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/version.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 /* temporary define until extent_map moves out of btrfs */
22 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
23 unsigned long extra_flags,
24 void (*ctor)(void *, struct kmem_cache *,
27 static struct kmem_cache *extent_state_cache;
28 static struct kmem_cache *extent_buffer_cache;
30 static LIST_HEAD(buffers);
31 static LIST_HEAD(states);
35 static spinlock_t leak_lock = SPIN_LOCK_UNLOCKED;
38 #define BUFFER_LRU_MAX 64
43 struct rb_node rb_node;
46 struct extent_page_data {
48 struct extent_io_tree *tree;
49 get_extent_t *get_extent;
52 int __init extent_io_init(void)
54 extent_state_cache = btrfs_cache_create("extent_state",
55 sizeof(struct extent_state), 0,
57 if (!extent_state_cache)
60 extent_buffer_cache = btrfs_cache_create("extent_buffers",
61 sizeof(struct extent_buffer), 0,
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("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state->start, state->end, state->state, state->tree, atomic_read(&state->refs));
80 list_del(&state->leak_list);
81 kmem_cache_free(extent_state_cache, state);
85 while (!list_empty(&buffers)) {
86 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
87 printk("buffer leak start %Lu len %lu refs %d\n", eb->start, eb->len, atomic_read(&eb->refs));
88 list_del(&eb->leak_list);
89 kmem_cache_free(extent_buffer_cache, eb);
91 if (extent_state_cache)
92 kmem_cache_destroy(extent_state_cache);
93 if (extent_buffer_cache)
94 kmem_cache_destroy(extent_buffer_cache);
97 void extent_io_tree_init(struct extent_io_tree *tree,
98 struct address_space *mapping, gfp_t mask)
100 tree->state.rb_node = NULL;
101 tree->buffer.rb_node = NULL;
103 tree->dirty_bytes = 0;
104 spin_lock_init(&tree->lock);
105 spin_lock_init(&tree->buffer_lock);
106 tree->mapping = mapping;
108 EXPORT_SYMBOL(extent_io_tree_init);
110 struct extent_state *alloc_extent_state(gfp_t mask)
112 struct extent_state *state;
117 state = kmem_cache_alloc(extent_state_cache, mask);
124 spin_lock_irqsave(&leak_lock, flags);
125 list_add(&state->leak_list, &states);
126 spin_unlock_irqrestore(&leak_lock, flags);
128 atomic_set(&state->refs, 1);
129 init_waitqueue_head(&state->wq);
132 EXPORT_SYMBOL(alloc_extent_state);
134 void free_extent_state(struct extent_state *state)
138 if (atomic_dec_and_test(&state->refs)) {
142 WARN_ON(state->tree);
144 spin_lock_irqsave(&leak_lock, flags);
145 list_del(&state->leak_list);
146 spin_unlock_irqrestore(&leak_lock, flags);
148 kmem_cache_free(extent_state_cache, state);
151 EXPORT_SYMBOL(free_extent_state);
153 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
154 struct rb_node *node)
156 struct rb_node ** p = &root->rb_node;
157 struct rb_node * parent = NULL;
158 struct tree_entry *entry;
162 entry = rb_entry(parent, struct tree_entry, rb_node);
164 if (offset < entry->start)
166 else if (offset > entry->end)
172 entry = rb_entry(node, struct tree_entry, rb_node);
173 rb_link_node(node, parent, p);
174 rb_insert_color(node, root);
178 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
179 struct rb_node **prev_ret,
180 struct rb_node **next_ret)
182 struct rb_root *root = &tree->state;
183 struct rb_node * n = root->rb_node;
184 struct rb_node *prev = NULL;
185 struct rb_node *orig_prev = NULL;
186 struct tree_entry *entry;
187 struct tree_entry *prev_entry = NULL;
190 entry = rb_entry(n, struct tree_entry, rb_node);
194 if (offset < entry->start)
196 else if (offset > entry->end)
205 while(prev && offset > prev_entry->end) {
206 prev = rb_next(prev);
207 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
214 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
215 while(prev && offset < prev_entry->start) {
216 prev = rb_prev(prev);
217 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
224 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
227 struct rb_node *prev = NULL;
230 ret = __etree_search(tree, offset, &prev, NULL);
237 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
238 u64 offset, struct rb_node *node)
240 struct rb_root *root = &tree->buffer;
241 struct rb_node ** p = &root->rb_node;
242 struct rb_node * parent = NULL;
243 struct extent_buffer *eb;
247 eb = rb_entry(parent, struct extent_buffer, rb_node);
249 if (offset < eb->start)
251 else if (offset > eb->start)
257 rb_link_node(node, parent, p);
258 rb_insert_color(node, root);
262 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
265 struct rb_root *root = &tree->buffer;
266 struct rb_node * n = root->rb_node;
267 struct extent_buffer *eb;
270 eb = rb_entry(n, struct extent_buffer, rb_node);
271 if (offset < eb->start)
273 else if (offset > eb->start)
282 * utility function to look for merge candidates inside a given range.
283 * Any extents with matching state are merged together into a single
284 * extent in the tree. Extents with EXTENT_IO in their state field
285 * are not merged because the end_io handlers need to be able to do
286 * operations on them without sleeping (or doing allocations/splits).
288 * This should be called with the tree lock held.
290 static int merge_state(struct extent_io_tree *tree,
291 struct extent_state *state)
293 struct extent_state *other;
294 struct rb_node *other_node;
296 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
299 other_node = rb_prev(&state->rb_node);
301 other = rb_entry(other_node, struct extent_state, rb_node);
302 if (other->end == state->start - 1 &&
303 other->state == state->state) {
304 state->start = other->start;
306 rb_erase(&other->rb_node, &tree->state);
307 free_extent_state(other);
310 other_node = rb_next(&state->rb_node);
312 other = rb_entry(other_node, struct extent_state, rb_node);
313 if (other->start == state->end + 1 &&
314 other->state == state->state) {
315 other->start = state->start;
317 rb_erase(&state->rb_node, &tree->state);
318 free_extent_state(state);
324 static void set_state_cb(struct extent_io_tree *tree,
325 struct extent_state *state,
328 if (tree->ops && tree->ops->set_bit_hook) {
329 tree->ops->set_bit_hook(tree->mapping->host, state->start,
330 state->end, state->state, bits);
334 static void clear_state_cb(struct extent_io_tree *tree,
335 struct extent_state *state,
338 if (tree->ops && tree->ops->set_bit_hook) {
339 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
340 state->end, state->state, bits);
345 * insert an extent_state struct into the tree. 'bits' are set on the
346 * struct before it is inserted.
348 * This may return -EEXIST if the extent is already there, in which case the
349 * state struct is freed.
351 * The tree lock is not taken internally. This is a utility function and
352 * probably isn't what you want to call (see set/clear_extent_bit).
354 static int insert_state(struct extent_io_tree *tree,
355 struct extent_state *state, u64 start, u64 end,
358 struct rb_node *node;
361 printk("end < start %Lu %Lu\n", end, start);
364 if (bits & EXTENT_DIRTY)
365 tree->dirty_bytes += end - start + 1;
366 set_state_cb(tree, state, bits);
367 state->state |= bits;
368 state->start = start;
370 node = tree_insert(&tree->state, end, &state->rb_node);
372 struct extent_state *found;
373 found = rb_entry(node, struct extent_state, rb_node);
374 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
375 free_extent_state(state);
379 merge_state(tree, state);
384 * split a given extent state struct in two, inserting the preallocated
385 * struct 'prealloc' as the newly created second half. 'split' indicates an
386 * offset inside 'orig' where it should be split.
389 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
390 * are two extent state structs in the tree:
391 * prealloc: [orig->start, split - 1]
392 * orig: [ split, orig->end ]
394 * The tree locks are not taken by this function. They need to be held
397 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
398 struct extent_state *prealloc, u64 split)
400 struct rb_node *node;
401 prealloc->start = orig->start;
402 prealloc->end = split - 1;
403 prealloc->state = orig->state;
406 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
408 struct extent_state *found;
409 found = rb_entry(node, struct extent_state, rb_node);
410 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
411 free_extent_state(prealloc);
414 prealloc->tree = tree;
419 * utility function to clear some bits in an extent state struct.
420 * it will optionally wake up any one waiting on this state (wake == 1), or
421 * forcibly remove the state from the tree (delete == 1).
423 * If no bits are set on the state struct after clearing things, the
424 * struct is freed and removed from the tree
426 static int clear_state_bit(struct extent_io_tree *tree,
427 struct extent_state *state, int bits, int wake,
430 int ret = state->state & bits;
432 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
433 u64 range = state->end - state->start + 1;
434 WARN_ON(range > tree->dirty_bytes);
435 tree->dirty_bytes -= range;
437 clear_state_cb(tree, state, bits);
438 state->state &= ~bits;
441 if (delete || state->state == 0) {
443 clear_state_cb(tree, state, state->state);
444 rb_erase(&state->rb_node, &tree->state);
446 free_extent_state(state);
451 merge_state(tree, state);
457 * clear some bits on a range in the tree. This may require splitting
458 * or inserting elements in the tree, so the gfp mask is used to
459 * indicate which allocations or sleeping are allowed.
461 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
462 * the given range from the tree regardless of state (ie for truncate).
464 * the range [start, end] is inclusive.
466 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
467 * bits were already set, or zero if none of the bits were already set.
