?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: kill btrfs_cache_create
[linux-2.6/linux-2.6-openrd.git] / fs / btrfs / extent_io.c
blobc33b54029d785303d598995da3255f8e031a1c3d
1 #include <linux/bitops.h>
2 #include <linux/slab.h>
3 #include <linux/bio.h>
4 #include <linux/mm.h>
5 #include <linux/gfp.h>
6 #include <linux/pagemap.h>
7 #include <linux/page-flags.h>
8 #include <linux/module.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include "extent_io.h"
15 #include "extent_map.h"
16 #include "compat.h"
17 #include "ctree.h"
18 #include "btrfs_inode.h"
19
20 static struct kmem_cache *extent_state_cache;
21 static struct kmem_cache *extent_buffer_cache;
22
23 static LIST_HEAD(buffers);
24 static LIST_HEAD(states);
25
26 #define LEAK_DEBUG 0
27 #if LEAK_DEBUG
28 static DEFINE_SPINLOCK(leak_lock);
29 #endif
30
31 #define BUFFER_LRU_MAX 64
32
33 struct tree_entry {
34 u64 start;
35 u64 end;
36 struct rb_node rb_node;
37 };
38
39 struct extent_page_data {
40 struct bio *bio;
41 struct extent_io_tree *tree;
42 get_extent_t *get_extent;
43
44 /* tells writepage not to lock the state bits for this range
45 * it still does the unlocking
46 */
47 unsigned int extent_locked:1;
48
49 /* tells the submit_bio code to use a WRITE_SYNC */
50 unsigned int sync_io:1;
51 };
52
53 int __init extent_io_init(void)
54 {
55 extent_state_cache = kmem_cache_create("extent_state",
56 sizeof(struct extent_state), 0,
57 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
58 if (!extent_state_cache)
59 return -ENOMEM;
60
61 extent_buffer_cache = kmem_cache_create("extent_buffers",
62 sizeof(struct extent_buffer), 0,
63 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
64 if (!extent_buffer_cache)
65 goto free_state_cache;
66 return 0;
67
68 free_state_cache:
69 kmem_cache_destroy(extent_state_cache);
70 return -ENOMEM;
71 }
72
73 void extent_io_exit(void)
74 {
75 struct extent_state *state;
76 struct extent_buffer *eb;
77
78 while (!list_empty(&states)) {
79 state = list_entry(states.next, struct extent_state, leak_list);
80 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
81 "state %lu in tree %p refs %d\n",
82 (unsigned long long)state->start,
83 (unsigned long long)state->end,
84 state->state, state->tree, atomic_read(&state->refs));
85 list_del(&state->leak_list);
86 kmem_cache_free(extent_state_cache, state);
87
88 }
89
90 while (!list_empty(&buffers)) {
91 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
92 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
93 "refs %d\n", (unsigned long long)eb->start,
94 eb->len, atomic_read(&eb->refs));
95 list_del(&eb->leak_list);
96 kmem_cache_free(extent_buffer_cache, eb);
97 }
98 if (extent_state_cache)
99 kmem_cache_destroy(extent_state_cache);
100 if (extent_buffer_cache)
101 kmem_cache_destroy(extent_buffer_cache);
102 }
103
104 void extent_io_tree_init(struct extent_io_tree *tree,
105 struct address_space *mapping, gfp_t mask)
106 {
107 tree->state.rb_node = NULL;
108 tree->buffer.rb_node = NULL;
109 tree->ops = NULL;
110 tree->dirty_bytes = 0;
111 spin_lock_init(&tree->lock);
112 spin_lock_init(&tree->buffer_lock);
113 tree->mapping = mapping;
114 }
115
116 static struct extent_state *alloc_extent_state(gfp_t mask)
117 {
118 struct extent_state *state;
119 #if LEAK_DEBUG
120 unsigned long flags;
121 #endif
122
123 state = kmem_cache_alloc(extent_state_cache, mask);
124 if (!state)
125 return state;
126 state->state = 0;
127 state->private = 0;
128 state->tree = NULL;
129 #if LEAK_DEBUG
130 spin_lock_irqsave(&leak_lock, flags);
131 list_add(&state->leak_list, &states);
132 spin_unlock_irqrestore(&leak_lock, flags);
133 #endif
134 atomic_set(&state->refs, 1);
135 init_waitqueue_head(&state->wq);
136 return state;
137 }
138
139 static void free_extent_state(struct extent_state *state)
140 {
141 if (!state)
142 return;
143 if (atomic_dec_and_test(&state->refs)) {
144 #if LEAK_DEBUG
145 unsigned long flags;
146 #endif
147 WARN_ON(state->tree);
148 #if LEAK_DEBUG
149 spin_lock_irqsave(&leak_lock, flags);
150 list_del(&state->leak_list);
151 spin_unlock_irqrestore(&leak_lock, flags);
152 #endif
153 kmem_cache_free(extent_state_cache, state);
154 }
155 }
156
157 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
158 struct rb_node *node)
159 {
160 struct rb_node **p = &root->rb_node;
161 struct rb_node *parent = NULL;
162 struct tree_entry *entry;
163
164 while (*p) {
165 parent = *p;
166 entry = rb_entry(parent, struct tree_entry, rb_node);
167
168 if (offset < entry->start)
169 p = &(*p)->rb_left;
170 else if (offset > entry->end)
171 p = &(*p)->rb_right;
172 else
173 return parent;
174 }
175
176 entry = rb_entry(node, struct tree_entry, rb_node);
177 rb_link_node(node, parent, p);
178 rb_insert_color(node, root);
179 return NULL;
180 }
181
182 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
183 struct rb_node **prev_ret,
184 struct rb_node **next_ret)
185 {
186 struct rb_root *root = &tree->state;
187 struct rb_node *n = root->rb_node;
188 struct rb_node *prev = NULL;
189 struct rb_node *orig_prev = NULL;
190 struct tree_entry *entry;
191 struct tree_entry *prev_entry = NULL;
192
193 while (n) {
194 entry = rb_entry(n, struct tree_entry, rb_node);
195 prev = n;
196 prev_entry = entry;
197
198 if (offset < entry->start)
199 n = n->rb_left;
200 else if (offset > entry->end)
201 n = n->rb_right;
202 else
203 return n;
204 }
205
206 if (prev_ret) {
207 orig_prev = prev;
208 while (prev && offset > prev_entry->end) {
209 prev = rb_next(prev);
210 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
211 }
212 *prev_ret = prev;
213 prev = orig_prev;
214 }
215
216 if (next_ret) {
217 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
218 while (prev && offset < prev_entry->start) {
219 prev = rb_prev(prev);
220 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
221 }
222 *next_ret = prev;
223 }
224 return NULL;
225 }
226
227 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
228 u64 offset)
229 {
230 struct rb_node *prev = NULL;
231 struct rb_node *ret;
232
233 ret = __etree_search(tree, offset, &prev, NULL);
234 if (!ret)
235 return prev;
236 return ret;
237 }
238
239 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
240 u64 offset, struct rb_node *node)
241 {
242 struct rb_root *root = &tree->buffer;
243 struct rb_node **p = &root->rb_node;
244 struct rb_node *parent = NULL;
245 struct extent_buffer *eb;
246
247 while (*p) {
248 parent = *p;
249 eb = rb_entry(parent, struct extent_buffer, rb_node);
250
251 if (offset < eb->start)
252 p = &(*p)->rb_left;
253 else if (offset > eb->start)
254 p = &(*p)->rb_right;
255 else
256 return eb;
257 }
258
259 rb_link_node(node, parent, p);
260 rb_insert_color(node, root);
261 return NULL;
262 }
263
264 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
265 u64 offset)
266 {
267 struct rb_root *root = &tree->buffer;
268 struct rb_node *n = root->rb_node;
269 struct extent_buffer *eb;
270
271 while (n) {
272 eb = rb_entry(n, struct extent_buffer, rb_node);
273 if (offset < eb->start)
274 n = n->rb_left;
275 else if (offset > eb->start)
276 n = n->rb_right;
277 else
278 return eb;
279 }
280 return NULL;
281 }
282
283 /*
284 * utility function to look for merge candidates inside a given range.
285 * Any extents with matching state are merged together into a single
286 * extent in the tree. Extents with EXTENT_IO in their state field
287 * are not merged because the end_io handlers need to be able to do
288 * operations on them without sleeping (or doing allocations/splits).
289 *
290 * This should be called with the tree lock held.
291 */
292 static int merge_state(struct extent_io_tree *tree,
293 struct extent_state *state)
294 {
295 struct extent_state *other;
296 struct rb_node *other_node;
297
298 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
299 return 0;
300
301 other_node = rb_prev(&state->rb_node);
302 if (other_node) {
303 other = rb_entry(other_node, struct extent_state, rb_node);
304 if (other->end == state->start - 1 &&
305 other->state == state->state) {
306 state->start = other->start;
307 other->tree = NULL;
308 rb_erase(&other->rb_node, &tree->state);
309 free_extent_state(other);
310 }
311 }
312 other_node = rb_next(&state->rb_node);
313 if (other_node) {
314 other = rb_entry(other_node, struct extent_state, rb_node);
315 if (other->start == state->end + 1 &&
316 other->state == state->state) {
317 other->start = state->start;
318 state->tree = NULL;
319 rb_erase(&state->rb_node, &tree->state);
320 free_extent_state(state);
321 }
322 }
323 return 0;
324 }
325
326 static void set_state_cb(struct extent_io_tree *tree,
327 struct extent_state *state,
328 unsigned long bits)
329 {
330 if (tree->ops && tree->ops->set_bit_hook) {
331 tree->ops->set_bit_hook(tree->mapping->host, state->start,
332 state->end, state->state, bits);
333 }
334 }
335
336 static void clear_state_cb(struct extent_io_tree *tree,
337 struct extent_state *state,
338 unsigned long bits)
339 {
340 if (tree->ops && tree->ops->clear_bit_hook) {
341 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
342 state->end, state->state, bits);
343 }
344 }
345
346 /*
347 * insert an extent_state struct into the tree. 'bits' are set on the
348 * struct before it is inserted.
349 *
350 * This may return -EEXIST if the extent is already there, in which case the
351 * state struct is freed.
352 *
353 * The tree lock is not taken internally. This is a utility function and
354 * probably isn't what you want to call (see set/clear_extent_bit).
355 */
356 static int insert_state(struct extent_io_tree *tree,
357 struct extent_state *state, u64 start, u64 end,
358 int bits)
359 {
360 struct rb_node *node;
361
362 if (end < start) {
363 printk(KERN_ERR "btrfs end < start %llu %llu\n",
364 (unsigned long long)end,
365 (unsigned long long)start);
366 WARN_ON(1);
367 }
368 if (bits & EXTENT_DIRTY)
369 tree->dirty_bytes += end - start + 1;
370 set_state_cb(tree, state, bits);
371 state->state |= bits;
372 state->start = start;
373 state->end = end;
374 node = tree_insert(&tree->state, end, &state->rb_node);
375 if (node) {
376 struct extent_state *found;
377 found = rb_entry(node, struct extent_state, rb_node);
378 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
379 "%llu %llu\n", (unsigned long long)found->start,
380 (unsigned long long)found->end,
381 (unsigned long long)start, (unsigned long long)end);
382 free_extent_state(state);
383 return -EEXIST;
384 }
385 state->tree = tree;
386 merge_state(tree, state);
387 return 0;
388 }
389
390 /*
391 * split a given extent state struct in two, inserting the preallocated
392 * struct 'prealloc' as the newly created second half. 'split' indicates an
393 * offset inside 'orig' where it should be split.
394 *
395 * Before calling,
396 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
397 * are two extent state structs in the tree:
398 * prealloc: [orig->start, split - 1]
399 * orig: [ split, orig->end ]
400 *
401 * The tree locks are not taken by this function. They need to be held
402 * by the caller.
403 */
404 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
405 struct extent_state *prealloc, u64 split)
406 {
407 struct rb_node *node;
408 prealloc->start = orig->start;
409 prealloc->end = split - 1;
410 prealloc->state = orig->state;
411 orig->start = split;
412
413 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
414 if (node) {
415 free_extent_state(prealloc);
416 return -EEXIST;
417 }
418 prealloc->tree = tree;
419 return 0;
420 }
421
422 /*
423 * utility function to clear some bits in an extent state struct.
424 * it will optionally wake up any one waiting on this state (wake == 1), or
425 * forcibly remove the state from the tree (delete == 1).
426 *
427 * If no bits are set on the state struct after clearing things, the
428 * struct is freed and removed from the tree
429 */
430 static int clear_state_bit(struct extent_io_tree *tree,
431 struct extent_state *state, int bits, int wake,
432 int delete)
433 {
434 int ret = state->state & bits;
435
436 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
437 u64 range = state->end - state->start + 1;
438 WARN_ON(range > tree->dirty_bytes);
439 tree->dirty_bytes -= range;
440 }
441 clear_state_cb(tree, state, bits);
442 state->state &= ~bits;
443 if (wake)
444 wake_up(&state->wq);
445 if (delete || state->state == 0) {
446 if (state->tree) {
447 clear_state_cb(tree, state, state->state);
448 rb_erase(&state->rb_node, &tree->state);
449 state->tree = NULL;
450 free_extent_state(state);
451 } else {
452 WARN_ON(1);
453 }
454 } else {
455 merge_state(tree, state);
456 }
457 return ret;
458 }
459
460 /*
461 * clear some bits on a range in the tree. This may require splitting
462 * or inserting elements in the tree, so the gfp mask is used to
463 * indicate which allocations or sleeping are allowed.
464 *
465 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
466 * the given range from the tree regardless of state (ie for truncate).
467 *
468 * the range [start, end] is inclusive.
