?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: make sure to flush queued bios if write_cache_pages waits
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / btrfs / extent_io.c
blobb40ba75f4483d205afb08265ddb4e6789f065404
1 #include <linux/bitops.h>
2 #include <linux/slab.h>
3 #include <linux/bio.h>
4 #include <linux/mm.h>
5 #include <linux/pagemap.h>
6 #include <linux/page-flags.h>
7 #include <linux/module.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
17 #include "compat.h"
18 #include "ctree.h"
19 #include "btrfs_inode.h"
20
21 static struct kmem_cache *extent_state_cache;
22 static struct kmem_cache *extent_buffer_cache;
23
24 static LIST_HEAD(buffers);
25 static LIST_HEAD(states);
26
27 #define LEAK_DEBUG 0
28 #if LEAK_DEBUG
29 static DEFINE_SPINLOCK(leak_lock);
30 #endif
31
32 #define BUFFER_LRU_MAX 64
33
34 struct tree_entry {
35 u64 start;
36 u64 end;
37 struct rb_node rb_node;
38 };
39
40 struct extent_page_data {
41 struct bio *bio;
42 struct extent_io_tree *tree;
43 get_extent_t *get_extent;
44
45 /* tells writepage not to lock the state bits for this range
46 * it still does the unlocking
47 */
48 unsigned int extent_locked:1;
49
50 /* tells the submit_bio code to use a WRITE_SYNC */
51 unsigned int sync_io:1;
52 };
53
54 int __init extent_io_init(void)
55 {
56 extent_state_cache = kmem_cache_create("extent_state",
57 sizeof(struct extent_state), 0,
58 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
59 if (!extent_state_cache)
60 return -ENOMEM;
61
62 extent_buffer_cache = kmem_cache_create("extent_buffers",
63 sizeof(struct extent_buffer), 0,
64 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
65 if (!extent_buffer_cache)
66 goto free_state_cache;
67 return 0;
68
69 free_state_cache:
70 kmem_cache_destroy(extent_state_cache);
71 return -ENOMEM;
72 }
73
74 void extent_io_exit(void)
75 {
76 struct extent_state *state;
77 struct extent_buffer *eb;
78
79 while (!list_empty(&states)) {
80 state = list_entry(states.next, struct extent_state, leak_list);
81 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
82 "state %lu in tree %p refs %d\n",
83 (unsigned long long)state->start,
84 (unsigned long long)state->end,
85 state->state, state->tree, atomic_read(&state->refs));
86 list_del(&state->leak_list);
87 kmem_cache_free(extent_state_cache, state);
88
89 }
90
91 while (!list_empty(&buffers)) {
92 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
93 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
94 "refs %d\n", (unsigned long long)eb->start,
95 eb->len, atomic_read(&eb->refs));
96 list_del(&eb->leak_list);
97 kmem_cache_free(extent_buffer_cache, eb);
98 }
99 if (extent_state_cache)
100 kmem_cache_destroy(extent_state_cache);
101 if (extent_buffer_cache)
102 kmem_cache_destroy(extent_buffer_cache);
103 }
104
105 void extent_io_tree_init(struct extent_io_tree *tree,
106 struct address_space *mapping)
107 {
108 tree->state = RB_ROOT;
109 INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
110 tree->ops = NULL;
111 tree->dirty_bytes = 0;
112 spin_lock_init(&tree->lock);
113 spin_lock_init(&tree->buffer_lock);
114 tree->mapping = mapping;
115 }
116
117 static struct extent_state *alloc_extent_state(gfp_t mask)
118 {
119 struct extent_state *state;
120 #if LEAK_DEBUG
121 unsigned long flags;
122 #endif
123
124 state = kmem_cache_alloc(extent_state_cache, mask);
125 if (!state)
126 return state;
127 state->state = 0;
128 state->private = 0;
129 state->tree = NULL;
130 #if LEAK_DEBUG
131 spin_lock_irqsave(&leak_lock, flags);
132 list_add(&state->leak_list, &states);
133 spin_unlock_irqrestore(&leak_lock, flags);
134 #endif
135 atomic_set(&state->refs, 1);
136 init_waitqueue_head(&state->wq);
137 return state;
138 }
139
140 void free_extent_state(struct extent_state *state)
141 {
142 if (!state)
143 return;
144 if (atomic_dec_and_test(&state->refs)) {
145 #if LEAK_DEBUG
146 unsigned long flags;
147 #endif
148 WARN_ON(state->tree);
149 #if LEAK_DEBUG
150 spin_lock_irqsave(&leak_lock, flags);
