]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/btrfs/extent_io.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: Use the extent map cache to find the logical disk block during data retries
[mirror_ubuntu-bionic-kernel.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_bits(struct extent_io_tree *tree, u64 start, u64 end,
797 int bits, gfp_t mask)
798 {
799 return set_extent_bit(tree, start, end, bits, 0, NULL,
800 mask);
801 }
802 EXPORT_SYMBOL(set_extent_bits);
803
804 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
805 int bits, gfp_t mask)
806 {
807 return clear_extent_bit(tree, start, end, bits, 0, 0, mask);
808 }
809 EXPORT_SYMBOL(clear_extent_bits);
810
811 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
812 gfp_t mask)
813 {
814 return set_extent_bit(tree, start, end,
815 EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL,
816 mask);
817 }
818 EXPORT_SYMBOL(set_extent_delalloc);
819
820 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
821 gfp_t mask)
822 {
823 return clear_extent_bit(tree, start, end,
824 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask);
825 }
826 EXPORT_SYMBOL(clear_extent_dirty);
827
828 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
829 gfp_t mask)
830 {
831 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
832 mask);
833 }
834 EXPORT_SYMBOL(set_extent_new);
835
836 int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
837 gfp_t mask)
838 {
839 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
840 }
841 EXPORT_SYMBOL(clear_extent_new);
842
843 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
844 gfp_t mask)
845 {
846 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
847 mask);
848 }
849 EXPORT_SYMBOL(set_extent_uptodate);
850
851 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
852 gfp_t mask)
853 {
854 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
855 }
856 EXPORT_SYMBOL(clear_extent_uptodate);
857
858 int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
859 gfp_t mask)
860 {
861 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
862 0, NULL, mask);
863 }
864 EXPORT_SYMBOL(set_extent_writeback);
865
866 int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
867 gfp_t mask)
868 {
869 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
870 }
871 EXPORT_SYMBOL(clear_extent_writeback);
872
873 int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end)
874 {
875 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
876 }
877 EXPORT_SYMBOL(wait_on_extent_writeback);
878
879 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
880 {
881 int err;
882 u64 failed_start;
883 while (1) {
884 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
885 &failed_start, mask);
886 if (err == -EEXIST && (mask & __GFP_WAIT)) {
887 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
888 start = failed_start;
889 } else {
890 break;
891 }
892 WARN_ON(start > end);
893 }
894 return err;
895 }
896 EXPORT_SYMBOL(lock_extent);
897
898 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end,
899 gfp_t mask)
900 {
901 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
902 }
903 EXPORT_SYMBOL(unlock_extent);
904
905 /*
906 * helper function to set pages and extents in the tree dirty
907 */
908 int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end)
909 {
910 unsigned long index = start >> PAGE_CACHE_SHIFT;
911 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
912 struct page *page;
913
914 while (index <= end_index) {
915 page = find_get_page(tree->mapping, index);
916 BUG_ON(!page);
917 __set_page_dirty_nobuffers(page);
918 page_cache_release(page);
919 index++;
920 }
921 set_extent_dirty(tree, start, end, GFP_NOFS);
922 return 0;
923 }
924 EXPORT_SYMBOL(set_range_dirty);
925
926 /*
927 * helper function to set both pages and extents in the tree writeback
928 */
929 int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
930 {
931 unsigned long index = start >> PAGE_CACHE_SHIFT;
932 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
933 struct page *page;
934
935 while (index <= end_index) {
936 page = find_get_page(tree->mapping, index);
937 BUG_ON(!page);
938 set_page_writeback(page);
939 page_cache_release(page);
940 index++;
941 }
942 set_extent_writeback(tree, start, end, GFP_NOFS);
943 return 0;
944 }
945 EXPORT_SYMBOL(set_range_writeback);
946
947 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
948 u64 *start_ret, u64 *end_ret, int bits)
949 {
950 struct rb_node *node;
951 struct extent_state *state;
952 int ret = 1;
953
954 spin_lock_irq(&tree->lock);
955 /*
956 * this search will find all the extents that end after
957 * our range starts.
958 */
959 node = tree_search(tree, start);
960 if (!node) {
961 goto out;
962 }
963
964 while(1) {
965 state = rb_entry(node, struct extent_state, rb_node);
966 if (state->end >= start && (state->state & bits)) {
967 *start_ret = state->start;
968 *end_ret = state->end;
969 ret = 0;
970 break;
971 }
972 node = rb_next(node);
973 if (!node)
974 break;
975 }
976 out:
977 spin_unlock_irq(&tree->lock);
978 return ret;
979 }
980 EXPORT_SYMBOL(find_first_extent_bit);
981
982 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
983 u64 start, int bits)
984 {
985 struct rb_node *node;
986 struct extent_state *state;
987
988 /*
989 * this search will find all the extents that end after
990 * our range starts.
991 */
992 node = tree_search(tree, start);
993 if (!node) {
994 goto out;
995 }
996
997 while(1) {
998 state = rb_entry(node, struct extent_state, rb_node);
999 if (state->end >= start && (state->state & bits)) {
1000 return state;
1001 }
1002 node = rb_next(node);
1003 if (!node)
1004 break;
1005 }
1006 out:
1007 return NULL;
1008 }
1009 EXPORT_SYMBOL(find_first_extent_bit_state);
1010
1011 u64 find_lock_delalloc_range(struct extent_io_tree *tree,
1012 u64 *start, u64 *end, u64 max_bytes)
1013 {
1014 struct rb_node *node;
1015 struct extent_state *state;
1016 u64 cur_start = *start;
1017 u64 found = 0;
1018 u64 total_bytes = 0;
1019
1020 spin_lock_irq(&tree->lock);
1021 /*
1022 * this search will find all the extents that end after
1023 * our range starts.
1024 */
1025 search_again:
1026 node = tree_search(tree, cur_start);
1027 if (!node) {
1028 if (!found)
1029 *end = (u64)-1;
1030 goto out;
1031 }
1032
1033 while(1) {
1034 state = rb_entry(node, struct extent_state, rb_node);
1035 if (found && state->start != cur_start) {
1036 goto out;
1037 }
1038 if (!(state->state & EXTENT_DELALLOC)) {
1039 if (!found)
1040 *end = state->end;
1041 goto out;
1042 }
1043 if (!found) {
1044 struct extent_state *prev_state;
1045 struct rb_node *prev_node = node;
1046 while(1) {
1047 prev_node = rb_prev(prev_node);
1048 if (!prev_node)
1049 break;
1050 prev_state = rb_entry(prev_node,
1051 struct extent_state,
1052 rb_node);
1053 if (!(prev_state->state & EXTENT_DELALLOC))
1054 break;
1055 state = prev_state;
1056 node = prev_node;
1057 }
1058 }
1059 if (state->state & EXTENT_LOCKED) {
1060 DEFINE_WAIT(wait);
1061 atomic_inc(&state->refs);
1062 prepare_to_wait(&state->wq, &wait,
1063 TASK_UNINTERRUPTIBLE);
1064 spin_unlock_irq(&tree->lock);
1065 schedule();
1066 spin_lock_irq(&tree->lock);
1067 finish_wait(&state->wq, &wait);
1068 free_extent_state(state);
1069 goto search_again;
1070 }
1071 set_state_cb(tree, state, EXTENT_LOCKED);
1072 state->state |= EXTENT_LOCKED;
1073 if (!found)
1074 *start = state->start;
1075 found++;
1076 *end = state->end;
1077 cur_start = state->end + 1;
1078 node = rb_next(node);
1079 if (!node)
1080 break;
1081 total_bytes += state->end - state->start + 1;
1082 if (total_bytes >= max_bytes)
1083 break;
1084 }
1085 out:
1086 spin_unlock_irq(&tree->lock);
1087 return found;
1088 }
1089
1090 u64 count_range_bits(struct extent_io_tree *tree,
1091 u64 *start, u64 search_end, u64 max_bytes,
1092 unsigned long bits)
1093 {
1094 struct rb_node *node;
1095 struct extent_state *state;
1096 u64 cur_start = *start;
1097 u64 total_bytes = 0;
1098 int found = 0;
1099
1100 if (search_end <= cur_start) {
1101 printk("search_end %Lu start %Lu\n", search_end, cur_start);
1102 WARN_ON(1);
1103 return 0;
1104 }
1105
1106 spin_lock_irq(&tree->lock);
1107 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1108 total_bytes = tree->dirty_bytes;
1109 goto out;
1110 }
1111 /*
1112 * this search will find all the extents that end after
1113 * our range starts.
1114 */
1115 node = tree_search(tree, cur_start);
1116 if (!node) {
1117 goto out;
1118 }
1119
1120 while(1) {
1121 state = rb_entry(node, struct extent_state, rb_node);
1122 if (state->start > search_end)
1123 break;
1124 if (state->end >= cur_start && (state->state & bits)) {
1125 total_bytes += min(search_end, state->end) + 1 -
1126 max(cur_start, state->start);
1127 if (total_bytes >= max_bytes)
1128 break;
1129 if (!found) {
1130 *start = state->start;
1131 found = 1;
1132 }
1133 }
1134 node = rb_next(node);
1135 if (!node)
1136 break;
1137 }
1138 out:
1139 spin_unlock_irq(&tree->lock);
1140 return total_bytes;
1141 }
1142 /*
1143 * helper function to lock both pages and extents in the tree.
1144 * pages must be locked first.
1145 */
1146 int lock_range(struct extent_io_tree *tree, u64 start, u64 end)
1147 {
1148 unsigned long index = start >> PAGE_CACHE_SHIFT;
1149 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1150 struct page *page;
1151 int err;
1152
1153 while (index <= end_index) {
1154 page = grab_cache_page(tree->mapping, index);
1155 if (!page) {
1156 err = -ENOMEM;
1157 goto failed;
1158 }
1159 if (IS_ERR(page)) {
1160 err = PTR_ERR(page);
1161 goto failed;
1162 }
1163 index++;
1164 }
1165 lock_extent(tree, start, end, GFP_NOFS);
1166 return 0;
1167
1168 failed:
1169 /*
1170 * we failed above in getting the page at 'index', so we undo here
1171 * up to but not including the page at 'index'
1172 */
1173 end_index = index;
1174 index = start >> PAGE_CACHE_SHIFT;
1175 while (index < end_index) {
1176 page = find_get_page(tree->mapping, index);
1177 unlock_page(page);
1178 page_cache_release(page);
1179 index++;
1180 }
1181 return err;
1182 }
1183 EXPORT_SYMBOL(lock_range);
1184
1185 /*
1186 * helper function to unlock both pages and extents in the tree.