469 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
470 int bits, int wake, int delete, gfp_t mask)
472 struct extent_state *state;
473 struct extent_state *prealloc = NULL;
474 struct rb_node *node;
480 if (!prealloc && (mask & __GFP_WAIT)) {
481 prealloc = alloc_extent_state(mask);
486 spin_lock_irqsave(&tree->lock, flags);
488 * this search will find the extents that end after
491 node = tree_search(tree, start);
494 state = rb_entry(node, struct extent_state, rb_node);
495 if (state->start > end)
497 WARN_ON(state->end < start);
500 * | ---- desired range ---- |
502 * | ------------- state -------------- |
504 * We need to split the extent we found, and may flip
505 * bits on second half.
507 * If the extent we found extends past our range, we
508 * just split and search again. It'll get split again
509 * the next time though.
511 * If the extent we found is inside our range, we clear
512 * the desired bit on it.
515 if (state->start < start) {
517 prealloc = alloc_extent_state(GFP_ATOMIC);
518 err = split_state(tree, state, prealloc, start);
519 BUG_ON(err == -EEXIST);
523 if (state->end <= end) {
524 start = state->end + 1;
525 set |= clear_state_bit(tree, state, bits,
528 start = state->start;
533 * | ---- desired range ---- |
535 * We need to split the extent, and clear the bit
538 if (state->start <= end && state->end > end) {
540 prealloc = alloc_extent_state(GFP_ATOMIC);
541 err = split_state(tree, state, prealloc, end + 1);
542 BUG_ON(err == -EEXIST);
546 set |= clear_state_bit(tree, prealloc, bits,
552 start = state->end + 1;
553 set |= clear_state_bit(tree, state, bits, wake, delete);
557 spin_unlock_irqrestore(&tree->lock, flags);
559 free_extent_state(prealloc);
566 spin_unlock_irqrestore(&tree->lock, flags);
567 if (mask & __GFP_WAIT)
571 EXPORT_SYMBOL(clear_extent_bit);
573 static int wait_on_state(struct extent_io_tree *tree,
574 struct extent_state *state)
577 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
578 spin_unlock_irq(&tree->lock);
580 spin_lock_irq(&tree->lock);
581 finish_wait(&state->wq, &wait);
586 * waits for one or more bits to clear on a range in the state tree.
587 * The range [start, end] is inclusive.
588 * The tree lock is taken by this function
590 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
592 struct extent_state *state;
593 struct rb_node *node;
595 spin_lock_irq(&tree->lock);
599 * this search will find all the extents that end after
602 node = tree_search(tree, start);
606 state = rb_entry(node, struct extent_state, rb_node);
608 if (state->start > end)
611 if (state->state & bits) {
612 start = state->start;
613 atomic_inc(&state->refs);
614 wait_on_state(tree, state);
615 free_extent_state(state);
618 start = state->end + 1;
623 if (need_resched()) {
624 spin_unlock_irq(&tree->lock);
626 spin_lock_irq(&tree->lock);
630 spin_unlock_irq(&tree->lock);
633 EXPORT_SYMBOL(wait_extent_bit);
635 static void set_state_bits(struct extent_io_tree *tree,
636 struct extent_state *state,
639 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
640 u64 range = state->end - state->start + 1;
641 tree->dirty_bytes += range;
643 set_state_cb(tree, state, bits);
644 state->state |= bits;
648 * set some bits on a range in the tree. This may require allocations
649 * or sleeping, so the gfp mask is used to indicate what is allowed.
651 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
652 * range already has the desired bits set. The start of the existing
653 * range is returned in failed_start in this case.
655 * [start, end] is inclusive
656 * This takes the tree lock.
658 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
659 int exclusive, u64 *failed_start, gfp_t mask)
661 struct extent_state *state;
662 struct extent_state *prealloc = NULL;
663 struct rb_node *node;
670 if (!prealloc && (mask & __GFP_WAIT)) {
671 prealloc = alloc_extent_state(mask);
676 spin_lock_irqsave(&tree->lock, flags);
678 * this search will find all the extents that end after
681 node = tree_search(tree, start);
683 err = insert_state(tree, prealloc, start, end, bits);
685 BUG_ON(err == -EEXIST);
689 state = rb_entry(node, struct extent_state, rb_node);
690 last_start = state->start;
691 last_end = state->end;
694 * | ---- desired range ---- |
697 * Just lock what we found and keep going
699 if (state->start == start && state->end <= end) {
700 set = state->state & bits;
701 if (set && exclusive) {
702 *failed_start = state->start;
706 set_state_bits(tree, state, bits);
707 start = state->end + 1;
708 merge_state(tree, state);
713 * | ---- desired range ---- |
716 * | ------------- state -------------- |
718 * We need to split the extent we found, and may flip bits on
721 * If the extent we found extends past our
722 * range, we just split and search again. It'll get split
723 * again the next time though.
725 * If the extent we found is inside our range, we set the
728 if (state->start < start) {
729 set = state->state & bits;
730 if (exclusive && set) {
731 *failed_start = start;
735 err = split_state(tree, state, prealloc, start);
736 BUG_ON(err == -EEXIST);
740 if (state->end <= end) {
741 set_state_bits(tree, state, bits);
742 start = state->end + 1;
743 merge_state(tree, state);
745 start = state->start;
750 * | ---- desired range ---- |
751 * | state | or | state |
753 * There's a hole, we need to insert something in it and
754 * ignore the extent we found.
756 if (state->start > start) {
758 if (end < last_start)
761 this_end = last_start -1;
762 err = insert_state(tree, prealloc, start, this_end,
765 BUG_ON(err == -EEXIST);
768 start = this_end + 1;
772 * | ---- desired range ---- |
774 * We need to split the extent, and set the bit
777 if (state->start <= end && state->end > end) {
778 set = state->state & bits;
779 if (exclusive && set) {
780 *failed_start = start;
784 err = split_state(tree, state, prealloc, end + 1);
785 BUG_ON(err == -EEXIST);
787 set_state_bits(tree, prealloc, bits);
788 merge_state(tree, prealloc);
796 spin_unlock_irqrestore(&tree->lock, flags);
798 free_extent_state(prealloc);
805 spin_unlock_irqrestore(&tree->lock, flags);
806 if (mask & __GFP_WAIT)
810 EXPORT_SYMBOL(set_extent_bit);
812 /* wrappers around set/clear extent bit */
813 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
816 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
819 EXPORT_SYMBOL(set_extent_dirty);
821 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
824 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
826 EXPORT_SYMBOL(set_extent_ordered);
828 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
829 int bits, gfp_t mask)
831 return set_extent_bit(tree, start, end, bits, 0, NULL,
834 EXPORT_SYMBOL(set_extent_bits);
836 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
837 int bits, gfp_t mask)
839 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
841 EXPORT_SYMBOL(clear_extent_bits);
843 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
846 return set_extent_bit(tree, start, end,
847 EXTENT_DELALLOC | EXTENT_DIRTY,
850 EXPORT_SYMBOL(set_extent_delalloc);
852 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
855 return clear_extent_bit(tree, start, end,
856 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
858 EXPORT_SYMBOL(clear_extent_dirty);
860 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
863 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
865 EXPORT_SYMBOL(clear_extent_ordered);
867 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
870 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
873 EXPORT_SYMBOL(set_extent_new);
875 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
878 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
880 EXPORT_SYMBOL(clear_extent_new);
882 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
885 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
888 EXPORT_SYMBOL(set_extent_uptodate);
890 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
893 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
895 EXPORT_SYMBOL(clear_extent_uptodate);
897 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
900 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
903 EXPORT_SYMBOL(set_extent_writeback);
905 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
908 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
910 EXPORT_SYMBOL(clear_extent_writeback);
912 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
914 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
916 EXPORT_SYMBOL(wait_on_extent_writeback);
919 * either insert or lock state struct between start and end use mask to tell
920 * us if waiting is desired.
922 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
927 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
928 &failed_start, mask);
929 if (err == -EEXIST && (mask & __GFP_WAIT)) {
930 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
931 start = failed_start;
935 WARN_ON(start > end);
939 EXPORT_SYMBOL(lock_extent);
941 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
944 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
946 EXPORT_SYMBOL(unlock_extent);
949 * helper function to set pages and extents in the tree dirty
951 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
953 unsigned long index = start >> PAGE_CACHE_SHIFT;
954 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
957 while (index <= end_index) {
958 page = find_get_page(tree->mapping, index);
960 __set_page_dirty_nobuffers(page);
961 page_cache_release(page);
964 set_extent_dirty(tree, start, end, GFP_NOFS);
967 EXPORT_SYMBOL(set_range_dirty);
970 * helper function to set both pages and extents in the tree writeback
972 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
974 unsigned long index = start >> PAGE_CACHE_SHIFT;
975 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
978 while (index <= end_index) {
979 page = find_get_page(tree->mapping, index);
981 set_page_writeback(page);
982 page_cache_release(page);
985 set_extent_writeback(tree, start, end, GFP_NOFS);
988 EXPORT_SYMBOL(set_range_writeback);
991 * find the first offset in the io tree with 'bits' set. zero is
992 * returned if we find something, and *start_ret and *end_ret are
993 * set to reflect the state struct that was found.