151 list_del(&state->leak_list);
152 spin_unlock_irqrestore(&leak_lock, flags);
153 #endif
154 kmem_cache_free(extent_state_cache, state);
155 }
156 }
157
158 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
159 struct rb_node *node)
160 {
161 struct rb_node **p = &root->rb_node;
162 struct rb_node *parent = NULL;
163 struct tree_entry *entry;
164
165 while (*p) {
166 parent = *p;
167 entry = rb_entry(parent, struct tree_entry, rb_node);
168
169 if (offset < entry->start)
170 p = &(*p)->rb_left;
171 else if (offset > entry->end)
172 p = &(*p)->rb_right;
173 else
174 return parent;
175 }
176
177 entry = rb_entry(node, struct tree_entry, rb_node);
178 rb_link_node(node, parent, p);
179 rb_insert_color(node, root);
180 return NULL;
181 }
182
183 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
184 struct rb_node **prev_ret,
185 struct rb_node **next_ret)
186 {
187 struct rb_root *root = &tree->state;
188 struct rb_node *n = root->rb_node;
189 struct rb_node *prev = NULL;
190 struct rb_node *orig_prev = NULL;
191 struct tree_entry *entry;
192 struct tree_entry *prev_entry = NULL;
193
194 while (n) {
195 entry = rb_entry(n, struct tree_entry, rb_node);
196 prev = n;
197 prev_entry = entry;
198
199 if (offset < entry->start)
200 n = n->rb_left;
201 else if (offset > entry->end)
202 n = n->rb_right;
203 else
204 return n;
205 }
206
207 if (prev_ret) {
208 orig_prev = prev;
209 while (prev && offset > prev_entry->end) {
210 prev = rb_next(prev);
211 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
212 }
213 *prev_ret = prev;
214 prev = orig_prev;
215 }
216
217 if (next_ret) {
218 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
219 while (prev && offset < prev_entry->start) {
220 prev = rb_prev(prev);
221 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
222 }
223 *next_ret = prev;
224 }
225 return NULL;
226 }
227
228 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
229 u64 offset)
230 {
231 struct rb_node *prev = NULL;
232 struct rb_node *ret;
233
234 ret = __etree_search(tree, offset, &prev, NULL);
235 if (!ret)
236 return prev;
237 return ret;
238 }
239
240 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
241 struct extent_state *other)
242 {
243 if (tree->ops && tree->ops->merge_extent_hook)
244 tree->ops->merge_extent_hook(tree->mapping->host, new,
245 other);
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 void 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 | EXTENT_BOUNDARY))
264 return;
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 merge_cb(tree, state, other);
272 state->start = other->start;
273 other->tree = NULL;
274 rb_erase(&other->rb_node, &tree->state);
275 free_extent_state(other);
276 }
277 }
278 other_node = rb_next(&state->rb_node);
279 if (other_node) {
280 other = rb_entry(other_node, struct extent_state, rb_node);
281 if (other->start == state->end + 1 &&
282 other->state == state->state) {
283 merge_cb(tree, state, other);
284 state->end = other->end;
285 other->tree = NULL;
286 rb_erase(&other->rb_node, &tree->state);
287 free_extent_state(other);
288 }
289 }
290 }
291
292 static void set_state_cb(struct extent_io_tree *tree,
293 struct extent_state *state, int *bits)
294 {
295 if (tree->ops && tree->ops->set_bit_hook)
296 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
297 }
298
299 static void clear_state_cb(struct extent_io_tree *tree,
300 struct extent_state *state, int *bits)
301 {
302 if (tree->ops && tree->ops->clear_bit_hook)
303 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
304 }
305
306 static void set_state_bits(struct extent_io_tree *tree,
307 struct extent_state *state, int *bits);
308
309 /*
310 * insert an extent_state struct into the tree. 'bits' are set on the
311 * struct before it is inserted.
312 *
313 * This may return -EEXIST if the extent is already there, in which case the
314 * state struct is freed.
315 *
316 * The tree lock is not taken internally. This is a utility function and
317 * probably isn't what you want to call (see set/clear_extent_bit).
