?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
tracing/x86: basic implementation of syscall tracing for x86
[linux/fpc-iii.git] / fs / btrfs / extent_io.c
blobebe6b29e60698156250708766c70bb74b903bfb6
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 /* temporary define until extent_map moves out of btrfs */
21 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
22 unsigned long extra_flags,
23 void (*ctor)(void *, struct kmem_cache *,
24 unsigned long));
25
26 static struct kmem_cache *extent_state_cache;
27 static struct kmem_cache *extent_buffer_cache;
28
29 static LIST_HEAD(buffers);
30 static LIST_HEAD(states);
31
32 #define LEAK_DEBUG 0
33 #if LEAK_DEBUG
34 static DEFINE_SPINLOCK(leak_lock);
35 #endif
36
37 #define BUFFER_LRU_MAX 64
38
39 struct tree_entry {
40 u64 start;
41 u64 end;
42 struct rb_node rb_node;
43 };
44
45 struct extent_page_data {
46 struct bio *bio;
47 struct extent_io_tree *tree;
48 get_extent_t *get_extent;
49
50 /* tells writepage not to lock the state bits for this range
51 * it still does the unlocking
52 */
53 int extent_locked;
54 };
55
56 int __init extent_io_init(void)
57 {
58 extent_state_cache = btrfs_cache_create("extent_state",
59 sizeof(struct extent_state), 0,
60 NULL);
61 if (!extent_state_cache)
62 return -ENOMEM;
63
64 extent_buffer_cache = btrfs_cache_create("extent_buffers",
65 sizeof(struct extent_buffer), 0,
66 NULL);
67 if (!extent_buffer_cache)
68 goto free_state_cache;
69 return 0;
70
71 free_state_cache:
72 kmem_cache_destroy(extent_state_cache);
73 return -ENOMEM;
74 }
75
76 void extent_io_exit(void)
77 {
78 struct extent_state *state;
79 struct extent_buffer *eb;
80
81 while (!list_empty(&states)) {
82 state = list_entry(states.next, struct extent_state, leak_list);
83 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
84 "state %lu in tree %p refs %d\n",
85 (unsigned long long)state->start,
86 (unsigned long long)state->end,
87 state->state, state->tree, atomic_read(&state->refs));
88 list_del(&state->leak_list);
89 kmem_cache_free(extent_state_cache, state);
90
91 }
92
93 while (!list_empty(&buffers)) {
94 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
95 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
96 "refs %d\n", (unsigned long long)eb->start,
97 eb->len, atomic_read(&eb->refs));
98 list_del(&eb->leak_list);
99 kmem_cache_free(extent_buffer_cache, eb);
100 }
101 if (extent_state_cache)
102 kmem_cache_destroy(extent_state_cache);
103 if (extent_buffer_cache)
104 kmem_cache_destroy(extent_buffer_cache);
105 }
106
107 void extent_io_tree_init(struct extent_io_tree *tree,
108 struct address_space *mapping, gfp_t mask)
109 {
110 tree->state.rb_node = NULL;
111 tree->buffer.rb_node = NULL;
112 tree->ops = NULL;
113 tree->dirty_bytes = 0;
114 spin_lock_init(&tree->lock);
115 spin_lock_init(&tree->buffer_lock);
116 tree->mapping = mapping;
117 }
118
119 static struct extent_state *alloc_extent_state(gfp_t mask)
120 {
121 struct extent_state *state;
122 #if LEAK_DEBUG
123 unsigned long flags;
124 #endif
125
126 state = kmem_cache_alloc(extent_state_cache, mask);
127 if (!state)
128 return state;
129 state->state = 0;
130 state->private = 0;
131 state->tree = NULL;
132 #if LEAK_DEBUG
133 spin_lock_irqsave(&leak_lock, flags);
134 list_add(&state->leak_list, &states);
135 spin_unlock_irqrestore(&leak_lock, flags);
136 #endif
137 atomic_set(&state->refs, 1);
138 init_waitqueue_head(&state->wq);
139 return state;
140 }
141
142 static void free_extent_state(struct extent_state *state)
143 {
144 if (!state)
145 return;
146 if (atomic_dec_and_test(&state->refs)) {
147 #if LEAK_DEBUG
148 unsigned long flags;
149 #endif
150 WARN_ON(state->tree);
151 #if LEAK_DEBUG
152 spin_lock_irqsave(&leak_lock, flags);
153 list_del(&state->leak_list);
154 spin_unlock_irqrestore(&leak_lock, flags);
155 #endif
156 kmem_cache_free(extent_state_cache, state);
157 }
158 }
159
160 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
161 struct rb_node *node)
162 {
163 struct rb_node **p = &root->rb_node;
164 struct rb_node *parent = NULL;
165 struct tree_entry *entry;
166
167 while (*p) {
168 parent = *p;
169 entry = rb_entry(parent, struct tree_entry, rb_node);
170
171 if (offset < entry->start)
172 p = &(*p)->rb_left;
173 else if (offset > entry->end)
174 p = &(*p)->rb_right;
175 else
176 return parent;
177 }
178
179 entry = rb_entry(node, struct tree_entry, rb_node);
180 rb_link_node(node, parent, p);
181 rb_insert_color(node, root);
182 return NULL;
183 }
184
185 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
186 struct rb_node **prev_ret,
187 struct rb_node **next_ret)
188 {
189 struct rb_root *root = &tree->state;
190 struct rb_node *n = root->rb_node;
191 struct rb_node *prev = NULL;
192 struct rb_node *orig_prev = NULL;
193 struct tree_entry *entry;
194 struct tree_entry *prev_entry = NULL;
195
196 while (n) {
197 entry = rb_entry(n, struct tree_entry, rb_node);
198 prev = n;
199 prev_entry = entry;
200
201 if (offset < entry->start)
202 n = n->rb_left;
203 else if (offset > entry->end)
204 n = n->rb_right;
205 else
206 return n;
207 }
208
209 if (prev_ret) {
210 orig_prev = prev;
211 while (prev && offset > prev_entry->end) {
212 prev = rb_next(prev);
213 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
214 }
215 *prev_ret = prev;
216 prev = orig_prev;
217 }
218
219 if (next_ret) {
220 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
221 while (prev && offset < prev_entry->start) {
222 prev = rb_prev(prev);
223 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
224 }
225 *next_ret = prev;
226 }
227 return NULL;
228 }
229
230 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
231 u64 offset)
232 {
233 struct rb_node *prev = NULL;
234 struct rb_node *ret;
235
236 ret = __etree_search(tree, offset, &prev, NULL);
237 if (!ret)
238 return prev;
239 return ret;
240 }
241
242 static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree,
243 u64 offset, struct rb_node *node)
244 {
245 struct rb_root *root = &tree->buffer;
246 struct rb_node **p = &root->rb_node;
247 struct rb_node *parent = NULL;
248 struct extent_buffer *eb;
249
250 while (*p) {
251 parent = *p;
252 eb = rb_entry(parent, struct extent_buffer, rb_node);
253
254 if (offset < eb->start)
255 p = &(*p)->rb_left;
256 else if (offset > eb->start)
257 p = &(*p)->rb_right;
258 else
259 return eb;
260 }
261
262 rb_link_node(node, parent, p);
263 rb_insert_color(node, root);
264 return NULL;
265 }
266
267 static struct extent_buffer *buffer_search(struct extent_io_tree *tree,
268 u64 offset)
269 {
270 struct rb_root *root = &tree->buffer;
271 struct rb_node *n = root->rb_node;
272 struct extent_buffer *eb;
273
274 while (n) {
275 eb = rb_entry(n, struct extent_buffer, rb_node);
276 if (offset < eb->start)
277 n = n->rb_left;
278 else if (offset > eb->start)
279 n = n->rb_right;
280 else
281 return eb;
282 }
283 return NULL;
284 }
285
286 /*
287 * utility function to look for merge candidates inside a given range.
288 * Any extents with matching state are merged together into a single
289 * extent in the tree. Extents with EXTENT_IO in their state field
290 * are not merged because the end_io handlers need to be able to do
291 * operations on them without sleeping (or doing allocations/splits).
292 *
293 * This should be called with the tree lock held.
294 */
295 static int merge_state(struct extent_io_tree *tree,
296 struct extent_state *state)
297 {
298 struct extent_state *other;
299 struct rb_node *other_node;
300
301 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
302 return 0;
303
304 other_node = rb_prev(&state->rb_node);
305 if (other_node) {
306 other = rb_entry(other_node, struct extent_state, rb_node);
307 if (other->end == state->start - 1 &&
308 other->state == state->state) {
309 state->start = other->start;
310 other->tree = NULL;
311 rb_erase(&other->rb_node, &tree->state);
312 free_extent_state(other);
313 }
314 }
315 other_node = rb_next(&state->rb_node);
316 if (other_node) {
317 other = rb_entry(other_node, struct extent_state, rb_node);
318 if (other->start == state->end + 1 &&
319 other->state == state->state) {
320 other->start = state->start;
321 state->tree = NULL;
322 rb_erase(&state->rb_node, &tree->state);
323 free_extent_state(state);
324 }
325 }
326 return 0;
327 }
328
329 static void set_state_cb(struct extent_io_tree *tree,
330 struct extent_state *state,
331 unsigned long bits)
332 {
333 if (tree->ops && tree->ops->set_bit_hook) {
334 tree->ops->set_bit_hook(tree->mapping->host, state->start,
335 state->end, state->state, bits);
336 }
337 }
338
339 static void clear_state_cb(struct extent_io_tree *tree,
340 struct extent_state *state,
341 unsigned long bits)
342 {
343 if (tree->ops && tree->ops->clear_bit_hook) {
344 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
345 state->end, state->state, bits);
346 }
347 }
348
349 /*
350 * insert an extent_state struct into the tree. 'bits' are set on the
351 * struct before it is inserted.
352 *
353 * This may return -EEXIST if the extent is already there, in which case the
354 * state struct is freed.
355 *
356 * The tree lock is not taken internally. This is a utility function and
357 * probably isn't what you want to call (see set/clear_extent_bit).
358 */
359 static int insert_state(struct extent_io_tree *tree,
360 struct extent_state *state, u64 start, u64 end,
361 int bits)
362 {
363 struct rb_node *node;
364
365 if (end < start) {
366 printk(KERN_ERR "btrfs end < start %llu %llu\n",
367 (unsigned long long)end,
368 (unsigned long long)start);
369 WARN_ON(1);
370 }
371 if (bits & EXTENT_DIRTY)
372 tree->dirty_bytes += end - start + 1;
373 set_state_cb(tree, state, bits);
374 state->state |= bits;
375 state->start = start;
376 state->end = end;
377 node = tree_insert(&tree->state, end, &state->rb_node);
378 if (node) {
379 struct extent_state *found;
380 found = rb_entry(node, struct extent_state, rb_node);
381 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
382 "%llu %llu\n", (unsigned long long)found->start,
383 (unsigned long long)found->end,
384 (unsigned long long)start, (unsigned long long)end);
385 free_extent_state(state);
386 return -EEXIST;
387 }
388 state->tree = tree;
389 merge_state(tree, state);
390 return 0;
391 }
392
393 /*
394 * split a given extent state struct in two, inserting the preallocated
395 * struct 'prealloc' as the newly created second half. 'split' indicates an
396 * offset inside 'orig' where it should be split.
397 *
398 * Before calling,
399 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
400 * are two extent state structs in the tree:
401 * prealloc: [orig->start, split - 1]
402 * orig: [ split, orig->end ]
403 *
404 * The tree locks are not taken by this function. They need to be held
405 * by the caller.
406 */
407 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
408 struct extent_state *prealloc, u64 split)
409 {
410 struct rb_node *node;
411 prealloc->start = orig->start;
412 prealloc->end = split - 1;
413 prealloc->state = orig->state;
414 orig->start = split;
415
416 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
417 if (node) {
418 free_extent_state(prealloc);
419 return -EEXIST;
420 }
421 prealloc->tree = tree;
422 return 0;
423 }
424
425 /*
426 * utility function to clear some bits in an extent state struct.
427 * it will optionally wake up any one waiting on this state (wake == 1), or
428 * forcibly remove the state from the tree (delete == 1).
429 *
430 * If no bits are set on the state struct after clearing things, the
431 * struct is freed and removed from the tree
432 */
433 static int clear_state_bit(struct extent_io_tree *tree,
434 struct extent_state *state, int bits, int wake,
435 int delete)
436 {
437 int ret = state->state & bits;
438
439 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
440 u64 range = state->end - state->start + 1;
441 WARN_ON(range > tree->dirty_bytes);
442 tree->dirty_bytes -= range;
443 }
444 clear_state_cb(tree, state, bits);
445 state->state &= ~bits;
446 if (wake)
447 wake_up(&state->wq);
448 if (delete || state->state == 0) {
449 if (state->tree) {
450 clear_state_cb(tree, state, state->state);
451 rb_erase(&state->rb_node, &tree->state);
452 state->tree = NULL;
453 free_extent_state(state);
454 } else {
455 WARN_ON(1);
456 }
457 } else {
458 merge_state(tree, state);
459 }
460 return ret;
461 }
462
463 /*
464 * clear some bits on a range in the tree. This may require splitting
465 * or inserting elements in the tree, so the gfp mask is used to
466 * indicate which allocations or sleeping are allowed.
467 *
468 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
469 * the given range from the tree regardless of state (ie for truncate).
470 *
471 * the range [start, end] is inclusive.
472 *
473 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
474 * bits were already set, or zero if none of the bits were already set.
