projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: New data=ordered implementation
[deliverable/linux.git] / fs / btrfs / extent_io.c
1 #include <linux/bitops.h>
2 #include <linux/slab.h>
3 #include <linux/bio.h>
4 #include <linux/mm.h>
5 #include <linux/gfp.h>
6 #include <linux/pagemap.h>
7 #include <linux/page-flags.h>
8 #include <linux/module.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/version.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
17
18 /* temporary define until extent_map moves out of btrfs */
19 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
20 unsigned long extra_flags,
21 void (*ctor)(void *, struct kmem_cache *,
22 unsigned long));
23
24 static struct kmem_cache *extent_state_cache;
25 static struct kmem_cache *extent_buffer_cache;
26
27 static LIST_HEAD(buffers);
28 static LIST_HEAD(states);
29 static spinlock_t leak_lock = SPIN_LOCK_UNLOCKED;
30
31 #define BUFFER_LRU_MAX 64
32
33 struct tree_entry {
34 u64 start;
35 u64 end;
36 struct rb_node rb_node;
37 };
38
39 struct extent_page_data {
40 struct bio *bio;
41 struct extent_io_tree *tree;
42 get_extent_t *get_extent;
43 };
44
45 int __init extent_io_init(void)
46 {
47 extent_state_cache = btrfs_cache_create("extent_state",
48 sizeof(struct extent_state), 0,
49 NULL);
50 if (!extent_state_cache)
51 return -ENOMEM;
52
53 extent_buffer_cache = btrfs_cache_create("extent_buffers",
54 sizeof(struct extent_buffer), 0,
55 NULL);
56 if (!extent_buffer_cache)
57 goto free_state_cache;
58 return 0;
59
60 free_state_cache:
61 kmem_cache_destroy(extent_state_cache);
62 return -ENOMEM;
63 }
64
65 void extent_io_exit(void)
66 {
67 struct extent_state *state;
68 struct extent_buffer *eb;
69
70 while (!list_empty(&states)) {
71 state = list_entry(states.next, struct extent_state, leak_list);
72 printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state->start, state->end, state->state, state->tree, atomic_read(&state->refs));
73 list_del(&state->leak_list);
74 kmem_cache_free(extent_state_cache, state);
75
76 }
77
78 while (!list_empty(&buffers)) {
79 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
80 printk("buffer leak start %Lu len %lu refs %d\n", eb->start, eb->len, atomic_read(&eb->refs));
81 list_del(&eb->leak_list);
82 kmem_cache_free(extent_buffer_cache, eb);
83 }
84 if (extent_state_cache)
85 kmem_cache_destroy(extent_state_cache);
86 if (extent_buffer_cache)
87 kmem_cache_destroy(extent_buffer_cache);
88 }
89
90 void extent_io_tree_init(struct extent_io_tree *tree,
91 struct address_space *mapping, gfp_t mask)
92 {
93 tree->state.rb_node = NULL;
94 tree->ops = NULL;
95 tree->dirty_bytes = 0;
96 spin_lock_init(&tree->lock);
97 spin_lock_init(&tree->lru_lock);
98 tree->mapping = mapping;
99 INIT_LIST_HEAD(&tree->buffer_lru);
100 tree->lru_size = 0;
101 tree->last = NULL;
102 }
103 EXPORT_SYMBOL(extent_io_tree_init);
104
105 void extent_io_tree_empty_lru(struct extent_io_tree *tree)
106 {
107 struct extent_buffer *eb;
108 while(!list_empty(&tree->buffer_lru)) {
109 eb = list_entry(tree->buffer_lru.next, struct extent_buffer,
110 lru);
111 list_del_init(&eb->lru);
112 free_extent_buffer(eb);
113 }
114 }
115 EXPORT_SYMBOL(extent_io_tree_empty_lru);
116
117 struct extent_state *alloc_extent_state(gfp_t mask)
118 {
119 struct extent_state *state;
120 unsigned long flags;
121
122 state = kmem_cache_alloc(extent_state_cache, mask);
123 if (!state)
124 return state;
125 state->state = 0;
126 state->private = 0;
127 state->tree = NULL;
128 spin_lock_irqsave(&leak_lock, flags);
129 list_add(&state->leak_list, &states);
130 spin_unlock_irqrestore(&leak_lock, flags);
131
132 atomic_set(&state->refs, 1);
133 init_waitqueue_head(&state->wq);
134 return state;
135 }
136 EXPORT_SYMBOL(alloc_extent_state);
137
138 void free_extent_state(struct extent_state *state)
139 {
140 if (!state)
141 return;
142 if (atomic_dec_and_test(&state->refs)) {
143 unsigned long flags;
144 WARN_ON(state->tree);
145 spin_lock_irqsave(&leak_lock, flags);
146 list_del(&state->leak_list);
147 spin_unlock_irqrestore(&leak_lock, flags);
148 kmem_cache_free(extent_state_cache, state);
149 }
150 }
151 EXPORT_SYMBOL(free_extent_state);
152
153 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
154 struct rb_node *node)
155 {
156 struct rb_node ** p = &root->rb_node;
157 struct rb_node * parent = NULL;
158 struct tree_entry *entry;
159
160 while(*p) {
161 parent = *p;
162 entry = rb_entry(parent, struct tree_entry, rb_node);
163
164 if (offset < entry->start)
165 p = &(*p)->rb_left;
166 else if (offset > entry->end)
167 p = &(*p)->rb_right;
168 else
169 return parent;
170 }
171
172 entry = rb_entry(node, struct tree_entry, rb_node);
173 rb_link_node(node, parent, p);
174 rb_insert_color(node, root);
175 return NULL;
176 }
177
178 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
179 struct rb_node **prev_ret,
180 struct rb_node **next_ret)
181 {
182 struct rb_root *root = &tree->state;
183 struct rb_node * n = root->rb_node;
184 struct rb_node *prev = NULL;
185 struct rb_node *orig_prev = NULL;
186 struct tree_entry *entry;
187 struct tree_entry *prev_entry = NULL;
188
189 if (tree->last) {
190 struct extent_state *state;
191 state = tree->last;
192 if (state->start <= offset && offset <= state->end)
193 return &tree->last->rb_node;
194 }
195 while(n) {
196 entry = rb_entry(n, struct tree_entry, rb_node);
197 prev = n;
198 prev_entry = entry;
199
200 if (offset < entry->start)
201 n = n->rb_left;
202 else if (offset > entry->end)
203 n = n->rb_right;
204 else {
205 tree->last = rb_entry(n, struct extent_state, rb_node);
206 return n;
207 }
208 }
209
210 if (prev_ret) {
211 orig_prev = prev;
212 while(prev && offset > prev_entry->end) {
213 prev = rb_next(prev);
214 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
215 }
216 *prev_ret = prev;
217 prev = orig_prev;
218 }
219
220 if (next_ret) {
221 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
222 while(prev && offset < prev_entry->start) {
223 prev = rb_prev(prev);
224 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
225 }
226 *next_ret = prev;
227 }
228 return NULL;
229 }
230
231 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
232 u64 offset)
233 {
234 struct rb_node *prev = NULL;
235 struct rb_node *ret;
236
237 ret = __etree_search(tree, offset, &prev, NULL);
238 if (!ret) {
239 if (prev) {
240 tree->last = rb_entry(prev, struct extent_state,
241 rb_node);
242 }
243 return prev;
244 }
245 return ret;
246 }
247
248 /*
249 * utility function to look for merge candidates inside a given range.
250 * Any extents with matching state are merged together into a single
251 * extent in the tree. Extents with EXTENT_IO in their state field
252 * are not merged because the end_io handlers need to be able to do
253 * operations on them without sleeping (or doing allocations/splits).
254 *
255 * This should be called with the tree lock held.
256 */
257 static int merge_state(struct extent_io_tree *tree,
258 struct extent_state *state)
259 {
260 struct extent_state *other;
261 struct rb_node *other_node;
262
263 if (state->state & EXTENT_IOBITS)
264 return 0;
265
266 other_node = rb_prev(&state->rb_node);
267 if (other_node) {
268 other = rb_entry(other_node, struct extent_state, rb_node);
269 if (other->end == state->start - 1 &&
270 other->state == state->state) {
271 state->start = other->start;
272 other->tree = NULL;
273 if (tree->last == other)
274 tree->last = state;
275 rb_erase(&other->rb_node, &tree->state);
276 free_extent_state(other);
277 }
278 }
279 other_node = rb_next(&state->rb_node);
280 if (other_node) {
281 other = rb_entry(other_node, struct extent_state, rb_node);
282 if (other->start == state->end + 1 &&
283 other->state == state->state) {
284 other->start = state->start;
285 state->tree = NULL;
286 if (tree->last == state)
287 tree->last = other;
288 rb_erase(&state->rb_node, &tree->state);
289 free_extent_state(state);
290 }
291 }
292 return 0;
293 }
294
295 static void set_state_cb(struct extent_io_tree *tree,
296 struct extent_state *state,
297 unsigned long bits)
298 {
299 if (tree->ops && tree->ops->set_bit_hook) {
300 tree->ops->set_bit_hook(tree->mapping->host, state->start,
301 state->end, state->state, bits);
302 }
303 }
304
305 static void clear_state_cb(struct extent_io_tree *tree,
306 struct extent_state *state,
307 unsigned long bits)
308 {
309 if (tree->ops && tree->ops->set_bit_hook) {
310 tree->ops->clear_bit_hook(tree->mapping->host, state->start,
311 state->end, state->state, bits);
312 }
313 }
314
315 /*
316 * insert an extent_state struct into the tree. 'bits' are set on the
317 * struct before it is inserted.
318 *
319 * This may return -EEXIST if the extent is already there, in which case the
320 * state struct is freed.
321 *
322 * The tree lock is not taken internally. This is a utility function and
323 * probably isn't what you want to call (see set/clear_extent_bit).
324 */
325 static int insert_state(struct extent_io_tree *tree,
326 struct extent_state *state, u64 start, u64 end,
327 int bits)
328 {
329 struct rb_node *node;
330
331 if (end < start) {
332 printk("end < start %Lu %Lu\n", end, start);
333 WARN_ON(1);
334 }
335 if (bits & EXTENT_DIRTY)
336 tree->dirty_bytes += end - start + 1;
337 set_state_cb(tree, state, bits);
338 state->state |= bits;
339 state->start = start;
340 state->end = end;
341 node = tree_insert(&tree->state, end, &state->rb_node);
342 if (node) {
343 struct extent_state *found;
344 found = rb_entry(node, struct extent_state, rb_node);
345 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
346 free_extent_state(state);
347 return -EEXIST;
348 }
349 state->tree = tree;
350 tree->last = state;
351 merge_state(tree, state);
352 return 0;
353 }
354
355 /*
356 * split a given extent state struct in two, inserting the preallocated
357 * struct 'prealloc' as the newly created second half. 'split' indicates an
358 * offset inside 'orig' where it should be split.
359 *
360 * Before calling,
361 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
362 * are two extent state structs in the tree:
363 * prealloc: [orig->start, split - 1]
364 * orig: [ split, orig->end ]
365 *
366 * The tree locks are not taken by this function. They need to be held
367 * by the caller.
368 */
369 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
370 struct extent_state *prealloc, u64 split)
371 {
372 struct rb_node *node;
373 prealloc->start = orig->start;
374 prealloc->end = split - 1;
375 prealloc->state = orig->state;
376 orig->start = split;
377
378 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
379 if (node) {
380 struct extent_state *found;
381 found = rb_entry(node, struct extent_state, rb_node);
382 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
383 free_extent_state(prealloc);
384 return -EEXIST;
385 }
386 prealloc->tree = tree;
387 return 0;
388 }
389
390 /*
391 * utility function to clear some bits in an extent state struct.
392 * it will optionally wake up any one waiting on this state (wake == 1), or
393 * forcibly remove the state from the tree (delete == 1).
394 *
395 * If no bits are set on the state struct after clearing things, the
396 * struct is freed and removed from the tree
397 */
398 static int clear_state_bit(struct extent_io_tree *tree,
399 struct extent_state *state, int bits, int wake,
400 int delete)
401 {
402 int ret = state->state & bits;
403
404 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
405 u64 range = state->end - state->start + 1;
406 WARN_ON(range > tree->dirty_bytes);
407 tree->dirty_bytes -= range;
408 }
409 clear_state_cb(tree, state, bits);
410 state->state &= ~bits;
411 if (wake)
412 wake_up(&state->wq);
413 if (delete || state->state == 0) {
414 if (state->tree) {
415 clear_state_cb(tree, state, state->state);
416 if (tree->last == state) {
417 tree->last = extent_state_next(state);
418 }
419 rb_erase(&state->rb_node, &tree->state);
420 state->tree = NULL;
421 free_extent_state(state);
422 } else {
423 WARN_ON(1);
424 }
425 } else {
426 merge_state(tree, state);
427 }
428 return ret;
429 }
430
431 /*
432 * clear some bits on a range in the tree. This may require splitting
433 * or inserting elements in the tree, so the gfp mask is used to
434 * indicate which allocations or sleeping are allowed.
435 *
436 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
437 * the given range from the tree regardless of state (ie for truncate).
438 *
439 * the range [start, end] is inclusive.