318 */
319 static int insert_state(struct extent_io_tree *tree,
320 struct extent_state *state, u64 start, u64 end,
321 int *bits)
322 {
323 struct rb_node *node;
324
325 if (end < start) {
326 printk(KERN_ERR "btrfs end < start %llu %llu\n",
327 (unsigned long long)end,
328 (unsigned long long)start);
329 WARN_ON(1);
330 }
331 state->start = start;
332 state->end = end;
333
334 set_state_bits(tree, state, bits);
335
336 node = tree_insert(&tree->state, end, &state->rb_node);
337 if (node) {
338 struct extent_state *found;
339 found = rb_entry(node, struct extent_state, rb_node);
340 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
341 "%llu %llu\n", (unsigned long long)found->start,
342 (unsigned long long)found->end,
343 (unsigned long long)start, (unsigned long long)end);
344 return -EEXIST;
345 }
346 state->tree = tree;
347 merge_state(tree, state);
348 return 0;
349 }
350
351 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
352 u64 split)
353 {
354 if (tree->ops && tree->ops->split_extent_hook)
355 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
356 }
357
358 /*
359 * split a given extent state struct in two, inserting the preallocated
360 * struct 'prealloc' as the newly created second half. 'split' indicates an
361 * offset inside 'orig' where it should be split.
362 *
363 * Before calling,
364 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
365 * are two extent state structs in the tree:
366 * prealloc: [orig->start, split - 1]
367 * orig: [ split, orig->end ]
368 *
369 * The tree locks are not taken by this function. They need to be held
370 * by the caller.
371 */
372 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
373 struct extent_state *prealloc, u64 split)
374 {
375 struct rb_node *node;
376
377 split_cb(tree, orig, split);
378
379 prealloc->start = orig->start;
380 prealloc->end = split - 1;
381 prealloc->state = orig->state;
382 orig->start = split;
383
384 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
385 if (node) {
386 free_extent_state(prealloc);
387 return -EEXIST;
388 }
389 prealloc->tree = tree;
390 return 0;
391 }
392
393 /*
394 * utility function to clear some bits in an extent state struct.
395 * it will optionally wake up any one waiting on this state (wake == 1), or
396 * forcibly remove the state from the tree (delete == 1).
397 *
398 * If no bits are set on the state struct after clearing things, the
399 * struct is freed and removed from the tree
400 */
401 static int clear_state_bit(struct extent_io_tree *tree,
402 struct extent_state *state,
403 int *bits, int wake)
404 {
405 int bits_to_clear = *bits & ~EXTENT_CTLBITS;
406 int ret = state->state & bits_to_clear;
407
408 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
409 u64 range = state->end - state->start + 1;
410 WARN_ON(range > tree->dirty_bytes);
411 tree->dirty_bytes -= range;
412 }
413 clear_state_cb(tree, state, bits);
414 state->state &= ~bits_to_clear;
415 if (wake)
416 wake_up(&state->wq);
417 if (state->state == 0) {
418 if (state->tree) {
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 static struct extent_state *
432 alloc_extent_state_atomic(struct extent_state *prealloc)
433 {
434 if (!prealloc)
435 prealloc = alloc_extent_state(GFP_ATOMIC);
436
437 return prealloc;
438 }
439
440 /*
441 * clear some bits on a range in the tree. This may require splitting
442 * or inserting elements in the tree, so the gfp mask is used to
443 * indicate which allocations or sleeping are allowed.
444 *
445 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
446 * the given range from the tree regardless of state (ie for truncate).
447 *
448 * the range [start, end] is inclusive.
449 *
450 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
451 * bits were already set, or zero if none of the bits were already set.
452 */
453 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
454 int bits, int wake, int delete,
455 struct extent_state **cached_state,
456 gfp_t mask)
457 {
458 struct extent_state *state;
459 struct extent_state *cached;
460 struct extent_state *prealloc = NULL;
461 struct rb_node *next_node;
462 struct rb_node *node;
463 u64 last_end;
464 int err;
465 int set = 0;
466 int clear = 0;
467
468 if (delete)
469 bits |= ~EXTENT_CTLBITS;
470 bits |= EXTENT_FIRST_DELALLOC;
471
472 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
473 clear = 1;
474 again:
475 if (!prealloc && (mask & __GFP_WAIT)) {
476 prealloc = alloc_extent_state(mask);
477 if (!prealloc)
478 return -ENOMEM;
479 }
480
481 spin_lock(&tree->lock);
482 if (cached_state) {
483 cached = *cached_state;
484
485 if (clear) {
486 *cached_state = NULL;
487 cached_state = NULL;
488 }
489
490 if (cached && cached->tree && cached->start <= start &&
491 cached->end > start) {
492 if (clear)
493 atomic_dec(&cached->refs);
494 state = cached;
495 goto hit_next;
496 }
497 if (clear)
498 free_extent_state(cached);
499 }
500 /*
501 * this search will find the extents that end after
502 * our range starts
503 */
504 node = tree_search(tree, start);
505 if (!node)
506 goto out;
507 state = rb_entry(node, struct extent_state, rb_node);
508 hit_next:
509 if (state->start > end)
510 goto out;
511 WARN_ON(state->end < start);
512 last_end = state->end;
513
514 /*
515 * | ---- desired range ---- |
516 * | state | or
517 * | ------------- state -------------- |
518 *
519 * We need to split the extent we found, and may flip
520 * bits on second half.
