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);
34 static spinlock_t leak_lock = SPIN_LOCK_UNLOCKED;
37 #define BUFFER_LRU_MAX 64
42 struct rb_node rb_node;
45 struct extent_page_data {
47 struct extent_io_tree *tree;
48 get_extent_t *get_extent;
51 int __init extent_io_init(void)
53 extent_state_cache = btrfs_cache_create("extent_state",
54 sizeof(struct extent_state), 0,
56 if (!extent_state_cache)
59 extent_buffer_cache = btrfs_cache_create("extent_buffers",
60 sizeof(struct extent_buffer), 0,
62 if (!extent_buffer_cache)
63 goto free_state_cache;
67 kmem_cache_destroy(extent_state_cache);
71 void extent_io_exit(void)
73 struct extent_state *state;
74 struct extent_buffer *eb;
76 while (!list_empty(&states)) {
77 state = list_entry(states.next, struct extent_state, leak_list);
78 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));
79 list_del(&state->leak_list);
80 kmem_cache_free(extent_state_cache, state);
84 while (!list_empty(&buffers)) {
85 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
86 printk("buffer leak start %Lu len %lu refs %d\n", eb->start, eb->len, atomic_read(&eb->refs));
87 list_del(&eb->leak_list);
88 kmem_cache_free(extent_buffer_cache, eb);
90 if (extent_state_cache)
91 kmem_cache_destroy(extent_state_cache);
92 if (extent_buffer_cache)
93 kmem_cache_destroy(extent_buffer_cache);
96 void extent_io_tree_init(struct extent_io_tree *tree,
97 struct address_space *mapping, gfp_t mask)
99 tree->state.rb_node = NULL;
100 tree->buffer.rb_node = NULL;
102 tree->dirty_bytes = 0;
103 spin_lock_init(&tree->lock);
104 spin_lock_init(&tree->buffer_lock);
105 tree->mapping = mapping;
107 EXPORT_SYMBOL(extent_io_tree_init);
109 struct extent_state *alloc_extent_state(gfp_t mask)
111 struct extent_state *state;
116 state = kmem_cache_alloc(extent_state_cache, mask);
123 spin_lock_irqsave(&leak_lock, flags);
124 list_add(&state->leak_list, &states);
125 spin_unlock_irqrestore(&leak_lock, flags);
127 atomic_set(&state->refs, 1);
128 init_waitqueue_head(&state->wq);
131 EXPORT_SYMBOL(alloc_extent_state);
133 void free_extent_state(struct extent_state *state)
137 if (atomic_dec_and_test(&state->refs)) {
141 WARN_ON(state->tree);
143 spin_lock_irqsave(&leak_lock, flags);
144 list_del(&state->leak_list);
145 spin_unlock_irqrestore(&leak_lock, flags);
147 kmem_cache_free(extent_state_cache, state);
150 EXPORT_SYMBOL(free_extent_state);
152 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
153 struct rb_node *node)
155 struct rb_node ** p = &root->rb_node;
156 struct rb_node * parent = NULL;
157 struct tree_entry *entry;
161 entry = rb_entry(parent, struct tree_entry, rb_node);
163 if (offset < entry->start)
165 else if (offset > entry->end)
171 entry = rb_entry(node, struct tree_entry, rb_node);
172 rb_link_node(node, parent, p);
173 rb_insert_color(node, root);
177 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
178 struct rb_node **prev_ret,
179 struct rb_node **next_ret)
181 struct rb_root *root = &tree->state;
182 struct rb_node * n = root->rb_node;
183 struct rb_node *prev = NULL;
184 struct rb_node *orig_prev = NULL;
185 struct tree_entry *entry;
186 struct tree_entry *prev_entry = NULL;
189 entry = rb_entry(n, struct tree_entry, rb_node);
193 if (offset < entry->start)
195 else if (offset > entry->end)
204 while(prev && offset > prev_entry->end) {
205 prev = rb_next(prev);
206 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
213 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
214 while(prev && offset < prev_entry->start) {
215 prev = rb_prev(prev);
216 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
223 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
226 struct rb_node *prev = NULL;
229 ret = __etree_search(tree, offset, &prev, NULL);
236 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
237 u64 offset, struct rb_node *node)
239 struct rb_root *root = &tree->buffer;
240 struct rb_node ** p = &root->rb_node;
241 struct rb_node * parent = NULL;
242 struct extent_buffer *eb;
246 eb = rb_entry(parent, struct extent_buffer, rb_node);
248 if (offset < eb->start)
250 else if (offset > eb->start)
256 rb_link_node(node, parent, p);
257 rb_insert_color(node, root);
261 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
264 struct rb_root *root = &tree->buffer;
265 struct rb_node * n = root->rb_node;
266 struct extent_buffer *eb;
269 eb = rb_entry(n, struct extent_buffer, rb_node);
270 if (offset < eb->start)
272 else if (offset > eb->start)
281 * utility function to look for merge candidates inside a given range.
282 * Any extents with matching state are merged together into a single
283 * extent in the tree. Extents with EXTENT_IO in their state field
284 * are not merged because the end_io handlers need to be able to do
285 * operations on them without sleeping (or doing allocations/splits).
287 * This should be called with the tree lock held.
289 static int merge_state(struct extent_io_tree *tree,
290 struct extent_state *state)
292 struct extent_state *other;
293 struct rb_node *other_node;
295 if (state->state & EXTENT_IOBITS)
298 other_node = rb_prev(&state->rb_node);
300 other = rb_entry(other_node, struct extent_state, rb_node);
301 if (other->end == state->start - 1 &&
302 other->state == state->state) {
303 state->start = other->start;
305 rb_erase(&other->rb_node, &tree->state);
306 free_extent_state(other);
309 other_node = rb_next(&state->rb_node);
311 other = rb_entry(other_node, struct extent_state, rb_node);
312 if (other->start == state->end + 1 &&
313 other->state == state->state) {
314 other->start = state->start;
316 rb_erase(&state->rb_node, &tree->state);
317 free_extent_state(state);
323 static void set_state_cb(struct extent_io_tree *tree,
324 struct extent_state *state,
327 if (tree->ops && tree->ops->set_bit_hook) {
328 tree->ops->set_bit_hook(tree->mapping->host, state->start,
329 state->end, state->state, bits);
333 static void clear_state_cb(struct extent_io_tree *tree,
334 struct extent_state *state,
337 if (tree->ops && tree->ops->set_bit_hook) {
338 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
339 state->end, state->state, bits);
344 * insert an extent_state struct into the tree. 'bits' are set on the
345 * struct before it is inserted.
347 * This may return -EEXIST if the extent is already there, in which case the
348 * state struct is freed.
350 * The tree lock is not taken internally. This is a utility function and
351 * probably isn't what you want to call (see set/clear_extent_bit).
353 static int insert_state(struct extent_io_tree *tree,
354 struct extent_state *state, u64 start, u64 end,
357 struct rb_node *node;
360 printk("end < start %Lu %Lu\n", end, start);
363 if (bits & EXTENT_DIRTY)
364 tree->dirty_bytes += end - start + 1;
365 set_state_cb(tree, state, bits);
366 state->state |= bits;
367 state->start = start;
369 node = tree_insert(&tree->state, end, &state->rb_node);
371 struct extent_state *found;
372 found = rb_entry(node, struct extent_state, rb_node);
373 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
374 free_extent_state(state);
378 merge_state(tree, state);
383 * split a given extent state struct in two, inserting the preallocated
384 * struct 'prealloc' as the newly created second half. 'split' indicates an
385 * offset inside 'orig' where it should be split.
388 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
389 * are two extent state structs in the tree:
390 * prealloc: [orig->start, split - 1]
391 * orig: [ split, orig->end ]
393 * The tree locks are not taken by this function. They need to be held
396 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
397 struct extent_state *prealloc, u64 split)
399 struct rb_node *node;
400 prealloc->start = orig->start;
401 prealloc->end = split - 1;
402 prealloc->state = orig->state;
405 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
407 struct extent_state *found;
408 found = rb_entry(node, struct extent_state, rb_node);
409 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
410 free_extent_state(prealloc);
413 prealloc->tree = tree;
418 * utility function to clear some bits in an extent state struct.
419 * it will optionally wake up any one waiting on this state (wake == 1), or
420 * forcibly remove the state from the tree (delete == 1).