440 *
441 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
442 * bits were already set, or zero if none of the bits were already set.
443 */
444 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
445 int bits, int wake, int delete, gfp_t mask)
446 {
447 struct extent_state *state;
448 struct extent_state *prealloc = NULL;
449 struct rb_node *node;
450 unsigned long flags;
451 int err;
452 int set = 0;
453
454 again:
455 if (!prealloc && (mask & __GFP_WAIT)) {
456 prealloc = alloc_extent_state(mask);
457 if (!prealloc)
458 return -ENOMEM;
459 }
460
461 spin_lock_irqsave(&tree->lock, flags);
462 /*
463 * this search will find the extents that end after
464 * our range starts
465 */
466 node = tree_search(tree, start);
467 if (!node)
468 goto out;
469 state = rb_entry(node, struct extent_state, rb_node);
470 if (state->start > end)
471 goto out;
472 WARN_ON(state->end < start);
473
474 /*
475 * | ---- desired range ---- |
476 * | state | or
477 * | ------------- state -------------- |
478 *
479 * We need to split the extent we found, and may flip
480 * bits on second half.
481 *
482 * If the extent we found extends past our range, we
483 * just split and search again. It'll get split again
484 * the next time though.
485 *
486 * If the extent we found is inside our range, we clear
487 * the desired bit on it.
488 */
489
490 if (state->start < start) {
491 if (!prealloc)
492 prealloc = alloc_extent_state(GFP_ATOMIC);
493 err = split_state(tree, state, prealloc, start);
494 BUG_ON(err == -EEXIST);
495 prealloc = NULL;
496 if (err)
497 goto out;
498 if (state->end <= end) {
499 start = state->end + 1;
500 set |= clear_state_bit(tree, state, bits,
501 wake, delete);
502 } else {
503 start = state->start;
504 }
505 goto search_again;
506 }
507 /*
508 * | ---- desired range ---- |
509 * | state |
510 * We need to split the extent, and clear the bit
511 * on the first half
512 */
513 if (state->start <= end && state->end > end) {
514 if (!prealloc)
515 prealloc = alloc_extent_state(GFP_ATOMIC);
516 err = split_state(tree, state, prealloc, end + 1);
517 BUG_ON(err == -EEXIST);
518
519 if (wake)
520 wake_up(&state->wq);
521 set |= clear_state_bit(tree, prealloc, bits,
522 wake, delete);
523 prealloc = NULL;
524 goto out;
525 }
526
527 start = state->end + 1;
528 set |= clear_state_bit(tree, state, bits, wake, delete);
529 goto search_again;
530
531 out:
532 spin_unlock_irqrestore(&tree->lock, flags);
533 if (prealloc)
534 free_extent_state(prealloc);
535
536 return set;
537
538 search_again:
539 if (start > end)
540 goto out;
541 spin_unlock_irqrestore(&tree->lock, flags);
542 if (mask & __GFP_WAIT)
543 cond_resched();
544 goto again;
545 }
546 EXPORT_SYMBOL(clear_extent_bit);
547
548 static int wait_on_state(struct extent_io_tree *tree,
549 struct extent_state *state)
550 {
551 DEFINE_WAIT(wait);
552 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
553 spin_unlock_irq(&tree->lock);
554 schedule();
555 spin_lock_irq(&tree->lock);
556 finish_wait(&state->wq, &wait);
557 return 0;
558 }
559
560 /*
561 * waits for one or more bits to clear on a range in the state tree.
562 * The range [start, end] is inclusive.
563 * The tree lock is taken by this function
564 */
565 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
566 {
567 struct extent_state *state;
568 struct rb_node *node;
569
570 spin_lock_irq(&tree->lock);
571 again:
572 while (1) {
573 /*
574 * this search will find all the extents that end after
575 * our range starts
576 */
577 node = tree_search(tree, start);
578 if (!node)
579 break;
580
581 state = rb_entry(node, struct extent_state, rb_node);
582
583 if (state->start > end)
584 goto out;
585
586 if (state->state & bits) {
587 start = state->start;
588 atomic_inc(&state->refs);
589 wait_on_state(tree, state);
590 free_extent_state(state);
591 goto again;
592 }
593 start = state->end + 1;
594
595 if (start > end)
596 break;
597
598 if (need_resched()) {
599 spin_unlock_irq(&tree->lock);
600 cond_resched();
601 spin_lock_irq(&tree->lock);
602 }
603 }
604 out:
605 spin_unlock_irq(&tree->lock);
606 return 0;
607 }
608 EXPORT_SYMBOL(wait_extent_bit);
609
610 static void set_state_bits(struct extent_io_tree *tree,
611 struct extent_state *state,
612 int bits)
613 {
614 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
615 u64 range = state->end - state->start + 1;
616 tree->dirty_bytes += range;
617 }
618 set_state_cb(tree, state, bits);
619 state->state |= bits;
620 }
621
622 /*
623 * set some bits on a range in the tree. This may require allocations
624 * or sleeping, so the gfp mask is used to indicate what is allowed.
625 *
626 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
627 * range already has the desired bits set. The start of the existing
628 * range is returned in failed_start in this case.
629 *
630 * [start, end] is inclusive
631 * This takes the tree lock.
632 */
633 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits,
634 int exclusive, u64 *failed_start, gfp_t mask)
635 {
636 struct extent_state *state;
637 struct extent_state *prealloc = NULL;
638 struct rb_node *node;
639 unsigned long flags;
640 int err = 0;
641 int set;
642 u64 last_start;
643 u64 last_end;
644 again:
645 if (!prealloc && (mask & __GFP_WAIT)) {
646 prealloc = alloc_extent_state(mask);
647 if (!prealloc)
648 return -ENOMEM;
649 }
650
651 spin_lock_irqsave(&tree->lock, flags);
652 /*
653 * this search will find all the extents that end after
654 * our range starts.
655 */
656 node = tree_search(tree, start);
657 if (!node) {
658 err = insert_state(tree, prealloc, start, end, bits);
659 prealloc = NULL;
660 BUG_ON(err == -EEXIST);
661 goto out;
662 }
663
664 state = rb_entry(node, struct extent_state, rb_node);
665 last_start = state->start;
666 last_end = state->end;
667
668 /*
669 * | ---- desired range ---- |
670 * | state |
671 *
672 * Just lock what we found and keep going
673 */
674 if (state->start == start && state->end <= end) {
675 set = state->state & bits;
676 if (set && exclusive) {
677 *failed_start = state->start;
678 err = -EEXIST;
679 goto out;
680 }
681 set_state_bits(tree, state, bits);
682 start = state->end + 1;
683 merge_state(tree, state);
684 goto search_again;
685 }
686
687 /*
688 * | ---- desired range ---- |
689 * | state |
690 * or
691 * | ------------- state -------------- |
692 *
693 * We need to split the extent we found, and may flip bits on
694 * second half.
695 *
696 * If the extent we found extends past our
697 * range, we just split and search again. It'll get split
698 * again the next time though.
699 *
700 * If the extent we found is inside our range, we set the
701 * desired bit on it.
702 */
703 if (state->start < start) {
704 set = state->state & bits;
705 if (exclusive && set) {
706 *failed_start = start;
707 err = -EEXIST;
708 goto out;
709 }
710 err = split_state(tree, state, prealloc, start);
711 BUG_ON(err == -EEXIST);
712 prealloc = NULL;
713 if (err)
714 goto out;
715 if (state->end <= end) {
716 set_state_bits(tree, state, bits);
717 start = state->end + 1;
718 merge_state(tree, state);
719 } else {
720 start = state->start;
721 }
722 goto search_again;
723 }
724 /*
725 * | ---- desired range ---- |
726 * | state | or | state |
727 *
728 * There's a hole, we need to insert something in it and
729 * ignore the extent we found.
730 */
731 if (state->start > start) {
732 u64 this_end;
733 if (end < last_start)
734 this_end = end;
735 else
736 this_end = last_start -1;
737 err = insert_state(tree, prealloc, start, this_end,
738 bits);
739 prealloc = NULL;
740 BUG_ON(err == -EEXIST);
741 if (err)
742 goto out;
743 start = this_end + 1;
744 goto search_again;
745 }
746 /*
747 * | ---- desired range ---- |
748 * | state |
749 * We need to split the extent, and set the bit
750 * on the first half
751 */
752 if (state->start <= end && state->end > end) {
753 set = state->state & bits;
754 if (exclusive && set) {
755 *failed_start = start;
756 err = -EEXIST;
757 goto out;
758 }
759 err = split_state(tree, state, prealloc, end + 1);
760 BUG_ON(err == -EEXIST);
761
762 set_state_bits(tree, prealloc, bits);
763 merge_state(tree, prealloc);
764 prealloc = NULL;
765 goto out;
766 }
767
768 goto search_again;
769
770 out:
771 spin_unlock_irqrestore(&tree->lock, flags);
772 if (prealloc)
773 free_extent_state(prealloc);
774
775 return err;
776
777 search_again:
778 if (start > end)
779 goto out;
780 spin_unlock_irqrestore(&tree->lock, flags);
781 if (mask & __GFP_WAIT)
782 cond_resched();
783 goto again;
784 }
785 EXPORT_SYMBOL(set_extent_bit);
786
787 /* wrappers around set/clear extent bit */
788 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
789 gfp_t mask)
790 {
791 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
792 mask);
793 }
794 EXPORT_SYMBOL(set_extent_dirty);
795
796 int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
797 gfp_t mask)
798 {
799 return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask);
800 }
801 EXPORT_SYMBOL(set_extent_ordered);
802
803 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
804 int bits, gfp_t mask)
805 {
806 return set_extent_bit(tree, start, end, bits, 0, NULL,
807 mask);
808 }
809 EXPORT_SYMBOL(set_extent_bits);
810
811 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
812 int bits, gfp_t mask)
813 {
814 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
815 }
816 EXPORT_SYMBOL(clear_extent_bits);
817
818 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
819 gfp_t mask)
820 {
821 return set_extent_bit(tree, start, end,
822 EXTENT_DELALLOC | EXTENT_DIRTY,
823 0, NULL, mask);
824 }
825 EXPORT_SYMBOL(set_extent_delalloc);
826
827 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
828 gfp_t mask)
829 {
830 return clear_extent_bit(tree, start, end,
831 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
832 }
833 EXPORT_SYMBOL(clear_extent_dirty);
834
835 int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end,
836 gfp_t mask)
837 {
838 return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask);
839 }
840 EXPORT_SYMBOL(clear_extent_ordered);
841
842 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
843 gfp_t mask)
844 {
845 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
846 mask);
847 }
848 EXPORT_SYMBOL(set_extent_new);
849
850 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
851 gfp_t mask)
852 {
853 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
854 }
855 EXPORT_SYMBOL(clear_extent_new);
856
857 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
858 gfp_t mask)
859 {
860 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
861 mask);
862 }
863 EXPORT_SYMBOL(set_extent_uptodate);
864
865 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
866 gfp_t mask)
867 {
868 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
869 }
870 EXPORT_SYMBOL(clear_extent_uptodate);
871
872 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
873 gfp_t mask)
874 {
875 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
876 0, NULL, mask);
877 }
878 EXPORT_SYMBOL(set_extent_writeback);
879
880 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
881 gfp_t mask)
882 {
883 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
884 }
885 EXPORT_SYMBOL(clear_extent_writeback);
886
887 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
888 {
889 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
890 }
891 EXPORT_SYMBOL(wait_on_extent_writeback);
892
893 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
894 {
895 int err;
896 u64 failed_start;
897 while (1) {
898 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
899 &failed_start, mask);
900 if (err == -EEXIST && (mask & __GFP_WAIT)) {
901 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
902 start = failed_start;
903 } else {
904 break;
905 }
906 WARN_ON(start > end);
907 }
908 return err;
909 }
910 EXPORT_SYMBOL(lock_extent);
911
912 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
913 gfp_t mask)
914 {
915 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
916 }
917 EXPORT_SYMBOL(unlock_extent);
918
919 /*
920 * helper function to set pages and extents in the tree dirty
921 */
922 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
923 {
924 unsigned long index = start >> PAGE_CACHE_SHIFT;
925 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
926 struct page *page;
927
928 while (index <= end_index) {
929 page = find_get_page(tree->mapping, index);
930 BUG_ON(!page);
931 __set_page_dirty_nobuffers(page);
932 page_cache_release(page);
933 index++;
934 }
935 set_extent_dirty(tree, start, end, GFP_NOFS);
936 return 0;
937 }
938 EXPORT_SYMBOL(set_range_dirty);
939
940 /*
941 * helper function to set both pages and extents in the tree writeback
942 */
943 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
944 {
945 unsigned long index = start >> PAGE_CACHE_SHIFT;
946 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
947 struct page *page;
948
949 while (index <= end_index) {
950 page = find_get_page(tree->mapping, index);
951 BUG_ON(!page);
952 set_page_writeback(page);
953 page_cache_release(page);
954 index++;
955 }
956 set_extent_writeback(tree, start, end, GFP_NOFS);
957 return 0;
958 }
959 EXPORT_SYMBOL(set_range_writeback);
960
961 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
962 u64 *start_ret, u64 *end_ret, int bits)
963 {
964 struct rb_node *node;
965 struct extent_state *state;
966 int ret = 1;
967
968 spin_lock_irq(&tree->lock);
969 /*
970 * this search will find all the extents that end after
971 * our range starts.