995 * If nothing was found, 1 is returned, < 0 on error
997 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
998 u64 *start_ret, u64 *end_ret, int bits)
1000 struct rb_node *node;
1001 struct extent_state *state;
1004 spin_lock_irq(&tree->lock);
1006 * this search will find all the extents that end after
1009 node = tree_search(tree, start);
1015 state = rb_entry(node, struct extent_state, rb_node);
1016 if (state->end >= start && (state->state & bits)) {
1017 *start_ret = state->start;
1018 *end_ret = state->end;
1022 node = rb_next(node);
1027 spin_unlock_irq(&tree->lock);
1030 EXPORT_SYMBOL(find_first_extent_bit);
1032 /* find the first state struct with 'bits' set after 'start', and
1033 * return it. tree->lock must be held. NULL will returned if
1034 * nothing was found after 'start'
1036 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1037 u64 start, int bits)
1039 struct rb_node *node;
1040 struct extent_state *state;
1043 * this search will find all the extents that end after
1046 node = tree_search(tree, start);
1052 state = rb_entry(node, struct extent_state, rb_node);
1053 if (state->end >= start && (state->state & bits)) {
1056 node = rb_next(node);
1063 EXPORT_SYMBOL(find_first_extent_bit_state);
1066 * find a contiguous range of bytes in the file marked as delalloc, not
1067 * more than 'max_bytes'. start and end are used to return the range,
1069 * 1 is returned if we find something, 0 if nothing was in the tree
1071 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1072 u64 *start, u64 *end, u64 max_bytes)
1074 struct rb_node *node;
1075 struct extent_state *state;
1076 u64 cur_start = *start;
1078 u64 total_bytes = 0;
1080 spin_lock_irq(&tree->lock);
1083 * this search will find all the extents that end after
1086 node = tree_search(tree, cur_start);
1094 state = rb_entry(node, struct extent_state, rb_node);
1095 if (found && (state->start != cur_start ||
1096 (state->state & EXTENT_BOUNDARY))) {
1099 if (!(state->state & EXTENT_DELALLOC)) {
1105 *start = state->start;
1108 cur_start = state->end + 1;
1109 node = rb_next(node);
1112 total_bytes += state->end - state->start + 1;
1113 if (total_bytes >= max_bytes)
1117 spin_unlock_irq(&tree->lock);
1121 static noinline int __unlock_for_delalloc(struct inode *inode,
1122 struct page *locked_page,
1126 struct page *pages[16];
1127 unsigned long index = start >> PAGE_CACHE_SHIFT;
1128 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1129 unsigned long nr_pages = end_index - index + 1;
1132 if (index == locked_page->index && end_index == index)
1135 while(nr_pages > 0) {
1136 ret = find_get_pages_contig(inode->i_mapping, index,
1137 min(nr_pages, ARRAY_SIZE(pages)), pages);
1138 for (i = 0; i < ret; i++) {
1139 if (pages[i] != locked_page)
1140 unlock_page(pages[i]);
1141 page_cache_release(pages[i]);
1150 static noinline int lock_delalloc_pages(struct inode *inode,
1151 struct page *locked_page,
1155 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1156 unsigned long start_index = index;
1157 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1158 unsigned long pages_locked = 0;
1159 struct page *pages[16];
1160 unsigned long nrpages;
1164 /* the caller is responsible for locking the start index */
1165 if (index == locked_page->index && index == end_index)
1168 /* skip the page at the start index */
1169 nrpages = end_index - index + 1;
1170 while(nrpages > 0) {
1171 ret = find_get_pages_contig(inode->i_mapping, index,
1172 min(nrpages, ARRAY_SIZE(pages)), pages);
1177 /* now we have an array of pages, lock them all */
1178 for (i = 0; i < ret; i++) {
1180 * the caller is taking responsibility for
1183 if (pages[i] != locked_page)
1184 lock_page(pages[i]);
1185 page_cache_release(pages[i]);
1187 pages_locked += ret;
1194 if (ret && pages_locked) {
1195 __unlock_for_delalloc(inode, locked_page,
1197 ((u64)(start_index + pages_locked - 1)) <<
1204 * find a contiguous range of bytes in the file marked as delalloc, not
1205 * more than 'max_bytes'. start and end are used to return the range,
1207 * 1 is returned if we find something, 0 if nothing was in the tree
1209 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1210 struct extent_io_tree *tree,
1211 struct page *locked_page,
1212 u64 *start, u64 *end,
1222 /* step one, find a bunch of delalloc bytes starting at start */
1223 delalloc_start = *start;
1225 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1228 *start = delalloc_start;
1229 *end = delalloc_end;
1234 * make sure to limit the number of pages we try to lock down
1237 if (delalloc_end + 1 - delalloc_start > max_bytes && loops) {
1238 delalloc_end = (delalloc_start + PAGE_CACHE_SIZE - 1) &
1239 ~((u64)PAGE_CACHE_SIZE - 1);
1241 /* step two, lock all the pages after the page that has start */
1242 ret = lock_delalloc_pages(inode, locked_page,
1243 delalloc_start, delalloc_end);
1244 if (ret == -EAGAIN) {
1245 /* some of the pages are gone, lets avoid looping by
1246 * shortening the size of the delalloc range we're searching
1249 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1250 max_bytes = PAGE_CACHE_SIZE - offset;
1260 /* step three, lock the state bits for the whole range */
1261 lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1263 /* then test to make sure it is all still delalloc */
1264 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1265 EXTENT_DELALLOC, 1);
1267 unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1268 __unlock_for_delalloc(inode, locked_page,
1269 delalloc_start, delalloc_end);
1273 *start = delalloc_start;
1274 *end = delalloc_end;
1279 int extent_clear_unlock_delalloc(struct inode *inode,
1280 struct extent_io_tree *tree,
1281 u64 start, u64 end, struct page *locked_page,
1282 int clear_dirty, int set_writeback,
1286 struct page *pages[16];
1287 unsigned long index = start >> PAGE_CACHE_SHIFT;
1288 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1289 unsigned long nr_pages = end_index - index + 1;
1291 int clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC;
1294 clear_bits |= EXTENT_DIRTY;
1296 clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
1298 while(nr_pages > 0) {
1299 ret = find_get_pages_contig(inode->i_mapping, index,
1300 min(nr_pages, ARRAY_SIZE(pages)), pages);
1301 for (i = 0; i < ret; i++) {
1302 if (pages[i] == locked_page) {
1303 page_cache_release(pages[i]);
1307 clear_page_dirty_for_io(pages[i]);
1309 set_page_writeback(pages[i]);
1311 end_page_writeback(pages[i]);
1312 unlock_page(pages[i]);
1313 page_cache_release(pages[i]);
1321 EXPORT_SYMBOL(extent_clear_unlock_delalloc);
1324 * count the number of bytes in the tree that have a given bit(s)
1325 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1326 * cached. The total number found is returned.
1328 u64 count_range_bits(struct extent_io_tree *tree,
1329 u64 *start, u64 search_end, u64 max_bytes,
1332 struct rb_node *node;
1333 struct extent_state *state;
1334 u64 cur_start = *start;
1335 u64 total_bytes = 0;
1338 if (search_end <= cur_start) {
1339 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1344 spin_lock_irq(&tree->lock);
1345 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1346 total_bytes = tree->dirty_bytes;
1350 * this search will find all the extents that end after
1353 node = tree_search(tree, cur_start);
1359 state = rb_entry(node, struct extent_state, rb_node);
1360 if (state->start > search_end)
1362 if (state->end >= cur_start && (state->state & bits)) {
1363 total_bytes += min(search_end, state->end) + 1 -
1364 max(cur_start, state->start);
1365 if (total_bytes >= max_bytes)
1368 *start = state->start;
1372 node = rb_next(node);
1377 spin_unlock_irq(&tree->lock);
1381 * helper function to lock both pages and extents in the tree.
1382 * pages must be locked first.
1384 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1386 unsigned long index = start >> PAGE_CACHE_SHIFT;
1387 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1391 while (index <= end_index) {
1392 page = grab_cache_page(tree->mapping, index);
1398 err = PTR_ERR(page);
1403 lock_extent(tree, start, end, GFP_NOFS);
1408 * we failed above in getting the page at 'index', so we undo here
1409 * up to but not including the page at 'index'
1412 index = start >> PAGE_CACHE_SHIFT;
1413 while (index < end_index) {
1414 page = find_get_page(tree->mapping, index);
1416 page_cache_release(page);
1421 EXPORT_SYMBOL(lock_range);
1424 * helper function to unlock both pages and extents in the tree.
1426 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1428 unsigned long index = start >> PAGE_CACHE_SHIFT;
1429 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1432 while (index <= end_index) {
1433 page = find_get_page(tree->mapping, index);
1435 page_cache_release(page);
1438 unlock_extent(tree, start, end, GFP_NOFS);
1441 EXPORT_SYMBOL(unlock_range);
1444 * set the private field for a given byte offset in the tree. If there isn't
1445 * an extent_state there already, this does nothing.
1447 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1449 struct rb_node *node;
1450 struct extent_state *state;
1453 spin_lock_irq(&tree->lock);
1455 * this search will find all the extents that end after
1458 node = tree_search(tree, start);
1463 state = rb_entry(node, struct extent_state, rb_node);
1464 if (state->start != start) {
1468 state->private = private;
1470 spin_unlock_irq(&tree->lock);
1474 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1476 struct rb_node *node;
1477 struct extent_state *state;
1480 spin_lock_irq(&tree->lock);
1482 * this search will find all the extents that end after
1485 node = tree_search(tree, start);
1490 state = rb_entry(node, struct extent_state, rb_node);
1491 if (state->start != start) {
1495 *private = state->private;
1497 spin_unlock_irq(&tree->lock);
1502 * searches a range in the state tree for a given mask.