469 *
470 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
471 * bits were already set, or zero if none of the bits were already set.
472 */
473 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
474 int bits, int wake, int delete, gfp_t mask)
475 {
476 struct extent_state *state;
477 struct extent_state *prealloc = NULL;
478 struct rb_node *node;
479 int err;
480 int set = 0;
481
482 again:
483 if (!prealloc && (mask & __GFP_WAIT)) {
484 prealloc = alloc_extent_state(mask);
485 if (!prealloc)
486 return -ENOMEM;
487 }
488
489 spin_lock(&tree->lock);
490 /*
491 * this search will find the extents that end after
492 * our range starts
493 */
494 node = tree_search(tree, start);
495 if (!node)
496 goto out;
497 state = rb_entry(node, struct extent_state, rb_node);
498 if (state->start > end)
499 goto out;
500 WARN_ON(state->end < start);
501
502 /*
503 * | ---- desired range ---- |
504 * | state | or
505 * | ------------- state -------------- |
506 *
507 * We need to split the extent we found, and may flip
508 * bits on second half.
509 *
510 * If the extent we found extends past our range, we
511 * just split and search again. It'll get split again
512 * the next time though.
513 *
514 * If the extent we found is inside our range, we clear
515 * the desired bit on it.
516 */
517
518 if (state->start < start) {
519 if (!prealloc)
520 prealloc = alloc_extent_state(GFP_ATOMIC);
521 err = split_state(tree, state, prealloc, start);
522 BUG_ON(err == -EEXIST);
523 prealloc = NULL;
524 if (err)
525 goto out;
526 if (state->end <= end) {
527 start = state->end + 1;
528 set |= clear_state_bit(tree, state, bits,
529 wake, delete);
530 } else {
531 start = state->start;
532 }
533 goto search_again;
534 }
535 /*
536 * | ---- desired range ---- |
537 * | state |
538 * We need to split the extent, and clear the bit
539 * on the first half
540 */
541 if (state->start <= end && state->end > end) {
542 if (!prealloc)
543 prealloc = alloc_extent_state(GFP_ATOMIC);
544 err = split_state(tree, state, prealloc, end + 1);
545 BUG_ON(err == -EEXIST);
546
547 if (wake)
548 wake_up(&state->wq);
549 set |= clear_state_bit(tree, prealloc, bits,
550 wake, delete);
551 prealloc = NULL;
552 goto out;
553 }
554
555 start = state->end + 1;
556 set |= clear_state_bit(tree, state, bits, wake, delete);
557 goto search_again;
558
559 out:
560 spin_unlock(&tree->lock);
561 if (prealloc)
562 free_extent_state(prealloc);
563
564 return set;
565
566 search_again:
567 if (start > end)
568 goto out;
569 spin_unlock(&tree->lock);
570 if (mask & __GFP_WAIT)
571 cond_resched();
572 goto again;
573 }
574
575 static int wait_on_state(struct extent_io_tree *tree,
576 struct extent_state *state)
577 __releases(tree->lock)
578 __acquires(tree->lock)
579 {
580 DEFINE_WAIT(wait);
581 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
582 spin_unlock(&tree->lock);
583 schedule();
584 spin_lock(&tree->lock);
585 finish_wait(&state->wq, &wait);
586 return 0;
587 }
588
589 /*
590 * waits for one or more bits to clear on a range in the state tree.
591 * The range [start, end] is inclusive.
592 * The tree lock is taken by this function
593 */
594 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
595 {
596 struct extent_state *state;
597 struct rb_node *node;
598
599 spin_lock(&tree->lock);
600 again:
601 while (1) {
602 /*
603 * this search will find all the extents that end after
604 * our range starts
605 */
606 node = tree_search(tree, start);
607 if (!node)
608 break;
609
610 state = rb_entry(node, struct extent_state, rb_node);
611
612 if (state->start > end)
613 goto out;
614
615 if (state->state & bits) {
616 start = state->start;
617 atomic_inc(&state->refs);
618 wait_on_state(tree, state);
619 free_extent_state(state);
620 goto again;
621 }
622 start = state->end + 1;
623
624 if (start > end)
625 break;
626
627 if (need_resched()) {
628 spin_unlock(&tree->lock);
629 cond_resched();
630 spin_lock(&tree->lock);
631 }
632 }
633 out:
634 spin_unlock(&tree->lock);
635 return 0;
636 }
637
638 static void set_state_bits(struct extent_io_tree *tree,
639 struct extent_state *state,
640 int bits)
641 {
642 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
643 u64 range = state->end - state->start + 1;
644 tree->dirty_bytes += range;
645 }
646 set_state_cb(tree, state, bits);
647 state->state |= bits;
648 }
649
650 /*
651 * set some bits on a range in the tree. This may require allocations
652 * or sleeping, so the gfp mask is used to indicate what is allowed.
653 *
654 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
655 * range already has the desired bits set. The start of the existing
656 * range is returned in failed_start in this case.
657 *
658 * [start, end] is inclusive
659 * This takes the tree lock.
660 */
661 static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
662 int bits, int exclusive, u64 *failed_start,
663 gfp_t mask)
664 {
665 struct extent_state *state;
666 struct extent_state *prealloc = NULL;
667 struct rb_node *node;
668 int err = 0;
669 int set;
670 u64 last_start;
671 u64 last_end;
672 again:
673 if (!prealloc && (mask & __GFP_WAIT)) {
674 prealloc = alloc_extent_state(mask);
675 if (!prealloc)
676 return -ENOMEM;
677 }
678
679 spin_lock(&tree->lock);
680 /*
681 * this search will find all the extents that end after
682 * our range starts.
683 */
684 node = tree_search(tree, start);
685 if (!node) {
686 err = insert_state(tree, prealloc, start, end, bits);
687 prealloc = NULL;
688 BUG_ON(err == -EEXIST);
689 goto out;
690 }
691
692 state = rb_entry(node, struct extent_state, rb_node);
693 last_start = state->start;
694 last_end = state->end;
695
696 /*
697 * | ---- desired range ---- |
698 * | state |
699 *
700 * Just lock what we found and keep going
701 */
702 if (state->start == start && state->end <= end) {
703 set = state->state & bits;
704 if (set && exclusive) {
705 *failed_start = state->start;
706 err = -EEXIST;
707 goto out;
708 }
709 set_state_bits(tree, state, bits);
710 start = state->end + 1;
711 merge_state(tree, state);
712 goto search_again;
713 }
714
715 /*
716 * | ---- desired range ---- |
717 * | state |
718 * or
719 * | ------------- state -------------- |
720 *
721 * We need to split the extent we found, and may flip bits on
722 * second half.
723 *
724 * If the extent we found extends past our
725 * range, we just split and search again. It'll get split
726 * again the next time though.
727 *
728 * If the extent we found is inside our range, we set the
729 * desired bit on it.
730 */
731 if (state->start < start) {
732 set = state->state & bits;
733 if (exclusive && set) {
734 *failed_start = start;
735 err = -EEXIST;
736 goto out;
737 }
738 err = split_state(tree, state, prealloc, start);
739 BUG_ON(err == -EEXIST);
740 prealloc = NULL;
741 if (err)
742 goto out;
743 if (state->end <= end) {
744 set_state_bits(tree, state, bits);
745 start = state->end + 1;
746 merge_state(tree, state);
747 } else {
748 start = state->start;
749 }
750 goto search_again;
751 }
752 /*
753 * | ---- desired range ---- |
754 * | state | or | state |
755 *
756 * There's a hole, we need to insert something in it and
757 * ignore the extent we found.
758 */
759 if (state->start > start) {
760 u64 this_end;
761 if (end < last_start)
762 this_end = end;
763 else
764 this_end = last_start - 1;
765 err = insert_state(tree, prealloc, start, this_end,
766 bits);
767 prealloc = NULL;
768 BUG_ON(err == -EEXIST);
769 if (err)
770 goto out;
771 start = this_end + 1;
772 goto search_again;
773 }
774 /*
775 * | ---- desired range ---- |
776 * | state |
777 * We need to split the extent, and set the bit
778 * on the first half
779 */
780 if (state->start <= end && state->end > end) {
781 set = state->state & bits;
782 if (exclusive && set) {
783 *failed_start = start;
784 err = -EEXIST;
785 goto out;
786 }
787 err = split_state(tree, state, prealloc, end + 1);
788 BUG_ON(err == -EEXIST);
789
790 set_state_bits(tree, prealloc, bits);
791 merge_state(tree, prealloc);
792 prealloc = NULL;
793 goto out;
794 }
795
796 goto search_again;
797
798 out:
799 spin_unlock(&tree->lock);
800 if (prealloc)
801 free_extent_state(prealloc);
802
803 return err;
804
805 search_again:
806 if (start > end)
807 goto out;
808 spin_unlock(&tree->lock);
809 if (mask & __GFP_WAIT)
810 cond_resched();
811 goto again;
812 }
813
814 /* wrappers around set/clear extent bit */
815 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
816 gfp_t mask)
817 {
818 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
819 mask);
820 }
821
822 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
823 gfp_t mask)
824 {
825 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
826 }
827
828 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
829 int bits, gfp_t mask)
830 {
831 return set_extent_bit(tree, start, end, bits, 0, NULL,
832 mask);
833 }
834
835 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
836 int bits, gfp_t mask)
837 {
838 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
839 }
840
841 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
842 gfp_t mask)
843 {
844 return set_extent_bit(tree, start, end,
845 EXTENT_DELALLOC | EXTENT_DIRTY,
846 0, NULL, mask);
847 }
848
849 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
850 gfp_t mask)
851 {
852 return clear_extent_bit(tree, start, end,
853 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
854 }
855
856 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
857 gfp_t mask)
858 {
859 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
860 }
861
862 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
863 gfp_t mask)
864 {
865 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
866 mask);
867 }
868
869 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
870 gfp_t mask)
871 {
872 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
873 }
874
875 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
876 gfp_t mask)
877 {
878 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
879 mask);
880 }
881
882 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
883 u64 end, gfp_t mask)
884 {
885 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
886 }
887
888 static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
889 gfp_t mask)
890 {
891 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
892 0, NULL, mask);
893 }
894
895 static int clear_extent_writeback(struct extent_io_tree *tree, u64 start,
896 u64 end, gfp_t mask)
897 {
898 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
899 }
900
901 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
902 {
903 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
904 }
905
906 /*
907 * either insert or lock state struct between start and end use mask to tell
908 * us if waiting is desired.
909 */
910 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
911 {
912 int err;
913 u64 failed_start;
914 while (1) {
915 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
916 &failed_start, mask);
917 if (err == -EEXIST && (mask & __GFP_WAIT)) {
918 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
919 start = failed_start;
920 } else {
921 break;
922 }
923 WARN_ON(start > end);
924 }
925 return err;
926 }
927
928 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
929 gfp_t mask)
930 {
931 int err;
932 u64 failed_start;
933
934 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
935 &failed_start, mask);
936 if (err == -EEXIST) {
937 if (failed_start > start)
938 clear_extent_bit(tree, start, failed_start - 1,
939 EXTENT_LOCKED, 1, 0, mask);
940 return 0;
941 }
942 return 1;
943 }
944
945 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
946 gfp_t mask)
947 {
948 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
949 }
950
951 /*
952 * helper function to set pages and extents in the tree dirty
953 */
954 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
955 {
956 unsigned long index = start >> PAGE_CACHE_SHIFT;
957 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
958 struct page *page;
959
960 while (index <= end_index) {
961 page = find_get_page(tree->mapping, index);
962 BUG_ON(!page);
963 __set_page_dirty_nobuffers(page);
964 page_cache_release(page);
965 index++;
966 }
967 set_extent_dirty(tree, start, end, GFP_NOFS);
968 return 0;
969 }
970
971 /*
972 * helper function to set both pages and extents in the tree writeback
973 */
974 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
975 {
976 unsigned long index = start >> PAGE_CACHE_SHIFT;
977 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
978 struct page *page;
979
980 while (index <= end_index) {
981 page = find_get_page(tree->mapping, index);
982 BUG_ON(!page);
983 set_page_writeback(page);
984 page_cache_release(page);
985 index++;
986 }
987 set_extent_writeback(tree, start, end, GFP_NOFS);
988 return 0;
989 }
990
991 /*
992 * find the first offset in the io tree with 'bits' set. zero is
993 * returned if we find something, and *start_ret and *end_ret are
994 * set to reflect the state struct that was found.
995 *
996 * If nothing was found, 1 is returned, < 0 on error
997 */
998 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
999 u64 *start_ret, u64 *end_ret, int bits)
1000 {
1001 struct rb_node *node;
1002 struct extent_state *state;
1003 int ret = 1;
1004
1005 spin_lock(&tree->lock);
1006 /*
1007 * this search will find all the extents that end after
1008 * our range starts.
1009 */
1010 node = tree_search(tree, start);
1011 if (!node)
1012 goto out;
1013
1014 while (1) {
1015 state = rb_entry(node, struct extent_state, rb_node);
1016 if (state->end >= start && (state->state & bits)) {
1017 *start_ret = state->start;
1018 *end_ret = state->end;
1019 ret = 0;
1020 break;
1021 }
1022 node = rb_next(node);
1023 if (!node)
1024 break;
1025 }
1026 out:
1027 spin_unlock(&tree->lock);
1028 return ret;
1029 }
1030
1031 /* find the first state struct with 'bits' set after 'start', and
1032 * return it. tree->lock must be held. NULL will returned if
1033 * nothing was found after 'start'
1034 */
1035 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1036 u64 start, int bits)
1037 {
1038 struct rb_node *node;
1039 struct extent_state *state;
1040
1041 /*
1042 * this search will find all the extents that end after
1043 * our range starts.