972 */
973 node = tree_search(tree, start);
974 if (!node) {
975 goto out;
976 }
977
978 while(1) {
979 state = rb_entry(node, struct extent_state, rb_node);
980 if (state->end >= start && (state->state & bits)) {
981 *start_ret = state->start;
982 *end_ret = state->end;
983 ret = 0;
984 break;
985 }
986 node = rb_next(node);
987 if (!node)
988 break;
989 }
990 out:
991 spin_unlock_irq(&tree->lock);
992 return ret;
993 }
994 EXPORT_SYMBOL(find_first_extent_bit);
995
996 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
997 u64 start, int bits)
998 {
999 struct rb_node *node;
1000 struct extent_state *state;
1001
1002 /*
1003 * this search will find all the extents that end after
1004 * our range starts.
1005 */
1006 node = tree_search(tree, start);
1007 if (!node) {
1008 goto out;
1009 }
1010
1011 while(1) {
1012 state = rb_entry(node, struct extent_state, rb_node);
1013 if (state->end >= start && (state->state & bits)) {
1014 return state;
1015 }
1016 node = rb_next(node);
1017 if (!node)
1018 break;
1019 }
1020 out:
1021 return NULL;
1022 }
1023 EXPORT_SYMBOL(find_first_extent_bit_state);
1024
1025 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1026 u64 *start, u64 *end, u64 max_bytes)
1027 {
1028 struct rb_node *node;
1029 struct extent_state *state;
1030 u64 cur_start = *start;
1031 u64 found = 0;
1032 u64 total_bytes = 0;
1033
1034 spin_lock_irq(&tree->lock);
1035 /*
1036 * this search will find all the extents that end after
1037 * our range starts.
1038 */
1039 search_again:
1040 node = tree_search(tree, cur_start);
1041 if (!node) {
1042 if (!found)
1043 *end = (u64)-1;
1044 goto out;
1045 }
1046
1047 while(1) {
1048 state = rb_entry(node, struct extent_state, rb_node);
1049 if (found && state->start != cur_start) {
1050 goto out;
1051 }
1052 if (!(state->state & EXTENT_DELALLOC)) {
1053 if (!found)
1054 *end = state->end;
1055 goto out;
1056 }
1057 if (!found) {
1058 struct extent_state *prev_state;
1059 struct rb_node *prev_node = node;
1060 while(1) {
1061 prev_node = rb_prev(prev_node);
1062 if (!prev_node)
1063 break;
1064 prev_state = rb_entry(prev_node,
1065 struct extent_state,
1066 rb_node);
1067 if (!(prev_state->state & EXTENT_DELALLOC))
1068 break;
1069 state = prev_state;
1070 node = prev_node;
1071 }
1072 }
1073 if (state->state & EXTENT_LOCKED) {
1074 DEFINE_WAIT(wait);
1075 atomic_inc(&state->refs);
1076 prepare_to_wait(&state->wq, &wait,
1077 TASK_UNINTERRUPTIBLE);
1078 spin_unlock_irq(&tree->lock);
1079 schedule();
1080 spin_lock_irq(&tree->lock);
1081 finish_wait(&state->wq, &wait);
1082 free_extent_state(state);
1083 goto search_again;
1084 }
1085 set_state_cb(tree, state, EXTENT_LOCKED);
1086 state->state |= EXTENT_LOCKED;
1087 if (!found)
1088 *start = state->start;
1089 found++;
1090 *end = state->end;
1091 cur_start = state->end + 1;
1092 node = rb_next(node);
1093 if (!node)
1094 break;
1095 total_bytes += state->end - state->start + 1;
1096 if (total_bytes >= max_bytes)
1097 break;
1098 }
1099 out:
1100 spin_unlock_irq(&tree->lock);
1101 return found;
1102 }
1103
1104 u64 count_range_bits(struct extent_io_tree *tree,
1105 u64 *start, u64 search_end, u64 max_bytes,
1106 unsigned long bits)
1107 {
1108 struct rb_node *node;
1109 struct extent_state *state;
1110 u64 cur_start = *start;
1111 u64 total_bytes = 0;
1112 int found = 0;
1113
1114 if (search_end <= cur_start) {
1115 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1116 WARN_ON(1);
1117 return 0;
1118 }
1119
1120 spin_lock_irq(&tree->lock);
1121 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1122 total_bytes = tree->dirty_bytes;
1123 goto out;
1124 }
1125 /*
1126 * this search will find all the extents that end after
1127 * our range starts.
1128 */
1129 node = tree_search(tree, cur_start);
1130 if (!node) {
1131 goto out;
1132 }
1133
1134 while(1) {
1135 state = rb_entry(node, struct extent_state, rb_node);
1136 if (state->start > search_end)
1137 break;
1138 if (state->end >= cur_start && (state->state & bits)) {
1139 total_bytes += min(search_end, state->end) + 1 -
1140 max(cur_start, state->start);
1141 if (total_bytes >= max_bytes)
1142 break;
1143 if (!found) {
1144 *start = state->start;
1145 found = 1;
1146 }
1147 }
1148 node = rb_next(node);
1149 if (!node)
1150 break;
1151 }
1152 out:
1153 spin_unlock_irq(&tree->lock);
1154 return total_bytes;
1155 }
1156 /*
1157 * helper function to lock both pages and extents in the tree.
1158 * pages must be locked first.
1159 */
1160 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1161 {
1162 unsigned long index = start >> PAGE_CACHE_SHIFT;
1163 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1164 struct page *page;
1165 int err;
1166
1167 while (index <= end_index) {
1168 page = grab_cache_page(tree->mapping, index);
1169 if (!page) {
1170 err = -ENOMEM;
1171 goto failed;
1172 }
1173 if (IS_ERR(page)) {
1174 err = PTR_ERR(page);
1175 goto failed;
1176 }
1177 index++;
1178 }
1179 lock_extent(tree, start, end, GFP_NOFS);
1180 return 0;
1181
1182 failed:
1183 /*
1184 * we failed above in getting the page at 'index', so we undo here
1185 * up to but not including the page at 'index'
1186 */
1187 end_index = index;
1188 index = start >> PAGE_CACHE_SHIFT;
1189 while (index < end_index) {
1190 page = find_get_page(tree->mapping, index);
1191 unlock_page(page);
1192 page_cache_release(page);
1193 index++;
1194 }
1195 return err;
1196 }
1197 EXPORT_SYMBOL(lock_range);
1198
1199 /*
1200 * helper function to unlock both pages and extents in the tree.
1201 */
1202 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1203 {
1204 unsigned long index = start >> PAGE_CACHE_SHIFT;
1205 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1206 struct page *page;
1207
1208 while (index <= end_index) {
1209 page = find_get_page(tree->mapping, index);
1210 unlock_page(page);
1211 page_cache_release(page);
1212 index++;
1213 }
1214 unlock_extent(tree, start, end, GFP_NOFS);
1215 return 0;
1216 }
1217 EXPORT_SYMBOL(unlock_range);
1218
1219 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1220 {
1221 struct rb_node *node;
1222 struct extent_state *state;
1223 int ret = 0;
1224
1225 spin_lock_irq(&tree->lock);
1226 /*
1227 * this search will find all the extents that end after
1228 * our range starts.
1229 */
1230 node = tree_search(tree, start);
1231 if (!node) {
1232 ret = -ENOENT;
1233 goto out;
1234 }
1235 state = rb_entry(node, struct extent_state, rb_node);
1236 if (state->start != start) {
1237 ret = -ENOENT;
1238 goto out;
1239 }
1240 state->private = private;
1241 out:
1242 spin_unlock_irq(&tree->lock);
1243 return ret;
1244 }
1245
1246 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1247 {
1248 struct rb_node *node;
1249 struct extent_state *state;
1250 int ret = 0;
1251
1252 spin_lock_irq(&tree->lock);
1253 /*
1254 * this search will find all the extents that end after
1255 * our range starts.
1256 */
1257 node = tree_search(tree, start);
1258 if (!node) {
1259 ret = -ENOENT;
1260 goto out;
1261 }
1262 state = rb_entry(node, struct extent_state, rb_node);
1263 if (state->start != start) {
1264 ret = -ENOENT;
1265 goto out;
1266 }
1267 *private = state->private;
1268 out:
1269 spin_unlock_irq(&tree->lock);
1270 return ret;
1271 }
1272
1273 /*
1274 * searches a range in the state tree for a given mask.
1275 * If 'filled' == 1, this returns 1 only if every extent in the tree
1276 * has the bits set. Otherwise, 1 is returned if any bit in the
1277 * range is found set.
1278 */
1279 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1280 int bits, int filled)
1281 {
1282 struct extent_state *state = NULL;
1283 struct rb_node *node;
1284 int bitset = 0;
1285 unsigned long flags;
1286
1287 spin_lock_irqsave(&tree->lock, flags);
1288 node = tree_search(tree, start);
1289 while (node && start <= end) {
1290 state = rb_entry(node, struct extent_state, rb_node);
1291
1292 if (filled && state->start > start) {
1293 bitset = 0;
1294 break;
1295 }
1296
1297 if (state->start > end)
1298 break;
1299
1300 if (state->state & bits) {
1301 bitset = 1;
1302 if (!filled)
1303 break;
1304 } else if (filled) {
1305 bitset = 0;
1306 break;
1307 }
1308 start = state->end + 1;
1309 if (start > end)
1310 break;
1311 node = rb_next(node);
1312 if (!node) {
1313 if (filled)
1314 bitset = 0;
1315 break;
1316 }
1317 }
1318 spin_unlock_irqrestore(&tree->lock, flags);
1319 return bitset;
1320 }
1321 EXPORT_SYMBOL(test_range_bit);
1322
1323 /*
1324 * helper function to set a given page up to date if all the
1325 * extents in the tree for that page are up to date
1326 */
1327 static int check_page_uptodate(struct extent_io_tree *tree,
1328 struct page *page)
1329 {
1330 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1331 u64 end = start + PAGE_CACHE_SIZE - 1;
1332 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1333 SetPageUptodate(page);
1334 return 0;
1335 }
1336
1337 /*
1338 * helper function to unlock a page if all the extents in the tree
1339 * for that page are unlocked
1340 */
1341 static int check_page_locked(struct extent_io_tree *tree,
1342 struct page *page)
1343 {
1344 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1345 u64 end = start + PAGE_CACHE_SIZE - 1;
1346 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1347 unlock_page(page);
1348 return 0;
1349 }
1350
1351 /*
1352 * helper function to end page writeback if all the extents
1353 * in the tree for that page are done with writeback
1354 */
1355 static int check_page_writeback(struct extent_io_tree *tree,
1356 struct page *page)
1357 {
1358 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1359 u64 end = start + PAGE_CACHE_SIZE - 1;
1360 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1361 end_page_writeback(page);
1362 return 0;
1363 }
1364
1365 /* lots and lots of room for performance fixes in the end_bio funcs */
1366
1367 /*
1368 * after a writepage IO is done, we need to:
1369 * clear the uptodate bits on error
1370 * clear the writeback bits in the extent tree for this IO
1371 * end_page_writeback if the page has no more pending IO
1372 *
1373 * Scheduling is not allowed, so the extent state tree is expected
1374 * to have one and only one object corresponding to this IO.
1375 */
1376 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1377 static void end_bio_extent_writepage(struct bio *bio, int err)
1378 #else
1379 static int end_bio_extent_writepage(struct bio *bio,
1380 unsigned int bytes_done, int err)
1381 #endif
1382 {
1383 int uptodate = err == 0;
1384 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1385 struct extent_state *state = bio->bi_private;
1386 struct extent_io_tree *tree = state->tree;
1387 struct rb_node *node;
1388 u64 start;
1389 u64 end;
1390 u64 cur;
1391 int whole_page;
1392 int ret;
1393 unsigned long flags;
1394
1395 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1396 if (bio->bi_size)
1397 return 1;
1398 #endif
1399 do {
1400 struct page *page = bvec->bv_page;
1401 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1402 bvec->bv_offset;
1403 end = start + bvec->bv_len - 1;
1404
1405 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1406 whole_page = 1;
1407 else
1408 whole_page = 0;
1409
1410 if (--bvec >= bio->bi_io_vec)
1411 prefetchw(&bvec->bv_page->flags);
1412 if (tree->ops && tree->ops->writepage_end_io_hook) {
1413 ret = tree->ops->writepage_end_io_hook(page, start,
1414 end, state, uptodate);
1415 if (ret)
1416 uptodate = 0;
1417 }
1418
1419 if (!uptodate && tree->ops &&
1420 tree->ops->writepage_io_failed_hook) {
1421 ret = tree->ops->writepage_io_failed_hook(bio, page,
1422 start, end, state);
1423 if (ret == 0) {
1424 state = NULL;
1425 uptodate = (err == 0);
1426 continue;
1427 }
1428 }
1429
1430 if (!uptodate) {
1431 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1432 ClearPageUptodate(page);
1433 SetPageError(page);
1434 }
1435
1436 /*
1437 * bios can get merged in funny ways, and so we need to
1438 * be careful with the state variable. We know the
1439 * state won't be merged with others because it has
1440 * WRITEBACK set, but we can't be sure each biovec is
1441 * sequential in the file. So, if our cached state
1442 * doesn't match the expected end, search the tree
1443 * for the correct one.