475 */
476 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
477 int bits, int wake, int delete, gfp_t mask)
478 {
479 struct extent_state *state;
480 struct extent_state *prealloc = NULL;
481 struct rb_node *node;
482 int err;
483 int set = 0;
484
485 again:
486 if (!prealloc && (mask & __GFP_WAIT)) {
487 prealloc = alloc_extent_state(mask);
488 if (!prealloc)
489 return -ENOMEM;
490 }
491
492 spin_lock(&tree->lock);
493 /*
494 * this search will find the extents that end after
495 * our range starts
496 */
497 node = tree_search(tree, start);
498 if (!node)
499 goto out;
500 state = rb_entry(node, struct extent_state, rb_node);
501 if (state->start > end)
502 goto out;
503 WARN_ON(state->end < start);
504
505 /*
506 * | ---- desired range ---- |
507 * | state | or
508 * | ------------- state -------------- |
509 *
510 * We need to split the extent we found, and may flip
511 * bits on second half.
512 *
513 * If the extent we found extends past our range, we
514 * just split and search again. It'll get split again
515 * the next time though.
516 *
517 * If the extent we found is inside our range, we clear
518 * the desired bit on it.
519 */
520
521 if (state->start < start) {
522 if (!prealloc)
523 prealloc = alloc_extent_state(GFP_ATOMIC);
524 err = split_state(tree, state, prealloc, start);
525 BUG_ON(err == -EEXIST);
526 prealloc = NULL;
527 if (err)
528 goto out;
529 if (state->end <= end) {
530 start = state->end + 1;
531 set |= clear_state_bit(tree, state, bits,
532 wake, delete);
533 } else {
534 start = state->start;
535 }
536 goto search_again;
537 }
538 /*
539 * | ---- desired range ---- |
540 * | state |
541 * We need to split the extent, and clear the bit
542 * on the first half
543 */
544 if (state->start <= end && state->end > end) {
545 if (!prealloc)
546 prealloc = alloc_extent_state(GFP_ATOMIC);
547 err = split_state(tree, state, prealloc, end + 1);
548 BUG_ON(err == -EEXIST);
549
550 if (wake)
551 wake_up(&state->wq);
552 set |= clear_state_bit(tree, prealloc, bits,
553 wake, delete);
554 prealloc = NULL;
555 goto out;
556 }
557
558 start = state->end + 1;
559 set |= clear_state_bit(tree, state, bits, wake, delete);
560 goto search_again;
561
562 out:
563 spin_unlock(&tree->lock);
564 if (prealloc)
565 free_extent_state(prealloc);
566
567 return set;
568
569 search_again:
570 if (start > end)
571 goto out;
572 spin_unlock(&tree->lock);
573 if (mask & __GFP_WAIT)
574 cond_resched();
575 goto again;
576 }
577
578 static int wait_on_state(struct extent_io_tree *tree,
579 struct extent_state *state)
580 __releases(tree->lock)
581 __acquires(tree->lock)
582 {
583 DEFINE_WAIT(wait);
584 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
585 spin_unlock(&tree->lock);
586 schedule();
587 spin_lock(&tree->lock);
588 finish_wait(&state->wq, &wait);
589 return 0;
590 }
591
592 /*
593 * waits for one or more bits to clear on a range in the state tree.
594 * The range [start, end] is inclusive.
595 * The tree lock is taken by this function
596 */
597 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
598 {
599 struct extent_state *state;
600 struct rb_node *node;
601
602 spin_lock(&tree->lock);
603 again:
604 while (1) {
605 /*
606 * this search will find all the extents that end after
607 * our range starts
608 */
609 node = tree_search(tree, start);
610 if (!node)
611 break;
612
613 state = rb_entry(node, struct extent_state, rb_node);
614
615 if (state->start > end)
616 goto out;
617
618 if (state->state & bits) {
619 start = state->start;
620 atomic_inc(&state->refs);
621 wait_on_state(tree, state);
622 free_extent_state(state);
623 goto again;
624 }
625 start = state->end + 1;
626
627 if (start > end)
628 break;
629
630 if (need_resched()) {
631 spin_unlock(&tree->lock);
632 cond_resched();
633 spin_lock(&tree->lock);
634 }
635 }
636 out:
637 spin_unlock(&tree->lock);
638 return 0;
639 }
640
641 static void set_state_bits(struct extent_io_tree *tree,
642 struct extent_state *state,
643 int bits)
644 {
645 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
646 u64 range = state->end - state->start + 1;
647 tree->dirty_bytes += range;
648 }
649 set_state_cb(tree, state, bits);
650 state->state |= bits;
651 }
652
653 /*
654 * set some bits on a range in the tree. This may require allocations
655 * or sleeping, so the gfp mask is used to indicate what is allowed.
656 *
657 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
658 * range already has the desired bits set. The start of the existing
659 * range is returned in failed_start in this case.
660 *
661 * [start, end] is inclusive
662 * This takes the tree lock.
663 */
664 static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
665 int bits, int exclusive, u64 *failed_start,
666 gfp_t mask)
667 {
668 struct extent_state *state;
669 struct extent_state *prealloc = NULL;
670 struct rb_node *node;
671 int err = 0;
672 int set;
673 u64 last_start;
674 u64 last_end;
675 again:
676 if (!prealloc && (mask & __GFP_WAIT)) {
677 prealloc = alloc_extent_state(mask);
678 if (!prealloc)
679 return -ENOMEM;
680 }
681
682 spin_lock(&tree->lock);
683 /*
684 * this search will find all the extents that end after
685 * our range starts.
686 */
687 node = tree_search(tree, start);
688 if (!node) {
689 err = insert_state(tree, prealloc, start, end, bits);
690 prealloc = NULL;
691 BUG_ON(err == -EEXIST);
692 goto out;
693 }
694
695 state = rb_entry(node, struct extent_state, rb_node);
696 last_start = state->start;
697 last_end = state->end;
698
699 /*
700 * | ---- desired range ---- |
701 * | state |
702 *
703 * Just lock what we found and keep going
704 */
705 if (state->start == start && state->end <= end) {
706 set = state->state & bits;
707 if (set && exclusive) {
708 *failed_start = state->start;
709 err = -EEXIST;
710 goto out;
711 }
712 set_state_bits(tree, state, bits);
713 start = state->end + 1;
714 merge_state(tree, state);
715 goto search_again;
716 }
717
718 /*
719 * | ---- desired range ---- |
720 * | state |
721 * or
722 * | ------------- state -------------- |
723 *
724 * We need to split the extent we found, and may flip bits on
725 * second half.
726 *
727 * If the extent we found extends past our
728 * range, we just split and search again. It'll get split
729 * again the next time though.
730 *
731 * If the extent we found is inside our range, we set the
732 * desired bit on it.
733 */
734 if (state->start < start) {
735 set = state->state & bits;
736 if (exclusive && set) {
737 *failed_start = start;
738 err = -EEXIST;
739 goto out;
740 }
741 err = split_state(tree, state, prealloc, start);
742 BUG_ON(err == -EEXIST);
743 prealloc = NULL;
744 if (err)
745 goto out;
746 if (state->end <= end) {
747 set_state_bits(tree, state, bits);
748 start = state->end + 1;
749 merge_state(tree, state);
750 } else {
751 start = state->start;
752 }
753 goto search_again;
754 }
755 /*
756 * | ---- desired range ---- |
757 * | state | or | state |
758 *
759 * There's a hole, we need to insert something in it and
760 * ignore the extent we found.
761 */
762 if (state->start > start) {
763 u64 this_end;
764 if (end < last_start)
765 this_end = end;
766 else
767 this_end = last_start - 1;
768 err = insert_state(tree, prealloc, start, this_end,
769 bits);
770 prealloc = NULL;
771 BUG_ON(err == -EEXIST);
772 if (err)
773 goto out;
774 start = this_end + 1;
775 goto search_again;
776 }
777 /*
778 * | ---- desired range ---- |
779 * | state |
780 * We need to split the extent, and set the bit
781 * on the first half
782 */
783 if (state->start <= end && state->end > end) {
784 set = state->state & bits;
785 if (exclusive && set) {
786 *failed_start = start;
787 err = -EEXIST;
788 goto out;
789 }
790 err = split_state(tree, state, prealloc, end + 1);
791 BUG_ON(err == -EEXIST);
792
793 set_state_bits(tree, prealloc, bits);
794 merge_state(tree, prealloc);
795 prealloc = NULL;
796 goto out;
797 }
798
799 goto search_again;
800
801 out:
802 spin_unlock(&tree->lock);
803 if (prealloc)
804 free_extent_state(prealloc);
805
806 return err;
807
808 search_again:
809 if (start > end)
810 goto out;
811 spin_unlock(&tree->lock);
812 if (mask & __GFP_WAIT)
813 cond_resched();
814 goto again;
815 }
816
817 /* wrappers around set/clear extent bit */
818 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
819 gfp_t mask)
820 {
821 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
822 mask);
823 }
824
825 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
826 gfp_t mask)
827 {
828 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
829 }
830
831 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
832 int bits, gfp_t mask)
833 {
834 return set_extent_bit(tree, start, end, bits, 0, NULL,
835 mask);
836 }
837
838 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
839 int bits, gfp_t mask)
840 {
841 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
842 }
843
844 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
845 gfp_t mask)
846 {
847 return set_extent_bit(tree, start, end,
848 EXTENT_DELALLOC | EXTENT_DIRTY,
849 0, NULL, mask);
850 }
851
852 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
853 gfp_t mask)
854 {
855 return clear_extent_bit(tree, start, end,
856 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
857 }
858
859 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
860 gfp_t mask)
861 {
862 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
863 }
864
865 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
866 gfp_t mask)
867 {
868 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
869 mask);
870 }
871
872 static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
873 gfp_t mask)
874 {
875 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
876 }
877
878 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
879 gfp_t mask)
880 {
881 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
882 mask);
883 }
884
885 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
886 u64 end, gfp_t mask)
887 {
888 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
889 }
890
891 static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
892 gfp_t mask)
893 {
894 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
895 0, NULL, mask);
896 }
897
898 static int clear_extent_writeback(struct extent_io_tree *tree, u64 start,
899 u64 end, gfp_t mask)
900 {
901 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
902 }
903
904 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
905 {
906 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
907 }
908
909 /*
910 * either insert or lock state struct between start and end use mask to tell
911 * us if waiting is desired.
912 */
913 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
914 {
915 int err;
916 u64 failed_start;
917 while (1) {
918 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
919 &failed_start, mask);
920 if (err == -EEXIST && (mask & __GFP_WAIT)) {
921 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
922 start = failed_start;
923 } else {
924 break;
925 }
926 WARN_ON(start > end);
927 }
928 return err;
929 }
930
931 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
932 gfp_t mask)
933 {
934 int err;
935 u64 failed_start;
936
937 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
938 &failed_start, mask);
939 if (err == -EEXIST) {
940 if (failed_start > start)
941 clear_extent_bit(tree, start, failed_start - 1,
942 EXTENT_LOCKED, 1, 0, mask);
943 return 0;
944 }
945 return 1;
946 }
947
948 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
949 gfp_t mask)
950 {
951 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
952 }
953
954 /*
955 * helper function to set pages and extents in the tree dirty
956 */
957 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
958 {
959 unsigned long index = start >> PAGE_CACHE_SHIFT;
960 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
961 struct page *page;
962
963 while (index <= end_index) {
964 page = find_get_page(tree->mapping, index);
965 BUG_ON(!page);
966 __set_page_dirty_nobuffers(page);
967 page_cache_release(page);
968 index++;
969 }
970 set_extent_dirty(tree, start, end, GFP_NOFS);
971 return 0;
972 }
973
974 /*
975 * helper function to set both pages and extents in the tree writeback
976 */
977 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
978 {
979 unsigned long index = start >> PAGE_CACHE_SHIFT;
980 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
981 struct page *page;
982
983 while (index <= end_index) {
984 page = find_get_page(tree->mapping, index);
985 BUG_ON(!page);
986 set_page_writeback(page);
987 page_cache_release(page);
988 index++;
989 }
990 set_extent_writeback(tree, start, end, GFP_NOFS);
991 return 0;
992 }
993
994 /*
995 * find the first offset in the io tree with 'bits' set. zero is
996 * returned if we find something, and *start_ret and *end_ret are
997 * set to reflect the state struct that was found.
998 *
999 * If nothing was found, 1 is returned, < 0 on error
1000 */
1001 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1002 u64 *start_ret, u64 *end_ret, int bits)
1003 {
1004 struct rb_node *node;
1005 struct extent_state *state;
1006 int ret = 1;
1007
1008 spin_lock(&tree->lock);
1009 /*
1010 * this search will find all the extents that end after
1011 * our range starts.
1012 */
1013 node = tree_search(tree, start);
1014 if (!node)
1015 goto out;
1016
1017 while (1) {
1018 state = rb_entry(node, struct extent_state, rb_node);
1019 if (state->end >= start && (state->state & bits)) {
1020 *start_ret = state->start;
1021 *end_ret = state->end;
1022 ret = 0;
1023 break;
1024 }
1025 node = rb_next(node);
1026 if (!node)
1027 break;
1028 }
1029 out:
1030 spin_unlock(&tree->lock);
1031 return ret;
1032 }
1033
1034 /* find the first state struct with 'bits' set after 'start', and
1035 * return it. tree->lock must be held. NULL will returned if
1036 * nothing was found after 'start'
1037 */
1038 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1039 u64 start, int bits)
1040 {
1041 struct rb_node *node;
1042 struct extent_state *state;
1043
1044 /*
1045 * this search will find all the extents that end after
1046 * our range starts.
