?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
uli526x: fix regions leak in driver probe error path.
[linux-2.6.git] / fs / btrfs / extent_io.c
blob2862454bcdb32fb78b435b1806d5db297a65f774
1 #include <linux/bitops.h>
2 #include <linux/slab.h>
3 #include <linux/bio.h>
4 #include <linux/mm.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
17 #include "compat.h"
18 #include "ctree.h"
19 #include "btrfs_inode.h"
20 #include "volumes.h"
21 #include "check-integrity.h"
22
23 static struct kmem_cache *extent_state_cache;
24 static struct kmem_cache *extent_buffer_cache;
25
26 static LIST_HEAD(buffers);
27 static LIST_HEAD(states);
28
29 #define LEAK_DEBUG 0
30 #if LEAK_DEBUG
31 static DEFINE_SPINLOCK(leak_lock);
32 #endif
33
34 #define BUFFER_LRU_MAX 64
35
36 struct tree_entry {
37 u64 start;
38 u64 end;
39 struct rb_node rb_node;
40 };
41
42 struct extent_page_data {
43 struct bio *bio;
44 struct extent_io_tree *tree;
45 get_extent_t *get_extent;
46
47 /* tells writepage not to lock the state bits for this range
48 * it still does the unlocking
49 */
50 unsigned int extent_locked:1;
51
52 /* tells the submit_bio code to use a WRITE_SYNC */
53 unsigned int sync_io:1;
54 };
55
56 int __init extent_io_init(void)
57 {
58 extent_state_cache = kmem_cache_create("extent_state",
59 sizeof(struct extent_state), 0,
60 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
61 if (!extent_state_cache)
62 return -ENOMEM;
63
64 extent_buffer_cache = kmem_cache_create("extent_buffers",
65 sizeof(struct extent_buffer), 0,
66 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, 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)
109 {
110 tree->state = RB_ROOT;
111 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
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 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 void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
243 struct extent_state *other)
244 {
245 if (tree->ops && tree->ops->merge_extent_hook)
246 tree->ops->merge_extent_hook(tree->mapping->host, new,
247 other);
248 }
249
250 /*
251 * utility function to look for merge candidates inside a given range.
252 * Any extents with matching state are merged together into a single
253 * extent in the tree. Extents with EXTENT_IO in their state field
254 * are not merged because the end_io handlers need to be able to do
255 * operations on them without sleeping (or doing allocations/splits).
256 *
257 * This should be called with the tree lock held.
258 */
259 static void merge_state(struct extent_io_tree *tree,
260 struct extent_state *state)
261 {
262 struct extent_state *other;
263 struct rb_node *other_node;
264
265 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
266 return;
267
268 other_node = rb_prev(&state->rb_node);
269 if (other_node) {
270 other = rb_entry(other_node, struct extent_state, rb_node);
271 if (other->end == state->start - 1 &&
272 other->state == state->state) {
273 merge_cb(tree, state, other);
274 state->start = other->start;
275 other->tree = NULL;
276 rb_erase(&other->rb_node, &tree->state);
277 free_extent_state(other);
278 }
279 }
280 other_node = rb_next(&state->rb_node);
281 if (other_node) {
282 other = rb_entry(other_node, struct extent_state, rb_node);
283 if (other->start == state->end + 1 &&
284 other->state == state->state) {
285 merge_cb(tree, state, other);
286 state->end = other->end;
287 other->tree = NULL;
288 rb_erase(&other->rb_node, &tree->state);
289 free_extent_state(other);
290 }
291 }
292 }
293
294 static void set_state_cb(struct extent_io_tree *tree,
295 struct extent_state *state, int *bits)
296 {
297 if (tree->ops && tree->ops->set_bit_hook)
298 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
299 }
300
301 static void clear_state_cb(struct extent_io_tree *tree,
302 struct extent_state *state, int *bits)
303 {
304 if (tree->ops && tree->ops->clear_bit_hook)
305 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
306 }
307
308 static void set_state_bits(struct extent_io_tree *tree,
309 struct extent_state *state, int *bits);
310
311 /*
312 * insert an extent_state struct into the tree. 'bits' are set on the
313 * struct before it is inserted.
314 *
315 * This may return -EEXIST if the extent is already there, in which case the
316 * state struct is freed.
317 *
318 * The tree lock is not taken internally. This is a utility function and
319 * probably isn't what you want to call (see set/clear_extent_bit).
320 */
321 static int insert_state(struct extent_io_tree *tree,
322 struct extent_state *state, u64 start, u64 end,
323 int *bits)
324 {
325 struct rb_node *node;
326
327 if (end < start) {
328 printk(KERN_ERR "btrfs end < start %llu %llu\n",
329 (unsigned long long)end,
330 (unsigned long long)start);
331 WARN_ON(1);
332 }
333 state->start = start;
334 state->end = end;
335
336 set_state_bits(tree, state, bits);
337
338 node = tree_insert(&tree->state, end, &state->rb_node);
339 if (node) {
340 struct extent_state *found;
341 found = rb_entry(node, struct extent_state, rb_node);
342 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
343 "%llu %llu\n", (unsigned long long)found->start,
344 (unsigned long long)found->end,
345 (unsigned long long)start, (unsigned long long)end);
346 return -EEXIST;
347 }
348 state->tree = tree;
349 merge_state(tree, state);
350 return 0;
351 }
352
353 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
354 u64 split)
355 {
356 if (tree->ops && tree->ops->split_extent_hook)
357 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
358 }
359
360 /*
361 * split a given extent state struct in two, inserting the preallocated
362 * struct 'prealloc' as the newly created second half. 'split' indicates an
363 * offset inside 'orig' where it should be split.
364 *
365 * Before calling,
366 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
367 * are two extent state structs in the tree:
368 * prealloc: [orig->start, split - 1]
369 * orig: [ split, orig->end ]
370 *
371 * The tree locks are not taken by this function. They need to be held
372 * by the caller.
373 */
374 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
375 struct extent_state *prealloc, u64 split)
376 {
377 struct rb_node *node;
378
379 split_cb(tree, orig, split);
380
381 prealloc->start = orig->start;
382 prealloc->end = split - 1;
383 prealloc->state = orig->state;
384 orig->start = split;
385
386 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
387 if (node) {
388 free_extent_state(prealloc);
389 return -EEXIST;
390 }
391 prealloc->tree = tree;
392 return 0;
393 }
394
395 /*
396 * utility function to clear some bits in an extent state struct.
397 * it will optionally wake up any one waiting on this state (wake == 1), or
398 * forcibly remove the state from the tree (delete == 1).
399 *
400 * If no bits are set on the state struct after clearing things, the
401 * struct is freed and removed from the tree
402 */
403 static int clear_state_bit(struct extent_io_tree *tree,
404 struct extent_state *state,
405 int *bits, int wake)
406 {
407 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
408 int ret = state->state & bits_to_clear;
409
410 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
411 u64 range = state->end - state->start + 1;
412 WARN_ON(range > tree->dirty_bytes);
413 tree->dirty_bytes -= range;
414 }
415 clear_state_cb(tree, state, bits);
416 state->state &= ~bits_to_clear;
417 if (wake)
418 wake_up(&state->wq);
419 if (state->state == 0) {
420 if (state->tree) {
421 rb_erase(&state->rb_node, &tree->state);
422 state->tree = NULL;
423 free_extent_state(state);
424 } else {
425 WARN_ON(1);
426 }
427 } else {
428 merge_state(tree, state);
429 }
430 return ret;
431 }
432
433 static struct extent_state *
434 alloc_extent_state_atomic(struct extent_state *prealloc)
435 {
436 if (!prealloc)
437 prealloc = alloc_extent_state(GFP_ATOMIC);
438
439 return prealloc;
440 }
441
442 /*
443 * clear some bits on a range in the tree. This may require splitting
444 * or inserting elements in the tree, so the gfp mask is used to
445 * indicate which allocations or sleeping are allowed.
446 *
447 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
448 * the given range from the tree regardless of state (ie for truncate).
449 *
450 * the range [start, end] is inclusive.
451 *
452 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
453 * bits were already set, or zero if none of the bits were already set.
454 */
455 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
456 int bits, int wake, int delete,
457 struct extent_state **cached_state,
458 gfp_t mask)
459 {
460 struct extent_state *state;
461 struct extent_state *cached;
462 struct extent_state *prealloc = NULL;
463 struct rb_node *next_node;
464 struct rb_node *node;
465 u64 last_end;
466 int err;
467 int set = 0;
468 int clear = 0;
469
470 if (delete)
471 bits |= ~EXTENT_CTLBITS;
472 bits |= EXTENT_FIRST_DELALLOC;
473
474 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
475 clear = 1;
476 again:
477 if (!prealloc && (mask & __GFP_WAIT)) {
478 prealloc = alloc_extent_state(mask);
479 if (!prealloc)
480 return -ENOMEM;
481 }
482
483 spin_lock(&tree->lock);
484 if (cached_state) {
485 cached = *cached_state;
486
487 if (clear) {
488 *cached_state = NULL;
489 cached_state = NULL;
490 }
491
492 if (cached && cached->tree && cached->start <= start &&
493 cached->end > start) {
494 if (clear)
495 atomic_dec(&cached->refs);
496 state = cached;
497 goto hit_next;
498 }
499 if (clear)
500 free_extent_state(cached);
501 }
502 /*
503 * this search will find the extents that end after
504 * our range starts
505 */
506 node = tree_search(tree, start);
507 if (!node)
508 goto out;
509 state = rb_entry(node, struct extent_state, rb_node);
510 hit_next:
511 if (state->start > end)
512 goto out;
513 WARN_ON(state->end < start);
514 last_end = state->end;
515
516 if (state->end < end && !need_resched())
517 next_node = rb_next(&state->rb_node);
518 else
519 next_node = NULL;
520
521 /* the state doesn't have the wanted bits, go ahead */
522 if (!(state->state & bits))
523 goto next;
524
525 /*
526 * | ---- desired range ---- |
527 * | state | or
528 * | ------------- state -------------- |
529 *
530 * We need to split the extent we found, and may flip
531 * bits on second half.
532 *
533 * If the extent we found extends past our range, we
534 * just split and search again. It'll get split again
535 * the next time though.
536 *
537 * If the extent we found is inside our range, we clear
538 * the desired bit on it.
539 */
540
541 if (state->start < start) {
542 prealloc = alloc_extent_state_atomic(prealloc);
543 BUG_ON(!prealloc);
544 err = split_state(tree, state, prealloc, start);
545 BUG_ON(err == -EEXIST);
546 prealloc = NULL;
547 if (err)
548 goto out;
549 if (state->end <= end) {
550 set |= clear_state_bit(tree, state, &bits, wake);
551 if (last_end == (u64)-1)
552 goto out;
553 start = last_end + 1;
554 }
555 goto search_again;
556 }
557 /*
558 * | ---- desired range ---- |
559 * | state |
560 * We need to split the extent, and clear the bit
561 * on the first half
562 */
563 if (state->start <= end && state->end > end) {
564 prealloc = alloc_extent_state_atomic(prealloc);
565 BUG_ON(!prealloc);
566 err = split_state(tree, state, prealloc, end + 1);
567 BUG_ON(err == -EEXIST);
568 if (wake)
569 wake_up(&state->wq);
570
571 set |= clear_state_bit(tree, prealloc, &bits, wake);
572
573 prealloc = NULL;
574 goto out;
575 }
576
577 set |= clear_state_bit(tree, state, &bits, wake);
578 next:
579 if (last_end == (u64)-1)
580 goto out;
581 start = last_end + 1;
582 if (start <= end && next_node) {
583 state = rb_entry(next_node, struct extent_state,
584 rb_node);
585 goto hit_next;
586 }
587 goto search_again;
588
589 out:
590 spin_unlock(&tree->lock);
591 if (prealloc)
592 free_extent_state(prealloc);
593
594 return set;
595
596 search_again:
597 if (start > end)
598 goto out;
599 spin_unlock(&tree->lock);
600 if (mask & __GFP_WAIT)
601 cond_resched();
602 goto again;
603 }
604
605 static int wait_on_state(struct extent_io_tree *tree,
606 struct extent_state *state)
607 __releases(tree->lock)
608 __acquires(tree->lock)
609 {
610 DEFINE_WAIT(wait);
611 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
612 spin_unlock(&tree->lock);
613 schedule();
614 spin_lock(&tree->lock);
615 finish_wait(&state->wq, &wait);
616 return 0;
617 }
618
619 /*
620 * waits for one or more bits to clear on a range in the state tree.
621 * The range [start, end] is inclusive.
622 * The tree lock is taken by this function
623 */
624 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
625 {
626 struct extent_state *state;
627 struct rb_node *node;
628
629 spin_lock(&tree->lock);
630 again:
631 while (1) {
632 /*
633 * this search will find all the extents that end after
634 * our range starts
635 */
636 node = tree_search(tree, start);
637 if (!node)
638 break;
639
640 state = rb_entry(node, struct extent_state, rb_node);
641
642 if (state->start > end)
643 goto out;
644
645 if (state->state & bits) {
646 start = state->start;
647 atomic_inc(&state->refs);
648 wait_on_state(tree, state);
649 free_extent_state(state);
650 goto again;
651 }
652 start = state->end + 1;
653
654 if (start > end)
655 break;
656
657 cond_resched_lock(&tree->lock);
658 }
659 out:
660 spin_unlock(&tree->lock);
661 return 0;
662 }
663
664 static void set_state_bits(struct extent_io_tree *tree,
665 struct extent_state *state,
666 int *bits)
667 {
668 int bits_to_set = *bits & ~EXTENT_CTLBITS;
669
670 set_state_cb(tree, state, bits);
671 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
672 u64 range = state->end - state->start + 1;
673 tree->dirty_bytes += range;
674 }
675 state->state |= bits_to_set;
676 }
677
678 static void cache_state(struct extent_state *state,
679 struct extent_state **cached_ptr)
680 {
681 if (cached_ptr && !(*cached_ptr)) {
682 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
683 *cached_ptr = state;
684 atomic_inc(&state->refs);
685 }
686 }
687 }
688
689 static void uncache_state(struct extent_state **cached_ptr)
690 {
691 if (cached_ptr && (*cached_ptr)) {
692 struct extent_state *state = *cached_ptr;
693 *cached_ptr = NULL;
694 free_extent_state(state);
695 }
696 }
697
698 /*
699 * set some bits on a range in the tree. This may require allocations or
700 * sleeping, so the gfp mask is used to indicate what is allowed.
701 *
702 * If any of the exclusive bits are set, this will fail with -EEXIST if some
703 * part of the range already has the desired bits set. The start of the
704 * existing range is returned in failed_start in this case.
705 *
706 * [start, end] is inclusive This takes the tree lock.
707 */
708
709 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
710 int bits, int exclusive_bits, u64 *failed_start,
711 struct extent_state **cached_state, gfp_t mask)
712 {
713 struct extent_state *state;
714 struct extent_state *prealloc = NULL;
715 struct rb_node *node;
716 int err = 0;
717 u64 last_start;
718 u64 last_end;
719
720 bits |= EXTENT_FIRST_DELALLOC;
721 again:
722 if (!prealloc && (mask & __GFP_WAIT)) {
723 prealloc = alloc_extent_state(mask);
724 BUG_ON(!prealloc);
725 }
726
727 spin_lock(&tree->lock);
728 if (cached_state && *cached_state) {
729 state = *cached_state;
730 if (state->start <= start && state->end > start &&
731 state->tree) {
732 node = &state->rb_node;
733 goto hit_next;
734 }
735 }
736 /*
737 * this search will find all the extents that end after
738 * our range starts.