1444 */
1445
1446 spin_lock_irqsave(&tree->lock, flags);
1447 if (!state || state->end != end) {
1448 state = NULL;
1449 node = __etree_search(tree, start, NULL, NULL);
1450 if (node) {
1451 state = rb_entry(node, struct extent_state,
1452 rb_node);
1453 if (state->end != end ||
1454 !(state->state & EXTENT_WRITEBACK))
1455 state = NULL;
1456 }
1457 if (!state) {
1458 spin_unlock_irqrestore(&tree->lock, flags);
1459 clear_extent_writeback(tree, start,
1460 end, GFP_ATOMIC);
1461 goto next_io;
1462 }
1463 }
1464 cur = end;
1465 while(1) {
1466 struct extent_state *clear = state;
1467 cur = state->start;
1468 node = rb_prev(&state->rb_node);
1469 if (node) {
1470 state = rb_entry(node,
1471 struct extent_state,
1472 rb_node);
1473 } else {
1474 state = NULL;
1475 }
1476
1477 clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1478 1, 0);
1479 if (cur == start)
1480 break;
1481 if (cur < start) {
1482 WARN_ON(1);
1483 break;
1484 }
1485 if (!node)
1486 break;
1487 }
1488 /* before releasing the lock, make sure the next state
1489 * variable has the expected bits set and corresponds
1490 * to the correct offsets in the file
1491 */
1492 if (state && (state->end + 1 != start ||
1493 !(state->state & EXTENT_WRITEBACK))) {
1494 state = NULL;
1495 }
1496 spin_unlock_irqrestore(&tree->lock, flags);
1497 next_io:
1498
1499 if (whole_page)
1500 end_page_writeback(page);
1501 else
1502 check_page_writeback(tree, page);
1503 } while (bvec >= bio->bi_io_vec);
1504 bio_put(bio);
1505 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1506 return 0;
1507 #endif
1508 }
1509
1510 /*
1511 * after a readpage IO is done, we need to:
1512 * clear the uptodate bits on error
1513 * set the uptodate bits if things worked
1514 * set the page up to date if all extents in the tree are uptodate
1515 * clear the lock bit in the extent tree
1516 * unlock the page if there are no other extents locked for it
1517 *
1518 * Scheduling is not allowed, so the extent state tree is expected
1519 * to have one and only one object corresponding to this IO.
1520 */
1521 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1522 static void end_bio_extent_readpage(struct bio *bio, int err)
1523 #else
1524 static int end_bio_extent_readpage(struct bio *bio,
1525 unsigned int bytes_done, int err)
1526 #endif
1527 {
1528 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1529 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1530 struct extent_state *state = bio->bi_private;
1531 struct extent_io_tree *tree = state->tree;
1532 struct rb_node *node;
1533 u64 start;
1534 u64 end;
1535 u64 cur;
1536 unsigned long flags;
1537 int whole_page;
1538 int ret;
1539
1540 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1541 if (bio->bi_size)
1542 return 1;
1543 #endif
1544
1545 do {
1546 struct page *page = bvec->bv_page;
1547 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1548 bvec->bv_offset;
1549 end = start + bvec->bv_len - 1;
1550
1551 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1552 whole_page = 1;
1553 else
1554 whole_page = 0;
1555
1556 if (--bvec >= bio->bi_io_vec)
1557 prefetchw(&bvec->bv_page->flags);
1558
1559 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1560 ret = tree->ops->readpage_end_io_hook(page, start, end,
1561 state);
1562 if (ret)
1563 uptodate = 0;
1564 }
1565 if (!uptodate && tree->ops &&
1566 tree->ops->readpage_io_failed_hook) {
1567 ret = tree->ops->readpage_io_failed_hook(bio, page,
1568 start, end, state);
1569 if (ret == 0) {
1570 state = NULL;
1571 uptodate =
1572 test_bit(BIO_UPTODATE, &bio->bi_flags);
1573 continue;
1574 }
1575 }
1576
1577 spin_lock_irqsave(&tree->lock, flags);
1578 if (!state || state->end != end) {
1579 state = NULL;
1580 node = __etree_search(tree, start, NULL, NULL);
1581 if (node) {
1582 state = rb_entry(node, struct extent_state,
1583 rb_node);
1584 if (state->end != end ||
1585 !(state->state & EXTENT_LOCKED))
1586 state = NULL;
1587 }
1588 if (!state) {
1589 spin_unlock_irqrestore(&tree->lock, flags);
1590 if (uptodate)
1591 set_extent_uptodate(tree, start, end,
1592 GFP_ATOMIC);
1593 unlock_extent(tree, start, end, GFP_ATOMIC);
1594 goto next_io;
1595 }
1596 }
1597
1598 cur = end;
1599 while(1) {
1600 struct extent_state *clear = state;
1601 cur = state->start;
1602 node = rb_prev(&state->rb_node);
1603 if (node) {
1604 state = rb_entry(node,
1605 struct extent_state,
1606 rb_node);
1607 } else {
1608 state = NULL;
1609 }
1610 if (uptodate) {
1611 set_state_cb(tree, clear, EXTENT_UPTODATE);
1612 clear->state |= EXTENT_UPTODATE;
1613 }
1614 clear_state_bit(tree, clear, EXTENT_LOCKED,
1615 1, 0);
1616 if (cur == start)
1617 break;
1618 if (cur < start) {
1619 WARN_ON(1);
1620 break;
1621 }
1622 if (!node)
1623 break;
1624 }
1625 /* before releasing the lock, make sure the next state
1626 * variable has the expected bits set and corresponds
1627 * to the correct offsets in the file
1628 */
1629 if (state && (state->end + 1 != start ||
1630 !(state->state & EXTENT_LOCKED))) {
1631 state = NULL;
1632 }
1633 spin_unlock_irqrestore(&tree->lock, flags);
1634 next_io:
1635 if (whole_page) {
1636 if (uptodate) {
1637 SetPageUptodate(page);
1638 } else {
1639 ClearPageUptodate(page);
1640 SetPageError(page);
1641 }
1642 unlock_page(page);
1643 } else {
1644 if (uptodate) {
1645 check_page_uptodate(tree, page);
1646 } else {
1647 ClearPageUptodate(page);
1648 SetPageError(page);
1649 }
1650 check_page_locked(tree, page);
1651 }
1652 } while (bvec >= bio->bi_io_vec);
1653
1654 bio_put(bio);
1655 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1656 return 0;
1657 #endif
1658 }
1659
1660 /*
1661 * IO done from prepare_write is pretty simple, we just unlock
1662 * the structs in the extent tree when done, and set the uptodate bits
1663 * as appropriate.
1664 */
1665 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1666 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1667 #else
1668 static int end_bio_extent_preparewrite(struct bio *bio,
1669 unsigned int bytes_done, int err)
1670 #endif
1671 {
1672 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1673 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1674 struct extent_state *state = bio->bi_private;
1675 struct extent_io_tree *tree = state->tree;
1676 u64 start;
1677 u64 end;
1678
1679 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1680 if (bio->bi_size)
1681 return 1;
1682 #endif
1683
1684 do {
1685 struct page *page = bvec->bv_page;
1686 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1687 bvec->bv_offset;
1688 end = start + bvec->bv_len - 1;
1689
1690 if (--bvec >= bio->bi_io_vec)
1691 prefetchw(&bvec->bv_page->flags);
1692
1693 if (uptodate) {
1694 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1695 } else {
1696 ClearPageUptodate(page);
1697 SetPageError(page);
1698 }
1699
1700 unlock_extent(tree, start, end, GFP_ATOMIC);
1701
1702 } while (bvec >= bio->bi_io_vec);
1703
1704 bio_put(bio);
1705 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1706 return 0;
1707 #endif
1708 }
1709
1710 static struct bio *
1711 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1712 gfp_t gfp_flags)
1713 {
1714 struct bio *bio;
1715
1716 bio = bio_alloc(gfp_flags, nr_vecs);
1717
1718 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1719 while (!bio && (nr_vecs /= 2))
1720 bio = bio_alloc(gfp_flags, nr_vecs);
1721 }
1722
1723 if (bio) {
1724 bio->bi_size = 0;
1725 bio->bi_bdev = bdev;
1726 bio->bi_sector = first_sector;
1727 }
1728 return bio;
1729 }
1730
1731 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1732 {
1733 int ret = 0;
1734 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1735 struct page *page = bvec->bv_page;
1736 struct extent_io_tree *tree = bio->bi_private;
1737 struct rb_node *node;
1738 struct extent_state *state;
1739 u64 start;
1740 u64 end;
1741
1742 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1743 end = start + bvec->bv_len - 1;
1744
1745 spin_lock_irq(&tree->lock);
1746 node = __etree_search(tree, start, NULL, NULL);
1747 BUG_ON(!node);
1748 state = rb_entry(node, struct extent_state, rb_node);
1749 while(state->end < end) {
1750 node = rb_next(node);
1751 state = rb_entry(node, struct extent_state, rb_node);
1752 }
1753 BUG_ON(state->end != end);
1754 spin_unlock_irq(&tree->lock);
1755
1756 bio->bi_private = state;
1757
1758 bio_get(bio);
1759
1760 if (tree->ops && tree->ops->submit_bio_hook)
1761 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1762 mirror_num);
1763 else
1764 submit_bio(rw, bio);
1765 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1766 ret = -EOPNOTSUPP;
1767 bio_put(bio);
1768 return ret;
1769 }
1770
1771 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1772 struct page *page, sector_t sector,
1773 size_t size, unsigned long offset,
1774 struct block_device *bdev,
1775 struct bio **bio_ret,
1776 unsigned long max_pages,
1777 bio_end_io_t end_io_func,
1778 int mirror_num)
1779 {
1780 int ret = 0;
1781 struct bio *bio;
1782 int nr;
1783
1784 if (bio_ret && *bio_ret) {
1785 bio = *bio_ret;
1786 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1787 (tree->ops && tree->ops->merge_bio_hook &&
1788 tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1789 bio_add_page(bio, page, size, offset) < size) {
1790 ret = submit_one_bio(rw, bio, mirror_num);
1791 bio = NULL;
1792 } else {
1793 return 0;
1794 }
1795 }
1796 nr = bio_get_nr_vecs(bdev);
1797 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1798 if (!bio) {
1799 printk("failed to allocate bio nr %d\n", nr);
1800 }
1801
1802
1803 bio_add_page(bio, page, size, offset);
1804 bio->bi_end_io = end_io_func;
1805 bio->bi_private = tree;
1806
1807 if (bio_ret) {
1808 *bio_ret = bio;
1809 } else {
1810 ret = submit_one_bio(rw, bio, mirror_num);
1811 }
1812
1813 return ret;
1814 }
1815
1816 void set_page_extent_mapped(struct page *page)
1817 {
1818 if (!PagePrivate(page)) {
1819 SetPagePrivate(page);
1820 WARN_ON(!page->mapping->a_ops->invalidatepage);
1821 set_page_private(page, EXTENT_PAGE_PRIVATE);
1822 page_cache_get(page);
1823 }
1824 }
1825
1826 void set_page_extent_head(struct page *page, unsigned long len)
1827 {
1828 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1829 }
1830
1831 /*
1832 * basic readpage implementation. Locked extent state structs are inserted
1833 * into the tree that are removed when the IO is done (by the end_io
1834 * handlers)
1835 */
1836 static int __extent_read_full_page(struct extent_io_tree *tree,
1837 struct page *page,
1838 get_extent_t *get_extent,
1839 struct bio **bio, int mirror_num)
