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"
18 /* temporary define until extent_map moves out of btrfs */
19 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
20 unsigned long extra_flags,
21 void (*ctor)(void *, struct kmem_cache *,
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
27 static LIST_HEAD(buffers);
28 static LIST_HEAD(states);
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;
44 int __init extent_io_init(void)
46 extent_state_cache = btrfs_cache_create("extent_state",
47 sizeof(struct extent_state), 0,
49 if (!extent_state_cache)
52 extent_buffer_cache = btrfs_cache_create("extent_buffers",
53 sizeof(struct extent_buffer), 0,
55 if (!extent_buffer_cache)
56 goto free_state_cache;
60 kmem_cache_destroy(extent_state_cache);
64 void extent_io_exit(void)
66 struct extent_state *state;
68 while (!list_empty(&states)) {
69 state = list_entry(states.next, struct extent_state, list);
70 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));
71 list_del(&state->list);
72 kmem_cache_free(extent_state_cache, state);
76 if (extent_state_cache)
77 kmem_cache_destroy(extent_state_cache);
78 if (extent_buffer_cache)
79 kmem_cache_destroy(extent_buffer_cache);
82 void extent_io_tree_init(struct extent_io_tree *tree,
83 struct address_space *mapping, gfp_t mask)
85 tree->state.rb_node = NULL;
87 tree->dirty_bytes = 0;
88 spin_lock_init(&tree->lock);
89 spin_lock_init(&tree->lru_lock);
90 tree->mapping = mapping;
91 INIT_LIST_HEAD(&tree->buffer_lru);
94 EXPORT_SYMBOL(extent_io_tree_init);
96 void extent_io_tree_empty_lru(struct extent_io_tree *tree)
98 struct extent_buffer *eb;
99 while(!list_empty(&tree->buffer_lru)) {
100 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
102 list_del_init(&eb->lru);
103 free_extent_buffer(eb);
106 EXPORT_SYMBOL(extent_io_tree_empty_lru);
108 struct extent_state *alloc_extent_state(gfp_t mask)
110 struct extent_state *state;
112 state = kmem_cache_alloc(extent_state_cache, mask);
113 if (!state || IS_ERR(state))
119 atomic_set(&state->refs, 1);
120 init_waitqueue_head(&state->wq);
123 EXPORT_SYMBOL(alloc_extent_state);
125 void free_extent_state(struct extent_state *state)
129 if (atomic_dec_and_test(&state->refs)) {
130 WARN_ON(state->tree);
131 kmem_cache_free(extent_state_cache, state);
134 EXPORT_SYMBOL(free_extent_state);
136 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
137 struct rb_node *node)
139 struct rb_node ** p = &root->rb_node;
140 struct rb_node * parent = NULL;
141 struct tree_entry *entry;
145 entry = rb_entry(parent, struct tree_entry, rb_node);
147 if (offset < entry->start)
149 else if (offset > entry->end)
155 entry = rb_entry(node, struct tree_entry, rb_node);
156 rb_link_node(node, parent, p);
157 rb_insert_color(node, root);
161 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
162 struct rb_node **prev_ret,
163 struct rb_node **next_ret)
165 struct rb_node * n = root->rb_node;
166 struct rb_node *prev = NULL;
167 struct rb_node *orig_prev = NULL;
168 struct tree_entry *entry;
169 struct tree_entry *prev_entry = NULL;
172 entry = rb_entry(n, struct tree_entry, rb_node);
176 if (offset < entry->start)
178 else if (offset > entry->end)
186 while(prev && offset > prev_entry->end) {
187 prev = rb_next(prev);
188 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
195 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
196 while(prev && offset < prev_entry->start) {
197 prev = rb_prev(prev);
198 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
205 static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
207 struct rb_node *prev = NULL;
210 ret = __tree_search(root, offset, &prev, NULL);
217 * utility function to look for merge candidates inside a given range.
218 * Any extents with matching state are merged together into a single
219 * extent in the tree. Extents with EXTENT_IO in their state field
220 * are not merged because the end_io handlers need to be able to do
221 * operations on them without sleeping (or doing allocations/splits).
223 * This should be called with the tree lock held.
225 static int merge_state(struct extent_io_tree *tree,
226 struct extent_state *state)
228 struct extent_state *other;
229 struct rb_node *other_node;
231 if (state->state & EXTENT_IOBITS)
234 other_node = rb_prev(&state->rb_node);
236 other = rb_entry(other_node, struct extent_state, rb_node);
237 if (other->end == state->start - 1 &&
238 other->state == state->state) {
239 state->start = other->start;
241 rb_erase(&other->rb_node, &tree->state);
242 free_extent_state(other);
245 other_node = rb_next(&state->rb_node);
247 other = rb_entry(other_node, struct extent_state, rb_node);
248 if (other->start == state->end + 1 &&
249 other->state == state->state) {
250 other->start = state->start;
252 rb_erase(&state->rb_node, &tree->state);
253 free_extent_state(state);
259 static void set_state_cb(struct extent_io_tree *tree,
260 struct extent_state *state,
263 if (tree->ops && tree->ops->set_bit_hook) {
264 tree->ops->set_bit_hook(tree->mapping->host, state->start,
265 state->end, state->state, bits);
269 static void clear_state_cb(struct extent_io_tree *tree,
270 struct extent_state *state,
273 if (tree->ops && tree->ops->set_bit_hook) {
274 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
275 state->end, state->state, bits);
280 * insert an extent_state struct into the tree. 'bits' are set on the
281 * struct before it is inserted.
283 * This may return -EEXIST if the extent is already there, in which case the
284 * state struct is freed.
286 * The tree lock is not taken internally. This is a utility function and
287 * probably isn't what you want to call (see set/clear_extent_bit).
289 static int insert_state(struct extent_io_tree *tree,
290 struct extent_state *state, u64 start, u64 end,
293 struct rb_node *node;
296 printk("end < start %Lu %Lu\n", end, start);
299 if (bits & EXTENT_DIRTY)
300 tree->dirty_bytes += end - start + 1;
301 set_state_cb(tree, state, bits);
302 state->state |= bits;
303 state->start = start;
305 node = tree_insert(&tree->state, end, &state->rb_node);
307 struct extent_state *found;
308 found = rb_entry(node, struct extent_state, rb_node);
309 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
310 free_extent_state(state);
314 merge_state(tree, state);
319 * split a given extent state struct in two, inserting the preallocated
320 * struct 'prealloc' as the newly created second half. 'split' indicates an
321 * offset inside 'orig' where it should be split.
324 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
325 * are two extent state structs in the tree:
326 * prealloc: [orig->start, split - 1]
327 * orig: [ split, orig->end ]
329 * The tree locks are not taken by this function. They need to be held
332 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
333 struct extent_state *prealloc, u64 split)
335 struct rb_node *node;
336 prealloc->start = orig->start;
337 prealloc->end = split - 1;
338 prealloc->state = orig->state;
341 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
343 struct extent_state *found;
344 found = rb_entry(node, struct extent_state, rb_node);
345 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
346 free_extent_state(prealloc);
349 prealloc->tree = tree;
354 * utility function to clear some bits in an extent state struct.
355 * it will optionally wake up any one waiting on this state (wake == 1), or
356 * forcibly remove the state from the tree (delete == 1).
358 * If no bits are set on the state struct after clearing things, the
359 * struct is freed and removed from the tree
361 static int clear_state_bit(struct extent_io_tree *tree,
362 struct extent_state *state, int bits, int wake,
365 int ret = state->state & bits;
367 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
368 u64 range = state->end - state->start + 1;
369 WARN_ON(range > tree->dirty_bytes);
370 tree->dirty_bytes -= range;
372 clear_state_cb(tree, state, bits);
373 state->state &= ~bits;
376 if (delete || state->state == 0) {
378 rb_erase(&state->rb_node, &tree->state);
380 free_extent_state(state);
385 merge_state(tree, state);
391 * clear some bits on a range in the tree. This may require splitting
392 * or inserting elements in the tree, so the gfp mask is used to
393 * indicate which allocations or sleeping are allowed.
395 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
396 * the given range from the tree regardless of state (ie for truncate).
398 * the range [start, end] is inclusive.
400 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
401 * bits were already set, or zero if none of the bits were already set.