422 * If no bits are set on the state struct after clearing things, the
423 * struct is freed and removed from the tree
425 static int clear_state_bit(struct extent_io_tree *tree,
426 struct extent_state *state, int bits, int wake,
429 int ret = state->state & bits;
431 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
432 u64 range = state->end - state->start + 1;
433 WARN_ON(range > tree->dirty_bytes);
434 tree->dirty_bytes -= range;
436 clear_state_cb(tree, state, bits);
437 state->state &= ~bits;
440 if (delete || state->state == 0) {
442 clear_state_cb(tree, state, state->state);
443 rb_erase(&state->rb_node, &tree->state);
445 free_extent_state(state);
450 merge_state(tree, state);
456 * clear some bits on a range in the tree. This may require splitting
457 * or inserting elements in the tree, so the gfp mask is used to
458 * indicate which allocations or sleeping are allowed.
460 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
461 * the given range from the tree regardless of state (ie for truncate).
463 * the range [start, end] is inclusive.
465 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
466 * bits were already set, or zero if none of the bits were already set.
468 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
469 int bits, int wake, int delete, gfp_t mask)
471 struct extent_state *state;
472 struct extent_state *prealloc = NULL;
473 struct rb_node *node;
479 if (!prealloc && (mask & __GFP_WAIT)) {
480 prealloc = alloc_extent_state(mask);
485 spin_lock_irqsave(&tree->lock, flags);
487 * this search will find the extents that end after
490 node = tree_search(tree, start);
493 state = rb_entry(node, struct extent_state, rb_node);
494 if (state->start > end)
496 WARN_ON(state->end < start);
499 * | ---- desired range ---- |
501 * | ------------- state -------------- |
503 * We need to split the extent we found, and may flip
504 * bits on second half.
506 * If the extent we found extends past our range, we
507 * just split and search again. It'll get split again
508 * the next time though.
510 * If the extent we found is inside our range, we clear
511 * the desired bit on it.
514 if (state->start < start) {
516 prealloc = alloc_extent_state(GFP_ATOMIC);
517 err = split_state(tree, state, prealloc, start);
518 BUG_ON(err == -EEXIST);
522 if (state->end <= end) {
523 start = state->end + 1;
524 set |= clear_state_bit(tree, state, bits,
527 start = state->start;
532 * | ---- desired range ---- |
534 * We need to split the extent, and clear the bit
537 if (state->start <= end && state->end > end) {
539 prealloc = alloc_extent_state(GFP_ATOMIC);
540 err = split_state(tree, state, prealloc, end + 1);
541 BUG_ON(err == -EEXIST);
545 set |= clear_state_bit(tree, prealloc, bits,
551 start = state->end + 1;
552 set |= clear_state_bit(tree, state, bits, wake, delete);
556 spin_unlock_irqrestore(&tree->lock, flags);
558 free_extent_state(prealloc);
565 spin_unlock_irqrestore(&tree->lock, flags);
566 if (mask & __GFP_WAIT)
570 EXPORT_SYMBOL(clear_extent_bit);
572 static int wait_on_state(struct extent_io_tree *tree,
573 struct extent_state *state)
576 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
577 spin_unlock_irq(&tree->lock);
579 spin_lock_irq(&tree->lock);
580 finish_wait(&state->wq, &wait);
585 * waits for one or more bits to clear on a range in the state tree.
586 * The range [start, end] is inclusive.
587 * The tree lock is taken by this function
589 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
591 struct extent_state *state;
592 struct rb_node *node;
594 spin_lock_irq(&tree->lock);
598 * this search will find all the extents that end after
601 node = tree_search(tree, start);
605 state = rb_entry(node, struct extent_state, rb_node);
607 if (state->start > end)
610 if (state->state & bits) {
611 start = state->start;
612 atomic_inc(&state->refs);
613 wait_on_state(tree, state);
614 free_extent_state(state);
617 start = state->end + 1;
622 if (need_resched()) {
623 spin_unlock_irq(&tree->lock);
625 spin_lock_irq(&tree->lock);
629 spin_unlock_irq(&tree->lock);
632 EXPORT_SYMBOL(wait_extent_bit);
634 static void set_state_bits(struct extent_io_tree *tree,
635 struct extent_state *state,
638 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
639 u64 range = state->end - state->start + 1;
640 tree->dirty_bytes += range;
642 set_state_cb(tree, state, bits);
643 state->state |= bits;
647 * set some bits on a range in the tree. This may require allocations
648 * or sleeping, so the gfp mask is used to indicate what is allowed.
650 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
651 * range already has the desired bits set. The start of the existing
652 * range is returned in failed_start in this case.
654 * [start, end] is inclusive
655 * This takes the tree lock.
657 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
658 int exclusive, u64 *failed_start, gfp_t mask)
660 struct extent_state *state;
661 struct extent_state *prealloc = NULL;
662 struct rb_node *node;
669 if (!prealloc && (mask & __GFP_WAIT)) {
670 prealloc = alloc_extent_state(mask);
675 spin_lock_irqsave(&tree->lock, flags);
677 * this search will find all the extents that end after
680 node = tree_search(tree, start);
682 err = insert_state(tree, prealloc, start, end, bits);
684 BUG_ON(err == -EEXIST);
688 state = rb_entry(node, struct extent_state, rb_node);
689 last_start = state->start;
690 last_end = state->end;
693 * | ---- desired range ---- |
696 * Just lock what we found and keep going
698 if (state->start == start && state->end <= end) {
699 set = state->state & bits;
700 if (set && exclusive) {
701 *failed_start = state->start;
705 set_state_bits(tree, state, bits);
706 start = state->end + 1;
707 merge_state(tree, state);
712 * | ---- desired range ---- |
715 * | ------------- state -------------- |
717 * We need to split the extent we found, and may flip bits on
720 * If the extent we found extends past our
721 * range, we just split and search again. It'll get split
722 * again the next time though.
724 * If the extent we found is inside our range, we set the
727 if (state->start < start) {
728 set = state->state & bits;
729 if (exclusive && set) {
730 *failed_start = start;
734 err = split_state(tree, state, prealloc, start);
735 BUG_ON(err == -EEXIST);
739 if (state->end <= end) {
740 set_state_bits(tree, state, bits);
741 start = state->end + 1;
742 merge_state(tree, state);
744 start = state->start;
749 * | ---- desired range ---- |
750 * | state | or | state |
752 * There's a hole, we need to insert something in it and
753 * ignore the extent we found.