521 *
522 * If the extent we found extends past our range, we
523 * just split and search again. It'll get split again
524 * the next time though.
525 *
526 * If the extent we found is inside our range, we clear
527 * the desired bit on it.
528 */
529
530 if (state->start < start) {
531 prealloc = alloc_extent_state_atomic(prealloc);
532 BUG_ON(!prealloc);
533 err = split_state(tree, state, prealloc, start);
534 BUG_ON(err == -EEXIST);
535 prealloc = NULL;
536 if (err)
537 goto out;
538 if (state->end <= end) {
539 set |= clear_state_bit(tree, state, &bits, wake);
540 if (last_end == (u64)-1)
541 goto out;
542 start = last_end + 1;
543 }
544 goto search_again;
545 }
546 /*
547 * | ---- desired range ---- |
548 * | state |
549 * We need to split the extent, and clear the bit
550 * on the first half
551 */
552 if (state->start <= end && state->end > end) {
553 prealloc = alloc_extent_state_atomic(prealloc);
554 BUG_ON(!prealloc);
555 err = split_state(tree, state, prealloc, end + 1);
556 BUG_ON(err == -EEXIST);
557 if (wake)
558 wake_up(&state->wq);
559
560 set |= clear_state_bit(tree, prealloc, &bits, wake);
561
562 prealloc = NULL;
563 goto out;
564 }
565
566 if (state->end < end && prealloc && !need_resched())
567 next_node = rb_next(&state->rb_node);
568 else
569 next_node = NULL;
570
571 set |= clear_state_bit(tree, state, &bits, wake);
572 if (last_end == (u64)-1)
573 goto out;
574 start = last_end + 1;
575 if (start <= end && next_node) {
576 state = rb_entry(next_node, struct extent_state,
577 rb_node);
578 if (state->start == start)
579 goto hit_next;
580 }
581 goto search_again;
582
583 out:
584 spin_unlock(&tree->lock);
585 if (prealloc)
586 free_extent_state(prealloc);
587
588 return set;
589
590 search_again:
591 if (start > end)
592 goto out;
593 spin_unlock(&tree->lock);
594 if (mask & __GFP_WAIT)
595 cond_resched();
596 goto again;
597 }
598
599 static int wait_on_state(struct extent_io_tree *tree,
600 struct extent_state *state)
601 __releases(tree->lock)
602 __acquires(tree->lock)
603 {
604 DEFINE_WAIT(wait);
605 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
606 spin_unlock(&tree->lock);
607 schedule();
608 spin_lock(&tree->lock);
609 finish_wait(&state->wq, &wait);
610 return 0;
611 }
612
613 /*
614 * waits for one or more bits to clear on a range in the state tree.
615 * The range [start, end] is inclusive.