1187 */
1188 int unlock_range(struct extent_io_tree *tree, u64 start, u64 end)
1189 {
1190 unsigned long index = start >> PAGE_CACHE_SHIFT;
1191 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1192 struct page *page;
1193
1194 while (index <= end_index) {
1195 page = find_get_page(tree->mapping, index);
1196 unlock_page(page);
1197 page_cache_release(page);
1198 index++;
1199 }
1200 unlock_extent(tree, start, end, GFP_NOFS);
1201 return 0;
1202 }
1203 EXPORT_SYMBOL(unlock_range);
1204
1205 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1206 {
1207 struct rb_node *node;
1208 struct extent_state *state;
1209 int ret = 0;
1210
1211 spin_lock_irq(&tree->lock);
1212 /*
1213 * this search will find all the extents that end after
1214 * our range starts.
1215 */
1216 node = tree_search(tree, start);
1217 if (!node) {
1218 ret = -ENOENT;
1219 goto out;
1220 }
1221 state = rb_entry(node, struct extent_state, rb_node);
1222 if (state->start != start) {
1223 ret = -ENOENT;
1224 goto out;
1225 }
1226 state->private = private;
1227 out:
1228 spin_unlock_irq(&tree->lock);
1229 return ret;
1230 }
1231
1232 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1233 {
1234 struct rb_node *node;
1235 struct extent_state *state;
1236 int ret = 0;
1237
1238 spin_lock_irq(&tree->lock);
1239 /*
1240 * this search will find all the extents that end after
1241 * our range starts.
1242 */
1243 node = tree_search(tree, start);
1244 if (!node) {
1245 ret = -ENOENT;
1246 goto out;
1247 }
1248 state = rb_entry(node, struct extent_state, rb_node);
1249 if (state->start != start) {
1250 ret = -ENOENT;
1251 goto out;
1252 }
1253 *private = state->private;
1254 out:
1255 spin_unlock_irq(&tree->lock);
1256 return ret;
1257 }
1258
1259 /*
1260 * searches a range in the state tree for a given mask.
1261 * If 'filled' == 1, this returns 1 only if every extent in the tree
1262 * has the bits set. Otherwise, 1 is returned if any bit in the
1263 * range is found set.
1264 */
1265 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1266 int bits, int filled)
1267 {
1268 struct extent_state *state = NULL;
1269 struct rb_node *node;
1270 int bitset = 0;
1271 unsigned long flags;
1272
1273 spin_lock_irqsave(&tree->lock, flags);
1274 node = tree_search(tree, start);
1275 while (node && start <= end) {
1276 state = rb_entry(node, struct extent_state, rb_node);
1277
1278 if (filled && state->start > start) {
1279 bitset = 0;
1280 break;
1281 }
1282
1283 if (state->start > end)
1284 break;
1285
1286 if (state->state & bits) {
1287 bitset = 1;
1288 if (!filled)
1289 break;
1290 } else if (filled) {
1291 bitset = 0;
1292 break;
1293 }
1294 start = state->end + 1;
1295 if (start > end)
1296 break;
1297 node = rb_next(node);
1298 if (!node) {
1299 if (filled)
1300 bitset = 0;
1301 break;
1302 }
1303 }
1304 spin_unlock_irqrestore(&tree->lock, flags);
1305 return bitset;
1306 }
1307 EXPORT_SYMBOL(test_range_bit);
1308
1309 /*
1310 * helper function to set a given page up to date if all the
1311 * extents in the tree for that page are up to date
1312 */
1313 static int check_page_uptodate(struct extent_io_tree *tree,
1314 struct page *page)
1315 {
1316 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1317 u64 end = start + PAGE_CACHE_SIZE - 1;
1318 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1319 SetPageUptodate(page);
1320 return 0;
1321 }
1322
1323 /*
1324 * helper function to unlock a page if all the extents in the tree
1325 * for that page are unlocked
1326 */
1327 static int check_page_locked(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_LOCKED, 0))
1333 unlock_page(page);
1334 return 0;
1335 }
1336
1337 /*
1338 * helper function to end page writeback if all the extents
1339 * in the tree for that page are done with writeback
1340 */
1341 static int check_page_writeback(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_WRITEBACK, 0))
1347 end_page_writeback(page);
1348 return 0;
1349 }
1350
1351 /* lots and lots of room for performance fixes in the end_bio funcs */
1352
1353 /*
1354 * after a writepage IO is done, we need to:
1355 * clear the uptodate bits on error
1356 * clear the writeback bits in the extent tree for this IO
1357 * end_page_writeback if the page has no more pending IO
1358 *
1359 * Scheduling is not allowed, so the extent state tree is expected
1360 * to have one and only one object corresponding to this IO.
1361 */
1362 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1363 static void end_bio_extent_writepage(struct bio *bio, int err)
1364 #else
1365 static int end_bio_extent_writepage(struct bio *bio,
1366 unsigned int bytes_done, int err)
1367 #endif
1368 {
1369 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1370 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1371 struct extent_state *state = bio->bi_private;
1372 struct extent_io_tree *tree = state->tree;
1373 struct rb_node *node;
1374 u64 start;
1375 u64 end;
1376 u64 cur;
1377 int whole_page;
1378 unsigned long flags;
1379
1380 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1381 if (bio->bi_size)
1382 return 1;
1383 #endif
1384 do {
1385 struct page *page = bvec->bv_page;
1386 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1387 bvec->bv_offset;
1388 end = start + bvec->bv_len - 1;
1389
1390 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1391 whole_page = 1;
1392 else
1393 whole_page = 0;
1394
1395 if (--bvec >= bio->bi_io_vec)
1396 prefetchw(&bvec->bv_page->flags);
1397
1398 if (!uptodate) {
1399 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1400 ClearPageUptodate(page);
1401 SetPageError(page);
1402 }
1403
1404 if (tree->ops && tree->ops->writepage_end_io_hook) {
1405 tree->ops->writepage_end_io_hook(page, start, end,
1406 state);
1407 }
1408
1409 /*
1410 * bios can get merged in funny ways, and so we need to
1411 * be careful with the state variable. We know the
1412 * state won't be merged with others because it has
1413 * WRITEBACK set, but we can't be sure each biovec is
1414 * sequential in the file. So, if our cached state
1415 * doesn't match the expected end, search the tree
1416 * for the correct one.
1417 */
1418
1419 spin_lock_irqsave(&tree->lock, flags);
1420 if (!state || state->end != end) {
1421 state = NULL;
1422 node = __etree_search(tree, start, NULL, NULL);
1423 if (node) {
1424 state = rb_entry(node, struct extent_state,
1425 rb_node);
1426 if (state->end != end ||
1427 !(state->state & EXTENT_WRITEBACK))
1428 state = NULL;
1429 }
1430 if (!state) {
1431 spin_unlock_irqrestore(&tree->lock, flags);
1432 clear_extent_writeback(tree, start,
1433 end, GFP_ATOMIC);
1434 goto next_io;
1435 }
1436 }
1437 cur = end;
1438 while(1) {
1439 struct extent_state *clear = state;
1440 cur = state->start;
1441 node = rb_prev(&state->rb_node);
1442 if (node) {
1443 state = rb_entry(node,
1444 struct extent_state,
1445 rb_node);
1446 } else {
1447 state = NULL;
1448 }
1449
1450 clear_state_bit(tree, clear, EXTENT_WRITEBACK,
1451 1, 0);
1452 if (cur == start)
1453 break;
1454 if (cur < start) {
1455 WARN_ON(1);
1456 break;
1457 }
1458 if (!node)
1459 break;
1460 }
1461 /* before releasing the lock, make sure the next state
1462 * variable has the expected bits set and corresponds
1463 * to the correct offsets in the file
1464 */
1465 if (state && (state->end + 1 != start ||
1466 !(state->state & EXTENT_WRITEBACK))) {
1467 state = NULL;
1468 }
1469 spin_unlock_irqrestore(&tree->lock, flags);
1470 next_io:
1471
1472 if (whole_page)
1473 end_page_writeback(page);
1474 else
1475 check_page_writeback(tree, page);
1476 } while (bvec >= bio->bi_io_vec);
1477 bio_put(bio);
1478 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1479 return 0;
1480 #endif
1481 }
1482
1483 /*
1484 * after a readpage IO is done, we need to:
1485 * clear the uptodate bits on error
1486 * set the uptodate bits if things worked
1487 * set the page up to date if all extents in the tree are uptodate
1488 * clear the lock bit in the extent tree
1489 * unlock the page if there are no other extents locked for it
1490 *
1491 * Scheduling is not allowed, so the extent state tree is expected
1492 * to have one and only one object corresponding to this IO.