755 if (state->start > start) {
757 if (end < last_start)
760 this_end = last_start -1;
761 err = insert_state(tree, prealloc, start, this_end,
764 BUG_ON(err == -EEXIST);
767 start = this_end + 1;
771 * | ---- desired range ---- |
773 * We need to split the extent, and set the bit
776 if (state->start <= end && state->end > end) {
777 set = state->state & bits;
778 if (exclusive && set) {
779 *failed_start = start;
783 err = split_state(tree, state, prealloc, end + 1);
784 BUG_ON(err == -EEXIST);
786 set_state_bits(tree, prealloc, bits);
787 merge_state(tree, prealloc);
795 spin_unlock_irqrestore(&tree->lock, flags);
797 free_extent_state(prealloc);
804 spin_unlock_irqrestore(&tree->lock, flags);
805 if (mask & __GFP_WAIT)
809 EXPORT_SYMBOL(set_extent_bit);
811 /* wrappers around set/clear extent bit */
812 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
815 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
818 EXPORT_SYMBOL(set_extent_dirty);
820 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
823 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
825 EXPORT_SYMBOL(set_extent_ordered);
827 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
828 int bits, gfp_t mask)
830 return set_extent_bit(tree, start, end, bits, 0, NULL,
833 EXPORT_SYMBOL(set_extent_bits);
835 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
836 int bits, gfp_t mask)
838 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
840 EXPORT_SYMBOL(clear_extent_bits);
842 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
845 return set_extent_bit(tree, start, end,
846 EXTENT_DELALLOC | EXTENT_DIRTY,
849 EXPORT_SYMBOL(set_extent_delalloc);
851 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
854 return clear_extent_bit(tree, start, end,
855 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
857 EXPORT_SYMBOL(clear_extent_dirty);
859 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
862 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
864 EXPORT_SYMBOL(clear_extent_ordered);
866 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
869 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
872 EXPORT_SYMBOL(set_extent_new);
874 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
877 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
879 EXPORT_SYMBOL(clear_extent_new);
881 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
884 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
887 EXPORT_SYMBOL(set_extent_uptodate);
889 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
892 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
894 EXPORT_SYMBOL(clear_extent_uptodate);
896 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
899 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
902 EXPORT_SYMBOL(set_extent_writeback);
904 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
907 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
909 EXPORT_SYMBOL(clear_extent_writeback);
911 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
913 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
915 EXPORT_SYMBOL(wait_on_extent_writeback);
917 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
922 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
923 &failed_start, mask);
924 if (err == -EEXIST && (mask & __GFP_WAIT)) {
925 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
926 start = failed_start;
930 WARN_ON(start > end);
934 EXPORT_SYMBOL(lock_extent);
936 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
939 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
941 EXPORT_SYMBOL(unlock_extent);
944 * helper function to set pages and extents in the tree dirty
946 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
948 unsigned long index = start >> PAGE_CACHE_SHIFT;
949 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
952 while (index <= end_index) {
953 page = find_get_page(tree->mapping, index);
955 __set_page_dirty_nobuffers(page);
956 page_cache_release(page);
959 set_extent_dirty(tree, start, end, GFP_NOFS);
962 EXPORT_SYMBOL(set_range_dirty);
965 * helper function to set both pages and extents in the tree writeback
967 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
969 unsigned long index = start >> PAGE_CACHE_SHIFT;
970 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
973 while (index <= end_index) {
974 page = find_get_page(tree->mapping, index);
976 set_page_writeback(page);
977 page_cache_release(page);
980 set_extent_writeback(tree, start, end, GFP_NOFS);
983 EXPORT_SYMBOL(set_range_writeback);
985 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
986 u64 *start_ret, u64 *end_ret, int bits)
988 struct rb_node *node;
989 struct extent_state *state;
992 spin_lock_irq(&tree->lock);
994 * this search will find all the extents that end after
997 node = tree_search(tree, start);
1003 state = rb_entry(node, struct extent_state, rb_node);
1004 if (state->end >= start && (state->state & bits)) {
1005 *start_ret = state->start;
1006 *end_ret = state->end;
1010 node = rb_next(node);
1015 spin_unlock_irq(&tree->lock);
1018 EXPORT_SYMBOL(find_first_extent_bit);
1020 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1021 u64 start, int bits)
1023 struct rb_node *node;
1024 struct extent_state *state;
1027 * this search will find all the extents that end after
1030 node = tree_search(tree, start);
1036 state = rb_entry(node, struct extent_state, rb_node);
1037 if (state->end >= start && (state->state & bits)) {
1040 node = rb_next(node);
1047 EXPORT_SYMBOL(find_first_extent_bit_state);
1049 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1050 u64 *start, u64 *end, u64 max_bytes)
1052 struct rb_node *node;
1053 struct extent_state *state;
1054 u64 cur_start = *start;
1056 u64 total_bytes = 0;
1058 spin_lock_irq(&tree->lock);
1060 * this search will find all the extents that end after
1064 node = tree_search(tree, cur_start);
1072 state = rb_entry(node, struct extent_state, rb_node);
1073 if (found && state->start != cur_start) {
1076 if (!(state->state & EXTENT_DELALLOC)) {
1082 struct extent_state *prev_state;
1083 struct rb_node *prev_node = node;
1085 prev_node = rb_prev(prev_node);
1088 prev_state = rb_entry(prev_node,
1089 struct extent_state,
1091 if (!(prev_state->state & EXTENT_DELALLOC))
1097 if (state->state & EXTENT_LOCKED) {
1099 atomic_inc(&state->refs);
1100 prepare_to_wait(&state->wq, &wait,
1101 TASK_UNINTERRUPTIBLE);
1102 spin_unlock_irq(&tree->lock);
1104 spin_lock_irq(&tree->lock);
1105 finish_wait(&state->wq, &wait);
1106 free_extent_state(state);
1109 set_state_cb(tree, state, EXTENT_LOCKED);
1110 state->state |= EXTENT_LOCKED;
1112 *start = state->start;
1115 cur_start = state->end + 1;
1116 node = rb_next(node);
1119 total_bytes += state->end - state->start + 1;
1120 if (total_bytes >= max_bytes)
1124 spin_unlock_irq(&tree->lock);
1128 u64 count_range_bits(struct extent_io_tree *tree,
1129 u64 *start, u64 search_end, u64 max_bytes,
1132 struct rb_node *node;
1133 struct extent_state *state;
1134 u64 cur_start = *start;
1135 u64 total_bytes = 0;
1138 if (search_end <= cur_start) {
1139 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1144 spin_lock_irq(&tree->lock);
1145 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1146 total_bytes = tree->dirty_bytes;
1150 * this search will find all the extents that end after
1153 node = tree_search(tree, cur_start);
1159 state = rb_entry(node, struct extent_state, rb_node);
1160 if (state->start > search_end)
1162 if (state->end >= cur_start && (state->state & bits)) {
1163 total_bytes += min(search_end, state->end) + 1 -
1164 max(cur_start, state->start);
1165 if (total_bytes >= max_bytes)
1168 *start = state->start;
1172 node = rb_next(node);
1177 spin_unlock_irq(&tree->lock);
1181 * helper function to lock both pages and extents in the tree.
1182 * pages must be locked first.
1184 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1186 unsigned long index = start >> PAGE_CACHE_SHIFT;
1187 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1191 while (index <= end_index) {
1192 page = grab_cache_page(tree->mapping, index);
1198 err = PTR_ERR(page);
1203 lock_extent(tree, start, end, GFP_NOFS);
1208 * we failed above in getting the page at 'index', so we undo here
1209 * up to but not including the page at 'index'
1212 index = start >> PAGE_CACHE_SHIFT;
1213 while (index < end_index) {
1214 page = find_get_page(tree->mapping, index);
1216 page_cache_release(page);
1221 EXPORT_SYMBOL(lock_range);
1224 * helper function to unlock both pages and extents in the tree.
1226 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1228 unsigned long index = start >> PAGE_CACHE_SHIFT;
1229 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1232 while (index <= end_index) {
1233 page = find_get_page(tree->mapping, index);
1235 page_cache_release(page);
1238 unlock_extent(tree, start, end, GFP_NOFS);
1241 EXPORT_SYMBOL(unlock_range);
1243 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1245 struct rb_node *node;
1246 struct extent_state *state;
1249 spin_lock_irq(&tree->lock);
1251 * this search will find all the extents that end after
1254 node = tree_search(tree, start);
1259 state = rb_entry(node, struct extent_state, rb_node);
1260 if (state->start != start) {
1264 state->private = private;
1266 spin_unlock_irq(&tree->lock);
1270 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1272 struct rb_node *node;
1273 struct extent_state *state;
1276 spin_lock_irq(&tree->lock);
1278 * this search will find all the extents that end after
1281 node = tree_search(tree, start);
1286 state = rb_entry(node, struct extent_state, rb_node);
1287 if (state->start != start) {
1291 *private = state->private;
1293 spin_unlock_irq(&tree->lock);
1298 * searches a range in the state tree for a given mask.
1299 * If 'filled' == 1, this returns 1 only if every extent in the tree
1300 * has the bits set. Otherwise, 1 is returned if any bit in the
1301 * range is found set.
1303 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1304 int bits, int filled)
1306 struct extent_state *state = NULL;
1307 struct rb_node *node;
1309 unsigned long flags;
1311 spin_lock_irqsave(&tree->lock, flags);
1312 node = tree_search(tree, start);
1313 while (node && start <= end) {
1314 state = rb_entry(node, struct extent_state, rb_node);
1316 if (filled && state->start > start) {
1321 if (state->start > end)
1324 if (state->state & bits) {
1328 } else if (filled) {
1332 start = state->end + 1;
1335 node = rb_next(node);
1342 spin_unlock_irqrestore(&tree->lock, flags);
1345 EXPORT_SYMBOL(test_range_bit);
1348 * helper function to set a given page up to date if all the
1349 * extents in the tree for that page are up to date
1351 static int check_page_uptodate(struct extent_io_tree *tree,
1354 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1355 u64 end = start + PAGE_CACHE_SIZE - 1;
1356 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1357 SetPageUptodate(page);
1362 * helper function to unlock a page if all the extents in the tree
1363 * for that page are unlocked
1365 static int check_page_locked(struct extent_io_tree *tree,
1368 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1369 u64 end = start + PAGE_CACHE_SIZE - 1;
1370 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1376 * helper function to end page writeback if all the extents
1377 * in the tree for that page are done with writeback
1379 static int check_page_writeback(struct extent_io_tree *tree,
1382 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1383 u64 end = start + PAGE_CACHE_SIZE - 1;
1384 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1385 end_page_writeback(page);
1389 /* lots and lots of room for performance fixes in the end_bio funcs */
1392 * after a writepage IO is done, we need to:
1393 * clear the uptodate bits on error
1394 * clear the writeback bits in the extent tree for this IO
1395 * end_page_writeback if the page has no more pending IO
1397 * Scheduling is not allowed, so the extent state tree is expected
1398 * to have one and only one object corresponding to this IO.