1047 */
1048 node = tree_search(tree, start);
1049 if (!node)
1050 goto out;
1051
1052 while (1) {
1053 state = rb_entry(node, struct extent_state, rb_node);
1054 if (state->end >= start && (state->state & bits))
1055 return state;
1056
1057 node = rb_next(node);
1058 if (!node)
1059 break;
1060 }
1061 out:
1062 return NULL;
1063 }
1064
1065 /*
1066 * find a contiguous range of bytes in the file marked as delalloc, not
1067 * more than 'max_bytes'. start and end are used to return the range,
1068 *
1069 * 1 is returned if we find something, 0 if nothing was in the tree
1070 */
1071 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1072 u64 *start, u64 *end, u64 max_bytes)
1073 {
1074 struct rb_node *node;
1075 struct extent_state *state;
1076 u64 cur_start = *start;
1077 u64 found = 0;
1078 u64 total_bytes = 0;
1079
1080 spin_lock(&tree->lock);
1081
1082 /*
1083 * this search will find all the extents that end after
1084 * our range starts.
1085 */
1086 node = tree_search(tree, cur_start);
1087 if (!node) {
1088 if (!found)
1089 *end = (u64)-1;
1090 goto out;
1091 }
1092
1093 while (1) {
1094 state = rb_entry(node, struct extent_state, rb_node);
1095 if (found && (state->start != cur_start ||
1096 (state->state & EXTENT_BOUNDARY))) {
1097 goto out;
1098 }
1099 if (!(state->state & EXTENT_DELALLOC)) {
1100 if (!found)
1101 *end = state->end;
1102 goto out;
1103 }
1104 if (!found)
1105 *start = state->start;
1106 found++;
1107 *end = state->end;
1108 cur_start = state->end + 1;
1109 node = rb_next(node);
1110 if (!node)
1111 break;
1112 total_bytes += state->end - state->start + 1;
1113 if (total_bytes >= max_bytes)
1114 break;
1115 }
1116 out:
1117 spin_unlock(&tree->lock);
1118 return found;
1119 }
1120
1121 static noinline int __unlock_for_delalloc(struct inode *inode,
1122 struct page *locked_page,
1123 u64 start, u64 end)
1124 {
1125 int ret;
1126 struct page *pages[16];
1127 unsigned long index = start >> PAGE_CACHE_SHIFT;
1128 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1129 unsigned long nr_pages = end_index - index + 1;
1130 int i;
1131
1132 if (index == locked_page->index && end_index == index)
1133 return 0;
1134
1135 while (nr_pages > 0) {
1136 ret = find_get_pages_contig(inode->i_mapping, index,
1137 min_t(unsigned long, nr_pages,
1138 ARRAY_SIZE(pages)), pages);
1139 for (i = 0; i < ret; i++) {
1140 if (pages[i] != locked_page)
1141 unlock_page(pages[i]);
1142 page_cache_release(pages[i]);
1143 }
1144 nr_pages -= ret;
1145 index += ret;
1146 cond_resched();
1147 }
1148 return 0;
1149 }
1150
1151 static noinline int lock_delalloc_pages(struct inode *inode,
1152 struct page *locked_page,
1153 u64 delalloc_start,
1154 u64 delalloc_end)
1155 {
1156 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1157 unsigned long start_index = index;
1158 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1159 unsigned long pages_locked = 0;
1160 struct page *pages[16];
1161 unsigned long nrpages;
1162 int ret;
1163 int i;
1164
1165 /* the caller is responsible for locking the start index */
1166 if (index == locked_page->index && index == end_index)
1167 return 0;
1168
1169 /* skip the page at the start index */
1170 nrpages = end_index - index + 1;
1171 while (nrpages > 0) {
1172 ret = find_get_pages_contig(inode->i_mapping, index,
1173 min_t(unsigned long,
1174 nrpages, ARRAY_SIZE(pages)), pages);
1175 if (ret == 0) {
1176 ret = -EAGAIN;
1177 goto done;
1178 }
1179 /* now we have an array of pages, lock them all */
1180 for (i = 0; i < ret; i++) {
1181 /*
1182 * the caller is taking responsibility for
1183 * locked_page
1184 */
1185 if (pages[i] != locked_page) {
1186 lock_page(pages[i]);
1187 if (!PageDirty(pages[i]) ||
1188 pages[i]->mapping != inode->i_mapping) {
1189 ret = -EAGAIN;
1190 unlock_page(pages[i]);
1191 page_cache_release(pages[i]);
1192 goto done;
1193 }
1194 }
1195 page_cache_release(pages[i]);
1196 pages_locked++;
1197 }
1198 nrpages -= ret;
1199 index += ret;
1200 cond_resched();
1201 }
1202 ret = 0;
1203 done:
1204 if (ret && pages_locked) {
1205 __unlock_for_delalloc(inode, locked_page,
1206 delalloc_start,
1207 ((u64)(start_index + pages_locked - 1)) <<
1208 PAGE_CACHE_SHIFT);
1209 }
1210 return ret;
1211 }
1212
1213 /*
1214 * find a contiguous range of bytes in the file marked as delalloc, not
1215 * more than 'max_bytes'. start and end are used to return the range,
1216 *
1217 * 1 is returned if we find something, 0 if nothing was in the tree
1218 */
1219 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1220 struct extent_io_tree *tree,
1221 struct page *locked_page,
1222 u64 *start, u64 *end,
1223 u64 max_bytes)
1224 {
1225 u64 delalloc_start;
1226 u64 delalloc_end;
1227 u64 found;
1228 int ret;
1229 int loops = 0;
1230
1231 again:
1232 /* step one, find a bunch of delalloc bytes starting at start */
1233 delalloc_start = *start;
1234 delalloc_end = 0;
1235 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1236 max_bytes);
1237 if (!found || delalloc_end <= *start) {
1238 *start = delalloc_start;
1239 *end = delalloc_end;
1240 return found;
1241 }
1242
1243 /*
1244 * start comes from the offset of locked_page. We have to lock
1245 * pages in order, so we can't process delalloc bytes before
1246 * locked_page
1247 */
1248 if (delalloc_start < *start)
1249 delalloc_start = *start;
1250
1251 /*
1252 * make sure to limit the number of pages we try to lock down
1253 * if we're looping.
1254 */
1255 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1256 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1257
1258 /* step two, lock all the pages after the page that has start */
1259 ret = lock_delalloc_pages(inode, locked_page,
1260 delalloc_start, delalloc_end);
1261 if (ret == -EAGAIN) {
1262 /* some of the pages are gone, lets avoid looping by
1263 * shortening the size of the delalloc range we're searching
1264 */
1265 if (!loops) {
1266 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1267 max_bytes = PAGE_CACHE_SIZE - offset;
1268 loops = 1;
1269 goto again;
1270 } else {
1271 found = 0;
1272 goto out_failed;
1273 }
1274 }
1275 BUG_ON(ret);
1276
1277 /* step three, lock the state bits for the whole range */
1278 lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1279
1280 /* then test to make sure it is all still delalloc */
1281 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1282 EXTENT_DELALLOC, 1);
1283 if (!ret) {
1284 unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS);
1285 __unlock_for_delalloc(inode, locked_page,
1286 delalloc_start, delalloc_end);
1287 cond_resched();
1288 goto again;
1289 }
1290 *start = delalloc_start;
1291 *end = delalloc_end;
1292 out_failed:
1293 return found;
1294 }
1295
1296 int extent_clear_unlock_delalloc(struct inode *inode,
1297 struct extent_io_tree *tree,
1298 u64 start, u64 end, struct page *locked_page,
1299 int unlock_pages,
1300 int clear_unlock,
1301 int clear_delalloc, int clear_dirty,
1302 int set_writeback,
1303 int end_writeback)
1304 {
1305 int ret;
1306 struct page *pages[16];
1307 unsigned long index = start >> PAGE_CACHE_SHIFT;
1308 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1309 unsigned long nr_pages = end_index - index + 1;
1310 int i;
1311 int clear_bits = 0;
1312
1313 if (clear_unlock)
1314 clear_bits |= EXTENT_LOCKED;
1315 if (clear_dirty)
1316 clear_bits |= EXTENT_DIRTY;
1317
1318 if (clear_delalloc)
1319 clear_bits |= EXTENT_DELALLOC;
1320
1321 clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
1322 if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
1323 return 0;
1324
1325 while (nr_pages > 0) {
1326 ret = find_get_pages_contig(inode->i_mapping, index,
1327 min_t(unsigned long,
1328 nr_pages, ARRAY_SIZE(pages)), pages);
1329 for (i = 0; i < ret; i++) {
1330 if (pages[i] == locked_page) {
1331 page_cache_release(pages[i]);
1332 continue;
1333 }
1334 if (clear_dirty)
1335 clear_page_dirty_for_io(pages[i]);
1336 if (set_writeback)
1337 set_page_writeback(pages[i]);
1338 if (end_writeback)
1339 end_page_writeback(pages[i]);
1340 if (unlock_pages)
1341 unlock_page(pages[i]);
1342 page_cache_release(pages[i]);
1343 }
1344 nr_pages -= ret;
1345 index += ret;
1346 cond_resched();
1347 }
1348 return 0;
1349 }
1350
1351 /*
1352 * count the number of bytes in the tree that have a given bit(s)
1353 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1354 * cached. The total number found is returned.
1355 */
1356 u64 count_range_bits(struct extent_io_tree *tree,
1357 u64 *start, u64 search_end, u64 max_bytes,
1358 unsigned long bits)
1359 {
1360 struct rb_node *node;
1361 struct extent_state *state;
1362 u64 cur_start = *start;
1363 u64 total_bytes = 0;
1364 int found = 0;
1365
1366 if (search_end <= cur_start) {
1367 WARN_ON(1);
1368 return 0;
1369 }
1370
1371 spin_lock(&tree->lock);
1372 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1373 total_bytes = tree->dirty_bytes;
1374 goto out;
1375 }
1376 /*
1377 * this search will find all the extents that end after
1378 * our range starts.
1379 */
1380 node = tree_search(tree, cur_start);
1381 if (!node)
1382 goto out;
1383
1384 while (1) {
1385 state = rb_entry(node, struct extent_state, rb_node);
1386 if (state->start > search_end)
1387 break;
1388 if (state->end >= cur_start && (state->state & bits)) {
1389 total_bytes += min(search_end, state->end) + 1 -
1390 max(cur_start, state->start);
1391 if (total_bytes >= max_bytes)
1392 break;
1393 if (!found) {
1394 *start = state->start;
1395 found = 1;
1396 }
1397 }
1398 node = rb_next(node);
1399 if (!node)
1400 break;
1401 }
1402 out:
1403 spin_unlock(&tree->lock);
1404 return total_bytes;
1405 }
1406
1407 #if 0
1408 /*
1409 * helper function to lock both pages and extents in the tree.
1410 * pages must be locked first.
1411 */
1412 static int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1413 {
1414 unsigned long index = start >> PAGE_CACHE_SHIFT;
1415 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1416 struct page *page;
1417 int err;
1418
1419 while (index <= end_index) {
1420 page = grab_cache_page(tree->mapping, index);
1421 if (!page) {
1422 err = -ENOMEM;
1423 goto failed;
1424 }
1425 if (IS_ERR(page)) {
1426 err = PTR_ERR(page);
1427 goto failed;
1428 }
1429 index++;
1430 }
1431 lock_extent(tree, start, end, GFP_NOFS);
1432 return 0;
1433
1434 failed:
1435 /*
1436 * we failed above in getting the page at 'index', so we undo here
1437 * up to but not including the page at 'index'
1438 */
1439 end_index = index;
1440 index = start >> PAGE_CACHE_SHIFT;
1441 while (index < end_index) {
1442 page = find_get_page(tree->mapping, index);
1443 unlock_page(page);
1444 page_cache_release(page);
1445 index++;
1446 }
1447 return err;
1448 }
1449
1450 /*
1451 * helper function to unlock both pages and extents in the tree.
1452 */
1453 static int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1454 {
1455 unsigned long index = start >> PAGE_CACHE_SHIFT;
1456 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1457 struct page *page;
1458
1459 while (index <= end_index) {
1460 page = find_get_page(tree->mapping, index);
1461 unlock_page(page);
1462 page_cache_release(page);
1463 index++;
1464 }
1465 unlock_extent(tree, start, end, GFP_NOFS);
1466 return 0;
1467 }
1468 #endif
1469
1470 /*
1471 * set the private field for a given byte offset in the tree. If there isn't
1472 * an extent_state there already, this does nothing.
1473 */
1474 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1475 {
1476 struct rb_node *node;
1477 struct extent_state *state;
1478 int ret = 0;
1479
1480 spin_lock(&tree->lock);
1481 /*
1482 * this search will find all the extents that end after
1483 * our range starts.
1484 */
1485 node = tree_search(tree, start);
1486 if (!node) {
1487 ret = -ENOENT;
1488 goto out;
1489 }
1490 state = rb_entry(node, struct extent_state, rb_node);
1491 if (state->start != start) {
1492 ret = -ENOENT;
1493 goto out;
1494 }
1495 state->private = private;
1496 out:
1497 spin_unlock(&tree->lock);
1498 return ret;
1499 }
1500
1501 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1502 {
1503 struct rb_node *node;
1504 struct extent_state *state;
1505 int ret = 0;
1506
1507 spin_lock(&tree->lock);
1508 /*
1509 * this search will find all the extents that end after
1510 * our range starts.
1511 */
1512 node = tree_search(tree, start);
1513 if (!node) {
1514 ret = -ENOENT;
1515 goto out;
1516 }
1517 state = rb_entry(node, struct extent_state, rb_node);
1518 if (state->start != start) {
1519 ret = -ENOENT;
1520 goto out;
1521 }
1522 *private = state->private;
1523 out:
1524 spin_unlock(&tree->lock);
1525 return ret;
1526 }
1527
1528 /*
1529 * searches a range in the state tree for a given mask.
1530 * If 'filled' == 1, this returns 1 only if every extent in the tree
1531 * has the bits set. Otherwise, 1 is returned if any bit in the
1532 * range is found set.