1503 * If 'filled' == 1, this returns 1 only if every extent in the tree
1504 * has the bits set. Otherwise, 1 is returned if any bit in the
1505 * range is found set.
1507 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1508 int bits, int filled)
1510 struct extent_state *state = NULL;
1511 struct rb_node *node;
1513 unsigned long flags;
1515 spin_lock_irqsave(&tree->lock, flags);
1516 node = tree_search(tree, start);
1517 while (node && start <= end) {
1518 state = rb_entry(node, struct extent_state, rb_node);
1520 if (filled && state->start > start) {
1525 if (state->start > end)
1528 if (state->state & bits) {
1532 } else if (filled) {
1536 start = state->end + 1;
1539 node = rb_next(node);
1546 spin_unlock_irqrestore(&tree->lock, flags);
1549 EXPORT_SYMBOL(test_range_bit);
1552 * helper function to set a given page up to date if all the
1553 * extents in the tree for that page are up to date
1555 static int check_page_uptodate(struct extent_io_tree *tree,
1558 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1559 u64 end = start + PAGE_CACHE_SIZE - 1;
1560 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1561 SetPageUptodate(page);
1566 * helper function to unlock a page if all the extents in the tree
1567 * for that page are unlocked
1569 static int check_page_locked(struct extent_io_tree *tree,
1572 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1573 u64 end = start + PAGE_CACHE_SIZE - 1;
1574 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1580 * helper function to end page writeback if all the extents
1581 * in the tree for that page are done with writeback
1583 static int check_page_writeback(struct extent_io_tree *tree,
1586 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1587 u64 end = start + PAGE_CACHE_SIZE - 1;
1588 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1589 end_page_writeback(page);
1593 /* lots and lots of room for performance fixes in the end_bio funcs */
1596 * after a writepage IO is done, we need to:
1597 * clear the uptodate bits on error
1598 * clear the writeback bits in the extent tree for this IO
1599 * end_page_writeback if the page has no more pending IO
1601 * Scheduling is not allowed, so the extent state tree is expected
1602 * to have one and only one object corresponding to this IO.
1604 static void end_bio_extent_writepage(struct bio *bio, int err)
1606 int uptodate = err == 0;
1607 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1608 struct extent_io_tree *tree;
1615 struct page *page = bvec->bv_page;
1616 tree = &BTRFS_I(page->mapping->host)->io_tree;
1618 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1620 end = start + bvec->bv_len - 1;
1622 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1627 if (--bvec >= bio->bi_io_vec)
1628 prefetchw(&bvec->bv_page->flags);
1629 if (tree->ops && tree->ops->writepage_end_io_hook) {
1630 ret = tree->ops->writepage_end_io_hook(page, start,
1631 end, NULL, uptodate);
1636 if (!uptodate && tree->ops &&
1637 tree->ops->writepage_io_failed_hook) {
1638 ret = tree->ops->writepage_io_failed_hook(bio, page,
1641 uptodate = (err == 0);
1647 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1648 ClearPageUptodate(page);
1652 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1655 end_page_writeback(page);
1657 check_page_writeback(tree, page);
1658 } while (bvec >= bio->bi_io_vec);
1664 * after a readpage IO is done, we need to:
1665 * clear the uptodate bits on error
1666 * set the uptodate bits if things worked
1667 * set the page up to date if all extents in the tree are uptodate
1668 * clear the lock bit in the extent tree
1669 * unlock the page if there are no other extents locked for it
1671 * Scheduling is not allowed, so the extent state tree is expected
1672 * to have one and only one object corresponding to this IO.
1674 static void end_bio_extent_readpage(struct bio *bio, int err)
1676 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1677 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1678 struct extent_io_tree *tree;
1685 struct page *page = bvec->bv_page;
1686 tree = &BTRFS_I(page->mapping->host)->io_tree;
1688 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1690 end = start + bvec->bv_len - 1;
1692 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1697 if (--bvec >= bio->bi_io_vec)
1698 prefetchw(&bvec->bv_page->flags);
1700 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1701 ret = tree->ops->readpage_end_io_hook(page, start, end,
1706 if (!uptodate && tree->ops &&
1707 tree->ops->readpage_io_failed_hook) {
1708 ret = tree->ops->readpage_io_failed_hook(bio, page,
1712 test_bit(BIO_UPTODATE, &bio->bi_flags);
1718 set_extent_uptodate(tree, start, end,
1720 unlock_extent(tree, start, end, GFP_ATOMIC);
1724 SetPageUptodate(page);
1726 ClearPageUptodate(page);
1732 check_page_uptodate(tree, page);
1734 ClearPageUptodate(page);
1737 check_page_locked(tree, page);
1739 } while (bvec >= bio->bi_io_vec);
1745 * IO done from prepare_write is pretty simple, we just unlock
1746 * the structs in the extent tree when done, and set the uptodate bits
1749 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1751 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1752 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1753 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 >= bio->bi_io_vec)
1766 prefetchw(&bvec->bv_page->flags);
1769 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1771 ClearPageUptodate(page);
1775 unlock_extent(tree, start, end, GFP_ATOMIC);
1777 } while (bvec >= bio->bi_io_vec);
1783 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1788 bio = bio_alloc(gfp_flags, nr_vecs);
1790 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1791 while (!bio && (nr_vecs /= 2))
1792 bio = bio_alloc(gfp_flags, nr_vecs);
1797 bio->bi_bdev = bdev;
1798 bio->bi_sector = first_sector;
1803 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1804 unsigned long bio_flags)
1807 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1808 struct page *page = bvec->bv_page;
1809 struct extent_io_tree *tree = bio->bi_private;
1813 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1814 end = start + bvec->bv_len - 1;
1816 bio->bi_private = NULL;
1820 if (tree->ops && tree->ops->submit_bio_hook)
1821 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1822 mirror_num, bio_flags);
1824 submit_bio(rw, bio);
1825 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1831 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1832 struct page *page, sector_t sector,
1833 size_t size, unsigned long offset,
1834 struct block_device *bdev,
1835 struct bio **bio_ret,
1836 unsigned long max_pages,
1837 bio_end_io_t end_io_func,
1839 unsigned long prev_bio_flags,
1840 unsigned long bio_flags)
1846 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1847 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1848 size_t page_size = min(size, PAGE_CACHE_SIZE);
1850 if (bio_ret && *bio_ret) {
1853 contig = bio->bi_sector == sector;
1855 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1858 if (prev_bio_flags != bio_flags || !contig ||
1859 (tree->ops && tree->ops->merge_bio_hook &&
1860 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1862 bio_add_page(bio, page, page_size, offset) < page_size) {
1863 ret = submit_one_bio(rw, bio, mirror_num,
1870 if (this_compressed)
1873 nr = bio_get_nr_vecs(bdev);
1875 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1877 printk("failed to allocate bio nr %d\n", nr);
1880 bio_add_page(bio, page, page_size, offset);
1881 bio->bi_end_io = end_io_func;
1882 bio->bi_private = tree;
1887 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1893 void set_page_extent_mapped(struct page *page)
1895 if (!PagePrivate(page)) {
1896 SetPagePrivate(page);
1897 page_cache_get(page);
1898 set_page_private(page, EXTENT_PAGE_PRIVATE);
1902 void set_page_extent_head(struct page *page, unsigned long len)
1904 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1908 * basic readpage implementation. Locked extent state structs are inserted
1909 * into the tree that are removed when the IO is done (by the end_io
1912 static int __extent_read_full_page(struct extent_io_tree *tree,
1914 get_extent_t *get_extent,
1915 struct bio **bio, int mirror_num,
1916 unsigned long *bio_flags)
1918 struct inode *inode = page->mapping->host;
1919 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1920 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1924 u64 last_byte = i_size_read(inode);
1928 struct extent_map *em;
1929 struct block_device *bdev;
1932 size_t page_offset = 0;
1934 size_t disk_io_size;
1935 size_t blocksize = inode->i_sb->s_blocksize;
1936 unsigned long this_bio_flag = 0;
1938 set_page_extent_mapped(page);
1941 lock_extent(tree, start, end, GFP_NOFS);