1044 */
1045 node = tree_search(tree, start);
1046 if (!node)
1047 goto out;
1048
1049 while (1) {
1050 state = rb_entry(node, struct extent_state, rb_node);
1051 if (state->end >= start && (state->state & bits))
1052 return state;
1053
1054 node = rb_next(node);
1055 if (!node)
1056 break;
1057 }
1058 out:
1059 return NULL;
1060 }
1061
1062 /*
1063 * find a contiguous range of bytes in the file marked as delalloc, not
1064 * more than 'max_bytes'. start and end are used to return the range,
1065 *
1066 * 1 is returned if we find something, 0 if nothing was in the tree
1067 */
1068 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1069 u64 *start, u64 *end, u64 max_bytes)
1070 {
1071 struct rb_node *node;
1072 struct extent_state *state;
1073 u64 cur_start = *start;
1074 u64 found = 0;
1075 u64 total_bytes = 0;
1076
1077 spin_lock(&tree->lock);
1078
1079 /*
1080 * this search will find all the extents that end after
1081 * our range starts.
1082 */
1083 node = tree_search(tree, cur_start);
1084 if (!node) {
1085 if (!found)
1086 *end = (u64)-1;
1087 goto out;
1088 }
1089
1090 while (1) {
1091 state = rb_entry(node, struct extent_state, rb_node);
1092 if (found && (state->start != cur_start ||
1093 (state->state & EXTENT_BOUNDARY))) {
1094 goto out;
1095 }
1096 if (!(state->state & EXTENT_DELALLOC)) {
1097 if (!found)
1098 *end = state->end;
1099 goto out;
1100 }
1101 if (!found)
1102 *start = state->start;
1103 found++;
1104 *end = state->end;
1105 cur_start = state->end + 1;
1106 node = rb_next(node);
1107 if (!node)
1108 break;
1109 total_bytes += state->end - state->start + 1;
1110 if (total_bytes >= max_bytes)
1111 break;
1112 }
1113 out:
1114 spin_unlock(&tree->lock);
1115 return found;
1116 }
1117
1118 static noinline int __unlock_for_delalloc(struct inode *inode,
1119 struct page *locked_page,
1120 u64 start, u64 end)
1121 {
1122 int ret;
1123 struct page *pages[16];
1124 unsigned long index = start >> PAGE_CACHE_SHIFT;
1125 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1126 unsigned long nr_pages = end_index - index + 1;
1127 int i;
1128
1129 if (index == locked_page->index && end_index == index)
1130 return 0;
1131
1132 while (nr_pages > 0) {
1133 ret = find_get_pages_contig(inode->i_mapping, index,
1134 min_t(unsigned long, nr_pages,
1135 ARRAY_SIZE(pages)), pages);
1136 for (i = 0; i < ret; i++) {
1137 if (pages[i] != locked_page)
1138 unlock_page(pages[i]);
1139 page_cache_release(pages[i]);
1140 }
1141 nr_pages -= ret;
1142 index += ret;
1143 cond_resched();
1144 }
1145 return 0;
1146 }
1147
1148 static noinline int lock_delalloc_pages(struct inode *inode,
1149 struct page *locked_page,
1150 u64 delalloc_start,
1151 u64 delalloc_end)
1152 {
1153 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1154 unsigned long start_index = index;
1155 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1156 unsigned long pages_locked = 0;
1157 struct page *pages[16];
1158 unsigned long nrpages;
1159 int ret;
1160 int i;
1161
1162 /* the caller is responsible for locking the start index */
1163 if (index == locked_page->index && index == end_index)
1164 return 0;
1165
1166 /* skip the page at the start index */
1167 nrpages = end_index - index + 1;
1168 while (nrpages > 0) {
1169 ret = find_get_pages_contig(inode->i_mapping, index,
1170 min_t(unsigned long,
1171 nrpages, ARRAY_SIZE(pages)), pages);
1172 if (ret == 0) {
1173 ret = -EAGAIN;
1174 goto done;
1175 }
1176 /* now we have an array of pages, lock them all */
1177 for (i = 0; i < ret; i++) {
1178 /*
1179 * the caller is taking responsibility for
1180 * locked_page
1181 */
1182 if (pages[i] != locked_page) {
1183 lock_page(pages[i]);
1184 if (!PageDirty(pages[i]) ||
1185 pages[i]->mapping != inode->i_mapping) {
1186 ret = -EAGAIN;
1187 unlock_page(pages[i]);
1188 page_cache_release(pages[i]);
1189 goto done;
1190 }
1191 }
1192 page_cache_release(pages[i]);
1193 pages_locked++;
1194 }
1195 nrpages -= ret;
1196 index += ret;
1197 cond_resched();
1198 }
1199 ret = 0;
1200 done:
1201 if (ret && pages_locked) {
1202 __unlock_for_delalloc(inode, locked_page,
1203 delalloc_start,
1204 ((u64)(start_index + pages_locked - 1)) <<
1205 PAGE_CACHE_SHIFT);
1206 }
1207 return ret;
1208 }
1209
1210 /*
1211 * find a contiguous range of bytes in the file marked as delalloc, not
1212 * more than 'max_bytes'. start and end are used to return the range,
1213 *
1214 * 1 is returned if we find something, 0 if nothing was in the tree
1215 */
1216 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1217 struct extent_io_tree *tree,
1218 struct page *locked_page,
1219 u64 *start, u64 *end,
1220 u64 max_bytes)
1221 {
1222 u64 delalloc_start;
1223 u64 delalloc_end;
1224 u64 found;
1225 int ret;
1226 int loops = 0;
1227
1228 again:
1229 /* step one, find a bunch of delalloc bytes starting at start */
1230 delalloc_start = *start;
1231 delalloc_end = 0;
1232 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1233 max_bytes);
1234 if (!found || delalloc_end <= *start) {
1235 *start = delalloc_start;
1236 *end = delalloc_end;
1237 return found;
1238 }
1239
1240 /*
1241 * start comes from the offset of locked_page. We have to lock
1242 * pages in order, so we can't process delalloc bytes before
1243 * locked_page
1244 */
1245 if (delalloc_start < *start)
1246 delalloc_start = *start;
1247
1248 /*
1249 * make sure to limit the number of pages we try to lock down
1250 * if we're looping.
1251 */
1252 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1253 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1254
1255 /* step two, lock all the pages after the page that has start */
1256 ret = lock_delalloc_pages(inode, locked_page,
1257 delalloc_start, delalloc_end);
1258 if (ret == -EAGAIN) {
1259 /* some of the pages are gone, lets avoid looping by
1260 * shortening the size of the delalloc range we're searching
1261 */
1262 if (!loops) {
1263 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1264 max_bytes = PAGE_CACHE_SIZE - offset;
1265 loops = 1;
1266 goto again;
1267 } else {
1268 found = 0;
1269 goto out_failed;
1270 }
1271 }
1272 BUG_ON(ret);
1273
1274 /* step three, lock the state bits for the whole range */
1275 lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1276
1277 /* then test to make sure it is all still delalloc */
1278 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1279 EXTENT_DELALLOC, 1);
1280 if (!ret) {
1281 unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1282 __unlock_for_delalloc(inode, locked_page,
1283 delalloc_start, delalloc_end);
1284 cond_resched();
1285 goto again;
1286 }
1287 *start = delalloc_start;
1288 *end = delalloc_end;
1289 out_failed:
1290 return found;
1291 }
1292
1293 int extent_clear_unlock_delalloc(struct inode *inode,
1294 struct extent_io_tree *tree,
1295 u64 start, u64 end, struct page *locked_page,
1296 int unlock_pages,
1297 int clear_unlock,
1298 int clear_delalloc, int clear_dirty,
1299 int set_writeback,
1300 int end_writeback)
1301 {
1302 int ret;
1303 struct page *pages[16];
1304 unsigned long index = start >> PAGE_CACHE_SHIFT;
1305 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1306 unsigned long nr_pages = end_index - index + 1;
1307 int i;
1308 int clear_bits = 0;
1309
1310 if (clear_unlock)
1311 clear_bits |= EXTENT_LOCKED;
1312 if (clear_dirty)
1313 clear_bits |= EXTENT_DIRTY;
1314
1315 if (clear_delalloc)
1316 clear_bits |= EXTENT_DELALLOC;
1317
1318 clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
1319 if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
1320 return 0;
1321
1322 while (nr_pages > 0) {
1323 ret = find_get_pages_contig(inode->i_mapping, index,
1324 min_t(unsigned long,
1325 nr_pages, ARRAY_SIZE(pages)), pages);
1326 for (i = 0; i < ret; i++) {
1327 if (pages[i] == locked_page) {
1328 page_cache_release(pages[i]);
1329 continue;
1330 }
1331 if (clear_dirty)
1332 clear_page_dirty_for_io(pages[i]);
1333 if (set_writeback)
1334 set_page_writeback(pages[i]);
1335 if (end_writeback)
1336 end_page_writeback(pages[i]);
1337 if (unlock_pages)
1338 unlock_page(pages[i]);
1339 page_cache_release(pages[i]);
1340 }
1341 nr_pages -= ret;
1342 index += ret;
1343 cond_resched();
1344 }
1345 return 0;
1346 }
1347
1348 /*
1349 * count the number of bytes in the tree that have a given bit(s)
1350 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1351 * cached. The total number found is returned.
1352 */
1353 u64 count_range_bits(struct extent_io_tree *tree,
1354 u64 *start, u64 search_end, u64 max_bytes,
1355 unsigned long bits)
1356 {
1357 struct rb_node *node;
1358 struct extent_state *state;
1359 u64 cur_start = *start;
1360 u64 total_bytes = 0;
1361 int found = 0;
1362
1363 if (search_end <= cur_start) {
1364 WARN_ON(1);
1365 return 0;
1366 }
1367
1368 spin_lock(&tree->lock);
1369 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1370 total_bytes = tree->dirty_bytes;
1371 goto out;
1372 }
1373 /*
1374 * this search will find all the extents that end after
1375 * our range starts.
1376 */
1377 node = tree_search(tree, cur_start);
1378 if (!node)
1379 goto out;
1380
1381 while (1) {
1382 state = rb_entry(node, struct extent_state, rb_node);
1383 if (state->start > search_end)
1384 break;
1385 if (state->end >= cur_start && (state->state & bits)) {
1386 total_bytes += min(search_end, state->end) + 1 -
1387 max(cur_start, state->start);
1388 if (total_bytes >= max_bytes)
1389 break;
1390 if (!found) {
1391 *start = state->start;
1392 found = 1;
1393 }
1394 }
1395 node = rb_next(node);
1396 if (!node)
1397 break;
1398 }
1399 out:
1400 spin_unlock(&tree->lock);
1401 return total_bytes;
1402 }
1403
1404 #if 0
1405 /*
1406 * helper function to lock both pages and extents in the tree.
1407 * pages must be locked first.
1408 */
1409 static int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1410 {
1411 unsigned long index = start >> PAGE_CACHE_SHIFT;
1412 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1413 struct page *page;
1414 int err;
1415
1416 while (index <= end_index) {
1417 page = grab_cache_page(tree->mapping, index);
1418 if (!page) {
1419 err = -ENOMEM;
1420 goto failed;
1421 }
1422 if (IS_ERR(page)) {
1423 err = PTR_ERR(page);
1424 goto failed;
1425 }
1426 index++;
1427 }
1428 lock_extent(tree, start, end, GFP_NOFS);
1429 return 0;
1430
1431 failed:
1432 /*
1433 * we failed above in getting the page at 'index', so we undo here
1434 * up to but not including the page at 'index'
1435 */
1436 end_index = index;
1437 index = start >> PAGE_CACHE_SHIFT;
1438 while (index < end_index) {
1439 page = find_get_page(tree->mapping, index);
1440 unlock_page(page);
1441 page_cache_release(page);
1442 index++;
1443 }
1444 return err;
1445 }
1446
1447 /*
1448 * helper function to unlock both pages and extents in the tree.
1449 */
1450 static int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1451 {
1452 unsigned long index = start >> PAGE_CACHE_SHIFT;
1453 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1454 struct page *page;
1455
1456 while (index <= end_index) {
1457 page = find_get_page(tree->mapping, index);
1458 unlock_page(page);
1459 page_cache_release(page);
1460 index++;
1461 }
1462 unlock_extent(tree, start, end, GFP_NOFS);
1463 return 0;
1464 }
1465 #endif
1466
1467 /*
1468 * set the private field for a given byte offset in the tree. If there isn't
1469 * an extent_state there already, this does nothing.
1470 */
1471 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1472 {
1473 struct rb_node *node;
1474 struct extent_state *state;
1475 int ret = 0;
1476
1477 spin_lock(&tree->lock);
1478 /*
1479 * this search will find all the extents that end after
1480 * our range starts.
1481 */
1482 node = tree_search(tree, start);
1483 if (!node) {
1484 ret = -ENOENT;
1485 goto out;
1486 }
1487 state = rb_entry(node, struct extent_state, rb_node);
1488 if (state->start != start) {
1489 ret = -ENOENT;
1490 goto out;
1491 }
1492 state->private = private;
1493 out:
1494 spin_unlock(&tree->lock);
1495 return ret;
1496 }
1497
1498 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1499 {
1500 struct rb_node *node;
1501 struct extent_state *state;
1502 int ret = 0;
1503
1504 spin_lock(&tree->lock);
1505 /*
1506 * this search will find all the extents that end after
1507 * our range starts.