1493 */
1494 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1495 static void end_bio_extent_readpage(struct bio *bio, int err)
1496 #else
1497 static int end_bio_extent_readpage(struct bio *bio,
1498 unsigned int bytes_done, int err)
1499 #endif
1500 {
1501 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1502 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1503 struct extent_state *state = bio->bi_private;
1504 struct extent_io_tree *tree = state->tree;
1505 struct rb_node *node;
1506 u64 start;
1507 u64 end;
1508 u64 cur;
1509 unsigned long flags;
1510 int whole_page;
1511 int ret;
1512
1513 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1514 if (bio->bi_size)
1515 return 1;
1516 #endif
1517
1518 do {
1519 struct page *page = bvec->bv_page;
1520 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1521 bvec->bv_offset;
1522 end = start + bvec->bv_len - 1;
1523
1524 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1525 whole_page = 1;
1526 else
1527 whole_page = 0;
1528
1529 if (--bvec >= bio->bi_io_vec)
1530 prefetchw(&bvec->bv_page->flags);
1531
1532 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1533 ret = tree->ops->readpage_end_io_hook(page, start, end,
1534 state);
1535 if (ret)
1536 uptodate = 0;
1537 }
1538 if (!uptodate && tree->ops &&
1539 tree->ops->readpage_io_failed_hook) {
1540 ret = tree->ops->readpage_io_failed_hook(bio, page,
1541 start, end, state);
1542 if (ret == 0) {
1543 state = NULL;
1544 uptodate =
1545 test_bit(BIO_UPTODATE, &bio->bi_flags);
1546 continue;
1547 }
1548 }
1549
1550 spin_lock_irqsave(&tree->lock, flags);
1551 if (!state || state->end != end) {
1552 state = NULL;
1553 node = __etree_search(tree, start, NULL, NULL);
1554 if (node) {
1555 state = rb_entry(node, struct extent_state,
1556 rb_node);
1557 if (state->end != end ||
1558 !(state->state & EXTENT_LOCKED))
1559 state = NULL;
1560 }
1561 if (!state) {
1562 spin_unlock_irqrestore(&tree->lock, flags);
1563 if (uptodate)
1564 set_extent_uptodate(tree, start, end,
1565 GFP_ATOMIC);
1566 unlock_extent(tree, start, end, GFP_ATOMIC);
1567 goto next_io;
1568 }
1569 }
1570
1571 cur = end;
1572 while(1) {
1573 struct extent_state *clear = state;
1574 cur = state->start;
1575 node = rb_prev(&state->rb_node);
1576 if (node) {
1577 state = rb_entry(node,
1578 struct extent_state,
1579 rb_node);
1580 } else {
1581 state = NULL;
1582 }
1583 if (uptodate) {
1584 set_state_cb(tree, clear, EXTENT_UPTODATE);
1585 clear->state |= EXTENT_UPTODATE;
1586 }
1587 clear_state_bit(tree, clear, EXTENT_LOCKED,
1588 1, 0);
1589 if (cur == start)
1590 break;
1591 if (cur < start) {
1592 WARN_ON(1);
1593 break;
1594 }
1595 if (!node)
1596 break;
1597 }
1598 /* before releasing the lock, make sure the next state
1599 * variable has the expected bits set and corresponds
1600 * to the correct offsets in the file
1601 */
1602 if (state && (state->end + 1 != start ||
1603 !(state->state & EXTENT_LOCKED))) {
1604 state = NULL;
1605 }
1606 spin_unlock_irqrestore(&tree->lock, flags);
1607 next_io:
1608 if (whole_page) {
1609 if (uptodate) {
1610 SetPageUptodate(page);
1611 } else {
1612 ClearPageUptodate(page);
1613 SetPageError(page);
1614 }
1615 unlock_page(page);
1616 } else {
1617 if (uptodate) {
1618 check_page_uptodate(tree, page);
1619 } else {
1620 ClearPageUptodate(page);
1621 SetPageError(page);
1622 }
1623 check_page_locked(tree, page);
1624 }
1625 } while (bvec >= bio->bi_io_vec);
1626
1627 bio_put(bio);
1628 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1629 return 0;
1630 #endif
1631 }
1632
1633 /*
1634 * IO done from prepare_write is pretty simple, we just unlock
1635 * the structs in the extent tree when done, and set the uptodate bits
1636 * as appropriate.
1637 */
1638 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1639 static void end_bio_extent_preparewrite(struct bio *bio, int err)
1640 #else
1641 static int end_bio_extent_preparewrite(struct bio *bio,
1642 unsigned int bytes_done, int err)
1643 #endif
1644 {
1645 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1646 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1647 struct extent_state *state = bio->bi_private;
1648 struct extent_io_tree *tree = state->tree;
1649 u64 start;
1650 u64 end;
1651
1652 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1653 if (bio->bi_size)
1654 return 1;
1655 #endif
1656
1657 do {
1658 struct page *page = bvec->bv_page;
1659 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1660 bvec->bv_offset;
1661 end = start + bvec->bv_len - 1;
1662
1663 if (--bvec >= bio->bi_io_vec)
1664 prefetchw(&bvec->bv_page->flags);
1665
1666 if (uptodate) {
1667 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1668 } else {
1669 ClearPageUptodate(page);
1670 SetPageError(page);
1671 }
1672
1673 unlock_extent(tree, start, end, GFP_ATOMIC);
1674
1675 } while (bvec >= bio->bi_io_vec);
1676
1677 bio_put(bio);
1678 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1679 return 0;
1680 #endif
1681 }
1682
1683 static struct bio *
1684 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1685 gfp_t gfp_flags)
1686 {
1687 struct bio *bio;
1688
1689 bio = bio_alloc(gfp_flags, nr_vecs);
1690
1691 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1692 while (!bio && (nr_vecs /= 2))
1693 bio = bio_alloc(gfp_flags, nr_vecs);
1694 }
1695
1696 if (bio) {
1697 bio->bi_bdev = bdev;
1698 bio->bi_sector = first_sector;
1699 }
1700 return bio;
1701 }
1702
1703 static int submit_one_bio(int rw, struct bio *bio, int mirror_num)
1704 {
1705 u64 maxsector;
1706 int ret = 0;
1707 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1708 struct page *page = bvec->bv_page;
1709 struct extent_io_tree *tree = bio->bi_private;
1710 struct rb_node *node;
1711 struct extent_state *state;
1712 u64 start;
1713 u64 end;
1714
1715 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1716 end = start + bvec->bv_len - 1;
1717
1718 spin_lock_irq(&tree->lock);
1719 node = __etree_search(tree, start, NULL, NULL);
1720 BUG_ON(!node);
1721 state = rb_entry(node, struct extent_state, rb_node);
1722 while(state->end < end) {
1723 node = rb_next(node);
1724 state = rb_entry(node, struct extent_state, rb_node);
1725 }
1726 BUG_ON(state->end != end);
1727 spin_unlock_irq(&tree->lock);
1728
1729 bio->bi_private = state;
1730
1731 bio_get(bio);
1732
1733 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1734 if (maxsector < bio->bi_sector) {
1735 printk("sector too large max %Lu got %llu\n", maxsector,
1736 (unsigned long long)bio->bi_sector);
1737 WARN_ON(1);
1738 }
1739 if (tree->ops && tree->ops->submit_bio_hook)
1740 tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1741 mirror_num);
1742 else
1743 submit_bio(rw, bio);
1744 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1745 ret = -EOPNOTSUPP;
1746 bio_put(bio);
1747 return ret;
1748 }
1749
1750 static int submit_extent_page(int rw, struct extent_io_tree *tree,
1751 struct page *page, sector_t sector,
1752 size_t size, unsigned long offset,
1753 struct block_device *bdev,
1754 struct bio **bio_ret,
1755 unsigned long max_pages,
1756 bio_end_io_t end_io_func,
1757 int mirror_num)
1758 {
1759 int ret = 0;
1760 struct bio *bio;
1761 int nr;
1762
1763 if (bio_ret && *bio_ret) {
1764 bio = *bio_ret;
1765 if (bio->bi_sector + (bio->bi_size >> 9) != sector ||
1766 (tree->ops && tree->ops->merge_bio_hook &&
1767 tree->ops->merge_bio_hook(page, offset, size, bio)) ||
1768 bio_add_page(bio, page, size, offset) < size) {
1769 ret = submit_one_bio(rw, bio, mirror_num);
1770 bio = NULL;
1771 } else {
1772 return 0;
1773 }
1774 }
1775 nr = bio_get_nr_vecs(bdev);
1776 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
1777 if (!bio) {
1778 printk("failed to allocate bio nr %d\n", nr);
1779 }
1780
1781
1782 bio_add_page(bio, page, size, offset);
1783 bio->bi_end_io = end_io_func;
1784 bio->bi_private = tree;
1785
1786 if (bio_ret) {
1787 *bio_ret = bio;
1788 } else {
1789 ret = submit_one_bio(rw, bio, mirror_num);
1790 }
1791
1792 return ret;
1793 }
1794
1795 void set_page_extent_mapped(struct page *page)
1796 {
1797 if (!PagePrivate(page)) {
1798 SetPagePrivate(page);
1799 WARN_ON(!page->mapping->a_ops->invalidatepage);
1800 set_page_private(page, EXTENT_PAGE_PRIVATE);
1801 page_cache_get(page);
1802 }
1803 }
1804
1805 void set_page_extent_head(struct page *page, unsigned long len)
1806 {
1807 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
1808 }
1809
1810 /*
1811 * basic readpage implementation. Locked extent state structs are inserted
1812 * into the tree that are removed when the IO is done (by the end_io
1813 * handlers)
1814 */
1815 static int __extent_read_full_page(struct extent_io_tree *tree,
1816 struct page *page,
1817 get_extent_t *get_extent,
1818 struct bio **bio, int mirror_num)
1819 {
1820 struct inode *inode = page->mapping->host;
1821 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1822 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1823 u64 end;
1824 u64 cur = start;
1825 u64 extent_offset;
1826 u64 last_byte = i_size_read(inode);
1827 u64 block_start;
1828 u64 cur_end;
1829 sector_t sector;
1830 struct extent_map *em;