1840 {
1841 struct inode *inode = page->mapping->host;
1842 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1843 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1844 u64 end;
1845 u64 cur = start;
1846 u64 extent_offset;
1847 u64 last_byte = i_size_read(inode);
1848 u64 block_start;
1849 u64 cur_end;
1850 sector_t sector;
1851 struct extent_map *em;
1852 struct block_device *bdev;
1853 int ret;
1854 int nr = 0;
1855 size_t page_offset = 0;
1856 size_t iosize;
1857 size_t blocksize = inode->i_sb->s_blocksize;
1858
1859 set_page_extent_mapped(page);
1860
1861 end = page_end;
1862 lock_extent(tree, start, end, GFP_NOFS);
1863
1864 while (cur <= end) {
1865 if (cur >= last_byte) {
1866 char *userpage;
1867 iosize = PAGE_CACHE_SIZE - page_offset;
1868 userpage = kmap_atomic(page, KM_USER0);
1869 memset(userpage + page_offset, 0, iosize);
1870 flush_dcache_page(page);
1871 kunmap_atomic(userpage, KM_USER0);
1872 set_extent_uptodate(tree, cur, cur + iosize - 1,
1873 GFP_NOFS);
1874 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1875 break;
1876 }
1877 em = get_extent(inode, page, page_offset, cur,
1878 end - cur + 1, 0);
1879 if (IS_ERR(em) || !em) {
1880 SetPageError(page);
1881 unlock_extent(tree, cur, end, GFP_NOFS);
1882 break;
1883 }
1884 extent_offset = cur - em->start;
1885 if (extent_map_end(em) <= cur) {
1886 printk("bad mapping em [%Lu %Lu] cur %Lu\n", em->start, extent_map_end(em), cur);
1887 }
1888 BUG_ON(extent_map_end(em) <= cur);
1889 if (end < cur) {
1890 printk("2bad mapping end %Lu cur %Lu\n", end, cur);
1891 }
1892 BUG_ON(end < cur);
1893
1894 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1895 cur_end = min(extent_map_end(em) - 1, end);
1896 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1897 sector = (em->block_start + extent_offset) >> 9;
1898 bdev = em->bdev;
1899 block_start = em->block_start;
1900 free_extent_map(em);
1901 em = NULL;
1902
1903 /* we've found a hole, just zero and go on */
1904 if (block_start == EXTENT_MAP_HOLE) {
1905 char *userpage;
1906 userpage = kmap_atomic(page, KM_USER0);
1907 memset(userpage + page_offset, 0, iosize);
1908 flush_dcache_page(page);
1909 kunmap_atomic(userpage, KM_USER0);
1910
1911 set_extent_uptodate(tree, cur, cur + iosize - 1,
1912 GFP_NOFS);
1913 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1914 cur = cur + iosize;
1915 page_offset += iosize;
1916 continue;
1917 }
1918 /* the get_extent function already copied into the page */
1919 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1920 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1921 cur = cur + iosize;
1922 page_offset += iosize;
1923 continue;
1924 }
1925 /* we have an inline extent but it didn't get marked up
1926 * to date. Error out
1927 */
1928 if (block_start == EXTENT_MAP_INLINE) {
1929 SetPageError(page);
1930 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1931 cur = cur + iosize;
1932 page_offset += iosize;
1933 continue;
1934 }
1935
1936 ret = 0;
1937 if (tree->ops && tree->ops->readpage_io_hook) {
1938 ret = tree->ops->readpage_io_hook(page, cur,
1939 cur + iosize - 1);
1940 }
1941 if (!ret) {
1942 unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1943 nr -= page->index;
1944 ret = submit_extent_page(READ, tree, page,
1945 sector, iosize, page_offset,
1946 bdev, bio, nr,
1947 end_bio_extent_readpage, mirror_num);
1948 }
1949 if (ret)
1950 SetPageError(page);
1951 cur = cur + iosize;
1952 page_offset += iosize;
1953 nr++;
1954 }
1955 if (!nr) {
1956 if (!PageError(page))
1957 SetPageUptodate(page);
1958 unlock_page(page);
1959 }
1960 return 0;
1961 }
1962
1963 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1964 get_extent_t *get_extent)
1965 {
1966 struct bio *bio = NULL;
1967 int ret;
1968
1969 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1970 if (bio)
1971 submit_one_bio(READ, bio, 0);
1972 return ret;
1973 }
1974 EXPORT_SYMBOL(extent_read_full_page);
1975
1976 /*
1977 * the writepage semantics are similar to regular writepage. extent
1978 * records are inserted to lock ranges in the tree, and as dirty areas
1979 * are found, they are marked writeback. Then the lock bits are removed
1980 * and the end_io handler clears the writeback ranges
1981 */
1982 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1983 void *data)
1984 {
1985 struct inode *inode = page->mapping->host;
1986 struct extent_page_data *epd = data;
1987 struct extent_io_tree *tree = epd->tree;
1988 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1989 u64 delalloc_start;
1990 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1991 u64 end;
1992 u64 cur = start;
1993 u64 extent_offset;
1994 u64 last_byte = i_size_read(inode);
1995 u64 block_start;
1996 u64 iosize;
1997 u64 unlock_start;
1998 sector_t sector;
1999 struct extent_map *em;
2000 struct block_device *bdev;
2001 int ret;
2002 int nr = 0;
2003 size_t page_offset = 0;
2004 size_t blocksize;
2005 loff_t i_size = i_size_read(inode);
2006 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2007 u64 nr_delalloc;
2008 u64 delalloc_end;
2009
2010 WARN_ON(!PageLocked(page));
2011 page_offset = i_size & (PAGE_CACHE_SIZE - 1);
2012 if (page->index > end_index ||
2013 (page->index == end_index && !page_offset)) {
2014 page->mapping->a_ops->invalidatepage(page, 0);
2015 unlock_page(page);
2016 return 0;
2017 }
2018
2019 if (page->index == end_index) {
2020 char *userpage;
2021
2022 userpage = kmap_atomic(page, KM_USER0);
2023 memset(userpage + page_offset, 0,
2024 PAGE_CACHE_SIZE - page_offset);
2025 kunmap_atomic(userpage, KM_USER0);
2026 flush_dcache_page(page);
2027 }
2028 page_offset = 0;
2029
2030 set_page_extent_mapped(page);
2031
2032 delalloc_start = start;
2033 delalloc_end = 0;
2034 while(delalloc_end < page_end) {
2035 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
2036 &delalloc_end,
2037 128 * 1024 * 1024);
2038 if (nr_delalloc == 0) {
2039 delalloc_start = delalloc_end + 1;
2040 continue;
2041 }
2042 tree->ops->fill_delalloc(inode, delalloc_start,
2043 delalloc_end);
2044 clear_extent_bit(tree, delalloc_start,
2045 delalloc_end,
2046 EXTENT_LOCKED | EXTENT_DELALLOC,
2047 1, 0, GFP_NOFS);
2048 delalloc_start = delalloc_end + 1;
2049 }
2050 lock_extent(tree, start, page_end, GFP_NOFS);
2051 unlock_start = start;
2052
2053 end = page_end;
2054 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2055 printk("found delalloc bits after lock_extent\n");
2056 }
2057
2058 if (last_byte <= start) {
2059 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2060 unlock_extent(tree, start, page_end, GFP_NOFS);
2061 if (tree->ops && tree->ops->writepage_end_io_hook)
2062 tree->ops->writepage_end_io_hook(page, start,
2063 page_end, NULL, 1);
2064 unlock_start = page_end + 1;
2065 goto done;
2066 }
2067
2068 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2069 blocksize = inode->i_sb->s_blocksize;
2070
2071 while (cur <= end) {
2072 if (cur >= last_byte) {
2073 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2074 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2075 if (tree->ops && tree->ops->writepage_end_io_hook)
2076 tree->ops->writepage_end_io_hook(page, cur,
2077 page_end, NULL, 1);
2078 unlock_start = page_end + 1;
2079 break;
2080 }
2081 em = epd->get_extent(inode, page, page_offset, cur,
2082 end - cur + 1, 1);
2083 if (IS_ERR(em) || !em) {
2084 SetPageError(page);
2085 break;
2086 }
2087
2088 extent_offset = cur - em->start;
2089 BUG_ON(extent_map_end(em) <= cur);
2090 BUG_ON(end < cur);
2091 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2092 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2093 sector = (em->block_start + extent_offset) >> 9;
2094 bdev = em->bdev;
2095 block_start = em->block_start;
2096 free_extent_map(em);
2097 em = NULL;
2098
2099 if (block_start == EXTENT_MAP_HOLE ||
2100 block_start == EXTENT_MAP_INLINE) {
2101 clear_extent_dirty(tree, cur,
2102 cur + iosize - 1, GFP_NOFS);
2103
2104 unlock_extent(tree, unlock_start, cur + iosize -1,
2105 GFP_NOFS);
2106 if (tree->ops && tree->ops->writepage_end_io_hook)
2107 tree->ops->writepage_end_io_hook(page, cur,
2108 cur + iosize - 1,
2109 NULL, 1);
2110 cur = cur + iosize;
2111 page_offset += iosize;
2112 unlock_start = cur;
2113 continue;
2114 }
2115
2116 /* leave this out until we have a page_mkwrite call */
2117 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2118 EXTENT_DIRTY, 0)) {
2119 cur = cur + iosize;
2120 page_offset += iosize;
2121 continue;
2122 }
2123 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2124 if (tree->ops && tree->ops->writepage_io_hook) {
2125 ret = tree->ops->writepage_io_hook(page, cur,
2126 cur + iosize - 1);
2127 } else {
2128 ret = 0;
2129 }
2130 if (ret) {
2131 SetPageError(page);
2132 } else {
2133 unsigned long max_nr = end_index + 1;
2134 set_range_writeback(tree, cur, cur + iosize - 1);
2135 if (!PageWriteback(page)) {
2136 printk("warning page %lu not writeback, "
2137 "cur %llu end %llu\n", page->index,
2138 (unsigned long long)cur,
2139 (unsigned long long)end);
2140 }
2141
2142 ret = submit_extent_page(WRITE, tree, page, sector,
2143 iosize, page_offset, bdev,
2144 &epd->bio, max_nr,
2145 end_bio_extent_writepage, 0);
2146 if (ret)
2147 SetPageError(page);
2148 }
2149 cur = cur + iosize;
2150 page_offset += iosize;
2151 nr++;
2152 }
2153 done:
2154 if (nr == 0) {
2155 /* make sure the mapping tag for page dirty gets cleared */
2156 set_page_writeback(page);
2157 end_page_writeback(page);
2158 }
2159 if (unlock_start <= page_end)
2160 unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
2161 unlock_page(page);
2162 return 0;
2163 }
2164
2165 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
2166 /* Taken directly from 2.6.23 for 2.6.18 back port */
2167 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2168 void *data);
2169
2170 /**
2171 * write_cache_pages - walk the list of dirty pages of the given address space
2172 * and write all of them.
2173 * @mapping: address space structure to write
2174 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2175 * @writepage: function called for each page
2176 * @data: data passed to writepage function
2177 *
2178 * If a page is already under I/O, write_cache_pages() skips it, even
2179 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2180 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2181 * and msync() need to guarantee that all the data which was dirty at the time
2182 * the call was made get new I/O started against them. If wbc->sync_mode is
2183 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2184 * existing IO to complete.