1508 */
1509 node = tree_search(tree, start);
1510 if (!node) {
1511 ret = -ENOENT;
1512 goto out;
1513 }
1514 state = rb_entry(node, struct extent_state, rb_node);
1515 if (state->start != start) {
1516 ret = -ENOENT;
1517 goto out;
1518 }
1519 *private = state->private;
1520 out:
1521 spin_unlock(&tree->lock);
1522 return ret;
1523 }
1524
1525 /*
1526 * searches a range in the state tree for a given mask.
1527 * If 'filled' == 1, this returns 1 only if every extent in the tree
1528 * has the bits set. Otherwise, 1 is returned if any bit in the
1529 * range is found set.
1530 */
1531 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1532 int bits, int filled)
1533 {
1534 struct extent_state *state = NULL;
1535 struct rb_node *node;
1536 int bitset = 0;
1537
1538 spin_lock(&tree->lock);
1539 node = tree_search(tree, start);
1540 while (node && start <= end) {
1541 state = rb_entry(node, struct extent_state, rb_node);
1542
1543 if (filled && state->start > start) {
1544 bitset = 0;
1545 break;
1546 }
1547
1548 if (state->start > end)
1549 break;
1550
1551 if (state->state & bits) {
1552 bitset = 1;
1553 if (!filled)
1554 break;
1555 } else if (filled) {
1556 bitset = 0;
1557 break;
1558 }
1559 start = state->end + 1;
1560 if (start > end)
1561 break;
1562 node = rb_next(node);
1563 if (!node) {
1564 if (filled)
1565 bitset = 0;
1566 break;
1567 }
1568 }
1569 spin_unlock(&tree->lock);
1570 return bitset;
1571 }
1572
1573 /*
1574 * helper function to set a given page up to date if all the
1575 * extents in the tree for that page are up to date
1576 */
1577 static int check_page_uptodate(struct extent_io_tree *tree,
1578 struct page *page)
1579 {
1580 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1581 u64 end = start + PAGE_CACHE_SIZE - 1;
1582 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1583 SetPageUptodate(page);
1584 return 0;
1585 }
1586
1587 /*
1588 * helper function to unlock a page if all the extents in the tree
1589 * for that page are unlocked
1590 */
1591 static int check_page_locked(struct extent_io_tree *tree,
1592 struct page *page)
1593 {
1594 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1595 u64 end = start + PAGE_CACHE_SIZE - 1;
1596 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1597 unlock_page(page);
1598 return 0;
1599 }
1600
1601 /*
1602 * helper function to end page writeback if all the extents
1603 * in the tree for that page are done with writeback
1604 */
1605 static int check_page_writeback(struct extent_io_tree *tree,
1606 struct page *page)
1607 {
1608 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1609 u64 end = start + PAGE_CACHE_SIZE - 1;
1610 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1611 end_page_writeback(page);
1612 return 0;
1613 }
1614
1615 /* lots and lots of room for performance fixes in the end_bio funcs */
1616
1617 /*
1618 * after a writepage IO is done, we need to:
1619 * clear the uptodate bits on error
1620 * clear the writeback bits in the extent tree for this IO
1621 * end_page_writeback if the page has no more pending IO
1622 *
1623 * Scheduling is not allowed, so the extent state tree is expected
1624 * to have one and only one object corresponding to this IO.
1625 */
1626 static void end_bio_extent_writepage(struct bio *bio, int err)
1627 {
1628 int uptodate = err == 0;
1629 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1630 struct extent_io_tree *tree;
1631 u64 start;
1632 u64 end;
1633 int whole_page;
1634 int ret;
1635
1636 do {
1637 struct page *page = bvec->bv_page;
1638 tree = &BTRFS_I(page->mapping->host)->io_tree;
1639
1640 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1641 bvec->bv_offset;
1642 end = start + bvec->bv_len - 1;
1643
1644 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1645 whole_page = 1;
1646 else
1647 whole_page = 0;
1648
1649 if (--bvec >= bio->bi_io_vec)
1650 prefetchw(&bvec->bv_page->flags);
1651 if (tree->ops && tree->ops->writepage_end_io_hook) {
1652 ret = tree->ops->writepage_end_io_hook(page, start,
1653 end, NULL, uptodate);
1654 if (ret)
1655 uptodate = 0;
1656 }
1657
1658 if (!uptodate && tree->ops &&
1659 tree->ops->writepage_io_failed_hook) {
1660 ret = tree->ops->writepage_io_failed_hook(bio, page,
1661 start, end, NULL);
1662 if (ret == 0) {
1663 uptodate = (err == 0);
1664 continue;
1665 }
1666 }
1667
1668 if (!uptodate) {
1669 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1670 ClearPageUptodate(page);
1671 SetPageError(page);
1672 }
1673
1674 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1675
1676 if (whole_page)
1677 end_page_writeback(page);
1678 else
1679 check_page_writeback(tree, page);
1680 } while (bvec >= bio->bi_io_vec);
1681
1682 bio_put(bio);
1683 }
1684
1685 /*
1686 * after a readpage IO is done, we need to:
1687 * clear the uptodate bits on error
1688 * set the uptodate bits if things worked
1689 * set the page up to date if all extents in the tree are uptodate
1690 * clear the lock bit in the extent tree
1691 * unlock the page if there are no other extents locked for it
1692 *
1693 * Scheduling is not allowed, so the extent state tree is expected
1694 * to have one and only one object corresponding to this IO.
1695 */
1696 static void end_bio_extent_readpage(struct bio *bio, int err)
1697 {
1698 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1699 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1700 struct extent_io_tree *tree;
1701 u64 start;
1702 u64 end;
1703 int whole_page;
1704 int ret;
1705
1706 if (err)
1707 uptodate = 0;
1708
1709 do {
1710 struct page *page = bvec->bv_page;
1711 tree = &BTRFS_I(page->mapping->host)->io_tree;
1712
1713 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1714 bvec->bv_offset;
1715 end = start + bvec->bv_len - 1;
1716
1717 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1718 whole_page = 1;
1719 else
1720 whole_page = 0;
1721
1722 if (--bvec >= bio->bi_io_vec)
1723 prefetchw(&bvec->bv_page->flags);
1724
1725 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1726 ret = tree->ops->readpage_end_io_hook(page, start, end,
1727 NULL);
1728 if (ret)
1729 uptodate = 0;
1730 }
1731 if (!uptodate && tree->ops &&
1732 tree->ops->readpage_io_failed_hook) {
1733 ret = tree->ops->readpage_io_failed_hook(bio, page,
1734 start, end, NULL);
1735 if (ret == 0) {
1736 uptodate =
1737 test_bit(BIO_UPTODATE, &bio->bi_flags);
1738 if (err)
1739 uptodate = 0;
1740 continue;
1741 }
1742 }
1743
1744 if (uptodate) {
1745 set_extent_uptodate(tree, start, end,
1746 GFP_ATOMIC);
1747 }
1748 unlock_extent(tree, start, end, GFP_ATOMIC);
1749
1750 if (whole_page) {
1751 if (uptodate) {
1752 SetPageUptodate(page);
1753 } else {
1754 ClearPageUptodate(page);
1755 SetPageError(page);
1756 }
1757 unlock_page(page);
1758 } else {
1759 if (uptodate) {
1760 check_page_uptodate(tree, page);
1761 } else {
1762 ClearPageUptodate(page);
1763 SetPageError(page);
1764 }
1765 check_page_locked(tree, page);
1766 }
1767 } while (bvec >= bio->bi_io_vec);
1768
1769 bio_put(bio);
1770 }
1771
1772 /*
1773 * IO done from prepare_write is pretty simple, we just unlock
1774 * the structs in the extent tree when done, and set the uptodate bits
1775 * as appropriate.
1776 */
1777 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1778 {
1779 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1780 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1781 struct extent_io_tree *tree;
1782 u64 start;
1783 u64 end;
1784
1785 do {
1786 struct page *page = bvec->bv_page;
1787 tree = &BTRFS_I(page->mapping->host)->io_tree;
1788
1789 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1790 bvec->bv_offset;
1791 end = start + bvec->bv_len - 1;
1792
1793 if (--bvec >= bio->bi_io_vec)
1794 prefetchw(&bvec->bv_page->flags);
1795
1796 if (uptodate) {
1797 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1798 } else {
1799 ClearPageUptodate(page);
1800 SetPageError(page);
1801 }
1802
1803 unlock_extent(tree, start, end, GFP_ATOMIC);
1804
1805 } while (bvec >= bio->bi_io_vec);
1806
1807 bio_put(bio);
1808 }
1809
1810 static struct bio *
1811 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1812 gfp_t gfp_flags)
1813 {
1814 struct bio *bio;
1815
1816 bio = bio_alloc(gfp_flags, nr_vecs);
1817
1818 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1819 while (!bio && (nr_vecs /= 2))
1820 bio = bio_alloc(gfp_flags, nr_vecs);
1821 }
1822
1823 if (bio) {
1824 bio->bi_size = 0;
1825 bio->bi_bdev = bdev;
1826 bio->bi_sector = first_sector;
1827 }
1828 return bio;
1829 }
1830
1831 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1832 unsigned long bio_flags)
1833 {
1834 int ret = 0;
1835 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1836 struct page *page = bvec->bv_page;
1837 struct extent_io_tree *tree = bio->bi_private;
1838 u64 start;
1839 u64 end;
1840
1841 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1842 end = start + bvec->bv_len - 1;
1843
1844 bio->bi_private = NULL;
1845
1846 bio_get(bio);
1847
1848 if (tree->ops && tree->ops->submit_bio_hook)
1849 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1850 mirror_num, bio_flags);
1851 else
1852 submit_bio(rw, bio);
1853 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1854 ret = -EOPNOTSUPP;
1855 bio_put(bio);
1856 return ret;
1857 }
1858
1859 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1860 struct page *page, sector_t sector,
1861 size_t size, unsigned long offset,
1862 struct block_device *bdev,
1863 struct bio **bio_ret,
1864 unsigned long max_pages,
1865 bio_end_io_t end_io_func,
1866 int mirror_num,
1867 unsigned long prev_bio_flags,
1868 unsigned long bio_flags)
1869 {
1870 int ret = 0;
1871 struct bio *bio;
1872 int nr;
1873 int contig = 0;
1874 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1875 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1876 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1877
1878 if (bio_ret && *bio_ret) {
1879 bio = *bio_ret;
1880 if (old_compressed)
1881 contig = bio->bi_sector == sector;
1882 else
1883 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1884 sector;
1885
1886 if (prev_bio_flags != bio_flags || !contig ||
1887 (tree->ops && tree->ops->merge_bio_hook &&
1888 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1889 bio_flags)) ||
1890 bio_add_page(bio, page, page_size, offset) < page_size) {
1891 ret = submit_one_bio(rw, bio, mirror_num,
1892 prev_bio_flags);
1893 bio = NULL;
1894 } else {
1895 return 0;
1896 }
1897 }
1898 if (this_compressed)
1899 nr = BIO_MAX_PAGES;
1900 else
1901 nr = bio_get_nr_vecs(bdev);
1902
1903 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1904
1905 bio_add_page(bio, page, page_size, offset);
1906 bio->bi_end_io = end_io_func;
1907 bio->bi_private = tree;
1908
1909 if (bio_ret)
1910 *bio_ret = bio;
1911 else
1912 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1913
1914 return ret;
1915 }
1916
1917 void set_page_extent_mapped(struct page *page)
1918 {
1919 if (!PagePrivate(page)) {
1920 SetPagePrivate(page);
1921 page_cache_get(page);
1922 set_page_private(page, EXTENT_PAGE_PRIVATE);
1923 }
1924 }
1925
1926 static void set_page_extent_head(struct page *page, unsigned long len)
1927 {
1928 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1929 }
1930
1931 /*
1932 * basic readpage implementation. Locked extent state structs are inserted
1933 * into the tree that are removed when the IO is done (by the end_io
1934 * handlers)
1935 */
1936 static int __extent_read_full_page(struct extent_io_tree *tree,
1937 struct page *page,
1938 get_extent_t *get_extent,
1939 struct bio **bio, int mirror_num,
1940 unsigned long *bio_flags)
1941 {
1942 struct inode *inode = page->mapping->host;
1943 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1944 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1945 u64 end;
1946 u64 cur = start;
1947 u64 extent_offset;
1948 u64 last_byte = i_size_read(inode);
1949 u64 block_start;
1950 u64 cur_end;
1951 sector_t sector;
1952 struct extent_map *em;
1953 struct block_device *bdev;
1954 int ret;
1955 int nr = 0;
1956 size_t page_offset = 0;
1957 size_t iosize;
1958 size_t disk_io_size;
1959 size_t blocksize = inode->i_sb->s_blocksize;
1960 unsigned long this_bio_flag = 0;
1961
1962 set_page_extent_mapped(page);
1963
1964 end = page_end;
1965 lock_extent(tree, start, end, GFP_NOFS);
1966
1967 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1968 char *userpage;