739 */
740 node = tree_search(tree, start);
741 if (!node) {
742 prealloc = alloc_extent_state_atomic(prealloc);
743 BUG_ON(!prealloc);
744 err = insert_state(tree, prealloc, start, end, &bits);
745 prealloc = NULL;
746 BUG_ON(err == -EEXIST);
747 goto out;
748 }
749 state = rb_entry(node, struct extent_state, rb_node);
750 hit_next:
751 last_start = state->start;
752 last_end = state->end;
753
754 /*
755 * | ---- desired range ---- |
756 * | state |
757 *
758 * Just lock what we found and keep going
759 */
760 if (state->start == start && state->end <= end) {
761 struct rb_node *next_node;
762 if (state->state & exclusive_bits) {
763 *failed_start = state->start;
764 err = -EEXIST;
765 goto out;
766 }
767
768 set_state_bits(tree, state, &bits);
769
770 cache_state(state, cached_state);
771 merge_state(tree, state);
772 if (last_end == (u64)-1)
773 goto out;
774
775 start = last_end + 1;
776 next_node = rb_next(&state->rb_node);
777 if (next_node && start < end && prealloc && !need_resched()) {
778 state = rb_entry(next_node, struct extent_state,
779 rb_node);
780 if (state->start == start)
781 goto hit_next;
782 }
783 goto search_again;
784 }
785
786 /*
787 * | ---- desired range ---- |
788 * | state |
789 * or
790 * | ------------- state -------------- |
791 *
792 * We need to split the extent we found, and may flip bits on
793 * second half.
794 *
795 * If the extent we found extends past our
796 * range, we just split and search again. It'll get split
797 * again the next time though.
798 *
799 * If the extent we found is inside our range, we set the
800 * desired bit on it.
801 */
802 if (state->start < start) {
803 if (state->state & exclusive_bits) {
804 *failed_start = start;
805 err = -EEXIST;
806 goto out;
807 }
808
809 prealloc = alloc_extent_state_atomic(prealloc);
810 BUG_ON(!prealloc);
811 err = split_state(tree, state, prealloc, start);
812 BUG_ON(err == -EEXIST);
813 prealloc = NULL;
814 if (err)
815 goto out;
816 if (state->end <= end) {
817 set_state_bits(tree, state, &bits);
818 cache_state(state, cached_state);
819 merge_state(tree, state);
820 if (last_end == (u64)-1)
821 goto out;
822 start = last_end + 1;
823 }
824 goto search_again;
825 }
826 /*
827 * | ---- desired range ---- |
828 * | state | or | state |
829 *
830 * There's a hole, we need to insert something in it and
831 * ignore the extent we found.
832 */
833 if (state->start > start) {
834 u64 this_end;
835 if (end < last_start)
836 this_end = end;
837 else
838 this_end = last_start - 1;
839
840 prealloc = alloc_extent_state_atomic(prealloc);
841 BUG_ON(!prealloc);
842
843 /*
844 * Avoid to free 'prealloc' if it can be merged with
845 * the later extent.
846 */
847 err = insert_state(tree, prealloc, start, this_end,
848 &bits);
849 BUG_ON(err == -EEXIST);
850 if (err) {
851 free_extent_state(prealloc);
852 prealloc = NULL;
853 goto out;
854 }
855 cache_state(prealloc, cached_state);
856 prealloc = NULL;
857 start = this_end + 1;
858 goto search_again;
859 }
860 /*
861 * | ---- desired range ---- |
862 * | state |
863 * We need to split the extent, and set the bit
864 * on the first half
865 */
866 if (state->start <= end && state->end > end) {
867 if (state->state & exclusive_bits) {
868 *failed_start = start;
869 err = -EEXIST;
870 goto out;
871 }
872
873 prealloc = alloc_extent_state_atomic(prealloc);
874 BUG_ON(!prealloc);
875 err = split_state(tree, state, prealloc, end + 1);
876 BUG_ON(err == -EEXIST);
877
878 set_state_bits(tree, prealloc, &bits);
879 cache_state(prealloc, cached_state);
880 merge_state(tree, prealloc);
881 prealloc = NULL;
882 goto out;
883 }
884
885 goto search_again;
886
887 out:
888 spin_unlock(&tree->lock);
889 if (prealloc)
890 free_extent_state(prealloc);
891
892 return err;
893
894 search_again:
895 if (start > end)
896 goto out;
897 spin_unlock(&tree->lock);
898 if (mask & __GFP_WAIT)
899 cond_resched();
900 goto again;
901 }
902
903 /**
904 * convert_extent - convert all bits in a given range from one bit to another
905 * @tree: the io tree to search
906 * @start: the start offset in bytes
907 * @end: the end offset in bytes (inclusive)
908 * @bits: the bits to set in this range
909 * @clear_bits: the bits to clear in this range
910 * @mask: the allocation mask
911 *
912 * This will go through and set bits for the given range. If any states exist
913 * already in this range they are set with the given bit and cleared of the
914 * clear_bits. This is only meant to be used by things that are mergeable, ie
915 * converting from say DELALLOC to DIRTY. This is not meant to be used with
916 * boundary bits like LOCK.
917 */
918 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
919 int bits, int clear_bits, gfp_t mask)
920 {
921 struct extent_state *state;
922 struct extent_state *prealloc = NULL;
923 struct rb_node *node;
924 int err = 0;
925 u64 last_start;
926 u64 last_end;
927
928 again:
929 if (!prealloc && (mask & __GFP_WAIT)) {
930 prealloc = alloc_extent_state(mask);
931 if (!prealloc)
932 return -ENOMEM;
933 }
934
935 spin_lock(&tree->lock);
936 /*
937 * this search will find all the extents that end after
938 * our range starts.
939 */
940 node = tree_search(tree, start);
941 if (!node) {
942 prealloc = alloc_extent_state_atomic(prealloc);
943 if (!prealloc) {
944 err = -ENOMEM;
945 goto out;
946 }
947 err = insert_state(tree, prealloc, start, end, &bits);
948 prealloc = NULL;
949 BUG_ON(err == -EEXIST);
950 goto out;
951 }
952 state = rb_entry(node, struct extent_state, rb_node);
953 hit_next:
954 last_start = state->start;
955 last_end = state->end;
956
957 /*
958 * | ---- desired range ---- |
959 * | state |
960 *
961 * Just lock what we found and keep going
962 */
963 if (state->start == start && state->end <= end) {
964 struct rb_node *next_node;
965
966 set_state_bits(tree, state, &bits);
967 clear_state_bit(tree, state, &clear_bits, 0);
968 if (last_end == (u64)-1)
969 goto out;
970
971 start = last_end + 1;
972 next_node = rb_next(&state->rb_node);
973 if (next_node && start < end && prealloc && !need_resched()) {
974 state = rb_entry(next_node, struct extent_state,
975 rb_node);
976 if (state->start == start)
977 goto hit_next;
978 }
979 goto search_again;
980 }
981
982 /*
983 * | ---- desired range ---- |
984 * | state |
985 * or
986 * | ------------- state -------------- |
987 *
988 * We need to split the extent we found, and may flip bits on
989 * second half.
990 *
991 * If the extent we found extends past our
992 * range, we just split and search again. It'll get split
993 * again the next time though.
994 *
995 * If the extent we found is inside our range, we set the
996 * desired bit on it.
997 */
998 if (state->start < start) {
999 prealloc = alloc_extent_state_atomic(prealloc);
1000 if (!prealloc) {
1001 err = -ENOMEM;
1002 goto out;
1003 }
1004 err = split_state(tree, state, prealloc, start);
1005 BUG_ON(err == -EEXIST);
1006 prealloc = NULL;
1007 if (err)
1008 goto out;
1009 if (state->end <= end) {
1010 set_state_bits(tree, state, &bits);
1011 clear_state_bit(tree, state, &clear_bits, 0);
1012 if (last_end == (u64)-1)
1013 goto out;
1014 start = last_end + 1;
1015 }
1016 goto search_again;
1017 }
1018 /*
1019 * | ---- desired range ---- |
1020 * | state | or | state |
1021 *
1022 * There's a hole, we need to insert something in it and
1023 * ignore the extent we found.
1024 */
1025 if (state->start > start) {
1026 u64 this_end;
1027 if (end < last_start)
1028 this_end = end;
1029 else
1030 this_end = last_start - 1;
1031
1032 prealloc = alloc_extent_state_atomic(prealloc);
1033 if (!prealloc) {
1034 err = -ENOMEM;
1035 goto out;
1036 }
1037
1038 /*
1039 * Avoid to free 'prealloc' if it can be merged with
1040 * the later extent.
1041 */
1042 err = insert_state(tree, prealloc, start, this_end,
1043 &bits);
1044 BUG_ON(err == -EEXIST);
1045 if (err) {
1046 free_extent_state(prealloc);
1047 prealloc = NULL;
1048 goto out;
1049 }
1050 prealloc = NULL;
1051 start = this_end + 1;
1052 goto search_again;
1053 }
1054 /*
1055 * | ---- desired range ---- |
1056 * | state |
1057 * We need to split the extent, and set the bit
1058 * on the first half
1059 */
1060 if (state->start <= end && state->end > end) {
1061 prealloc = alloc_extent_state_atomic(prealloc);
1062 if (!prealloc) {
1063 err = -ENOMEM;
1064 goto out;
1065 }
1066
1067 err = split_state(tree, state, prealloc, end + 1);
1068 BUG_ON(err == -EEXIST);
1069
1070 set_state_bits(tree, prealloc, &bits);
1071 clear_state_bit(tree, prealloc, &clear_bits, 0);
1072 prealloc = NULL;
1073 goto out;
1074 }
1075
1076 goto search_again;
1077
1078 out:
1079 spin_unlock(&tree->lock);
1080 if (prealloc)
1081 free_extent_state(prealloc);
1082
1083 return err;
1084
1085 search_again:
1086 if (start > end)
1087 goto out;
1088 spin_unlock(&tree->lock);
1089 if (mask & __GFP_WAIT)
1090 cond_resched();
1091 goto again;
1092 }
1093
1094 /* wrappers around set/clear extent bit */
1095 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1096 gfp_t mask)
1097 {
1098 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
1099 NULL, mask);
1100 }
1101
1102 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1103 int bits, gfp_t mask)
1104 {
1105 return set_extent_bit(tree, start, end, bits, 0, NULL,
1106 NULL, mask);
1107 }
1108
1109 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1110 int bits, gfp_t mask)
1111 {
1112 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1113 }
1114
1115 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1116 struct extent_state **cached_state, gfp_t mask)
1117 {
1118 return set_extent_bit(tree, start, end,
1119 EXTENT_DELALLOC | EXTENT_UPTODATE,
1120 0, NULL, cached_state, mask);
1121 }
1122
1123 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1124 gfp_t mask)
1125 {
1126 return clear_extent_bit(tree, start, end,
1127 EXTENT_DIRTY | EXTENT_DELALLOC |
1128 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1129 }
1130
1131 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1132 gfp_t mask)
1133 {
1134 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
1135 NULL, mask);
1136 }
1137
1138 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1139 struct extent_state **cached_state, gfp_t mask)
1140 {
1141 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1142 NULL, cached_state, mask);
1143 }
1144
1145 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
1146 u64 end, struct extent_state **cached_state,
1147 gfp_t mask)
1148 {
1149 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1150 cached_state, mask);
1151 }
1152
1153 /*
1154 * either insert or lock state struct between start and end use mask to tell
1155 * us if waiting is desired.
1156 */
1157 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1158 int bits, struct extent_state **cached_state, gfp_t mask)
1159 {
1160 int err;
1161 u64 failed_start;
1162 while (1) {
1163 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1164 EXTENT_LOCKED, &failed_start,
1165 cached_state, mask);
1166 if (err == -EEXIST && (mask & __GFP_WAIT)) {
1167 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1168 start = failed_start;
1169 } else {
1170 break;
1171 }
1172 WARN_ON(start > end);
1173 }
1174 return err;
1175 }
1176
1177 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1178 {
1179 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1180 }
1181
1182 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1183 gfp_t mask)
1184 {
1185 int err;
1186 u64 failed_start;
1187
1188 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1189 &failed_start, NULL, mask);
1190 if (err == -EEXIST) {
1191 if (failed_start > start)
1192 clear_extent_bit(tree, start, failed_start - 1,
1193 EXTENT_LOCKED, 1, 0, NULL, mask);
1194 return 0;
1195 }
1196 return 1;
1197 }
1198
1199 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1200 struct extent_state **cached, gfp_t mask)
1201 {
1202 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1203 mask);
1204 }
1205
1206 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1207 {
1208 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1209 mask);
1210 }
1211
1212 /*
1213 * helper function to set both pages and extents in the tree writeback
1214 */
1215 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1216 {
1217 unsigned long index = start >> PAGE_CACHE_SHIFT;
1218 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1219 struct page *page;
1220
1221 while (index <= end_index) {
1222 page = find_get_page(tree->mapping, index);
1223 BUG_ON(!page);
1224 set_page_writeback(page);
1225 page_cache_release(page);
1226 index++;
1227 }
1228 return 0;
1229 }
1230
1231 /* find the first state struct with 'bits' set after 'start', and
1232 * return it. tree->lock must be held. NULL will returned if
1233 * nothing was found after 'start'
1234 */
1235 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1236 u64 start, int bits)
1237 {
1238 struct rb_node *node;
1239 struct extent_state *state;
1240
1241 /*
1242 * this search will find all the extents that end after
1243 * our range starts.
1244 */
1245 node = tree_search(tree, start);
1246 if (!node)
1247 goto out;
1248
1249 while (1) {
1250 state = rb_entry(node, struct extent_state, rb_node);
1251 if (state->end >= start && (state->state & bits))
1252 return state;
1253
1254 node = rb_next(node);
1255 if (!node)
1256 break;
1257 }
1258 out:
1259 return NULL;
1260 }
1261
1262 /*
1263 * find the first offset in the io tree with 'bits' set. zero is
1264 * returned if we find something, and *start_ret and *end_ret are
1265 * set to reflect the state struct that was found.
1266 *
1267 * If nothing was found, 1 is returned, < 0 on error
1268 */
1269 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1270 u64 *start_ret, u64 *end_ret, int bits)
1271 {
1272 struct extent_state *state;
1273 int ret = 1;
1274
1275 spin_lock(&tree->lock);
1276 state = find_first_extent_bit_state(tree, start, bits);
1277 if (state) {
1278 *start_ret = state->start;
1279 *end_ret = state->end;
1280 ret = 0;
1281 }
1282 spin_unlock(&tree->lock);
1283 return ret;
1284 }
1285
1286 /*
1287 * find a contiguous range of bytes in the file marked as delalloc, not
1288 * more than 'max_bytes'. start and end are used to return the range,
1289 *
1290 * 1 is returned if we find something, 0 if nothing was in the tree
1291 */
1292 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1293 u64 *start, u64 *end, u64 max_bytes,
1294 struct extent_state **cached_state)
1295 {
1296 struct rb_node *node;
1297 struct extent_state *state;
1298 u64 cur_start = *start;
1299 u64 found = 0;
1300 u64 total_bytes = 0;
1301
1302 spin_lock(&tree->lock);
1303
1304 /*
1305 * this search will find all the extents that end after
1306 * our range starts.
