_SHIFT;
3042 mapping = eb->first_page->mapping;
3047 * extent_buffer_page is only called after pinning the page
3048 * by increasing the reference count. So we know the page must
3049 * be in the radix tree.
3052 p = radix_tree_lookup(&mapping->page_tree, i);
3058 static inline unsigned long num_extent_pages(u64 start, u64 len)
3060 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3061 (start >> PAGE_CACHE_SHIFT);
3064 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3069 struct extent_buffer *eb = NULL;
3071 unsigned long flags;
3074 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3077 spin_lock_init(&eb->lock);
3078 init_waitqueue_head(&eb->lock_wq);
3081 spin_lock_irqsave(&leak_lock, flags);
3082 list_add(&eb->leak_list, &buffers);
3083 spin_unlock_irqrestore(&leak_lock, flags);
3085 atomic_set(&eb->refs, 1);
3090 static void __free_extent_buffer(struct extent_buffer *eb)
3093 unsigned long flags;
3094 spin_lock_irqsave(&leak_lock, flags);
3095 list_del(&eb->leak_list);
3096 spin_unlock_irqrestore(&leak_lock, flags);
3098 kmem_cache_free(extent_buffer_cache, eb);
3102 * Helper for releasing extent buffer page.
3104 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3105 unsigned long start_idx)
3107 unsigned long index;
3110 if (!eb->first_page)
3113 index = num_extent_pages(eb->start, eb->len);
3114 if (start_idx >= index)
3119 page = extent_buffer_page(eb, index);
3121 page_cache_release(page);
3122 } while (index != start_idx);
3126 * Helper for releasing the extent buffer.
3128 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3130 btrfs_release_extent_buffer_page(eb, 0);
3131 __free_extent_buffer(eb);
3134 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3135 u64 start, unsigned long len,
3139 unsigned long num_pages = num_extent_pages(start, len);
3141 unsigned long index = start >> PAGE_CACHE_SHIFT;
3142 struct extent_buffer *eb;
3143 struct extent_buffer *exists = NULL;
3145 struct address_space *mapping = tree->mapping;
3150 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3151 if (eb && atomic_inc_not_zero(&eb->refs)) {
3153 mark_page_accessed(eb->first_page);
3158 eb = __alloc_extent_buffer(tree, start, len, mask);
3163 eb->first_page = page0;
3166 page_cache_get(page0);
3167 mark_page_accessed(page0);
3168 set_page_extent_mapped(page0);
3169 set_page_extent_head(page0, len);
3170 uptodate = PageUptodate(page0);
3174 for (; i < num_pages; i++, index++) {
3175 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
3180 set_page_extent_mapped(p);
3181 mark_page_accessed(p);
3184 set_page_extent_head(p, len);
3186 set_page_private(p, EXTENT_PAGE_PRIVATE);
3188 if (!PageUptodate(p))
3193 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3195 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3199 spin_lock(&tree->buffer_lock);
3200 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3201 if (ret == -EEXIST) {
3202 exists = radix_tree_lookup(&tree->buffer,
3203 start >> PAGE_CACHE_SHIFT);
3204 /* add one reference for the caller */
3205 atomic_inc(&exists->refs);
3206 spin_unlock(&tree->buffer_lock);
3207 radix_tree_preload_end();
3210 /* add one reference for the tree */
3211 atomic_inc(&eb->refs);
3212 spin_unlock(&tree->buffer_lock);
3213 radix_tree_preload_end();
3217 if (!atomic_dec_and_test(&eb->refs))
3219 btrfs_release_extent_buffer(eb);
3223 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3224 u64 start, unsigned long len,
3227 struct extent_buffer *eb;
3230 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3231 if (eb && atomic_inc_not_zero(&eb->refs)) {
3233 mark_page_accessed(eb->first_page);
3241 void free_extent_buffer(struct extent_buffer *eb)
3246 if (!atomic_dec_and_test(&eb->refs))
3252 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3253 struct extent_buffer *eb)
3256 unsigned long num_pages;
3259 num_pages = num_extent_pages(eb->start, eb->len);
3261 for (i = 0; i < num_pages; i++) {
3262 page = extent_buffer_page(eb, i);
3263 if (!PageDirty(page))
3268 set_page_extent_head(page, eb->len);
3270 set_page_private(page, EXTENT_PAGE_PRIVATE);
3272 clear_page_dirty_for_io(page);
3273 spin_lock_irq(&page->mapping->tree_lock);
3274 if (!PageDirty(page)) {
3275 radix_tree_tag_clear(&page->mapping->page_tree,