1400 static void end_bio_extent_writepage(struct bio *bio, int err)
1402 int uptodate = err == 0;
1403 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1404 struct extent_io_tree *tree;
1411 struct page *page = bvec->bv_page;
1412 tree = &BTRFS_I(page->mapping->host)->io_tree;
1414 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1416 end = start + bvec->bv_len - 1;
1418 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1423 if (--bvec >= bio->bi_io_vec)
1424 prefetchw(&bvec->bv_page->flags);
1425 if (tree->ops && tree->ops->writepage_end_io_hook) {
1426 ret = tree->ops->writepage_end_io_hook(page, start,
1427 end, NULL, uptodate);
1432 if (!uptodate && tree->ops &&
1433 tree->ops->writepage_io_failed_hook) {
1434 ret = tree->ops->writepage_io_failed_hook(bio, page,
1437 uptodate = (err == 0);
1443 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1444 ClearPageUptodate(page);
1448 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1451 end_page_writeback(page);
1453 check_page_writeback(tree, page);
1454 } while (bvec >= bio->bi_io_vec);
1460 * after a readpage IO is done, we need to:
1461 * clear the uptodate bits on error
1462 * set the uptodate bits if things worked
1463 * set the page up to date if all extents in the tree are uptodate
1464 * clear the lock bit in the extent tree
1465 * unlock the page if there are no other extents locked for it
1467 * Scheduling is not allowed, so the extent state tree is expected
1468 * to have one and only one object corresponding to this IO.
1470 static void end_bio_extent_readpage(struct bio *bio, int err)
1472 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1473 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1474 struct extent_io_tree *tree;
1481 struct page *page = bvec->bv_page;
1482 tree = &BTRFS_I(page->mapping->host)->io_tree;
1484 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1486 end = start + bvec->bv_len - 1;
1488 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1493 if (--bvec >= bio->bi_io_vec)
1494 prefetchw(&bvec->bv_page->flags);
1496 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1497 ret = tree->ops->readpage_end_io_hook(page, start, end,
1502 if (!uptodate && tree->ops &&
1503 tree->ops->readpage_io_failed_hook) {
1504 ret = tree->ops->readpage_io_failed_hook(bio, page,
1508 test_bit(BIO_UPTODATE, &bio->bi_flags);
1514 set_extent_uptodate(tree, start, end,
1516 unlock_extent(tree, start, end, GFP_ATOMIC);
1520 SetPageUptodate(page);
1522 ClearPageUptodate(page);
1528 check_page_uptodate(tree, page);
1530 ClearPageUptodate(page);
1533 check_page_locked(tree, page);
1535 } while (bvec >= bio->bi_io_vec);
1541 * IO done from prepare_write is pretty simple, we just unlock
1542 * the structs in the extent tree when done, and set the uptodate bits
1545 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1547 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1548 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1549 struct extent_io_tree *tree;
1554 struct page *page = bvec->bv_page;
1555 tree = &BTRFS_I(page->mapping->host)->io_tree;
1557 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1559 end = start + bvec->bv_len - 1;
1561 if (--bvec >= bio->bi_io_vec)
1562 prefetchw(&bvec->bv_page->flags);
1565 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1567 ClearPageUptodate(page);
1571 unlock_extent(tree, start, end, GFP_ATOMIC);
1573 } while (bvec >= bio->bi_io_vec);
1579 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1584 bio = bio_alloc(gfp_flags, nr_vecs);
1586 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1587 while (!bio && (nr_vecs /= 2))
1588 bio = bio_alloc(gfp_flags, nr_vecs);
1593 bio->bi_bdev = bdev;
1594 bio->bi_sector = first_sector;
1599 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1602 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1603 struct page *page = bvec->bv_page;
1604 struct extent_io_tree *tree = bio->bi_private;
1605 struct rb_node *node;
1606 struct extent_state *state;
1610 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1611 end = start + bvec->bv_len - 1;
1613 spin_lock_irq(&tree->lock);
1614 node = __etree_search(tree, start, NULL, NULL);
1616 state = rb_entry(node, struct extent_state, rb_node);
1617 while(state->end < end) {
1618 node = rb_next(node);
1619 state = rb_entry(node, struct extent_state, rb_node);
1621 BUG_ON(state->end != end);
1622 spin_unlock_irq(&tree->lock);
1624 bio->bi_private = NULL;
1628 if (tree->ops && tree->ops->submit_bio_hook)
1629 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1632 submit_bio(rw, bio);
1633 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1639 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1640 struct page *page, sector_t sector,
1641 size_t size, unsigned long offset,
1642 struct block_device *bdev,
1643 struct bio **bio_ret,
1644 unsigned long max_pages,
1645 bio_end_io_t end_io_func,
1652 if (bio_ret && *bio_ret) {
1654 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1655 (tree->ops && tree->ops->merge_bio_hook &&
1656 tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1657 bio_add_page(bio, page, size, offset) < size) {
1658 ret = submit_one_bio(rw, bio, mirror_num);
1664 nr = bio_get_nr_vecs(bdev);
1665 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1667 printk("failed to allocate bio nr %d\n", nr);
1671 bio_add_page(bio, page, size, offset);
1672 bio->bi_end_io = end_io_func;
1673 bio->bi_private = tree;
1678 ret = submit_one_bio(rw, bio, mirror_num);
1684 void set_page_extent_mapped(struct page *page)
1686 if (!PagePrivate(page)) {
1687 SetPagePrivate(page);
1688 page_cache_get(page);
1689 set_page_private(page, EXTENT_PAGE_PRIVATE);
1693 void set_page_extent_head(struct page *page, unsigned long len)
1695 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1699 * basic readpage implementation. Locked extent state structs are inserted
1700 * into the tree that are removed when the IO is done (by the end_io
1703 static int __extent_read_full_page(struct extent_io_tree *tree,
1705 get_extent_t *get_extent,
1706 struct bio **bio, int mirror_num)
1708 struct inode *inode = page->mapping->host;
1709 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1710 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1714 u64 last_byte = i_size_read(inode);
1718 struct extent_map *em;
1719 struct block_device *bdev;
1722 size_t page_offset = 0;
1724 size_t blocksize = inode->i_sb->s_blocksize;
1726 set_page_extent_mapped(page);
1729 lock_extent(tree, start, end, GFP_NOFS);
1731 while (cur <= end) {
1732 if (cur >= last_byte) {
1734 iosize = PAGE_CACHE_SIZE - page_offset;
1735 userpage = kmap_atomic(page, KM_USER0);
1736 memset(userpage + page_offset, 0, iosize);
1737 flush_dcache_page(page);
1738 kunmap_atomic(userpage, KM_USER0);
1739 set_extent_uptodate(tree, cur, cur + iosize - 1,
1741 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1744 em = get_extent(inode, page, page_offset, cur,
1746 if (IS_ERR(em) || !em) {
1748 unlock_extent(tree, cur, end, GFP_NOFS);
1751 extent_offset = cur - em->start;
1752 if (extent_map_end(em) <= cur) {
1753 printk("bad mapping em [%Lu %Lu] cur %Lu\n", em->start, extent_map_end(em), cur);
1755 BUG_ON(extent_map_end(em) <= cur);
1757 printk("2bad mapping end %Lu cur %Lu\n", end, cur);
1761 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1762 cur_end = min(extent_map_end(em) - 1, end);
1763 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1764 sector = (em->block_start + extent_offset) >> 9;
1766 block_start = em->block_start;
1767 free_extent_map(em);
1770 /* we've found a hole, just zero and go on */
1771 if (block_start == EXTENT_MAP_HOLE) {
1773 userpage = kmap_atomic(page, KM_USER0);
1774 memset(userpage + page_offset, 0, iosize);
1775 flush_dcache_page(page);
1776 kunmap_atomic(userpage, KM_USER0);
1778 set_extent_uptodate(tree, cur, cur + iosize - 1,
1780 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1782 page_offset += iosize;
1785 /* the get_extent function already copied into the page */
1786 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1787 check_page_uptodate(tree, page);
1788 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1790 page_offset += iosize;
1793 /* we have an inline extent but it didn't get marked up
1794 * to date. Error out
1796 if (block_start == EXTENT_MAP_INLINE) {
1798 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1800 page_offset += iosize;