403 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
404 int bits, int wake, int delete, gfp_t mask)
406 struct extent_state *state;
407 struct extent_state *prealloc = NULL;
408 struct rb_node *node;
414 if (!prealloc && (mask & __GFP_WAIT)) {
415 prealloc = alloc_extent_state(mask);
420 spin_lock_irqsave(&tree->lock, flags);
422 * this search will find the extents that end after
425 node = tree_search(&tree->state, start);
428 state = rb_entry(node, struct extent_state, rb_node);
429 if (state->start > end)
431 WARN_ON(state->end < start);
434 * | ---- desired range ---- |
436 * | ------------- state -------------- |
438 * We need to split the extent we found, and may flip
439 * bits on second half.
441 * If the extent we found extends past our range, we
442 * just split and search again. It'll get split again
443 * the next time though.
445 * If the extent we found is inside our range, we clear
446 * the desired bit on it.
449 if (state->start < start) {
451 prealloc = alloc_extent_state(GFP_ATOMIC);
452 err = split_state(tree, state, prealloc, start);
453 BUG_ON(err == -EEXIST);
457 if (state->end <= end) {
458 start = state->end + 1;
459 set |= clear_state_bit(tree, state, bits,
462 start = state->start;
467 * | ---- desired range ---- |
469 * We need to split the extent, and clear the bit
472 if (state->start <= end && state->end > end) {
474 prealloc = alloc_extent_state(GFP_ATOMIC);
475 err = split_state(tree, state, prealloc, end + 1);
476 BUG_ON(err == -EEXIST);
480 set |= clear_state_bit(tree, prealloc, bits,
486 start = state->end + 1;
487 set |= clear_state_bit(tree, state, bits, wake, delete);
491 spin_unlock_irqrestore(&tree->lock, flags);
493 free_extent_state(prealloc);
500 spin_unlock_irqrestore(&tree->lock, flags);
501 if (mask & __GFP_WAIT)
505 EXPORT_SYMBOL(clear_extent_bit);
507 static int wait_on_state(struct extent_io_tree *tree,
508 struct extent_state *state)
511 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
512 spin_unlock_irq(&tree->lock);
514 spin_lock_irq(&tree->lock);
515 finish_wait(&state->wq, &wait);
520 * waits for one or more bits to clear on a range in the state tree.
521 * The range [start, end] is inclusive.
522 * The tree lock is taken by this function
524 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
526 struct extent_state *state;
527 struct rb_node *node;
529 spin_lock_irq(&tree->lock);
533 * this search will find all the extents that end after
536 node = tree_search(&tree->state, start);
540 state = rb_entry(node, struct extent_state, rb_node);
542 if (state->start > end)
545 if (state->state & bits) {
546 start = state->start;
547 atomic_inc(&state->refs);
548 wait_on_state(tree, state);
549 free_extent_state(state);
552 start = state->end + 1;
557 if (need_resched()) {
558 spin_unlock_irq(&tree->lock);
560 spin_lock_irq(&tree->lock);
564 spin_unlock_irq(&tree->lock);
567 EXPORT_SYMBOL(wait_extent_bit);
569 static void set_state_bits(struct extent_io_tree *tree,
570 struct extent_state *state,
573 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
574 u64 range = state->end - state->start + 1;
575 tree->dirty_bytes += range;
577 set_state_cb(tree, state, bits);
578 state->state |= bits;
582 * set some bits on a range in the tree. This may require allocations
583 * or sleeping, so the gfp mask is used to indicate what is allowed.
585 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
586 * range already has the desired bits set. The start of the existing
587 * range is returned in failed_start in this case.
589 * [start, end] is inclusive
590 * This takes the tree lock.
592 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
593 int exclusive, u64 *failed_start, gfp_t mask)
595 struct extent_state *state;
596 struct extent_state *prealloc = NULL;
597 struct rb_node *node;
604 if (!prealloc && (mask & __GFP_WAIT)) {
605 prealloc = alloc_extent_state(mask);
610 spin_lock_irqsave(&tree->lock, flags);
612 * this search will find all the extents that end after
615 node = tree_search(&tree->state, start);
617 err = insert_state(tree, prealloc, start, end, bits);
619 BUG_ON(err == -EEXIST);
623 state = rb_entry(node, struct extent_state, rb_node);
624 last_start = state->start;
625 last_end = state->end;
628 * | ---- desired range ---- |
631 * Just lock what we found and keep going
633 if (state->start == start && state->end <= end) {
634 set = state->state & bits;
635 if (set && exclusive) {
636 *failed_start = state->start;
640 set_state_bits(tree, state, bits);
641 start = state->end + 1;
642 merge_state(tree, state);
647 * | ---- desired range ---- |
650 * | ------------- state -------------- |
652 * We need to split the extent we found, and may flip bits on
655 * If the extent we found extends past our
656 * range, we just split and search again. It'll get split
657 * again the next time though.
659 * If the extent we found is inside our range, we set the
662 if (state->start < start) {
663 set = state->state & bits;
664 if (exclusive && set) {
665 *failed_start = start;
669 err = split_state(tree, state, prealloc, start);
670 BUG_ON(err == -EEXIST);
674 if (state->end <= end) {
675 set_state_bits(tree, state, bits);
676 start = state->end + 1;
677 merge_state(tree, state);
679 start = state->start;
684 * | ---- desired range ---- |
685 * | state | or | state |
687 * There's a hole, we need to insert something in it and
688 * ignore the extent we found.
690 if (state->start > start) {
692 if (end < last_start)
695 this_end = last_start -1;
696 err = insert_state(tree, prealloc, start, this_end,
699 BUG_ON(err == -EEXIST);
702 start = this_end + 1;
706 * | ---- desired range ---- |
708 * We need to split the extent, and set the bit
711 if (state->start <= end && state->end > end) {
712 set = state->state & bits;
713 if (exclusive && set) {
714 *failed_start = start;
718 err = split_state(tree, state, prealloc, end + 1);
719 BUG_ON(err == -EEXIST);
721 set_state_bits(tree, prealloc, bits);
722 merge_state(tree, prealloc);
730 spin_unlock_irqrestore(&tree->lock, flags);
732 free_extent_state(prealloc);
739 spin_unlock_irqrestore(&tree->lock, flags);
740 if (mask & __GFP_WAIT)
744 EXPORT_SYMBOL(set_extent_bit);
746 /* wrappers around set/clear extent bit */
747 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
750 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
753 EXPORT_SYMBOL(set_extent_dirty);
755 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
756 int bits, gfp_t mask)
758 return set_extent_bit(tree, start, end, bits, 0, NULL,
761 EXPORT_SYMBOL(set_extent_bits);
763 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
764 int bits, gfp_t mask)
766 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
768 EXPORT_SYMBOL(clear_extent_bits);
770 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
773 return set_extent_bit(tree, start, end,
774 EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
777 EXPORT_SYMBOL(set_extent_delalloc);
779 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
782 return clear_extent_bit(tree, start, end,
783 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
785 EXPORT_SYMBOL(clear_extent_dirty);
787 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
790 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
793 EXPORT_SYMBOL(set_extent_new);
795 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
798 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
800 EXPORT_SYMBOL(clear_extent_new);
802 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
805 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
808 EXPORT_SYMBOL(set_extent_uptodate);
810 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
813 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
815 EXPORT_SYMBOL(clear_extent_uptodate);
817 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
820 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
823 EXPORT_SYMBOL(set_extent_writeback);
825 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
828 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
830 EXPORT_SYMBOL(clear_extent_writeback);
832 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
834 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
836 EXPORT_SYMBOL(wait_on_extent_writeback);
838 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
843 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
844 &failed_start, mask);
845 if (err == -EEXIST && (mask & __GFP_WAIT)) {
846 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
847 start = failed_start;
851 WARN_ON(start > end);
855 EXPORT_SYMBOL(lock_extent);
857 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
860 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
862 EXPORT_SYMBOL(unlock_extent);
865 * helper function to set pages and extents in the tree dirty
867 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
869 unsigned long index = start >> PAGE_CACHE_SHIFT;
870 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
873 while (index <= end_index) {
874 page = find_get_page(tree->mapping, index);
876 __set_page_dirty_nobuffers(page);
877 page_cache_release(page);
880 set_extent_dirty(tree, start, end, GFP_NOFS);