1831 struct block_device *bdev;
1832 int ret;
1833 int nr = 0;
1834 size_t page_offset = 0;
1835 size_t iosize;
1836 size_t blocksize = inode->i_sb->s_blocksize;
1837
1838 set_page_extent_mapped(page);
1839
1840 end = page_end;
1841 lock_extent(tree, start, end, GFP_NOFS);
1842
1843 while (cur <= end) {
1844 if (cur >= last_byte) {
1845 char *userpage;
1846 iosize = PAGE_CACHE_SIZE - page_offset;
1847 userpage = kmap_atomic(page, KM_USER0);
1848 memset(userpage + page_offset, 0, iosize);
1849 flush_dcache_page(page);
1850 kunmap_atomic(userpage, KM_USER0);
1851 set_extent_uptodate(tree, cur, cur + iosize - 1,
1852 GFP_NOFS);
1853 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1854 break;
1855 }
1856 em = get_extent(inode, page, page_offset, cur,
1857 end - cur + 1, 0);
1858 if (IS_ERR(em) || !em) {
1859 SetPageError(page);
1860 unlock_extent(tree, cur, end, GFP_NOFS);
1861 break;
1862 }
1863
1864 extent_offset = cur - em->start;
1865 BUG_ON(extent_map_end(em) <= cur);
1866 BUG_ON(end < cur);
1867
1868 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1869 cur_end = min(extent_map_end(em) - 1, end);
1870 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1871 sector = (em->block_start + extent_offset) >> 9;
1872 bdev = em->bdev;
1873 block_start = em->block_start;
1874 free_extent_map(em);
1875 em = NULL;
1876
1877 /* we've found a hole, just zero and go on */
1878 if (block_start == EXTENT_MAP_HOLE) {
1879 char *userpage;
1880 userpage = kmap_atomic(page, KM_USER0);
1881 memset(userpage + page_offset, 0, iosize);
1882 flush_dcache_page(page);
1883 kunmap_atomic(userpage, KM_USER0);
1884
1885 set_extent_uptodate(tree, cur, cur + iosize - 1,
1886 GFP_NOFS);
1887 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1888 cur = cur + iosize;
1889 page_offset += iosize;
1890 continue;
1891 }
1892 /* the get_extent function already copied into the page */
1893 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1894 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1895 cur = cur + iosize;
1896 page_offset += iosize;
1897 continue;
1898 }
1899 /* we have an inline extent but it didn't get marked up
1900 * to date. Error out
1901 */
1902 if (block_start == EXTENT_MAP_INLINE) {
1903 SetPageError(page);
1904 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1905 cur = cur + iosize;
1906 page_offset += iosize;
1907 continue;
1908 }
1909
1910 ret = 0;
1911 if (tree->ops && tree->ops->readpage_io_hook) {
1912 ret = tree->ops->readpage_io_hook(page, cur,
1913 cur + iosize - 1);
1914 }
1915 if (!ret) {
1916 unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
1917 nr -= page->index;
1918 ret = submit_extent_page(READ, tree, page,
1919 sector, iosize, page_offset,
1920 bdev, bio, nr,
1921 end_bio_extent_readpage, mirror_num);
1922 }
1923 if (ret)
1924 SetPageError(page);
1925 cur = cur + iosize;
1926 page_offset += iosize;
1927 nr++;
1928 }
1929 if (!nr) {
1930 if (!PageError(page))
1931 SetPageUptodate(page);
1932 unlock_page(page);
1933 }
1934 return 0;
1935 }
1936
1937 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
1938 get_extent_t *get_extent)
1939 {
1940 struct bio *bio = NULL;
1941 int ret;
1942
1943 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0);
1944 if (bio)
1945 submit_one_bio(READ, bio, 0);
1946 return ret;
1947 }
1948 EXPORT_SYMBOL(extent_read_full_page);
1949
1950 /*
1951 * the writepage semantics are similar to regular writepage. extent
1952 * records are inserted to lock ranges in the tree, and as dirty areas
1953 * are found, they are marked writeback. Then the lock bits are removed
1954 * and the end_io handler clears the writeback ranges
1955 */
1956 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
1957 void *data)
1958 {
1959 struct inode *inode = page->mapping->host;
1960 struct extent_page_data *epd = data;
1961 struct extent_io_tree *tree = epd->tree;
1962 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1963 u64 delalloc_start;
1964 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1965 u64 end;
1966 u64 cur = start;
1967 u64 extent_offset;
1968 u64 last_byte = i_size_read(inode);
1969 u64 block_start;
1970 u64 iosize;
1971 sector_t sector;
1972 struct extent_map *em;
1973 struct block_device *bdev;
1974 int ret;
1975 int nr = 0;
1976 size_t page_offset = 0;
1977 size_t blocksize;
1978 loff_t i_size = i_size_read(inode);
1979 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1980 u64 nr_delalloc;
1981 u64 delalloc_end;
1982
1983 WARN_ON(!PageLocked(page));
1984 if (page->index > end_index) {
1985 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1986 unlock_page(page);
1987 return 0;
1988 }
1989
1990 if (page->index == end_index) {
1991 char *userpage;
1992
1993 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1994
1995 userpage = kmap_atomic(page, KM_USER0);
1996 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset);
1997 flush_dcache_page(page);
1998 kunmap_atomic(userpage, KM_USER0);
1999 }
2000
2001 set_page_extent_mapped(page);
2002
2003 delalloc_start = start;
2004 delalloc_end = 0;
2005 while(delalloc_end < page_end) {
2006 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start,
2007 &delalloc_end,
2008 128 * 1024 * 1024);
2009 if (nr_delalloc == 0) {
2010 delalloc_start = delalloc_end + 1;
2011 continue;
2012 }
2013 tree->ops->fill_delalloc(inode, delalloc_start,
2014 delalloc_end);
2015 clear_extent_bit(tree, delalloc_start,
2016 delalloc_end,
2017 EXTENT_LOCKED | EXTENT_DELALLOC,
2018 1, 0, GFP_NOFS);
2019 delalloc_start = delalloc_end + 1;
2020 }
2021 lock_extent(tree, start, page_end, GFP_NOFS);
2022
2023 end = page_end;
2024 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
2025 printk("found delalloc bits after lock_extent\n");
2026 }
2027
2028 if (last_byte <= start) {
2029 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
2030 goto done;
2031 }
2032
2033 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
2034 blocksize = inode->i_sb->s_blocksize;
2035
2036 while (cur <= end) {
2037 if (cur >= last_byte) {
2038 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
2039 break;
2040 }
2041 em = epd->get_extent(inode, page, page_offset, cur,
2042 end - cur + 1, 1);
2043 if (IS_ERR(em) || !em) {
2044 SetPageError(page);
2045 break;
2046 }
2047
2048 extent_offset = cur - em->start;
2049 BUG_ON(extent_map_end(em) <= cur);
2050 BUG_ON(end < cur);
2051 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2052 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2053 sector = (em->block_start + extent_offset) >> 9;
2054 bdev = em->bdev;
2055 block_start = em->block_start;
2056 free_extent_map(em);
2057 em = NULL;
2058
2059 if (block_start == EXTENT_MAP_HOLE ||
2060 block_start == EXTENT_MAP_INLINE) {
2061 clear_extent_dirty(tree, cur,
2062 cur + iosize - 1, GFP_NOFS);
2063 cur = cur + iosize;
2064 page_offset += iosize;
2065 continue;
2066 }
2067
2068 /* leave this out until we have a page_mkwrite call */
2069 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2070 EXTENT_DIRTY, 0)) {
2071 cur = cur + iosize;
2072 page_offset += iosize;
2073 continue;
2074 }
2075 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
2076 if (tree->ops && tree->ops->writepage_io_hook) {
2077 ret = tree->ops->writepage_io_hook(page, cur,
2078 cur + iosize - 1);
2079 } else {
2080 ret = 0;
2081 }
2082 if (ret)
2083 SetPageError(page);
2084 else {
2085 unsigned long max_nr = end_index + 1;
2086 set_range_writeback(tree, cur, cur + iosize - 1);
2087 if (!PageWriteback(page)) {
2088 printk("warning page %lu not writeback, "
2089 "cur %llu end %llu\n", page->index,
2090 (unsigned long long)cur,
2091 (unsigned long long)end);
2092 }
2093
2094 ret = submit_extent_page(WRITE, tree, page, sector,
2095 iosize, page_offset, bdev,
2096 &epd->bio, max_nr,
2097 end_bio_extent_writepage, 0);
2098 if (ret)
2099 SetPageError(page);
2100 }
2101 cur = cur + iosize;
2102 page_offset += iosize;
2103 nr++;
2104 }
2105 done:
2106 if (nr == 0) {
2107 /* make sure the mapping tag for page dirty gets cleared */
2108 set_page_writeback(page);
2109 end_page_writeback(page);
2110 }
2111 unlock_extent(tree, start, page_end, GFP_NOFS);
2112 unlock_page(page);
2113 return 0;
2114 }
2115
2116 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,20)
2117 /* Taken directly from 2.6.23 for 2.6.18 back port */
2118 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
2119 void *data);
2120
2121 /**
2122 * write_cache_pages - walk the list of dirty pages of the given address space
2123 * and write all of them.
2124 * @mapping: address space structure to write
2125 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2126 * @writepage: function called for each page
2127 * @data: data passed to writepage function
2128 *
2129 * If a page is already under I/O, write_cache_pages() skips it, even
2130 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2131 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2132 * and msync() need to guarantee that all the data which was dirty at the time
2133 * the call was made get new I/O started against them. If wbc->sync_mode is
2134 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2135 * existing IO to complete.