1805 if (tree->ops && tree->ops->readpage_io_hook) {
1806 ret = tree->ops->readpage_io_hook(page, cur,
1810 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1812 ret = submit_extent_page(READ, tree, page,
1813 sector, iosize, page_offset,
1815 end_bio_extent_readpage, mirror_num);
1821 page_offset += iosize;
1824 if (!PageError(page))
1825 SetPageUptodate(page);
1831 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1832 get_extent_t *get_extent)
1834 struct bio *bio = NULL;
1837 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1839 submit_one_bio(READ, bio, 0);
1842 EXPORT_SYMBOL(extent_read_full_page);
1845 * the writepage semantics are similar to regular writepage. extent
1846 * records are inserted to lock ranges in the tree, and as dirty areas
1847 * are found, they are marked writeback. Then the lock bits are removed
1848 * and the end_io handler clears the writeback ranges
1850 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1853 struct inode *inode = page->mapping->host;
1854 struct extent_page_data *epd = data;
1855 struct extent_io_tree *tree = epd->tree;
1856 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1858 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1862 u64 last_byte = i_size_read(inode);
1867 struct extent_map *em;
1868 struct block_device *bdev;
1871 size_t pg_offset = 0;
1873 loff_t i_size = i_size_read(inode);
1874 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1878 WARN_ON(!PageLocked(page));
1879 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
1880 if (page->index > end_index ||
1881 (page->index == end_index && !pg_offset)) {
1882 page->mapping->a_ops->invalidatepage(page, 0);
1887 if (page->index == end_index) {
1890 userpage = kmap_atomic(page, KM_USER0);
1891 memset(userpage + pg_offset, 0,
1892 PAGE_CACHE_SIZE - pg_offset);
1893 kunmap_atomic(userpage, KM_USER0);
1894 flush_dcache_page(page);
1898 set_page_extent_mapped(page);
1900 delalloc_start = start;
1902 while(delalloc_end < page_end) {
1903 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
1906 if (nr_delalloc == 0) {
1907 delalloc_start = delalloc_end + 1;
1910 tree->ops->fill_delalloc(inode, delalloc_start,
1912 clear_extent_bit(tree, delalloc_start,
1914 EXTENT_LOCKED | EXTENT_DELALLOC,
1916 delalloc_start = delalloc_end + 1;
1918 lock_extent(tree, start, page_end, GFP_NOFS);
1919 unlock_start = start;
1921 if (tree->ops && tree->ops->writepage_start_hook) {
1922 ret = tree->ops->writepage_start_hook(page, start, page_end);
1923 if (ret == -EAGAIN) {
1924 unlock_extent(tree, start, page_end, GFP_NOFS);
1925 redirty_page_for_writepage(wbc, page);
1932 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
1933 printk("found delalloc bits after lock_extent\n");
1936 if (last_byte <= start) {
1937 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1938 unlock_extent(tree, start, page_end, GFP_NOFS);
1939 if (tree->ops && tree->ops->writepage_end_io_hook)
1940 tree->ops->writepage_end_io_hook(page, start,
1942 unlock_start = page_end + 1;
1946 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1947 blocksize = inode->i_sb->s_blocksize;
1949 while (cur <= end) {
1950 if (cur >= last_byte) {
1951 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1952 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
1953 if (tree->ops && tree->ops->writepage_end_io_hook)
1954 tree->ops->writepage_end_io_hook(page, cur,
1956 unlock_start = page_end + 1;
1959 em = epd->get_extent(inode, page, pg_offset, cur,
1961 if (IS_ERR(em) || !em) {
1966 extent_offset = cur - em->start;
1967 BUG_ON(extent_map_end(em) <= cur);
1969 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1970 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1971 sector = (em->block_start + extent_offset) >> 9;
1973 block_start = em->block_start;
1974 free_extent_map(em);
1977 if (block_start == EXTENT_MAP_HOLE ||
1978 block_start == EXTENT_MAP_INLINE) {
1979 clear_extent_dirty(tree, cur,
1980 cur + iosize - 1, GFP_NOFS);
1982 unlock_extent(tree, unlock_start, cur + iosize -1,
1985 if (tree->ops && tree->ops->writepage_end_io_hook)
1986 tree->ops->writepage_end_io_hook(page, cur,
1990 pg_offset += iosize;
1995 /* leave this out until we have a page_mkwrite call */
1996 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1999 pg_offset += iosize;
2002 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2003 if (tree->ops && tree->ops->writepage_io_hook) {
2004 ret = tree->ops->writepage_io_hook(page, cur,
2012 unsigned long max_nr = end_index + 1;
2014 set_range_writeback(tree, cur, cur + iosize - 1);
2015 if (!PageWriteback(page)) {
2016 printk("warning page %lu not writeback, "
2017 "cur %llu end %llu\n", page->index,
2018 (unsigned long long)cur,
2019 (unsigned long long)end);
2022 ret = submit_extent_page(WRITE, tree, page, sector,
2023 iosize, pg_offset, bdev,
2025 end_bio_extent_writepage, 0);
2030 pg_offset += iosize;
2035 /* make sure the mapping tag for page dirty gets cleared */
2036 set_page_writeback(page);
2037 end_page_writeback(page);
2039 if (unlock_start <= page_end)
2040 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2046 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2047 * @mapping: address space structure to write
2048 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2049 * @writepage: function called for each page
2050 * @data: data passed to writepage function
2052 * If a page is already under I/O, write_cache_pages() skips it, even
2053 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2054 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2055 * and msync() need to guarantee that all the data which was dirty at the time
2056 * the call was made get new I/O started against them. If wbc->sync_mode is
2057 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2058 * existing IO to complete.
2060 int extent_write_cache_pages(struct extent_io_tree *tree,
2061 struct address_space *mapping,
2062 struct writeback_control *wbc,
2063 writepage_t writepage, void *data)
2065 struct backing_dev_info *bdi = mapping->backing_dev_info;
2068 struct pagevec pvec;
2071 pgoff_t end; /* Inclusive */
2073 int range_whole = 0;
2075 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2076 wbc->encountered_congestion = 1;
2080 pagevec_init(&pvec, 0);
2081 if (wbc->range_cyclic) {
2082 index = mapping->writeback_index; /* Start from prev offset */
2085 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2086 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2087 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2092 while (!done && (index <= end) &&
2093 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2094 PAGECACHE_TAG_DIRTY,
2095 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2099 for (i = 0; i < nr_pages; i++) {
2100 struct page *page = pvec.pages[i];
2103 * At this point we hold neither mapping->tree_lock nor
2104 * lock on the page itself: the page may be truncated or
2105 * invalidated (changing page->mapping to NULL), or even
2106 * swizzled back from swapper_space to tmpfs file
2109 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2110 tree->ops->write_cache_pages_lock_hook(page);
2114 if (unlikely(page->mapping != mapping)) {
2119 if (!wbc->range_cyclic && page->index > end) {
2125 if (wbc->sync_mode != WB_SYNC_NONE)
2126 wait_on_page_writeback(page);
2128 if (PageWriteback(page) ||
2129 !clear_page_dirty_for_io(page)) {
2134 ret = (*writepage)(page, wbc, data);
2136 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2140 if (ret || (--(wbc->nr_to_write) <= 0))
2142 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2143 wbc->encountered_congestion = 1;
2147 pagevec_release(&pvec);
2150 if (!scanned && !done) {
2152 * We hit the last page and there is more work to be done: wrap
2153 * back to the start of the file
2159 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2160 mapping->writeback_index = index;
2162 if (wbc->range_cont)
2163 wbc->range_start = index << PAGE_CACHE_SHIFT;
2166 EXPORT_SYMBOL(extent_write_cache_pages);
2168 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2169 get_extent_t *get_extent,
2170 struct writeback_control *wbc)
2173 struct address_space *mapping = page->mapping;
2174 struct extent_page_data epd = {
2177 .get_extent = get_extent,
2179 struct writeback_control wbc_writepages = {
2181 .sync_mode = WB_SYNC_NONE,
2182 .older_than_this = NULL,
2184 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2185 .range_end = (loff_t)-1,