1969 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
1970
1971 if (zero_offset) {
1972 iosize = PAGE_CACHE_SIZE - zero_offset;
1973 userpage = kmap_atomic(page, KM_USER0);
1974 memset(userpage + zero_offset, 0, iosize);
1975 flush_dcache_page(page);
1976 kunmap_atomic(userpage, KM_USER0);
1977 }
1978 }
1979 while (cur <= end) {
1980 if (cur >= last_byte) {
1981 char *userpage;
1982 iosize = PAGE_CACHE_SIZE - page_offset;
1983 userpage = kmap_atomic(page, KM_USER0);
1984 memset(userpage + page_offset, 0, iosize);
1985 flush_dcache_page(page);
1986 kunmap_atomic(userpage, KM_USER0);
1987 set_extent_uptodate(tree, cur, cur + iosize - 1,
1988 GFP_NOFS);
1989 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1990 break;
1991 }
1992 em = get_extent(inode, page, page_offset, cur,
1993 end - cur + 1, 0);
1994 if (IS_ERR(em) || !em) {
1995 SetPageError(page);
1996 unlock_extent(tree, cur, end, GFP_NOFS);
1997 break;
1998 }
1999 extent_offset = cur - em->start;
2000 BUG_ON(extent_map_end(em) <= cur);
2001 BUG_ON(end < cur);
2002
2003 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2004 this_bio_flag = EXTENT_BIO_COMPRESSED;
2005
2006 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2007 cur_end = min(extent_map_end(em) - 1, end);
2008 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2009 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2010 disk_io_size = em->block_len;
2011 sector = em->block_start >> 9;
2012 } else {
2013 sector = (em->block_start + extent_offset) >> 9;
2014 disk_io_size = iosize;
2015 }
2016 bdev = em->bdev;
2017 block_start = em->block_start;
2018 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2019 block_start = EXTENT_MAP_HOLE;
2020 free_extent_map(em);
2021 em = NULL;
2022
2023 /* we've found a hole, just zero and go on */
2024 if (block_start == EXTENT_MAP_HOLE) {
2025 char *userpage;
2026 userpage = kmap_atomic(page, KM_USER0);
2027 memset(userpage + page_offset, 0, iosize);
2028 flush_dcache_page(page);
2029 kunmap_atomic(userpage, KM_USER0);
2030
2031 set_extent_uptodate(tree, cur, cur + iosize - 1,
2032 GFP_NOFS);
2033 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2034 cur = cur + iosize;
2035 page_offset += iosize;
2036 continue;
2037 }
2038 /* the get_extent function already copied into the page */
2039 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
2040 check_page_uptodate(tree, page);
2041 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2042 cur = cur + iosize;
2043 page_offset += iosize;
2044 continue;
2045 }
2046 /* we have an inline extent but it didn't get marked up
2047 * to date. Error out
2048 */
2049 if (block_start == EXTENT_MAP_INLINE) {
2050 SetPageError(page);
2051 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2052 cur = cur + iosize;
2053 page_offset += iosize;
2054 continue;
2055 }
2056
2057 ret = 0;
2058 if (tree->ops && tree->ops->readpage_io_hook) {
2059 ret = tree->ops->readpage_io_hook(page, cur,
2060 cur + iosize - 1);
2061 }
2062 if (!ret) {
2063 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2064 pnr -= page->index;
2065 ret = submit_extent_page(READ, tree, page,
2066 sector, disk_io_size, page_offset,
2067 bdev, bio, pnr,
2068 end_bio_extent_readpage, mirror_num,
2069 *bio_flags,
2070 this_bio_flag);
2071 nr++;
2072 *bio_flags = this_bio_flag;
2073 }
2074 if (ret)
2075 SetPageError(page);
2076 cur = cur + iosize;
2077 page_offset += iosize;
2078 }
2079 if (!nr) {
2080 if (!PageError(page))
2081 SetPageUptodate(page);
2082 unlock_page(page);
2083 }
2084 return 0;
2085 }
2086
2087 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2088 get_extent_t *get_extent)
2089 {
2090 struct bio *bio = NULL;
2091 unsigned long bio_flags = 0;
2092 int ret;
2093
2094 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2095 &bio_flags);
2096 if (bio)
2097 submit_one_bio(READ, bio, 0, bio_flags);
2098 return ret;
2099 }
2100
2101 static noinline void update_nr_written(struct page *page,
2102 struct writeback_control *wbc,
2103 unsigned long nr_written)
2104 {
2105 wbc->nr_to_write -= nr_written;
2106 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2107 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2108 page->mapping->writeback_index = page->index + nr_written;
2109 }
2110
2111 /*
2112 * the writepage semantics are similar to regular writepage. extent
2113 * records are inserted to lock ranges in the tree, and as dirty areas
2114 * are found, they are marked writeback. Then the lock bits are removed
2115 * and the end_io handler clears the writeback ranges
2116 */
2117 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2118 void *data)
2119 {
2120 struct inode *inode = page->mapping->host;
2121 struct extent_page_data *epd = data;
2122 struct extent_io_tree *tree = epd->tree;
2123 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2124 u64 delalloc_start;
2125 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2126 u64 end;
2127 u64 cur = start;
2128 u64 extent_offset;
2129 u64 last_byte = i_size_read(inode);
2130 u64 block_start;
2131 u64 iosize;
2132 u64 unlock_start;
2133 sector_t sector;
2134 struct extent_map *em;
2135 struct block_device *bdev;
2136 int ret;
2137 int nr = 0;
2138 size_t pg_offset = 0;
2139 size_t blocksize;
2140 loff_t i_size = i_size_read(inode);
2141 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2142 u64 nr_delalloc;
2143 u64 delalloc_end;
2144 int page_started;
2145 int compressed;
2146 int write_flags;
2147 unsigned long nr_written = 0;
2148
2149 if (wbc->sync_mode == WB_SYNC_ALL)
2150 write_flags = WRITE_SYNC_PLUG;
2151 else
2152 write_flags = WRITE;
2153
2154 WARN_ON(!PageLocked(page));
2155 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2156 if (page->index > end_index ||
2157 (page->index == end_index && !pg_offset)) {
2158 page->mapping->a_ops->invalidatepage(page, 0);
2159 unlock_page(page);
2160 return 0;
2161 }
2162
2163 if (page->index == end_index) {
2164 char *userpage;
2165
2166 userpage = kmap_atomic(page, KM_USER0);
2167 memset(userpage + pg_offset, 0,
2168 PAGE_CACHE_SIZE - pg_offset);
2169 kunmap_atomic(userpage, KM_USER0);
2170 flush_dcache_page(page);
2171 }
2172 pg_offset = 0;
2173
2174 set_page_extent_mapped(page);
2175
2176 delalloc_start = start;
2177 delalloc_end = 0;
2178 page_started = 0;
2179 if (!epd->extent_locked) {
2180 /*
2181 * make sure the wbc mapping index is at least updated
2182 * to this page.
2183 */
2184 update_nr_written(page, wbc, 0);
2185
2186 while (delalloc_end < page_end) {
2187 nr_delalloc = find_lock_delalloc_range(inode, tree,
2188 page,
2189 &delalloc_start,
2190 &delalloc_end,
2191 128 * 1024 * 1024);
2192 if (nr_delalloc == 0) {
2193 delalloc_start = delalloc_end + 1;
2194 continue;
2195 }
2196 tree->ops->fill_delalloc(inode, page, delalloc_start,
2197 delalloc_end, &page_started,
2198 &nr_written);
2199 delalloc_start = delalloc_end + 1;
2200 }
2201
2202 /* did the fill delalloc function already unlock and start
2203 * the IO?
2204 */
2205 if (page_started) {
2206 ret = 0;
2207 /*
2208 * we've unlocked the page, so we can't update
2209 * the mapping's writeback index, just update
2210 * nr_to_write.
2211 */
2212 wbc->nr_to_write -= nr_written;
2213 goto done_unlocked;
2214 }
2215 }
2216 lock_extent(tree, start, page_end, GFP_NOFS);
2217
2218 unlock_start = start;
2219
2220 if (tree->ops && tree->ops->writepage_start_hook) {
2221 ret = tree->ops->writepage_start_hook(page, start,
2222 page_end);
2223 if (ret == -EAGAIN) {
2224 unlock_extent(tree, start, page_end, GFP_NOFS);
2225 redirty_page_for_writepage(wbc, page);
2226 update_nr_written(page, wbc, nr_written);
2227 unlock_page(page);
2228 ret = 0;
2229 goto done_unlocked;
2230 }
2231 }
2232
2233 /*
2234 * we don't want to touch the inode after unlocking the page,
2235 * so we update the mapping writeback index now
2236 */
2237 update_nr_written(page, wbc, nr_written + 1);
2238
2239 end = page_end;
2240 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
2241 printk(KERN_ERR "btrfs delalloc bits after lock_extent\n");
2242
2243 if (last_byte <= start) {
2244 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2245 unlock_extent(tree, start, page_end, GFP_NOFS);
2246 if (tree->ops && tree->ops->writepage_end_io_hook)
2247 tree->ops->writepage_end_io_hook(page, start,
2248 page_end, NULL, 1);
2249 unlock_start = page_end + 1;
2250 goto done;
2251 }
2252
2253 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2254 blocksize = inode->i_sb->s_blocksize;
2255
2256 while (cur <= end) {
2257 if (cur >= last_byte) {
2258 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2259 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2260 if (tree->ops && tree->ops->writepage_end_io_hook)
2261 tree->ops->writepage_end_io_hook(page, cur,
2262 page_end, NULL, 1);
2263 unlock_start = page_end + 1;
2264 break;
2265 }
2266 em = epd->get_extent(inode, page, pg_offset, cur,
2267 end - cur + 1, 1);
2268 if (IS_ERR(em) || !em) {
2269 SetPageError(page);
2270 break;
2271 }
2272
2273 extent_offset = cur - em->start;
2274 BUG_ON(extent_map_end(em) <= cur);
2275 BUG_ON(end < cur);
2276 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2277 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2278 sector = (em->block_start + extent_offset) >> 9;
2279 bdev = em->bdev;
2280 block_start = em->block_start;
2281 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2282 free_extent_map(em);
2283 em = NULL;
2284
2285 /*
2286 * compressed and inline extents are written through other
2287 * paths in the FS
2288 */
2289 if (compressed || block_start == EXTENT_MAP_HOLE ||
2290 block_start == EXTENT_MAP_INLINE) {
2291 clear_extent_dirty(tree, cur,
2292 cur + iosize - 1, GFP_NOFS);
2293
2294 unlock_extent(tree, unlock_start, cur + iosize - 1,
2295 GFP_NOFS);
2296
2297 /*
2298 * end_io notification does not happen here for
2299 * compressed extents
2300 */
2301 if (!compressed && tree->ops &&
2302 tree->ops->writepage_end_io_hook)
2303 tree->ops->writepage_end_io_hook(page, cur,
2304 cur + iosize - 1,
2305 NULL, 1);
2306 else if (compressed) {
2307 /* we don't want to end_page_writeback on
2308 * a compressed extent. this happens
2309 * elsewhere
2310 */
2311 nr++;
2312 }
2313
2314 cur += iosize;
2315 pg_offset += iosize;
2316 unlock_start = cur;
2317 continue;
2318 }
2319 /* leave this out until we have a page_mkwrite call */
2320 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2321 EXTENT_DIRTY, 0)) {
2322 cur = cur + iosize;
2323 pg_offset += iosize;
2324 continue;
2325 }
2326
2327 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2328 if (tree->ops && tree->ops->writepage_io_hook) {
2329 ret = tree->ops->writepage_io_hook(page, cur,
2330 cur + iosize - 1);
2331 } else {
2332 ret = 0;
2333 }
2334 if (ret) {
2335 SetPageError(page);
2336 } else {
2337 unsigned long max_nr = end_index + 1;
2338
2339 set_range_writeback(tree, cur, cur + iosize - 1);
2340 if (!PageWriteback(page)) {
2341 printk(KERN_ERR "btrfs warning page %lu not "
2342 "writeback, cur %llu end %llu\n",
2343 page->index, (unsigned long long)cur,
2344 (unsigned long long)end);
2345 }
2346
2347 ret = submit_extent_page(write_flags, tree, page,
2348 sector, iosize, pg_offset,
2349 bdev, &epd->bio, max_nr,
2350 end_bio_extent_writepage,
2351 0, 0, 0);
2352 if (ret)
2353 SetPageError(page);
2354 }
2355 cur = cur + iosize;
2356 pg_offset += iosize;
2357 nr++;
2358 }
2359 done:
2360 if (nr == 0) {
2361 /* make sure the mapping tag for page dirty gets cleared */
2362 set_page_writeback(page);
2363 end_page_writeback(page);
2364 }
2365 if (unlock_start <= page_end)
2366 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2367 unlock_page(page);
2368
2369 done_unlocked:
2370
2371 return 0;
2372 }
2373
2374 /**
2375 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2376 * @mapping: address space structure to write
2377 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2378 * @writepage: function called for each page
2379 * @data: data passed to writepage function
2380 *
2381 * If a page is already under I/O, write_cache_pages() skips it, even
2382 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2383 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2384 * and msync() need to guarantee that all the data which was dirty at the time
2385 * the call was made get new I/O started against them. If wbc->sync_mode is
2386 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2387 * existing IO to complete.