883 EXPORT_SYMBOL(set_range_dirty);
886 * helper function to set both pages and extents in the tree writeback
888 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
890 unsigned long index = start >> PAGE_CACHE_SHIFT;
891 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
894 while (index <= end_index) {
895 page = find_get_page(tree->mapping, index);
897 set_page_writeback(page);
898 page_cache_release(page);
901 set_extent_writeback(tree, start, end, GFP_NOFS);
904 EXPORT_SYMBOL(set_range_writeback);
906 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
907 u64 *start_ret, u64 *end_ret, int bits)
909 struct rb_node *node;
910 struct extent_state *state;
913 spin_lock_irq(&tree->lock);
915 * this search will find all the extents that end after
918 node = tree_search(&tree->state, start);
919 if (!node || IS_ERR(node)) {
924 state = rb_entry(node, struct extent_state, rb_node);
925 if (state->end >= start && (state->state & bits)) {
926 *start_ret = state->start;
927 *end_ret = state->end;
931 node = rb_next(node);
936 spin_unlock_irq(&tree->lock);
939 EXPORT_SYMBOL(find_first_extent_bit);
941 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
942 u64 *start, u64 *end, u64 max_bytes)
944 struct rb_node *node;
945 struct extent_state *state;
946 u64 cur_start = *start;
950 spin_lock_irq(&tree->lock);
952 * this search will find all the extents that end after
956 node = tree_search(&tree->state, cur_start);
957 if (!node || IS_ERR(node)) {
963 state = rb_entry(node, struct extent_state, rb_node);
964 if (found && state->start != cur_start) {
967 if (!(state->state & EXTENT_DELALLOC)) {
973 struct extent_state *prev_state;
974 struct rb_node *prev_node = node;
976 prev_node = rb_prev(prev_node);
979 prev_state = rb_entry(prev_node,
982 if (!(prev_state->state & EXTENT_DELALLOC))
988 if (state->state & EXTENT_LOCKED) {
990 atomic_inc(&state->refs);
991 prepare_to_wait(&state->wq, &wait,
992 TASK_UNINTERRUPTIBLE);
993 spin_unlock_irq(&tree->lock);
995 spin_lock_irq(&tree->lock);
996 finish_wait(&state->wq, &wait);
997 free_extent_state(state);
1000 set_state_cb(tree, state, EXTENT_LOCKED);
1001 state->state |= EXTENT_LOCKED;
1003 *start = state->start;
1006 cur_start = state->end + 1;
1007 node = rb_next(node);
1010 total_bytes += state->end - state->start + 1;
1011 if (total_bytes >= max_bytes)
1015 spin_unlock_irq(&tree->lock);
1019 u64 count_range_bits(struct extent_io_tree *tree,
1020 u64 *start, u64 search_end, u64 max_bytes,
1023 struct rb_node *node;
1024 struct extent_state *state;
1025 u64 cur_start = *start;
1026 u64 total_bytes = 0;
1029 if (search_end <= cur_start) {
1030 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1035 spin_lock_irq(&tree->lock);
1036 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1037 total_bytes = tree->dirty_bytes;
1041 * this search will find all the extents that end after
1044 node = tree_search(&tree->state, cur_start);
1045 if (!node || IS_ERR(node)) {
1050 state = rb_entry(node, struct extent_state, rb_node);
1051 if (state->start > search_end)
1053 if (state->end >= cur_start && (state->state & bits)) {
1054 total_bytes += min(search_end, state->end) + 1 -
1055 max(cur_start, state->start);
1056 if (total_bytes >= max_bytes)
1059 *start = state->start;
1063 node = rb_next(node);
1068 spin_unlock_irq(&tree->lock);
1072 * helper function to lock both pages and extents in the tree.
1073 * pages must be locked first.
1075 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1077 unsigned long index = start >> PAGE_CACHE_SHIFT;
1078 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1082 while (index <= end_index) {
1083 page = grab_cache_page(tree->mapping, index);
1089 err = PTR_ERR(page);
1094 lock_extent(tree, start, end, GFP_NOFS);
1099 * we failed above in getting the page at 'index', so we undo here
1100 * up to but not including the page at 'index'
1103 index = start >> PAGE_CACHE_SHIFT;
1104 while (index < end_index) {
1105 page = find_get_page(tree->mapping, index);
1107 page_cache_release(page);
1112 EXPORT_SYMBOL(lock_range);
1115 * helper function to unlock both pages and extents in the tree.
1117 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1119 unsigned long index = start >> PAGE_CACHE_SHIFT;
1120 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1123 while (index <= end_index) {
1124 page = find_get_page(tree->mapping, index);
1126 page_cache_release(page);
1129 unlock_extent(tree, start, end, GFP_NOFS);
1132 EXPORT_SYMBOL(unlock_range);
1134 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1136 struct rb_node *node;
1137 struct extent_state *state;
1140 spin_lock_irq(&tree->lock);
1142 * this search will find all the extents that end after
1145 node = tree_search(&tree->state, start);
1146 if (!node || IS_ERR(node)) {
1150 state = rb_entry(node, struct extent_state, rb_node);
1151 if (state->start != start) {
1155 state->private = private;
1157 spin_unlock_irq(&tree->lock);
1161 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1163 struct rb_node *node;
1164 struct extent_state *state;
1167 spin_lock_irq(&tree->lock);
1169 * this search will find all the extents that end after
1172 node = tree_search(&tree->state, start);
1173 if (!node || IS_ERR(node)) {
1177 state = rb_entry(node, struct extent_state, rb_node);
1178 if (state->start != start) {
1182 *private = state->private;
1184 spin_unlock_irq(&tree->lock);
1189 * searches a range in the state tree for a given mask.
1190 * If 'filled' == 1, this returns 1 only if every extent in the tree
1191 * has the bits set. Otherwise, 1 is returned if any bit in the
1192 * range is found set.
1194 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1195 int bits, int filled)
1197 struct extent_state *state = NULL;
1198 struct rb_node *node;
1200 unsigned long flags;
1202 spin_lock_irqsave(&tree->lock, flags);
1203 node = tree_search(&tree->state, start);
1204 while (node && start <= end) {
1205 state = rb_entry(node, struct extent_state, rb_node);
1207 if (filled && state->start > start) {
1212 if (state->start > end)
1215 if (state->state & bits) {
1219 } else if (filled) {
1223 start = state->end + 1;
1226 node = rb_next(node);
1233 spin_unlock_irqrestore(&tree->lock, flags);
1236 EXPORT_SYMBOL(test_range_bit);
1239 * helper function to set a given page up to date if all the
1240 * extents in the tree for that page are up to date
1242 static int check_page_uptodate(struct extent_io_tree *tree,
1245 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1246 u64 end = start + PAGE_CACHE_SIZE - 1;
1247 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1248 SetPageUptodate(page);
1253 * helper function to unlock a page if all the extents in the tree
1254 * for that page are unlocked
1256 static int check_page_locked(struct extent_io_tree *tree,
1259 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1260 u64 end = start + PAGE_CACHE_SIZE - 1;
1261 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1267 * helper function to end page writeback if all the extents
1268 * in the tree for that page are done with writeback
1270 static int check_page_writeback(struct extent_io_tree *tree,
1273 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1274 u64 end = start + PAGE_CACHE_SIZE - 1;
1275 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1276 end_page_writeback(page);
1280 /* lots and lots of room for performance fixes in the end_bio funcs */
1283 * after a writepage IO is done, we need to:
1284 * clear the uptodate bits on error
1285 * clear the writeback bits in the extent tree for this IO
1286 * end_page_writeback if the page has no more pending IO
1288 * Scheduling is not allowed, so the extent state tree is expected
1289 * to have one and only one object corresponding to this IO.
1291 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1292 static void end_bio_extent_writepage(struct bio *bio, int err)
1294 static int end_bio_extent_writepage(struct bio *bio,
1295 unsigned int bytes_done, int err)
1298 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1299 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1300 struct extent_state *state = bio->bi_private;
1301 struct extent_io_tree *tree = state->tree;
1302 struct rb_node *node;
1307 unsigned long flags;
1309 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1314 struct page *page = bvec->bv_page;
1315 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1317 end = start + bvec->bv_len - 1;
1319 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1324 if (--bvec >= bio->bi_io_vec)
1325 prefetchw(&bvec->bv_page->flags);
1328 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1329 ClearPageUptodate(page);
1333 if (tree->ops && tree->ops->writepage_end_io_hook) {
1334 tree->ops->writepage_end_io_hook(page, start, end,
1339 * bios can get merged in funny ways, and so we need to
1340 * be careful with the state variable. We know the
1341 * state won't be merged with others because it has
1342 * WRITEBACK set, but we can't be sure each biovec is
1343 * sequential in the file. So, if our cached state
1344 * doesn't match the expected end, search the tree
1345 * for the correct one.