616 * The tree lock is taken by this function
617 */
618 int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits)
619 {
620 struct extent_state *state;
621 struct rb_node *node;
622
623 spin_lock(&tree->lock);
624 again:
625 while (1) {
626 /*
627 * this search will find all the extents that end after
628 * our range starts
629 */
630 node = tree_search(tree, start);
631 if (!node)
632 break;
633
634 state = rb_entry(node, struct extent_state, rb_node);
635
636 if (state->start > end)
637 goto out;
638
639 if (state->state & bits) {
640 start = state->start;
641 atomic_inc(&state->refs);
642 wait_on_state(tree, state);
643 free_extent_state(state);
644 goto again;
645 }
646 start = state->end + 1;
647
648 if (start > end)
649 break;
650
651 cond_resched_lock(&tree->lock);
652 }
653 out:
654 spin_unlock(&tree->lock);
655 return 0;
656 }
657
658 static void set_state_bits(struct extent_io_tree *tree,
659 struct extent_state *state,
660 int *bits)
661 {
662 int bits_to_set = *bits & ~EXTENT_CTLBITS;
663
664 set_state_cb(tree, state, bits);
665 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
666 u64 range = state->end - state->start + 1;
667 tree->dirty_bytes += range;
668 }
669 state->state |= bits_to_set;
670 }
671
672 static void cache_state(struct extent_state *state,
673 struct extent_state **cached_ptr)
674 {
675 if (cached_ptr && !(*cached_ptr)) {
676 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
677 *cached_ptr = state;
678 atomic_inc(&state->refs);
679 }
680 }
681 }
682
683 static void uncache_state(struct extent_state **cached_ptr)
684 {
685 if (cached_ptr && (*cached_ptr)) {
686 struct extent_state *state = *cached_ptr;
687 *cached_ptr = NULL;
688 free_extent_state(state);
689 }
690 }
691
692 /*
693 * set some bits on a range in the tree. This may require allocations or
694 * sleeping, so the gfp mask is used to indicate what is allowed.
695 *
696 * If any of the exclusive bits are set, this will fail with -EEXIST if some
697 * part of the range already has the desired bits set. The start of the
698 * existing range is returned in failed_start in this case.
699 *
700 * [start, end] is inclusive This takes the tree lock.
701 */
702
703 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
704 int bits, int exclusive_bits, u64 *failed_start,
705 struct extent_state **cached_state, gfp_t mask)
706 {
707 struct extent_state *state;
708 struct extent_state *prealloc = NULL;
709 struct rb_node *node;
710 int err = 0;
711 u64 last_start;
712 u64 last_end;
713
714 bits |= EXTENT_FIRST_DELALLOC;
715 again:
716 if (!prealloc && (mask & __GFP_WAIT)) {
717 prealloc = alloc_extent_state(mask);
718 BUG_ON(!prealloc);
719 }
720
721 spin_lock(&tree->lock);
722 if (cached_state && *cached_state) {
723 state = *cached_state;
724 if (state->start <= start && state->end > start &&
725 state->tree) {
726 node = &state->rb_node;
727 goto hit_next;
728 }
729 }
730 /*
731 * this search will find all the extents that end after
732 * our range starts.
733 */
734 node = tree_search(tree, start);
735 if (!node) {
736 prealloc = alloc_extent_state_atomic(prealloc);
737 BUG_ON(!prealloc);
738 err = insert_state(tree, prealloc, start, end, &bits);
739 prealloc = NULL;
740 BUG_ON(err == -EEXIST);
741 goto out;
742 }
743 state = rb_entry(node, struct extent_state, rb_node);
744 hit_next:
745 last_start = state->start;
746 last_end = state->end;
747
748 /*
749 * | ---- desired range ---- |
750 * | state |
751 *
752 * Just lock what we found and keep going
753 */
754 if (state->start == start && state->end <= end) {
755 struct rb_node *next_node;
756 if (state->state & exclusive_bits) {
757 *failed_start = state->start;
758 err = -EEXIST;
759 goto out;
760 }
761
762 set_state_bits(tree, state, &bits);
763
764 cache_state(state, cached_state);
765 merge_state(tree, state);
766 if (last_end == (u64)-1)
767 goto out;
768
769 start = last_end + 1;
770 next_node = rb_next(&state->rb_node);
771 if (next_node && start < end && prealloc && !need_resched()) {
772 state = rb_entry(next_node, struct extent_state,
773 rb_node);
774 if (state->start == start)
775 goto hit_next;
776 }
777 goto search_again;
778 }
779
780 /*
781 * | ---- desired range ---- |
782 * | state |
783 * or
784 * | ------------- state -------------- |
785 *
786 * We need to split the extent we found, and may flip bits on
787 * second half.
788 *
789 * If the extent we found extends past our
790 * range, we just split and search again. It'll get split
791 * again the next time though.
792 *
793 * If the extent we found is inside our range, we set the
794 * desired bit on it.
795 */
796 if (state->start < start) {
797 if (state->state & exclusive_bits) {
798 *failed_start = start;
799 err = -EEXIST;
800 goto out;
801 }
802
803 prealloc = alloc_extent_state_atomic(prealloc);
804 BUG_ON(!prealloc);
805 err = split_state(tree, state, prealloc, start);
806 BUG_ON(err == -EEXIST);
807 prealloc = NULL;
808 if (err)
809 goto out;
810 if (state->end <= end) {
811 set_state_bits(tree, state, &bits);
812 cache_state(state, cached_state);
813 merge_state(tree, state);
814 if (last_end == (u64)-1)
815 goto out;
816 start = last_end + 1;
817 }
818 goto search_again;
819 }
820 /*
821 * | ---- desired range ---- |
822 * | state | or | state |
823 *
824 * There's a hole, we need to insert something in it and
825 * ignore the extent we found.