1307 */
1308 node = tree_search(tree, cur_start);
1309 if (!node) {
1310 if (!found)
1311 *end = (u64)-1;
1312 goto out;
1313 }
1314
1315 while (1) {
1316 state = rb_entry(node, struct extent_state, rb_node);
1317 if (found && (state->start != cur_start ||
1318 (state->state & EXTENT_BOUNDARY))) {
1319 goto out;
1320 }
1321 if (!(state->state & EXTENT_DELALLOC)) {
1322 if (!found)
1323 *end = state->end;
1324 goto out;
1325 }
1326 if (!found) {
1327 *start = state->start;
1328 *cached_state = state;
1329 atomic_inc(&state->refs);
1330 }
1331 found++;
1332 *end = state->end;
1333 cur_start = state->end + 1;
1334 node = rb_next(node);
1335 if (!node)
1336 break;
1337 total_bytes += state->end - state->start + 1;
1338 if (total_bytes >= max_bytes)
1339 break;
1340 }
1341 out:
1342 spin_unlock(&tree->lock);
1343 return found;
1344 }
1345
1346 static noinline int __unlock_for_delalloc(struct inode *inode,
1347 struct page *locked_page,
1348 u64 start, u64 end)
1349 {
1350 int ret;
1351 struct page *pages[16];
1352 unsigned long index = start >> PAGE_CACHE_SHIFT;
1353 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1354 unsigned long nr_pages = end_index - index + 1;
1355 int i;
1356
1357 if (index == locked_page->index && end_index == index)
1358 return 0;
1359
1360 while (nr_pages > 0) {
1361 ret = find_get_pages_contig(inode->i_mapping, index,
1362 min_t(unsigned long, nr_pages,
1363 ARRAY_SIZE(pages)), pages);
1364 for (i = 0; i < ret; i++) {
1365 if (pages[i] != locked_page)
1366 unlock_page(pages[i]);
1367 page_cache_release(pages[i]);
1368 }
1369 nr_pages -= ret;
1370 index += ret;
1371 cond_resched();
1372 }
1373 return 0;
1374 }
1375
1376 static noinline int lock_delalloc_pages(struct inode *inode,
1377 struct page *locked_page,
1378 u64 delalloc_start,
1379 u64 delalloc_end)
1380 {
1381 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1382 unsigned long start_index = index;
1383 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1384 unsigned long pages_locked = 0;
1385 struct page *pages[16];
1386 unsigned long nrpages;
1387 int ret;
1388 int i;
1389
1390 /* the caller is responsible for locking the start index */
1391 if (index == locked_page->index && index == end_index)
1392 return 0;
1393
1394 /* skip the page at the start index */
1395 nrpages = end_index - index + 1;
1396 while (nrpages > 0) {
1397 ret = find_get_pages_contig(inode->i_mapping, index,
1398 min_t(unsigned long,
1399 nrpages, ARRAY_SIZE(pages)), pages);
1400 if (ret == 0) {
1401 ret = -EAGAIN;
1402 goto done;
1403 }
1404 /* now we have an array of pages, lock them all */
1405 for (i = 0; i < ret; i++) {
1406 /*
1407 * the caller is taking responsibility for
1408 * locked_page
1409 */
1410 if (pages[i] != locked_page) {
1411 lock_page(pages[i]);
1412 if (!PageDirty(pages[i]) ||
1413 pages[i]->mapping != inode->i_mapping) {
1414 ret = -EAGAIN;
1415 unlock_page(pages[i]);
1416 page_cache_release(pages[i]);
1417 goto done;
1418 }
1419 }
1420 page_cache_release(pages[i]);
1421 pages_locked++;
1422 }
1423 nrpages -= ret;
1424 index += ret;
1425 cond_resched();
1426 }
1427 ret = 0;
1428 done:
1429 if (ret && pages_locked) {
1430 __unlock_for_delalloc(inode, locked_page,
1431 delalloc_start,
1432 ((u64)(start_index + pages_locked - 1)) <<
1433 PAGE_CACHE_SHIFT);
1434 }
1435 return ret;
1436 }
1437
1438 /*
1439 * find a contiguous range of bytes in the file marked as delalloc, not
1440 * more than 'max_bytes'. start and end are used to return the range,
1441 *
1442 * 1 is returned if we find something, 0 if nothing was in the tree
1443 */
1444 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1445 struct extent_io_tree *tree,
1446 struct page *locked_page,
1447 u64 *start, u64 *end,
1448 u64 max_bytes)
1449 {
1450 u64 delalloc_start;
1451 u64 delalloc_end;
1452 u64 found;
1453 struct extent_state *cached_state = NULL;
1454 int ret;
1455 int loops = 0;
1456
1457 again:
1458 /* step one, find a bunch of delalloc bytes starting at start */
1459 delalloc_start = *start;
1460 delalloc_end = 0;
1461 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1462 max_bytes, &cached_state);
1463 if (!found || delalloc_end <= *start) {
1464 *start = delalloc_start;
1465 *end = delalloc_end;
1466 free_extent_state(cached_state);
1467 return found;
1468 }
1469
1470 /*
1471 * start comes from the offset of locked_page. We have to lock
1472 * pages in order, so we can't process delalloc bytes before
1473 * locked_page
1474 */
1475 if (delalloc_start < *start)
1476 delalloc_start = *start;
1477
1478 /*
1479 * make sure to limit the number of pages we try to lock down
1480 * if we're looping.
1481 */
1482 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1483 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1484
1485 /* step two, lock all the pages after the page that has start */
1486 ret = lock_delalloc_pages(inode, locked_page,
1487 delalloc_start, delalloc_end);
1488 if (ret == -EAGAIN) {
1489 /* some of the pages are gone, lets avoid looping by
1490 * shortening the size of the delalloc range we're searching
1491 */
1492 free_extent_state(cached_state);
1493 if (!loops) {
1494 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1495 max_bytes = PAGE_CACHE_SIZE - offset;
1496 loops = 1;
1497 goto again;
1498 } else {
1499 found = 0;
1500 goto out_failed;
1501 }
1502 }
1503 BUG_ON(ret);
1504
1505 /* step three, lock the state bits for the whole range */
1506 lock_extent_bits(tree, delalloc_start, delalloc_end,
1507 0, &cached_state, GFP_NOFS);
1508
1509 /* then test to make sure it is all still delalloc */
1510 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1511 EXTENT_DELALLOC, 1, cached_state);
1512 if (!ret) {
1513 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1514 &cached_state, GFP_NOFS);
1515 __unlock_for_delalloc(inode, locked_page,
1516 delalloc_start, delalloc_end);
1517 cond_resched();
1518 goto again;
1519 }
1520 free_extent_state(cached_state);
1521 *start = delalloc_start;
1522 *end = delalloc_end;
1523 out_failed:
1524 return found;
1525 }
1526
1527 int extent_clear_unlock_delalloc(struct inode *inode,
1528 struct extent_io_tree *tree,
1529 u64 start, u64 end, struct page *locked_page,
1530 unsigned long op)
1531 {
1532 int ret;
1533 struct page *pages[16];
1534 unsigned long index = start >> PAGE_CACHE_SHIFT;
1535 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1536 unsigned long nr_pages = end_index - index + 1;
1537 int i;
1538 int clear_bits = 0;
1539
1540 if (op & EXTENT_CLEAR_UNLOCK)
1541 clear_bits |= EXTENT_LOCKED;
1542 if (op & EXTENT_CLEAR_DIRTY)
1543 clear_bits |= EXTENT_DIRTY;
1544
1545 if (op & EXTENT_CLEAR_DELALLOC)
1546 clear_bits |= EXTENT_DELALLOC;
1547
1548 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1549 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1550 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1551 EXTENT_SET_PRIVATE2)))
1552 return 0;
1553
1554 while (nr_pages > 0) {
1555 ret = find_get_pages_contig(inode->i_mapping, index,
1556 min_t(unsigned long,
1557 nr_pages, ARRAY_SIZE(pages)), pages);
1558 for (i = 0; i < ret; i++) {
1559
1560 if (op & EXTENT_SET_PRIVATE2)
1561 SetPagePrivate2(pages[i]);
1562
1563 if (pages[i] == locked_page) {
1564 page_cache_release(pages[i]);
1565 continue;
1566 }
1567 if (op & EXTENT_CLEAR_DIRTY)
1568 clear_page_dirty_for_io(pages[i]);
1569 if (op & EXTENT_SET_WRITEBACK)
1570 set_page_writeback(pages[i]);
1571 if (op & EXTENT_END_WRITEBACK)
1572 end_page_writeback(pages[i]);
1573 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1574 unlock_page(pages[i]);
1575 page_cache_release(pages[i]);
1576 }
1577 nr_pages -= ret;
1578 index += ret;
1579 cond_resched();
1580 }
1581 return 0;
1582 }
1583
1584 /*
1585 * count the number of bytes in the tree that have a given bit(s)
1586 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1587 * cached. The total number found is returned.
1588 */
1589 u64 count_range_bits(struct extent_io_tree *tree,
1590 u64 *start, u64 search_end, u64 max_bytes,
1591 unsigned long bits, int contig)
1592 {
1593 struct rb_node *node;
1594 struct extent_state *state;
1595 u64 cur_start = *start;
1596 u64 total_bytes = 0;
1597 u64 last = 0;
1598 int found = 0;
1599
1600 if (search_end <= cur_start) {
1601 WARN_ON(1);
1602 return 0;
1603 }
1604
1605 spin_lock(&tree->lock);
1606 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1607 total_bytes = tree->dirty_bytes;
1608 goto out;
1609 }
1610 /*
1611 * this search will find all the extents that end after
1612 * our range starts.
1613 */
1614 node = tree_search(tree, cur_start);
1615 if (!node)
1616 goto out;
1617
1618 while (1) {
1619 state = rb_entry(node, struct extent_state, rb_node);
1620 if (state->start > search_end)
1621 break;
1622 if (contig && found && state->start > last + 1)
1623 break;
1624 if (state->end >= cur_start && (state->state & bits) == bits) {
1625 total_bytes += min(search_end, state->end) + 1 -
1626 max(cur_start, state->start);
1627 if (total_bytes >= max_bytes)
1628 break;
1629 if (!found) {
1630 *start = max(cur_start, state->start);
1631 found = 1;
1632 }
1633 last = state->end;
1634 } else if (contig && found) {
1635 break;
1636 }
1637 node = rb_next(node);
1638 if (!node)
1639 break;
1640 }
1641 out:
1642 spin_unlock(&tree->lock);
1643 return total_bytes;
1644 }
1645
1646 /*
1647 * set the private field for a given byte offset in the tree. If there isn't
1648 * an extent_state there already, this does nothing.
1649 */
1650 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1651 {
1652 struct rb_node *node;
1653 struct extent_state *state;
1654 int ret = 0;
1655
1656 spin_lock(&tree->lock);
1657 /*
1658 * this search will find all the extents that end after
1659 * our range starts.
1660 */
1661 node = tree_search(tree, start);
1662 if (!node) {
1663 ret = -ENOENT;
1664 goto out;
1665 }
1666 state = rb_entry(node, struct extent_state, rb_node);
1667 if (state->start != start) {
1668 ret = -ENOENT;
1669 goto out;
1670 }
1671 state->private = private;
1672 out:
1673 spin_unlock(&tree->lock);
1674 return ret;
1675 }
1676
1677 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1678 {
1679 struct rb_node *node;
1680 struct extent_state *state;
1681 int ret = 0;
1682
1683 spin_lock(&tree->lock);
1684 /*
1685 * this search will find all the extents that end after
1686 * our range starts.
1687 */
1688 node = tree_search(tree, start);
1689 if (!node) {
1690 ret = -ENOENT;
1691 goto out;
1692 }
1693 state = rb_entry(node, struct extent_state, rb_node);
1694 if (state->start != start) {
1695 ret = -ENOENT;
1696 goto out;
1697 }
1698 *private = state->private;
1699 out:
1700 spin_unlock(&tree->lock);
1701 return ret;
1702 }
1703
1704 /*
1705 * searches a range in the state tree for a given mask.
1706 * If 'filled' == 1, this returns 1 only if every extent in the tree
1707 * has the bits set. Otherwise, 1 is returned if any bit in the
1708 * range is found set.
1709 */
1710 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1711 int bits, int filled, struct extent_state *cached)
1712 {
1713 struct extent_state *state = NULL;
1714 struct rb_node *node;
1715 int bitset = 0;
1716
1717 spin_lock(&tree->lock);
1718 if (cached && cached->tree && cached->start <= start &&
1719 cached->end > start)
1720 node = &cached->rb_node;
1721 else
1722 node = tree_search(tree, start);
1723 while (node && start <= end) {
1724 state = rb_entry(node, struct extent_state, rb_node);
1725
1726 if (filled && state->start > start) {
1727 bitset = 0;
1728 break;
1729 }
1730
1731 if (state->start > end)
1732 break;
1733
1734 if (state->state & bits) {
1735 bitset = 1;
1736 if (!filled)
1737 break;
1738 } else if (filled) {
1739 bitset = 0;
1740 break;
1741 }
1742
1743 if (state->end == (u64)-1)
1744 break;
1745
1746 start = state->end + 1;
1747 if (start > end)
1748 break;
1749 node = rb_next(node);
1750 if (!node) {
1751 if (filled)
1752 bitset = 0;
1753 break;
1754 }
1755 }
1756 spin_unlock(&tree->lock);
1757 return bitset;
1758 }
1759
1760 /*
1761 * helper function to set a given page up to date if all the
1762 * extents in the tree for that page are up to date
1763 */
1764 static int check_page_uptodate(struct extent_io_tree *tree,
1765 struct page *page)
1766 {
1767 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1768 u64 end = start + PAGE_CACHE_SIZE - 1;
1769 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1770 SetPageUptodate(page);
1771 return 0;
1772 }
1773
1774 /*
1775 * helper function to unlock a page if all the extents in the tree
1776 * for that page are unlocked
1777 */
1778 static int check_page_locked(struct extent_io_tree *tree,
1779 struct page *page)
1780 {
1781 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1782 u64 end = start + PAGE_CACHE_SIZE - 1;
1783 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1784 unlock_page(page);
1785 return 0;
1786 }
1787
1788 /*
1789 * helper function to end page writeback if all the extents
1790 * in the tree for that page are done with writeback
1791 */
1792 static int check_page_writeback(struct extent_io_tree *tree,
1793 struct page *page)
1794 {
1795 end_page_writeback(page);
1796 return 0;
1797 }
1798
1799 /*
1800 * When IO fails, either with EIO or csum verification fails, we
1801 * try other mirrors that might have a good copy of the data. This
1802 * io_failure_record is used to record state as we go through all the
1803 * mirrors. If another mirror has good data, the page is set up to date
1804 * and things continue. If a good mirror can't be found, the original
1805 * bio end_io callback is called to indicate things have failed.
1806 */
1807 struct io_failure_record {
1808 struct page *page;
1809 u64 start;
1810 u64 len;
1811 u64 logical;
1812 unsigned long bio_flags;
1813 int this_mirror;
1814 int failed_mirror;
1815 int in_validation;
1816 };
1817
1818 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1819 int did_repair)
1820 {
1821 int ret;
1822 int err = 0;
1823 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1824
1825 set_state_private(failure_tree, rec->start, 0);
1826 ret = clear_extent_bits(failure_tree, rec->start,
1827 rec->start + rec->len - 1,
1828 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1829 if (ret)
1830 err = ret;
1831
1832 if (did_repair) {
1833 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1834 rec->start + rec->len - 1,
1835 EXTENT_DAMAGED, GFP_NOFS);
1836 if (ret && !err)
1837 err = ret;
1838 }
1839
1840 kfree(rec);
1841 return err;
1842 }
1843
1844 static void repair_io_failure_callback(struct bio *bio, int err)
1845 {
1846 complete(bio->bi_private);
1847 }
1848
1849 /*
1850 * this bypasses the standard btrfs submit functions deliberately, as
1851 * the standard behavior is to write all copies in a raid setup. here we only
1852 * want to write the one bad copy. so we do the mapping for ourselves and issue
1853 * submit_bio directly.
1854 * to avoid any synchonization issues, wait for the data after writing, which
1855 * actually prevents the read that triggered the error from finishing.