1348 spin_lock_irqsave(&tree->lock, flags);
1349 if (!state || state->end != end) {
1351 node = __tree_search(&tree->state, start, NULL, NULL);
1353 state = rb_entry(node, struct extent_state,
1355 if (state->end != end ||
1356 !(state->state & EXTENT_WRITEBACK))
1360 spin_unlock_irqrestore(&tree->lock, flags);
1361 clear_extent_writeback(tree, start,
1368 struct extent_state *clear = state;
1370 node = rb_prev(&state->rb_node);
1372 state = rb_entry(node,
1373 struct extent_state,
1379 clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1390 /* before releasing the lock, make sure the next state
1391 * variable has the expected bits set and corresponds
1392 * to the correct offsets in the file
1394 if (state && (state->end + 1 != start ||
1395 !state->state & EXTENT_WRITEBACK)) {
1398 spin_unlock_irqrestore(&tree->lock, flags);
1402 end_page_writeback(page);
1404 check_page_writeback(tree, page);
1405 } while (bvec >= bio->bi_io_vec);
1407 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1413 * after a readpage IO is done, we need to:
1414 * clear the uptodate bits on error
1415 * set the uptodate bits if things worked
1416 * set the page up to date if all extents in the tree are uptodate
1417 * clear the lock bit in the extent tree
1418 * unlock the page if there are no other extents locked for it
1420 * Scheduling is not allowed, so the extent state tree is expected
1421 * to have one and only one object corresponding to this IO.
1423 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1424 static void end_bio_extent_readpage(struct bio *bio, int err)
1426 static int end_bio_extent_readpage(struct bio *bio,
1427 unsigned int bytes_done, int err)
1430 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1431 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1432 struct extent_state *state = bio->bi_private;
1433 struct extent_io_tree *tree = state->tree;
1434 struct rb_node *node;
1438 unsigned long flags;
1442 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1448 struct page *page = bvec->bv_page;
1449 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1451 end = start + bvec->bv_len - 1;
1453 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1458 if (--bvec >= bio->bi_io_vec)
1459 prefetchw(&bvec->bv_page->flags);
1461 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1462 ret = tree->ops->readpage_end_io_hook(page, start, end,
1468 spin_lock_irqsave(&tree->lock, flags);
1469 if (!state || state->end != end) {
1471 node = __tree_search(&tree->state, start, NULL, NULL);
1473 state = rb_entry(node, struct extent_state,
1475 if (state->end != end ||
1476 !(state->state & EXTENT_LOCKED))
1480 spin_unlock_irqrestore(&tree->lock, flags);
1481 set_extent_uptodate(tree, start, end,
1483 unlock_extent(tree, start, end, GFP_ATOMIC);
1490 struct extent_state *clear = state;
1492 node = rb_prev(&state->rb_node);
1494 state = rb_entry(node,
1495 struct extent_state,
1500 set_state_cb(tree, clear, EXTENT_UPTODATE);
1501 clear->state |= EXTENT_UPTODATE;
1502 clear_state_bit(tree, clear, EXTENT_LOCKED,
1513 /* before releasing the lock, make sure the next state
1514 * variable has the expected bits set and corresponds
1515 * to the correct offsets in the file
1517 if (state && (state->end + 1 != start ||
1518 !state->state & EXTENT_WRITEBACK)) {
1521 spin_unlock_irqrestore(&tree->lock, flags);
1525 SetPageUptodate(page);
1527 ClearPageUptodate(page);
1533 check_page_uptodate(tree, page);
1535 ClearPageUptodate(page);
1538 check_page_locked(tree, page);
1540 } while (bvec >= bio->bi_io_vec);
1543 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1549 * IO done from prepare_write is pretty simple, we just unlock
1550 * the structs in the extent tree when done, and set the uptodate bits
1553 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1554 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1556 static int end_bio_extent_preparewrite(struct bio *bio,
1557 unsigned int bytes_done, int err)
1560 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1561 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1562 struct extent_state *state = bio->bi_private;
1563 struct extent_io_tree *tree = state->tree;
1567 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1573 struct page *page = bvec->bv_page;
1574 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1576 end = start + bvec->bv_len - 1;
1578 if (--bvec >= bio->bi_io_vec)
1579 prefetchw(&bvec->bv_page->flags);
1582 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1584 ClearPageUptodate(page);
1588 unlock_extent(tree, start, end, GFP_ATOMIC);
1590 } while (bvec >= bio->bi_io_vec);
1593 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1599 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1604 bio = bio_alloc(gfp_flags, nr_vecs);
1606 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1607 while (!bio && (nr_vecs /= 2))
1608 bio = bio_alloc(gfp_flags, nr_vecs);
1612 bio->bi_bdev = bdev;
1613 bio->bi_sector = first_sector;
1618 static int submit_one_bio(int rw, struct bio *bio)
1622 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1623 struct page *page = bvec->bv_page;
1624 struct extent_io_tree *tree = bio->bi_private;
1625 struct rb_node *node;
1626 struct extent_state *state;
1630 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1631 end = start + bvec->bv_len - 1;
1633 spin_lock_irq(&tree->lock);
1634 node = __tree_search(&tree->state, start, NULL, NULL);
1636 state = rb_entry(node, struct extent_state, rb_node);
1637 while(state->end < end) {
1638 node = rb_next(node);
1639 state = rb_entry(node, struct extent_state, rb_node);
1641 BUG_ON(state->end != end);
1642 spin_unlock_irq(&tree->lock);
1644 bio->bi_private = state;
1648 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1649 if (maxsector < bio->bi_sector) {
1650 printk("sector too large max %Lu got %llu\n", maxsector,
1651 (unsigned long long)bio->bi_sector);
1655 submit_bio(rw, bio);
1656 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1662 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1663 struct page *page, sector_t sector,
1664 size_t size, unsigned long offset,
1665 struct block_device *bdev,
1666 struct bio **bio_ret,
1667 unsigned long max_pages,
1668 bio_end_io_t end_io_func)
1674 if (bio_ret && *bio_ret) {
1676 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1677 bio_add_page(bio, page, size, offset) < size) {
1678 ret = submit_one_bio(rw, bio);
1684 nr = min_t(int, max_pages, bio_get_nr_vecs(bdev));
1685 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1687 printk("failed to allocate bio nr %d\n", nr);
1691 bio_add_page(bio, page, size, offset);
1692 bio->bi_end_io = end_io_func;
1693 bio->bi_private = tree;
1698 ret = submit_one_bio(rw, bio);
1704 void set_page_extent_mapped(struct page *page)
1706 if (!PagePrivate(page)) {
1707 SetPagePrivate(page);
1708 WARN_ON(!page->mapping->a_ops->invalidatepage);
1709 set_page_private(page, EXTENT_PAGE_PRIVATE);
1710 page_cache_get(page);
1714 void set_page_extent_head(struct page *page, unsigned long len)
1716 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1720 * basic readpage implementation. Locked extent state structs are inserted
1721 * into the tree that are removed when the IO is done (by the end_io
1724 static int __extent_read_full_page(struct extent_io_tree *tree,
1726 get_extent_t *get_extent,
1729 struct inode *inode = page->mapping->host;
1730 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1731 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1735 u64 last_byte = i_size_read(inode);
1739 struct extent_map *em;
1740 struct block_device *bdev;
1743 size_t page_offset = 0;
1745 size_t blocksize = inode->i_sb->s_blocksize;
1747 set_page_extent_mapped(page);
1750 lock_extent(tree, start, end, GFP_NOFS);