3277 PAGECACHE_TAG_DIRTY);
3279 spin_unlock_irq(&page->mapping->tree_lock);
3285 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3286 struct extent_buffer *eb)
3288 return wait_on_extent_writeback(tree, eb->start,
3289 eb->start + eb->len - 1);
3292 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3293 struct extent_buffer *eb)
3296 unsigned long num_pages;
3299 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3300 num_pages = num_extent_pages(eb->start, eb->len);
3301 for (i = 0; i < num_pages; i++)
3302 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3306 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3307 struct extent_buffer *eb,
3308 struct extent_state **cached_state)
3312 unsigned long num_pages;
3314 num_pages = num_extent_pages(eb->start, eb->len);
3315 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3317 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3318 cached_state, GFP_NOFS);
3319 for (i = 0; i < num_pages; i++) {
3320 page = extent_buffer_page(eb, i);
3322 ClearPageUptodate(page);
3327 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3328 struct extent_buffer *eb)
3332 unsigned long num_pages;
3334 num_pages = num_extent_pages(eb->start, eb->len);
3336 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3338 for (i = 0; i < num_pages; i++) {
3339 page = extent_buffer_page(eb, i);
3340 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3341 ((i == num_pages - 1) &&
3342 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3343 check_page_uptodate(tree, page);
3346 SetPageUptodate(page);
3351 int extent_range_uptodate(struct extent_io_tree *tree,
3356 int pg_uptodate = 1;
3358 unsigned long index;
3360 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL);
3363 while (start <= end) {
3364 index = start >> PAGE_CACHE_SHIFT;
3365 page = find_get_page(tree->mapping, index);
3366 uptodate = PageUptodate(page);
3367 page_cache_release(page);
3372 start += PAGE_CACHE_SIZE;
3377 int extent_buffer_uptodate(struct extent_io_tree *tree,
3378 struct extent_buffer *eb,
3379 struct extent_state *cached_state)
3382 unsigned long num_pages;
3385 int pg_uptodate = 1;
3387 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3390 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3391 EXTENT_UPTODATE, 1, cached_state);
3395 num_pages = num_extent_pages(eb->start, eb->len);
3396 for (i = 0; i < num_pages; i++) {
3397 page = extent_buffer_page(eb, i);
3398 if (!PageUptodate(page)) {
3406 int read_extent_buffer_pages(struct extent_io_tree *tree,
3407 struct extent_buffer *eb,
3408 u64 start, int wait,
3409 get_extent_t *get_extent, int mirror_num)
3412 unsigned long start_i;
3416 int locked_pages = 0;
3417 int all_uptodate = 1;
3418 int inc_all_pages = 0;
3419 unsigned long num_pages;
3420 struct bio *bio = NULL;
3421 unsigned long bio_flags = 0;
3423 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3426 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3427 EXTENT_UPTODATE, 1, NULL)) {
3432 WARN_ON(start < eb->start);
3433 start_i = (start >> PAGE_CACHE_SHIFT) -
3434 (eb->start >> PAGE_CACHE_SHIFT);
3439 num_pages = num_extent_pages(eb->start, eb->len);
3440 for (i = start_i; i < num_pages; i++) {
3441 page = extent_buffer_page(eb, i);
3443 if (!trylock_page(page))
3449 if (!PageUptodate(page))
3454 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3458 for (i = start_i; i < num_pages; i++) {
3459 page = extent_buffer_page(eb, i);
3461 page_cache_get(page);
3462 if (!PageUptodate(page)) {
3465 ClearPageError(page);
3466 err = __extent_read_full_page(tree, page,
3468 mirror_num, &bio_flags);
3477 submit_one_bio(READ, bio, mirror_num, bio_flags);
3482 for (i = start_i; i < num_pages; i++) {
3483 page = extent_buffer_page(eb, i);
3484 wait_on_page_locked(page);
3485 if (!PageUptodate(page))
3490 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3495 while (locked_pages > 0) {
3496 page = extent_buffer_page(eb, i);
3504 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3505 unsigned long start,
3512 char *dst = (char *)dstv;
3513 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3514 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3516 WARN_ON(start > eb->len);
3517 WARN_ON(start + len > eb->start + eb->len);