1533 */
1534 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1535 int bits, int filled)
1536 {
1537 struct extent_state *state = NULL;
1538 struct rb_node *node;
1539 int bitset = 0;
1540
1541 spin_lock(&tree->lock);
1542 node = tree_search(tree, start);
1543 while (node && start <= end) {
1544 state = rb_entry(node, struct extent_state, rb_node);
1545
1546 if (filled && state->start > start) {
1547 bitset = 0;
1548 break;
1549 }
1550
1551 if (state->start > end)
1552 break;
1553
1554 if (state->state & bits) {
1555 bitset = 1;
1556 if (!filled)
1557 break;
1558 } else if (filled) {
1559 bitset = 0;
1560 break;
1561 }
1562 start = state->end + 1;
1563 if (start > end)
1564 break;
1565 node = rb_next(node);
1566 if (!node) {
1567 if (filled)
1568 bitset = 0;
1569 break;
1570 }
1571 }
1572 spin_unlock(&tree->lock);
1573 return bitset;
1574 }
1575
1576 /*
1577 * helper function to set a given page up to date if all the
1578 * extents in the tree for that page are up to date
1579 */
1580 static int check_page_uptodate(struct extent_io_tree *tree,
1581 struct page *page)
1582 {
1583 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1584 u64 end = start + PAGE_CACHE_SIZE - 1;
1585 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1586 SetPageUptodate(page);
1587 return 0;
1588 }
1589
1590 /*
1591 * helper function to unlock a page if all the extents in the tree
1592 * for that page are unlocked
1593 */
1594 static int check_page_locked(struct extent_io_tree *tree,
1595 struct page *page)
1596 {
1597 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1598 u64 end = start + PAGE_CACHE_SIZE - 1;
1599 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1600 unlock_page(page);
1601 return 0;
1602 }
1603
1604 /*
1605 * helper function to end page writeback if all the extents
1606 * in the tree for that page are done with writeback
1607 */
1608 static int check_page_writeback(struct extent_io_tree *tree,
1609 struct page *page)
1610 {
1611 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1612 u64 end = start + PAGE_CACHE_SIZE - 1;
1613 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1614 end_page_writeback(page);
1615 return 0;
1616 }
1617
1618 /* lots and lots of room for performance fixes in the end_bio funcs */
1619
1620 /*
1621 * after a writepage IO is done, we need to:
1622 * clear the uptodate bits on error
1623 * clear the writeback bits in the extent tree for this IO
1624 * end_page_writeback if the page has no more pending IO
1625 *
1626 * Scheduling is not allowed, so the extent state tree is expected
1627 * to have one and only one object corresponding to this IO.
1628 */
1629 static void end_bio_extent_writepage(struct bio *bio, int err)
1630 {
1631 int uptodate = err == 0;
1632 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1633 struct extent_io_tree *tree;
1634 u64 start;
1635 u64 end;
1636 int whole_page;
1637 int ret;
1638
1639 do {
1640 struct page *page = bvec->bv_page;
1641 tree = &BTRFS_I(page->mapping->host)->io_tree;
1642
1643 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1644 bvec->bv_offset;
1645 end = start + bvec->bv_len - 1;
1646
1647 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1648 whole_page = 1;
1649 else
1650 whole_page = 0;
1651
1652 if (--bvec >= bio->bi_io_vec)
1653 prefetchw(&bvec->bv_page->flags);
1654 if (tree->ops && tree->ops->writepage_end_io_hook) {
1655 ret = tree->ops->writepage_end_io_hook(page, start,
1656 end, NULL, uptodate);
1657 if (ret)
1658 uptodate = 0;
1659 }
1660
1661 if (!uptodate && tree->ops &&
1662 tree->ops->writepage_io_failed_hook) {
1663 ret = tree->ops->writepage_io_failed_hook(bio, page,
1664 start, end, NULL);
1665 if (ret == 0) {
1666 uptodate = (err == 0);
1667 continue;
1668 }
1669 }
1670
1671 if (!uptodate) {
1672 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1673 ClearPageUptodate(page);
1674 SetPageError(page);
1675 }
1676
1677 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1678
1679 if (whole_page)
1680 end_page_writeback(page);
1681 else
1682 check_page_writeback(tree, page);
1683 } while (bvec >= bio->bi_io_vec);
1684
1685 bio_put(bio);
1686 }
1687
1688 /*
1689 * after a readpage IO is done, we need to:
1690 * clear the uptodate bits on error
1691 * set the uptodate bits if things worked
1692 * set the page up to date if all extents in the tree are uptodate
1693 * clear the lock bit in the extent tree
1694 * unlock the page if there are no other extents locked for it
1695 *
1696 * Scheduling is not allowed, so the extent state tree is expected
1697 * to have one and only one object corresponding to this IO.
1698 */
1699 static void end_bio_extent_readpage(struct bio *bio, int err)
1700 {
1701 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1702 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1703 struct extent_io_tree *tree;
1704 u64 start;
1705 u64 end;
1706 int whole_page;
1707 int ret;
1708
1709 if (err)
1710 uptodate = 0;
1711
1712 do {
1713 struct page *page = bvec->bv_page;
1714 tree = &BTRFS_I(page->mapping->host)->io_tree;
1715
1716 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1717 bvec->bv_offset;
1718 end = start + bvec->bv_len - 1;
1719
1720 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1721 whole_page = 1;
1722 else
1723 whole_page = 0;
1724
1725 if (--bvec >= bio->bi_io_vec)
1726 prefetchw(&bvec->bv_page->flags);
1727
1728 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1729 ret = tree->ops->readpage_end_io_hook(page, start, end,
1730 NULL);
1731 if (ret)
1732 uptodate = 0;
1733 }
1734 if (!uptodate && tree->ops &&
1735 tree->ops->readpage_io_failed_hook) {
1736 ret = tree->ops->readpage_io_failed_hook(bio, page,
1737 start, end, NULL);
1738 if (ret == 0) {
1739 uptodate =
1740 test_bit(BIO_UPTODATE, &bio->bi_flags);
1741 if (err)
1742 uptodate = 0;
1743 continue;
1744 }
1745 }
1746
1747 if (uptodate) {
1748 set_extent_uptodate(tree, start, end,
1749 GFP_ATOMIC);
1750 }
1751 unlock_extent(tree, start, end, GFP_ATOMIC);
1752
1753 if (whole_page) {
1754 if (uptodate) {
1755 SetPageUptodate(page);
1756 } else {
1757 ClearPageUptodate(page);
1758 SetPageError(page);
1759 }
1760 unlock_page(page);
1761 } else {
1762 if (uptodate) {
1763 check_page_uptodate(tree, page);
1764 } else {
1765 ClearPageUptodate(page);
1766 SetPageError(page);
1767 }
1768 check_page_locked(tree, page);
1769 }
1770 } while (bvec >= bio->bi_io_vec);
1771
1772 bio_put(bio);
1773 }
1774
1775 /*
1776 * IO done from prepare_write is pretty simple, we just unlock
1777 * the structs in the extent tree when done, and set the uptodate bits
1778 * as appropriate.
1779 */
1780 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1781 {
1782 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1783 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1784 struct extent_io_tree *tree;
1785 u64 start;
1786 u64 end;
1787
1788 do {
1789 struct page *page = bvec->bv_page;
1790 tree = &BTRFS_I(page->mapping->host)->io_tree;
1791
1792 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1793 bvec->bv_offset;
1794 end = start + bvec->bv_len - 1;
1795
1796 if (--bvec >= bio->bi_io_vec)
1797 prefetchw(&bvec->bv_page->flags);
1798
1799 if (uptodate) {
1800 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1801 } else {
1802 ClearPageUptodate(page);
1803 SetPageError(page);
1804 }
1805
1806 unlock_extent(tree, start, end, GFP_ATOMIC);
1807
1808 } while (bvec >= bio->bi_io_vec);
1809
1810 bio_put(bio);
1811 }
1812
1813 static struct bio *
1814 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1815 gfp_t gfp_flags)
1816 {
1817 struct bio *bio;
1818
1819 bio = bio_alloc(gfp_flags, nr_vecs);
1820
1821 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1822 while (!bio && (nr_vecs /= 2))
1823 bio = bio_alloc(gfp_flags, nr_vecs);
1824 }
1825
1826 if (bio) {
1827 bio->bi_size = 0;
1828 bio->bi_bdev = bdev;
1829 bio->bi_sector = first_sector;
1830 }
1831 return bio;
1832 }
1833
1834 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1835 unsigned long bio_flags)
1836 {
1837 int ret = 0;
1838 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1839 struct page *page = bvec->bv_page;
1840 struct extent_io_tree *tree = bio->bi_private;
1841 u64 start;
1842 u64 end;
1843
1844 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1845 end = start + bvec->bv_len - 1;
1846
1847 bio->bi_private = NULL;
1848
1849 bio_get(bio);
1850
1851 if (tree->ops && tree->ops->submit_bio_hook)
1852 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1853 mirror_num, bio_flags);
1854 else
1855 submit_bio(rw, bio);
1856 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1857 ret = -EOPNOTSUPP;
1858 bio_put(bio);
1859 return ret;
1860 }
1861
1862 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1863 struct page *page, sector_t sector,
1864 size_t size, unsigned long offset,
1865 struct block_device *bdev,
1866 struct bio **bio_ret,
1867 unsigned long max_pages,
1868 bio_end_io_t end_io_func,
1869 int mirror_num,
1870 unsigned long prev_bio_flags,
1871 unsigned long bio_flags)
1872 {
1873 int ret = 0;
1874 struct bio *bio;
1875 int nr;
1876 int contig = 0;
1877 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
1878 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
1879 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
1880
1881 if (bio_ret && *bio_ret) {
1882 bio = *bio_ret;
1883 if (old_compressed)
1884 contig = bio->bi_sector == sector;
1885 else
1886 contig = bio->bi_sector + (bio->bi_size >> 9) ==
1887 sector;
1888
1889 if (prev_bio_flags != bio_flags || !contig ||
1890 (tree->ops && tree->ops->merge_bio_hook &&
1891 tree->ops->merge_bio_hook(page, offset, page_size, bio,
1892 bio_flags)) ||
1893 bio_add_page(bio, page, page_size, offset) < page_size) {
1894 ret = submit_one_bio(rw, bio, mirror_num,
1895 prev_bio_flags);
1896 bio = NULL;
1897 } else {
1898 return 0;
1899 }
1900 }
1901 if (this_compressed)
1902 nr = BIO_MAX_PAGES;
1903 else
1904 nr = bio_get_nr_vecs(bdev);
1905
1906 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1907
1908 bio_add_page(bio, page, page_size, offset);
1909 bio->bi_end_io = end_io_func;
1910 bio->bi_private = tree;
1911
1912 if (bio_ret)
1913 *bio_ret = bio;
1914 else
1915 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
1916
1917 return ret;
1918 }
1919
1920 void set_page_extent_mapped(struct page *page)
1921 {
1922 if (!PagePrivate(page)) {
1923 SetPagePrivate(page);
1924 page_cache_get(page);
1925 set_page_private(page, EXTENT_PAGE_PRIVATE);
1926 }
1927 }
1928
1929 static void set_page_extent_head(struct page *page, unsigned long len)
1930 {
1931 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1932 }
1933
1934 /*
1935 * basic readpage implementation. Locked extent state structs are inserted
1936 * into the tree that are removed when the IO is done (by the end_io
1937 * handlers)
1938 */
1939 static int __extent_read_full_page(struct extent_io_tree *tree,
1940 struct page *page,
1941 get_extent_t *get_extent,
1942 struct bio **bio, int mirror_num,
1943 unsigned long *bio_flags)
1944 {
1945 struct inode *inode = page->mapping->host;
1946 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1947 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1948 u64 end;
1949 u64 cur = start;
1950 u64 extent_offset;
1951 u64 last_byte = i_size_read(inode);
1952 u64 block_start;
1953 u64 cur_end;
1954 sector_t sector;
1955 struct extent_map *em;
1956 struct block_device *bdev;
1957 int ret;
1958 int nr = 0;
1959 size_t page_offset = 0;
1960 size_t iosize;
1961 size_t disk_io_size;
1962 size_t blocksize = inode->i_sb->s_blocksize;
1963 unsigned long this_bio_flag = 0;
1964
1965 set_page_extent_mapped(page);
1966
1967 end = page_end;
1968 lock_extent(tree, start, end, GFP_NOFS);