2189 ret = __extent_writepage(page, wbc, &epd);
2191 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2192 __extent_writepage, &epd);
2194 submit_one_bio(WRITE, epd.bio, 0);
2198 EXPORT_SYMBOL(extent_write_full_page);
2201 int extent_writepages(struct extent_io_tree *tree,
2202 struct address_space *mapping,
2203 get_extent_t *get_extent,
2204 struct writeback_control *wbc)
2207 struct extent_page_data epd = {
2210 .get_extent = get_extent,
2213 ret = extent_write_cache_pages(tree, mapping, wbc,
2214 __extent_writepage, &epd);
2216 submit_one_bio(WRITE, epd.bio, 0);
2220 EXPORT_SYMBOL(extent_writepages);
2222 int extent_readpages(struct extent_io_tree *tree,
2223 struct address_space *mapping,
2224 struct list_head *pages, unsigned nr_pages,
2225 get_extent_t get_extent)
2227 struct bio *bio = NULL;
2229 struct pagevec pvec;
2231 pagevec_init(&pvec, 0);
2232 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2233 struct page *page = list_entry(pages->prev, struct page, lru);
2235 prefetchw(&page->flags);
2236 list_del(&page->lru);
2238 * what we want to do here is call add_to_page_cache_lru,
2239 * but that isn't exported, so we reproduce it here
2241 if (!add_to_page_cache(page, mapping,
2242 page->index, GFP_KERNEL)) {
2244 /* open coding of lru_cache_add, also not exported */
2245 page_cache_get(page);
2246 if (!pagevec_add(&pvec, page))
2247 __pagevec_lru_add(&pvec);
2248 __extent_read_full_page(tree, page, get_extent,
2251 page_cache_release(page);
2253 if (pagevec_count(&pvec))
2254 __pagevec_lru_add(&pvec);
2255 BUG_ON(!list_empty(pages));
2257 submit_one_bio(READ, bio, 0);
2260 EXPORT_SYMBOL(extent_readpages);
2263 * basic invalidatepage code, this waits on any locked or writeback
2264 * ranges corresponding to the page, and then deletes any extent state
2265 * records from the tree
2267 int extent_invalidatepage(struct extent_io_tree *tree,
2268 struct page *page, unsigned long offset)
2270 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2271 u64 end = start + PAGE_CACHE_SIZE - 1;
2272 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2274 start += (offset + blocksize -1) & ~(blocksize - 1);
2278 lock_extent(tree, start, end, GFP_NOFS);
2279 wait_on_extent_writeback(tree, start, end);
2280 clear_extent_bit(tree, start, end,
2281 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2285 EXPORT_SYMBOL(extent_invalidatepage);
2288 * simple commit_write call, set_range_dirty is used to mark both
2289 * the pages and the extent records as dirty
2291 int extent_commit_write(struct extent_io_tree *tree,
2292 struct inode *inode, struct page *page,
2293 unsigned from, unsigned to)
2295 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2297 set_page_extent_mapped(page);
2298 set_page_dirty(page);
2300 if (pos > inode->i_size) {
2301 i_size_write(inode, pos);
2302 mark_inode_dirty(inode);
2306 EXPORT_SYMBOL(extent_commit_write);
2308 int extent_prepare_write(struct extent_io_tree *tree,
2309 struct inode *inode, struct page *page,
2310 unsigned from, unsigned to, get_extent_t *get_extent)
2312 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2313 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2315 u64 orig_block_start;
2318 struct extent_map *em;
2319 unsigned blocksize = 1 << inode->i_blkbits;
2320 size_t page_offset = 0;
2321 size_t block_off_start;
2322 size_t block_off_end;
2328 set_page_extent_mapped(page);
2330 block_start = (page_start + from) & ~((u64)blocksize - 1);
2331 block_end = (page_start + to - 1) | (blocksize - 1);
2332 orig_block_start = block_start;
2334 lock_extent(tree, page_start, page_end, GFP_NOFS);
2335 while(block_start <= block_end) {
2336 em = get_extent(inode, page, page_offset, block_start,
2337 block_end - block_start + 1, 1);
2338 if (IS_ERR(em) || !em) {
2341 cur_end = min(block_end, extent_map_end(em) - 1);
2342 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2343 block_off_end = block_off_start + blocksize;
2344 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2346 if (!PageUptodate(page) && isnew &&
2347 (block_off_end > to || block_off_start < from)) {
2350 kaddr = kmap_atomic(page, KM_USER0);
2351 if (block_off_end > to)
2352 memset(kaddr + to, 0, block_off_end - to);
2353 if (block_off_start < from)
2354 memset(kaddr + block_off_start, 0,
2355 from - block_off_start);
2356 flush_dcache_page(page);
2357 kunmap_atomic(kaddr, KM_USER0);
2359 if ((em->block_start != EXTENT_MAP_HOLE &&
2360 em->block_start != EXTENT_MAP_INLINE) &&
2361 !isnew && !PageUptodate(page) &&
2362 (block_off_end > to || block_off_start < from) &&
2363 !test_range_bit(tree, block_start, cur_end,
2364 EXTENT_UPTODATE, 1)) {
2366 u64 extent_offset = block_start - em->start;
2368 sector = (em->block_start + extent_offset) >> 9;
2369 iosize = (cur_end - block_start + blocksize) &
2370 ~((u64)blocksize - 1);
2372 * we've already got the extent locked, but we
2373 * need to split the state such that our end_bio
2374 * handler can clear the lock.
2376 set_extent_bit(tree, block_start,
2377 block_start + iosize - 1,
2378 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2379 ret = submit_extent_page(READ, tree, page,
2380 sector, iosize, page_offset, em->bdev,
2382 end_bio_extent_preparewrite, 0);
2384 block_start = block_start + iosize;
2386 set_extent_uptodate(tree, block_start, cur_end,
2388 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2389 block_start = cur_end + 1;
2391 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2392 free_extent_map(em);
2395 wait_extent_bit(tree, orig_block_start,
2396 block_end, EXTENT_LOCKED);
2398 check_page_uptodate(tree, page);
2400 /* FIXME, zero out newly allocated blocks on error */
2403 EXPORT_SYMBOL(extent_prepare_write);
2406 * a helper for releasepage, this tests for areas of the page that
2407 * are locked or under IO and drops the related state bits if it is safe
2410 int try_release_extent_state(struct extent_map_tree *map,
2411 struct extent_io_tree *tree, struct page *page,
2414 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2415 u64 end = start + PAGE_CACHE_SIZE - 1;
2418 if (test_range_bit(tree, start, end,
2419 EXTENT_IOBITS | EXTENT_ORDERED, 0))
2422 if ((mask & GFP_NOFS) == GFP_NOFS)
2424 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2429 EXPORT_SYMBOL(try_release_extent_state);
2432 * a helper for releasepage. As long as there are no locked extents
2433 * in the range corresponding to the page, both state records and extent
2434 * map records are removed
2436 int try_release_extent_mapping(struct extent_map_tree *map,
2437 struct extent_io_tree *tree, struct page *page,
2440 struct extent_map *em;
2441 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2442 u64 end = start + PAGE_CACHE_SIZE - 1;
2444 if ((mask & __GFP_WAIT) &&
2445 page->mapping->host->i_size > 16 * 1024 * 1024) {
2447 while (start <= end) {
2448 len = end - start + 1;
2449 spin_lock(&map->lock);
2450 em = lookup_extent_mapping(map, start, len);
2451 if (!em || IS_ERR(em)) {
2452 spin_unlock(&map->lock);
2455 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2456 em->start != start) {
2457 spin_unlock(&map->lock);
2458 free_extent_map(em);
2461 if (!test_range_bit(tree, em->start,
2462 extent_map_end(em) - 1,
2463 EXTENT_LOCKED, 0)) {
2464 remove_extent_mapping(map, em);
2465 /* once for the rb tree */
2466 free_extent_map(em);
2468 start = extent_map_end(em);
2469 spin_unlock(&map->lock);
2472 free_extent_map(em);
2475 return try_release_extent_state(map, tree, page, mask);
2477 EXPORT_SYMBOL(try_release_extent_mapping);
2479 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2480 get_extent_t *get_extent)
2482 struct inode *inode = mapping->host;
2483 u64 start = iblock << inode->i_blkbits;
2484 sector_t sector = 0;
2485 struct extent_map *em;
2487 em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2488 if (!em || IS_ERR(em))
2491 if (em->block_start == EXTENT_MAP_INLINE ||
2492 em->block_start == EXTENT_MAP_HOLE)
2495 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2497 free_extent_map(em);
2501 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2505 struct address_space *mapping;
2508 return eb->first_page;
2509 i += eb->start >> PAGE_CACHE_SHIFT;
2510 mapping = eb->first_page->mapping;
2515 * extent_buffer_page is only called after pinning the page
2516 * by increasing the reference count. So we know the page must
2517 * be in the radix tree.