2185 */
2186 static int write_cache_pages(struct address_space *mapping,
2187 struct writeback_control *wbc, writepage_t writepage,
2188 void *data)
2189 {
2190 struct backing_dev_info *bdi = mapping->backing_dev_info;
2191 int ret = 0;
2192 int done = 0;
2193 struct pagevec pvec;
2194 int nr_pages;
2195 pgoff_t index;
2196 pgoff_t end; /* Inclusive */
2197 int scanned = 0;
2198 int range_whole = 0;
2199
2200 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2201 wbc->encountered_congestion = 1;
2202 return 0;
2203 }
2204
2205 pagevec_init(&pvec, 0);
2206 if (wbc->range_cyclic) {
2207 index = mapping->writeback_index; /* Start from prev offset */
2208 end = -1;
2209 } else {
2210 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2211 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2212 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2213 range_whole = 1;
2214 scanned = 1;
2215 }
2216 retry:
2217 while (!done && (index <= end) &&
2218 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2219 PAGECACHE_TAG_DIRTY,
2220 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2221 unsigned i;
2222
2223 scanned = 1;
2224 for (i = 0; i < nr_pages; i++) {
2225 struct page *page = pvec.pages[i];
2226
2227 /*
2228 * At this point we hold neither mapping->tree_lock nor
2229 * lock on the page itself: the page may be truncated or
2230 * invalidated (changing page->mapping to NULL), or even
2231 * swizzled back from swapper_space to tmpfs file
2232 * mapping
2233 */
2234 lock_page(page);
2235
2236 if (unlikely(page->mapping != mapping)) {
2237 unlock_page(page);
2238 continue;
2239 }
2240
2241 if (!wbc->range_cyclic && page->index > end) {
2242 done = 1;
2243 unlock_page(page);
2244 continue;
2245 }
2246
2247 if (wbc->sync_mode != WB_SYNC_NONE)
2248 wait_on_page_writeback(page);
2249
2250 if (PageWriteback(page) ||
2251 !clear_page_dirty_for_io(page)) {
2252 unlock_page(page);
2253 continue;
2254 }
2255
2256 ret = (*writepage)(page, wbc, data);
2257
2258 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2259 unlock_page(page);
2260 ret = 0;
2261 }
2262 if (ret || (--(wbc->nr_to_write) <= 0))
2263 done = 1;
2264 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2265 wbc->encountered_congestion = 1;
2266 done = 1;
2267 }
2268 }
2269 pagevec_release(&pvec);
2270 cond_resched();
2271 }
2272 if (!scanned && !done) {
2273 /*
2274 * We hit the last page and there is more work to be done: wrap
2275 * back to the start of the file
2276 */
2277 scanned = 1;
2278 index = 0;
2279 goto retry;
2280 }
2281 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2282 mapping->writeback_index = index;
2283 return ret;
2284 }
2285 #endif
2286
2287 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2288 get_extent_t *get_extent,
2289 struct writeback_control *wbc)
2290 {
2291 int ret;
2292 struct address_space *mapping = page->mapping;
2293 struct extent_page_data epd = {
2294 .bio = NULL,
2295 .tree = tree,
2296 .get_extent = get_extent,
2297 };
2298 struct writeback_control wbc_writepages = {
2299 .bdi = wbc->bdi,
2300 .sync_mode = WB_SYNC_NONE,
2301 .older_than_this = NULL,
2302 .nr_to_write = 64,
2303 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2304 .range_end = (loff_t)-1,
2305 };
2306
2307
2308 ret = __extent_writepage(page, wbc, &epd);
2309
2310 write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2311 if (epd.bio) {
2312 submit_one_bio(WRITE, epd.bio, 0);
2313 }
2314 return ret;
2315 }
2316 EXPORT_SYMBOL(extent_write_full_page);
2317
2318
2319 int extent_writepages(struct extent_io_tree *tree,
2320 struct address_space *mapping,
2321 get_extent_t *get_extent,
2322 struct writeback_control *wbc)
2323 {
2324 int ret = 0;
2325 struct extent_page_data epd = {
2326 .bio = NULL,
2327 .tree = tree,
2328 .get_extent = get_extent,
2329 };
2330
2331 ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2332 if (epd.bio) {
2333 submit_one_bio(WRITE, epd.bio, 0);
2334 }
2335 return ret;
2336 }
2337 EXPORT_SYMBOL(extent_writepages);
2338
2339 int extent_readpages(struct extent_io_tree *tree,
2340 struct address_space *mapping,
2341 struct list_head *pages, unsigned nr_pages,
2342 get_extent_t get_extent)
2343 {
2344 struct bio *bio = NULL;
2345 unsigned page_idx;
2346 struct pagevec pvec;
2347
2348 pagevec_init(&pvec, 0);
2349 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2350 struct page *page = list_entry(pages->prev, struct page, lru);
2351
2352 prefetchw(&page->flags);
2353 list_del(&page->lru);
2354 /*
2355 * what we want to do here is call add_to_page_cache_lru,
2356 * but that isn't exported, so we reproduce it here
2357 */
2358 if (!add_to_page_cache(page, mapping,
2359 page->index, GFP_KERNEL)) {
2360
2361 /* open coding of lru_cache_add, also not exported */
2362 page_cache_get(page);
2363 if (!pagevec_add(&pvec, page))
2364 __pagevec_lru_add(&pvec);
2365 __extent_read_full_page(tree, page, get_extent,
2366 &bio, 0);
2367 }
2368 page_cache_release(page);
2369 }
2370 if (pagevec_count(&pvec))
2371 __pagevec_lru_add(&pvec);
2372 BUG_ON(!list_empty(pages));
2373 if (bio)
2374 submit_one_bio(READ, bio, 0);
2375 return 0;
2376 }
2377 EXPORT_SYMBOL(extent_readpages);
2378
2379 /*
2380 * basic invalidatepage code, this waits on any locked or writeback
2381 * ranges corresponding to the page, and then deletes any extent state
2382 * records from the tree
2383 */
2384 int extent_invalidatepage(struct extent_io_tree *tree,
2385 struct page *page, unsigned long offset)
2386 {
2387 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2388 u64 end = start + PAGE_CACHE_SIZE - 1;
2389 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2390
2391 start += (offset + blocksize -1) & ~(blocksize - 1);
2392 if (start > end)
2393 return 0;
2394
2395 lock_extent(tree, start, end, GFP_NOFS);
2396 wait_on_extent_writeback(tree, start, end);
2397 clear_extent_bit(tree, start, end,
2398 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2399 1, 1, GFP_NOFS);
2400 return 0;
2401 }
2402 EXPORT_SYMBOL(extent_invalidatepage);
2403
2404 /*
2405 * simple commit_write call, set_range_dirty is used to mark both
2406 * the pages and the extent records as dirty
2407 */
2408 int extent_commit_write(struct extent_io_tree *tree,
2409 struct inode *inode, struct page *page,
2410 unsigned from, unsigned to)
2411 {
2412 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2413
2414 set_page_extent_mapped(page);
2415 set_page_dirty(page);
2416
2417 if (pos > inode->i_size) {
2418 i_size_write(inode, pos);
2419 mark_inode_dirty(inode);
2420 }
2421 return 0;
2422 }
2423 EXPORT_SYMBOL(extent_commit_write);
2424
2425 int extent_prepare_write(struct extent_io_tree *tree,
2426 struct inode *inode, struct page *page,
2427 unsigned from, unsigned to, get_extent_t *get_extent)
2428 {
2429 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2430 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2431 u64 block_start;
2432 u64 orig_block_start;
2433 u64 block_end;
2434 u64 cur_end;
2435 struct extent_map *em;
2436 unsigned blocksize = 1 << inode->i_blkbits;
2437 size_t page_offset = 0;
2438 size_t block_off_start;
2439 size_t block_off_end;
2440 int err = 0;
2441 int iocount = 0;
2442 int ret = 0;
2443 int isnew;
2444
2445 set_page_extent_mapped(page);
2446
2447 block_start = (page_start + from) & ~((u64)blocksize - 1);
2448 block_end = (page_start + to - 1) | (blocksize - 1);
2449 orig_block_start = block_start;
2450
2451 lock_extent(tree, page_start, page_end, GFP_NOFS);
2452 while(block_start <= block_end) {
2453 em = get_extent(inode, page, page_offset, block_start,
2454 block_end - block_start + 1, 1);
2455 if (IS_ERR(em) || !em) {
2456 goto err;
2457 }
2458 cur_end = min(block_end, extent_map_end(em) - 1);
2459 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2460 block_off_end = block_off_start + blocksize;
2461 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2462
2463 if (!PageUptodate(page) && isnew &&
2464 (block_off_end > to || block_off_start < from)) {
2465 void *kaddr;
2466
2467 kaddr = kmap_atomic(page, KM_USER0);
2468 if (block_off_end > to)
2469 memset(kaddr + to, 0, block_off_end - to);
2470 if (block_off_start < from)
2471 memset(kaddr + block_off_start, 0,
2472 from - block_off_start);
2473 flush_dcache_page(page);
2474 kunmap_atomic(kaddr, KM_USER0);
2475 }
2476 if ((em->block_start != EXTENT_MAP_HOLE &&
2477 em->block_start != EXTENT_MAP_INLINE) &&
2478 !isnew && !PageUptodate(page) &&
2479 (block_off_end > to || block_off_start < from) &&
2480 !test_range_bit(tree, block_start, cur_end,
2481 EXTENT_UPTODATE, 1)) {
2482 u64 sector;
2483 u64 extent_offset = block_start - em->start;
2484 size_t iosize;
2485 sector = (em->block_start + extent_offset) >> 9;
2486 iosize = (cur_end - block_start + blocksize) &
2487 ~((u64)blocksize - 1);
2488 /*
2489 * we've already got the extent locked, but we
2490 * need to split the state such that our end_bio
2491 * handler can clear the lock.
2492 */
2493 set_extent_bit(tree, block_start,
2494 block_start + iosize - 1,
2495 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2496 ret = submit_extent_page(READ, tree, page,
2497 sector, iosize, page_offset, em->bdev,
2498 NULL, 1,
2499 end_bio_extent_preparewrite, 0);
2500 iocount++;
2501 block_start = block_start + iosize;
2502 } else {
2503 set_extent_uptodate(tree, block_start, cur_end,
2504 GFP_NOFS);
2505 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2506 block_start = cur_end + 1;
2507 }
2508 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2509 free_extent_map(em);
2510 }
2511 if (iocount) {
2512 wait_extent_bit(tree, orig_block_start,
2513 block_end, EXTENT_LOCKED);
2514 }
2515 check_page_uptodate(tree, page);
2516 err:
2517 /* FIXME, zero out newly allocated blocks on error */
2518 return err;
2519 }
2520 EXPORT_SYMBOL(extent_prepare_write);
2521
2522 /*
2523 * a helper for releasepage, this tests for areas of the page that
2524 * are locked or under IO and drops the related state bits if it is safe
2525 * to drop the page.
2526 */
2527 int try_release_extent_state(struct extent_map_tree *map,
2528 struct extent_io_tree *tree, struct page *page,
2529 gfp_t mask)
2530 {
2531 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2532 u64 end = start + PAGE_CACHE_SIZE - 1;
2533 int ret = 1;
2534
2535 if (test_range_bit(tree, start, end, EXTENT_IOBITS, 0))
2536 ret = 0;
2537 else {
2538 if ((mask & GFP_NOFS) == GFP_NOFS)
2539 mask = GFP_NOFS;
2540 clear_extent_bit(tree, start, end, EXTENT_UPTODATE,
2541 1, 1, mask);
2542 }
2543 return ret;
2544 }
2545 EXPORT_SYMBOL(try_release_extent_state);
2546
2547 /*
2548 * a helper for releasepage. As long as there are no locked extents
2549 * in the range corresponding to the page, both state records and extent
2550 * map records are removed
2551 */
2552 int try_release_extent_mapping(struct extent_map_tree *map,
2553 struct extent_io_tree *tree, struct page *page,
2554 gfp_t mask)
2555 {
2556 struct extent_map *em;
2557 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2558 u64 end = start + PAGE_CACHE_SIZE - 1;
2559
2560 if ((mask & __GFP_WAIT) &&
2561 page->mapping->host->i_size > 16 * 1024 * 1024) {
2562 u64 len;
2563 while (start <= end) {
2564 len = end - start + 1;
2565 spin_lock(&map->lock);
2566 em = lookup_extent_mapping(map, start, len);
2567 if (!em || IS_ERR(em)) {
2568 spin_unlock(&map->lock);
2569 break;
2570 }
2571 if (em->start != start) {
2572 spin_unlock(&map->lock);
2573 free_extent_map(em);
2574 break;
2575 }
2576 if (!test_range_bit(tree, em->start,
2577 extent_map_end(em) - 1,
2578 EXTENT_LOCKED, 0)) {
2579 remove_extent_mapping(map, em);
2580 /* once for the rb tree */
2581 free_extent_map(em);
2582 }
2583 start = extent_map_end(em);
2584 spin_unlock(&map->lock);
2585
2586 /* once for us */
2587 free_extent_map(em);
2588 }
2589 }
2590 return try_release_extent_state(map, tree, page, mask);
2591 }
2592 EXPORT_SYMBOL(try_release_extent_mapping);
2593
2594 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2595 get_extent_t *get_extent)
2596 {
2597 struct inode *inode = mapping->host;
2598 u64 start = iblock << inode->i_blkbits;
2599 sector_t sector = 0;
2600 struct extent_map *em;
2601
2602 em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2603 if (!em || IS_ERR(em))
2604 return 0;
2605
2606 if (em->block_start == EXTENT_MAP_INLINE ||
2607 em->block_start == EXTENT_MAP_HOLE)
2608 goto out;
2609
2610 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2611 out:
2612 free_extent_map(em);
2613 return sector;
2614 }
2615
2616 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2617 {
2618 if (list_empty(&eb->lru)) {
2619 extent_buffer_get(eb);
2620 list_add(&eb->lru, &tree->buffer_lru);
2621 tree->lru_size++;
2622 if (tree->lru_size >= BUFFER_LRU_MAX) {