2136 */
2137 static int write_cache_pages(struct address_space *mapping,
2138 struct writeback_control *wbc, writepage_t writepage,
2139 void *data)
2140 {
2141 struct backing_dev_info *bdi = mapping->backing_dev_info;
2142 int ret = 0;
2143 int done = 0;
2144 struct pagevec pvec;
2145 int nr_pages;
2146 pgoff_t index;
2147 pgoff_t end; /* Inclusive */
2148 int scanned = 0;
2149 int range_whole = 0;
2150
2151 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2152 wbc->encountered_congestion = 1;
2153 return 0;
2154 }
2155
2156 pagevec_init(&pvec, 0);
2157 if (wbc->range_cyclic) {
2158 index = mapping->writeback_index; /* Start from prev offset */
2159 end = -1;
2160 } else {
2161 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2162 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2163 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2164 range_whole = 1;
2165 scanned = 1;
2166 }
2167 retry:
2168 while (!done && (index <= end) &&
2169 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2170 PAGECACHE_TAG_DIRTY,
2171 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2172 unsigned i;
2173
2174 scanned = 1;
2175 for (i = 0; i < nr_pages; i++) {
2176 struct page *page = pvec.pages[i];
2177
2178 /*
2179 * At this point we hold neither mapping->tree_lock nor
2180 * lock on the page itself: the page may be truncated or
2181 * invalidated (changing page->mapping to NULL), or even
2182 * swizzled back from swapper_space to tmpfs file
2183 * mapping
2184 */
2185 lock_page(page);
2186
2187 if (unlikely(page->mapping != mapping)) {
2188 unlock_page(page);
2189 continue;
2190 }
2191
2192 if (!wbc->range_cyclic && page->index > end) {
2193 done = 1;
2194 unlock_page(page);
2195 continue;
2196 }
2197
2198 if (wbc->sync_mode != WB_SYNC_NONE)
2199 wait_on_page_writeback(page);
2200
2201 if (PageWriteback(page) ||
2202 !clear_page_dirty_for_io(page)) {
2203 unlock_page(page);
2204 continue;
2205 }
2206
2207 ret = (*writepage)(page, wbc, data);
2208
2209 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2210 unlock_page(page);
2211 ret = 0;
2212 }
2213 if (ret || (--(wbc->nr_to_write) <= 0))
2214 done = 1;
2215 if (wbc->nonblocking && bdi_write_congested(bdi)) {
2216 wbc->encountered_congestion = 1;
2217 done = 1;
2218 }
2219 }
2220 pagevec_release(&pvec);
2221 cond_resched();
2222 }
2223 if (!scanned && !done) {
2224 /*
2225 * We hit the last page and there is more work to be done: wrap
2226 * back to the start of the file
2227 */
2228 scanned = 1;
2229 index = 0;
2230 goto retry;
2231 }
2232 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2233 mapping->writeback_index = index;
2234 return ret;
2235 }
2236 #endif
2237
2238 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2239 get_extent_t *get_extent,
2240 struct writeback_control *wbc)
2241 {
2242 int ret;
2243 struct address_space *mapping = page->mapping;
2244 struct extent_page_data epd = {
2245 .bio = NULL,
2246 .tree = tree,
2247 .get_extent = get_extent,
2248 };
2249 struct writeback_control wbc_writepages = {
2250 .bdi = wbc->bdi,
2251 .sync_mode = WB_SYNC_NONE,
2252 .older_than_this = NULL,
2253 .nr_to_write = 64,
2254 .range_start = page_offset(page) + PAGE_CACHE_SIZE,
2255 .range_end = (loff_t)-1,
2256 };
2257
2258
2259 ret = __extent_writepage(page, wbc, &epd);
2260
2261 write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd);
2262 if (epd.bio) {
2263 submit_one_bio(WRITE, epd.bio, 0);
2264 }
2265 return ret;
2266 }
2267 EXPORT_SYMBOL(extent_write_full_page);
2268
2269
2270 int extent_writepages(struct extent_io_tree *tree,
2271 struct address_space *mapping,
2272 get_extent_t *get_extent,
2273 struct writeback_control *wbc)
2274 {
2275 int ret = 0;
2276 struct extent_page_data epd = {
2277 .bio = NULL,
2278 .tree = tree,
2279 .get_extent = get_extent,
2280 };
2281
2282 ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd);
2283 if (epd.bio) {
2284 submit_one_bio(WRITE, epd.bio, 0);
2285 }
2286 return ret;
2287 }
2288 EXPORT_SYMBOL(extent_writepages);
2289
2290 int extent_readpages(struct extent_io_tree *tree,
2291 struct address_space *mapping,
2292 struct list_head *pages, unsigned nr_pages,
2293 get_extent_t get_extent)
2294 {
2295 struct bio *bio = NULL;
2296 unsigned page_idx;
2297 struct pagevec pvec;
2298
2299 pagevec_init(&pvec, 0);
2300 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2301 struct page *page = list_entry(pages->prev, struct page, lru);
2302
2303 prefetchw(&page->flags);
2304 list_del(&page->lru);
2305 /*
2306 * what we want to do here is call add_to_page_cache_lru,
2307 * but that isn't exported, so we reproduce it here
2308 */
2309 if (!add_to_page_cache(page, mapping,
2310 page->index, GFP_KERNEL)) {
2311
2312 /* open coding of lru_cache_add, also not exported */
2313 page_cache_get(page);
2314 if (!pagevec_add(&pvec, page))
2315 __pagevec_lru_add(&pvec);
2316 __extent_read_full_page(tree, page, get_extent,
2317 &bio, 0);
2318 }
2319 page_cache_release(page);
2320 }
2321 if (pagevec_count(&pvec))
2322 __pagevec_lru_add(&pvec);
2323 BUG_ON(!list_empty(pages));
2324 if (bio)
2325 submit_one_bio(READ, bio, 0);
2326 return 0;
2327 }
2328 EXPORT_SYMBOL(extent_readpages);
2329
2330 /*
2331 * basic invalidatepage code, this waits on any locked or writeback
2332 * ranges corresponding to the page, and then deletes any extent state
2333 * records from the tree
2334 */
2335 int extent_invalidatepage(struct extent_io_tree *tree,
2336 struct page *page, unsigned long offset)
2337 {
2338 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2339 u64 end = start + PAGE_CACHE_SIZE - 1;
2340 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2341
2342 start += (offset + blocksize -1) & ~(blocksize - 1);
2343 if (start > end)
2344 return 0;
2345
2346 lock_extent(tree, start, end, GFP_NOFS);
2347 wait_on_extent_writeback(tree, start, end);
2348 clear_extent_bit(tree, start, end,
2349 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
2350 1, 1, GFP_NOFS);
2351 return 0;
2352 }
2353 EXPORT_SYMBOL(extent_invalidatepage);
2354
2355 /*
2356 * simple commit_write call, set_range_dirty is used to mark both
2357 * the pages and the extent records as dirty
2358 */
2359 int extent_commit_write(struct extent_io_tree *tree,
2360 struct inode *inode, struct page *page,
2361 unsigned from, unsigned to)
2362 {
2363 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2364
2365 set_page_extent_mapped(page);
2366 set_page_dirty(page);
2367
2368 if (pos > inode->i_size) {
2369 i_size_write(inode, pos);
2370 mark_inode_dirty(inode);
2371 }
2372 return 0;
2373 }
2374 EXPORT_SYMBOL(extent_commit_write);
2375
2376 int extent_prepare_write(struct extent_io_tree *tree,
2377 struct inode *inode, struct page *page,
2378 unsigned from, unsigned to, get_extent_t *get_extent)
2379 {
2380 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2381 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2382 u64 block_start;
2383 u64 orig_block_start;
2384 u64 block_end;
2385 u64 cur_end;
2386 struct extent_map *em;
2387 unsigned blocksize = 1 << inode->i_blkbits;
2388 size_t page_offset = 0;
2389 size_t block_off_start;
2390 size_t block_off_end;
2391 int err = 0;
2392 int iocount = 0;
2393 int ret = 0;
2394 int isnew;
2395
2396 set_page_extent_mapped(page);
2397
2398 block_start = (page_start + from) & ~((u64)blocksize - 1);
2399 block_end = (page_start + to - 1) | (blocksize - 1);
2400 orig_block_start = block_start;
2401
2402 lock_extent(tree, page_start, page_end, GFP_NOFS);
2403 while(block_start <= block_end) {
2404 em = get_extent(inode, page, page_offset, block_start,
2405 block_end - block_start + 1, 1);
2406 if (IS_ERR(em) || !em) {
2407 goto err;
2408 }
2409 cur_end = min(block_end, extent_map_end(em) - 1);
2410 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
2411 block_off_end = block_off_start + blocksize;
2412 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
2413
2414 if (!PageUptodate(page) && isnew &&
2415 (block_off_end > to || block_off_start < from)) {
2416 void *kaddr;
2417
2418 kaddr = kmap_atomic(page, KM_USER0);
2419 if (block_off_end > to)
2420 memset(kaddr + to, 0, block_off_end - to);
2421 if (block_off_start < from)
2422 memset(kaddr + block_off_start, 0,
2423 from - block_off_start);
2424 flush_dcache_page(page);
2425 kunmap_atomic(kaddr, KM_USER0);
2426 }
2427 if ((em->block_start != EXTENT_MAP_HOLE &&
2428 em->block_start != EXTENT_MAP_INLINE) &&
2429 !isnew && !PageUptodate(page) &&
2430 (block_off_end > to || block_off_start < from) &&
2431 !test_range_bit(tree, block_start, cur_end,
2432 EXTENT_UPTODATE, 1)) {
2433 u64 sector;
2434 u64 extent_offset = block_start - em->start;
2435 size_t iosize;
2436 sector = (em->block_start + extent_offset) >> 9;
2437 iosize = (cur_end - block_start + blocksize) &
2438 ~((u64)blocksize - 1);
2439 /*
2440 * we've already got the extent locked, but we
2441 * need to split the state such that our end_bio
2442 * handler can clear the lock.