1943 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1945 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
1948 iosize = PAGE_CACHE_SIZE - zero_offset;
1949 userpage = kmap_atomic(page, KM_USER0);
1950 memset(userpage + zero_offset, 0, iosize);
1951 flush_dcache_page(page);
1952 kunmap_atomic(userpage, KM_USER0);
1955 while (cur <= end) {
1956 if (cur >= last_byte) {
1958 iosize = PAGE_CACHE_SIZE - page_offset;
1959 userpage = kmap_atomic(page, KM_USER0);
1960 memset(userpage + page_offset, 0, iosize);
1961 flush_dcache_page(page);
1962 kunmap_atomic(userpage, KM_USER0);
1963 set_extent_uptodate(tree, cur, cur + iosize - 1,
1965 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1968 em = get_extent(inode, page, page_offset, cur,
1970 if (IS_ERR(em) || !em) {
1972 unlock_extent(tree, cur, end, GFP_NOFS);
1975 extent_offset = cur - em->start;
1976 if (extent_map_end(em) <= cur) {
1977 printk("bad mapping em [%Lu %Lu] cur %Lu\n", em->start, extent_map_end(em), cur);
1979 BUG_ON(extent_map_end(em) <= cur);
1981 printk("2bad mapping end %Lu cur %Lu\n", end, cur);
1985 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1986 this_bio_flag = EXTENT_BIO_COMPRESSED;
1988 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1989 cur_end = min(extent_map_end(em) - 1, end);
1990 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1991 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
1992 disk_io_size = em->block_len;
1993 sector = em->block_start >> 9;
1995 sector = (em->block_start + extent_offset) >> 9;
1996 disk_io_size = iosize;
1999 block_start = em->block_start;
2000 free_extent_map(em);
2003 /* we've found a hole, just zero and go on */
2004 if (block_start == EXTENT_MAP_HOLE) {
2006 userpage = kmap_atomic(page, KM_USER0);
2007 memset(userpage + page_offset, 0, iosize);
2008 flush_dcache_page(page);
2009 kunmap_atomic(userpage, KM_USER0);
2011 set_extent_uptodate(tree, cur, cur + iosize - 1,
2013 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2015 page_offset += iosize;
2018 /* the get_extent function already copied into the page */
2019 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
2020 check_page_uptodate(tree, page);
2021 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2023 page_offset += iosize;
2026 /* we have an inline extent but it didn't get marked up
2027 * to date. Error out
2029 if (block_start == EXTENT_MAP_INLINE) {
2031 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2033 page_offset += iosize;
2038 if (tree->ops && tree->ops->readpage_io_hook) {
2039 ret = tree->ops->readpage_io_hook(page, cur,
2043 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2045 ret = submit_extent_page(READ, tree, page,
2046 sector, disk_io_size, page_offset,
2048 end_bio_extent_readpage, mirror_num,
2052 *bio_flags = this_bio_flag;
2057 page_offset += iosize;
2060 if (!PageError(page))
2061 SetPageUptodate(page);
2067 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2068 get_extent_t *get_extent)
2070 struct bio *bio = NULL;
2071 unsigned long bio_flags = 0;
2074 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2077 submit_one_bio(READ, bio, 0, bio_flags);
2080 EXPORT_SYMBOL(extent_read_full_page);
2083 * the writepage semantics are similar to regular writepage. extent
2084 * records are inserted to lock ranges in the tree, and as dirty areas
2085 * are found, they are marked writeback. Then the lock bits are removed
2086 * and the end_io handler clears the writeback ranges
2088 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2091 struct inode *inode = page->mapping->host;
2092 struct extent_page_data *epd = data;
2093 struct extent_io_tree *tree = epd->tree;
2094 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2096 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2100 u64 last_byte = i_size_read(inode);
2105 struct extent_map *em;
2106 struct block_device *bdev;
2109 size_t pg_offset = 0;
2111 loff_t i_size = i_size_read(inode);
2112 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2118 WARN_ON(!PageLocked(page));
2119 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2120 if (page->index > end_index ||
2121 (page->index == end_index && !pg_offset)) {
2122 page->mapping->a_ops->invalidatepage(page, 0);
2127 if (page->index == end_index) {
2130 userpage = kmap_atomic(page, KM_USER0);
2131 memset(userpage + pg_offset, 0,
2132 PAGE_CACHE_SIZE - pg_offset);
2133 kunmap_atomic(userpage, KM_USER0);
2134 flush_dcache_page(page);
2138 set_page_extent_mapped(page);
2140 delalloc_start = start;
2143 while(delalloc_end < page_end) {
2144 nr_delalloc = find_lock_delalloc_range(inode, tree,
2149 if (nr_delalloc == 0) {
2150 delalloc_start = delalloc_end + 1;
2153 tree->ops->fill_delalloc(inode, page, delalloc_start,
2154 delalloc_end, &page_started);
2155 delalloc_start = delalloc_end + 1;
2158 /* did the fill delalloc function already unlock and start the IO? */
2163 lock_extent(tree, start, page_end, GFP_NOFS);
2164 unlock_start = start;
2166 if (tree->ops && tree->ops->writepage_start_hook) {
2167 ret = tree->ops->writepage_start_hook(page, start,
2169 if (ret == -EAGAIN) {
2170 unlock_extent(tree, start, page_end, GFP_NOFS);
2171 redirty_page_for_writepage(wbc, page);
2178 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2179 printk("found delalloc bits after lock_extent\n");
2182 if (last_byte <= start) {
2183 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2184 unlock_extent(tree, start, page_end, GFP_NOFS);
2185 if (tree->ops && tree->ops->writepage_end_io_hook)
2186 tree->ops->writepage_end_io_hook(page, start,
2188 unlock_start = page_end + 1;
2192 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2193 blocksize = inode->i_sb->s_blocksize;
2195 while (cur <= end) {
2196 if (cur >= last_byte) {
2197 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2198 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2199 if (tree->ops && tree->ops->writepage_end_io_hook)
2200 tree->ops->writepage_end_io_hook(page, cur,
2202 unlock_start = page_end + 1;
2205 em = epd->get_extent(inode, page, pg_offset, cur,
2207 if (IS_ERR(em) || !em) {
2212 extent_offset = cur - em->start;
2213 BUG_ON(extent_map_end(em) <= cur);
2215 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2216 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2217 sector = (em->block_start + extent_offset) >> 9;
2219 block_start = em->block_start;
2220 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2221 free_extent_map(em);
2225 * compressed and inline extents are written through other
2228 if (compressed || block_start == EXTENT_MAP_HOLE ||
2229 block_start == EXTENT_MAP_INLINE) {
2230 clear_extent_dirty(tree, cur,
2231 cur + iosize - 1, GFP_NOFS);
2233 unlock_extent(tree, unlock_start, cur + iosize -1,
2237 * end_io notification does not happen here for
2238 * compressed extents
2240 if (!compressed && tree->ops &&
2241 tree->ops->writepage_end_io_hook)
2242 tree->ops->writepage_end_io_hook(page, cur,
2245 else if (compressed) {
2246 /* we don't want to end_page_writeback on
2247 * a compressed extent. this happens
2254 pg_offset += iosize;
2258 /* leave this out until we have a page_mkwrite call */
2259 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2262 pg_offset += iosize;
2266 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2267 if (tree->ops && tree->ops->writepage_io_hook) {
2268 ret = tree->ops->writepage_io_hook(page, cur,
2276 unsigned long max_nr = end_index + 1;
2278 set_range_writeback(tree, cur, cur + iosize - 1);
2279 if (!PageWriteback(page)) {
2280 printk("warning page %lu not writeback, "
2281 "cur %llu end %llu\n", page->index,
2282 (unsigned long long)cur,
2283 (unsigned long long)end);
2286 ret = submit_extent_page(WRITE, tree, page, sector,
2287 iosize, pg_offset, bdev,
2289 end_bio_extent_writepage,
2295 pg_offset += iosize;
2300 /* make sure the mapping tag for page dirty gets cleared */
2301 set_page_writeback(page);
2302 end_page_writeback(page);
2304 if (unlock_start <= page_end)
2305 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2311 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2312 * @mapping: address space structure to write
2313 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2314 * @writepage: function called for each page
2315 * @data: data passed to writepage function
2317 * If a page is already under I/O, write_cache_pages() skips it, even
2318 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2319 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2320 * and msync() need to guarantee that all the data which was dirty at the time
2321 * the call was made get new I/O started against them. If wbc->sync_mode is
2322 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2323 * existing IO to complete.