1752 while (cur <= end) {
1753 if (cur >= last_byte) {
1755 iosize = PAGE_CACHE_SIZE - page_offset;
1756 userpage = kmap_atomic(page, KM_USER0);
1757 memset(userpage + page_offset, 0, iosize);
1758 flush_dcache_page(page);
1759 kunmap_atomic(userpage, KM_USER0);
1760 set_extent_uptodate(tree, cur, cur + iosize - 1,
1762 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1765 em = get_extent(inode, page, page_offset, cur,
1767 if (IS_ERR(em) || !em) {
1769 unlock_extent(tree, cur, end, GFP_NOFS);
1773 extent_offset = cur - em->start;
1774 BUG_ON(extent_map_end(em) <= cur);
1777 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1778 cur_end = min(extent_map_end(em) - 1, end);
1779 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1780 sector = (em->block_start + extent_offset) >> 9;
1782 block_start = em->block_start;
1783 free_extent_map(em);
1786 /* we've found a hole, just zero and go on */
1787 if (block_start == EXTENT_MAP_HOLE) {
1789 userpage = kmap_atomic(page, KM_USER0);
1790 memset(userpage + page_offset, 0, iosize);
1791 flush_dcache_page(page);
1792 kunmap_atomic(userpage, KM_USER0);
1794 set_extent_uptodate(tree, cur, cur + iosize - 1,
1796 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1798 page_offset += iosize;
1801 /* the get_extent function already copied into the page */
1802 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1803 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1805 page_offset += iosize;
1808 /* we have an inline extent but it didn't get marked up
1809 * to date. Error out
1811 if (block_start == EXTENT_MAP_INLINE) {
1813 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1815 page_offset += iosize;
1820 if (tree->ops && tree->ops->readpage_io_hook) {
1821 ret = tree->ops->readpage_io_hook(page, cur,
1825 unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1827 ret = submit_extent_page(READ, tree, page,
1828 sector, iosize, page_offset,
1830 end_bio_extent_readpage);
1835 page_offset += iosize;
1839 if (!PageError(page))
1840 SetPageUptodate(page);
1846 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1847 get_extent_t *get_extent)
1849 struct bio *bio = NULL;
1852 ret = __extent_read_full_page(tree, page, get_extent, &bio);
1854 submit_one_bio(READ, bio);
1857 EXPORT_SYMBOL(extent_read_full_page);
1860 * the writepage semantics are similar to regular writepage. extent
1861 * records are inserted to lock ranges in the tree, and as dirty areas
1862 * are found, they are marked writeback. Then the lock bits are removed
1863 * and the end_io handler clears the writeback ranges
1865 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1868 struct inode *inode = page->mapping->host;
1869 struct extent_page_data *epd = data;
1870 struct extent_io_tree *tree = epd->tree;
1871 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1873 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1877 u64 last_byte = i_size_read(inode);
1881 struct extent_map *em;
1882 struct block_device *bdev;
1885 size_t page_offset = 0;
1887 loff_t i_size = i_size_read(inode);
1888 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1892 WARN_ON(!PageLocked(page));
1893 if (page->index > end_index) {
1894 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1899 if (page->index == end_index) {
1902 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1904 userpage = kmap_atomic(page, KM_USER0);
1905 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset);
1906 flush_dcache_page(page);
1907 kunmap_atomic(userpage, KM_USER0);
1910 set_page_extent_mapped(page);
1912 delalloc_start = start;
1914 while(delalloc_end < page_end) {
1915 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
1918 if (nr_delalloc == 0) {
1919 delalloc_start = delalloc_end + 1;
1922 tree->ops->fill_delalloc(inode, delalloc_start,
1924 clear_extent_bit(tree, delalloc_start,
1926 EXTENT_LOCKED | EXTENT_DELALLOC,
1928 delalloc_start = delalloc_end + 1;
1930 lock_extent(tree, start, page_end, GFP_NOFS);
1933 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1934 printk("found delalloc bits after lock_extent\n");
1937 if (last_byte <= start) {
1938 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1942 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1943 blocksize = inode->i_sb->s_blocksize;
1945 while (cur <= end) {
1946 if (cur >= last_byte) {
1947 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1950 em = epd->get_extent(inode, page, page_offset, cur,
1952 if (IS_ERR(em) || !em) {
1957 extent_offset = cur - em->start;
1958 BUG_ON(extent_map_end(em) <= cur);
1960 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1961 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1962 sector = (em->block_start + extent_offset) >> 9;
1964 block_start = em->block_start;
1965 free_extent_map(em);
1968 if (block_start == EXTENT_MAP_HOLE ||
1969 block_start == EXTENT_MAP_INLINE) {
1970 clear_extent_dirty(tree, cur,
1971 cur + iosize - 1, GFP_NOFS);
1973 page_offset += iosize;
1977 /* leave this out until we have a page_mkwrite call */
1978 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1981 page_offset += iosize;
1984 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1985 if (tree->ops && tree->ops->writepage_io_hook) {
1986 ret = tree->ops->writepage_io_hook(page, cur,
1994 unsigned long max_nr = end_index + 1;
1995 set_range_writeback(tree, cur, cur + iosize - 1);
1996 if (!PageWriteback(page)) {
1997 printk("warning page %lu not writeback, "
1998 "cur %llu end %llu\n", page->index,
1999 (unsigned long long)cur,
2000 (unsigned long long)end);
2003 ret = submit_extent_page(WRITE, tree, page, sector,
2004 iosize, page_offset, bdev,
2006 end_bio_extent_writepage);
2011 page_offset += iosize;
2016 /* make sure the mapping tag for page dirty gets cleared */
2017 set_page_writeback(page);
2018 end_page_writeback(page);
2020 unlock_extent(tree, start, page_end, GFP_NOFS);
2025 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2027 /* Taken directly from 2.6.23 for 2.6.18 back port */
2028 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2032 * write_cache_pages - walk the list of dirty pages of the given address space
2033 * and write all of them.
2034 * @mapping: address space structure to write
2035 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2036 * @writepage: function called for each page
2037 * @data: data passed to writepage function
2039 * If a page is already under I/O, write_cache_pages() skips it, even
2040 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2041 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2042 * and msync() need to guarantee that all the data which was dirty at the time
2043 * the call was made get new I/O started against them. If wbc->sync_mode is
2044 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2045 * existing IO to complete.
2047 static int write_cache_pages(struct address_space *mapping,
2048 struct writeback_control *wbc, writepage_t writepage,
2051 struct backing_dev_info *bdi = mapping->backing_dev_info;
2054 struct pagevec pvec;
2057 pgoff_t end; /* Inclusive */
2059 int range_whole = 0;
2061 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2062 wbc->encountered_congestion = 1;
2066 pagevec_init(&pvec, 0);
2067 if (wbc->range_cyclic) {
2068 index = mapping->writeback_index; /* Start from prev offset */
2071 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2072 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2073 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2078 while (!done && (index <= end) &&
2079 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2080 PAGECACHE_TAG_DIRTY,
2081 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2085 for (i = 0; i < nr_pages; i++) {
2086 struct page *page = pvec.pages[i];