2443 */
2444 set_extent_bit(tree, block_start,
2445 block_start + iosize - 1,
2446 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
2447 ret = submit_extent_page(READ, tree, page,
2448 sector, iosize, page_offset, em->bdev,
2449 NULL, 1,
2450 end_bio_extent_preparewrite, 0);
2451 iocount++;
2452 block_start = block_start + iosize;
2453 } else {
2454 set_extent_uptodate(tree, block_start, cur_end,
2455 GFP_NOFS);
2456 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
2457 block_start = cur_end + 1;
2458 }
2459 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
2460 free_extent_map(em);
2461 }
2462 if (iocount) {
2463 wait_extent_bit(tree, orig_block_start,
2464 block_end, EXTENT_LOCKED);
2465 }
2466 check_page_uptodate(tree, page);
2467 err:
2468 /* FIXME, zero out newly allocated blocks on error */
2469 return err;
2470 }
2471 EXPORT_SYMBOL(extent_prepare_write);
2472
2473 /*
2474 * a helper for releasepage. As long as there are no locked extents
2475 * in the range corresponding to the page, both state records and extent
2476 * map records are removed
2477 */
2478 int try_release_extent_mapping(struct extent_map_tree *map,
2479 struct extent_io_tree *tree, struct page *page,
2480 gfp_t mask)
2481 {
2482 struct extent_map *em;
2483 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2484 u64 end = start + PAGE_CACHE_SIZE - 1;
2485 u64 orig_start = start;
2486 int ret = 1;
2487 if ((mask & __GFP_WAIT) &&
2488 page->mapping->host->i_size > 16 * 1024 * 1024) {
2489 u64 len;
2490 while (start <= end) {
2491 len = end - start + 1;
2492 spin_lock(&map->lock);
2493 em = lookup_extent_mapping(map, start, len);
2494 if (!em || IS_ERR(em)) {
2495 spin_unlock(&map->lock);
2496 break;
2497 }
2498 if (em->start != start) {
2499 spin_unlock(&map->lock);
2500 free_extent_map(em);
2501 break;
2502 }
2503 if (!test_range_bit(tree, em->start,
2504 extent_map_end(em) - 1,
2505 EXTENT_LOCKED, 0)) {
2506 remove_extent_mapping(map, em);
2507 /* once for the rb tree */
2508 free_extent_map(em);
2509 }
2510 start = extent_map_end(em);
2511 spin_unlock(&map->lock);
2512
2513 /* once for us */
2514 free_extent_map(em);
2515 }
2516 }
2517 if (test_range_bit(tree, orig_start, end, EXTENT_IOBITS, 0))
2518 ret = 0;
2519 else {
2520 if ((mask & GFP_NOFS) == GFP_NOFS)
2521 mask = GFP_NOFS;
2522 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
2523 1, 1, mask);
2524 }
2525 return ret;
2526 }
2527 EXPORT_SYMBOL(try_release_extent_mapping);
2528
2529 sector_t extent_bmap(struct address_space *mapping, sector_t iblock,
2530 get_extent_t *get_extent)
2531 {
2532 struct inode *inode = mapping->host;
2533 u64 start = iblock << inode->i_blkbits;
2534 sector_t sector = 0;
2535 struct extent_map *em;
2536
2537 em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0);
2538 if (!em || IS_ERR(em))
2539 return 0;
2540
2541 if (em->block_start == EXTENT_MAP_INLINE ||
2542 em->block_start == EXTENT_MAP_HOLE)
2543 goto out;
2544
2545 sector = (em->block_start + start - em->start) >> inode->i_blkbits;
2546 out:
2547 free_extent_map(em);
2548 return sector;
2549 }
2550
2551 static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb)
2552 {
2553 if (list_empty(&eb->lru)) {
2554 extent_buffer_get(eb);
2555 list_add(&eb->lru, &tree->buffer_lru);
2556 tree->lru_size++;
2557 if (tree->lru_size >= BUFFER_LRU_MAX) {
2558 struct extent_buffer *rm;
2559 rm = list_entry(tree->buffer_lru.prev,
2560 struct extent_buffer, lru);
2561 tree->lru_size--;
2562 list_del_init(&rm->lru);
2563 free_extent_buffer(rm);
2564 }
2565 } else
2566 list_move(&eb->lru, &tree->buffer_lru);
2567 return 0;
2568 }
2569 static struct extent_buffer *find_lru(struct extent_io_tree *tree,
2570 u64 start, unsigned long len)
2571 {
2572 struct list_head *lru = &tree->buffer_lru;
2573 struct list_head *cur = lru->next;
2574 struct extent_buffer *eb;
2575
2576 if (list_empty(lru))
2577 return NULL;
2578
2579 do {
2580 eb = list_entry(cur, struct extent_buffer, lru);
2581 if (eb->start == start && eb->len == len) {
2582 extent_buffer_get(eb);
2583 return eb;
2584 }
2585 cur = cur->next;
2586 } while (cur != lru);
2587 return NULL;
2588 }
2589
2590 static inline unsigned long num_extent_pages(u64 start, u64 len)
2591 {
2592 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
2593 (start >> PAGE_CACHE_SHIFT);
2594 }
2595
2596 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2597 unsigned long i)
2598 {
2599 struct page *p;
2600 struct address_space *mapping;
2601
2602 if (i == 0)
2603 return eb->first_page;
2604 i += eb->start >> PAGE_CACHE_SHIFT;
2605 mapping = eb->first_page->mapping;
2606 read_lock_irq(&mapping->tree_lock);
2607 p = radix_tree_lookup(&mapping->page_tree, i);
2608 read_unlock_irq(&mapping->tree_lock);
2609 return p;
2610 }
2611
2612 int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2613 {
2614 unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2615 struct page *page;
2616 unsigned long i;
2617
2618 if (num_pages == 1)
2619 return 0;
2620 for (i = 1; i < num_pages; i++) {
2621 page = extent_buffer_page(eb, i);
2622 page_cache_release(page);
2623 }
2624 return 0;
2625 }
2626
2627
2628 int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2629 unsigned long len)
2630 {
2631 struct list_head *lru = &tree->buffer_lru;
2632 struct list_head *cur = lru->next;
2633 struct extent_buffer *eb;
2634 int found = 0;
2635
2636 spin_lock(&tree->lru_lock);
2637 if (list_empty(lru))
2638 goto out;
2639
2640 do {
2641 eb = list_entry(cur, struct extent_buffer, lru);
2642 if (eb->start <= start && eb->start + eb->len > start) {
2643 eb->flags &= ~EXTENT_UPTODATE;
2644 }
2645 cur = cur->next;
2646 } while (cur != lru);
2647 out:
2648 spin_unlock(&tree->lru_lock);
2649 return found;
2650 }
2651
2652 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
2653 u64 start,
2654 unsigned long len,
2655 gfp_t mask)
2656 {
2657 struct extent_buffer *eb = NULL;
2658 unsigned long flags;
2659
2660 spin_lock(&tree->lru_lock);
2661 eb = find_lru(tree, start, len);
2662 spin_unlock(&tree->lru_lock);
2663 if (eb) {
2664 return eb;
2665 }
2666
2667 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
2668 INIT_LIST_HEAD(&eb->lru);
2669 eb->start = start;
2670 eb->len = len;
2671 spin_lock_irqsave(&leak_lock, flags);
2672 list_add(&eb->leak_list, &buffers);
2673 spin_unlock_irqrestore(&leak_lock, flags);
2674 atomic_set(&eb->refs, 1);
2675
2676 return eb;
2677 }
2678
2679 static void __free_extent_buffer(struct extent_buffer *eb)
2680 {
2681 unsigned long flags;
2682 spin_lock_irqsave(&leak_lock, flags);
2683 list_del(&eb->leak_list);
2684 spin_unlock_irqrestore(&leak_lock, flags);
2685 kmem_cache_free(extent_buffer_cache, eb);
2686 }
2687
2688 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2689 u64 start, unsigned long len,
2690 struct page *page0,
2691 gfp_t mask)
2692 {
2693 unsigned long num_pages = num_extent_pages(start, len);
2694 unsigned long i;
2695 unsigned long index = start >> PAGE_CACHE_SHIFT;
2696 struct extent_buffer *eb;
2697 struct page *p;
2698 struct address_space *mapping = tree->mapping;
2699 int uptodate = 1;
2700
2701 eb = __alloc_extent_buffer(tree, start, len, mask);
2702 if (!eb)
2703 return NULL;
2704
2705 if (eb->flags & EXTENT_BUFFER_FILLED)
2706 goto lru_add;
2707
2708 if (page0) {
2709 eb->first_page = page0;
2710 i = 1;
2711 index++;
2712 page_cache_get(page0);
2713 mark_page_accessed(page0);
2714 set_page_extent_mapped(page0);
2715 set_page_extent_head(page0, len);
2716 uptodate = PageUptodate(page0);
2717 } else {
2718 i = 0;
2719 }
2720 for (; i < num_pages; i++, index++) {
2721 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM);
2722 if (!p) {
2723 WARN_ON(1);
2724 goto fail;
2725 }
2726 set_page_extent_mapped(p);
2727 mark_page_accessed(p);
2728 if (i == 0) {
2729 eb->first_page = p;
2730 set_page_extent_head(p, len);
2731 } else {
2732 set_page_private(p, EXTENT_PAGE_PRIVATE);
2733 }
2734 if (!PageUptodate(p))
2735 uptodate = 0;
2736 unlock_page(p);
2737 }
2738 if (uptodate)
2739 eb->flags |= EXTENT_UPTODATE;
2740 eb->flags |= EXTENT_BUFFER_FILLED;
2741
2742 lru_add:
2743 spin_lock(&tree->lru_lock);
2744 add_lru(tree, eb);
2745 spin_unlock(&tree->lru_lock);
2746 return eb;
2747
2748 fail:
2749 spin_lock(&tree->lru_lock);
2750 list_del_init(&eb->lru);
2751 spin_unlock(&tree->lru_lock);
2752 if (!atomic_dec_and_test(&eb->refs))
2753 return NULL;
2754 for (index = 1; index < i; index++) {
2755 page_cache_release(extent_buffer_page(eb, index));
2756 }
2757 if (i > 0)
2758 page_cache_release(extent_buffer_page(eb, 0));
2759 __free_extent_buffer(eb);
2760 return NULL;
2761 }
2762 EXPORT_SYMBOL(alloc_extent_buffer);
2763
2764 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
2765 u64 start, unsigned long len,
2766 gfp_t mask)
2767 {
2768 unsigned long num_pages = num_extent_pages(start, len);
2769 unsigned long i;
2770 unsigned long index = start >> PAGE_CACHE_SHIFT;
2771 struct extent_buffer *eb;
2772 struct page *p;
2773 struct address_space *mapping = tree->mapping;
2774 int uptodate = 1;
2775
2776 eb = __alloc_extent_buffer(tree, start, len, mask);
2777 if (!eb)
2778 return NULL;
2779
2780 if (eb->flags & EXTENT_BUFFER_FILLED)
2781 goto lru_add;
2782
2783 for (i = 0; i < num_pages; i++, index++) {
2784 p = find_lock_page(mapping, index);
2785 if (!p) {
2786 goto fail;
2787 }
2788 set_page_extent_mapped(p);
2789 mark_page_accessed(p);
2790
2791 if (i == 0) {
2792 eb->first_page = p;
2793 set_page_extent_head(p, len);
2794 } else {
2795 set_page_private(p, EXTENT_PAGE_PRIVATE);
2796 }
2797
2798 if (!PageUptodate(p))
2799 uptodate = 0;
2800 unlock_page(p);
2801 }
2802 if (uptodate)
2803 eb->flags |= EXTENT_UPTODATE;
2804 eb->flags |= EXTENT_BUFFER_FILLED;
2805
2806 lru_add:
2807 spin_lock(&tree->lru_lock);
2808 add_lru(tree, eb);
2809 spin_unlock(&tree->lru_lock);
2810 return eb;
2811 fail:
2812 spin_lock(&tree->lru_lock);
2813 list_del_init(&eb->lru);
2814 spin_unlock(&tree->lru_lock);
2815 if (!atomic_dec_and_test(&eb->refs))
2816 return NULL;
2817 for (index = 1; index < i; index++) {
2818 page_cache_release(extent_buffer_page(eb, index));
2819 }
2820 if (i > 0)
2821 page_cache_release(extent_buffer_page(eb, 0));
2822 __free_extent_buffer(eb);
2823 return NULL;
2824 }
2825 EXPORT_SYMBOL(find_extent_buffer);
2826
2827 void free_extent_buffer(struct extent_buffer *eb)
2828 {
2829 unsigned long i;
2830 unsigned long num_pages;
2831
2832 if (!eb)
2833 return;
2834
2835 if (!atomic_dec_and_test(&eb->refs))
2836 return;
2837
2838 WARN_ON(!list_empty(&eb->lru));
2839 num_pages = num_extent_pages(eb->start, eb->len);
2840
2841 for (i = 1; i < num_pages; i++) {
2842 page_cache_release(extent_buffer_page(eb, i));
2843 }
2844 page_cache_release(extent_buffer_page(eb, 0));
2845 __free_extent_buffer(eb);
2846 }
2847 EXPORT_SYMBOL(free_extent_buffer);
2848
2849 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
2850 struct extent_buffer *eb)
2851 {
2852 int set;
2853 unsigned long i;
2854 unsigned long num_pages;
2855 struct page *page;
2856
2857 u64 start = eb->start;
2858 u64 end = start + eb->len - 1;
2859
2860 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
2861 num_pages = num_extent_pages(eb->start, eb->len);
2862
2863 for (i = 0; i < num_pages; i++) {
2864 page = extent_buffer_page(eb, i);
2865 lock_page(page);
2866 if (i == 0)
2867 set_page_extent_head(page, eb->len);
2868 else
2869 set_page_private(page, EXTENT_PAGE_PRIVATE);
2870
2871 /*
2872 * if we're on the last page or the first page and the
2873 * block isn't aligned on a page boundary, do extra checks
2874 * to make sure we don't clean page that is partially dirty
2875 */
2876 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2877 ((i == num_pages - 1) &&
2878 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2879 start = (u64)page->index << PAGE_CACHE_SHIFT;
2880 end = start + PAGE_CACHE_SIZE - 1;
2881 if (test_range_bit(tree, start, end,
2882 EXTENT_DIRTY, 0)) {
2883 unlock_page(page);
2884 continue;
2885 }
2886 }
2887 clear_page_dirty_for_io(page);
2888 read_lock_irq(&page->mapping->tree_lock);
2889 if (!PageDirty(page)) {
2890 radix_tree_tag_clear(&page->mapping->page_tree,
2891 page_index(page),
2892 PAGECACHE_TAG_DIRTY);
2893 }
2894 read_unlock_irq(&page->mapping->tree_lock);
2895 unlock_page(page);
2896 }
2897 return 0;
2898 }
2899 EXPORT_SYMBOL(clear_extent_buffer_dirty);
2900
2901 int wait_on_extent_buffer_writeback(struct extent_io_tree *tree,
2902 struct extent_buffer *eb)
2903 {
2904 return wait_on_extent_writeback(tree, eb->start,
2905 eb->start + eb->len - 1);
2906 }
2907 EXPORT_SYMBOL(wait_on_extent_buffer_writeback);
2908
2909 int set_extent_buffer_dirty(struct extent_io_tree *tree,
2910 struct extent_buffer *eb)
2911 {
2912 unsigned long i;
2913 unsigned long num_pages;
2914
2915 num_pages = num_extent_pages(eb->start, eb->len);
2916 for (i = 0; i < num_pages; i++) {
2917 struct page *page = extent_buffer_page(eb, i);
2918 /* writepage may need to do something special for the
2919 * first page, we have to make sure page->private is
2920 * properly set. releasepage may drop page->private
2921 * on us if the page isn't already dirty.