2325 int extent_write_cache_pages(struct extent_io_tree *tree,
2326 struct address_space *mapping,
2327 struct writeback_control *wbc,
2328 writepage_t writepage, void *data)
2330 struct backing_dev_info *bdi = mapping->backing_dev_info;
2333 struct pagevec pvec;
2336 pgoff_t end; /* Inclusive */
2338 int range_whole = 0;
2340 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2341 wbc->encountered_congestion = 1;
2345 pagevec_init(&pvec, 0);
2346 if (wbc->range_cyclic) {
2347 index = mapping->writeback_index; /* Start from prev offset */
2350 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2351 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2352 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2357 while (!done && (index <= end) &&
2358 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2359 PAGECACHE_TAG_DIRTY,
2360 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2364 for (i = 0; i < nr_pages; i++) {
2365 struct page *page = pvec.pages[i];
2368 * At this point we hold neither mapping->tree_lock nor
2369 * lock on the page itself: the page may be truncated or
2370 * invalidated (changing page->mapping to NULL), or even
2371 * swizzled back from swapper_space to tmpfs file
2374 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2375 tree->ops->write_cache_pages_lock_hook(page);
2379 if (unlikely(page->mapping != mapping)) {
2384 if (!wbc->range_cyclic && page->index > end) {
2390 if (wbc->sync_mode != WB_SYNC_NONE)
2391 wait_on_page_writeback(page);
2393 if (PageWriteback(page) ||
2394 !clear_page_dirty_for_io(page)) {
2399 ret = (*writepage)(page, wbc, data);
2401 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2405 if (ret || (--(wbc->nr_to_write) <= 0))
2407 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2408 wbc->encountered_congestion = 1;
2412 pagevec_release(&pvec);
2415 if (!scanned && !done) {
2417 * We hit the last page and there is more work to be done: wrap
2418 * back to the start of the file
2424 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2425 mapping->writeback_index = index;
2427 if (wbc->range_cont)
2428 wbc->range_start = index << PAGE_CACHE_SHIFT;
2431 EXPORT_SYMBOL(extent_write_cache_pages);
2433 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2434 get_extent_t *get_extent,
2435 struct writeback_control *wbc)
2438 struct address_space *mapping = page->mapping;
2439 struct extent_page_data epd = {
2442 .get_extent = get_extent,
2444 struct writeback_control wbc_writepages = {
2446 .sync_mode = WB_SYNC_NONE,
2447 .older_than_this = NULL,
2449 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2450 .range_end = (loff_t)-1,
2454 ret = __extent_writepage(page, wbc, &epd);
2456 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2457 __extent_writepage, &epd);
2459 submit_one_bio(WRITE, epd.bio, 0, 0);
2463 EXPORT_SYMBOL(extent_write_full_page);
2466 int extent_writepages(struct extent_io_tree *tree,
2467 struct address_space *mapping,
2468 get_extent_t *get_extent,
2469 struct writeback_control *wbc)
2472 struct extent_page_data epd = {
2475 .get_extent = get_extent,
2478 ret = extent_write_cache_pages(tree, mapping, wbc,
2479 __extent_writepage, &epd);
2481 submit_one_bio(WRITE, epd.bio, 0, 0);
2485 EXPORT_SYMBOL(extent_writepages);
2487 int extent_readpages(struct extent_io_tree *tree,
2488 struct address_space *mapping,
2489 struct list_head *pages, unsigned nr_pages,
2490 get_extent_t get_extent)
2492 struct bio *bio = NULL;
2494 struct pagevec pvec;
2495 unsigned long bio_flags = 0;
2497 pagevec_init(&pvec, 0);
2498 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2499 struct page *page = list_entry(pages->prev, struct page, lru);
2501 prefetchw(&page->flags);
2502 list_del(&page->lru);
2504 * what we want to do here is call add_to_page_cache_lru,
2505 * but that isn't exported, so we reproduce it here
2507 if (!add_to_page_cache(page, mapping,
2508 page->index, GFP_KERNEL)) {
2510 /* open coding of lru_cache_add, also not exported */
2511 page_cache_get(page);
2512 if (!pagevec_add(&pvec, page))
2513 __pagevec_lru_add(&pvec);
2514 __extent_read_full_page(tree, page, get_extent,
2515 &bio, 0, &bio_flags);
2517 page_cache_release(page);
2519 if (pagevec_count(&pvec))
2520 __pagevec_lru_add(&pvec);
2521 BUG_ON(!list_empty(pages));
2523 submit_one_bio(READ, bio, 0, bio_flags);
2526 EXPORT_SYMBOL(extent_readpages);
2529 * basic invalidatepage code, this waits on any locked or writeback
2530 * ranges corresponding to the page, and then deletes any extent state
2531 * records from the tree
2533 int extent_invalidatepage(struct extent_io_tree *tree,
2534 struct page *page, unsigned long offset)
2536 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2537 u64 end = start + PAGE_CACHE_SIZE - 1;
2538 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2540 start += (offset + blocksize -1) & ~(blocksize - 1);
2544 lock_extent(tree, start, end, GFP_NOFS);
2545 wait_on_extent_writeback(tree, start, end);
2546 clear_extent_bit(tree, start, end,
2547 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2551 EXPORT_SYMBOL(extent_invalidatepage);
2554 * simple commit_write call, set_range_dirty is used to mark both
2555 * the pages and the extent records as dirty
2557 int extent_commit_write(struct extent_io_tree *tree,
2558 struct inode *inode, struct page *page,
2559 unsigned from, unsigned to)
2561 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2563 set_page_extent_mapped(page);
2564 set_page_dirty(page);
2566 if (pos > inode->i_size) {
2567 i_size_write(inode, pos);
2568 mark_inode_dirty(inode);
2572 EXPORT_SYMBOL(extent_commit_write);
2574 int extent_prepare_write(struct extent_io_tree *tree,
2575 struct inode *inode, struct page *page,
2576 unsigned from, unsigned to, get_extent_t *get_extent)
2578 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2579 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2581 u64 orig_block_start;
2584 struct extent_map *em;
2585 unsigned blocksize = 1 << inode->i_blkbits;
2586 size_t page_offset = 0;
2587 size_t block_off_start;
2588 size_t block_off_end;
2594 set_page_extent_mapped(page);
2596 block_start = (page_start + from) & ~((u64)blocksize - 1);
2597 block_end = (page_start + to - 1) | (blocksize - 1);
2598 orig_block_start = block_start;
2600 lock_extent(tree, page_start, page_end, GFP_NOFS);
2601 while(block_start <= block_end) {
2602 em = get_extent(inode, page, page_offset, block_start,
2603 block_end - block_start + 1, 1);
2604 if (IS_ERR(em) || !em) {
2607 cur_end = min(block_end, extent_map_end(em) - 1);
2608 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2609 block_off_end = block_off_start + blocksize;
2610 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2612 if (!PageUptodate(page) && isnew &&
2613 (block_off_end > to || block_off_start < from)) {
2616 kaddr = kmap_atomic(page, KM_USER0);
2617 if (block_off_end > to)
2618 memset(kaddr + to, 0, block_off_end - to);
2619 if (block_off_start < from)
2620 memset(kaddr + block_off_start, 0,
2621 from - block_off_start);
2622 flush_dcache_page(page);
2623 kunmap_atomic(kaddr, KM_USER0);
2625 if ((em->block_start != EXTENT_MAP_HOLE &&
2626 em->block_start != EXTENT_MAP_INLINE) &&
2627 !isnew && !PageUptodate(page) &&
2628 (block_off_end > to || block_off_start < from) &&
2629 !test_range_bit(tree, block_start, cur_end,
2630 EXTENT_UPTODATE, 1)) {
2632 u64 extent_offset = block_start - em->start;
2634 sector = (em->block_start + extent_offset) >> 9;
2635 iosize = (cur_end - block_start + blocksize) &
2636 ~((u64)blocksize - 1);
2638 * we've already got the extent locked, but we
2639 * need to split the state such that our end_bio
2640 * handler can clear the lock.
2642 set_extent_bit(tree, block_start,
2643 block_start + iosize - 1,
2644 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2645 ret = submit_extent_page(READ, tree, page,
2646 sector, iosize, page_offset, em->bdev,
2648 end_bio_extent_preparewrite, 0,
2651 block_start = block_start + iosize;
2653 set_extent_uptodate(tree, block_start, cur_end,
2655 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2656 block_start = cur_end + 1;
2658 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2659 free_extent_map(em);
2662 wait_extent_bit(tree, orig_block_start,
2663 block_end, EXTENT_LOCKED);
2665 check_page_uptodate(tree, page);
2667 /* FIXME, zero out newly allocated blocks on error */
2670 EXPORT_SYMBOL(extent_prepare_write);
2673 * a helper for releasepage, this tests for areas of the page that
2674 * are locked or under IO and drops the related state bits if it is safe
2677 int try_release_extent_state(struct extent_map_tree *map,
2678 struct extent_io_tree *tree, struct page *page,
2681 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2682 u64 end = start + PAGE_CACHE_SIZE - 1;
2685 if (test_range_bit(tree, start, end,
2686 EXTENT_IOBITS | EXTENT_ORDERED, 0))
2689 if ((mask & GFP_NOFS) == GFP_NOFS)
2691 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2696 EXPORT_SYMBOL(try_release_extent_state);
2699 * a helper for releasepage. As long as there are no locked extents
2700 * in the range corresponding to the page, both state records and extent
2701 * map records are removed
2703 int try_release_extent_mapping(struct extent_map_tree *map,
2704 struct extent_io_tree *tree, struct page *page,
2707 struct extent_map *em;
2708 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2709 u64 end = start + PAGE_CACHE_SIZE - 1;
2711 if ((mask & __GFP_WAIT) &&
2712 page->mapping->host->i_size > 16 * 1024 * 1024) {
2714 while (start <= end) {
2715 len = end - start + 1;
2716 spin_lock(&map->lock);
2717 em = lookup_extent_mapping(map, start, len);
2718 if (!em || IS_ERR(em)) {
2719 spin_unlock(&map->lock);
2722 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2723 em->start != start) {
2724 spin_unlock(&map->lock);
2725 free_extent_map(em);
2728 if (!test_range_bit(tree, em->start,
2729 extent_map_end(em) - 1,
2730 EXTENT_LOCKED | EXTENT_WRITEBACK |
2733 remove_extent_mapping(map, em);
2734 /* once for the rb tree */
2735 free_extent_map(em);
2737 start = extent_map_end(em);
2738 spin_unlock(&map->lock);
2741 free_extent_map(em);
2744 return try_release_extent_state(map, tree, page, mask);
2746 EXPORT_SYMBOL(try_release_extent_mapping);
2748 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2749 get_extent_t *get_extent)
2751 struct inode *inode = mapping->host;
2752 u64 start = iblock << inode->i_blkbits;
2753 sector_t sector = 0;
2754 struct extent_map *em;
2756 em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2757 if (!em || IS_ERR(em))
2760 if (em->block_start == EXTENT_MAP_INLINE ||
2761 em->block_start == EXTENT_MAP_HOLE)
2764 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2766 free_extent_map(em);
2770 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2774 struct address_space *mapping;
2777 return eb->first_page;
2778 i += eb->start >> PAGE_CACHE_SHIFT;
2779 mapping = eb->first_page->mapping;
2784 * extent_buffer_page is only called after pinning the page
2785 * by increasing the reference count. So we know the page must
2786 * be in the radix tree.