3519 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3522 page = extent_buffer_page(eb, i);
3524 cur = min(len, (PAGE_CACHE_SIZE - offset));
3525 kaddr = kmap_atomic(page, KM_USER1);
3526 memcpy(dst, kaddr + offset, cur);
3527 kunmap_atomic(kaddr, KM_USER1);
3536 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3537 unsigned long min_len, char **token, char **map,
3538 unsigned long *map_start,
3539 unsigned long *map_len, int km)
3541 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3544 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3545 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3546 unsigned long end_i = (start_offset + start + min_len - 1) >>
3553 offset = start_offset;
3557 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3560 if (start + min_len > eb->len) {
3561 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3562 "wanted %lu %lu\n", (unsigned long long)eb->start,
3563 eb->len, start, min_len);
3567 p = extent_buffer_page(eb, i);
3568 kaddr = kmap_atomic(p, km);
3570 *map = kaddr + offset;
3571 *map_len = PAGE_CACHE_SIZE - offset;
3575 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3576 unsigned long min_len,
3577 char **token, char **map,
3578 unsigned long *map_start,
3579 unsigned long *map_len, int km)
3583 if (eb->map_token) {
3584 unmap_extent_buffer(eb, eb->map_token, km);
3585 eb->map_token = NULL;
3588 err = map_private_extent_buffer(eb, start, min_len, token, map,
3589 map_start, map_len, km);
3591 eb->map_token = *token;
3593 eb->map_start = *map_start;
3594 eb->map_len = *map_len;
3599 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3601 kunmap_atomic(token, km);
3604 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3605 unsigned long start,
3612 char *ptr = (char *)ptrv;
3613 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3614 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3617 WARN_ON(start > eb->len);
3618 WARN_ON(start + len > eb->start + eb->len);
3620 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3623 page = extent_buffer_page(eb, i);
3625 cur = min(len, (PAGE_CACHE_SIZE - offset));
3627 kaddr = kmap_atomic(page, KM_USER0);
3628 ret = memcmp(ptr, kaddr + offset, cur);
3629 kunmap_atomic(kaddr, KM_USER0);
3641 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3642 unsigned long start, unsigned long len)
3648 char *src = (char *)srcv;
3649 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3650 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3652 WARN_ON(start > eb->len);
3653 WARN_ON(start + len > eb->start + eb->len);
3655 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3658 page = extent_buffer_page(eb, i);
3659 WARN_ON(!PageUptodate(page));
3661 cur = min(len, PAGE_CACHE_SIZE - offset);
3662 kaddr = kmap_atomic(page, KM_USER1);
3663 memcpy(kaddr + offset, src, cur);
3664 kunmap_atomic(kaddr, KM_USER1);
3673 void memset_extent_buffer(struct extent_buffer *eb, char c,
3674 unsigned long start, unsigned long len)
3680 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3681 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3683 WARN_ON(start > eb->len);
3684 WARN_ON(start + len > eb->start + eb->len);
3686 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3689 page = extent_buffer_page(eb, i);
3690 WARN_ON(!PageUptodate(page));
3692 cur = min(len, PAGE_CACHE_SIZE - offset);
3693 kaddr = kmap_atomic(page, KM_USER0);
3694 memset(kaddr + offset, c, cur);
3695 kunmap_atomic(kaddr, KM_USER0);
3703 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3704 unsigned long dst_offset, unsigned long src_offset,
3707 u64 dst_len = dst->len;
3712 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3713 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3715 WARN_ON(src->len != dst_len);
3717 offset = (start_offset + dst_offset) &
3718 ((unsigned long)PAGE_CACHE_SIZE - 1);
3721 page = extent_buffer_page(dst, i);
3722 WARN_ON(!PageUptodate(page));
3724 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3726 kaddr = kmap_atomic(page, KM_USER0);
3727 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3728 kunmap_atomic(kaddr, KM_USER0);
3737 static void move_pages(struct page *dst_page, struct page *src_page,
3738 unsigned long dst_off, unsigned long src_off,