2922 */
2923 if (i == 0) {
2924 lock_page(page);
2925 set_page_extent_head(page, eb->len);
2926 } else if (PagePrivate(page) &&
2927 page->private != EXTENT_PAGE_PRIVATE) {
2928 lock_page(page);
2929 set_page_extent_mapped(page);
2930 unlock_page(page);
2931 }
2932 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
2933 if (i == 0)
2934 unlock_page(page);
2935 }
2936 return set_extent_dirty(tree, eb->start,
2937 eb->start + eb->len - 1, GFP_NOFS);
2938 }
2939 EXPORT_SYMBOL(set_extent_buffer_dirty);
2940
2941 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2942 struct extent_buffer *eb)
2943 {
2944 unsigned long i;
2945 struct page *page;
2946 unsigned long num_pages;
2947
2948 num_pages = num_extent_pages(eb->start, eb->len);
2949
2950 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
2951 GFP_NOFS);
2952 for (i = 0; i < num_pages; i++) {
2953 page = extent_buffer_page(eb, i);
2954 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
2955 ((i == num_pages - 1) &&
2956 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
2957 check_page_uptodate(tree, page);
2958 continue;
2959 }
2960 SetPageUptodate(page);
2961 }
2962 return 0;
2963 }
2964 EXPORT_SYMBOL(set_extent_buffer_uptodate);
2965
2966 int extent_range_uptodate(struct extent_io_tree *tree,
2967 u64 start, u64 end)
2968 {
2969 struct page *page;
2970 int ret;
2971 int pg_uptodate = 1;
2972 int uptodate;
2973 unsigned long index;
2974
2975 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2976 if (ret)
2977 return 1;
2978 while(start <= end) {
2979 index = start >> PAGE_CACHE_SHIFT;
2980 page = find_get_page(tree->mapping, index);
2981 uptodate = PageUptodate(page);
2982 page_cache_release(page);
2983 if (!uptodate) {
2984 pg_uptodate = 0;
2985 break;
2986 }
2987 start += PAGE_CACHE_SIZE;
2988 }
2989 return pg_uptodate;
2990 }
2991
2992 int extent_buffer_uptodate(struct extent_io_tree *tree,
2993 struct extent_buffer *eb)
2994 {
2995 int ret = 0;
2996 int ret2;
2997 unsigned long num_pages;
2998 unsigned long i;
2999 struct page *page;
3000 int pg_uptodate = 1;
3001
3002 if (eb->flags & EXTENT_UPTODATE)
3003 ret = 1;
3004
3005 ret2 = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3006 EXTENT_UPTODATE, 1);
3007
3008 num_pages = num_extent_pages(eb->start, eb->len);
3009 for (i = 0; i < num_pages; i++) {
3010 page = extent_buffer_page(eb, i);
3011 if (!PageUptodate(page)) {
3012 pg_uptodate = 0;
3013 break;
3014 }
3015 }
3016 if ((ret || ret2) && !pg_uptodate) {
3017 printk("uptodate error2 eb %Lu ret %d ret2 %d pg_uptodate %d\n", eb->start, ret, ret2, pg_uptodate);
3018 WARN_ON(1);
3019 }
3020 return (ret || ret2);
3021 }
3022 EXPORT_SYMBOL(extent_buffer_uptodate);
3023
3024 int read_extent_buffer_pages(struct extent_io_tree *tree,
3025 struct extent_buffer *eb,
3026 u64 start, int wait,
3027 get_extent_t *get_extent, int mirror_num)
3028 {
3029 unsigned long i;
3030 unsigned long start_i;
3031 struct page *page;
3032 int err;
3033 int ret = 0;
3034 int locked_pages = 0;
3035 int all_uptodate = 1;
3036 int inc_all_pages = 0;
3037 unsigned long num_pages;
3038 struct bio *bio = NULL;
3039
3040 if (eb->flags & EXTENT_UPTODATE)
3041 return 0;
3042
3043 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3044 EXTENT_UPTODATE, 1)) {
3045 return 0;
3046 }
3047
3048 if (start) {
3049 WARN_ON(start < eb->start);
3050 start_i = (start >> PAGE_CACHE_SHIFT) -
3051 (eb->start >> PAGE_CACHE_SHIFT);
3052 } else {
3053 start_i = 0;
3054 }
3055
3056 num_pages = num_extent_pages(eb->start, eb->len);
3057 for (i = start_i; i < num_pages; i++) {
3058 page = extent_buffer_page(eb, i);
3059 if (!wait) {
3060 if (TestSetPageLocked(page))
3061 goto unlock_exit;
3062 } else {
3063 lock_page(page);
3064 }
3065 locked_pages++;
3066 if (!PageUptodate(page)) {
3067 all_uptodate = 0;
3068 }
3069 }
3070 if (all_uptodate) {
3071 if (start_i == 0)
3072 eb->flags |= EXTENT_UPTODATE;
3073 goto unlock_exit;
3074 }
3075
3076 for (i = start_i; i < num_pages; i++) {
3077 page = extent_buffer_page(eb, i);
3078 if (inc_all_pages)
3079 page_cache_get(page);
3080 if (!PageUptodate(page)) {
3081 if (start_i == 0)
3082 inc_all_pages = 1;
3083 ClearPageError(page);
3084 err = __extent_read_full_page(tree, page,
3085 get_extent, &bio,
3086 mirror_num);
3087 if (err) {
3088 ret = err;
3089 }
3090 } else {
3091 unlock_page(page);
3092 }
3093 }
3094
3095 if (bio)
3096 submit_one_bio(READ, bio, mirror_num);
3097
3098 if (ret || !wait) {
3099 return ret;
3100 }
3101 for (i = start_i; i < num_pages; i++) {
3102 page = extent_buffer_page(eb, i);
3103 wait_on_page_locked(page);
3104 if (!PageUptodate(page)) {
3105 ret = -EIO;
3106 }
3107 }
3108 if (!ret)
3109 eb->flags |= EXTENT_UPTODATE;
3110 return ret;
3111
3112 unlock_exit:
3113 i = start_i;
3114 while(locked_pages > 0) {
3115 page = extent_buffer_page(eb, i);
3116 i++;
3117 unlock_page(page);
3118 locked_pages--;
3119 }
3120 return ret;
3121 }
3122 EXPORT_SYMBOL(read_extent_buffer_pages);
3123
3124 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3125 unsigned long start,
3126 unsigned long len)
3127 {
3128 size_t cur;
3129 size_t offset;
3130 struct page *page;
3131 char *kaddr;
3132 char *dst = (char *)dstv;
3133 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3134 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3135
3136 WARN_ON(start > eb->len);
3137 WARN_ON(start + len > eb->start + eb->len);
3138
3139 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3140
3141 while(len > 0) {
3142 page = extent_buffer_page(eb, i);
3143
3144 cur = min(len, (PAGE_CACHE_SIZE - offset));
3145 kaddr = kmap_atomic(page, KM_USER1);
3146 memcpy(dst, kaddr + offset, cur);
3147 kunmap_atomic(kaddr, KM_USER1);
3148
3149 dst += cur;
3150 len -= cur;
3151 offset = 0;
3152 i++;
3153 }
3154 }
3155 EXPORT_SYMBOL(read_extent_buffer);
3156
3157 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3158 unsigned long min_len, char **token, char **map,
3159 unsigned long *map_start,
3160 unsigned long *map_len, int km)
3161 {
3162 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3163 char *kaddr;
3164 struct page *p;
3165 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3166 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3167 unsigned long end_i = (start_offset + start + min_len - 1) >>
3168 PAGE_CACHE_SHIFT;
3169
3170 if (i != end_i)
3171 return -EINVAL;
3172
3173 if (i == 0) {
3174 offset = start_offset;
3175 *map_start = 0;
3176 } else {
3177 offset = 0;
3178 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3179 }
3180 if (start + min_len > eb->len) {
3181 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len);
3182 WARN_ON(1);
3183 }
3184
3185 p = extent_buffer_page(eb, i);
3186 kaddr = kmap_atomic(p, km);
3187 *token = kaddr;
3188 *map = kaddr + offset;
3189 *map_len = PAGE_CACHE_SIZE - offset;
3190 return 0;
3191 }
3192 EXPORT_SYMBOL(map_private_extent_buffer);
3193
3194 int map_extent_buffer(struct extent_buffer *eb, unsigned long start,
3195 unsigned long min_len,
3196 char **token, char **map,
3197 unsigned long *map_start,
3198 unsigned long *map_len, int km)
3199 {
3200 int err;
3201 int save = 0;
3202 if (eb->map_token) {
3203 unmap_extent_buffer(eb, eb->map_token, km);
3204 eb->map_token = NULL;
3205 save = 1;
3206 }
3207 err = map_private_extent_buffer(eb, start, min_len, token, map,
3208 map_start, map_len, km);
3209 if (!err && save) {
3210 eb->map_token = *token;
3211 eb->kaddr = *map;
3212 eb->map_start = *map_start;
3213 eb->map_len = *map_len;
3214 }
3215 return err;
3216 }
3217 EXPORT_SYMBOL(map_extent_buffer);
3218
3219 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km)
3220 {
3221 kunmap_atomic(token, km);
3222 }
3223 EXPORT_SYMBOL(unmap_extent_buffer);
3224