2520 p = radix_tree_lookup(&mapping->page_tree, i);
2526 static inline unsigned long num_extent_pages(u64 start, u64 len)
2528 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2529 (start >> PAGE_CACHE_SHIFT);
2532 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2537 struct extent_buffer *eb = NULL;
2539 unsigned long flags;
2542 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2545 mutex_init(&eb->mutex);
2547 spin_lock_irqsave(&leak_lock, flags);
2548 list_add(&eb->leak_list, &buffers);
2549 spin_unlock_irqrestore(&leak_lock, flags);
2551 atomic_set(&eb->refs, 1);
2556 static void __free_extent_buffer(struct extent_buffer *eb)
2559 unsigned long flags;
2560 spin_lock_irqsave(&leak_lock, flags);
2561 list_del(&eb->leak_list);
2562 spin_unlock_irqrestore(&leak_lock, flags);
2564 kmem_cache_free(extent_buffer_cache, eb);
2567 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2568 u64 start, unsigned long len,
2572 unsigned long num_pages = num_extent_pages(start, len);
2574 unsigned long index = start >> PAGE_CACHE_SHIFT;
2575 struct extent_buffer *eb;
2576 struct extent_buffer *exists = NULL;
2578 struct address_space *mapping = tree->mapping;
2581 spin_lock(&tree->buffer_lock);
2582 eb = buffer_search(tree, start);
2584 atomic_inc(&eb->refs);
2585 spin_unlock(&tree->buffer_lock);
2586 mark_page_accessed(eb->first_page);
2589 spin_unlock(&tree->buffer_lock);
2591 eb = __alloc_extent_buffer(tree, start, len, mask);
2596 eb->first_page = page0;
2599 page_cache_get(page0);
2600 mark_page_accessed(page0);
2601 set_page_extent_mapped(page0);
2602 set_page_extent_head(page0, len);
2603 uptodate = PageUptodate(page0);
2607 for (; i < num_pages; i++, index++) {
2608 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2613 set_page_extent_mapped(p);
2614 mark_page_accessed(p);
2617 set_page_extent_head(p, len);
2619 set_page_private(p, EXTENT_PAGE_PRIVATE);
2621 if (!PageUptodate(p))
2626 eb->flags |= EXTENT_UPTODATE;
2627 eb->flags |= EXTENT_BUFFER_FILLED;
2629 spin_lock(&tree->buffer_lock);
2630 exists = buffer_tree_insert(tree, start, &eb->rb_node);
2632 /* add one reference for the caller */
2633 atomic_inc(&exists->refs);
2634 spin_unlock(&tree->buffer_lock);
2637 spin_unlock(&tree->buffer_lock);
2639 /* add one reference for the tree */
2640 atomic_inc(&eb->refs);
2644 if (!atomic_dec_and_test(&eb->refs))
2646 for (index = 1; index < i; index++)
2647 page_cache_release(extent_buffer_page(eb, index));
2648 page_cache_release(extent_buffer_page(eb, 0));
2649 __free_extent_buffer(eb);
2652 EXPORT_SYMBOL(alloc_extent_buffer);
2654 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2655 u64 start, unsigned long len,
2658 struct extent_buffer *eb;
2660 spin_lock(&tree->buffer_lock);
2661 eb = buffer_search(tree, start);
2663 atomic_inc(&eb->refs);
2664 spin_unlock(&tree->buffer_lock);
2667 mark_page_accessed(eb->first_page);
2671 EXPORT_SYMBOL(find_extent_buffer);
2673 void free_extent_buffer(struct extent_buffer *eb)
2678 if (!atomic_dec_and_test(&eb->refs))
2683 EXPORT_SYMBOL(free_extent_buffer);
2685 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2686 struct extent_buffer *eb)
2690 unsigned long num_pages;
2693 u64 start = eb->start;
2694 u64 end = start + eb->len - 1;
2696 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2697 num_pages = num_extent_pages(eb->start, eb->len);
2699 for (i = 0; i < num_pages; i++) {
2700 page = extent_buffer_page(eb, i);
2703 set_page_extent_head(page, eb->len);
2705 set_page_private(page, EXTENT_PAGE_PRIVATE);
2708 * if we're on the last page or the first page and the
2709 * block isn't aligned on a page boundary, do extra checks
2710 * to make sure we don't clean page that is partially dirty
2712 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2713 ((i == num_pages - 1) &&
2714 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2715 start = (u64)page->index << PAGE_CACHE_SHIFT;
2716 end = start + PAGE_CACHE_SIZE - 1;
2717 if (test_range_bit(tree, start, end,
2723 clear_page_dirty_for_io(page);
2724 spin_lock_irq(&page->mapping->tree_lock);
2725 if (!PageDirty(page)) {
2726 radix_tree_tag_clear(&page->mapping->page_tree,
2728 PAGECACHE_TAG_DIRTY);
2730 spin_unlock_irq(&page->mapping->tree_lock);
2735 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2737 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2738 struct extent_buffer *eb)
2740 return wait_on_extent_writeback(tree, eb->start,
2741 eb->start + eb->len - 1);
2743 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2745 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2746 struct extent_buffer *eb)
2749 unsigned long num_pages;
2751 num_pages = num_extent_pages(eb->start, eb->len);
2752 for (i = 0; i < num_pages; i++) {
2753 struct page *page = extent_buffer_page(eb, i);
2754 /* writepage may need to do something special for the
2755 * first page, we have to make sure page->private is
2756 * properly set. releasepage may drop page->private
2757 * on us if the page isn't already dirty.
2761 set_page_extent_head(page, eb->len);
2762 } else if (PagePrivate(page) &&
2763 page->private != EXTENT_PAGE_PRIVATE) {
2764 set_page_extent_mapped(page);
2766 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2767 set_extent_dirty(tree, page_offset(page),
2768 page_offset(page) + PAGE_CACHE_SIZE -1,
2774 EXPORT_SYMBOL(set_extent_buffer_dirty);
2776 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
2777 struct extent_buffer *eb)
2781 unsigned long num_pages;
2783 num_pages = num_extent_pages(eb->start, eb->len);
2784 eb->flags &= ~EXTENT_UPTODATE;
2786 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2788 for (i = 0; i < num_pages; i++) {
2789 page = extent_buffer_page(eb, i);
2791 ClearPageUptodate(page);
2796 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2797 struct extent_buffer *eb)
2801 unsigned long num_pages;
2803 num_pages = num_extent_pages(eb->start, eb->len);
2805 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2807 for (i = 0; i < num_pages; i++) {
2808 page = extent_buffer_page(eb, i);
2809 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2810 ((i == num_pages - 1) &&
2811 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2812 check_page_uptodate(tree, page);
2815 SetPageUptodate(page);
2819 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2821 int extent_range_uptodate(struct extent_io_tree *tree,
2826 int pg_uptodate = 1;
2828 unsigned long index;
2830 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2833 while(start <= end) {
2834 index = start >> PAGE_CACHE_SHIFT;
2835 page = find_get_page(tree->mapping, index);
2836 uptodate = PageUptodate(page);
2837 page_cache_release(page);
2842 start += PAGE_CACHE_SIZE;
2847 int extent_buffer_uptodate(struct extent_io_tree *tree,
2848 struct extent_buffer *eb)
2851 unsigned long num_pages;
2854 int pg_uptodate = 1;
2856 if (eb->flags & EXTENT_UPTODATE)
2859 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2860 EXTENT_UPTODATE, 1);
2864 num_pages = num_extent_pages(eb->start, eb->len);
2865 for (i = 0; i < num_pages; i++) {
2866 page = extent_buffer_page(eb, i);
2867 if (!PageUptodate(page)) {
2874 EXPORT_SYMBOL(extent_buffer_uptodate);
2876 int read_extent_buffer_pages(struct extent_io_tree *tree,
2877 struct extent_buffer *eb,
2878 u64 start, int wait,
2879 get_extent_t *get_extent, int mirror_num)
2882 unsigned long start_i;
2886 int locked_pages = 0;
2887 int all_uptodate = 1;
2888 int inc_all_pages = 0;
2889 unsigned long num_pages;
2890 struct bio *bio = NULL;
2892 if (eb->flags & EXTENT_UPTODATE)
2895 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2896 EXTENT_UPTODATE, 1)) {
2901 WARN_ON(start < eb->start);
2902 start_i = (start >> PAGE_CACHE_SHIFT) -
2903 (eb->start >> PAGE_CACHE_SHIFT);
2908 num_pages = num_extent_pages(eb->start, eb->len);
2909 for (i = start_i; i < num_pages; i++) {
2910 page = extent_buffer_page(eb, i);
2912 if (!trylock_page(page))
2918 if (!PageUptodate(page)) {
2924 eb->flags |= EXTENT_UPTODATE;
2926 printk("all up to date but ret is %d\n", ret);
2931 for (i = start_i; i < num_pages; i++) {
2932 page = extent_buffer_page(eb, i);
2934 page_cache_get(page);
2935 if (!PageUptodate(page)) {
2938 ClearPageError(page);
2939 err = __extent_read_full_page(tree, page,
2944 printk("err %d from __extent_read_full_page\n", ret);
2952 submit_one_bio(READ, bio, mirror_num);
2956 printk("ret %d wait %d returning\n", ret, wait);
2959 for (i = start_i; i < num_pages; i++) {
2960 page = extent_buffer_page(eb, i);
2961 wait_on_page_locked(page);
2962 if (!PageUptodate(page)) {
2963 printk("page not uptodate after wait_on_page_locked\n");
2968 eb->flags |= EXTENT_UPTODATE;
2973 while(locked_pages > 0) {
2974 page = extent_buffer_page(eb, i);
2981 EXPORT_SYMBOL(read_extent_buffer_pages);