1856 * currently, there can be no more than two copies of every data bit. thus,
1857 * exactly one rewrite is required.
1858 */
1859 int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start,
1860 u64 length, u64 logical, struct page *page,
1861 int mirror_num)
1862 {
1863 struct bio *bio;
1864 struct btrfs_device *dev;
1865 DECLARE_COMPLETION_ONSTACK(compl);
1866 u64 map_length = 0;
1867 u64 sector;
1868 struct btrfs_bio *bbio = NULL;
1869 int ret;
1870
1871 BUG_ON(!mirror_num);
1872
1873 bio = bio_alloc(GFP_NOFS, 1);
1874 if (!bio)
1875 return -EIO;
1876 bio->bi_private = &compl;
1877 bio->bi_end_io = repair_io_failure_callback;
1878 bio->bi_size = 0;
1879 map_length = length;
1880
1881 ret = btrfs_map_block(map_tree, WRITE, logical,
1882 &map_length, &bbio, mirror_num);
1883 if (ret) {
1884 bio_put(bio);
1885 return -EIO;
1886 }
1887 BUG_ON(mirror_num != bbio->mirror_num);
1888 sector = bbio->stripes[mirror_num-1].physical >> 9;
1889 bio->bi_sector = sector;
1890 dev = bbio->stripes[mirror_num-1].dev;
1891 kfree(bbio);
1892 if (!dev || !dev->bdev || !dev->writeable) {
1893 bio_put(bio);
1894 return -EIO;
1895 }
1896 bio->bi_bdev = dev->bdev;
1897 bio_add_page(bio, page, length, start-page_offset(page));
1898 btrfsic_submit_bio(WRITE_SYNC, bio);
1899 wait_for_completion(&compl);
1900
1901 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
1902 /* try to remap that extent elsewhere? */
1903 bio_put(bio);
1904 return -EIO;
1905 }
1906
1907 printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s "
1908 "sector %llu)\n", page->mapping->host->i_ino, start,
1909 dev->name, sector);
1910
1911 bio_put(bio);
1912 return 0;
1913 }
1914
1915 /*
1916 * each time an IO finishes, we do a fast check in the IO failure tree
1917 * to see if we need to process or clean up an io_failure_record
1918 */
1919 static int clean_io_failure(u64 start, struct page *page)
1920 {
1921 u64 private;
1922 u64 private_failure;
1923 struct io_failure_record *failrec;
1924 struct btrfs_mapping_tree *map_tree;
1925 struct extent_state *state;
1926 int num_copies;
1927 int did_repair = 0;
1928 int ret;
1929 struct inode *inode = page->mapping->host;
1930
1931 private = 0;
1932 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1933 (u64)-1, 1, EXTENT_DIRTY, 0);
1934 if (!ret)
1935 return 0;
1936
1937 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
1938 &private_failure);
1939 if (ret)
1940 return 0;
1941
1942 failrec = (struct io_failure_record *)(unsigned long) private_failure;
1943 BUG_ON(!failrec->this_mirror);
1944
1945 if (failrec->in_validation) {
1946 /* there was no real error, just free the record */
1947 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
1948 failrec->start);
1949 did_repair = 1;
1950 goto out;
1951 }
1952
1953 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1954 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1955 failrec->start,
1956 EXTENT_LOCKED);
1957 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1958
1959 if (state && state->start == failrec->start) {
1960 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
1961 num_copies = btrfs_num_copies(map_tree, failrec->logical,
1962 failrec->len);
1963 if (num_copies > 1) {
1964 ret = repair_io_failure(map_tree, start, failrec->len,
1965 failrec->logical, page,
1966 failrec->failed_mirror);
1967 did_repair = !ret;
1968 }
1969 }
1970
1971 out:
1972 if (!ret)
1973 ret = free_io_failure(inode, failrec, did_repair);
1974
1975 return ret;
1976 }
1977
1978 /*
1979 * this is a generic handler for readpage errors (default
1980 * readpage_io_failed_hook). if other copies exist, read those and write back
1981 * good data to the failed position. does not investigate in remapping the
1982 * failed extent elsewhere, hoping the device will be smart enough to do this as
1983 * needed
1984 */
1985
1986 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
1987 u64 start, u64 end, int failed_mirror,
1988 struct extent_state *state)
1989 {
1990 struct io_failure_record *failrec = NULL;
1991 u64 private;
1992 struct extent_map *em;
1993 struct inode *inode = page->mapping->host;
1994 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1995 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1996 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1997 struct bio *bio;
1998 int num_copies;
1999 int ret;
2000 int read_mode;
2001 u64 logical;
2002
2003 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2004
2005 ret = get_state_private(failure_tree, start, &private);
2006 if (ret) {
2007 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2008 if (!failrec)
2009 return -ENOMEM;
2010 failrec->start = start;
2011 failrec->len = end - start + 1;
2012 failrec->this_mirror = 0;
2013 failrec->bio_flags = 0;
2014 failrec->in_validation = 0;
2015
2016 read_lock(&em_tree->lock);
2017 em = lookup_extent_mapping(em_tree, start, failrec->len);
2018 if (!em) {
2019 read_unlock(&em_tree->lock);
2020 kfree(failrec);
2021 return -EIO;
2022 }
2023
2024 if (em->start > start || em->start + em->len < start) {
2025 free_extent_map(em);
2026 em = NULL;
2027 }
2028 read_unlock(&em_tree->lock);
2029
2030 if (!em || IS_ERR(em)) {
2031 kfree(failrec);
2032 return -EIO;
2033 }
2034 logical = start - em->start;
2035 logical = em->block_start + logical;
2036 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2037 logical = em->block_start;
2038 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2039 extent_set_compress_type(&failrec->bio_flags,
2040 em->compress_type);
2041 }
2042 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2043 "len=%llu\n", logical, start, failrec->len);
2044 failrec->logical = logical;
2045 free_extent_map(em);
2046
2047 /* set the bits in the private failure tree */
2048 ret = set_extent_bits(failure_tree, start, end,
2049 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2050 if (ret >= 0)
2051 ret = set_state_private(failure_tree, start,
2052 (u64)(unsigned long)failrec);
2053 /* set the bits in the inode's tree */
2054 if (ret >= 0)
2055 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2056 GFP_NOFS);
2057 if (ret < 0) {
2058 kfree(failrec);
2059 return ret;
2060 }
2061 } else {
2062 failrec = (struct io_failure_record *)(unsigned long)private;
2063 pr_debug("bio_readpage_error: (found) logical=%llu, "
2064 "start=%llu, len=%llu, validation=%d\n",
2065 failrec->logical, failrec->start, failrec->len,
2066 failrec->in_validation);
2067 /*
2068 * when data can be on disk more than twice, add to failrec here
2069 * (e.g. with a list for failed_mirror) to make
2070 * clean_io_failure() clean all those errors at once.
2071 */
2072 }
2073 num_copies = btrfs_num_copies(
2074 &BTRFS_I(inode)->root->fs_info->mapping_tree,
2075 failrec->logical, failrec->len);
2076 if (num_copies == 1) {
2077 /*
2078 * we only have a single copy of the data, so don't bother with
2079 * all the retry and error correction code that follows. no
2080 * matter what the error is, it is very likely to persist.
2081 */
2082 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2083 "state=%p, num_copies=%d, next_mirror %d, "
2084 "failed_mirror %d\n", state, num_copies,
2085 failrec->this_mirror, failed_mirror);
2086 free_io_failure(inode, failrec, 0);
2087 return -EIO;
2088 }
2089
2090 if (!state) {
2091 spin_lock(&tree->lock);
2092 state = find_first_extent_bit_state(tree, failrec->start,
2093 EXTENT_LOCKED);
2094 if (state && state->start != failrec->start)
2095 state = NULL;
2096 spin_unlock(&tree->lock);
2097 }
2098
2099 /*
2100 * there are two premises:
2101 * a) deliver good data to the caller
2102 * b) correct the bad sectors on disk
2103 */
2104 if (failed_bio->bi_vcnt > 1) {
2105 /*
2106 * to fulfill b), we need to know the exact failing sectors, as
2107 * we don't want to rewrite any more than the failed ones. thus,
2108 * we need separate read requests for the failed bio
2109 *
2110 * if the following BUG_ON triggers, our validation request got
2111 * merged. we need separate requests for our algorithm to work.
2112 */
2113 BUG_ON(failrec->in_validation);
2114 failrec->in_validation = 1;
2115 failrec->this_mirror = failed_mirror;
2116 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2117 } else {
2118 /*
2119 * we're ready to fulfill a) and b) alongside. get a good copy
2120 * of the failed sector and if we succeed, we have setup
2121 * everything for repair_io_failure to do the rest for us.
2122 */
2123 if (failrec->in_validation) {
2124 BUG_ON(failrec->this_mirror != failed_mirror);
2125 failrec->in_validation = 0;
2126 failrec->this_mirror = 0;
2127 }
2128 failrec->failed_mirror = failed_mirror;
2129 failrec->this_mirror++;
2130 if (failrec->this_mirror == failed_mirror)
2131 failrec->this_mirror++;
2132 read_mode = READ_SYNC;
2133 }
2134
2135 if (!state || failrec->this_mirror > num_copies) {
2136 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2137 "next_mirror %d, failed_mirror %d\n", state,
2138 num_copies, failrec->this_mirror, failed_mirror);
2139 free_io_failure(inode, failrec, 0);
2140 return -EIO;
2141 }
2142
2143 bio = bio_alloc(GFP_NOFS, 1);
2144 bio->bi_private = state;
2145 bio->bi_end_io = failed_bio->bi_end_io;
2146 bio->bi_sector = failrec->logical >> 9;
2147 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2148 bio->bi_size = 0;
2149
2150 bio_add_page(bio, page, failrec->len, start - page_offset(page));
2151
2152 pr_debug("bio_readpage_error: submitting new read[%#x] to "
2153 "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2154 failrec->this_mirror, num_copies, failrec->in_validation);
2155
2156 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2157 failrec->this_mirror,
2158 failrec->bio_flags, 0);
2159 return ret;
2160 }
2161
2162 /* lots and lots of room for performance fixes in the end_bio funcs */
2163
2164 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2165 {
2166 int uptodate = (err == 0);
2167 struct extent_io_tree *tree;
2168 int ret;
2169
2170 tree = &BTRFS_I(page->mapping->host)->io_tree;
2171
2172 if (tree->ops && tree->ops->writepage_end_io_hook) {
2173 ret = tree->ops->writepage_end_io_hook(page, start,
2174 end, NULL, uptodate);
2175 if (ret)
2176 uptodate = 0;
2177 }
2178
2179 if (!uptodate && tree->ops &&
2180 tree->ops->writepage_io_failed_hook) {
2181 ret = tree->ops->writepage_io_failed_hook(NULL, page,
2182 start, end, NULL);
2183 /* Writeback already completed */
2184 if (ret == 0)
2185 return 1;
2186 }
2187
2188 if (!uptodate) {
2189 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
2190 ClearPageUptodate(page);
2191 SetPageError(page);
2192 }
2193 return 0;
2194 }
2195
2196 /*
2197 * after a writepage IO is done, we need to:
2198 * clear the uptodate bits on error
2199 * clear the writeback bits in the extent tree for this IO
2200 * end_page_writeback if the page has no more pending IO
2201 *
2202 * Scheduling is not allowed, so the extent state tree is expected
2203 * to have one and only one object corresponding to this IO.
2204 */
2205 static void end_bio_extent_writepage(struct bio *bio, int err)
2206 {
2207 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2208 struct extent_io_tree *tree;
2209 u64 start;
2210 u64 end;
2211 int whole_page;
2212
2213 do {
2214 struct page *page = bvec->bv_page;
2215 tree = &BTRFS_I(page->mapping->host)->io_tree;
2216
2217 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2218 bvec->bv_offset;
2219 end = start + bvec->bv_len - 1;
2220
2221 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2222 whole_page = 1;
2223 else
2224 whole_page = 0;
2225
2226 if (--bvec >= bio->bi_io_vec)
2227 prefetchw(&bvec->bv_page->flags);
2228
2229 if (end_extent_writepage(page, err, start, end))
2230 continue;
2231
2232 if (whole_page)
2233 end_page_writeback(page);
2234 else
2235 check_page_writeback(tree, page);
2236 } while (bvec >= bio->bi_io_vec);
2237
2238 bio_put(bio);
2239 }
2240
2241 /*
2242 * after a readpage IO is done, we need to:
2243 * clear the uptodate bits on error
2244 * set the uptodate bits if things worked
2245 * set the page up to date if all extents in the tree are uptodate
2246 * clear the lock bit in the extent tree
2247 * unlock the page if there are no other extents locked for it
2248 *
2249 * Scheduling is not allowed, so the extent state tree is expected
2250 * to have one and only one object corresponding to this IO.
2251 */
2252 static void end_bio_extent_readpage(struct bio *bio, int err)
2253 {
2254 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2255 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2256 struct bio_vec *bvec = bio->bi_io_vec;
2257 struct extent_io_tree *tree;
2258 u64 start;
2259 u64 end;
2260 int whole_page;
2261 int ret;
2262
2263 if (err)
2264 uptodate = 0;
2265
2266 do {
2267 struct page *page = bvec->bv_page;
2268 struct extent_state *cached = NULL;
2269 struct extent_state *state;
2270
2271 pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, "
2272 "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err,
2273 (long int)bio->bi_bdev);
2274 tree = &BTRFS_I(page->mapping->host)->io_tree;
2275
2276 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
2277 bvec->bv_offset;
2278 end = start + bvec->bv_len - 1;
2279
2280 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
2281 whole_page = 1;
2282 else
2283 whole_page = 0;
2284
2285 if (++bvec <= bvec_end)
2286 prefetchw(&bvec->bv_page->flags);
2287
2288 spin_lock(&tree->lock);
2289 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2290 if (state && state->start == start) {
2291 /*
2292 * take a reference on the state, unlock will drop
2293 * the ref
2294 */
2295 cache_state(state, &cached);
2296 }
2297 spin_unlock(&tree->lock);
2298
2299 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2300 ret = tree->ops->readpage_end_io_hook(page, start, end,
2301 state);
2302 if (ret)
2303 uptodate = 0;
2304 else
2305 clean_io_failure(start, page);
2306 }
2307 if (!uptodate) {
2308 int failed_mirror;
2309 failed_mirror = (int)(unsigned long)bio->bi_bdev;
2310 /*
2311 * The generic bio_readpage_error handles errors the
2312 * following way: If possible, new read requests are
2313 * created and submitted and will end up in
2314 * end_bio_extent_readpage as well (if we're lucky, not
2315 * in the !uptodate case). In that case it returns 0 and
2316 * we just go on with the next page in our bio. If it
2317 * can't handle the error it will return -EIO and we
2318 * remain responsible for that page.