2089 * At this point we hold neither mapping->tree_lock nor
2090 * lock on the page itself: the page may be truncated or
2091 * invalidated (changing page->mapping to NULL), or even
2092 * swizzled back from swapper_space to tmpfs file
2097 if (unlikely(page->mapping != mapping)) {
2102 if (!wbc->range_cyclic && page->index > end) {
2108 if (wbc->sync_mode != WB_SYNC_NONE)
2109 wait_on_page_writeback(page);
2111 if (PageWriteback(page) ||
2112 !clear_page_dirty_for_io(page)) {
2117 ret = (*writepage)(page, wbc, data);
2119 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2123 if (ret || (--(wbc->nr_to_write) <= 0))
2125 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2126 wbc->encountered_congestion = 1;
2130 pagevec_release(&pvec);
2133 if (!scanned && !done) {
2135 * We hit the last page and there is more work to be done: wrap
2136 * back to the start of the file
2142 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2143 mapping->writeback_index = index;
2148 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2149 get_extent_t *get_extent,
2150 struct writeback_control *wbc)
2153 struct address_space *mapping = page->mapping;
2154 struct extent_page_data epd = {
2157 .get_extent = get_extent,
2159 struct writeback_control wbc_writepages = {
2161 .sync_mode = WB_SYNC_NONE,
2162 .older_than_this = NULL,
2164 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2165 .range_end = (loff_t)-1,
2169 ret = __extent_writepage(page, wbc, &epd);
2171 write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2173 submit_one_bio(WRITE, epd.bio);
2177 EXPORT_SYMBOL(extent_write_full_page);
2180 int extent_writepages(struct extent_io_tree *tree,
2181 struct address_space *mapping,
2182 get_extent_t *get_extent,
2183 struct writeback_control *wbc)
2186 struct extent_page_data epd = {
2189 .get_extent = get_extent,
2192 ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2194 submit_one_bio(WRITE, epd.bio);
2198 EXPORT_SYMBOL(extent_writepages);
2200 int extent_readpages(struct extent_io_tree *tree,
2201 struct address_space *mapping,
2202 struct list_head *pages, unsigned nr_pages,
2203 get_extent_t get_extent)
2205 struct bio *bio = NULL;
2207 struct pagevec pvec;
2209 pagevec_init(&pvec, 0);
2210 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2211 struct page *page = list_entry(pages->prev, struct page, lru);
2213 prefetchw(&page->flags);
2214 list_del(&page->lru);
2216 * what we want to do here is call add_to_page_cache_lru,
2217 * but that isn't exported, so we reproduce it here
2219 if (!add_to_page_cache(page, mapping,
2220 page->index, GFP_KERNEL)) {
2222 /* open coding of lru_cache_add, also not exported */
2223 page_cache_get(page);
2224 if (!pagevec_add(&pvec, page))
2225 __pagevec_lru_add(&pvec);
2226 __extent_read_full_page(tree, page, get_extent, &bio);
2228 page_cache_release(page);
2230 if (pagevec_count(&pvec))
2231 __pagevec_lru_add(&pvec);
2232 BUG_ON(!list_empty(pages));
2234 submit_one_bio(READ, bio);
2237 EXPORT_SYMBOL(extent_readpages);
2240 * basic invalidatepage code, this waits on any locked or writeback
2241 * ranges corresponding to the page, and then deletes any extent state
2242 * records from the tree
2244 int extent_invalidatepage(struct extent_io_tree *tree,
2245 struct page *page, unsigned long offset)
2247 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2248 u64 end = start + PAGE_CACHE_SIZE - 1;
2249 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2251 start += (offset + blocksize -1) & ~(blocksize - 1);
2255 lock_extent(tree, start, end, GFP_NOFS);
2256 wait_on_extent_writeback(tree, start, end);
2257 clear_extent_bit(tree, start, end,
2258 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2262 EXPORT_SYMBOL(extent_invalidatepage);
2265 * simple commit_write call, set_range_dirty is used to mark both
2266 * the pages and the extent records as dirty
2268 int extent_commit_write(struct extent_io_tree *tree,
2269 struct inode *inode, struct page *page,
2270 unsigned from, unsigned to)
2272 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2274 set_page_extent_mapped(page);
2275 set_page_dirty(page);
2277 if (pos > inode->i_size) {
2278 i_size_write(inode, pos);
2279 mark_inode_dirty(inode);
2283 EXPORT_SYMBOL(extent_commit_write);
2285 int extent_prepare_write(struct extent_io_tree *tree,
2286 struct inode *inode, struct page *page,
2287 unsigned from, unsigned to, get_extent_t *get_extent)
2289 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2290 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2292 u64 orig_block_start;
2295 struct extent_map *em;
2296 unsigned blocksize = 1 << inode->i_blkbits;
2297 size_t page_offset = 0;
2298 size_t block_off_start;
2299 size_t block_off_end;
2305 set_page_extent_mapped(page);
2307 block_start = (page_start + from) & ~((u64)blocksize - 1);
2308 block_end = (page_start + to - 1) | (blocksize - 1);
2309 orig_block_start = block_start;
2311 lock_extent(tree, page_start, page_end, GFP_NOFS);
2312 while(block_start <= block_end) {
2313 em = get_extent(inode, page, page_offset, block_start,
2314 block_end - block_start + 1, 1);
2315 if (IS_ERR(em) || !em) {
2318 cur_end = min(block_end, extent_map_end(em) - 1);
2319 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2320 block_off_end = block_off_start + blocksize;
2321 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2323 if (!PageUptodate(page) && isnew &&
2324 (block_off_end > to || block_off_start < from)) {
2327 kaddr = kmap_atomic(page, KM_USER0);
2328 if (block_off_end > to)
2329 memset(kaddr + to, 0, block_off_end - to);
2330 if (block_off_start < from)
2331 memset(kaddr + block_off_start, 0,
2332 from - block_off_start);
2333 flush_dcache_page(page);
2334 kunmap_atomic(kaddr, KM_USER0);
2336 if ((em->block_start != EXTENT_MAP_HOLE &&
2337 em->block_start != EXTENT_MAP_INLINE) &&
2338 !isnew && !PageUptodate(page) &&
2339 (block_off_end > to || block_off_start < from) &&
2340 !test_range_bit(tree, block_start, cur_end,
2341 EXTENT_UPTODATE, 1)) {
2343 u64 extent_offset = block_start - em->start;
2345 sector = (em->block_start + extent_offset) >> 9;
2346 iosize = (cur_end - block_start + blocksize) &
2347 ~((u64)blocksize - 1);
2349 * we've already got the extent locked, but we
2350 * need to split the state such that our end_bio
2351 * handler can clear the lock.
2353 set_extent_bit(tree, block_start,
2354 block_start + iosize - 1,
2355 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2356 ret = submit_extent_page(READ, tree, page,
2357 sector, iosize, page_offset, em->bdev,
2359 end_bio_extent_preparewrite);
2361 block_start = block_start + iosize;
2363 set_extent_uptodate(tree, block_start, cur_end,
2365 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2366 block_start = cur_end + 1;
2368 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2369 free_extent_map(em);
2372 wait_extent_bit(tree, orig_block_start,
2373 block_end, EXTENT_LOCKED);
2375 check_page_uptodate(tree, page);
2377 /* FIXME, zero out newly allocated blocks on error */
2380 EXPORT_SYMBOL(extent_prepare_write);
2383 * a helper for releasepage. As long as there are no locked extents
2384 * in the range corresponding to the page, both state records and extent
2385 * map records are removed
2387 int try_release_extent_mapping(struct extent_map_tree *map,
2388 struct extent_io_tree *tree, struct page *page,
2391 struct extent_map *em;
2392 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2393 u64 end = start + PAGE_CACHE_SIZE - 1;
2394 u64 orig_start = start;
2397 if ((mask & __GFP_WAIT) &&
2398 page->mapping->host->i_size > 16 * 1024 * 1024) {
2399 while (start <= end) {
2400 spin_lock(&map->lock);
2401 em = lookup_extent_mapping(map, start, end);
2402 if (!em || IS_ERR(em)) {
2403 spin_unlock(&map->lock);