1969
1970 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
1971 char *userpage;
1972 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
1973
1974 if (zero_offset) {
1975 iosize = PAGE_CACHE_SIZE - zero_offset;
1976 userpage = kmap_atomic(page, KM_USER0);
1977 memset(userpage + zero_offset, 0, iosize);
1978 flush_dcache_page(page);
1979 kunmap_atomic(userpage, KM_USER0);
1980 }
1981 }
1982 while (cur <= end) {
1983 if (cur >= last_byte) {
1984 char *userpage;
1985 iosize = PAGE_CACHE_SIZE - page_offset;
1986 userpage = kmap_atomic(page, KM_USER0);
1987 memset(userpage + page_offset, 0, iosize);
1988 flush_dcache_page(page);
1989 kunmap_atomic(userpage, KM_USER0);
1990 set_extent_uptodate(tree, cur, cur + iosize - 1,
1991 GFP_NOFS);
1992 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1993 break;
1994 }
1995 em = get_extent(inode, page, page_offset, cur,
1996 end - cur + 1, 0);
1997 if (IS_ERR(em) || !em) {
1998 SetPageError(page);
1999 unlock_extent(tree, cur, end, GFP_NOFS);
2000 break;
2001 }
2002 extent_offset = cur - em->start;
2003 BUG_ON(extent_map_end(em) <= cur);
2004 BUG_ON(end < cur);
2005
2006 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2007 this_bio_flag = EXTENT_BIO_COMPRESSED;
2008
2009 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2010 cur_end = min(extent_map_end(em) - 1, end);
2011 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2012 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2013 disk_io_size = em->block_len;
2014 sector = em->block_start >> 9;
2015 } else {
2016 sector = (em->block_start + extent_offset) >> 9;
2017 disk_io_size = iosize;
2018 }
2019 bdev = em->bdev;
2020 block_start = em->block_start;
2021 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2022 block_start = EXTENT_MAP_HOLE;
2023 free_extent_map(em);
2024 em = NULL;
2025
2026 /* we've found a hole, just zero and go on */
2027 if (block_start == EXTENT_MAP_HOLE) {
2028 char *userpage;
2029 userpage = kmap_atomic(page, KM_USER0);
2030 memset(userpage + page_offset, 0, iosize);
2031 flush_dcache_page(page);
2032 kunmap_atomic(userpage, KM_USER0);
2033
2034 set_extent_uptodate(tree, cur, cur + iosize - 1,
2035 GFP_NOFS);
2036 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2037 cur = cur + iosize;
2038 page_offset += iosize;
2039 continue;
2040 }
2041 /* the get_extent function already copied into the page */
2042 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
2043 check_page_uptodate(tree, page);
2044 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2045 cur = cur + iosize;
2046 page_offset += iosize;
2047 continue;
2048 }
2049 /* we have an inline extent but it didn't get marked up
2050 * to date. Error out
2051 */
2052 if (block_start == EXTENT_MAP_INLINE) {
2053 SetPageError(page);
2054 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2055 cur = cur + iosize;
2056 page_offset += iosize;
2057 continue;
2058 }
2059
2060 ret = 0;
2061 if (tree->ops && tree->ops->readpage_io_hook) {
2062 ret = tree->ops->readpage_io_hook(page, cur,
2063 cur + iosize - 1);
2064 }
2065 if (!ret) {
2066 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2067 pnr -= page->index;
2068 ret = submit_extent_page(READ, tree, page,
2069 sector, disk_io_size, page_offset,
2070 bdev, bio, pnr,
2071 end_bio_extent_readpage, mirror_num,
2072 *bio_flags,
2073 this_bio_flag);
2074 nr++;
2075 *bio_flags = this_bio_flag;
2076 }
2077 if (ret)
2078 SetPageError(page);
2079 cur = cur + iosize;
2080 page_offset += iosize;
2081 }
2082 if (!nr) {
2083 if (!PageError(page))
2084 SetPageUptodate(page);
2085 unlock_page(page);
2086 }
2087 return 0;
2088 }
2089
2090 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2091 get_extent_t *get_extent)
2092 {
2093 struct bio *bio = NULL;
2094 unsigned long bio_flags = 0;
2095 int ret;
2096
2097 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2098 &bio_flags);
2099 if (bio)
2100 submit_one_bio(READ, bio, 0, bio_flags);
2101 return ret;
2102 }
2103
2104 /*
2105 * the writepage semantics are similar to regular writepage. extent
2106 * records are inserted to lock ranges in the tree, and as dirty areas
2107 * are found, they are marked writeback. Then the lock bits are removed
2108 * and the end_io handler clears the writeback ranges
2109 */
2110 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2111 void *data)
2112 {
2113 struct inode *inode = page->mapping->host;
2114 struct extent_page_data *epd = data;
2115 struct extent_io_tree *tree = epd->tree;
2116 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2117 u64 delalloc_start;
2118 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2119 u64 end;
2120 u64 cur = start;
2121 u64 extent_offset;
2122 u64 last_byte = i_size_read(inode);
2123 u64 block_start;
2124 u64 iosize;
2125 u64 unlock_start;
2126 sector_t sector;
2127 struct extent_map *em;
2128 struct block_device *bdev;
2129 int ret;
2130 int nr = 0;
2131 size_t pg_offset = 0;
2132 size_t blocksize;
2133 loff_t i_size = i_size_read(inode);
2134 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2135 u64 nr_delalloc;
2136 u64 delalloc_end;
2137 int page_started;
2138 int compressed;
2139 unsigned long nr_written = 0;
2140
2141 WARN_ON(!PageLocked(page));
2142 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2143 if (page->index > end_index ||
2144 (page->index == end_index && !pg_offset)) {
2145 page->mapping->a_ops->invalidatepage(page, 0);
2146 unlock_page(page);
2147 return 0;
2148 }
2149
2150 if (page->index == end_index) {
2151 char *userpage;
2152
2153 userpage = kmap_atomic(page, KM_USER0);
2154 memset(userpage + pg_offset, 0,
2155 PAGE_CACHE_SIZE - pg_offset);
2156 kunmap_atomic(userpage, KM_USER0);
2157 flush_dcache_page(page);
2158 }
2159 pg_offset = 0;
2160
2161 set_page_extent_mapped(page);
2162
2163 delalloc_start = start;
2164 delalloc_end = 0;
2165 page_started = 0;
2166 if (!epd->extent_locked) {
2167 while (delalloc_end < page_end) {
2168 nr_delalloc = find_lock_delalloc_range(inode, tree,
2169 page,
2170 &delalloc_start,
2171 &delalloc_end,
2172 128 * 1024 * 1024);
2173 if (nr_delalloc == 0) {
2174 delalloc_start = delalloc_end + 1;
2175 continue;
2176 }
2177 tree->ops->fill_delalloc(inode, page, delalloc_start,
2178 delalloc_end, &page_started,
2179 &nr_written);
2180 delalloc_start = delalloc_end + 1;
2181 }
2182
2183 /* did the fill delalloc function already unlock and start
2184 * the IO?
2185 */
2186 if (page_started) {
2187 ret = 0;
2188 goto update_nr_written;
2189 }
2190 }
2191 lock_extent(tree, start, page_end, GFP_NOFS);
2192
2193 unlock_start = start;
2194
2195 if (tree->ops && tree->ops->writepage_start_hook) {
2196 ret = tree->ops->writepage_start_hook(page, start,
2197 page_end);
2198 if (ret == -EAGAIN) {
2199 unlock_extent(tree, start, page_end, GFP_NOFS);
2200 redirty_page_for_writepage(wbc, page);
2201 unlock_page(page);
2202 ret = 0;
2203 goto update_nr_written;
2204 }
2205 }
2206
2207 nr_written++;
2208
2209 end = page_end;
2210 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0))
2211 printk(KERN_ERR "btrfs delalloc bits after lock_extent\n");
2212
2213 if (last_byte <= start) {
2214 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2215 unlock_extent(tree, start, page_end, GFP_NOFS);
2216 if (tree->ops && tree->ops->writepage_end_io_hook)
2217 tree->ops->writepage_end_io_hook(page, start,
2218 page_end, NULL, 1);
2219 unlock_start = page_end + 1;
2220 goto done;
2221 }
2222
2223 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2224 blocksize = inode->i_sb->s_blocksize;
2225
2226 while (cur <= end) {
2227 if (cur >= last_byte) {
2228 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2229 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2230 if (tree->ops && tree->ops->writepage_end_io_hook)
2231 tree->ops->writepage_end_io_hook(page, cur,
2232 page_end, NULL, 1);
2233 unlock_start = page_end + 1;
2234 break;
2235 }
2236 em = epd->get_extent(inode, page, pg_offset, cur,
2237 end - cur + 1, 1);
2238 if (IS_ERR(em) || !em) {
2239 SetPageError(page);
2240 break;
2241 }
2242
2243 extent_offset = cur - em->start;
2244 BUG_ON(extent_map_end(em) <= cur);
2245 BUG_ON(end < cur);
2246 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2247 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2248 sector = (em->block_start + extent_offset) >> 9;
2249 bdev = em->bdev;
2250 block_start = em->block_start;
2251 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2252 free_extent_map(em);
2253 em = NULL;
2254
2255 /*
2256 * compressed and inline extents are written through other
2257 * paths in the FS
2258 */
2259 if (compressed || block_start == EXTENT_MAP_HOLE ||
2260 block_start == EXTENT_MAP_INLINE) {
2261 clear_extent_dirty(tree, cur,
2262 cur + iosize - 1, GFP_NOFS);
2263
2264 unlock_extent(tree, unlock_start, cur + iosize - 1,
2265 GFP_NOFS);
2266
2267 /*
2268 * end_io notification does not happen here for
2269 * compressed extents
2270 */
2271 if (!compressed && tree->ops &&
2272 tree->ops->writepage_end_io_hook)
2273 tree->ops->writepage_end_io_hook(page, cur,
2274 cur + iosize - 1,
2275 NULL, 1);
2276 else if (compressed) {
2277 /* we don't want to end_page_writeback on
2278 * a compressed extent. this happens
2279 * elsewhere
2280 */
2281 nr++;
2282 }
2283
2284 cur += iosize;
2285 pg_offset += iosize;
2286 unlock_start = cur;
2287 continue;
2288 }
2289 /* leave this out until we have a page_mkwrite call */
2290 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2291 EXTENT_DIRTY, 0)) {
2292 cur = cur + iosize;
2293 pg_offset += iosize;
2294 continue;
2295 }
2296
2297 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2298 if (tree->ops && tree->ops->writepage_io_hook) {
2299 ret = tree->ops->writepage_io_hook(page, cur,
2300 cur + iosize - 1);
2301 } else {
2302 ret = 0;
2303 }
2304 if (ret) {
2305 SetPageError(page);
2306 } else {
2307 unsigned long max_nr = end_index + 1;
2308
2309 set_range_writeback(tree, cur, cur + iosize - 1);
2310 if (!PageWriteback(page)) {
2311 printk(KERN_ERR "btrfs warning page %lu not "
2312 "writeback, cur %llu end %llu\n",
2313 page->index, (unsigned long long)cur,
2314 (unsigned long long)end);
2315 }
2316
2317 ret = submit_extent_page(WRITE, tree, page, sector,
2318 iosize, pg_offset, bdev,
2319 &epd->bio, max_nr,
2320 end_bio_extent_writepage,
2321 0, 0, 0);
2322 if (ret)
2323 SetPageError(page);
2324 }
2325 cur = cur + iosize;
2326 pg_offset += iosize;
2327 nr++;
2328 }
2329 done:
2330 if (nr == 0) {
2331 /* make sure the mapping tag for page dirty gets cleared */
2332 set_page_writeback(page);
2333 end_page_writeback(page);
2334 }
2335 if (unlock_start <= page_end)
2336 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2337 unlock_page(page);
2338
2339 update_nr_written:
2340 wbc->nr_to_write -= nr_written;
2341 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2342 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2343 page->mapping->writeback_index = page->index + nr_written;
2344 return 0;
2345 }
2346
2347 /**
2348 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2349 * @mapping: address space structure to write
2350 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2351 * @writepage: function called for each page
2352 * @data: data passed to writepage function
2353 *
2354 * If a page is already under I/O, write_cache_pages() skips it, even
2355 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2356 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2357 * and msync() need to guarantee that all the data which was dirty at the time
2358 * the call was made get new I/O started against them. If wbc->sync_mode is
2359 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2360 * existing IO to complete.