3741 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3742 if (dst_page == src_page) {
3743 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3745 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3746 char *p = dst_kaddr + dst_off + len;
3747 char *s = src_kaddr + src_off + len;
3752 kunmap_atomic(src_kaddr, KM_USER1);
3754 kunmap_atomic(dst_kaddr, KM_USER0);
3757 static void copy_pages(struct page *dst_page, struct page *src_page,
3758 unsigned long dst_off, unsigned long src_off,
3761 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3764 if (dst_page != src_page)
3765 src_kaddr = kmap_atomic(src_page, KM_USER1);
3767 src_kaddr = dst_kaddr;
3769 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3770 kunmap_atomic(dst_kaddr, KM_USER0);
3771 if (dst_page != src_page)
3772 kunmap_atomic(src_kaddr, KM_USER1);
3775 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3776 unsigned long src_offset, unsigned long len)
3779 size_t dst_off_in_page;
3780 size_t src_off_in_page;
3781 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3782 unsigned long dst_i;
3783 unsigned long src_i;
3785 if (src_offset + len > dst->len) {
3786 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3787 "len %lu dst len %lu\n", src_offset, len, dst->len);
3790 if (dst_offset + len > dst->len) {
3791 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3792 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3797 dst_off_in_page = (start_offset + dst_offset) &
3798 ((unsigned long)PAGE_CACHE_SIZE - 1);
3799 src_off_in_page = (start_offset + src_offset) &
3800 ((unsigned long)PAGE_CACHE_SIZE - 1);
3802 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3803 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3805 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3807 cur = min_t(unsigned long, cur,
3808 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3810 copy_pages(extent_buffer_page(dst, dst_i),
3811 extent_buffer_page(dst, src_i),
3812 dst_off_in_page, src_off_in_page, cur);
3820 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3821 unsigned long src_offset, unsigned long len)
3824 size_t dst_off_in_page;
3825 size_t src_off_in_page;
3826 unsigned long dst_end = dst_offset + len - 1;
3827 unsigned long src_end = src_offset + len - 1;
3828 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3829 unsigned long dst_i;
3830 unsigned long src_i;
3832 if (src_offset + len > dst->len) {
3833 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3834 "len %lu len %lu\n", src_offset, len, dst->len);
3837 if (dst_offset + len > dst->len) {
3838 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3839 "len %lu len %lu\n", dst_offset, len, dst->len);
3842 if (dst_offset < src_offset) {
3843 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3847 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3848 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3850 dst_off_in_page = (start_offset + dst_end) &
3851 ((unsigned long)PAGE_CACHE_SIZE - 1);
3852 src_off_in_page = (start_offset + src_end) &
3853 ((unsigned long)PAGE_CACHE_SIZE - 1);
3855 cur = min_t(unsigned long, len, src_off_in_page + 1);
3856 cur = min(cur, dst_off_in_page + 1);
3857 move_pages(extent_buffer_page(dst, dst_i),
3858 extent_buffer_page(dst, src_i),
3859 dst_off_in_page - cur + 1,
3860 src_off_in_page - cur + 1, cur);
3868 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3870 struct extent_buffer *eb =
3871 container_of(head, struct extent_buffer, rcu_head);
3873 btrfs_release_extent_buffer(eb);
3876 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3878 u64 start = page_offset(page);
3879 struct extent_buffer *eb;
3882 spin_lock(&tree->buffer_lock);
3883 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3885 spin_unlock(&tree->buffer_lock);
3889 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3895 * set @eb->refs to 0 if it is already 1, and then release the @eb.
3898 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
3903 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3905 spin_unlock(&tree->buffer_lock);
3907 /* at this point we can safely release the extent buffer */
3908 if (atomic_read(&eb->refs) == 0)
3909 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);