2406 if (em->start != start) {
2407 spin_unlock(&map->lock);
2408 free_extent_map(em);
2411 if (!test_range_bit(tree, em->start,
2412 extent_map_end(em) - 1,
2413 EXTENT_LOCKED, 0)) {
2414 remove_extent_mapping(map, em);
2415 /* once for the rb tree */
2416 free_extent_map(em);
2418 start = extent_map_end(em);
2419 spin_unlock(&map->lock);
2422 free_extent_map(em);
2425 if (test_range_bit(tree, orig_start, end, EXTENT_IOBITS, 0))
2428 if ((mask & GFP_NOFS) == GFP_NOFS)
2430 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2435 EXPORT_SYMBOL(try_release_extent_mapping);
2437 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2438 get_extent_t *get_extent)
2440 struct inode *inode = mapping->host;
2441 u64 start = iblock << inode->i_blkbits;
2442 sector_t sector = 0;
2443 struct extent_map *em;
2445 em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2446 if (!em || IS_ERR(em))
2449 if (em->block_start == EXTENT_MAP_INLINE ||
2450 em->block_start == EXTENT_MAP_HOLE)
2453 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2455 free_extent_map(em);
2459 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2461 if (list_empty(&eb->lru)) {
2462 extent_buffer_get(eb);
2463 list_add(&eb->lru, &tree->buffer_lru);
2465 if (tree->lru_size >= BUFFER_LRU_MAX) {
2466 struct extent_buffer *rm;
2467 rm = list_entry(tree->buffer_lru.prev,
2468 struct extent_buffer, lru);
2470 list_del_init(&rm->lru);
2471 free_extent_buffer(rm);
2474 list_move(&eb->lru, &tree->buffer_lru);
2477 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2478 u64 start, unsigned long len)
2480 struct list_head *lru = &tree->buffer_lru;
2481 struct list_head *cur = lru->next;
2482 struct extent_buffer *eb;
2484 if (list_empty(lru))
2488 eb = list_entry(cur, struct extent_buffer, lru);
2489 if (eb->start == start && eb->len == len) {
2490 extent_buffer_get(eb);
2494 } while (cur != lru);
2498 static inline unsigned long num_extent_pages(u64 start, u64 len)
2500 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2501 (start >> PAGE_CACHE_SHIFT);
2504 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2508 struct address_space *mapping;
2511 return eb->first_page;
2512 i += eb->start >> PAGE_CACHE_SHIFT;
2513 mapping = eb->first_page->mapping;
2514 read_lock_irq(&mapping->tree_lock);
2515 p = radix_tree_lookup(&mapping->page_tree, i);
2516 read_unlock_irq(&mapping->tree_lock);
2520 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2525 struct extent_buffer *eb = NULL;
2527 spin_lock(&tree->lru_lock);
2528 eb = find_lru(tree, start, len);
2529 spin_unlock(&tree->lru_lock);
2534 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2535 INIT_LIST_HEAD(&eb->lru);
2538 atomic_set(&eb->refs, 1);
2543 static void __free_extent_buffer(struct extent_buffer *eb)
2545 kmem_cache_free(extent_buffer_cache, eb);
2548 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2549 u64 start, unsigned long len,
2553 unsigned long num_pages = num_extent_pages(start, len);
2555 unsigned long index = start >> PAGE_CACHE_SHIFT;
2556 struct extent_buffer *eb;
2558 struct address_space *mapping = tree->mapping;
2561 eb = __alloc_extent_buffer(tree, start, len, mask);
2562 if (!eb || IS_ERR(eb))
2565 if (eb->flags & EXTENT_BUFFER_FILLED)
2569 eb->first_page = page0;
2572 page_cache_get(page0);
2573 mark_page_accessed(page0);
2574 set_page_extent_mapped(page0);
2575 WARN_ON(!PageUptodate(page0));
2576 set_page_extent_head(page0, len);
2580 for (; i < num_pages; i++, index++) {
2581 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2586 set_page_extent_mapped(p);
2587 mark_page_accessed(p);
2590 set_page_extent_head(p, len);
2592 set_page_private(p, EXTENT_PAGE_PRIVATE);
2594 if (!PageUptodate(p))
2599 eb->flags |= EXTENT_UPTODATE;
2600 eb->flags |= EXTENT_BUFFER_FILLED;
2603 spin_lock(&tree->lru_lock);
2605 spin_unlock(&tree->lru_lock);
2609 spin_lock(&tree->lru_lock);
2610 list_del_init(&eb->lru);
2611 spin_unlock(&tree->lru_lock);
2612 if (!atomic_dec_and_test(&eb->refs))
2614 for (index = 1; index < i; index++) {
2615 page_cache_release(extent_buffer_page(eb, index));
2618 page_cache_release(extent_buffer_page(eb, 0));
2619 __free_extent_buffer(eb);
2622 EXPORT_SYMBOL(alloc_extent_buffer);
2624 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2625 u64 start, unsigned long len,
2628 unsigned long num_pages = num_extent_pages(start, len);
2630 unsigned long index = start >> PAGE_CACHE_SHIFT;
2631 struct extent_buffer *eb;
2633 struct address_space *mapping = tree->mapping;
2636 eb = __alloc_extent_buffer(tree, start, len, mask);
2637 if (!eb || IS_ERR(eb))
2640 if (eb->flags & EXTENT_BUFFER_FILLED)
2643 for (i = 0; i < num_pages; i++, index++) {
2644 p = find_lock_page(mapping, index);
2648 set_page_extent_mapped(p);
2649 mark_page_accessed(p);
2653 set_page_extent_head(p, len);
2655 set_page_private(p, EXTENT_PAGE_PRIVATE);
2658 if (!PageUptodate(p))
2663 eb->flags |= EXTENT_UPTODATE;
2664 eb->flags |= EXTENT_BUFFER_FILLED;
2667 spin_lock(&tree->lru_lock);
2669 spin_unlock(&tree->lru_lock);
2672 spin_lock(&tree->lru_lock);
2673 list_del_init(&eb->lru);
2674 spin_unlock(&tree->lru_lock);
2675 if (!atomic_dec_and_test(&eb->refs))
2677 for (index = 1; index < i; index++) {
2678 page_cache_release(extent_buffer_page(eb, index));
2681 page_cache_release(extent_buffer_page(eb, 0));
2682 __free_extent_buffer(eb);
2685 EXPORT_SYMBOL(find_extent_buffer);
2687 void free_extent_buffer(struct extent_buffer *eb)
2690 unsigned long num_pages;
2695 if (!atomic_dec_and_test(&eb->refs))
2698 WARN_ON(!list_empty(&eb->lru));
2699 num_pages = num_extent_pages(eb->start, eb->len);
2701 for (i = 1; i < num_pages; i++) {
2702 page_cache_release(extent_buffer_page(eb, i));
2704 page_cache_release(extent_buffer_page(eb, 0));
2705 __free_extent_buffer(eb);
2707 EXPORT_SYMBOL(free_extent_buffer);
2709 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2710 struct extent_buffer *eb)
2714 unsigned long num_pages;
2717 u64 start = eb->start;
2718 u64 end = start + eb->len - 1;
2720 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2721 num_pages = num_extent_pages(eb->start, eb->len);
2723 for (i = 0; i < num_pages; i++) {
2724 page = extent_buffer_page(eb, i);
2727 set_page_extent_head(page, eb->len);
2729 set_page_private(page, EXTENT_PAGE_PRIVATE);
2732 * if we're on the last page or the first page and the
2733 * block isn't aligned on a page boundary, do extra checks
2734 * to make sure we don't clean page that is partially dirty
2736 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2737 ((i == num_pages - 1) &&
2738 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2739 start = (u64)page->index << PAGE_CACHE_SHIFT;
2740 end = start + PAGE_CACHE_SIZE - 1;
2741 if (test_range_bit(tree, start, end,
2747 clear_page_dirty_for_io(page);
2748 read_lock_irq(&page->mapping->tree_lock);
2749 if (!PageDirty(page)) {
2750 radix_tree_tag_clear(&page->mapping->page_tree,
2752 PAGECACHE_TAG_DIRTY);
2754 read_unlock_irq(&page->mapping->tree_lock);
2759 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2761 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2762 struct extent_buffer *eb)
2764 return wait_on_extent_writeback(tree, eb->start,
2765 eb->start + eb->len - 1);
2767 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2769 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2770 struct extent_buffer *eb)
2773 unsigned long num_pages;
2775 num_pages = num_extent_pages(eb->start, eb->len);
2776 for (i = 0; i < num_pages; i++) {
2777 struct page *page = extent_buffer_page(eb, i);
2778 /* writepage may need to do something special for the
2779 * first page, we have to make sure page->private is
2780 * properly set. releasepage may drop page->private
2781 * on us if the page isn't already dirty.