826 */
827 if (state->start > start) {
828 u64 this_end;
829 if (end < last_start)
830 this_end = end;
831 else
832 this_end = last_start - 1;
833
834 prealloc = alloc_extent_state_atomic(prealloc);
835 BUG_ON(!prealloc);
836
837 /*
838 * Avoid to free 'prealloc' if it can be merged with
839 * the later extent.
840 */
841 err = insert_state(tree, prealloc, start, this_end,
842 &bits);
843 BUG_ON(err == -EEXIST);
844 if (err) {
845 free_extent_state(prealloc);
846 prealloc = NULL;
847 goto out;
848 }
849 cache_state(prealloc, cached_state);
850 prealloc = NULL;
851 start = this_end + 1;
852 goto search_again;
853 }
854 /*
855 * | ---- desired range ---- |
856 * | state |
857 * We need to split the extent, and set the bit
858 * on the first half
859 */
860 if (state->start <= end && state->end > end) {
861 if (state->state & exclusive_bits) {
862 *failed_start = start;
863 err = -EEXIST;
864 goto out;
865 }
866
867 prealloc = alloc_extent_state_atomic(prealloc);
868 BUG_ON(!prealloc);
869 err = split_state(tree, state, prealloc, end + 1);
870 BUG_ON(err == -EEXIST);
871
872 set_state_bits(tree, prealloc, &bits);
873 cache_state(prealloc, cached_state);
874 merge_state(tree, prealloc);
875 prealloc = NULL;
876 goto out;
877 }
878
879 goto search_again;
880
881 out:
882 spin_unlock(&tree->lock);
883 if (prealloc)
884 free_extent_state(prealloc);
885
886 return err;
887
888 search_again:
889 if (start > end)
890 goto out;
891 spin_unlock(&tree->lock);
892 if (mask & __GFP_WAIT)
893 cond_resched();
894 goto again;
895 }
896
897 /**
898 * convert_extent - convert all bits in a given range from one bit to another
899 * @tree: the io tree to search
900 * @start: the start offset in bytes
901 * @end: the end offset in bytes (inclusive)
902 * @bits: the bits to set in this range
903 * @clear_bits: the bits to clear in this range
904 * @mask: the allocation mask
905 *
906 * This will go through and set bits for the given range. If any states exist
907 * already in this range they are set with the given bit and cleared of the
908 * clear_bits. This is only meant to be used by things that are mergeable, ie
909 * converting from say DELALLOC to DIRTY. This is not meant to be used with
910 * boundary bits like LOCK.
911 */
912 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
913 int bits, int clear_bits, gfp_t mask)
914 {
915 struct extent_state *state;
916 struct extent_state *prealloc = NULL;
917 struct rb_node *node;
918 int err = 0;
919 u64 last_start;
920 u64 last_end;
921
922 again:
923 if (!prealloc && (mask & __GFP_WAIT)) {
924 prealloc = alloc_extent_state(mask);
925 if (!prealloc)
926 return -ENOMEM;
927 }
928
929 spin_lock(&tree->lock);
930 /*
931 * this search will find all the extents that end after
932 * our range starts.
933 */
934 node = tree_search(tree, start);
935 if (!node) {
936 prealloc = alloc_extent_state_atomic(prealloc);
937 if (!prealloc)
938 return -ENOMEM;
939 err = insert_state(tree, prealloc, start, end, &bits);
940 prealloc = NULL;
941 BUG_ON(err == -EEXIST);
942 goto out;
943 }
944 state = rb_entry(node, struct extent_state, rb_node);
945 hit_next:
946 last_start = state->start;
947 last_end = state->end;
948
949 /*
950 * | ---- desired range ---- |
951 * | state |
952 *
953 * Just lock what we found and keep going
954 */
955 if (state->start == start && state->end <= end) {
956 struct rb_node *next_node;
957
958 set_state_bits(tree, state, &bits);
959 clear_state_bit(tree, state, &clear_bits, 0);
960
961 merge_state(tree, state);
962 if (last_end == (u64)-1)
963 goto out;
964
965 start = last_end + 1;
966 next_node = rb_next(&state->rb_node);
967 if (next_node && start < end && prealloc && !need_resched()) {
968 state = rb_entry(next_node, struct extent_state,
969 rb_node);
970 if (state->start == start)
971 goto hit_next;
972 }
973 goto search_again;
974 }
975
976 /*
977 * | ---- desired range ---- |
978 * | state |
979 * or
980 * | ------------- state -------------- |
981 *
982 * We need to split the extent we found, and may flip bits on
983 * second half.