2319 */
2320 ret = bio_readpage_error(bio, page, start, end,
2321 failed_mirror, NULL);
2322 if (ret == 0) {
2323 error_handled:
2324 uptodate =
2325 test_bit(BIO_UPTODATE, &bio->bi_flags);
2326 if (err)
2327 uptodate = 0;
2328 uncache_state(&cached);
2329 continue;
2330 }
2331 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2332 ret = tree->ops->readpage_io_failed_hook(
2333 bio, page, start, end,
2334 failed_mirror, state);
2335 if (ret == 0)
2336 goto error_handled;
2337 }
2338 }
2339
2340 if (uptodate) {
2341 set_extent_uptodate(tree, start, end, &cached,
2342 GFP_ATOMIC);
2343 }
2344 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2345
2346 if (whole_page) {
2347 if (uptodate) {
2348 SetPageUptodate(page);
2349 } else {
2350 ClearPageUptodate(page);
2351 SetPageError(page);
2352 }
2353 unlock_page(page);
2354 } else {
2355 if (uptodate) {
2356 check_page_uptodate(tree, page);
2357 } else {
2358 ClearPageUptodate(page);
2359 SetPageError(page);
2360 }
2361 check_page_locked(tree, page);
2362 }
2363 } while (bvec <= bvec_end);
2364
2365 bio_put(bio);
2366 }
2367
2368 struct bio *
2369 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2370 gfp_t gfp_flags)
2371 {
2372 struct bio *bio;
2373
2374 bio = bio_alloc(gfp_flags, nr_vecs);
2375
2376 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2377 while (!bio && (nr_vecs /= 2))
2378 bio = bio_alloc(gfp_flags, nr_vecs);
2379 }
2380
2381 if (bio) {
2382 bio->bi_size = 0;
2383 bio->bi_bdev = bdev;
2384 bio->bi_sector = first_sector;
2385 }
2386 return bio;
2387 }
2388
2389 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
2390 unsigned long bio_flags)
2391 {
2392 int ret = 0;
2393 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2394 struct page *page = bvec->bv_page;
2395 struct extent_io_tree *tree = bio->bi_private;
2396 u64 start;
2397
2398 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
2399
2400 bio->bi_private = NULL;
2401
2402 bio_get(bio);
2403
2404 if (tree->ops && tree->ops->submit_bio_hook)
2405 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2406 mirror_num, bio_flags, start);
2407 else
2408 btrfsic_submit_bio(rw, bio);
2409
2410 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2411 ret = -EOPNOTSUPP;
2412 bio_put(bio);
2413 return ret;
2414 }
2415
2416 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2417 struct page *page, sector_t sector,
2418 size_t size, unsigned long offset,
2419 struct block_device *bdev,
2420 struct bio **bio_ret,
2421 unsigned long max_pages,
2422 bio_end_io_t end_io_func,
2423 int mirror_num,
2424 unsigned long prev_bio_flags,
2425 unsigned long bio_flags)
2426 {
2427 int ret = 0;
2428 struct bio *bio;
2429 int nr;
2430 int contig = 0;
2431 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2432 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2433 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2434
2435 if (bio_ret && *bio_ret) {
2436 bio = *bio_ret;
2437 if (old_compressed)
2438 contig = bio->bi_sector == sector;
2439 else
2440 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2441 sector;
2442
2443 if (prev_bio_flags != bio_flags || !contig ||
2444 (tree->ops && tree->ops->merge_bio_hook &&
2445 tree->ops->merge_bio_hook(page, offset, page_size, bio,
2446 bio_flags)) ||
2447 bio_add_page(bio, page, page_size, offset) < page_size) {
2448 ret = submit_one_bio(rw, bio, mirror_num,
2449 prev_bio_flags);
2450 bio = NULL;
2451 } else {
2452 return 0;
2453 }
2454 }
2455 if (this_compressed)
2456 nr = BIO_MAX_PAGES;
2457 else
2458 nr = bio_get_nr_vecs(bdev);
2459
2460 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2461 if (!bio)
2462 return -ENOMEM;
2463
2464 bio_add_page(bio, page, page_size, offset);
2465 bio->bi_end_io = end_io_func;
2466 bio->bi_private = tree;
2467
2468 if (bio_ret)
2469 *bio_ret = bio;
2470 else
2471 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2472
2473 return ret;
2474 }
2475
2476 void set_page_extent_mapped(struct page *page)
2477 {
2478 if (!PagePrivate(page)) {
2479 SetPagePrivate(page);
2480 page_cache_get(page);
2481 set_page_private(page, EXTENT_PAGE_PRIVATE);
2482 }
2483 }
2484
2485 static void set_page_extent_head(struct page *page, unsigned long len)
2486 {
2487 WARN_ON(!PagePrivate(page));
2488 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
2489 }
2490
2491 /*
2492 * basic readpage implementation. Locked extent state structs are inserted
2493 * into the tree that are removed when the IO is done (by the end_io
2494 * handlers)
2495 */
2496 static int __extent_read_full_page(struct extent_io_tree *tree,
2497 struct page *page,
2498 get_extent_t *get_extent,
2499 struct bio **bio, int mirror_num,
2500 unsigned long *bio_flags)
2501 {
2502 struct inode *inode = page->mapping->host;
2503 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2504 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2505 u64 end;
2506 u64 cur = start;
2507 u64 extent_offset;
2508 u64 last_byte = i_size_read(inode);
2509 u64 block_start;
2510 u64 cur_end;
2511 sector_t sector;
2512 struct extent_map *em;
2513 struct block_device *bdev;
2514 struct btrfs_ordered_extent *ordered;
2515 int ret;
2516 int nr = 0;
2517 size_t pg_offset = 0;
2518 size_t iosize;
2519 size_t disk_io_size;
2520 size_t blocksize = inode->i_sb->s_blocksize;
2521 unsigned long this_bio_flag = 0;
2522
2523 set_page_extent_mapped(page);
2524
2525 if (!PageUptodate(page)) {
2526 if (cleancache_get_page(page) == 0) {
2527 BUG_ON(blocksize != PAGE_SIZE);
2528 goto out;
2529 }
2530 }
2531
2532 end = page_end;
2533 while (1) {
2534 lock_extent(tree, start, end, GFP_NOFS);
2535 ordered = btrfs_lookup_ordered_extent(inode, start);
2536 if (!ordered)
2537 break;
2538 unlock_extent(tree, start, end, GFP_NOFS);
2539 btrfs_start_ordered_extent(inode, ordered, 1);
2540 btrfs_put_ordered_extent(ordered);
2541 }
2542
2543 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2544 char *userpage;
2545 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2546
2547 if (zero_offset) {
2548 iosize = PAGE_CACHE_SIZE - zero_offset;
2549 userpage = kmap_atomic(page);
2550 memset(userpage + zero_offset, 0, iosize);
2551 flush_dcache_page(page);
2552 kunmap_atomic(userpage);
2553 }
2554 }
2555 while (cur <= end) {
2556 if (cur >= last_byte) {
2557 char *userpage;
2558 struct extent_state *cached = NULL;
2559
2560 iosize = PAGE_CACHE_SIZE - pg_offset;
2561 userpage = kmap_atomic(page);
2562 memset(userpage + pg_offset, 0, iosize);
2563 flush_dcache_page(page);
2564 kunmap_atomic(userpage);
2565 set_extent_uptodate(tree, cur, cur + iosize - 1,
2566 &cached, GFP_NOFS);
2567 unlock_extent_cached(tree, cur, cur + iosize - 1,
2568 &cached, GFP_NOFS);
2569 break;
2570 }
2571 em = get_extent(inode, page, pg_offset, cur,
2572 end - cur + 1, 0);
2573 if (IS_ERR_OR_NULL(em)) {
2574 SetPageError(page);
2575 unlock_extent(tree, cur, end, GFP_NOFS);
2576 break;
2577 }
2578 extent_offset = cur - em->start;
2579 BUG_ON(extent_map_end(em) <= cur);
2580 BUG_ON(end < cur);
2581
2582 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2583 this_bio_flag = EXTENT_BIO_COMPRESSED;
2584 extent_set_compress_type(&this_bio_flag,
2585 em->compress_type);
2586 }
2587
2588 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2589 cur_end = min(extent_map_end(em) - 1, end);
2590 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2591 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2592 disk_io_size = em->block_len;
2593 sector = em->block_start >> 9;
2594 } else {
2595 sector = (em->block_start + extent_offset) >> 9;
2596 disk_io_size = iosize;
2597 }
2598 bdev = em->bdev;
2599 block_start = em->block_start;
2600 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2601 block_start = EXTENT_MAP_HOLE;
2602 free_extent_map(em);
2603 em = NULL;
2604
2605 /* we've found a hole, just zero and go on */
2606 if (block_start == EXTENT_MAP_HOLE) {
2607 char *userpage;
2608 struct extent_state *cached = NULL;
2609
2610 userpage = kmap_atomic(page);
2611 memset(userpage + pg_offset, 0, iosize);
2612 flush_dcache_page(page);
2613 kunmap_atomic(userpage);
2614
2615 set_extent_uptodate(tree, cur, cur + iosize - 1,
2616 &cached, GFP_NOFS);
2617 unlock_extent_cached(tree, cur, cur + iosize - 1,
2618 &cached, GFP_NOFS);
2619 cur = cur + iosize;
2620 pg_offset += iosize;
2621 continue;
2622 }
2623 /* the get_extent function already copied into the page */
2624 if (test_range_bit(tree, cur, cur_end,
2625 EXTENT_UPTODATE, 1, NULL)) {
2626 check_page_uptodate(tree, page);
2627 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2628 cur = cur + iosize;
2629 pg_offset += iosize;
2630 continue;
2631 }
2632 /* we have an inline extent but it didn't get marked up
2633 * to date. Error out
2634 */
2635 if (block_start == EXTENT_MAP_INLINE) {
2636 SetPageError(page);
2637 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2638 cur = cur + iosize;
2639 pg_offset += iosize;
2640 continue;
2641 }
2642
2643 ret = 0;
2644 if (tree->ops && tree->ops->readpage_io_hook) {
2645 ret = tree->ops->readpage_io_hook(page, cur,
2646 cur + iosize - 1);
2647 }
2648 if (!ret) {
2649 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2650 pnr -= page->index;
2651 ret = submit_extent_page(READ, tree, page,
2652 sector, disk_io_size, pg_offset,
2653 bdev, bio, pnr,
2654 end_bio_extent_readpage, mirror_num,
2655 *bio_flags,
2656 this_bio_flag);
2657 nr++;
2658 *bio_flags = this_bio_flag;
2659 }
2660 if (ret)
2661 SetPageError(page);
2662 cur = cur + iosize;
2663 pg_offset += iosize;
2664 }
2665 out:
2666 if (!nr) {
2667 if (!PageError(page))
2668 SetPageUptodate(page);
2669 unlock_page(page);
2670 }
2671 return 0;
2672 }
2673
2674 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2675 get_extent_t *get_extent, int mirror_num)
2676 {
2677 struct bio *bio = NULL;
2678 unsigned long bio_flags = 0;
2679 int ret;
2680
2681 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2682 &bio_flags);
2683 if (bio)
2684 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2685 return ret;
2686 }
2687
2688 static noinline void update_nr_written(struct page *page,
2689 struct writeback_control *wbc,
2690 unsigned long nr_written)
2691 {
2692 wbc->nr_to_write -= nr_written;
2693 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2694 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2695 page->mapping->writeback_index = page->index + nr_written;
2696 }
2697
2698 /*
2699 * the writepage semantics are similar to regular writepage. extent
2700 * records are inserted to lock ranges in the tree, and as dirty areas
2701 * are found, they are marked writeback. Then the lock bits are removed
2702 * and the end_io handler clears the writeback ranges
2703 */
2704 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2705 void *data)
2706 {
2707 struct inode *inode = page->mapping->host;
2708 struct extent_page_data *epd = data;
2709 struct extent_io_tree *tree = epd->tree;
2710 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2711 u64 delalloc_start;
2712 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2713 u64 end;
2714 u64 cur = start;
2715 u64 extent_offset;
2716 u64 last_byte = i_size_read(inode);
2717 u64 block_start;
2718 u64 iosize;
2719 sector_t sector;
2720 struct extent_state *cached_state = NULL;
2721 struct extent_map *em;
2722 struct block_device *bdev;
2723 int ret;
2724 int nr = 0;
2725 size_t pg_offset = 0;
2726 size_t blocksize;
2727 loff_t i_size = i_size_read(inode);
2728 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2729 u64 nr_delalloc;
2730 u64 delalloc_end;
2731 int page_started;
2732 int compressed;
2733 int write_flags;
2734 unsigned long nr_written = 0;
2735 bool fill_delalloc = true;
2736
2737 if (wbc->sync_mode == WB_SYNC_ALL)
2738 write_flags = WRITE_SYNC;
2739 else
2740 write_flags = WRITE;
2741
2742 trace___extent_writepage(page, inode, wbc);
2743
2744 WARN_ON(!PageLocked(page));
2745
2746 ClearPageError(page);
2747
2748 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2749 if (page->index > end_index ||
2750 (page->index == end_index && !pg_offset)) {
2751 page->mapping->a_ops->invalidatepage(page, 0);
2752 unlock_page(page);
2753 return 0;
2754 }
2755
2756 if (page->index == end_index) {
2757 char *userpage;
2758
2759 userpage = kmap_atomic(page);
2760 memset(userpage + pg_offset, 0,
2761 PAGE_CACHE_SIZE - pg_offset);
2762 kunmap_atomic(userpage);
2763 flush_dcache_page(page);
2764 }
2765 pg_offset = 0;
2766
2767 set_page_extent_mapped(page);
2768
2769 if (!tree->ops || !tree->ops->fill_delalloc)
2770 fill_delalloc = false;
2771
2772 delalloc_start = start;
2773 delalloc_end = 0;
2774 page_started = 0;
2775 if (!epd->extent_locked && fill_delalloc) {
2776 u64 delalloc_to_write = 0;
2777 /*
2778 * make sure the wbc mapping index is at least updated
2779 * to this page.
2780 */
2781 update_nr_written(page, wbc, 0);
2782
2783 while (delalloc_end < page_end) {
2784 nr_delalloc = find_lock_delalloc_range(inode, tree,
2785 page,
2786 &delalloc_start,
2787 &delalloc_end,
2788 128 * 1024 * 1024);
2789 if (nr_delalloc == 0) {
2790 delalloc_start = delalloc_end + 1;
2791 continue;
2792 }
2793 ret = tree->ops->fill_delalloc(inode, page,
2794 delalloc_start,
2795 delalloc_end,
2796 &page_started,
2797 &nr_written);
2798 BUG_ON(ret);
2799 /*
2800 * delalloc_end is already one less than the total
2801 * length, so we don't subtract one from
2802 * PAGE_CACHE_SIZE
2803 */
2804 delalloc_to_write += (delalloc_end - delalloc_start +
2805 PAGE_CACHE_SIZE) >>
2806 PAGE_CACHE_SHIFT;
2807 delalloc_start = delalloc_end + 1;
2808 }
2809 if (wbc->nr_to_write < delalloc_to_write) {
2810 int thresh = 8192;
2811
2812 if (delalloc_to_write < thresh * 2)
2813 thresh = delalloc_to_write;
2814 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2815 thresh);
2816 }
2817
2818 /* did the fill delalloc function already unlock and start
2819 * the IO?
2820 */
2821 if (page_started) {
2822 ret = 0;
2823 /*
2824 * we've unlocked the page, so we can't update
2825 * the mapping's writeback index, just update
2826 * nr_to_write.