2785 set_page_extent_head(page, eb->len);
2786 } else if (PagePrivate(page) &&
2787 page->private != EXTENT_PAGE_PRIVATE) {
2789 set_page_extent_mapped(page);
2792 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2796 return set_extent_dirty(tree, eb->start,
2797 eb->start + eb->len - 1, GFP_NOFS);
2799 EXPORT_SYMBOL(set_extent_buffer_dirty);
2801 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2802 struct extent_buffer *eb)
2806 unsigned long num_pages;
2808 num_pages = num_extent_pages(eb->start, eb->len);
2810 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2812 for (i = 0; i < num_pages; i++) {
2813 page = extent_buffer_page(eb, i);
2814 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2815 ((i == num_pages - 1) &&
2816 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2817 check_page_uptodate(tree, page);
2820 SetPageUptodate(page);
2824 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2826 int extent_buffer_uptodate(struct extent_io_tree *tree,
2827 struct extent_buffer *eb)
2829 if (eb->flags & EXTENT_UPTODATE)
2831 return test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2832 EXTENT_UPTODATE, 1);
2834 EXPORT_SYMBOL(extent_buffer_uptodate);
2836 int read_extent_buffer_pages(struct extent_io_tree *tree,
2837 struct extent_buffer *eb,
2842 unsigned long start_i;
2846 unsigned long num_pages;
2848 if (eb->flags & EXTENT_UPTODATE)
2851 if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2852 EXTENT_UPTODATE, 1)) {
2857 WARN_ON(start < eb->start);
2858 start_i = (start >> PAGE_CACHE_SHIFT) -
2859 (eb->start >> PAGE_CACHE_SHIFT);
2864 num_pages = num_extent_pages(eb->start, eb->len);
2865 for (i = start_i; i < num_pages; i++) {
2866 page = extent_buffer_page(eb, i);
2867 if (PageUptodate(page)) {
2871 if (TestSetPageLocked(page)) {
2877 if (!PageUptodate(page)) {
2878 err = page->mapping->a_ops->readpage(NULL, page);
2890 for (i = start_i; i < num_pages; i++) {
2891 page = extent_buffer_page(eb, i);
2892 wait_on_page_locked(page);
2893 if (!PageUptodate(page)) {
2898 eb->flags |= EXTENT_UPTODATE;
2901 EXPORT_SYMBOL(read_extent_buffer_pages);
2903 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2904 unsigned long start,
2911 char *dst = (char *)dstv;
2912 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2913 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2914 unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2916 WARN_ON(start > eb->len);
2917 WARN_ON(start + len > eb->start + eb->len);
2919 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
2922 page = extent_buffer_page(eb, i);
2923 if (!PageUptodate(page)) {
2924 printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len);
2927 WARN_ON(!PageUptodate(page));
2929 cur = min(len, (PAGE_CACHE_SIZE - offset));
2930 kaddr = kmap_atomic(page, KM_USER1);
2931 memcpy(dst, kaddr + offset, cur);
2932 kunmap_atomic(kaddr, KM_USER1);
2940 EXPORT_SYMBOL(read_extent_buffer);
2942 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
2943 unsigned long min_len, char **token, char **map,
2944 unsigned long *map_start,
2945 unsigned long *map_len, int km)
2947 size_t offset = start & (PAGE_CACHE_SIZE - 1);
2950 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2951 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2952 unsigned long end_i = (start_offset + start + min_len - 1) >>
2959 offset = start_offset;
2963 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
2965 if (start + min_len > eb->len) {
2966 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
2970 p = extent_buffer_page(eb, i);
2971 WARN_ON(!PageUptodate(p));
2972 kaddr = kmap_atomic(p, km);
2974 *map = kaddr + offset;
2975 *map_len = PAGE_CACHE_SIZE - offset;
2978 EXPORT_SYMBOL(map_private_extent_buffer);
2980 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
2981 unsigned long min_len,
2982 char **token, char **map,
2983 unsigned long *map_start,
2984 unsigned long *map_len, int km)
2988 if (eb->map_token) {
2989 unmap_extent_buffer(eb, eb->map_token, km);
2990 eb->map_token = NULL;
2993 err = map_private_extent_buffer(eb, start, min_len, token, map,
2994 map_start, map_len, km);
2996 eb->map_token = *token;
2998 eb->map_start = *map_start;
2999 eb->map_len = *map_len;
3003 EXPORT_SYMBOL(map_extent_buffer);
3005 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3007 kunmap_atomic(token, km);
3009 EXPORT_SYMBOL(unmap_extent_buffer);
3011 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3012 unsigned long start,
3019 char *ptr = (char *)ptrv;
3020 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3021 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3024 WARN_ON(start > eb->len);
3025 WARN_ON(start + len > eb->start + eb->len);
3027 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3030 page = extent_buffer_page(eb, i);
3031 WARN_ON(!PageUptodate(page));
3033 cur = min(len, (PAGE_CACHE_SIZE - offset));
3035 kaddr = kmap_atomic(page, KM_USER0);
3036 ret = memcmp(ptr, kaddr + offset, cur);
3037 kunmap_atomic(kaddr, KM_USER0);
3048 EXPORT_SYMBOL(memcmp_extent_buffer);
3050 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3051 unsigned long start, unsigned long len)
3057 char *src = (char *)srcv;
3058 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3059 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3061 WARN_ON(start > eb->len);
3062 WARN_ON(start + len > eb->start + eb->len);
3064 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3067 page = extent_buffer_page(eb, i);
3068 WARN_ON(!PageUptodate(page));
3070 cur = min(len, PAGE_CACHE_SIZE - offset);
3071 kaddr = kmap_atomic(page, KM_USER1);
3072 memcpy(kaddr + offset, src, cur);
3073 kunmap_atomic(kaddr, KM_USER1);
3081 EXPORT_SYMBOL(write_extent_buffer);
3083 void memset_extent_buffer(struct extent_buffer *eb, char c,
3084 unsigned long start, unsigned long len)
3090 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3091 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3093 WARN_ON(start > eb->len);
3094 WARN_ON(start + len > eb->start + eb->len);
3096 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3099 page = extent_buffer_page(eb, i);
3100 WARN_ON(!PageUptodate(page));
3102 cur = min(len, PAGE_CACHE_SIZE - offset);
3103 kaddr = kmap_atomic(page, KM_USER0);
3104 memset(kaddr + offset, c, cur);
3105 kunmap_atomic(kaddr, KM_USER0);
3112 EXPORT_SYMBOL(memset_extent_buffer);
3114 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3115 unsigned long dst_offset, unsigned long src_offset,
3118 u64 dst_len = dst->len;
3123 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3124 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3126 WARN_ON(src->len != dst_len);
3128 offset = (start_offset + dst_offset) &
3129 ((unsigned long)PAGE_CACHE_SIZE - 1);
3132 page = extent_buffer_page(dst, i);
3133 WARN_ON(!PageUptodate(page));
3135 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3137 kaddr = kmap_atomic(page, KM_USER0);
3138 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3139 kunmap_atomic(kaddr, KM_USER0);
3147 EXPORT_SYMBOL(copy_extent_buffer);
3149 static void move_pages(struct page *dst_page, struct page *src_page,
3150 unsigned long dst_off, unsigned long src_off,
3153 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3154 if (dst_page == src_page) {
3155 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3157 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3158 char *p = dst_kaddr + dst_off + len;
3159 char *s = src_kaddr + src_off + len;
3164 kunmap_atomic(src_kaddr, KM_USER1);
3166 kunmap_atomic(dst_kaddr, KM_USER0);
3169 static void copy_pages(struct page *dst_page, struct page *src_page,
3170 unsigned long dst_off, unsigned long src_off,
3173 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3176 if (dst_page != src_page)
3177 src_kaddr = kmap_atomic(src_page, KM_USER1);
3179 src_kaddr = dst_kaddr;
3181 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3182 kunmap_atomic(dst_kaddr, KM_USER0);
3183 if (dst_page != src_page)
3184 kunmap_atomic(src_kaddr, KM_USER1);
3187 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3188 unsigned long src_offset, unsigned long len)
3191 size_t dst_off_in_page;
3192 size_t src_off_in_page;
3193 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3194 unsigned long dst_i;
3195 unsigned long src_i;
3197 if (src_offset + len > dst->len) {
3198 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3199 src_offset, len, dst->len);
3202 if (dst_offset + len > dst->len) {
3203 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3204 dst_offset, len, dst->len);
3209 dst_off_in_page = (start_offset + dst_offset) &
3210 ((unsigned long)PAGE_CACHE_SIZE - 1);
3211 src_off_in_page = (start_offset + src_offset) &
3212 ((unsigned long)PAGE_CACHE_SIZE - 1);
3214 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3215 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3217 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3219 cur = min_t(unsigned long, cur,
3220 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3222 copy_pages(extent_buffer_page(dst, dst_i),
3223 extent_buffer_page(dst, src_i),
3224 dst_off_in_page, src_off_in_page, cur);
3231 EXPORT_SYMBOL(memcpy_extent_buffer);
3233 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3234 unsigned long src_offset, unsigned long len)
3237 size_t dst_off_in_page;
3238 size_t src_off_in_page;
3239 unsigned long dst_end = dst_offset + len - 1;
3240 unsigned long src_end = src_offset + len - 1;
3241 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3242 unsigned long dst_i;
3243 unsigned long src_i;
3245 if (src_offset + len > dst->len) {
3246 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3247 src_offset, len, dst->len);
3250 if (dst_offset + len > dst->len) {
3251 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3252 dst_offset, len, dst->len);
3255 if (dst_offset < src_offset) {
3256 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3260 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3261 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3263 dst_off_in_page = (start_offset + dst_end) &
3264 ((unsigned long)PAGE_CACHE_SIZE - 1);
3265 src_off_in_page = (start_offset + src_end) &
3266 ((unsigned long)PAGE_CACHE_SIZE - 1);
3268 cur = min_t(unsigned long, len, src_off_in_page + 1);
3269 cur = min(cur, dst_off_in_page + 1);
3270 move_pages(extent_buffer_page(dst, dst_i),
3271 extent_buffer_page(dst, src_i),
3272 dst_off_in_page - cur + 1,
3273 src_off_in_page - cur + 1, cur);
3280 EXPORT_SYMBOL(memmove_extent_buffer);