984 *
985 * If the extent we found extends past our
986 * range, we just split and search again. It'll get split
987 * again the next time though.
988 *
989 * If the extent we found is inside our range, we set the
990 * desired bit on it.
991 */
992 if (state->start < start) {
993 prealloc = alloc_extent_state_atomic(prealloc);
994 if (!prealloc)
995 return -ENOMEM;
996 err = split_state(tree, state, prealloc, start);
997 BUG_ON(err == -EEXIST);
998 prealloc = NULL;
999 if (err)
1000 goto out;
1001 if (state->end <= end) {
1002 set_state_bits(tree, state, &bits);
1003 clear_state_bit(tree, state, &clear_bits, 0);
1004 merge_state(tree, state);
1005 if (last_end == (u64)-1)
1006 goto out;
1007 start = last_end + 1;
1008 }
1009 goto search_again;
1010 }
1011 /*
1012 * | ---- desired range ---- |
1013 * | state | or | state |
1014 *
1015 * There's a hole, we need to insert something in it and
1016 * ignore the extent we found.
1017 */
1018 if (state->start > start) {
1019 u64 this_end;
1020 if (end < last_start)
1021 this_end = end;
1022 else
1023 this_end = last_start - 1;
1024
1025 prealloc = alloc_extent_state_atomic(prealloc);
1026 if (!prealloc)
1027 return -ENOMEM;
1028
1029 /*
1030 * Avoid to free 'prealloc' if it can be merged with
1031 * the later extent.
1032 */
1033 err = insert_state(tree, prealloc, start, this_end,
1034 &bits);
1035 BUG_ON(err == -EEXIST);
1036 if (err) {
1037 free_extent_state(prealloc);
1038 prealloc = NULL;
1039 goto out;
1040 }
1041 prealloc = NULL;
1042 start = this_end + 1;
1043 goto search_again;
1044 }
1045 /*
1046 * | ---- desired range ---- |
1047 * | state |
1048 * We need to split the extent, and set the bit
1049 * on the first half
1050 */
1051 if (state->start <= end && state->end > end) {
1052 prealloc = alloc_extent_state_atomic(prealloc);
1053 if (!prealloc)
1054 return -ENOMEM;
1055
1056 err = split_state(tree, state, prealloc, end + 1);
1057 BUG_ON(err == -EEXIST);
1058
1059 set_state_bits(tree, prealloc, &bits);
1060 clear_state_bit(tree, prealloc, &clear_bits, 0);
1061
1062 merge_state(tree, prealloc);
1063 prealloc = NULL;
1064 goto out;
1065 }
1066
1067 goto search_again;
1068
1069 out:
1070 spin_unlock(&tree->lock);
1071 if (prealloc)
1072 free_extent_state(prealloc);
1073
1074 return err;
1075
1076 search_again:
1077 if (start > end)
1078 goto out;
1079 spin_unlock(&tree->lock);
1080 if (mask & __GFP_WAIT)
1081 cond_resched();
1082 goto again;
1083 }
1084
1085 /* wrappers around set/clear extent bit */
1086 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1087 gfp_t mask)
1088 {
1089 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
1090 NULL, mask);
1091 }
1092
1093 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1094 int bits, gfp_t mask)
1095 {
1096 return set_extent_bit(tree, start, end, bits, 0, NULL,
1097 NULL, mask);
1098 }
1099
1100 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1101 int bits, gfp_t mask)
1102 {
1103 return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1104 }
1105
1106 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1107 struct extent_state **cached_state, gfp_t mask)
1108 {
1109 return set_extent_bit(tree, start, end,
1110 EXTENT_DELALLOC | EXTENT_UPTODATE,
1111 0, NULL, cached_state, mask);
1112 }
1113
1114 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1115 gfp_t mask)
1116 {
1117 return clear_extent_bit(tree, start, end,
1118 EXTENT_DIRTY | EXTENT_DELALLOC |
1119 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1120 }
1121
1122 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1123 gfp_t mask)
1124 {
1125 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
1126 NULL, mask);
1127 }
1128
1129 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1130 struct extent_state **cached_state, gfp_t mask)
1131 {
1132 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0,
1133 NULL, cached_state, mask);
1134 }
1135
1136 static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
1137 u64 end, struct extent_state **cached_state,
1138 gfp_t mask)
1139 {
1140 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1141 cached_state, mask);
1142 }
1143
1144 /*
1145 * either insert or lock state struct between start and end use mask to tell
1146 * us if waiting is desired.