2827 */
2828 wbc->nr_to_write -= nr_written;
2829 goto done_unlocked;
2830 }
2831 }
2832 if (tree->ops && tree->ops->writepage_start_hook) {
2833 ret = tree->ops->writepage_start_hook(page, start,
2834 page_end);
2835 if (ret) {
2836 /* Fixup worker will requeue */
2837 if (ret == -EBUSY)
2838 wbc->pages_skipped++;
2839 else
2840 redirty_page_for_writepage(wbc, page);
2841 update_nr_written(page, wbc, nr_written);
2842 unlock_page(page);
2843 ret = 0;
2844 goto done_unlocked;
2845 }
2846 }
2847
2848 /*
2849 * we don't want to touch the inode after unlocking the page,
2850 * so we update the mapping writeback index now
2851 */
2852 update_nr_written(page, wbc, nr_written + 1);
2853
2854 end = page_end;
2855 if (last_byte <= start) {
2856 if (tree->ops && tree->ops->writepage_end_io_hook)
2857 tree->ops->writepage_end_io_hook(page, start,
2858 page_end, NULL, 1);
2859 goto done;
2860 }
2861
2862 blocksize = inode->i_sb->s_blocksize;
2863
2864 while (cur <= end) {
2865 if (cur >= last_byte) {
2866 if (tree->ops && tree->ops->writepage_end_io_hook)
2867 tree->ops->writepage_end_io_hook(page, cur,
2868 page_end, NULL, 1);
2869 break;
2870 }
2871 em = epd->get_extent(inode, page, pg_offset, cur,
2872 end - cur + 1, 1);
2873 if (IS_ERR_OR_NULL(em)) {
2874 SetPageError(page);
2875 break;
2876 }
2877
2878 extent_offset = cur - em->start;
2879 BUG_ON(extent_map_end(em) <= cur);
2880 BUG_ON(end < cur);
2881 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2882 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2883 sector = (em->block_start + extent_offset) >> 9;
2884 bdev = em->bdev;
2885 block_start = em->block_start;
2886 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2887 free_extent_map(em);
2888 em = NULL;
2889
2890 /*
2891 * compressed and inline extents are written through other
2892 * paths in the FS
2893 */
2894 if (compressed || block_start == EXTENT_MAP_HOLE ||
2895 block_start == EXTENT_MAP_INLINE) {
2896 /*
2897 * end_io notification does not happen here for
2898 * compressed extents
2899 */
2900 if (!compressed && tree->ops &&
2901 tree->ops->writepage_end_io_hook)
2902 tree->ops->writepage_end_io_hook(page, cur,
2903 cur + iosize - 1,
2904 NULL, 1);
2905 else if (compressed) {
2906 /* we don't want to end_page_writeback on
2907 * a compressed extent. this happens
2908 * elsewhere
2909 */
2910 nr++;
2911 }
2912
2913 cur += iosize;
2914 pg_offset += iosize;
2915 continue;
2916 }
2917 /* leave this out until we have a page_mkwrite call */
2918 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2919 EXTENT_DIRTY, 0, NULL)) {
2920 cur = cur + iosize;
2921 pg_offset += iosize;
2922 continue;
2923 }
2924
2925 if (tree->ops && tree->ops->writepage_io_hook) {
2926 ret = tree->ops->writepage_io_hook(page, cur,
2927 cur + iosize - 1);
2928 } else {
2929 ret = 0;
2930 }
2931 if (ret) {
2932 SetPageError(page);
2933 } else {
2934 unsigned long max_nr = end_index + 1;
2935
2936 set_range_writeback(tree, cur, cur + iosize - 1);
2937 if (!PageWriteback(page)) {
2938 printk(KERN_ERR "btrfs warning page %lu not "
2939 "writeback, cur %llu end %llu\n",
2940 page->index, (unsigned long long)cur,
2941 (unsigned long long)end);
2942 }
2943
2944 ret = submit_extent_page(write_flags, tree, page,
2945 sector, iosize, pg_offset,
2946 bdev, &epd->bio, max_nr,
2947 end_bio_extent_writepage,
2948 0, 0, 0);
2949 if (ret)
2950 SetPageError(page);
2951 }
2952 cur = cur + iosize;
2953 pg_offset += iosize;
2954 nr++;
2955 }
2956 done:
2957 if (nr == 0) {
2958 /* make sure the mapping tag for page dirty gets cleared */
2959 set_page_writeback(page);
2960 end_page_writeback(page);
2961 }
2962 unlock_page(page);
2963
2964 done_unlocked:
2965
2966 /* drop our reference on any cached states */
2967 free_extent_state(cached_state);
2968 return 0;
2969 }
2970
2971 /**
2972 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2973 * @mapping: address space structure to write
2974 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2975 * @writepage: function called for each page
2976 * @data: data passed to writepage function
2977 *
2978 * If a page is already under I/O, write_cache_pages() skips it, even
2979 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2980 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2981 * and msync() need to guarantee that all the data which was dirty at the time
2982 * the call was made get new I/O started against them. If wbc->sync_mode is
2983 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2984 * existing IO to complete.
2985 */
2986 static int extent_write_cache_pages(struct extent_io_tree *tree,
2987 struct address_space *mapping,
2988 struct writeback_control *wbc,
2989 writepage_t writepage, void *data,
2990 void (*flush_fn)(void *))
2991 {
2992 int ret = 0;
2993 int done = 0;
2994 int nr_to_write_done = 0;
2995 struct pagevec pvec;
2996 int nr_pages;
2997 pgoff_t index;
2998 pgoff_t end; /* Inclusive */
2999 int scanned = 0;
3000 int tag;
3001
3002 pagevec_init(&pvec, 0);
3003 if (wbc->range_cyclic) {
3004 index = mapping->writeback_index; /* Start from prev offset */
3005 end = -1;
3006 } else {
3007 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3008 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3009 scanned = 1;
3010 }
3011 if (wbc->sync_mode == WB_SYNC_ALL)
3012 tag = PAGECACHE_TAG_TOWRITE;
3013 else
3014 tag = PAGECACHE_TAG_DIRTY;
3015 retry:
3016 if (wbc->sync_mode == WB_SYNC_ALL)
3017 tag_pages_for_writeback(mapping, index, end);
3018 while (!done && !nr_to_write_done && (index <= end) &&
3019 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3020 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3021 unsigned i;
3022
3023 scanned = 1;
3024 for (i = 0; i < nr_pages; i++) {
3025 struct page *page = pvec.pages[i];
3026
3027 /*
3028 * At this point we hold neither mapping->tree_lock nor
3029 * lock on the page itself: the page may be truncated or
3030 * invalidated (changing page->mapping to NULL), or even
3031 * swizzled back from swapper_space to tmpfs file
3032 * mapping
3033 */
3034 if (tree->ops &&
3035 tree->ops->write_cache_pages_lock_hook) {
3036 tree->ops->write_cache_pages_lock_hook(page,
3037 data, flush_fn);
3038 } else {
3039 if (!trylock_page(page)) {
3040 flush_fn(data);
3041 lock_page(page);
3042 }
3043 }
3044
3045 if (unlikely(page->mapping != mapping)) {
3046 unlock_page(page);
3047 continue;
3048 }
3049
3050 if (!wbc->range_cyclic && page->index > end) {
3051 done = 1;
3052 unlock_page(page);
3053 continue;
3054 }
3055
3056 if (wbc->sync_mode != WB_SYNC_NONE) {
3057 if (PageWriteback(page))
3058 flush_fn(data);
3059 wait_on_page_writeback(page);
3060 }
3061
3062 if (PageWriteback(page) ||
3063 !clear_page_dirty_for_io(page)) {
3064 unlock_page(page);
3065 continue;
3066 }
3067
3068 ret = (*writepage)(page, wbc, data);
3069
3070 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3071 unlock_page(page);
3072 ret = 0;
3073 }
3074 if (ret)
3075 done = 1;
3076
3077 /*
3078 * the filesystem may choose to bump up nr_to_write.
3079 * We have to make sure to honor the new nr_to_write
3080 * at any time
3081 */
3082 nr_to_write_done = wbc->nr_to_write <= 0;
3083 }
3084 pagevec_release(&pvec);
3085 cond_resched();
3086 }
3087 if (!scanned && !done) {
3088 /*
3089 * We hit the last page and there is more work to be done: wrap
3090 * back to the start of the file
3091 */
3092 scanned = 1;
3093 index = 0;
3094 goto retry;
3095 }
3096 return ret;
3097 }
3098
3099 static void flush_epd_write_bio(struct extent_page_data *epd)
3100 {
3101 if (epd->bio) {
3102 if (epd->sync_io)
3103 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
3104 else
3105 submit_one_bio(WRITE, epd->bio, 0, 0);
3106 epd->bio = NULL;
3107 }
3108 }
3109
3110 static noinline void flush_write_bio(void *data)
3111 {
3112 struct extent_page_data *epd = data;
3113 flush_epd_write_bio(epd);
3114 }
3115
3116 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3117 get_extent_t *get_extent,
3118 struct writeback_control *wbc)
3119 {
3120 int ret;
3121 struct extent_page_data epd = {
3122 .bio = NULL,
3123 .tree = tree,
3124 .get_extent = get_extent,
3125 .extent_locked = 0,
3126 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3127 };
3128
3129 ret = __extent_writepage(page, wbc, &epd);
3130
3131 flush_epd_write_bio(&epd);
3132 return ret;
3133 }
3134
3135 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3136 u64 start, u64 end, get_extent_t *get_extent,
3137 int mode)
3138 {
3139 int ret = 0;
3140 struct address_space *mapping = inode->i_mapping;
3141 struct page *page;
3142 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3143 PAGE_CACHE_SHIFT;
3144
3145 struct extent_page_data epd = {
3146 .bio = NULL,
3147 .tree = tree,
3148 .get_extent = get_extent,
3149 .extent_locked = 1,
3150 .sync_io = mode == WB_SYNC_ALL,
3151 };
3152 struct writeback_control wbc_writepages = {
3153 .sync_mode = mode,
3154 .nr_to_write = nr_pages * 2,
3155 .range_start = start,
3156 .range_end = end + 1,
3157 };
3158
3159 while (start <= end) {
3160 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3161 if (clear_page_dirty_for_io(page))
3162 ret = __extent_writepage(page, &wbc_writepages, &epd);
3163 else {
3164 if (tree->ops && tree->ops->writepage_end_io_hook)
3165 tree->ops->writepage_end_io_hook(page, start,
3166 start + PAGE_CACHE_SIZE - 1,
3167 NULL, 1);
3168 unlock_page(page);
3169 }
3170 page_cache_release(page);
3171 start += PAGE_CACHE_SIZE;
3172 }
3173
3174 flush_epd_write_bio(&epd);
3175 return ret;
3176 }
3177
3178 int extent_writepages(struct extent_io_tree *tree,
3179 struct address_space *mapping,
3180 get_extent_t *get_extent,
3181 struct writeback_control *wbc)
3182 {
3183 int ret = 0;
3184 struct extent_page_data epd = {
3185 .bio = NULL,
3186 .tree = tree,
3187 .get_extent = get_extent,
3188 .extent_locked = 0,
3189 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3190 };
3191
3192 ret = extent_write_cache_pages(tree, mapping, wbc,
3193 __extent_writepage, &epd,
3194 flush_write_bio);
3195 flush_epd_write_bio(&epd);
3196 return ret;
3197 }
3198
3199 int extent_readpages(struct extent_io_tree *tree,
3200 struct address_space *mapping,
3201 struct list_head *pages, unsigned nr_pages,
3202 get_extent_t get_extent)
3203 {
3204 struct bio *bio = NULL;
3205 unsigned page_idx;
3206 unsigned long bio_flags = 0;
3207
3208 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3209 struct page *page = list_entry(pages->prev, struct page, lru);
3210
3211 prefetchw(&page->flags);
3212 list_del(&page->lru);
3213 if (!add_to_page_cache_lru(page, mapping,
3214 page->index, GFP_NOFS)) {
3215 __extent_read_full_page(tree, page, get_extent,
3216 &bio, 0, &bio_flags);
3217 }
3218 page_cache_release(page);
3219 }
3220 BUG_ON(!list_empty(pages));
3221 if (bio)
3222 submit_one_bio(READ, bio, 0, bio_flags);
3223 return 0;
3224 }
3225
3226 /*
3227 * basic invalidatepage code, this waits on any locked or writeback
3228 * ranges corresponding to the page, and then deletes any extent state
3229 * records from the tree
3230 */
3231 int extent_invalidatepage(struct extent_io_tree *tree,
3232 struct page *page, unsigned long offset)
3233 {
3234 struct extent_state *cached_state = NULL;
3235 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
3236 u64 end = start + PAGE_CACHE_SIZE - 1;
3237 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3238
3239 start += (offset + blocksize - 1) & ~(blocksize - 1);
3240 if (start > end)
3241 return 0;
3242
3243 lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS);
3244 wait_on_page_writeback(page);
3245 clear_extent_bit(tree, start, end,
3246 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3247 EXTENT_DO_ACCOUNTING,
3248 1, 1, &cached_state, GFP_NOFS);
3249 return 0;
3250 }
3251
3252 /*
3253 * a helper for releasepage, this tests for areas of the page that
3254 * are locked or under IO and drops the related state bits if it is safe
3255 * to drop the page.
3256 */
3257 int try_release_extent_state(struct extent_map_tree *map,
3258 struct extent_io_tree *tree, struct page *page,
3259 gfp_t mask)
3260 {
3261 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3262 u64 end = start + PAGE_CACHE_SIZE - 1;
3263 int ret = 1;
3264
3265 if (test_range_bit(tree, start, end,
3266 EXTENT_IOBITS, 0, NULL))
3267 ret = 0;
3268 else {
3269 if ((mask & GFP_NOFS) == GFP_NOFS)
3270 mask = GFP_NOFS;
3271 /*
3272 * at this point we can safely clear everything except the
3273 * locked bit and the nodatasum bit
3274 */
3275 ret = clear_extent_bit(tree, start, end,
3276 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3277 0, 0, NULL, mask);
3278
3279 /* if clear_extent_bit failed for enomem reasons,
3280 * we can't allow the release to continue.
3281 */
3282 if (ret < 0)
3283 ret = 0;
3284 else
3285 ret = 1;
3286 }
3287 return ret;
3288 }
3289
3290 /*
3291 * a helper for releasepage. As long as there are no locked extents
3292 * in the range corresponding to the page, both state records and extent
3293 * map records are removed
3294 */
3295 int try_release_extent_mapping(struct extent_map_tree *map,
3296 struct extent_io_tree *tree, struct page *page,
3297 gfp_t mask)
3298 {
3299 struct extent_map *em;
3300 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
3301 u64 end = start + PAGE_CACHE_SIZE - 1;
3302
3303 if ((mask & __GFP_WAIT) &&
3304 page->mapping->host->i_size > 16 * 1024 * 1024) {
3305 u64 len;
3306 while (start <= end) {
3307 len = end - start + 1;
3308 write_lock(&map->lock);
3309 em = lookup_extent_mapping(map, start, len);
3310 if (!em) {
3311 write_unlock(&map->lock);
3312 break;
3313 }
3314 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3315 em->start != start) {
3316 write_unlock(&map->lock);
3317 free_extent_map(em);
3318 break;
3319 }
3320 if (!test_range_bit(tree, em->start,
3321 extent_map_end(em) - 1,
3322 EXTENT_LOCKED | EXTENT_WRITEBACK,
3323 0, NULL)) {
3324 remove_extent_mapping(map, em);
3325 /* once for the rb tree */
3326 free_extent_map(em);
3327 }
3328 start = extent_map_end(em);
3329 write_unlock(&map->lock);
3330
3331 /* once for us */
3332 free_extent_map(em);
3333 }
3334 }
3335 return try_release_extent_state(map, tree, page, mask);
3336 }
3337
3338 /*
3339 * helper function for fiemap, which doesn't want to see any holes.