2388 */
2389 static int extent_write_cache_pages(struct extent_io_tree *tree,
2390 struct address_space *mapping,
2391 struct writeback_control *wbc,
2392 writepage_t writepage, void *data,
2393 void (*flush_fn)(void *))
2394 {
2395 struct backing_dev_info *bdi = mapping->backing_dev_info;
2396 int ret = 0;
2397 int done = 0;
2398 struct pagevec pvec;
2399 int nr_pages;
2400 pgoff_t index;
2401 pgoff_t end; /* Inclusive */
2402 int scanned = 0;
2403 int range_whole = 0;
2404
2405 pagevec_init(&pvec, 0);
2406 if (wbc->range_cyclic) {
2407 index = mapping->writeback_index; /* Start from prev offset */
2408 end = -1;
2409 } else {
2410 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2411 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2412 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2413 range_whole = 1;
2414 scanned = 1;
2415 }
2416 retry:
2417 while (!done && (index <= end) &&
2418 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2419 PAGECACHE_TAG_DIRTY, min(end - index,
2420 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2421 unsigned i;
2422
2423 scanned = 1;
2424 for (i = 0; i < nr_pages; i++) {
2425 struct page *page = pvec.pages[i];
2426
2427 /*
2428 * At this point we hold neither mapping->tree_lock nor
2429 * lock on the page itself: the page may be truncated or
2430 * invalidated (changing page->mapping to NULL), or even
2431 * swizzled back from swapper_space to tmpfs file
2432 * mapping
2433 */
2434 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2435 tree->ops->write_cache_pages_lock_hook(page);
2436 else
2437 lock_page(page);
2438
2439 if (unlikely(page->mapping != mapping)) {
2440 unlock_page(page);
2441 continue;
2442 }
2443
2444 if (!wbc->range_cyclic && page->index > end) {
2445 done = 1;
2446 unlock_page(page);
2447 continue;
2448 }
2449
2450 if (wbc->sync_mode != WB_SYNC_NONE) {
2451 if (PageWriteback(page))
2452 flush_fn(data);
2453 wait_on_page_writeback(page);
2454 }
2455
2456 if (PageWriteback(page) ||
2457 !clear_page_dirty_for_io(page)) {
2458 unlock_page(page);
2459 continue;
2460 }
2461
2462 ret = (*writepage)(page, wbc, data);
2463
2464 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2465 unlock_page(page);
2466 ret = 0;
2467 }
2468 if (ret || wbc->nr_to_write <= 0)
2469 done = 1;
2470 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2471 wbc->encountered_congestion = 1;
2472 done = 1;
2473 }
2474 }
2475 pagevec_release(&pvec);
2476 cond_resched();
2477 }
2478 if (!scanned && !done) {
2479 /*
2480 * We hit the last page and there is more work to be done: wrap
2481 * back to the start of the file
2482 */
2483 scanned = 1;
2484 index = 0;
2485 goto retry;
2486 }
2487 return ret;
2488 }
2489
2490 static void flush_epd_write_bio(struct extent_page_data *epd)
2491 {
2492 if (epd->bio) {
2493 if (epd->sync_io)
2494 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2495 else
2496 submit_one_bio(WRITE, epd->bio, 0, 0);
2497 epd->bio = NULL;
2498 }
2499 }
2500
2501 static noinline void flush_write_bio(void *data)
2502 {
2503 struct extent_page_data *epd = data;
2504 flush_epd_write_bio(epd);
2505 }
2506
2507 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2508 get_extent_t *get_extent,
2509 struct writeback_control *wbc)
2510 {
2511 int ret;
2512 struct address_space *mapping = page->mapping;
2513 struct extent_page_data epd = {
2514 .bio = NULL,
2515 .tree = tree,
2516 .get_extent = get_extent,
2517 .extent_locked = 0,
2518 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2519 };
2520 struct writeback_control wbc_writepages = {
2521 .bdi = wbc->bdi,
2522 .sync_mode = wbc->sync_mode,
2523 .older_than_this = NULL,
2524 .nr_to_write = 64,
2525 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2526 .range_end = (loff_t)-1,
2527 };
2528
2529 ret = __extent_writepage(page, wbc, &epd);
2530
2531 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2532 __extent_writepage, &epd, flush_write_bio);
2533 flush_epd_write_bio(&epd);
2534 return ret;
2535 }
2536
2537 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2538 u64 start, u64 end, get_extent_t *get_extent,
2539 int mode)
2540 {
2541 int ret = 0;
2542 struct address_space *mapping = inode->i_mapping;
2543 struct page *page;
2544 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2545 PAGE_CACHE_SHIFT;
2546
2547 struct extent_page_data epd = {
2548 .bio = NULL,
2549 .tree = tree,
2550 .get_extent = get_extent,
2551 .extent_locked = 1,
2552 .sync_io = mode == WB_SYNC_ALL,
2553 };
2554 struct writeback_control wbc_writepages = {
2555 .bdi = inode->i_mapping->backing_dev_info,
2556 .sync_mode = mode,
2557 .older_than_this = NULL,
2558 .nr_to_write = nr_pages * 2,
2559 .range_start = start,
2560 .range_end = end + 1,
2561 };
2562
2563 while (start <= end) {
2564 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2565 if (clear_page_dirty_for_io(page))
2566 ret = __extent_writepage(page, &wbc_writepages, &epd);
2567 else {
2568 if (tree->ops && tree->ops->writepage_end_io_hook)
2569 tree->ops->writepage_end_io_hook(page, start,
2570 start + PAGE_CACHE_SIZE - 1,
2571 NULL, 1);
2572 unlock_page(page);
2573 }
2574 page_cache_release(page);
2575 start += PAGE_CACHE_SIZE;
2576 }
2577
2578 flush_epd_write_bio(&epd);
2579 return ret;
2580 }
2581
2582 int extent_writepages(struct extent_io_tree *tree,
2583 struct address_space *mapping,
2584 get_extent_t *get_extent,
2585 struct writeback_control *wbc)
2586 {
2587 int ret = 0;
2588 struct extent_page_data epd = {
2589 .bio = NULL,
2590 .tree = tree,
2591 .get_extent = get_extent,
2592 .extent_locked = 0,
2593 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2594 };
2595
2596 ret = extent_write_cache_pages(tree, mapping, wbc,
2597 __extent_writepage, &epd,
2598 flush_write_bio);
2599 flush_epd_write_bio(&epd);
2600 return ret;
2601 }
2602
2603 int extent_readpages(struct extent_io_tree *tree,
2604 struct address_space *mapping,
2605 struct list_head *pages, unsigned nr_pages,
2606 get_extent_t get_extent)
2607 {
2608 struct bio *bio = NULL;
2609 unsigned page_idx;
2610 struct pagevec pvec;
2611 unsigned long bio_flags = 0;
2612
2613 pagevec_init(&pvec, 0);
2614 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2615 struct page *page = list_entry(pages->prev, struct page, lru);
2616
2617 prefetchw(&page->flags);
2618 list_del(&page->lru);
2619 /*
2620 * what we want to do here is call add_to_page_cache_lru,
2621 * but that isn't exported, so we reproduce it here
2622 */
2623 if (!add_to_page_cache(page, mapping,
2624 page->index, GFP_KERNEL)) {
2625
2626 /* open coding of lru_cache_add, also not exported */
2627 page_cache_get(page);
2628 if (!pagevec_add(&pvec, page))
2629 __pagevec_lru_add_file(&pvec);
2630 __extent_read_full_page(tree, page, get_extent,
2631 &bio, 0, &bio_flags);
2632 }
2633 page_cache_release(page);
2634 }
2635 if (pagevec_count(&pvec))
2636 __pagevec_lru_add_file(&pvec);
2637 BUG_ON(!list_empty(pages));
2638 if (bio)
2639 submit_one_bio(READ, bio, 0, bio_flags);
2640 return 0;
2641 }
2642
2643 /*
2644 * basic invalidatepage code, this waits on any locked or writeback
2645 * ranges corresponding to the page, and then deletes any extent state
2646 * records from the tree
2647 */
2648 int extent_invalidatepage(struct extent_io_tree *tree,
2649 struct page *page, unsigned long offset)
2650 {
2651 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2652 u64 end = start + PAGE_CACHE_SIZE - 1;
2653 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2654
2655 start += (offset + blocksize - 1) & ~(blocksize - 1);
2656 if (start > end)
2657 return 0;
2658
2659 lock_extent(tree, start, end, GFP_NOFS);
2660 wait_on_extent_writeback(tree, start, end);
2661 clear_extent_bit(tree, start, end,
2662 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2663 1, 1, GFP_NOFS);
2664 return 0;
2665 }
2666
2667 /*
2668 * simple commit_write call, set_range_dirty is used to mark both
2669 * the pages and the extent records as dirty
2670 */
2671 int extent_commit_write(struct extent_io_tree *tree,
2672 struct inode *inode, struct page *page,
2673 unsigned from, unsigned to)
2674 {
2675 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2676
2677 set_page_extent_mapped(page);
2678 set_page_dirty(page);
2679
2680 if (pos > inode->i_size) {
2681 i_size_write(inode, pos);
2682 mark_inode_dirty(inode);
2683 }
2684 return 0;
2685 }
2686
2687 int extent_prepare_write(struct extent_io_tree *tree,
2688 struct inode *inode, struct page *page,
2689 unsigned from, unsigned to, get_extent_t *get_extent)
2690 {
2691 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2692 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2693 u64 block_start;
2694 u64 orig_block_start;
2695 u64 block_end;
2696 u64 cur_end;
2697 struct extent_map *em;
2698 unsigned blocksize = 1 << inode->i_blkbits;
2699 size_t page_offset = 0;
2700 size_t block_off_start;
2701 size_t block_off_end;
2702 int err = 0;
2703 int iocount = 0;
2704 int ret = 0;
2705 int isnew;
2706
2707 set_page_extent_mapped(page);
2708
2709 block_start = (page_start + from) & ~((u64)blocksize - 1);
2710 block_end = (page_start + to - 1) | (blocksize - 1);
2711 orig_block_start = block_start;
2712
2713 lock_extent(tree, page_start, page_end, GFP_NOFS);
2714 while (block_start <= block_end) {
2715 em = get_extent(inode, page, page_offset, block_start,
2716 block_end - block_start + 1, 1);
2717 if (IS_ERR(em) || !em)
2718 goto err;
2719
2720 cur_end = min(block_end, extent_map_end(em) - 1);
2721 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2722 block_off_end = block_off_start + blocksize;
2723 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2724
2725 if (!PageUptodate(page) && isnew &&
2726 (block_off_end > to || block_off_start < from)) {
2727 void *kaddr;
2728
2729 kaddr = kmap_atomic(page, KM_USER0);
2730 if (block_off_end > to)
2731 memset(kaddr + to, 0, block_off_end - to);
2732 if (block_off_start < from)
2733 memset(kaddr + block_off_start, 0,
2734 from - block_off_start);
2735 flush_dcache_page(page);
2736 kunmap_atomic(kaddr, KM_USER0);
2737 }
2738 if ((em->block_start != EXTENT_MAP_HOLE &&
2739 em->block_start != EXTENT_MAP_INLINE) &&
2740 !isnew && !PageUptodate(page) &&
2741 (block_off_end > to || block_off_start < from) &&
2742 !test_range_bit(tree, block_start, cur_end,
2743 EXTENT_UPTODATE, 1)) {
2744 u64 sector;
2745 u64 extent_offset = block_start - em->start;
2746 size_t iosize;
2747 sector = (em->block_start + extent_offset) >> 9;
2748 iosize = (cur_end - block_start + blocksize) &
2749 ~((u64)blocksize - 1);
2750 /*
2751 * we've already got the extent locked, but we
2752 * need to split the state such that our end_bio
2753 * handler can clear the lock.
2754 */
2755 set_extent_bit(tree, block_start,
2756 block_start + iosize - 1,
2757 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2758 ret = submit_extent_page(READ, tree, page,
2759 sector, iosize, page_offset, em->bdev,
2760 NULL, 1,
2761 end_bio_extent_preparewrite, 0,
2762 0, 0);
2763 iocount++;
2764 block_start = block_start + iosize;
2765 } else {
2766 set_extent_uptodate(tree, block_start, cur_end,
2767 GFP_NOFS);
2768 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2769 block_start = cur_end + 1;
2770 }
2771 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2772 free_extent_map(em);
2773 }
2774 if (iocount) {
2775 wait_extent_bit(tree, orig_block_start,
2776 block_end, EXTENT_LOCKED);
2777 }
2778 check_page_uptodate(tree, page);
2779 err:
2780 /* FIXME, zero out newly allocated blocks on error */
2781 return err;
2782 }
2783
2784 /*
2785 * a helper for releasepage, this tests for areas of the page that
2786 * are locked or under IO and drops the related state bits if it is safe
2787 * to drop the page.