2623 struct extent_buffer *rm;
2624 rm = list_entry(tree->buffer_lru.prev,
2625 struct extent_buffer, lru);
2626 tree->lru_size--;
2627 list_del_init(&rm->lru);
2628 free_extent_buffer(rm);
2629 }
2630 } else
2631 list_move(&eb->lru, &tree->buffer_lru);
2632 return 0;
2633 }
2634 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2635 u64 start, unsigned long len)
2636 {
2637 struct list_head *lru = &tree->buffer_lru;
2638 struct list_head *cur = lru->next;
2639 struct extent_buffer *eb;
2640
2641 if (list_empty(lru))
2642 return NULL;
2643
2644 do {
2645 eb = list_entry(cur, struct extent_buffer, lru);
2646 if (eb->start == start && eb->len == len) {
2647 extent_buffer_get(eb);
2648 return eb;
2649 }
2650 cur = cur->next;
2651 } while (cur != lru);
2652 return NULL;
2653 }
2654
2655 static inline unsigned long num_extent_pages(u64 start, u64 len)
2656 {
2657 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2658 (start >> PAGE_CACHE_SHIFT);
2659 }
2660
2661 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2662 unsigned long i)
2663 {
2664 struct page *p;
2665 struct address_space *mapping;
2666
2667 if (i == 0)
2668 return eb->first_page;
2669 i += eb->start >> PAGE_CACHE_SHIFT;
2670 mapping = eb->first_page->mapping;
2671 read_lock_irq(&mapping->tree_lock);
2672 p = radix_tree_lookup(&mapping->page_tree, i);
2673 read_unlock_irq(&mapping->tree_lock);
2674 return p;
2675 }
2676
2677 int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2678 {
2679 unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2680 struct page *page;
2681 unsigned long i;
2682
2683 if (num_pages == 1)
2684 return 0;
2685 for (i = 1; i < num_pages; i++) {
2686 page = extent_buffer_page(eb, i);
2687 page_cache_release(page);
2688 }
2689 return 0;
2690 }
2691
2692
2693 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2694 unsigned long len)
2695 {
2696 struct list_head *lru = &tree->buffer_lru;
2697 struct list_head *cur = lru->next;
2698 struct extent_buffer *eb;
2699 int found = 0;
2700
2701 spin_lock(&tree->lru_lock);
2702 if (list_empty(lru))
2703 goto out;
2704
2705 do {
2706 eb = list_entry(cur, struct extent_buffer, lru);
2707 if (eb->start <= start && eb->start + eb->len > start) {
2708 eb->flags &= ~EXTENT_UPTODATE;
2709 }
2710 cur = cur->next;
2711 } while (cur != lru);
2712 out:
2713 spin_unlock(&tree->lru_lock);
2714 return found;
2715 }
2716
2717 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2718 u64 start,
2719 unsigned long len,
2720 gfp_t mask)
2721 {
2722 struct extent_buffer *eb = NULL;
2723 unsigned long flags;
2724
2725 spin_lock(&tree->lru_lock);
2726 eb = find_lru(tree, start, len);
2727 spin_unlock(&tree->lru_lock);
2728 if (eb) {
2729 return eb;
2730 }
2731
2732 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2733 INIT_LIST_HEAD(&eb->lru);
2734 eb->start = start;
2735 eb->len = len;
2736 spin_lock_irqsave(&leak_lock, flags);
2737 list_add(&eb->leak_list, &buffers);
2738 spin_unlock_irqrestore(&leak_lock, flags);
2739 atomic_set(&eb->refs, 1);
2740
2741 return eb;
2742 }
2743
2744 static void __free_extent_buffer(struct extent_buffer *eb)
2745 {
2746 unsigned long flags;
2747 spin_lock_irqsave(&leak_lock, flags);
2748 list_del(&eb->leak_list);
2749 spin_unlock_irqrestore(&leak_lock, flags);
2750 kmem_cache_free(extent_buffer_cache, eb);
2751 }
2752
2753 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2754 u64 start, unsigned long len,
2755 struct page *page0,
2756 gfp_t mask)
2757 {
2758 unsigned long num_pages = num_extent_pages(start, len);
2759 unsigned long i;
2760 unsigned long index = start >> PAGE_CACHE_SHIFT;
2761 struct extent_buffer *eb;
2762 struct page *p;
2763 struct address_space *mapping = tree->mapping;
2764 int uptodate = 1;
2765
2766 eb = __alloc_extent_buffer(tree, start, len, mask);
2767 if (!eb)
2768 return NULL;
2769
2770 if (eb->flags & EXTENT_BUFFER_FILLED)
2771 goto lru_add;
2772
2773 if (page0) {
2774 eb->first_page = page0;
2775 i = 1;
2776 index++;
2777 page_cache_get(page0);
2778 mark_page_accessed(page0);
2779 set_page_extent_mapped(page0);
2780 set_page_extent_head(page0, len);
2781 uptodate = PageUptodate(page0);
2782 } else {
2783 i = 0;
2784 }
2785 for (; i < num_pages; i++, index++) {
2786 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2787 if (!p) {
2788 WARN_ON(1);
2789 goto fail;
2790 }
2791 set_page_extent_mapped(p);
2792 mark_page_accessed(p);
2793 if (i == 0) {
2794 eb->first_page = p;
2795 set_page_extent_head(p, len);
2796 } else {
2797 set_page_private(p, EXTENT_PAGE_PRIVATE);
2798 }
2799 if (!PageUptodate(p))
2800 uptodate = 0;
2801 unlock_page(p);
2802 }
2803 if (uptodate)
2804 eb->flags |= EXTENT_UPTODATE;
2805 eb->flags |= EXTENT_BUFFER_FILLED;
2806
2807 lru_add:
2808 spin_lock(&tree->lru_lock);
2809 add_lru(tree, eb);
2810 spin_unlock(&tree->lru_lock);
2811 return eb;
2812
2813 fail:
2814 spin_lock(&tree->lru_lock);
2815 list_del_init(&eb->lru);
2816 spin_unlock(&tree->lru_lock);
2817 if (!atomic_dec_and_test(&eb->refs))
2818 return NULL;
2819 for (index = 1; index < i; index++) {
2820 page_cache_release(extent_buffer_page(eb, index));
2821 }
2822 if (i > 0)
2823 page_cache_release(extent_buffer_page(eb, 0));
2824 __free_extent_buffer(eb);
2825 return NULL;
2826 }
2827 EXPORT_SYMBOL(alloc_extent_buffer);
2828
2829 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2830 u64 start, unsigned long len,
2831 gfp_t mask)
2832 {
2833 unsigned long num_pages = num_extent_pages(start, len);
2834 unsigned long i;
2835 unsigned long index = start >> PAGE_CACHE_SHIFT;
2836 struct extent_buffer *eb;
2837 struct page *p;
2838 struct address_space *mapping = tree->mapping;
2839 int uptodate = 1;
2840
2841 eb = __alloc_extent_buffer(tree, start, len, mask);
2842 if (!eb)
2843 return NULL;
2844
2845 if (eb->flags & EXTENT_BUFFER_FILLED)
2846 goto lru_add;
2847
2848 for (i = 0; i < num_pages; i++, index++) {
2849 p = find_get_page(mapping, index);
2850 if (!p) {
2851 goto fail;
2852 }
2853 if (TestSetPageLocked(p)) {
2854 page_cache_release(p);
2855 goto fail;
2856 }
2857
2858 set_page_extent_mapped(p);
2859 mark_page_accessed(p);
2860
2861 if (i == 0) {
2862 eb->first_page = p;
2863 set_page_extent_head(p, len);
2864 } else {
2865 set_page_private(p, EXTENT_PAGE_PRIVATE);
2866 }
2867
2868 if (!PageUptodate(p))
2869 uptodate = 0;
2870 unlock_page(p);
2871 }
2872 if (uptodate)
2873 eb->flags |= EXTENT_UPTODATE;
2874 eb->flags |= EXTENT_BUFFER_FILLED;
2875
2876 lru_add:
2877 spin_lock(&tree->lru_lock);
2878 add_lru(tree, eb);
2879 spin_unlock(&tree->lru_lock);
2880 return eb;
2881 fail:
2882 spin_lock(&tree->lru_lock);
2883 list_del_init(&eb->lru);
2884 spin_unlock(&tree->lru_lock);
2885 if (!atomic_dec_and_test(&eb->refs))
2886 return NULL;
2887 for (index = 1; index < i; index++) {
2888 page_cache_release(extent_buffer_page(eb, index));
2889 }
2890 if (i > 0)
2891 page_cache_release(extent_buffer_page(eb, 0));
2892 __free_extent_buffer(eb);
2893 return NULL;
2894 }
2895 EXPORT_SYMBOL(find_extent_buffer);
2896
2897 void free_extent_buffer(struct extent_buffer *eb)
2898 {
2899 unsigned long i;
2900 unsigned long num_pages;
2901
2902 if (!eb)
2903 return;
2904
2905 if (!atomic_dec_and_test(&eb->refs))
2906 return;
2907
2908 WARN_ON(!list_empty(&eb->lru));
2909 num_pages = num_extent_pages(eb->start, eb->len);
2910
2911 for (i = 1; i < num_pages; i++) {
2912 page_cache_release(extent_buffer_page(eb, i));
2913 }
2914 page_cache_release(extent_buffer_page(eb, 0));
2915 __free_extent_buffer(eb);
2916 }
2917 EXPORT_SYMBOL(free_extent_buffer);
2918
2919 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2920 struct extent_buffer *eb)
2921 {
2922 int set;
2923 unsigned long i;
2924 unsigned long num_pages;
2925 struct page *page;
2926
2927 u64 start = eb->start;
2928 u64 end = start + eb->len - 1;
2929
2930 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2931 num_pages = num_extent_pages(eb->start, eb->len);
2932
2933 for (i = 0; i < num_pages; i++) {
2934 page = extent_buffer_page(eb, i);
2935 if (i == 0)
2936 set_page_extent_head(page, eb->len);
2937 else
2938 set_page_private(page, EXTENT_PAGE_PRIVATE);
2939
2940 /*
2941 * if we're on the last page or the first page and the
2942 * block isn't aligned on a page boundary, do extra checks
2943 * to make sure we don't clean page that is partially dirty
2944 */
2945 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2946 ((i == num_pages - 1) &&
2947 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2948 start = (u64)page->index << PAGE_CACHE_SHIFT;
2949 end = start + PAGE_CACHE_SIZE - 1;
2950 if (test_range_bit(tree, start, end,
2951 EXTENT_DIRTY, 0)) {
2952 continue;
2953 }
2954 }
2955 clear_page_dirty_for_io(page);
2956 read_lock_irq(&page->mapping->tree_lock);
2957 if (!PageDirty(page)) {
2958 radix_tree_tag_clear(&page->mapping->page_tree,
2959 page_index(page),
2960 PAGECACHE_TAG_DIRTY);
2961 }
2962 read_unlock_irq(&page->mapping->tree_lock);
2963 }
2964 return 0;
2965 }
2966 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2967
2968 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2969 struct extent_buffer *eb)
2970 {
2971 return wait_on_extent_writeback(tree, eb->start,
2972 eb->start + eb->len - 1);
2973 }
2974 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2975
2976 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2977 struct extent_buffer *eb)
2978 {
2979 unsigned long i;
2980 unsigned long num_pages;
2981
2982 num_pages = num_extent_pages(eb->start, eb->len);
2983 for (i = 0; i < num_pages; i++) {
2984 struct page *page = extent_buffer_page(eb, i);
2985 /* writepage may need to do something special for the
2986 * first page, we have to make sure page->private is
2987 * properly set. releasepage may drop page->private
2988 * on us if the page isn't already dirty.
2989 */
2990 if (i == 0) {
2991 set_page_extent_head(page, eb->len);
2992 } else if (PagePrivate(page) &&
2993 page->private != EXTENT_PAGE_PRIVATE) {
2994 set_page_extent_mapped(page);
2995 }
2996 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2997 }
2998 return set_extent_dirty(tree, eb->start,
2999 eb->start + eb->len - 1, GFP_NOFS);
3000 }
3001 EXPORT_SYMBOL(set_extent_buffer_dirty);
3002
3003 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3004 struct extent_buffer *eb)
3005 {
3006 unsigned long i;
3007 struct page *page;
3008 unsigned long num_pages;
3009
3010 num_pages = num_extent_pages(eb->start, eb->len);
3011 eb->flags &= ~EXTENT_UPTODATE;
3012
3013 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3014 GFP_NOFS);
3015 for (i = 0; i < num_pages; i++) {
3016 page = extent_buffer_page(eb, i);
3017 ClearPageUptodate(page);
3018 }
3019 return 0;
3020 }
3021
3022 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3023 struct extent_buffer *eb)
3024 {
3025 unsigned long i;
3026 struct page *page;
3027 unsigned long num_pages;
3028
3029 num_pages = num_extent_pages(eb->start, eb->len);
3030
3031 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3032 GFP_NOFS);
3033 for (i = 0; i < num_pages; i++) {
3034 page = extent_buffer_page(eb, i);
3035 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3036 ((i == num_pages - 1) &&
3037 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3038 check_page_uptodate(tree, page);
3039 continue;
3040 }
3041 SetPageUptodate(page);
3042 }
3043 return 0;
3044 }
3045 EXPORT_SYMBOL(set_extent_buffer_uptodate);
3046
3047 int extent_range_uptodate(struct extent_io_tree *tree,
3048 u64 start, u64 end)
3049 {
3050 struct page *page;
3051 int ret;
3052 int pg_uptodate = 1;
3053 int uptodate;
3054 unsigned long index;
3055
3056 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
3057 if (ret)
3058 return 1;
3059 while(start <= end) {
3060 index = start >> PAGE_CACHE_SHIFT;
3061 page = find_get_page(tree->mapping, index);
3062 uptodate = PageUptodate(page);
3063 page_cache_release(page);
3064 if (!uptodate) {
3065 pg_uptodate = 0;
3066 break;
3067 }
3068 start += PAGE_CACHE_SIZE;
3069 }
3070 return pg_uptodate;
3071 }
3072
3073 int extent_buffer_uptodate(struct extent_io_tree *tree,
3074 struct extent_buffer *eb)
3075 {
3076 int ret = 0;
3077 unsigned long num_pages;
3078 unsigned long i;
3079 struct page *page;
3080 int pg_uptodate = 1;
3081
3082 if (eb->flags & EXTENT_UPTODATE)
3083 return 1;
3084
3085 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3086 EXTENT_UPTODATE, 1);
3087 if (ret)
3088 return ret;
3089
3090 num_pages = num_extent_pages(eb->start, eb->len);
3091 for (i = 0; i < num_pages; i++) {
3092 page = extent_buffer_page(eb, i);
3093 if (!PageUptodate(page)) {
3094 pg_uptodate = 0;
3095 break;
3096 }
3097 }
3098 return pg_uptodate;
3099 }
3100 EXPORT_SYMBOL(extent_buffer_uptodate);
3101