3340 * This maps until we find something past 'last'
3341 */
3342 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3343 u64 offset,
3344 u64 last,
3345 get_extent_t *get_extent)
3346 {
3347 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3348 struct extent_map *em;
3349 u64 len;
3350
3351 if (offset >= last)
3352 return NULL;
3353
3354 while(1) {
3355 len = last - offset;
3356 if (len == 0)
3357 break;
3358 len = (len + sectorsize - 1) & ~(sectorsize - 1);
3359 em = get_extent(inode, NULL, 0, offset, len, 0);
3360 if (IS_ERR_OR_NULL(em))
3361 return em;
3362
3363 /* if this isn't a hole return it */
3364 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3365 em->block_start != EXTENT_MAP_HOLE) {
3366 return em;
3367 }
3368
3369 /* this is a hole, advance to the next extent */
3370 offset = extent_map_end(em);
3371 free_extent_map(em);
3372 if (offset >= last)
3373 break;
3374 }
3375 return NULL;
3376 }
3377
3378 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3379 __u64 start, __u64 len, get_extent_t *get_extent)
3380 {
3381 int ret = 0;
3382 u64 off = start;
3383 u64 max = start + len;
3384 u32 flags = 0;
3385 u32 found_type;
3386 u64 last;
3387 u64 last_for_get_extent = 0;
3388 u64 disko = 0;
3389 u64 isize = i_size_read(inode);
3390 struct btrfs_key found_key;
3391 struct extent_map *em = NULL;
3392 struct extent_state *cached_state = NULL;
3393 struct btrfs_path *path;
3394 struct btrfs_file_extent_item *item;
3395 int end = 0;
3396 u64 em_start = 0;
3397 u64 em_len = 0;
3398 u64 em_end = 0;
3399 unsigned long emflags;
3400
3401 if (len == 0)
3402 return -EINVAL;
3403
3404 path = btrfs_alloc_path();
3405 if (!path)
3406 return -ENOMEM;
3407 path->leave_spinning = 1;
3408
3409 start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3410 len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3411
3412 /*
3413 * lookup the last file extent. We're not using i_size here
3414 * because there might be preallocation past i_size
3415 */
3416 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3417 path, btrfs_ino(inode), -1, 0);
3418 if (ret < 0) {
3419 btrfs_free_path(path);
3420 return ret;
3421 }
3422 WARN_ON(!ret);
3423 path->slots[0]--;
3424 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3425 struct btrfs_file_extent_item);
3426 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3427 found_type = btrfs_key_type(&found_key);
3428
3429 /* No extents, but there might be delalloc bits */
3430 if (found_key.objectid != btrfs_ino(inode) ||
3431 found_type != BTRFS_EXTENT_DATA_KEY) {
3432 /* have to trust i_size as the end */
3433 last = (u64)-1;
3434 last_for_get_extent = isize;
3435 } else {
3436 /*
3437 * remember the start of the last extent. There are a
3438 * bunch of different factors that go into the length of the
3439 * extent, so its much less complex to remember where it started
3440 */
3441 last = found_key.offset;
3442 last_for_get_extent = last + 1;
3443 }
3444 btrfs_free_path(path);
3445
3446 /*
3447 * we might have some extents allocated but more delalloc past those
3448 * extents. so, we trust isize unless the start of the last extent is
3449 * beyond isize
3450 */
3451 if (last < isize) {
3452 last = (u64)-1;
3453 last_for_get_extent = isize;
3454 }
3455
3456 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3457 &cached_state, GFP_NOFS);
3458
3459 em = get_extent_skip_holes(inode, start, last_for_get_extent,
3460 get_extent);
3461 if (!em)
3462 goto out;
3463 if (IS_ERR(em)) {
3464 ret = PTR_ERR(em);
3465 goto out;
3466 }
3467
3468 while (!end) {
3469 u64 offset_in_extent;
3470
3471 /* break if the extent we found is outside the range */
3472 if (em->start >= max || extent_map_end(em) < off)
3473 break;
3474
3475 /*
3476 * get_extent may return an extent that starts before our
3477 * requested range. We have to make sure the ranges
3478 * we return to fiemap always move forward and don't
3479 * overlap, so adjust the offsets here
3480 */
3481 em_start = max(em->start, off);
3482
3483 /*
3484 * record the offset from the start of the extent
3485 * for adjusting the disk offset below
3486 */
3487 offset_in_extent = em_start - em->start;
3488 em_end = extent_map_end(em);
3489 em_len = em_end - em_start;
3490 emflags = em->flags;
3491 disko = 0;
3492 flags = 0;
3493
3494 /*
3495 * bump off for our next call to get_extent
3496 */
3497 off = extent_map_end(em);
3498 if (off >= max)
3499 end = 1;
3500
3501 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3502 end = 1;
3503 flags |= FIEMAP_EXTENT_LAST;
3504 } else if (em->block_start == EXTENT_MAP_INLINE) {
3505 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3506 FIEMAP_EXTENT_NOT_ALIGNED);
3507 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3508 flags |= (FIEMAP_EXTENT_DELALLOC |
3509 FIEMAP_EXTENT_UNKNOWN);
3510 } else {
3511 disko = em->block_start + offset_in_extent;
3512 }
3513 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3514 flags |= FIEMAP_EXTENT_ENCODED;
3515
3516 free_extent_map(em);
3517 em = NULL;
3518 if ((em_start >= last) || em_len == (u64)-1 ||
3519 (last == (u64)-1 && isize <= em_end)) {
3520 flags |= FIEMAP_EXTENT_LAST;
3521 end = 1;
3522 }
3523
3524 /* now scan forward to see if this is really the last extent. */
3525 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3526 get_extent);
3527 if (IS_ERR(em)) {
3528 ret = PTR_ERR(em);
3529 goto out;
3530 }
3531 if (!em) {
3532 flags |= FIEMAP_EXTENT_LAST;
3533 end = 1;
3534 }
3535 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3536 em_len, flags);
3537 if (ret)
3538 goto out_free;
3539 }
3540 out_free:
3541 free_extent_map(em);
3542 out:
3543 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3544 &cached_state, GFP_NOFS);
3545 return ret;
3546 }
3547
3548 inline struct page *extent_buffer_page(struct extent_buffer *eb,
3549 unsigned long i)
3550 {
3551 struct page *p;
3552 struct address_space *mapping;
3553
3554 if (i == 0)
3555 return eb->first_page;
3556 i += eb->start >> PAGE_CACHE_SHIFT;
3557 mapping = eb->first_page->mapping;
3558 if (!mapping)
3559 return NULL;
3560
3561 /*
3562 * extent_buffer_page is only called after pinning the page
3563 * by increasing the reference count. So we know the page must
3564 * be in the radix tree.
3565 */
3566 rcu_read_lock();
3567 p = radix_tree_lookup(&mapping->page_tree, i);
3568 rcu_read_unlock();
3569
3570 return p;
3571 }
3572
3573 inline unsigned long num_extent_pages(u64 start, u64 len)
3574 {
3575 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3576 (start >> PAGE_CACHE_SHIFT);
3577 }
3578
3579 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3580 u64 start,
3581 unsigned long len,
3582 gfp_t mask)
3583 {
3584 struct extent_buffer *eb = NULL;
3585 #if LEAK_DEBUG
3586 unsigned long flags;
3587 #endif
3588
3589 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3590 if (eb == NULL)
3591 return NULL;
3592 eb->start = start;
3593 eb->len = len;
3594 rwlock_init(&eb->lock);
3595 atomic_set(&eb->write_locks, 0);
3596 atomic_set(&eb->read_locks, 0);
3597 atomic_set(&eb->blocking_readers, 0);
3598 atomic_set(&eb->blocking_writers, 0);
3599 atomic_set(&eb->spinning_readers, 0);
3600 atomic_set(&eb->spinning_writers, 0);
3601 eb->lock_nested = 0;
3602 init_waitqueue_head(&eb->write_lock_wq);
3603 init_waitqueue_head(&eb->read_lock_wq);
3604
3605 #if LEAK_DEBUG
3606 spin_lock_irqsave(&leak_lock, flags);
3607 list_add(&eb->leak_list, &buffers);
3608 spin_unlock_irqrestore(&leak_lock, flags);
3609 #endif
3610 atomic_set(&eb->refs, 1);
3611
3612 return eb;
3613 }
3614
3615 static void __free_extent_buffer(struct extent_buffer *eb)
3616 {
3617 #if LEAK_DEBUG
3618 unsigned long flags;
3619 spin_lock_irqsave(&leak_lock, flags);
3620 list_del(&eb->leak_list);
3621 spin_unlock_irqrestore(&leak_lock, flags);
3622 #endif
3623 kmem_cache_free(extent_buffer_cache, eb);
3624 }
3625
3626 /*
3627 * Helper for releasing extent buffer page.
3628 */
3629 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3630 unsigned long start_idx)
3631 {
3632 unsigned long index;
3633 struct page *page;
3634
3635 if (!eb->first_page)
3636 return;
3637
3638 index = num_extent_pages(eb->start, eb->len);
3639 if (start_idx >= index)
3640 return;
3641
3642 do {
3643 index--;
3644 page = extent_buffer_page(eb, index);
3645 if (page)
3646 page_cache_release(page);
3647 } while (index != start_idx);
3648 }
3649
3650 /*
3651 * Helper for releasing the extent buffer.
3652 */
3653 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3654 {
3655 btrfs_release_extent_buffer_page(eb, 0);
3656 __free_extent_buffer(eb);
3657 }
3658
3659 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3660 u64 start, unsigned long len,
3661 struct page *page0)
3662 {
3663 unsigned long num_pages = num_extent_pages(start, len);
3664 unsigned long i;
3665 unsigned long index = start >> PAGE_CACHE_SHIFT;
3666 struct extent_buffer *eb;
3667 struct extent_buffer *exists = NULL;
3668 struct page *p;
3669 struct address_space *mapping = tree->mapping;
3670 int uptodate = 1;
3671 int ret;
3672
3673 rcu_read_lock();
3674 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3675 if (eb && atomic_inc_not_zero(&eb->refs)) {
3676 rcu_read_unlock();
3677 mark_page_accessed(eb->first_page);
3678 return eb;
3679 }
3680 rcu_read_unlock();
3681
3682 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
3683 if (!eb)
3684 return NULL;
3685
3686 if (page0) {
3687 eb->first_page = page0;
3688 i = 1;
3689 index++;
3690 page_cache_get(page0);
3691 mark_page_accessed(page0);
3692 set_page_extent_mapped(page0);
3693 set_page_extent_head(page0, len);
3694 uptodate = PageUptodate(page0);
3695 } else {
3696 i = 0;
3697 }
3698 for (; i < num_pages; i++, index++) {
3699 p = find_or_create_page(mapping, index, GFP_NOFS);
3700 if (!p) {
3701 WARN_ON(1);
3702 goto free_eb;
3703 }
3704 set_page_extent_mapped(p);
3705 mark_page_accessed(p);
3706 if (i == 0) {
3707 eb->first_page = p;
3708 set_page_extent_head(p, len);
3709 } else {
3710 set_page_private(p, EXTENT_PAGE_PRIVATE);
3711 }
3712 if (!PageUptodate(p))
3713 uptodate = 0;
3714
3715 /*
3716 * see below about how we avoid a nasty race with release page
3717 * and why we unlock later
3718 */
3719 if (i != 0)
3720 unlock_page(p);
3721 }
3722 if (uptodate)
3723 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3724
3725 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3726 if (ret)
3727 goto free_eb;
3728
3729 spin_lock(&tree->buffer_lock);
3730 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3731 if (ret == -EEXIST) {
3732 exists = radix_tree_lookup(&tree->buffer,
3733 start >> PAGE_CACHE_SHIFT);
3734 /* add one reference for the caller */
3735 atomic_inc(&exists->refs);
3736 spin_unlock(&tree->buffer_lock);
3737 radix_tree_preload_end();
3738 goto free_eb;
3739 }
3740 /* add one reference for the tree */
3741 atomic_inc(&eb->refs);
3742 spin_unlock(&tree->buffer_lock);
3743 radix_tree_preload_end();
3744
3745 /*
3746 * there is a race where release page may have
3747 * tried to find this extent buffer in the radix
3748 * but failed. It will tell the VM it is safe to
3749 * reclaim the, and it will clear the page private bit.