2789 p = radix_tree_lookup(&mapping->page_tree, i);
2795 static inline unsigned long num_extent_pages(u64 start, u64 len)
2797 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2798 (start >> PAGE_CACHE_SHIFT);
2801 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2806 struct extent_buffer *eb = NULL;
2808 unsigned long flags;
2811 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2814 mutex_init(&eb->mutex);
2816 spin_lock_irqsave(&leak_lock, flags);
2817 list_add(&eb->leak_list, &buffers);
2818 spin_unlock_irqrestore(&leak_lock, flags);
2820 atomic_set(&eb->refs, 1);
2825 static void __free_extent_buffer(struct extent_buffer *eb)
2828 unsigned long flags;
2829 spin_lock_irqsave(&leak_lock, flags);
2830 list_del(&eb->leak_list);
2831 spin_unlock_irqrestore(&leak_lock, flags);
2833 kmem_cache_free(extent_buffer_cache, eb);
2836 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2837 u64 start, unsigned long len,
2841 unsigned long num_pages = num_extent_pages(start, len);
2843 unsigned long index = start >> PAGE_CACHE_SHIFT;
2844 struct extent_buffer *eb;
2845 struct extent_buffer *exists = NULL;
2847 struct address_space *mapping = tree->mapping;
2850 spin_lock(&tree->buffer_lock);
2851 eb = buffer_search(tree, start);
2853 atomic_inc(&eb->refs);
2854 spin_unlock(&tree->buffer_lock);
2855 mark_page_accessed(eb->first_page);
2858 spin_unlock(&tree->buffer_lock);
2860 eb = __alloc_extent_buffer(tree, start, len, mask);
2865 eb->first_page = page0;
2868 page_cache_get(page0);
2869 mark_page_accessed(page0);
2870 set_page_extent_mapped(page0);
2871 set_page_extent_head(page0, len);
2872 uptodate = PageUptodate(page0);
2876 for (; i < num_pages; i++, index++) {
2877 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2882 set_page_extent_mapped(p);
2883 mark_page_accessed(p);
2886 set_page_extent_head(p, len);
2888 set_page_private(p, EXTENT_PAGE_PRIVATE);
2890 if (!PageUptodate(p))
2895 eb->flags |= EXTENT_UPTODATE;
2896 eb->flags |= EXTENT_BUFFER_FILLED;
2898 spin_lock(&tree->buffer_lock);
2899 exists = buffer_tree_insert(tree, start, &eb->rb_node);
2901 /* add one reference for the caller */
2902 atomic_inc(&exists->refs);
2903 spin_unlock(&tree->buffer_lock);
2906 spin_unlock(&tree->buffer_lock);
2908 /* add one reference for the tree */
2909 atomic_inc(&eb->refs);
2913 if (!atomic_dec_and_test(&eb->refs))
2915 for (index = 1; index < i; index++)
2916 page_cache_release(extent_buffer_page(eb, index));
2917 page_cache_release(extent_buffer_page(eb, 0));
2918 __free_extent_buffer(eb);
2921 EXPORT_SYMBOL(alloc_extent_buffer);
2923 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2924 u64 start, unsigned long len,
2927 struct extent_buffer *eb;
2929 spin_lock(&tree->buffer_lock);
2930 eb = buffer_search(tree, start);
2932 atomic_inc(&eb->refs);
2933 spin_unlock(&tree->buffer_lock);
2936 mark_page_accessed(eb->first_page);
2940 EXPORT_SYMBOL(find_extent_buffer);
2942 void free_extent_buffer(struct extent_buffer *eb)
2947 if (!atomic_dec_and_test(&eb->refs))
2952 EXPORT_SYMBOL(free_extent_buffer);
2954 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2955 struct extent_buffer *eb)
2959 unsigned long num_pages;
2962 u64 start = eb->start;
2963 u64 end = start + eb->len - 1;
2965 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2966 num_pages = num_extent_pages(eb->start, eb->len);
2968 for (i = 0; i < num_pages; i++) {
2969 page = extent_buffer_page(eb, i);
2972 set_page_extent_head(page, eb->len);
2974 set_page_private(page, EXTENT_PAGE_PRIVATE);
2977 * if we're on the last page or the first page and the
2978 * block isn't aligned on a page boundary, do extra checks
2979 * to make sure we don't clean page that is partially dirty
2981 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2982 ((i == num_pages - 1) &&
2983 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2984 start = (u64)page->index << PAGE_CACHE_SHIFT;
2985 end = start + PAGE_CACHE_SIZE - 1;
2986 if (test_range_bit(tree, start, end,
2992 clear_page_dirty_for_io(page);
2993 spin_lock_irq(&page->mapping->tree_lock);
2994 if (!PageDirty(page)) {
2995 radix_tree_tag_clear(&page->mapping->page_tree,
2997 PAGECACHE_TAG_DIRTY);
2999 spin_unlock_irq(&page->mapping->tree_lock);
3004 EXPORT_SYMBOL(clear_extent_buffer_dirty);
3006 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3007 struct extent_buffer *eb)
3009 return wait_on_extent_writeback(tree, eb->start,
3010 eb->start + eb->len - 1);
3012 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
3014 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3015 struct extent_buffer *eb)
3018 unsigned long num_pages;
3020 num_pages = num_extent_pages(eb->start, eb->len);
3021 for (i = 0; i < num_pages; i++) {
3022 struct page *page = extent_buffer_page(eb, i);
3023 /* writepage may need to do something special for the
3024 * first page, we have to make sure page->private is
3025 * properly set. releasepage may drop page->private
3026 * on us if the page isn't already dirty.
3030 set_page_extent_head(page, eb->len);
3031 } else if (PagePrivate(page) &&
3032 page->private != EXTENT_PAGE_PRIVATE) {
3033 set_page_extent_mapped(page);
3035 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3036 set_extent_dirty(tree, page_offset(page),
3037 page_offset(page) + PAGE_CACHE_SIZE -1,
3043 EXPORT_SYMBOL(set_extent_buffer_dirty);
3045 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3046 struct extent_buffer *eb)
3050 unsigned long num_pages;
3052 num_pages = num_extent_pages(eb->start, eb->len);
3053 eb->flags &= ~EXTENT_UPTODATE;
3055 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3057 for (i = 0; i < num_pages; i++) {
3058 page = extent_buffer_page(eb, i);
3060 ClearPageUptodate(page);
3065 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3066 struct extent_buffer *eb)
3070 unsigned long num_pages;
3072 num_pages = num_extent_pages(eb->start, eb->len);
3074 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3076 for (i = 0; i < num_pages; i++) {
3077 page = extent_buffer_page(eb, i);
3078 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3079 ((i == num_pages - 1) &&
3080 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3081 check_page_uptodate(tree, page);
3084 SetPageUptodate(page);
3088 EXPORT_SYMBOL(set_extent_buffer_uptodate);
3090 int extent_range_uptodate(struct extent_io_tree *tree,
3095 int pg_uptodate = 1;
3097 unsigned long index;
3099 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3102 while(start <= end) {
3103 index = start >> PAGE_CACHE_SHIFT;
3104 page = find_get_page(tree->mapping, index);
3105 uptodate = PageUptodate(page);
3106 page_cache_release(page);
3111 start += PAGE_CACHE_SIZE;
3116 int extent_buffer_uptodate(struct extent_io_tree *tree,
3117 struct extent_buffer *eb)
3120 unsigned long num_pages;
3123 int pg_uptodate = 1;
3125 if (eb->flags & EXTENT_UPTODATE)
3128 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3129 EXTENT_UPTODATE, 1);
3133 num_pages = num_extent_pages(eb->start, eb->len);
3134 for (i = 0; i < num_pages; i++) {
3135 page = extent_buffer_page(eb, i);
3136 if (!PageUptodate(page)) {
3143 EXPORT_SYMBOL(extent_buffer_uptodate);
3145 int read_extent_buffer_pages(struct extent_io_tree *tree,
3146 struct extent_buffer *eb,
3147 u64 start, int wait,
3148 get_extent_t *get_extent, int mirror_num)
3151 unsigned long start_i;
3155 int locked_pages = 0;
3156 int all_uptodate = 1;
3157 int inc_all_pages = 0;
3158 unsigned long num_pages;
3159 struct bio *bio = NULL;
3160 unsigned long bio_flags = 0;
3162 if (eb->flags & EXTENT_UPTODATE)
3165 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3166 EXTENT_UPTODATE, 1)) {
3171 WARN_ON(start < eb->start);
3172 start_i = (start >> PAGE_CACHE_SHIFT) -
3173 (eb->start >> PAGE_CACHE_SHIFT);
3178 num_pages = num_extent_pages(eb->start, eb->len);
3179 for (i = start_i; i < num_pages; i++) {
3180 page = extent_buffer_page(eb, i);
3182 if (!trylock_page(page))
3188 if (!PageUptodate(page)) {
3194 eb->flags |= EXTENT_UPTODATE;
3196 printk("all up to date but ret is %d\n", ret);
3201 for (i = start_i; i < num_pages; i++) {
3202 page = extent_buffer_page(eb, i);
3204 page_cache_get(page);
3205 if (!PageUptodate(page)) {
3208 ClearPageError(page);
3209 err = __extent_read_full_page(tree, page,
3211 mirror_num, &bio_flags);
3214 printk("err %d from __extent_read_full_page\n", ret);
3222 submit_one_bio(READ, bio, mirror_num, bio_flags);
3226 printk("ret %d wait %d returning\n", ret, wait);
3229 for (i = start_i; i < num_pages; i++) {
3230 page = extent_buffer_page(eb, i);
3231 wait_on_page_locked(page);
3232 if (!PageUptodate(page)) {
3233 printk("page not uptodate after wait_on_page_locked\n");
3238 eb->flags |= EXTENT_UPTODATE;
3243 while(locked_pages > 0) {
3244 page = extent_buffer_page(eb, i);