3225 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3226 unsigned long start,
3227 unsigned long len)
3228 {
3229 size_t cur;
3230 size_t offset;
3231 struct page *page;
3232 char *kaddr;
3233 char *ptr = (char *)ptrv;
3234 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3235 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3236 int ret = 0;
3237
3238 WARN_ON(start > eb->len);
3239 WARN_ON(start + len > eb->start + eb->len);
3240
3241 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3242
3243 while(len > 0) {
3244 page = extent_buffer_page(eb, i);
3245
3246 cur = min(len, (PAGE_CACHE_SIZE - offset));
3247
3248 kaddr = kmap_atomic(page, KM_USER0);
3249 ret = memcmp(ptr, kaddr + offset, cur);
3250 kunmap_atomic(kaddr, KM_USER0);
3251 if (ret)
3252 break;
3253
3254 ptr += cur;
3255 len -= cur;
3256 offset = 0;
3257 i++;
3258 }
3259 return ret;
3260 }
3261 EXPORT_SYMBOL(memcmp_extent_buffer);
3262
3263 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3264 unsigned long start, unsigned long len)
3265 {
3266 size_t cur;
3267 size_t offset;
3268 struct page *page;
3269 char *kaddr;
3270 char *src = (char *)srcv;
3271 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3272 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3273
3274 WARN_ON(start > eb->len);
3275 WARN_ON(start + len > eb->start + eb->len);
3276
3277 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3278
3279 while(len > 0) {
3280 page = extent_buffer_page(eb, i);
3281 WARN_ON(!PageUptodate(page));
3282
3283 cur = min(len, PAGE_CACHE_SIZE - offset);
3284 kaddr = kmap_atomic(page, KM_USER1);
3285 memcpy(kaddr + offset, src, cur);
3286 kunmap_atomic(kaddr, KM_USER1);
3287
3288 src += cur;
3289 len -= cur;
3290 offset = 0;
3291 i++;
3292 }
3293 }
3294 EXPORT_SYMBOL(write_extent_buffer);
3295
3296 void memset_extent_buffer(struct extent_buffer *eb, char c,
3297 unsigned long start, unsigned long len)
3298 {
3299 size_t cur;
3300 size_t offset;
3301 struct page *page;
3302 char *kaddr;
3303 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3304 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3305
3306 WARN_ON(start > eb->len);
3307 WARN_ON(start + len > eb->start + eb->len);
3308
3309 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3310
3311 while(len > 0) {
3312 page = extent_buffer_page(eb, i);
3313 WARN_ON(!PageUptodate(page));
3314
3315 cur = min(len, PAGE_CACHE_SIZE - offset);
3316 kaddr = kmap_atomic(page, KM_USER0);
3317 memset(kaddr + offset, c, cur);
3318 kunmap_atomic(kaddr, KM_USER0);
3319
3320 len -= cur;
3321 offset = 0;
3322 i++;
3323 }
3324 }
3325 EXPORT_SYMBOL(memset_extent_buffer);
3326
3327 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3328 unsigned long dst_offset, unsigned long src_offset,
3329 unsigned long len)
3330 {
3331 u64 dst_len = dst->len;
3332 size_t cur;
3333 size_t offset;
3334 struct page *page;
3335 char *kaddr;
3336 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3337 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3338
3339 WARN_ON(src->len != dst_len);
3340
3341 offset = (start_offset + dst_offset) &
3342 ((unsigned long)PAGE_CACHE_SIZE - 1);
3343
3344 while(len > 0) {
3345 page = extent_buffer_page(dst, i);
3346 WARN_ON(!PageUptodate(page));
3347
3348 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3349
3350 kaddr = kmap_atomic(page, KM_USER0);
3351 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3352 kunmap_atomic(kaddr, KM_USER0);
3353
3354 src_offset += cur;
3355 len -= cur;
3356 offset = 0;
3357 i++;
3358 }
3359 }
3360 EXPORT_SYMBOL(copy_extent_buffer);
3361
3362 static void move_pages(struct page *dst_page, struct page *src_page,
3363 unsigned long dst_off, unsigned long src_off,
3364 unsigned long len)
3365 {
3366 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3367 if (dst_page == src_page) {
3368 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3369 } else {
3370 char *src_kaddr = kmap_atomic(src_page, KM_USER1);
3371 char *p = dst_kaddr + dst_off + len;
3372 char *s = src_kaddr + src_off + len;
3373
3374 while (len--)
3375 *--p = *--s;
3376
3377 kunmap_atomic(src_kaddr, KM_USER1);
3378 }
3379 kunmap_atomic(dst_kaddr, KM_USER0);
3380 }
3381
3382 static void copy_pages(struct page *dst_page, struct page *src_page,
3383 unsigned long dst_off, unsigned long src_off,
3384 unsigned long len)
3385 {
3386 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0);
3387 char *src_kaddr;
3388
3389 if (dst_page != src_page)
3390 src_kaddr = kmap_atomic(src_page, KM_USER1);
3391 else
3392 src_kaddr = dst_kaddr;
3393
3394 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3395 kunmap_atomic(dst_kaddr, KM_USER0);
3396 if (dst_page != src_page)
3397 kunmap_atomic(src_kaddr, KM_USER1);
3398 }
3399
3400 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3401 unsigned long src_offset, unsigned long len)
3402 {
3403 size_t cur;
3404 size_t dst_off_in_page;
3405 size_t src_off_in_page;
3406 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3407 unsigned long dst_i;
3408 unsigned long src_i;
3409
3410 if (src_offset + len > dst->len) {
3411 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3412 src_offset, len, dst->len);
3413 BUG_ON(1);
3414 }
3415 if (dst_offset + len > dst->len) {
3416 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3417 dst_offset, len, dst->len);
3418 BUG_ON(1);
3419 }
3420
3421 while(len > 0) {
3422 dst_off_in_page = (start_offset + dst_offset) &
3423 ((unsigned long)PAGE_CACHE_SIZE - 1);
3424 src_off_in_page = (start_offset + src_offset) &
3425 ((unsigned long)PAGE_CACHE_SIZE - 1);
3426
3427 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3428 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3429
3430 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3431 src_off_in_page));
3432 cur = min_t(unsigned long, cur,
3433 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3434
3435 copy_pages(extent_buffer_page(dst, dst_i),
3436 extent_buffer_page(dst, src_i),
3437 dst_off_in_page, src_off_in_page, cur);
3438
3439 src_offset += cur;
3440 dst_offset += cur;
3441 len -= cur;
3442 }
3443 }
3444 EXPORT_SYMBOL(memcpy_extent_buffer);
3445
3446 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3447 unsigned long src_offset, unsigned long len)
3448 {
3449 size_t cur;
3450 size_t dst_off_in_page;
3451 size_t src_off_in_page;
3452 unsigned long dst_end = dst_offset + len - 1;
3453 unsigned long src_end = src_offset + len - 1;
3454 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3455 unsigned long dst_i;
3456 unsigned long src_i;
3457
3458 if (src_offset + len > dst->len) {
3459 printk("memmove bogus src_offset %lu move len %lu len %lu\n",
3460 src_offset, len, dst->len);
3461 BUG_ON(1);
3462 }
3463 if (dst_offset + len > dst->len) {
3464 printk("memmove bogus dst_offset %lu move len %lu len %lu\n",
3465 dst_offset, len, dst->len);
3466 BUG_ON(1);
3467 }
3468 if (dst_offset < src_offset) {
3469 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3470 return;
3471 }
3472 while(len > 0) {
3473 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3474 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3475
3476 dst_off_in_page = (start_offset + dst_end) &
3477 ((unsigned long)PAGE_CACHE_SIZE - 1);
3478 src_off_in_page = (start_offset + src_end) &
3479 ((unsigned long)PAGE_CACHE_SIZE - 1);
3480
3481 cur = min_t(unsigned long, len, src_off_in_page + 1);
3482 cur = min(cur, dst_off_in_page + 1);
3483 move_pages(extent_buffer_page(dst, dst_i),
3484 extent_buffer_page(dst, src_i),
3485 dst_off_in_page - cur + 1,
3486 src_off_in_page - cur + 1, cur);
3487
3488 dst_end -= cur;
3489 src_end -= cur;
3490 len -= cur;
3491 }
3492 }
3493 EXPORT_SYMBOL(memmove_extent_buffer);
ubuntu-bionic-kernel mirror