2361 */
2362 static int extent_write_cache_pages(struct extent_io_tree *tree,
2363 struct address_space *mapping,
2364 struct writeback_control *wbc,
2365 writepage_t writepage, void *data,
2366 void (*flush_fn)(void *))
2367 {
2368 struct backing_dev_info *bdi = mapping->backing_dev_info;
2369 int ret = 0;
2370 int done = 0;
2371 struct pagevec pvec;
2372 int nr_pages;
2373 pgoff_t index;
2374 pgoff_t end; /* Inclusive */
2375 int scanned = 0;
2376 int range_whole = 0;
2377
2378 pagevec_init(&pvec, 0);
2379 if (wbc->range_cyclic) {
2380 index = mapping->writeback_index; /* Start from prev offset */
2381 end = -1;
2382 } else {
2383 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2384 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2385 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2386 range_whole = 1;
2387 scanned = 1;
2388 }
2389 retry:
2390 while (!done && (index <= end) &&
2391 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2392 PAGECACHE_TAG_DIRTY, min(end - index,
2393 (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2394 unsigned i;
2395
2396 scanned = 1;
2397 for (i = 0; i < nr_pages; i++) {
2398 struct page *page = pvec.pages[i];
2399
2400 /*
2401 * At this point we hold neither mapping->tree_lock nor
2402 * lock on the page itself: the page may be truncated or
2403 * invalidated (changing page->mapping to NULL), or even
2404 * swizzled back from swapper_space to tmpfs file
2405 * mapping
2406 */
2407 if (tree->ops && tree->ops->write_cache_pages_lock_hook)
2408 tree->ops->write_cache_pages_lock_hook(page);
2409 else
2410 lock_page(page);
2411
2412 if (unlikely(page->mapping != mapping)) {
2413 unlock_page(page);
2414 continue;
2415 }
2416
2417 if (!wbc->range_cyclic && page->index > end) {
2418 done = 1;
2419 unlock_page(page);
2420 continue;
2421 }
2422
2423 if (wbc->sync_mode != WB_SYNC_NONE) {
2424 if (PageWriteback(page))
2425 flush_fn(data);
2426 wait_on_page_writeback(page);
2427 }
2428
2429 if (PageWriteback(page) ||
2430 !clear_page_dirty_for_io(page)) {
2431 unlock_page(page);
2432 continue;
2433 }
2434
2435 ret = (*writepage)(page, wbc, data);
2436
2437 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2438 unlock_page(page);
2439 ret = 0;
2440 }
2441 if (ret || wbc->nr_to_write <= 0)
2442 done = 1;
2443 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2444 wbc->encountered_congestion = 1;
2445 done = 1;
2446 }
2447 }
2448 pagevec_release(&pvec);
2449 cond_resched();
2450 }
2451 if (!scanned && !done) {
2452 /*
2453 * We hit the last page and there is more work to be done: wrap
2454 * back to the start of the file
2455 */
2456 scanned = 1;
2457 index = 0;
2458 goto retry;
2459 }
2460 return ret;
2461 }
2462
2463 static noinline void flush_write_bio(void *data)
2464 {
2465 struct extent_page_data *epd = data;
2466 if (epd->bio) {
2467 submit_one_bio(WRITE, epd->bio, 0, 0);
2468 epd->bio = NULL;
2469 }
2470 }
2471
2472 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2473 get_extent_t *get_extent,
2474 struct writeback_control *wbc)
2475 {
2476 int ret;
2477 struct address_space *mapping = page->mapping;
2478 struct extent_page_data epd = {
2479 .bio = NULL,
2480 .tree = tree,
2481 .get_extent = get_extent,
2482 .extent_locked = 0,
2483 };
2484 struct writeback_control wbc_writepages = {
2485 .bdi = wbc->bdi,
2486 .sync_mode = WB_SYNC_NONE,
2487 .older_than_this = NULL,
2488 .nr_to_write = 64,
2489 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2490 .range_end = (loff_t)-1,
2491 };
2492
2493
2494 ret = __extent_writepage(page, wbc, &epd);
2495
2496 extent_write_cache_pages(tree, mapping, &wbc_writepages,
2497 __extent_writepage, &epd, flush_write_bio);
2498 if (epd.bio)
2499 submit_one_bio(WRITE, epd.bio, 0, 0);
2500 return ret;
2501 }
2502
2503 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2504 u64 start, u64 end, get_extent_t *get_extent,
2505 int mode)
2506 {
2507 int ret = 0;
2508 struct address_space *mapping = inode->i_mapping;
2509 struct page *page;
2510 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2511 PAGE_CACHE_SHIFT;
2512
2513 struct extent_page_data epd = {
2514 .bio = NULL,
2515 .tree = tree,
2516 .get_extent = get_extent,
2517 .extent_locked = 1,
2518 };
2519 struct writeback_control wbc_writepages = {
2520 .bdi = inode->i_mapping->backing_dev_info,
2521 .sync_mode = mode,
2522 .older_than_this = NULL,
2523 .nr_to_write = nr_pages * 2,
2524 .range_start = start,
2525 .range_end = end + 1,
2526 };
2527
2528 while (start <= end) {
2529 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2530 if (clear_page_dirty_for_io(page))
2531 ret = __extent_writepage(page, &wbc_writepages, &epd);
2532 else {
2533 if (tree->ops && tree->ops->writepage_end_io_hook)
2534 tree->ops->writepage_end_io_hook(page, start,
2535 start + PAGE_CACHE_SIZE - 1,
2536 NULL, 1);
2537 unlock_page(page);
2538 }
2539 page_cache_release(page);
2540 start += PAGE_CACHE_SIZE;
2541 }
2542
2543 if (epd.bio)
2544 submit_one_bio(WRITE, epd.bio, 0, 0);
2545 return ret;
2546 }
2547
2548 int extent_writepages(struct extent_io_tree *tree,
2549 struct address_space *mapping,
2550 get_extent_t *get_extent,
2551 struct writeback_control *wbc)
2552 {
2553 int ret = 0;
2554 struct extent_page_data epd = {
2555 .bio = NULL,
2556 .tree = tree,
2557 .get_extent = get_extent,
2558 .extent_locked = 0,
2559 };
2560
2561 ret = extent_write_cache_pages(tree, mapping, wbc,
2562 __extent_writepage, &epd,
2563 flush_write_bio);
2564 if (epd.bio)
2565 submit_one_bio(WRITE, epd.bio, 0, 0);
2566 return ret;
2567 }
2568
2569 int extent_readpages(struct extent_io_tree *tree,
2570 struct address_space *mapping,
2571 struct list_head *pages, unsigned nr_pages,
2572 get_extent_t get_extent)
2573 {
2574 struct bio *bio = NULL;
2575 unsigned page_idx;
2576 struct pagevec pvec;
2577 unsigned long bio_flags = 0;
2578
2579 pagevec_init(&pvec, 0);
2580 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2581 struct page *page = list_entry(pages->prev, struct page, lru);
2582
2583 prefetchw(&page->flags);
2584 list_del(&page->lru);
2585 /*
2586 * what we want to do here is call add_to_page_cache_lru,
2587 * but that isn't exported, so we reproduce it here
2588 */
2589 if (!add_to_page_cache(page, mapping,
2590 page->index, GFP_KERNEL)) {
2591
2592 /* open coding of lru_cache_add, also not exported */
2593 page_cache_get(page);
2594 if (!pagevec_add(&pvec, page))
2595 __pagevec_lru_add_file(&pvec);
2596 __extent_read_full_page(tree, page, get_extent,
2597 &bio, 0, &bio_flags);
2598 }
2599 page_cache_release(page);
2600 }
2601 if (pagevec_count(&pvec))
2602 __pagevec_lru_add_file(&pvec);
2603 BUG_ON(!list_empty(pages));
2604 if (bio)
2605 submit_one_bio(READ, bio, 0, bio_flags);
2606 return 0;
2607 }
2608
2609 /*
2610 * basic invalidatepage code, this waits on any locked or writeback
2611 * ranges corresponding to the page, and then deletes any extent state
2612 * records from the tree
2613 */
2614 int extent_invalidatepage(struct extent_io_tree *tree,
2615 struct page *page, unsigned long offset)
2616 {
2617 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2618 u64 end = start + PAGE_CACHE_SIZE - 1;
2619 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2620
2621 start += (offset + blocksize - 1) & ~(blocksize - 1);
2622 if (start > end)
2623 return 0;
2624
2625 lock_extent(tree, start, end, GFP_NOFS);
2626 wait_on_extent_writeback(tree, start, end);
2627 clear_extent_bit(tree, start, end,
2628 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2629 1, 1, GFP_NOFS);
2630 return 0;
2631 }
2632
2633 /*
2634 * simple commit_write call, set_range_dirty is used to mark both
2635 * the pages and the extent records as dirty
2636 */
2637 int extent_commit_write(struct extent_io_tree *tree,
2638 struct inode *inode, struct page *page,
2639 unsigned from, unsigned to)
2640 {
2641 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2642
2643 set_page_extent_mapped(page);
2644 set_page_dirty(page);
2645
2646 if (pos > inode->i_size) {
2647 i_size_write(inode, pos);
2648 mark_inode_dirty(inode);
2649 }
2650 return 0;
2651 }
2652
2653 int extent_prepare_write(struct extent_io_tree *tree,
2654 struct inode *inode, struct page *page,
2655 unsigned from, unsigned to, get_extent_t *get_extent)
2656 {
2657 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2658 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2659 u64 block_start;
2660 u64 orig_block_start;
2661 u64 block_end;
2662 u64 cur_end;
2663 struct extent_map *em;
2664 unsigned blocksize = 1 << inode->i_blkbits;
2665 size_t page_offset = 0;
2666 size_t block_off_start;
2667 size_t block_off_end;
2668 int err = 0;
2669 int iocount = 0;
2670 int ret = 0;
2671 int isnew;
2672
2673 set_page_extent_mapped(page);
2674
2675 block_start = (page_start + from) & ~((u64)blocksize - 1);
2676 block_end = (page_start + to - 1) | (blocksize - 1);
2677 orig_block_start = block_start;
2678
2679 lock_extent(tree, page_start, page_end, GFP_NOFS);
2680 while (block_start <= block_end) {
2681 em = get_extent(inode, page, page_offset, block_start,
2682 block_end - block_start + 1, 1);
2683 if (IS_ERR(em) || !em)
2684 goto err;
2685
2686 cur_end = min(block_end, extent_map_end(em) - 1);
2687 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2688 block_off_end = block_off_start + blocksize;
2689 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2690
2691 if (!PageUptodate(page) && isnew &&
2692 (block_off_end > to || block_off_start < from)) {
2693 void *kaddr;
2694
2695 kaddr = kmap_atomic(page, KM_USER0);
2696 if (block_off_end > to)
2697 memset(kaddr + to, 0, block_off_end - to);
2698 if (block_off_start < from)
2699 memset(kaddr + block_off_start, 0,
2700 from - block_off_start);
2701 flush_dcache_page(page);
2702 kunmap_atomic(kaddr, KM_USER0);
2703 }
2704 if ((em->block_start != EXTENT_MAP_HOLE &&
2705 em->block_start != EXTENT_MAP_INLINE) &&
2706 !isnew && !PageUptodate(page) &&
2707 (block_off_end > to || block_off_start < from) &&
2708 !test_range_bit(tree, block_start, cur_end,
2709 EXTENT_UPTODATE, 1)) {
2710 u64 sector;
2711 u64 extent_offset = block_start - em->start;
2712 size_t iosize;
2713 sector = (em->block_start + extent_offset) >> 9;
2714 iosize = (cur_end - block_start + blocksize) &
2715 ~((u64)blocksize - 1);
2716 /*
2717 * we've already got the extent locked, but we
2718 * need to split the state such that our end_bio
2719 * handler can clear the lock.
2720 */
2721 set_extent_bit(tree, block_start,
2722 block_start + iosize - 1,
2723 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2724 ret = submit_extent_page(READ, tree, page,
2725 sector, iosize, page_offset, em->bdev,
2726 NULL, 1,
2727 end_bio_extent_preparewrite, 0,
2728 0, 0);
2729 iocount++;
2730 block_start = block_start + iosize;
2731 } else {
2732 set_extent_uptodate(tree, block_start, cur_end,
2733 GFP_NOFS);
2734 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2735 block_start = cur_end + 1;
2736 }
2737 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2738 free_extent_map(em);
2739 }
2740 if (iocount) {
2741 wait_extent_bit(tree, orig_block_start,
2742 block_end, EXTENT_LOCKED);
2743 }
2744 check_page_uptodate(tree, page);
2745 err:
2746 /* FIXME, zero out newly allocated blocks on error */
2747 return err;
2748 }
2749
2750 /*
2751 * a helper for releasepage, this tests for areas of the page that
2752 * are locked or under IO and drops the related state bits if it is safe
2753 * to drop the page.
2754 */
2755 int try_release_extent_state(struct extent_map_tree *map,
2756 struct extent_io_tree *tree, struct page *page,
2757 gfp_t mask)
2758 {
2759 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2760 u64 end = start + PAGE_CACHE_SIZE - 1;
2761 int ret = 1;
2762
2763 if (test_range_bit(tree, start, end,
2764 EXTENT_IOBITS | EXTENT_ORDERED, 0))
2765 ret = 0;
2766 else {
2767 if ((mask & GFP_NOFS) == GFP_NOFS)
2768 mask = GFP_NOFS;
2769 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2770 1, 1, mask);
2771 }
2772 return ret;
2773 }
2774
2775 /*
2776 * a helper for releasepage. As long as there are no locked extents
2777 * in the range corresponding to the page, both state records and extent
2778 * map records are removed
2779 */
2780 int try_release_extent_mapping(struct extent_map_tree *map,
2781 struct extent_io_tree *tree, struct page *page,
2782 gfp_t mask)
2783 {
2784 struct extent_map *em;
2785 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2786 u64 end = start + PAGE_CACHE_SIZE - 1;
2787
2788 if ((mask & __GFP_WAIT) &&
2789 page->mapping->host->i_size > 16 * 1024 * 1024) {
2790 u64 len;
2791 while (start <= end) {
2792 len = end - start + 1;
2793 spin_lock(&map->lock);
2794 em = lookup_extent_mapping(map, start, len);
2795 if (!em || IS_ERR(em)) {
2796 spin_unlock(&map->lock);
2797 break;
2798 }
2799 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2800 em->start != start) {
2801 spin_unlock(&map->lock);
2802 free_extent_map(em);
2803 break;
2804 }
2805 if (!test_range_bit(tree, em->start,
2806 extent_map_end(em) - 1,
2807 EXTENT_LOCKED | EXTENT_WRITEBACK |
2808 EXTENT_ORDERED,
2809 0)) {
2810 remove_extent_mapping(map, em);
2811 /* once for the rb tree */
2812 free_extent_map(em);
2813 }
2814 start = extent_map_end(em);
2815 spin_unlock(&map->lock);
2816
2817 /* once for us */
2818 free_extent_map(em);
2819 }
2820 }
2821 return try_release_extent_state(map, tree, page, mask);
2822 }
2823
2824 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2825 get_extent_t *get_extent)
2826 {
2827 struct inode *inode = mapping->host;
2828 u64 start = iblock << inode->i_blkbits;
2829 sector_t sector = 0;
2830 size_t blksize = (1 << inode->i_blkbits);
2831 struct extent_map *em;
2832
2833 lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2834 GFP_NOFS);
2835 em = get_extent(inode, NULL, 0, start, blksize, 0);
2836 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1,
2837 GFP_NOFS);
2838 if (!em || IS_ERR(em))
2839 return 0;
2840
2841 if (em->block_start > EXTENT_MAP_LAST_BYTE)
2842 goto out;
2843
2844 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2845 out:
2846 free_extent_map(em);
2847 return sector;
2848 }
2849
2850 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2851 __u64 start, __u64 len, get_extent_t *get_extent)
2852 {
2853 int ret;
2854 u64 off = start;
2855 u64 max = start + len;
2856 u32 flags = 0;
2857 u64 disko = 0;
2858 struct extent_map *em = NULL;
2859 int end = 0;
2860 u64 em_start = 0, em_len = 0;
2861 unsigned long emflags;
2862 ret = 0;
2863
2864 if (len == 0)
2865 return -EINVAL;
2866
2867 lock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2868 GFP_NOFS);
2869 em = get_extent(inode, NULL, 0, off, max - off, 0);
2870 if (!em)
2871 goto out;
2872 if (IS_ERR(em)) {
2873 ret = PTR_ERR(em);
2874 goto out;
2875 }
2876 while (!end) {
2877 off = em->start + em->len;
2878 if (off >= max)
2879 end = 1;
2880
2881 em_start = em->start;
2882 em_len = em->len;
2883
2884 disko = 0;
2885 flags = 0;
2886
2887 switch (em->block_start) {
2888 case EXTENT_MAP_LAST_BYTE:
2889 end = 1;
2890 flags |= FIEMAP_EXTENT_LAST;
2891 break;
2892 case EXTENT_MAP_HOLE:
2893 flags |= FIEMAP_EXTENT_UNWRITTEN;
2894 break;
2895 case EXTENT_MAP_INLINE:
2896 flags |= (FIEMAP_EXTENT_DATA_INLINE |
2897 FIEMAP_EXTENT_NOT_ALIGNED);
2898 break;
2899 case EXTENT_MAP_DELALLOC:
2900 flags |= (FIEMAP_EXTENT_DELALLOC |
2901 FIEMAP_EXTENT_UNKNOWN);
2902 break;
2903 default:
2904 disko = em->block_start;
2905 break;
2906 }
2907 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
2908 flags |= FIEMAP_EXTENT_ENCODED;
2909
2910 emflags = em->flags;
2911 free_extent_map(em);
2912 em = NULL;
2913
2914 if (!end) {
2915 em = get_extent(inode, NULL, 0, off, max - off, 0);
2916 if (!em)
2917 goto out;
2918 if (IS_ERR(em)) {
2919 ret = PTR_ERR(em);
2920 goto out;
2921 }
2922 emflags = em->flags;
2923 }
2924 if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) {
2925 flags |= FIEMAP_EXTENT_LAST;
2926 end = 1;
2927 }
2928
2929 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
2930 em_len, flags);
2931 if (ret)
2932 goto out_free;
2933 }
2934 out_free:
2935 free_extent_map(em);
2936 out:
2937 unlock_extent(&BTRFS_I(inode)->io_tree, start, start + len,
2938 GFP_NOFS);
2939 return ret;
2940 }
2941
2942 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2943 unsigned long i)
2944 {
2945 struct page *p;
2946 struct address_space *mapping;
2947
2948 if (i == 0)
2949 return eb->first_page;
2950 i += eb->start >> PAGE_CACHE_SHIFT;
2951 mapping = eb->first_page->mapping;
2952 if (!mapping)
2953 return NULL;
2954
2955 /*
2956 * extent_buffer_page is only called after pinning the page
2957 * by increasing the reference count. So we know the page must
2958 * be in the radix tree.