3251 EXPORT_SYMBOL(read_extent_buffer_pages);
3253 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3254 unsigned long start,
3261 char *dst = (char *)dstv;
3262 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3263 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3265 WARN_ON(start > eb->len);
3266 WARN_ON(start + len > eb->start + eb->len);
3268 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3271 page = extent_buffer_page(eb, i);
3273 cur = min(len, (PAGE_CACHE_SIZE - offset));
3274 kaddr = kmap_atomic(page, KM_USER1);
3275 memcpy(dst, kaddr + offset, cur);
3276 kunmap_atomic(kaddr, KM_USER1);
3284 EXPORT_SYMBOL(read_extent_buffer);
3286 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3287 unsigned long min_len, char **token, char **map,
3288 unsigned long *map_start,
3289 unsigned long *map_len, int km)
3291 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3294 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3295 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3296 unsigned long end_i = (start_offset + start + min_len - 1) >>
3303 offset = start_offset;
3307 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3309 if (start + min_len > eb->len) {
3310 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3314 p = extent_buffer_page(eb, i);
3315 kaddr = kmap_atomic(p, km);
3317 *map = kaddr + offset;
3318 *map_len = PAGE_CACHE_SIZE - offset;
3321 EXPORT_SYMBOL(map_private_extent_buffer);
3323 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3324 unsigned long min_len,
3325 char **token, char **map,
3326 unsigned long *map_start,
3327 unsigned long *map_len, int km)
3331 if (eb->map_token) {
3332 unmap_extent_buffer(eb, eb->map_token, km);
3333 eb->map_token = NULL;
3336 err = map_private_extent_buffer(eb, start, min_len, token, map,
3337 map_start, map_len, km);
3339 eb->map_token = *token;
3341 eb->map_start = *map_start;
3342 eb->map_len = *map_len;
3346 EXPORT_SYMBOL(map_extent_buffer);
3348 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3350 kunmap_atomic(token, km);
3352 EXPORT_SYMBOL(unmap_extent_buffer);
3354 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3355 unsigned long start,
3362 char *ptr = (char *)ptrv;
3363 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3364 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3367 WARN_ON(start > eb->len);
3368 WARN_ON(start + len > eb->start + eb->len);
3370 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3373 page = extent_buffer_page(eb, i);
3375 cur = min(len, (PAGE_CACHE_SIZE - offset));
3377 kaddr = kmap_atomic(page, KM_USER0);
3378 ret = memcmp(ptr, kaddr + offset, cur);
3379 kunmap_atomic(kaddr, KM_USER0);
3390 EXPORT_SYMBOL(memcmp_extent_buffer);
3392 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3393 unsigned long start, unsigned long len)
3399 char *src = (char *)srcv;
3400 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3401 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3403 WARN_ON(start > eb->len);
3404 WARN_ON(start + len > eb->start + eb->len);
3406 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3409 page = extent_buffer_page(eb, i);
3410 WARN_ON(!PageUptodate(page));
3412 cur = min(len, PAGE_CACHE_SIZE - offset);
3413 kaddr = kmap_atomic(page, KM_USER1);
3414 memcpy(kaddr + offset, src, cur);
3415 kunmap_atomic(kaddr, KM_USER1);
3423 EXPORT_SYMBOL(write_extent_buffer);
3425 void memset_extent_buffer(struct extent_buffer *eb, char c,
3426 unsigned long start, unsigned long len)
3432 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3433 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3435 WARN_ON(start > eb->len);
3436 WARN_ON(start + len > eb->start + eb->len);
3438 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3441 page = extent_buffer_page(eb, i);
3442 WARN_ON(!PageUptodate(page));
3444 cur = min(len, PAGE_CACHE_SIZE - offset);
3445 kaddr = kmap_atomic(page, KM_USER0);
3446 memset(kaddr + offset, c, cur);
3447 kunmap_atomic(kaddr, KM_USER0);
3454 EXPORT_SYMBOL(memset_extent_buffer);
3456 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3457 unsigned long dst_offset, unsigned long src_offset,
3460 u64 dst_len = dst->len;
3465 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3466 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3468 WARN_ON(src->len != dst_len);
3470 offset = (start_offset + dst_offset) &
3471 ((unsigned long)PAGE_CACHE_SIZE - 1);
3474 page = extent_buffer_page(dst, i);
3475 WARN_ON(!PageUptodate(page));
3477 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3479 kaddr = kmap_atomic(page, KM_USER0);
3480 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3481 kunmap_atomic(kaddr, KM_USER0);
3489 EXPORT_SYMBOL(copy_extent_buffer);
3491 static void move_pages(struct page *dst_page, struct page *src_page,
3492 unsigned long dst_off, unsigned long src_off,
3495 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3496 if (dst_page == src_page) {
3497 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3499 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3500 char *p = dst_kaddr + dst_off + len;
3501 char *s = src_kaddr + src_off + len;
3506 kunmap_atomic(src_kaddr, KM_USER1);
3508 kunmap_atomic(dst_kaddr, KM_USER0);
3511 static void copy_pages(struct page *dst_page, struct page *src_page,
3512 unsigned long dst_off, unsigned long src_off,
3515 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3518 if (dst_page != src_page)
3519 src_kaddr = kmap_atomic(src_page, KM_USER1);
3521 src_kaddr = dst_kaddr;
3523 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3524 kunmap_atomic(dst_kaddr, KM_USER0);
3525 if (dst_page != src_page)
3526 kunmap_atomic(src_kaddr, KM_USER1);
3529 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3530 unsigned long src_offset, unsigned long len)
3533 size_t dst_off_in_page;
3534 size_t src_off_in_page;
3535 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3536 unsigned long dst_i;
3537 unsigned long src_i;
3539 if (src_offset + len > dst->len) {
3540 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3541 src_offset, len, dst->len);
3544 if (dst_offset + len > dst->len) {
3545 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3546 dst_offset, len, dst->len);
3551 dst_off_in_page = (start_offset + dst_offset) &
3552 ((unsigned long)PAGE_CACHE_SIZE - 1);
3553 src_off_in_page = (start_offset + src_offset) &
3554 ((unsigned long)PAGE_CACHE_SIZE - 1);
3556 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3557 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3559 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3561 cur = min_t(unsigned long, cur,
3562 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3564 copy_pages(extent_buffer_page(dst, dst_i),
3565 extent_buffer_page(dst, src_i),
3566 dst_off_in_page, src_off_in_page, cur);
3573 EXPORT_SYMBOL(memcpy_extent_buffer);
3575 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3576 unsigned long src_offset, unsigned long len)
3579 size_t dst_off_in_page;
3580 size_t src_off_in_page;
3581 unsigned long dst_end = dst_offset + len - 1;
3582 unsigned long src_end = src_offset + len - 1;
3583 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3584 unsigned long dst_i;
3585 unsigned long src_i;
3587 if (src_offset + len > dst->len) {
3588 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3589 src_offset, len, dst->len);
3592 if (dst_offset + len > dst->len) {
3593 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3594 dst_offset, len, dst->len);
3597 if (dst_offset < src_offset) {
3598 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3602 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3603 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3605 dst_off_in_page = (start_offset + dst_end) &
3606 ((unsigned long)PAGE_CACHE_SIZE - 1);
3607 src_off_in_page = (start_offset + src_end) &
3608 ((unsigned long)PAGE_CACHE_SIZE - 1);
3610 cur = min_t(unsigned long, len, src_off_in_page + 1);
3611 cur = min(cur, dst_off_in_page + 1);
3612 move_pages(extent_buffer_page(dst, dst_i),
3613 extent_buffer_page(dst, src_i),
3614 dst_off_in_page - cur + 1,
3615 src_off_in_page - cur + 1, cur);
3622 EXPORT_SYMBOL(memmove_extent_buffer);
3624 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3626 u64 start = page_offset(page);
3627 struct extent_buffer *eb;
3630 unsigned long num_pages;
3632 spin_lock(&tree->buffer_lock);
3633 eb = buffer_search(tree, start);
3637 if (atomic_read(&eb->refs) > 1) {
3641 /* at this point we can safely release the extent buffer */
3642 num_pages = num_extent_pages(eb->start, eb->len);
3643 for (i = 0; i < num_pages; i++)
3644 page_cache_release(extent_buffer_page(eb, i));
3645 rb_erase(&eb->rb_node, &tree->buffer);
3646 __free_extent_buffer(eb);
3648 spin_unlock(&tree->buffer_lock);
3651 EXPORT_SYMBOL(try_release_extent_buffer);