2983 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
2984 unsigned long start,
2991 char *dst = (char *)dstv;
2992 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
2993 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
2995 WARN_ON(start > eb->len);
2996 WARN_ON(start + len > eb->start + eb->len);
2998 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3001 page = extent_buffer_page(eb, i);
3003 cur = min(len, (PAGE_CACHE_SIZE - offset));
3004 kaddr = kmap_atomic(page, KM_USER1);
3005 memcpy(dst, kaddr + offset, cur);
3006 kunmap_atomic(kaddr, KM_USER1);
3014 EXPORT_SYMBOL(read_extent_buffer);
3016 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3017 unsigned long min_len, char **token, char **map,
3018 unsigned long *map_start,
3019 unsigned long *map_len, int km)
3021 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3024 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3025 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3026 unsigned long end_i = (start_offset + start + min_len - 1) >>
3033 offset = start_offset;
3037 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3039 if (start + min_len > eb->len) {
3040 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3044 p = extent_buffer_page(eb, i);
3045 kaddr = kmap_atomic(p, km);
3047 *map = kaddr + offset;
3048 *map_len = PAGE_CACHE_SIZE - offset;
3051 EXPORT_SYMBOL(map_private_extent_buffer);
3053 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3054 unsigned long min_len,
3055 char **token, char **map,
3056 unsigned long *map_start,
3057 unsigned long *map_len, int km)
3061 if (eb->map_token) {
3062 unmap_extent_buffer(eb, eb->map_token, km);
3063 eb->map_token = NULL;
3066 err = map_private_extent_buffer(eb, start, min_len, token, map,
3067 map_start, map_len, km);
3069 eb->map_token = *token;
3071 eb->map_start = *map_start;
3072 eb->map_len = *map_len;
3076 EXPORT_SYMBOL(map_extent_buffer);
3078 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3080 kunmap_atomic(token, km);
3082 EXPORT_SYMBOL(unmap_extent_buffer);
3084 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3085 unsigned long start,
3092 char *ptr = (char *)ptrv;
3093 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3094 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3097 WARN_ON(start > eb->len);
3098 WARN_ON(start + len > eb->start + eb->len);
3100 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3103 page = extent_buffer_page(eb, i);
3105 cur = min(len, (PAGE_CACHE_SIZE - offset));
3107 kaddr = kmap_atomic(page, KM_USER0);
3108 ret = memcmp(ptr, kaddr + offset, cur);
3109 kunmap_atomic(kaddr, KM_USER0);
3120 EXPORT_SYMBOL(memcmp_extent_buffer);
3122 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3123 unsigned long start, unsigned long len)
3129 char *src = (char *)srcv;
3130 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3131 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3133 WARN_ON(start > eb->len);
3134 WARN_ON(start + len > eb->start + eb->len);
3136 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3139 page = extent_buffer_page(eb, i);
3140 WARN_ON(!PageUptodate(page));
3142 cur = min(len, PAGE_CACHE_SIZE - offset);
3143 kaddr = kmap_atomic(page, KM_USER1);
3144 memcpy(kaddr + offset, src, cur);
3145 kunmap_atomic(kaddr, KM_USER1);
3153 EXPORT_SYMBOL(write_extent_buffer);
3155 void memset_extent_buffer(struct extent_buffer *eb, char c,
3156 unsigned long start, unsigned long len)
3162 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3163 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3165 WARN_ON(start > eb->len);
3166 WARN_ON(start + len > eb->start + eb->len);
3168 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3171 page = extent_buffer_page(eb, i);
3172 WARN_ON(!PageUptodate(page));
3174 cur = min(len, PAGE_CACHE_SIZE - offset);
3175 kaddr = kmap_atomic(page, KM_USER0);
3176 memset(kaddr + offset, c, cur);
3177 kunmap_atomic(kaddr, KM_USER0);
3184 EXPORT_SYMBOL(memset_extent_buffer);
3186 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3187 unsigned long dst_offset, unsigned long src_offset,
3190 u64 dst_len = dst->len;
3195 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3196 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3198 WARN_ON(src->len != dst_len);
3200 offset = (start_offset + dst_offset) &
3201 ((unsigned long)PAGE_CACHE_SIZE - 1);
3204 page = extent_buffer_page(dst, i);
3205 WARN_ON(!PageUptodate(page));
3207 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3209 kaddr = kmap_atomic(page, KM_USER0);
3210 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3211 kunmap_atomic(kaddr, KM_USER0);
3219 EXPORT_SYMBOL(copy_extent_buffer);
3221 static void move_pages(struct page *dst_page, struct page *src_page,
3222 unsigned long dst_off, unsigned long src_off,
3225 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3226 if (dst_page == src_page) {
3227 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3229 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3230 char *p = dst_kaddr + dst_off + len;
3231 char *s = src_kaddr + src_off + len;
3236 kunmap_atomic(src_kaddr, KM_USER1);
3238 kunmap_atomic(dst_kaddr, KM_USER0);
3241 static void copy_pages(struct page *dst_page, struct page *src_page,
3242 unsigned long dst_off, unsigned long src_off,
3245 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3248 if (dst_page != src_page)
3249 src_kaddr = kmap_atomic(src_page, KM_USER1);
3251 src_kaddr = dst_kaddr;
3253 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3254 kunmap_atomic(dst_kaddr, KM_USER0);
3255 if (dst_page != src_page)
3256 kunmap_atomic(src_kaddr, KM_USER1);
3259 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3260 unsigned long src_offset, unsigned long len)
3263 size_t dst_off_in_page;
3264 size_t src_off_in_page;
3265 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3266 unsigned long dst_i;
3267 unsigned long src_i;
3269 if (src_offset + len > dst->len) {
3270 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3271 src_offset, len, dst->len);
3274 if (dst_offset + len > dst->len) {
3275 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3276 dst_offset, len, dst->len);
3281 dst_off_in_page = (start_offset + dst_offset) &
3282 ((unsigned long)PAGE_CACHE_SIZE - 1);
3283 src_off_in_page = (start_offset + src_offset) &
3284 ((unsigned long)PAGE_CACHE_SIZE - 1);
3286 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3287 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3289 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3291 cur = min_t(unsigned long, cur,
3292 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3294 copy_pages(extent_buffer_page(dst, dst_i),
3295 extent_buffer_page(dst, src_i),
3296 dst_off_in_page, src_off_in_page, cur);
3303 EXPORT_SYMBOL(memcpy_extent_buffer);
3305 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3306 unsigned long src_offset, unsigned long len)
3309 size_t dst_off_in_page;
3310 size_t src_off_in_page;
3311 unsigned long dst_end = dst_offset + len - 1;
3312 unsigned long src_end = src_offset + len - 1;
3313 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3314 unsigned long dst_i;
3315 unsigned long src_i;
3317 if (src_offset + len > dst->len) {
3318 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3319 src_offset, len, dst->len);
3322 if (dst_offset + len > dst->len) {
3323 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3324 dst_offset, len, dst->len);
3327 if (dst_offset < src_offset) {
3328 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3332 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3333 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3335 dst_off_in_page = (start_offset + dst_end) &
3336 ((unsigned long)PAGE_CACHE_SIZE - 1);
3337 src_off_in_page = (start_offset + src_end) &
3338 ((unsigned long)PAGE_CACHE_SIZE - 1);
3340 cur = min_t(unsigned long, len, src_off_in_page + 1);
3341 cur = min(cur, dst_off_in_page + 1);
3342 move_pages(extent_buffer_page(dst, dst_i),
3343 extent_buffer_page(dst, src_i),
3344 dst_off_in_page - cur + 1,
3345 src_off_in_page - cur + 1, cur);
3352 EXPORT_SYMBOL(memmove_extent_buffer);
3354 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3356 u64 start = page_offset(page);
3357 struct extent_buffer *eb;
3360 unsigned long num_pages;
3362 spin_lock(&tree->buffer_lock);
3363 eb = buffer_search(tree, start);
3367 if (atomic_read(&eb->refs) > 1) {
3371 /* at this point we can safely release the extent buffer */
3372 num_pages = num_extent_pages(eb->start, eb->len);
3373 for (i = 0; i < num_pages; i++)
3374 page_cache_release(extent_buffer_page(eb, i));
3375 rb_erase(&eb->rb_node, &tree->buffer);
3376 __free_extent_buffer(eb);
3378 spin_unlock(&tree->buffer_lock);
3381 EXPORT_SYMBOL(try_release_extent_buffer);