3102 int read_extent_buffer_pages(struct extent_io_tree *tree,
3103 struct extent_buffer *eb,
3104 u64 start, int wait,
3105 get_extent_t *get_extent, int mirror_num)
3106 {
3107 unsigned long i;
3108 unsigned long start_i;
3109 struct page *page;
3110 int err;
3111 int ret = 0;
3112 int locked_pages = 0;
3113 int all_uptodate = 1;
3114 int inc_all_pages = 0;
3115 unsigned long num_pages;
3116 struct bio *bio = NULL;
3117
3118 if (eb->flags & EXTENT_UPTODATE)
3119 return 0;
3120
3121 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3122 EXTENT_UPTODATE, 1)) {
3123 return 0;
3124 }
3125
3126 if (start) {
3127 WARN_ON(start < eb->start);
3128 start_i = (start >> PAGE_CACHE_SHIFT) -
3129 (eb->start >> PAGE_CACHE_SHIFT);
3130 } else {
3131 start_i = 0;
3132 }
3133
3134 num_pages = num_extent_pages(eb->start, eb->len);
3135 for (i = start_i; i < num_pages; i++) {
3136 page = extent_buffer_page(eb, i);
3137 if (!wait) {
3138 if (TestSetPageLocked(page))
3139 goto unlock_exit;
3140 } else {
3141 lock_page(page);
3142 }
3143 locked_pages++;
3144 if (!PageUptodate(page)) {
3145 all_uptodate = 0;
3146 }
3147 }
3148 if (all_uptodate) {
3149 if (start_i == 0)
3150 eb->flags |= EXTENT_UPTODATE;
3151 goto unlock_exit;
3152 }
3153
3154 for (i = start_i; i < num_pages; i++) {
3155 page = extent_buffer_page(eb, i);
3156 if (inc_all_pages)
3157 page_cache_get(page);
3158 if (!PageUptodate(page)) {
3159 if (start_i == 0)
3160 inc_all_pages = 1;
3161 ClearPageError(page);
3162 err = __extent_read_full_page(tree, page,
3163 get_extent, &bio,
3164 mirror_num);
3165 if (err) {
3166 ret = err;
3167 }
3168 } else {
3169 unlock_page(page);
3170 }
3171 }
3172
3173 if (bio)
3174 submit_one_bio(READ, bio, mirror_num);
3175
3176 if (ret || !wait) {
3177 return ret;
3178 }
3179 for (i = start_i; i < num_pages; i++) {
3180 page = extent_buffer_page(eb, i);
3181 wait_on_page_locked(page);
3182 if (!PageUptodate(page)) {
3183 ret = -EIO;
3184 }
3185 }
3186 if (!ret)
3187 eb->flags |= EXTENT_UPTODATE;
3188 return ret;
3189
3190 unlock_exit:
3191 i = start_i;
3192 while(locked_pages > 0) {
3193 page = extent_buffer_page(eb, i);
3194 i++;
3195 unlock_page(page);
3196 locked_pages--;
3197 }
3198 return ret;
3199 }
3200 EXPORT_SYMBOL(read_extent_buffer_pages);
3201
3202 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3203 unsigned long start,
3204 unsigned long len)
3205 {
3206 size_t cur;
3207 size_t offset;
3208 struct page *page;
3209 char *kaddr;
3210 char *dst = (char *)dstv;
3211 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3212 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3213
3214 WARN_ON(start > eb->len);
3215 WARN_ON(start + len > eb->start + eb->len);
3216
3217 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3218
3219 while(len > 0) {
3220 page = extent_buffer_page(eb, i);
3221
3222 cur = min(len, (PAGE_CACHE_SIZE - offset));
3223 kaddr = kmap_atomic(page, KM_USER1);
3224 memcpy(dst, kaddr + offset, cur);
3225 kunmap_atomic(kaddr, KM_USER1);
3226
3227 dst += cur;
3228 len -= cur;
3229 offset = 0;
3230 i++;
3231 }
3232 }
3233 EXPORT_SYMBOL(read_extent_buffer);
3234
3235 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3236 unsigned long min_len, char **token, char **map,
3237 unsigned long *map_start,
3238 unsigned long *map_len, int km)
3239 {
3240 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3241 char *kaddr;
3242 struct page *p;
3243 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3244 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3245 unsigned long end_i = (start_offset + start + min_len - 1) >>
3246 PAGE_CACHE_SHIFT;
3247
3248 if (i != end_i)
3249 return -EINVAL;
3250
3251 if (i == 0) {
3252 offset = start_offset;
3253 *map_start = 0;
3254 } else {
3255 offset = 0;
3256 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3257 }
3258 if (start + min_len > eb->len) {
3259 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3260 WARN_ON(1);
3261 }
3262
3263 p = extent_buffer_page(eb, i);
3264 kaddr = kmap_atomic(p, km);
3265 *token = kaddr;
3266 *map = kaddr + offset;
3267 *map_len = PAGE_CACHE_SIZE - offset;
3268 return 0;
3269 }
3270 EXPORT_SYMBOL(map_private_extent_buffer);
3271
3272 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3273 unsigned long min_len,
3274 char **token, char **map,
3275 unsigned long *map_start,
3276 unsigned long *map_len, int km)
3277 {
3278 int err;
3279 int save = 0;
3280 if (eb->map_token) {
3281 unmap_extent_buffer(eb, eb->map_token, km);
3282 eb->map_token = NULL;
3283 save = 1;
3284 }
3285 err = map_private_extent_buffer(eb, start, min_len, token, map,
3286 map_start, map_len, km);
3287 if (!err && save) {
3288 eb->map_token = *token;
3289 eb->kaddr = *map;
3290 eb->map_start = *map_start;
3291 eb->map_len = *map_len;
3292 }
3293 return err;
3294 }
3295 EXPORT_SYMBOL(map_extent_buffer);
3296
3297 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3298 {
3299 kunmap_atomic(token, km);
3300 }
3301 EXPORT_SYMBOL(unmap_extent_buffer);
3302
3303 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3304 unsigned long start,
3305 unsigned long len)
3306 {
3307 size_t cur;
3308 size_t offset;
3309 struct page *page;
3310 char *kaddr;
3311 char *ptr = (char *)ptrv;
3312 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3313 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3314 int ret = 0;
3315
3316 WARN_ON(start > eb->len);
3317 WARN_ON(start + len > eb->start + eb->len);
3318
3319 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3320
3321 while(len > 0) {
3322 page = extent_buffer_page(eb, i);
3323
3324 cur = min(len, (PAGE_CACHE_SIZE - offset));
3325
3326 kaddr = kmap_atomic(page, KM_USER0);
3327 ret = memcmp(ptr, kaddr + offset, cur);
3328 kunmap_atomic(kaddr, KM_USER0);
3329 if (ret)
3330 break;
3331
3332 ptr += cur;
3333 len -= cur;
3334 offset = 0;
3335 i++;
3336 }
3337 return ret;
3338 }
3339 EXPORT_SYMBOL(memcmp_extent_buffer);
3340
3341 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3342 unsigned long start, unsigned long len)
3343 {
3344 size_t cur;
3345 size_t offset;
3346 struct page *page;
3347 char *kaddr;
3348 char *src = (char *)srcv;
3349 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3350 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3351
3352 WARN_ON(start > eb->len);
3353 WARN_ON(start + len > eb->start + eb->len);
3354
3355 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3356
3357 while(len > 0) {
3358 page = extent_buffer_page(eb, i);
3359 WARN_ON(!PageUptodate(page));
3360
3361 cur = min(len, PAGE_CACHE_SIZE - offset);
3362 kaddr = kmap_atomic(page, KM_USER1);
3363 memcpy(kaddr + offset, src, cur);
3364 kunmap_atomic(kaddr, KM_USER1);
3365
3366 src += cur;
3367 len -= cur;
3368 offset = 0;
3369 i++;
3370 }
3371 }
3372 EXPORT_SYMBOL(write_extent_buffer);
3373
3374 void memset_extent_buffer(struct extent_buffer *eb, char c,
3375 unsigned long start, unsigned long len)
3376 {
3377 size_t cur;
3378 size_t offset;
3379 struct page *page;
3380 char *kaddr;
3381 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3382 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3383
3384 WARN_ON(start > eb->len);
3385 WARN_ON(start + len > eb->start + eb->len);
3386
3387 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3388
3389 while(len > 0) {
3390 page = extent_buffer_page(eb, i);
3391 WARN_ON(!PageUptodate(page));
3392
3393 cur = min(len, PAGE_CACHE_SIZE - offset);
3394 kaddr = kmap_atomic(page, KM_USER0);
3395 memset(kaddr + offset, c, cur);
3396 kunmap_atomic(kaddr, KM_USER0);
3397
3398 len -= cur;
3399 offset = 0;
3400 i++;
3401 }
3402 }
3403 EXPORT_SYMBOL(memset_extent_buffer);
3404
3405 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3406 unsigned long dst_offset, unsigned long src_offset,
3407 unsigned long len)
3408 {
3409 u64 dst_len = dst->len;
3410 size_t cur;
3411 size_t offset;
3412 struct page *page;
3413 char *kaddr;
3414 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3415 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3416
3417 WARN_ON(src->len != dst_len);
3418
3419 offset = (start_offset + dst_offset) &
3420 ((unsigned long)PAGE_CACHE_SIZE - 1);
3421
3422 while(len > 0) {
3423 page = extent_buffer_page(dst, i);
3424 WARN_ON(!PageUptodate(page));
3425
3426 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3427
3428 kaddr = kmap_atomic(page, KM_USER0);
3429 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3430 kunmap_atomic(kaddr, KM_USER0);
3431
3432 src_offset += cur;
3433 len -= cur;
3434 offset = 0;
3435 i++;
3436 }
3437 }
3438 EXPORT_SYMBOL(copy_extent_buffer);
3439
3440 static void move_pages(struct page *dst_page, struct page *src_page,
3441 unsigned long dst_off, unsigned long src_off,
3442 unsigned long len)
3443 {
3444 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3445 if (dst_page == src_page) {
3446 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3447 } else {
3448 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3449 char *p = dst_kaddr + dst_off + len;
3450 char *s = src_kaddr + src_off + len;
3451
3452 while (len--)
3453 *--p = *--s;
3454
3455 kunmap_atomic(src_kaddr, KM_USER1);
3456 }
3457 kunmap_atomic(dst_kaddr, KM_USER0);
3458 }
3459
3460 static void copy_pages(struct page *dst_page, struct page *src_page,
3461 unsigned long dst_off, unsigned long src_off,
3462 unsigned long len)
3463 {
3464 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3465 char *src_kaddr;
3466
3467 if (dst_page != src_page)
3468 src_kaddr = kmap_atomic(src_page, KM_USER1);
3469 else
3470 src_kaddr = dst_kaddr;
3471
3472 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3473 kunmap_atomic(dst_kaddr, KM_USER0);
3474 if (dst_page != src_page)
3475 kunmap_atomic(src_kaddr, KM_USER1);
3476 }
3477
3478 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3479 unsigned long src_offset, unsigned long len)
3480 {
3481 size_t cur;
3482 size_t dst_off_in_page;
3483 size_t src_off_in_page;
3484 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3485 unsigned long dst_i;
3486 unsigned long src_i;
3487
3488 if (src_offset + len > dst->len) {
3489 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3490 src_offset, len, dst->len);
3491 BUG_ON(1);
3492 }
3493 if (dst_offset + len > dst->len) {
3494 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3495 dst_offset, len, dst->len);
3496 BUG_ON(1);
3497 }
3498
3499 while(len > 0) {
3500 dst_off_in_page = (start_offset + dst_offset) &
3501 ((unsigned long)PAGE_CACHE_SIZE - 1);
3502 src_off_in_page = (start_offset + src_offset) &
3503 ((unsigned long)PAGE_CACHE_SIZE - 1);
3504
3505 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3506 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3507
3508 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3509 src_off_in_page));
3510 cur = min_t(unsigned long, cur,
3511 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3512
3513 copy_pages(extent_buffer_page(dst, dst_i),
3514 extent_buffer_page(dst, src_i),
3515 dst_off_in_page, src_off_in_page, cur);
3516
3517 src_offset += cur;
3518 dst_offset += cur;
3519 len -= cur;
3520 }
3521 }
3522 EXPORT_SYMBOL(memcpy_extent_buffer);
3523
3524 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3525 unsigned long src_offset, unsigned long len)
3526 {
3527 size_t cur;
3528 size_t dst_off_in_page;
3529 size_t src_off_in_page;
3530 unsigned long dst_end = dst_offset + len - 1;
3531 unsigned long src_end = src_offset + len - 1;
3532 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3533 unsigned long dst_i;
3534 unsigned long src_i;
3535
3536 if (src_offset + len > dst->len) {
3537 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3538 src_offset, len, dst->len);
3539 BUG_ON(1);
3540 }
3541 if (dst_offset + len > dst->len) {
3542 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3543 dst_offset, len, dst->len);
3544 BUG_ON(1);
3545 }
3546 if (dst_offset < src_offset) {
3547 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3548 return;
3549 }
3550 while(len > 0) {
3551 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3552 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3553
3554 dst_off_in_page = (start_offset + dst_end) &
3555 ((unsigned long)PAGE_CACHE_SIZE - 1);
3556 src_off_in_page = (start_offset + src_end) &
3557 ((unsigned long)PAGE_CACHE_SIZE - 1);
3558
3559 cur = min_t(unsigned long, len, src_off_in_page + 1);
3560 cur = min(cur, dst_off_in_page + 1);
3561 move_pages(extent_buffer_page(dst, dst_i),
3562 extent_buffer_page(dst, src_i),
3563 dst_off_in_page - cur + 1,
3564 src_off_in_page - cur + 1, cur);
3565
3566 dst_end -= cur;
3567 src_end -= cur;
3568 len -= cur;
3569 }
3570 }
3571 EXPORT_SYMBOL(memmove_extent_buffer);
Linux kernel - Deliverables
This page took 0.177213 seconds and 6 git commands to generate.