3750 * We must make sure to set the page private bit properly
3751 * after the extent buffer is in the radix tree so
3752 * it doesn't get lost
3753 */
3754 set_page_extent_mapped(eb->first_page);
3755 set_page_extent_head(eb->first_page, eb->len);
3756 if (!page0)
3757 unlock_page(eb->first_page);
3758 return eb;
3759
3760 free_eb:
3761 if (eb->first_page && !page0)
3762 unlock_page(eb->first_page);
3763
3764 if (!atomic_dec_and_test(&eb->refs))
3765 return exists;
3766 btrfs_release_extent_buffer(eb);
3767 return exists;
3768 }
3769
3770 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3771 u64 start, unsigned long len)
3772 {
3773 struct extent_buffer *eb;
3774
3775 rcu_read_lock();
3776 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3777 if (eb && atomic_inc_not_zero(&eb->refs)) {
3778 rcu_read_unlock();
3779 mark_page_accessed(eb->first_page);
3780 return eb;
3781 }
3782 rcu_read_unlock();
3783
3784 return NULL;
3785 }
3786
3787 void free_extent_buffer(struct extent_buffer *eb)
3788 {
3789 if (!eb)
3790 return;
3791
3792 if (!atomic_dec_and_test(&eb->refs))
3793 return;
3794
3795 WARN_ON(1);
3796 }
3797
3798 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3799 struct extent_buffer *eb)
3800 {
3801 unsigned long i;
3802 unsigned long num_pages;
3803 struct page *page;
3804
3805 num_pages = num_extent_pages(eb->start, eb->len);
3806
3807 for (i = 0; i < num_pages; i++) {
3808 page = extent_buffer_page(eb, i);
3809 if (!PageDirty(page))
3810 continue;
3811
3812 lock_page(page);
3813 WARN_ON(!PagePrivate(page));
3814
3815 set_page_extent_mapped(page);
3816 if (i == 0)
3817 set_page_extent_head(page, eb->len);
3818
3819 clear_page_dirty_for_io(page);
3820 spin_lock_irq(&page->mapping->tree_lock);
3821 if (!PageDirty(page)) {
3822 radix_tree_tag_clear(&page->mapping->page_tree,
3823 page_index(page),
3824 PAGECACHE_TAG_DIRTY);
3825 }
3826 spin_unlock_irq(&page->mapping->tree_lock);
3827 ClearPageError(page);
3828 unlock_page(page);
3829 }
3830 return 0;
3831 }
3832
3833 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3834 struct extent_buffer *eb)
3835 {
3836 unsigned long i;
3837 unsigned long num_pages;
3838 int was_dirty = 0;
3839
3840 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3841 num_pages = num_extent_pages(eb->start, eb->len);
3842 for (i = 0; i < num_pages; i++)
3843 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3844 return was_dirty;
3845 }
3846
3847 static int __eb_straddles_pages(u64 start, u64 len)
3848 {
3849 if (len < PAGE_CACHE_SIZE)
3850 return 1;
3851 if (start & (PAGE_CACHE_SIZE - 1))
3852 return 1;
3853 if ((start + len) & (PAGE_CACHE_SIZE - 1))
3854 return 1;
3855 return 0;
3856 }
3857
3858 static int eb_straddles_pages(struct extent_buffer *eb)
3859 {
3860 return __eb_straddles_pages(eb->start, eb->len);
3861 }
3862
3863 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3864 struct extent_buffer *eb,
3865 struct extent_state **cached_state)
3866 {
3867 unsigned long i;
3868 struct page *page;
3869 unsigned long num_pages;
3870
3871 num_pages = num_extent_pages(eb->start, eb->len);
3872 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3873
3874 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3875 cached_state, GFP_NOFS);
3876
3877 for (i = 0; i < num_pages; i++) {
3878 page = extent_buffer_page(eb, i);
3879 if (page)
3880 ClearPageUptodate(page);
3881 }
3882 return 0;
3883 }
3884
3885 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3886 struct extent_buffer *eb)
3887 {
3888 unsigned long i;
3889 struct page *page;
3890 unsigned long num_pages;
3891
3892 num_pages = num_extent_pages(eb->start, eb->len);
3893
3894 if (eb_straddles_pages(eb)) {
3895 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3896 NULL, GFP_NOFS);
3897 }
3898 for (i = 0; i < num_pages; i++) {
3899 page = extent_buffer_page(eb, i);
3900 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3901 ((i == num_pages - 1) &&
3902 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3903 check_page_uptodate(tree, page);
3904 continue;
3905 }
3906 SetPageUptodate(page);
3907 }
3908 return 0;
3909 }
3910
3911 int extent_range_uptodate(struct extent_io_tree *tree,
3912 u64 start, u64 end)
3913 {
3914 struct page *page;
3915 int ret;
3916 int pg_uptodate = 1;
3917 int uptodate;
3918 unsigned long index;
3919
3920 if (__eb_straddles_pages(start, end - start + 1)) {
3921 ret = test_range_bit(tree, start, end,
3922 EXTENT_UPTODATE, 1, NULL);
3923 if (ret)
3924 return 1;
3925 }
3926 while (start <= end) {
3927 index = start >> PAGE_CACHE_SHIFT;
3928 page = find_get_page(tree->mapping, index);
3929 if (!page)
3930 return 1;
3931 uptodate = PageUptodate(page);
3932 page_cache_release(page);
3933 if (!uptodate) {
3934 pg_uptodate = 0;
3935 break;
3936 }
3937 start += PAGE_CACHE_SIZE;
3938 }
3939 return pg_uptodate;
3940 }
3941
3942 int extent_buffer_uptodate(struct extent_io_tree *tree,
3943 struct extent_buffer *eb,
3944 struct extent_state *cached_state)
3945 {
3946 int ret = 0;
3947 unsigned long num_pages;
3948 unsigned long i;
3949 struct page *page;
3950 int pg_uptodate = 1;
3951
3952 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3953 return 1;
3954
3955 if (eb_straddles_pages(eb)) {
3956 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3957 EXTENT_UPTODATE, 1, cached_state);
3958 if (ret)
3959 return ret;
3960 }
3961
3962 num_pages = num_extent_pages(eb->start, eb->len);
3963 for (i = 0; i < num_pages; i++) {
3964 page = extent_buffer_page(eb, i);
3965 if (!PageUptodate(page)) {
3966 pg_uptodate = 0;
3967 break;
3968 }
3969 }
3970 return pg_uptodate;
3971 }
3972
3973 int read_extent_buffer_pages(struct extent_io_tree *tree,
3974 struct extent_buffer *eb, u64 start, int wait,
3975 get_extent_t *get_extent, int mirror_num)
3976 {
3977 unsigned long i;
3978 unsigned long start_i;
3979 struct page *page;
3980 int err;
3981 int ret = 0;
3982 int locked_pages = 0;
3983 int all_uptodate = 1;
3984 int inc_all_pages = 0;
3985 unsigned long num_pages;
3986 struct bio *bio = NULL;
3987 unsigned long bio_flags = 0;
3988
3989 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3990 return 0;
3991
3992 if (eb_straddles_pages(eb)) {
3993 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3994 EXTENT_UPTODATE, 1, NULL)) {
3995 return 0;
3996 }
3997 }
3998
3999 if (start) {
4000 WARN_ON(start < eb->start);
4001 start_i = (start >> PAGE_CACHE_SHIFT) -
4002 (eb->start >> PAGE_CACHE_SHIFT);
4003 } else {
4004 start_i = 0;
4005 }
4006
4007 num_pages = num_extent_pages(eb->start, eb->len);
4008 for (i = start_i; i < num_pages; i++) {
4009 page = extent_buffer_page(eb, i);
4010 if (wait == WAIT_NONE) {
4011 if (!trylock_page(page))
4012 goto unlock_exit;
4013 } else {
4014 lock_page(page);
4015 }
4016 locked_pages++;
4017 if (!PageUptodate(page))
4018 all_uptodate = 0;
4019 }
4020 if (all_uptodate) {
4021 if (start_i == 0)
4022 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4023 goto unlock_exit;
4024 }
4025
4026 for (i = start_i; i < num_pages; i++) {
4027 page = extent_buffer_page(eb, i);
4028
4029 WARN_ON(!PagePrivate(page));
4030
4031 set_page_extent_mapped(page);
4032 if (i == 0)
4033 set_page_extent_head(page, eb->len);
4034
4035 if (inc_all_pages)
4036 page_cache_get(page);
4037 if (!PageUptodate(page)) {
4038 if (start_i == 0)
4039 inc_all_pages = 1;
4040 ClearPageError(page);
4041 err = __extent_read_full_page(tree, page,
4042 get_extent, &bio,
4043 mirror_num, &bio_flags);
4044 if (err)
4045 ret = err;
4046 } else {
4047 unlock_page(page);
4048 }
4049 }
4050
4051 if (bio)
4052 submit_one_bio(READ, bio, mirror_num, bio_flags);
4053
4054 if (ret || wait != WAIT_COMPLETE)
4055 return ret;
4056
4057 for (i = start_i; i < num_pages; i++) {
4058 page = extent_buffer_page(eb, i);
4059 wait_on_page_locked(page);
4060 if (!PageUptodate(page))
4061 ret = -EIO;
4062 }
4063
4064 if (!ret)
4065 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4066 return ret;
4067
4068 unlock_exit:
4069 i = start_i;
4070 while (locked_pages > 0) {
4071 page = extent_buffer_page(eb, i);
4072 i++;
4073 unlock_page(page);
4074 locked_pages--;
4075 }
4076 return ret;
4077 }
4078
4079 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4080 unsigned long start,
4081 unsigned long len)
4082 {
4083 size_t cur;
4084 size_t offset;
4085 struct page *page;
4086 char *kaddr;
4087 char *dst = (char *)dstv;
4088 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4089 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4090
4091 WARN_ON(start > eb->len);
4092 WARN_ON(start + len > eb->start + eb->len);
4093
4094 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4095
4096 while (len > 0) {
4097 page = extent_buffer_page(eb, i);
4098
4099 cur = min(len, (PAGE_CACHE_SIZE - offset));
4100 kaddr = page_address(page);
4101 memcpy(dst, kaddr + offset, cur);
4102
4103 dst += cur;
4104 len -= cur;
4105 offset = 0;
4106 i++;
4107 }
4108 }
4109
4110 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4111 unsigned long min_len, char **map,
4112 unsigned long *map_start,
4113 unsigned long *map_len)
4114 {
4115 size_t offset = start & (PAGE_CACHE_SIZE - 1);
4116 char *kaddr;
4117 struct page *p;
4118 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4119 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4120 unsigned long end_i = (start_offset + start + min_len - 1) >>
4121 PAGE_CACHE_SHIFT;
4122
4123 if (i != end_i)
4124 return -EINVAL;
4125
4126 if (i == 0) {
4127 offset = start_offset;
4128 *map_start = 0;
4129 } else {
4130 offset = 0;
4131 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4132 }
4133
4134 if (start + min_len > eb->len) {
4135 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4136 "wanted %lu %lu\n", (unsigned long long)eb->start,
4137 eb->len, start, min_len);
4138 WARN_ON(1);
4139 return -EINVAL;
4140 }
4141
4142 p = extent_buffer_page(eb, i);
4143 kaddr = page_address(p);
4144 *map = kaddr + offset;
4145 *map_len = PAGE_CACHE_SIZE - offset;
4146 return 0;
4147 }
4148
4149 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4150 unsigned long start,
4151 unsigned long len)
4152 {
4153 size_t cur;
4154 size_t offset;
4155 struct page *page;
4156 char *kaddr;
4157 char *ptr = (char *)ptrv;
4158 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4159 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4160 int ret = 0;
4161
4162 WARN_ON(start > eb->len);
4163 WARN_ON(start + len > eb->start + eb->len);
4164
4165 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4166
4167 while (len > 0) {
4168 page = extent_buffer_page(eb, i);
4169
4170 cur = min(len, (PAGE_CACHE_SIZE - offset));
4171
4172 kaddr = page_address(page);
4173 ret = memcmp(ptr, kaddr + offset, cur);
4174 if (ret)
4175 break;
4176
4177 ptr += cur;
4178 len -= cur;
4179 offset = 0;
4180 i++;
4181 }
4182 return ret;
4183 }
4184
4185 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4186 unsigned long start, unsigned long len)
4187 {
4188 size_t cur;
4189 size_t offset;
4190 struct page *page;
4191 char *kaddr;
4192 char *src = (char *)srcv;
4193 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4194 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4195
4196 WARN_ON(start > eb->len);
4197 WARN_ON(start + len > eb->start + eb->len);
4198
4199 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4200
4201 while (len > 0) {
4202 page = extent_buffer_page(eb, i);
4203 WARN_ON(!PageUptodate(page));
4204
4205 cur = min(len, PAGE_CACHE_SIZE - offset);
4206 kaddr = page_address(page);
4207 memcpy(kaddr + offset, src, cur);
4208
4209 src += cur;
4210 len -= cur;
4211 offset = 0;
4212 i++;
4213 }
4214 }
4215
4216 void memset_extent_buffer(struct extent_buffer *eb, char c,
4217 unsigned long start, unsigned long len)
4218 {
4219 size_t cur;
4220 size_t offset;
4221 struct page *page;
4222 char *kaddr;
4223 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4224 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4225
4226 WARN_ON(start > eb->len);
4227 WARN_ON(start + len > eb->start + eb->len);
4228
4229 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4230
4231 while (len > 0) {
4232 page = extent_buffer_page(eb, i);
4233 WARN_ON(!PageUptodate(page));
4234
4235 cur = min(len, PAGE_CACHE_SIZE - offset);
4236 kaddr = page_address(page);
4237 memset(kaddr + offset, c, cur);
4238
4239 len -= cur;
4240 offset = 0;
4241 i++;
4242 }
4243 }
4244
4245 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4246 unsigned long dst_offset, unsigned long src_offset,
4247 unsigned long len)
4248 {
4249 u64 dst_len = dst->len;
4250 size_t cur;
4251 size_t offset;
4252 struct page *page;
4253 char *kaddr;
4254 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4255 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4256
4257 WARN_ON(src->len != dst_len);
4258
4259 offset = (start_offset + dst_offset) &
4260 ((unsigned long)PAGE_CACHE_SIZE - 1);
4261
4262 while (len > 0) {
4263 page = extent_buffer_page(dst, i);
4264 WARN_ON(!PageUptodate(page));
4265
4266 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4267
4268 kaddr = page_address(page);
4269 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4270
4271 src_offset += cur;
4272 len -= cur;
4273 offset = 0;
4274 i++;
4275 }
4276 }
4277
4278 static void move_pages(struct page *dst_page, struct page *src_page,
4279 unsigned long dst_off, unsigned long src_off,
4280 unsigned long len)
4281 {
4282 char *dst_kaddr = page_address(dst_page);
4283 if (dst_page == src_page) {
4284 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4285 } else {
4286 char *src_kaddr = page_address(src_page);
4287 char *p = dst_kaddr + dst_off + len;
4288 char *s = src_kaddr + src_off + len;
4289
4290 while (len--)
4291 *--p = *--s;
4292 }
4293 }
4294
4295 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4296 {
4297 unsigned long distance = (src > dst) ? src - dst : dst - src;
4298 return distance < len;
4299 }
4300
4301 static void copy_pages(struct page *dst_page, struct page *src_page,
4302 unsigned long dst_off, unsigned long src_off,
4303 unsigned long len)
4304 {
4305 char *dst_kaddr = page_address(dst_page);
4306 char *src_kaddr;
4307
4308 if (dst_page != src_page) {
4309 src_kaddr = page_address(src_page);
4310 } else {
4311 src_kaddr = dst_kaddr;
4312 BUG_ON(areas_overlap(src_off, dst_off, len));
4313 }
4314
4315 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4316 }
4317
4318 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4319 unsigned long src_offset, unsigned long len)
4320 {
4321 size_t cur;
4322 size_t dst_off_in_page;
4323 size_t src_off_in_page;
4324 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4325 unsigned long dst_i;
4326 unsigned long src_i;
4327
4328 if (src_offset + len > dst->len) {
4329 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4330 "len %lu dst len %lu\n", src_offset, len, dst->len);
4331 BUG_ON(1);
4332 }
4333 if (dst_offset + len > dst->len) {
4334 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4335 "len %lu dst len %lu\n", dst_offset, len, dst->len);
4336 BUG_ON(1);
4337 }
4338
4339 while (len > 0) {
4340 dst_off_in_page = (start_offset + dst_offset) &
4341 ((unsigned long)PAGE_CACHE_SIZE - 1);
4342 src_off_in_page = (start_offset + src_offset) &
4343 ((unsigned long)PAGE_CACHE_SIZE - 1);
4344
4345 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4346 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
4347
4348 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
4349 src_off_in_page));
4350 cur = min_t(unsigned long, cur,
4351 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
4352
4353 copy_pages(extent_buffer_page(dst, dst_i),
4354 extent_buffer_page(dst, src_i),
4355 dst_off_in_page, src_off_in_page, cur);
4356
4357 src_offset += cur;
4358 dst_offset += cur;
4359 len -= cur;
4360 }
4361 }
4362
4363 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4364 unsigned long src_offset, unsigned long len)
4365 {
4366 size_t cur;
4367 size_t dst_off_in_page;
4368 size_t src_off_in_page;
4369 unsigned long dst_end = dst_offset + len - 1;
4370 unsigned long src_end = src_offset + len - 1;
4371 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4372 unsigned long dst_i;
4373 unsigned long src_i;
4374
4375 if (src_offset + len > dst->len) {
4376 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4377 "len %lu len %lu\n", src_offset, len, dst->len);
4378 BUG_ON(1);
4379 }
4380 if (dst_offset + len > dst->len) {
4381 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4382 "len %lu len %lu\n", dst_offset, len, dst->len);
4383 BUG_ON(1);
4384 }
4385 if (!areas_overlap(src_offset, dst_offset, len)) {
4386 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4387 return;
4388 }
4389 while (len > 0) {
4390 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
4391 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
4392
4393 dst_off_in_page = (start_offset + dst_end) &
4394 ((unsigned long)PAGE_CACHE_SIZE - 1);
4395 src_off_in_page = (start_offset + src_end) &
4396 ((unsigned long)PAGE_CACHE_SIZE - 1);
4397
4398 cur = min_t(unsigned long, len, src_off_in_page + 1);
4399 cur = min(cur, dst_off_in_page + 1);
4400 move_pages(extent_buffer_page(dst, dst_i),
4401 extent_buffer_page(dst, src_i),
4402 dst_off_in_page - cur + 1,
4403 src_off_in_page - cur + 1, cur);
4404
4405 dst_end -= cur;
4406 src_end -= cur;
4407 len -= cur;
4408 }
4409 }
4410
4411 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4412 {
4413 struct extent_buffer *eb =
4414 container_of(head, struct extent_buffer, rcu_head);
4415
4416 btrfs_release_extent_buffer(eb);
4417 }
4418
4419 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
4420 {
4421 u64 start = page_offset(page);
4422 struct extent_buffer *eb;
4423 int ret = 1;
4424
4425 spin_lock(&tree->buffer_lock);
4426 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4427 if (!eb) {
4428 spin_unlock(&tree->buffer_lock);
4429 return ret;
4430 }
4431
4432 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4433 ret = 0;
4434 goto out;
4435 }
4436
4437 /*
4438 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4439 * Or go back.
4440 */
4441 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
4442 ret = 0;
4443 goto out;
4444 }
4445
4446 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4447 out:
4448 spin_unlock(&tree->buffer_lock);
4449
4450 /* at this point we can safely release the extent buffer */
4451 if (atomic_read(&eb->refs) == 0)
4452 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4453 return ret;
4454 }
Linus' kernel tree