2959 */
2960 rcu_read_lock();
2961 p = radix_tree_lookup(&mapping->page_tree, i);
2962 rcu_read_unlock();
2963
2964 return p;
2965 }
2966
2967 static inline unsigned long num_extent_pages(u64 start, u64 len)
2968 {
2969 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2970 (start >> PAGE_CACHE_SHIFT);
2971 }
2972
2973 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2974 u64 start,
2975 unsigned long len,
2976 gfp_t mask)
2977 {
2978 struct extent_buffer *eb = NULL;
2979 #if LEAK_DEBUG
2980 unsigned long flags;
2981 #endif
2982
2983 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2984 eb->start = start;
2985 eb->len = len;
2986 spin_lock_init(&eb->lock);
2987 init_waitqueue_head(&eb->lock_wq);
2988
2989 #if LEAK_DEBUG
2990 spin_lock_irqsave(&leak_lock, flags);
2991 list_add(&eb->leak_list, &buffers);
2992 spin_unlock_irqrestore(&leak_lock, flags);
2993 #endif
2994 atomic_set(&eb->refs, 1);
2995
2996 return eb;
2997 }
2998
2999 static void __free_extent_buffer(struct extent_buffer *eb)
3000 {
3001 #if LEAK_DEBUG
3002 unsigned long flags;
3003 spin_lock_irqsave(&leak_lock, flags);
3004 list_del(&eb->leak_list);
3005 spin_unlock_irqrestore(&leak_lock, flags);
3006 #endif
3007 kmem_cache_free(extent_buffer_cache, eb);
3008 }
3009
3010 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3011 u64 start, unsigned long len,
3012 struct page *page0,
3013 gfp_t mask)
3014 {
3015 unsigned long num_pages = num_extent_pages(start, len);
3016 unsigned long i;
3017 unsigned long index = start >> PAGE_CACHE_SHIFT;
3018 struct extent_buffer *eb;
3019 struct extent_buffer *exists = NULL;
3020 struct page *p;
3021 struct address_space *mapping = tree->mapping;
3022 int uptodate = 1;
3023
3024 spin_lock(&tree->buffer_lock);
3025 eb = buffer_search(tree, start);
3026 if (eb) {
3027 atomic_inc(&eb->refs);
3028 spin_unlock(&tree->buffer_lock);
3029 mark_page_accessed(eb->first_page);
3030 return eb;
3031 }
3032 spin_unlock(&tree->buffer_lock);
3033
3034 eb = __alloc_extent_buffer(tree, start, len, mask);
3035 if (!eb)
3036 return NULL;
3037
3038 if (page0) {
3039 eb->first_page = page0;
3040 i = 1;
3041 index++;
3042 page_cache_get(page0);
3043 mark_page_accessed(page0);
3044 set_page_extent_mapped(page0);
3045 set_page_extent_head(page0, len);
3046 uptodate = PageUptodate(page0);
3047 } else {
3048 i = 0;
3049 }
3050 for (; i < num_pages; i++, index++) {
3051 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
3052 if (!p) {
3053 WARN_ON(1);
3054 goto free_eb;
3055 }
3056 set_page_extent_mapped(p);
3057 mark_page_accessed(p);
3058 if (i == 0) {
3059 eb->first_page = p;
3060 set_page_extent_head(p, len);
3061 } else {
3062 set_page_private(p, EXTENT_PAGE_PRIVATE);
3063 }
3064 if (!PageUptodate(p))
3065 uptodate = 0;
3066 unlock_page(p);
3067 }
3068 if (uptodate)
3069 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3070
3071 spin_lock(&tree->buffer_lock);
3072 exists = buffer_tree_insert(tree, start, &eb->rb_node);
3073 if (exists) {
3074 /* add one reference for the caller */
3075 atomic_inc(&exists->refs);
3076 spin_unlock(&tree->buffer_lock);
3077 goto free_eb;
3078 }
3079 spin_unlock(&tree->buffer_lock);
3080
3081 /* add one reference for the tree */
3082 atomic_inc(&eb->refs);
3083 return eb;
3084
3085 free_eb:
3086 if (!atomic_dec_and_test(&eb->refs))
3087 return exists;
3088 for (index = 1; index < i; index++)
3089 page_cache_release(extent_buffer_page(eb, index));
3090 page_cache_release(extent_buffer_page(eb, 0));
3091 __free_extent_buffer(eb);
3092 return exists;
3093 }
3094
3095 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3096 u64 start, unsigned long len,
3097 gfp_t mask)
3098 {
3099 struct extent_buffer *eb;
3100
3101 spin_lock(&tree->buffer_lock);
3102 eb = buffer_search(tree, start);
3103 if (eb)
3104 atomic_inc(&eb->refs);
3105 spin_unlock(&tree->buffer_lock);
3106
3107 if (eb)
3108 mark_page_accessed(eb->first_page);
3109
3110 return eb;
3111 }
3112
3113 void free_extent_buffer(struct extent_buffer *eb)
3114 {
3115 if (!eb)
3116 return;
3117
3118 if (!atomic_dec_and_test(&eb->refs))
3119 return;
3120
3121 WARN_ON(1);
3122 }
3123
3124 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3125 struct extent_buffer *eb)
3126 {
3127 int set;
3128 unsigned long i;
3129 unsigned long num_pages;
3130 struct page *page;
3131
3132 u64 start = eb->start;
3133 u64 end = start + eb->len - 1;
3134
3135 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
3136 num_pages = num_extent_pages(eb->start, eb->len);
3137
3138 for (i = 0; i < num_pages; i++) {
3139 page = extent_buffer_page(eb, i);
3140 if (!set && !PageDirty(page))
3141 continue;
3142
3143 lock_page(page);
3144 if (i == 0)
3145 set_page_extent_head(page, eb->len);
3146 else
3147 set_page_private(page, EXTENT_PAGE_PRIVATE);
3148
3149 /*
3150 * if we're on the last page or the first page and the
3151 * block isn't aligned on a page boundary, do extra checks
3152 * to make sure we don't clean page that is partially dirty
3153 */
3154 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3155 ((i == num_pages - 1) &&
3156 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3157 start = (u64)page->index << PAGE_CACHE_SHIFT;
3158 end = start + PAGE_CACHE_SIZE - 1;
3159 if (test_range_bit(tree, start, end,
3160 EXTENT_DIRTY, 0)) {
3161 unlock_page(page);
3162 continue;
3163 }
3164 }
3165 clear_page_dirty_for_io(page);
3166 spin_lock_irq(&page->mapping->tree_lock);
3167 if (!PageDirty(page)) {
3168 radix_tree_tag_clear(&page->mapping->page_tree,
3169 page_index(page),
3170 PAGECACHE_TAG_DIRTY);
3171 }
3172 spin_unlock_irq(&page->mapping->tree_lock);
3173 unlock_page(page);
3174 }
3175 return 0;
3176 }
3177
3178 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
3179 struct extent_buffer *eb)
3180 {
3181 return wait_on_extent_writeback(tree, eb->start,
3182 eb->start + eb->len - 1);
3183 }
3184
3185 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3186 struct extent_buffer *eb)
3187 {
3188 unsigned long i;
3189 unsigned long num_pages;
3190
3191 num_pages = num_extent_pages(eb->start, eb->len);
3192 for (i = 0; i < num_pages; i++) {
3193 struct page *page = extent_buffer_page(eb, i);
3194 /* writepage may need to do something special for the
3195 * first page, we have to make sure page->private is
3196 * properly set. releasepage may drop page->private
3197 * on us if the page isn't already dirty.