2788 */
2789 int try_release_extent_state(struct extent_map_tree *map,
2790 struct extent_io_tree *tree, struct page *page,
2791 gfp_t mask)
2792 {
2793 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2794 u64 end = start + PAGE_CACHE_SIZE - 1;
2795 int ret = 1;
2796
2797 if (test_range_bit(tree, start, end,
2798 EXTENT_IOBITS | EXTENT_ORDERED, 0))
2799 ret = 0;
2800 else {
2801 if ((mask & GFP_NOFS) == GFP_NOFS)
2802 mask = GFP_NOFS;
2803 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2804 1, 1, mask);
2805 }
2806 return ret;
2807 }
2808
2809 /*
2810 * a helper for releasepage. As long as there are no locked extents
2811 * in the range corresponding to the page, both state records and extent
2812 * map records are removed
2813 */
2814 int try_release_extent_mapping(struct extent_map_tree *map,
2815 struct extent_io_tree *tree, struct page *page,
2816 gfp_t mask)
2817 {
2818 struct extent_map *em;
2819 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2820 u64 end = start + PAGE_CACHE_SIZE - 1;
2821
2822 if ((mask & __GFP_WAIT) &&
2823 page->mapping->host->i_size > 16 * 1024 * 1024) {
2824 u64 len;
2825 while (start <= end) {
2826 len = end - start + 1;
2827 spin_lock(&map->lock);
2828 em = lookup_extent_mapping(map, start, len);
2829 if (!em || IS_ERR(em)) {
2830 spin_unlock(&map->lock);
2831 break;
2832 }
2833 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2834 em->start != start) {
2835 spin_unlock(&map->lock);
2836 free_extent_map(em);
2837 break;
2838 }
2839 if (!test_range_bit(tree, em->start,
2840 extent_map_end(em) - 1,
2841 EXTENT_LOCKED | EXTENT_WRITEBACK |
2842 EXTENT_ORDERED,
2843 0)) {
2844 remove_extent_mapping(map, em);
2845 /* once for the rb tree */
2846 free_extent_map(em);
2847 }
2848 start = extent_map_end(em);
2849 spin_unlock(&map->lock);
2850
2851 /* once for us */
2852 free_extent_map(em);
2853 }
2854 }
2855 return try_release_extent_state(map, tree, page, mask);
2856 }
2857
2858 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2859 get_extent_t *get_extent)
2860 {
2861 struct inode *inode = mapping->host;
2862 u64 start = iblock << inode->i_blkbits;
2863 sector_t sector = 0;
2864 size_t blksize = (1 << inode->i_blkbits);
2865 struct extent_map *em;
2866
2867 lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2868 GFP_NOFS);
2869 em = get_extent(inode, NULL, 0, start, blksize, 0);
2870 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2871 GFP_NOFS);
2872 if (!em || IS_ERR(em))
2873 return 0;
2874
2875 if (em->block_start > EXTENT_MAP_LAST_BYTE)
2876 goto out;
2877
2878 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2879 out:
2880 free_extent_map(em);
2881 return sector;
2882 }
2883
2884 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2885 __u64 start, __u64 len, get_extent_t *get_extent)
2886 {
2887 int ret;
2888 u64 off = start;
2889 u64 max = start + len;
2890 u32 flags = 0;
2891 u64 disko = 0;
2892 struct extent_map *em = NULL;
2893 int end = 0;
2894 u64 em_start = 0, em_len = 0;
2895 unsigned long emflags;
2896 ret = 0;
2897
2898 if (len == 0)
2899 return -EINVAL;
2900
2901 lock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2902 GFP_NOFS);
2903 em = get_extent(inode, NULL, 0, off, max - off, 0);
2904 if (!em)
2905 goto out;
2906 if (IS_ERR(em)) {
2907 ret = PTR_ERR(em);
2908 goto out;
2909 }
2910 while (!end) {
2911 off = em->start + em->len;
2912 if (off >= max)
2913 end = 1;
2914
2915 em_start = em->start;
2916 em_len = em->len;
2917
2918 disko = 0;
2919 flags = 0;
2920
2921 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
2922 end = 1;
2923 flags |= FIEMAP_EXTENT_LAST;
2924 } else if (em->block_start == EXTENT_MAP_HOLE) {
2925 flags |= FIEMAP_EXTENT_UNWRITTEN;
2926 } else if (em->block_start == EXTENT_MAP_INLINE) {
2927 flags |= (FIEMAP_EXTENT_DATA_INLINE |
2928 FIEMAP_EXTENT_NOT_ALIGNED);
2929 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
2930 flags |= (FIEMAP_EXTENT_DELALLOC |
2931 FIEMAP_EXTENT_UNKNOWN);
2932 } else {
2933 disko = em->block_start;
2934 }
2935 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2936 flags |= FIEMAP_EXTENT_ENCODED;
2937
2938 emflags = em->flags;
2939 free_extent_map(em);
2940 em = NULL;
2941
2942 if (!end) {
2943 em = get_extent(inode, NULL, 0, off, max - off, 0);
2944 if (!em)
2945 goto out;
2946 if (IS_ERR(em)) {
2947 ret = PTR_ERR(em);
2948 goto out;
2949 }
2950 emflags = em->flags;
2951 }
2952 if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) {
2953 flags |= FIEMAP_EXTENT_LAST;
2954 end = 1;
2955 }
2956
2957 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
2958 em_len, flags);
2959 if (ret)
2960 goto out_free;
2961 }
2962 out_free:
2963 free_extent_map(em);
2964 out:
2965 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2966 GFP_NOFS);
2967 return ret;
2968 }
2969
2970 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2971 unsigned long i)
2972 {
2973 struct page *p;
2974 struct address_space *mapping;
2975
2976 if (i == 0)
2977 return eb->first_page;
2978 i += eb->start >> PAGE_CACHE_SHIFT;
2979 mapping = eb->first_page->mapping;
2980 if (!mapping)
2981 return NULL;
2982
2983 /*
2984 * extent_buffer_page is only called after pinning the page
2985 * by increasing the reference count. So we know the page must
2986 * be in the radix tree.
2987 */
2988 rcu_read_lock();
2989 p = radix_tree_lookup(&mapping->page_tree, i);
2990 rcu_read_unlock();
2991
2992 return p;
2993 }
2994
2995 static inline unsigned long num_extent_pages(u64 start, u64 len)
2996 {
2997 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2998 (start >> PAGE_CACHE_SHIFT);
2999 }
3000
3001 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3002 u64 start,
3003 unsigned long len,
3004 gfp_t mask)
3005 {
3006 struct extent_buffer *eb = NULL;
3007 #if LEAK_DEBUG
3008 unsigned long flags;
3009 #endif
3010
3011 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3012 eb->start = start;
3013 eb->len = len;
3014 spin_lock_init(&eb->lock);
3015 init_waitqueue_head(&eb->lock_wq);
3016
3017 #if LEAK_DEBUG
3018 spin_lock_irqsave(&leak_lock, flags);
3019 list_add(&eb->leak_list, &buffers);
3020 spin_unlock_irqrestore(&leak_lock, flags);
3021 #endif
3022 atomic_set(&eb->refs, 1);
3023
3024 return eb;
3025 }
3026
3027 static void __free_extent_buffer(struct extent_buffer *eb)
3028 {
3029 #if LEAK_DEBUG
3030 unsigned long flags;
3031 spin_lock_irqsave(&leak_lock, flags);
3032 list_del(&eb->leak_list);
3033 spin_unlock_irqrestore(&leak_lock, flags);
3034 #endif
3035 kmem_cache_free(extent_buffer_cache, eb);
3036 }
3037
3038 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3039 u64 start, unsigned long len,
3040 struct page *page0,
3041 gfp_t mask)
3042 {
3043 unsigned long num_pages = num_extent_pages(start, len);
3044 unsigned long i;
3045 unsigned long index = start >> PAGE_CACHE_SHIFT;
3046 struct extent_buffer *eb;
3047 struct extent_buffer *exists = NULL;
3048 struct page *p;
3049 struct address_space *mapping = tree->mapping;
3050 int uptodate = 1;
3051
3052 spin_lock(&tree->buffer_lock);
3053 eb = buffer_search(tree, start);
3054 if (eb) {
3055 atomic_inc(&eb->refs);
3056 spin_unlock(&tree->buffer_lock);
3057 mark_page_accessed(eb->first_page);
3058 return eb;
3059 }
3060 spin_unlock(&tree->buffer_lock);
3061
3062 eb = __alloc_extent_buffer(tree, start, len, mask);
3063 if (!eb)
3064 return NULL;
3065
3066 if (page0) {
3067 eb->first_page = page0;
3068 i = 1;
3069 index++;
3070 page_cache_get(page0);
3071 mark_page_accessed(page0);
3072 set_page_extent_mapped(page0);
3073 set_page_extent_head(page0, len);
3074 uptodate = PageUptodate(page0);
3075 } else {
3076 i = 0;
3077 }
3078 for (; i < num_pages; i++, index++) {
3079 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
3080 if (!p) {
3081 WARN_ON(1);
3082 goto free_eb;
3083 }
3084 set_page_extent_mapped(p);
3085 mark_page_accessed(p);
3086 if (i == 0) {
3087 eb->first_page = p;
3088 set_page_extent_head(p, len);
3089 } else {
3090 set_page_private(p, EXTENT_PAGE_PRIVATE);
3091 }
3092 if (!PageUptodate(p))
3093 uptodate = 0;
3094 unlock_page(p);
3095 }
3096 if (uptodate)
3097 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3098
3099 spin_lock(&tree->buffer_lock);
3100 exists = buffer_tree_insert(tree, start, &eb->rb_node);
3101 if (exists) {
3102 /* add one reference for the caller */
3103 atomic_inc(&exists->refs);
3104 spin_unlock(&tree->buffer_lock);
3105 goto free_eb;
3106 }
3107 spin_unlock(&tree->buffer_lock);
3108
3109 /* add one reference for the tree */
3110 atomic_inc(&eb->refs);
3111 return eb;
3112
3113 free_eb:
3114 if (!atomic_dec_and_test(&eb->refs))
3115 return exists;
3116 for (index = 1; index < i; index++)
3117 page_cache_release(extent_buffer_page(eb, index));
3118 page_cache_release(extent_buffer_page(eb, 0));
3119 __free_extent_buffer(eb);
3120 return exists;
3121 }
3122
3123 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3124 u64 start, unsigned long len,
3125 gfp_t mask)
3126 {
3127 struct extent_buffer *eb;
3128
3129 spin_lock(&tree->buffer_lock);
3130 eb = buffer_search(tree, start);
3131 if (eb)
3132 atomic_inc(&eb->refs);
3133 spin_unlock(&tree->buffer_lock);
3134
3135 if (eb)
3136 mark_page_accessed(eb->first_page);
3137
3138 return eb;
3139 }
3140
3141 void free_extent_buffer(struct extent_buffer *eb)
3142 {
3143 if (!eb)
3144 return;
3145
3146 if (!atomic_dec_and_test(&eb->refs))
3147 return;
3148
3149 WARN_ON(1);
3150 }
3151
3152 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3153 struct extent_buffer *eb)
3154 {
3155 unsigned long i;
3156 unsigned long num_pages;
3157 struct page *page;
3158
3159 num_pages = num_extent_pages(eb->start, eb->len);
3160
3161 for (i = 0; i < num_pages; i++) {
3162 page = extent_buffer_page(eb, i);
3163 if (!PageDirty(page))
3164 continue;
3165
3166 lock_page(page);
3167 if (i == 0)
3168 set_page_extent_head(page, eb->len);
3169 else
3170 set_page_private(page, EXTENT_PAGE_PRIVATE);
3171
3172 clear_page_dirty_for_io(page);
3173 spin_lock_irq(&page->mapping->tree_lock);
3174 if (!PageDirty(page)) {
3175 radix_tree_tag_clear(&page->mapping->page_tree,
3176 page_index(page),
3177 PAGECACHE_TAG_DIRTY);
3178 }
3179 spin_unlock_irq(&page->mapping->tree_lock);
3180 unlock_page(page);
3181 }
3182 return 0;
3183 }
3184
3185 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3186 struct extent_buffer *eb)
3187 {
3188 return wait_on_extent_writeback(tree, eb->start,
3189 eb->start + eb->len - 1);
3190 }
3191
3192 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3193 struct extent_buffer *eb)
3194 {
3195 unsigned long i;
3196 unsigned long num_pages;
3197 int was_dirty = 0;
3198
3199 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3200 num_pages = num_extent_pages(eb->start, eb->len);
3201 for (i = 0; i < num_pages; i++)
3202 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3203 return was_dirty;
3204 }
3205
3206 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3207 struct extent_buffer *eb)
3208 {
3209 unsigned long i;
3210 struct page *page;
3211 unsigned long num_pages;
3212
3213 num_pages = num_extent_pages(eb->start, eb->len);
3214 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3215
3216 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3217 GFP_NOFS);
3218 for (i = 0; i < num_pages; i++) {
3219 page = extent_buffer_page(eb, i);
3220 if (page)
3221 ClearPageUptodate(page);
3222 }
3223 return 0;
3224 }
3225
3226 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3227 struct extent_buffer *eb)
3228 {
3229 unsigned long i;
3230 struct page *page;
3231 unsigned long num_pages;
3232
3233 num_pages = num_extent_pages(eb->start, eb->len);
3234
3235 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3236 GFP_NOFS);
3237 for (i = 0; i < num_pages; i++) {
3238 page = extent_buffer_page(eb, i);
3239 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3240 ((i == num_pages - 1) &&
3241 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3242 check_page_uptodate(tree, page);
3243 continue;
3244 }
3245 SetPageUptodate(page);
3246 }
3247 return 0;
3248 }
3249
3250 int extent_range_uptodate(struct extent_io_tree *tree,
3251 u64 start, u64 end)
3252 {
3253 struct page *page;
3254 int ret;
3255 int pg_uptodate = 1;
3256 int uptodate;
3257 unsigned long index;
3258
3259 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3260 if (ret)
3261 return 1;
3262 while (start <= end) {
3263 index = start >> PAGE_CACHE_SHIFT;
3264 page = find_get_page(tree->mapping, index);
3265 uptodate = PageUptodate(page);
3266 page_cache_release(page);
3267 if (!uptodate) {
3268 pg_uptodate = 0;
3269 break;
3270 }
3271 start += PAGE_CACHE_SIZE;
3272 }
3273 return pg_uptodate;
3274 }
3275
3276 int extent_buffer_uptodate(struct extent_io_tree *tree,
3277 struct extent_buffer *eb)
3278 {
3279 int ret = 0;
3280 unsigned long num_pages;
3281 unsigned long i;
3282 struct page *page;
3283 int pg_uptodate = 1;
3284
3285 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3286 return 1;
3287
3288 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3289 EXTENT_UPTODATE, 1);
3290 if (ret)
3291 return ret;
3292
3293 num_pages = num_extent_pages(eb->start, eb->len);
3294 for (i = 0; i < num_pages; i++) {
3295 page = extent_buffer_page(eb, i);
3296 if (!PageUptodate(page)) {
3297 pg_uptodate = 0;
3298 break;
3299 }
3300 }
3301 return pg_uptodate;
3302 }
3303