3198 */
3199 lock_page(page);
3200 if (i == 0) {
3201 set_page_extent_head(page, eb->len);
3202 } else if (PagePrivate(page) &&
3203 page->private != EXTENT_PAGE_PRIVATE) {
3204 set_page_extent_mapped(page);
3205 }
3206 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3207 set_extent_dirty(tree, page_offset(page),
3208 page_offset(page) + PAGE_CACHE_SIZE - 1,
3209 GFP_NOFS);
3210 unlock_page(page);
3211 }
3212 return 0;
3213 }
3214
3215 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3216 struct extent_buffer *eb)
3217 {
3218 unsigned long i;
3219 struct page *page;
3220 unsigned long num_pages;
3221
3222 num_pages = num_extent_pages(eb->start, eb->len);
3223 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3224
3225 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3226 GFP_NOFS);
3227 for (i = 0; i < num_pages; i++) {
3228 page = extent_buffer_page(eb, i);
3229 if (page)
3230 ClearPageUptodate(page);
3231 }
3232 return 0;
3233 }
3234
3235 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3236 struct extent_buffer *eb)
3237 {
3238 unsigned long i;
3239 struct page *page;
3240 unsigned long num_pages;
3241
3242 num_pages = num_extent_pages(eb->start, eb->len);
3243
3244 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3245 GFP_NOFS);
3246 for (i = 0; i < num_pages; i++) {
3247 page = extent_buffer_page(eb, i);
3248 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3249 ((i == num_pages - 1) &&
3250 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3251 check_page_uptodate(tree, page);
3252 continue;
3253 }
3254 SetPageUptodate(page);
3255 }
3256 return 0;
3257 }
3258
3259 int extent_range_uptodate(struct extent_io_tree *tree,
3260 u64 start, u64 end)
3261 {
3262 struct page *page;
3263 int ret;
3264 int pg_uptodate = 1;
3265 int uptodate;
3266 unsigned long index;
3267
3268 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3269 if (ret)
3270 return 1;
3271 while (start <= end) {
3272 index = start >> PAGE_CACHE_SHIFT;
3273 page = find_get_page(tree->mapping, index);
3274 uptodate = PageUptodate(page);
3275 page_cache_release(page);
3276 if (!uptodate) {
3277 pg_uptodate = 0;
3278 break;
3279 }
3280 start += PAGE_CACHE_SIZE;
3281 }
3282 return pg_uptodate;
3283 }
3284
3285 int extent_buffer_uptodate(struct extent_io_tree *tree,
3286 struct extent_buffer *eb)
3287 {
3288 int ret = 0;
3289 unsigned long num_pages;
3290 unsigned long i;
3291 struct page *page;
3292 int pg_uptodate = 1;
3293
3294 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3295 return 1;
3296
3297 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3298 EXTENT_UPTODATE, 1);
3299 if (ret)
3300 return ret;
3301
3302 num_pages = num_extent_pages(eb->start, eb->len);
3303 for (i = 0; i < num_pages; i++) {
3304 page = extent_buffer_page(eb, i);
3305 if (!PageUptodate(page)) {
3306 pg_uptodate = 0;
3307 break;
3308 }
3309 }
3310 return pg_uptodate;
3311 }
3312
3313 int read_extent_buffer_pages(struct extent_io_tree *tree,
3314 struct extent_buffer *eb,
3315 u64 start, int wait,
3316 get_extent_t *get_extent, int mirror_num)
3317 {
3318 unsigned long i;
3319 unsigned long start_i;
3320 struct page *page;
3321 int err;
3322 int ret = 0;
3323 int locked_pages = 0;
3324 int all_uptodate = 1;
3325 int inc_all_pages = 0;
3326 unsigned long num_pages;
3327 struct bio *bio = NULL;
3328 unsigned long bio_flags = 0;
3329
3330 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3331 return 0;
3332
3333 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3334 EXTENT_UPTODATE, 1)) {
3335 return 0;
3336 }
3337
3338 if (start) {
3339 WARN_ON(start < eb->start);
3340 start_i = (start >> PAGE_CACHE_SHIFT) -
3341 (eb->start >> PAGE_CACHE_SHIFT);
3342 } else {
3343 start_i = 0;
3344 }
3345
3346 num_pages = num_extent_pages(eb->start, eb->len);
3347 for (i = start_i; i < num_pages; i++) {
3348 page = extent_buffer_page(eb, i);
3349 if (!wait) {
3350 if (!trylock_page(page))
3351 goto unlock_exit;
3352 } else {
3353 lock_page(page);
3354 }
3355 locked_pages++;
3356 if (!PageUptodate(page))
3357 all_uptodate = 0;
3358 }
3359 if (all_uptodate) {
3360 if (start_i == 0)
3361 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3362 goto unlock_exit;
3363 }
3364
3365 for (i = start_i; i < num_pages; i++) {
3366 page = extent_buffer_page(eb, i);
3367 if (inc_all_pages)
3368 page_cache_get(page);
3369 if (!PageUptodate(page)) {
3370 if (start_i == 0)
3371 inc_all_pages = 1;
3372 ClearPageError(page);
3373 err = __extent_read_full_page(tree, page,
3374 get_extent, &bio,
3375 mirror_num, &bio_flags);
3376 if (err)
3377 ret = err;
3378 } else {
3379 unlock_page(page);
3380 }
3381 }
3382
3383 if (bio)
3384 submit_one_bio(READ, bio, mirror_num, bio_flags);
3385
3386 if (ret || !wait)
3387 return ret;
3388
3389 for (i = start_i; i < num_pages; i++) {
3390 page = extent_buffer_page(eb, i);
3391 wait_on_page_locked(page);
3392 if (!PageUptodate(page))
3393 ret = -EIO;
3394 }
3395
3396 if (!ret)
3397 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3398 return ret;
3399
3400 unlock_exit:
3401 i = start_i;
3402 while (locked_pages > 0) {
3403 page = extent_buffer_page(eb, i);
3404 i++;
3405 unlock_page(page);
3406 locked_pages--;
3407 }
3408 return ret;
3409 }
3410
3411 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3412 unsigned long start,
3413 unsigned long len)
3414 {
3415 size_t cur;
3416 size_t offset;
3417 struct page *page;
3418 char *kaddr;
3419 char *dst = (char *)dstv;
3420 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3421 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3422
3423 WARN_ON(start > eb->len);
3424 WARN_ON(start + len > eb->start + eb->len);
3425
3426 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3427
3428 while (len > 0) {
3429 page = extent_buffer_page(eb, i);
3430
3431 cur = min(len, (PAGE_CACHE_SIZE - offset));
3432 kaddr = kmap_atomic(page, KM_USER1);
3433 memcpy(dst, kaddr + offset, cur);
3434 kunmap_atomic(kaddr, KM_USER1);
3435
3436 dst += cur;
3437 len -= cur;
3438 offset = 0;
3439 i++;
3440 }
3441 }
3442
3443 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3444 unsigned long min_len, char **token, char **map,
3445 unsigned long *map_start,
3446 unsigned long *map_len, int km)
3447 {
3448 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3449 char *kaddr;
3450 struct page *p;
3451 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3452 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3453 unsigned long end_i = (start_offset + start + min_len - 1) >>
3454 PAGE_CACHE_SHIFT;
3455
3456 if (i != end_i)
3457 return -EINVAL;
3458
3459 if (i == 0) {
3460 offset = start_offset;
3461 *map_start = 0;
3462 } else {
3463 offset = 0;
3464 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3465 }
3466
3467 if (start + min_len > eb->len) {
3468 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3469 "wanted %lu %lu\n", (unsigned long long)eb->start,
3470 eb->len, start, min_len);
3471 WARN_ON(1);
3472 }
3473
3474 p = extent_buffer_page(eb, i);
3475 kaddr = kmap_atomic(p, km);
3476 *token = kaddr;
3477 *map = kaddr + offset;
3478 *map_len = PAGE_CACHE_SIZE - offset;
3479 return 0;
3480 }
3481
3482 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3483 unsigned long min_len,
3484 char **token, char **map,
3485 unsigned long *map_start,
3486 unsigned long *map_len, int km)
3487 {
3488 int err;
3489 int save = 0;
3490 if (eb->map_token) {
3491 unmap_extent_buffer(eb, eb->map_token, km);
3492 eb->map_token = NULL;
3493 save = 1;
3494 }
3495 err = map_private_extent_buffer(eb, start, min_len, token, map,
3496 map_start, map_len, km);
3497 if (!err && save) {
3498 eb->map_token = *token;
3499 eb->kaddr = *map;
3500 eb->map_start = *map_start;
3501 eb->map_len = *map_len;
3502 }
3503 return err;
3504 }
3505
3506 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3507 {
3508 kunmap_atomic(token, km);
3509 }
3510
3511 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3512 unsigned long start,
3513 unsigned long len)
3514 {
3515 size_t cur;
3516 size_t offset;
3517 struct page *page;
3518 char *kaddr;
3519 char *ptr = (char *)ptrv;
3520 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3521 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3522 int ret = 0;
3523
3524 WARN_ON(start > eb->len);
3525 WARN_ON(start + len > eb->start + eb->len);
3526
3527 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3528
3529 while (len > 0) {
3530 page = extent_buffer_page(eb, i);
3531
3532 cur = min(len, (PAGE_CACHE_SIZE - offset));
3533
3534 kaddr = kmap_atomic(page, KM_USER0);
3535 ret = memcmp(ptr, kaddr + offset, cur);
3536 kunmap_atomic(kaddr, KM_USER0);
3537 if (ret)
3538 break;
3539
3540 ptr += cur;
3541 len -= cur;
3542 offset = 0;
3543 i++;
3544 }
3545 return ret;
3546 }
3547
3548 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3549 unsigned long start, unsigned long len)
3550 {
3551 size_t cur;
3552 size_t offset;
3553 struct page *page;
3554 char *kaddr;
3555 char *src = (char *)srcv;
3556 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3557 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3558
3559 WARN_ON(start > eb->len);
3560 WARN_ON(start + len > eb->start + eb->len);
3561
3562 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3563
3564 while (len > 0) {
3565 page = extent_buffer_page(eb, i);
3566 WARN_ON(!PageUptodate(page));
3567
3568 cur = min(len, PAGE_CACHE_SIZE - offset);
3569 kaddr = kmap_atomic(page, KM_USER1);
3570 memcpy(kaddr + offset, src, cur);
3571 kunmap_atomic(kaddr, KM_USER1);
3572
3573 src += cur;
3574 len -= cur;
3575 offset = 0;
3576 i++;
3577 }
3578 }
3579
3580 void memset_extent_buffer(struct extent_buffer *eb, char c,
3581 unsigned long start, unsigned long len)
3582 {
3583 size_t cur;
3584 size_t offset;
3585 struct page *page;
3586 char *kaddr;
3587 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3588 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3589
3590 WARN_ON(start > eb->len);
3591 WARN_ON(start + len > eb->start + eb->len);
3592
3593 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3594
3595 while (len > 0) {
3596 page = extent_buffer_page(eb, i);
3597 WARN_ON(!PageUptodate(page));
3598
3599 cur = min(len, PAGE_CACHE_SIZE - offset);
3600 kaddr = kmap_atomic(page, KM_USER0);
3601 memset(kaddr + offset, c, cur);
3602 kunmap_atomic(kaddr, KM_USER0);
3603
3604 len -= cur;
3605 offset = 0;
3606 i++;
3607 }
3608 }
3609
3610 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3611 unsigned long dst_offset, unsigned long src_offset,
3612 unsigned long len)
3613 {
3614 u64 dst_len = dst->len;
3615 size_t cur;
3616 size_t offset;
3617 struct page *page;
3618 char *kaddr;
3619 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3620 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3621
3622 WARN_ON(src->len != dst_len);
3623
3624 offset = (start_offset + dst_offset) &
3625 ((unsigned long)PAGE_CACHE_SIZE - 1);
3626
3627 while (len > 0) {
3628 page = extent_buffer_page(dst, i);
3629 WARN_ON(!PageUptodate(page));
3630
3631 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3632
3633 kaddr = kmap_atomic(page, KM_USER0);
3634 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3635 kunmap_atomic(kaddr, KM_USER0);
3636
3637 src_offset += cur;
3638 len -= cur;
3639 offset = 0;
3640 i++;
3641 }
3642 }
3643
3644 static void move_pages(struct page *dst_page, struct page *src_page,
3645 unsigned long dst_off, unsigned long src_off,
3646 unsigned long len)
3647 {
3648 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3649 if (dst_page == src_page) {
3650 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3651 } else {
3652 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3653 char *p = dst_kaddr + dst_off + len;
3654 char *s = src_kaddr + src_off + len;
3655
3656 while (len--)
3657 *--p = *--s;
3658
3659 kunmap_atomic(src_kaddr, KM_USER1);
3660 }
3661 kunmap_atomic(dst_kaddr, KM_USER0);
3662 }
3663
3664 static void copy_pages(struct page *dst_page, struct page *src_page,
3665 unsigned long dst_off, unsigned long src_off,
3666 unsigned long len)
3667 {
3668 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3669 char *src_kaddr;
3670
3671 if (dst_page != src_page)
3672 src_kaddr = kmap_atomic(src_page, KM_USER1);
3673 else
3674 src_kaddr = dst_kaddr;
3675
3676 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3677 kunmap_atomic(dst_kaddr, KM_USER0);
3678 if (dst_page != src_page)
3679 kunmap_atomic(src_kaddr, KM_USER1);
3680 }
3681
3682 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3683 unsigned long src_offset, unsigned long len)
3684 {
3685 size_t cur;
3686 size_t dst_off_in_page;
3687 size_t src_off_in_page;
3688 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3689 unsigned long dst_i;
3690 unsigned long src_i;
3691
3692 if (src_offset + len > dst->len) {
3693 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3694 "len %lu dst len %lu\n", src_offset, len, dst->len);
3695 BUG_ON(1);
3696 }
3697 if (dst_offset + len > dst->len) {
3698 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3699 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3700 BUG_ON(1);
3701 }
3702
3703 while (len > 0) {
3704 dst_off_in_page = (start_offset + dst_offset) &
3705 ((unsigned long)PAGE_CACHE_SIZE - 1);
3706 src_off_in_page = (start_offset + src_offset) &
3707 ((unsigned long)PAGE_CACHE_SIZE - 1);
3708
3709 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3710 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3711
3712 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3713 src_off_in_page));
3714 cur = min_t(unsigned long, cur,
3715 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3716
3717 copy_pages(extent_buffer_page(dst, dst_i),
3718 extent_buffer_page(dst, src_i),
3719 dst_off_in_page, src_off_in_page, cur);
3720
3721 src_offset += cur;
3722 dst_offset += cur;
3723 len -= cur;
3724 }
3725 }
3726
3727 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3728 unsigned long src_offset, unsigned long len)
3729 {
3730 size_t cur;
3731 size_t dst_off_in_page;
3732 size_t src_off_in_page;
3733 unsigned long dst_end = dst_offset + len - 1;
3734 unsigned long src_end = src_offset + len - 1;
3735 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3736 unsigned long dst_i;
3737 unsigned long src_i;
3738
3739 if (src_offset + len > dst->len) {
3740 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3741 "len %lu len %lu\n", src_offset, len, dst->len);
3742 BUG_ON(1);
3743 }
3744 if (dst_offset + len > dst->len) {
3745 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3746 "len %lu len %lu\n", dst_offset, len, dst->len);
3747 BUG_ON(1);
3748 }
3749 if (dst_offset < src_offset) {
3750 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3751 return;
3752 }
3753 while (len > 0) {
3754 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3755 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3756
3757 dst_off_in_page = (start_offset + dst_end) &
3758 ((unsigned long)PAGE_CACHE_SIZE - 1);
3759 src_off_in_page = (start_offset + src_end) &
3760 ((unsigned long)PAGE_CACHE_SIZE - 1);
3761
3762 cur = min_t(unsigned long, len, src_off_in_page + 1);
3763 cur = min(cur, dst_off_in_page + 1);
3764 move_pages(extent_buffer_page(dst, dst_i),
3765 extent_buffer_page(dst, src_i),
3766 dst_off_in_page - cur + 1,
3767 src_off_in_page - cur + 1, cur);
3768
3769 dst_end -= cur;
3770 src_end -= cur;
3771 len -= cur;
3772 }
3773 }
3774
3775 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3776 {
3777 u64 start = page_offset(page);
3778 struct extent_buffer *eb;
3779 int ret = 1;
3780 unsigned long i;
3781 unsigned long num_pages;
3782
3783 spin_lock(&tree->buffer_lock);
3784 eb = buffer_search(tree, start);
3785 if (!eb)
3786 goto out;
3787
3788 if (atomic_read(&eb->refs) > 1) {
3789 ret = 0;
3790 goto out;
3791 }
3792 /* at this point we can safely release the extent buffer */
3793 num_pages = num_extent_pages(eb->start, eb->len);
3794 for (i = 0; i < num_pages; i++)
3795 page_cache_release(extent_buffer_page(eb, i));
3796 rb_erase(&eb->rb_node, &tree->buffer);
3797 __free_extent_buffer(eb);
3798 out:
3799 spin_unlock(&tree->buffer_lock);
3800 return ret;
3801 }
Linux with added FPC-III support