3304 int read_extent_buffer_pages(struct extent_io_tree *tree,
3305 struct extent_buffer *eb,
3306 u64 start, int wait,
3307 get_extent_t *get_extent, int mirror_num)
3308 {
3309 unsigned long i;
3310 unsigned long start_i;
3311 struct page *page;
3312 int err;
3313 int ret = 0;
3314 int locked_pages = 0;
3315 int all_uptodate = 1;
3316 int inc_all_pages = 0;
3317 unsigned long num_pages;
3318 struct bio *bio = NULL;
3319 unsigned long bio_flags = 0;
3320
3321 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3322 return 0;
3323
3324 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3325 EXTENT_UPTODATE, 1)) {
3326 return 0;
3327 }
3328
3329 if (start) {
3330 WARN_ON(start < eb->start);
3331 start_i = (start >> PAGE_CACHE_SHIFT) -
3332 (eb->start >> PAGE_CACHE_SHIFT);
3333 } else {
3334 start_i = 0;
3335 }
3336
3337 num_pages = num_extent_pages(eb->start, eb->len);
3338 for (i = start_i; i < num_pages; i++) {
3339 page = extent_buffer_page(eb, i);
3340 if (!wait) {
3341 if (!trylock_page(page))
3342 goto unlock_exit;
3343 } else {
3344 lock_page(page);
3345 }
3346 locked_pages++;
3347 if (!PageUptodate(page))
3348 all_uptodate = 0;
3349 }
3350 if (all_uptodate) {
3351 if (start_i == 0)
3352 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3353 goto unlock_exit;
3354 }
3355
3356 for (i = start_i; i < num_pages; i++) {
3357 page = extent_buffer_page(eb, i);
3358 if (inc_all_pages)
3359 page_cache_get(page);
3360 if (!PageUptodate(page)) {
3361 if (start_i == 0)
3362 inc_all_pages = 1;
3363 ClearPageError(page);
3364 err = __extent_read_full_page(tree, page,
3365 get_extent, &bio,
3366 mirror_num, &bio_flags);
3367 if (err)
3368 ret = err;
3369 } else {
3370 unlock_page(page);
3371 }
3372 }
3373
3374 if (bio)
3375 submit_one_bio(READ, bio, mirror_num, bio_flags);
3376
3377 if (ret || !wait)
3378 return ret;
3379
3380 for (i = start_i; i < num_pages; i++) {
3381 page = extent_buffer_page(eb, i);
3382 wait_on_page_locked(page);
3383 if (!PageUptodate(page))
3384 ret = -EIO;
3385 }
3386
3387 if (!ret)
3388 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3389 return ret;
3390
3391 unlock_exit:
3392 i = start_i;
3393 while (locked_pages > 0) {
3394 page = extent_buffer_page(eb, i);
3395 i++;
3396 unlock_page(page);
3397 locked_pages--;
3398 }
3399 return ret;
3400 }
3401
3402 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3403 unsigned long start,
3404 unsigned long len)
3405 {
3406 size_t cur;
3407 size_t offset;
3408 struct page *page;
3409 char *kaddr;
3410 char *dst = (char *)dstv;
3411 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3412 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3413
3414 WARN_ON(start > eb->len);
3415 WARN_ON(start + len > eb->start + eb->len);
3416
3417 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3418
3419 while (len > 0) {
3420 page = extent_buffer_page(eb, i);
3421
3422 cur = min(len, (PAGE_CACHE_SIZE - offset));
3423 kaddr = kmap_atomic(page, KM_USER1);
3424 memcpy(dst, kaddr + offset, cur);
3425 kunmap_atomic(kaddr, KM_USER1);
3426
3427 dst += cur;
3428 len -= cur;
3429 offset = 0;
3430 i++;
3431 }
3432 }
3433
3434 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3435 unsigned long min_len, char **token, char **map,
3436 unsigned long *map_start,
3437 unsigned long *map_len, int km)
3438 {
3439 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3440 char *kaddr;
3441 struct page *p;
3442 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3443 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3444 unsigned long end_i = (start_offset + start + min_len - 1) >>
3445 PAGE_CACHE_SHIFT;
3446
3447 if (i != end_i)
3448 return -EINVAL;
3449
3450 if (i == 0) {
3451 offset = start_offset;
3452 *map_start = 0;
3453 } else {
3454 offset = 0;
3455 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3456 }
3457
3458 if (start + min_len > eb->len) {
3459 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3460 "wanted %lu %lu\n", (unsigned long long)eb->start,
3461 eb->len, start, min_len);
3462 WARN_ON(1);
3463 }
3464
3465 p = extent_buffer_page(eb, i);
3466 kaddr = kmap_atomic(p, km);
3467 *token = kaddr;
3468 *map = kaddr + offset;
3469 *map_len = PAGE_CACHE_SIZE - offset;
3470 return 0;
3471 }
3472
3473 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3474 unsigned long min_len,
3475 char **token, char **map,
3476 unsigned long *map_start,
3477 unsigned long *map_len, int km)
3478 {
3479 int err;
3480 int save = 0;
3481 if (eb->map_token) {
3482 unmap_extent_buffer(eb, eb->map_token, km);
3483 eb->map_token = NULL;
3484 save = 1;
3485 }
3486 err = map_private_extent_buffer(eb, start, min_len, token, map,
3487 map_start, map_len, km);
3488 if (!err && save) {
3489 eb->map_token = *token;
3490 eb->kaddr = *map;
3491 eb->map_start = *map_start;
3492 eb->map_len = *map_len;
3493 }
3494 return err;
3495 }
3496
3497 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3498 {
3499 kunmap_atomic(token, km);
3500 }
3501
3502 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3503 unsigned long start,
3504 unsigned long len)
3505 {
3506 size_t cur;
3507 size_t offset;
3508 struct page *page;
3509 char *kaddr;
3510 char *ptr = (char *)ptrv;
3511 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3512 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3513 int ret = 0;
3514
3515 WARN_ON(start > eb->len);
3516 WARN_ON(start + len > eb->start + eb->len);
3517
3518 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3519
3520 while (len > 0) {
3521 page = extent_buffer_page(eb, i);
3522
3523 cur = min(len, (PAGE_CACHE_SIZE - offset));
3524
3525 kaddr = kmap_atomic(page, KM_USER0);
3526 ret = memcmp(ptr, kaddr + offset, cur);
3527 kunmap_atomic(kaddr, KM_USER0);
3528 if (ret)
3529 break;
3530
3531 ptr += cur;
3532 len -= cur;
3533 offset = 0;
3534 i++;
3535 }
3536 return ret;
3537 }
3538
3539 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3540 unsigned long start, unsigned long len)
3541 {
3542 size_t cur;
3543 size_t offset;
3544 struct page *page;
3545 char *kaddr;
3546 char *src = (char *)srcv;
3547 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3548 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3549
3550 WARN_ON(start > eb->len);
3551 WARN_ON(start + len > eb->start + eb->len);
3552
3553 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3554
3555 while (len > 0) {
3556 page = extent_buffer_page(eb, i);
3557 WARN_ON(!PageUptodate(page));
3558
3559 cur = min(len, PAGE_CACHE_SIZE - offset);
3560 kaddr = kmap_atomic(page, KM_USER1);
3561 memcpy(kaddr + offset, src, cur);
3562 kunmap_atomic(kaddr, KM_USER1);
3563
3564 src += cur;
3565 len -= cur;
3566 offset = 0;
3567 i++;
3568 }
3569 }
3570
3571 void memset_extent_buffer(struct extent_buffer *eb, char c,
3572 unsigned long start, unsigned long len)
3573 {
3574 size_t cur;
3575 size_t offset;
3576 struct page *page;
3577 char *kaddr;
3578 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3579 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3580
3581 WARN_ON(start > eb->len);
3582 WARN_ON(start + len > eb->start + eb->len);
3583
3584 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3585
3586 while (len > 0) {
3587 page = extent_buffer_page(eb, i);
3588 WARN_ON(!PageUptodate(page));
3589
3590 cur = min(len, PAGE_CACHE_SIZE - offset);
3591 kaddr = kmap_atomic(page, KM_USER0);
3592 memset(kaddr + offset, c, cur);
3593 kunmap_atomic(kaddr, KM_USER0);
3594
3595 len -= cur;
3596 offset = 0;
3597 i++;
3598 }
3599 }
3600
3601 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3602 unsigned long dst_offset, unsigned long src_offset,
3603 unsigned long len)
3604 {
3605 u64 dst_len = dst->len;
3606 size_t cur;
3607 size_t offset;
3608 struct page *page;
3609 char *kaddr;
3610 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3611 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3612
3613 WARN_ON(src->len != dst_len);
3614
3615 offset = (start_offset + dst_offset) &
3616 ((unsigned long)PAGE_CACHE_SIZE - 1);
3617
3618 while (len > 0) {
3619 page = extent_buffer_page(dst, i);
3620 WARN_ON(!PageUptodate(page));
3621
3622 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3623
3624 kaddr = kmap_atomic(page, KM_USER0);
3625 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3626 kunmap_atomic(kaddr, KM_USER0);
3627
3628 src_offset += cur;
3629 len -= cur;
3630 offset = 0;
3631 i++;
3632 }
3633 }
3634
3635 static void move_pages(struct page *dst_page, struct page *src_page,
3636 unsigned long dst_off, unsigned long src_off,
3637 unsigned long len)
3638 {
3639 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3640 if (dst_page == src_page) {
3641 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3642 } else {
3643 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3644 char *p = dst_kaddr + dst_off + len;
3645 char *s = src_kaddr + src_off + len;
3646
3647 while (len--)
3648 *--p = *--s;
3649
3650 kunmap_atomic(src_kaddr, KM_USER1);
3651 }
3652 kunmap_atomic(dst_kaddr, KM_USER0);
3653 }
3654
3655 static void copy_pages(struct page *dst_page, struct page *src_page,
3656 unsigned long dst_off, unsigned long src_off,
3657 unsigned long len)
3658 {
3659 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3660 char *src_kaddr;
3661
3662 if (dst_page != src_page)
3663 src_kaddr = kmap_atomic(src_page, KM_USER1);
3664 else
3665 src_kaddr = dst_kaddr;
3666
3667 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3668 kunmap_atomic(dst_kaddr, KM_USER0);
3669 if (dst_page != src_page)
3670 kunmap_atomic(src_kaddr, KM_USER1);
3671 }
3672
3673 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3674 unsigned long src_offset, unsigned long len)
3675 {
3676 size_t cur;
3677 size_t dst_off_in_page;
3678 size_t src_off_in_page;
3679 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3680 unsigned long dst_i;
3681 unsigned long src_i;
3682
3683 if (src_offset + len > dst->len) {
3684 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3685 "len %lu dst len %lu\n", src_offset, len, dst->len);
3686 BUG_ON(1);
3687 }
3688 if (dst_offset + len > dst->len) {
3689 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3690 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3691 BUG_ON(1);
3692 }
3693
3694 while (len > 0) {
3695 dst_off_in_page = (start_offset + dst_offset) &
3696 ((unsigned long)PAGE_CACHE_SIZE - 1);
3697 src_off_in_page = (start_offset + src_offset) &
3698 ((unsigned long)PAGE_CACHE_SIZE - 1);
3699
3700 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3701 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3702
3703 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3704 src_off_in_page));
3705 cur = min_t(unsigned long, cur,
3706 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3707
3708 copy_pages(extent_buffer_page(dst, dst_i),
3709 extent_buffer_page(dst, src_i),
3710 dst_off_in_page, src_off_in_page, cur);
3711
3712 src_offset += cur;
3713 dst_offset += cur;
3714 len -= cur;
3715 }
3716 }
3717
3718 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3719 unsigned long src_offset, unsigned long len)
3720 {
3721 size_t cur;
3722 size_t dst_off_in_page;
3723 size_t src_off_in_page;
3724 unsigned long dst_end = dst_offset + len - 1;
3725 unsigned long src_end = src_offset + len - 1;
3726 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3727 unsigned long dst_i;
3728 unsigned long src_i;
3729
3730 if (src_offset + len > dst->len) {
3731 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3732 "len %lu len %lu\n", src_offset, len, dst->len);
3733 BUG_ON(1);
3734 }
3735 if (dst_offset + len > dst->len) {
3736 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3737 "len %lu len %lu\n", dst_offset, len, dst->len);
3738 BUG_ON(1);
3739 }
3740 if (dst_offset < src_offset) {
3741 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3742 return;
3743 }
3744 while (len > 0) {
3745 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3746 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3747
3748 dst_off_in_page = (start_offset + dst_end) &
3749 ((unsigned long)PAGE_CACHE_SIZE - 1);
3750 src_off_in_page = (start_offset + src_end) &
3751 ((unsigned long)PAGE_CACHE_SIZE - 1);
3752
3753 cur = min_t(unsigned long, len, src_off_in_page + 1);
3754 cur = min(cur, dst_off_in_page + 1);
3755 move_pages(extent_buffer_page(dst, dst_i),
3756 extent_buffer_page(dst, src_i),
3757 dst_off_in_page - cur + 1,
3758 src_off_in_page - cur + 1, cur);
3759
3760 dst_end -= cur;
3761 src_end -= cur;
3762 len -= cur;
3763 }
3764 }
3765
3766 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3767 {
3768 u64 start = page_offset(page);
3769 struct extent_buffer *eb;
3770 int ret = 1;
3771 unsigned long i;
3772 unsigned long num_pages;
3773
3774 spin_lock(&tree->buffer_lock);
3775 eb = buffer_search(tree, start);
3776 if (!eb)
3777 goto out;
3778
3779 if (atomic_read(&eb->refs) > 1) {
3780 ret = 0;
3781 goto out;
3782 }
3783 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3784 ret = 0;
3785 goto out;
3786 }
3787 /* at this point we can safely release the extent buffer */
3788 num_pages = num_extent_pages(eb->start, eb->len);
3789 for (i = 0; i < num_pages; i++)
3790 page_cache_release(extent_buffer_page(eb, i));
3791 rb_erase(&eb->rb_node, &tree->buffer);
3792 __free_extent_buffer(eb);
3793 out:
3794 spin_unlock(&tree->buffer_lock);
3795 return ret;
3796 }
linux 2.6 git repository for openrd