1147 */
1148 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1149 int bits, struct extent_state **cached_state, gfp_t mask)
1150 {
1151 int err;
1152 u64 failed_start;
1153 while (1) {
1154 err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1155 EXTENT_LOCKED, &failed_start,
1156 cached_state, mask);
1157 if (err == -EEXIST && (mask & __GFP_WAIT)) {
1158 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1159 start = failed_start;
1160 } else {
1161 break;
1162 }
1163 WARN_ON(start > end);
1164 }
1165 return err;
1166 }
1167
1168 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1169 {
1170 return lock_extent_bits(tree, start, end, 0, NULL, mask);
1171 }
1172
1173 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
1174 gfp_t mask)
1175 {
1176 int err;
1177 u64 failed_start;
1178
1179 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1180 &failed_start, NULL, mask);
1181 if (err == -EEXIST) {
1182 if (failed_start > start)
1183 clear_extent_bit(tree, start, failed_start - 1,
1184 EXTENT_LOCKED, 1, 0, NULL, mask);
1185 return 0;
1186 }
1187 return 1;
1188 }
1189
1190 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1191 struct extent_state **cached, gfp_t mask)
1192 {
1193 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1194 mask);
1195 }
1196
1197 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask)
1198 {
1199 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1200 mask);
1201 }
1202
1203 /*
1204 * helper function to set both pages and extents in the tree writeback
1205 */
1206 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1207 {
1208 unsigned long index = start >> PAGE_CACHE_SHIFT;
1209 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1210 struct page *page;
1211
1212 while (index <= end_index) {
1213 page = find_get_page(tree->mapping, index);
1214 BUG_ON(!page);
1215 set_page_writeback(page);
1216 page_cache_release(page);
1217 index++;
1218 }
1219 return 0;
1220 }
1221
1222 /* find the first state struct with 'bits' set after 'start', and
1223 * return it. tree->lock must be held. NULL will returned if
1224 * nothing was found after 'start'
1225 */
1226 struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
1227 u64 start, int bits)
1228 {
1229 struct rb_node *node;
1230 struct extent_state *state;
1231
1232 /*
1233 * this search will find all the extents that end after
1234 * our range starts.
1235 */
1236 node = tree_search(tree, start);
1237 if (!node)
1238 goto out;
1239
1240 while (1) {
1241 state = rb_entry(node, struct extent_state, rb_node);
1242 if (state->end >= start && (state->state & bits))
1243 return state;
1244
1245 node = rb_next(node);
1246 if (!node)
1247 break;
1248 }
1249 out:
1250 return NULL;
1251 }
1252
1253 /*
1254 * find the first offset in the io tree with 'bits' set. zero is
1255 * returned if we find something, and *start_ret and *end_ret are
1256 * set to reflect the state struct that was found.
1257 *
1258 * If nothing was found, 1 is returned, < 0 on error
1259 */
1260 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1261 u64 *start_ret, u64 *end_ret, int bits)
1262 {
1263 struct extent_state *state;
1264 int ret = 1;
1265
1266 spin_lock(&tree->lock);
1267 state = find_first_extent_bit_state(tree, start, bits);
1268 if (state) {
1269 *start_ret = state->start;
1270 *end_ret = state->end;
1271 ret = 0;
1272 }
1273 spin_unlock(&tree->lock);
1274 return ret;
1275 }
1276
1277 /*
1278 * find a contiguous range of bytes in the file marked as delalloc, not
1279 * more than 'max_bytes'. start and end are used to return the range,
1280 *
1281 * 1 is returned if we find something, 0 if nothing was in the tree
1282 */
1283 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1284 u64 *start, u64 *end, u64 max_bytes,
1285 struct extent_state **cached_state)
1286 {
1287 struct rb_node *node;
1288 struct extent_state *state;
1289 u64 cur_start = *start;
1290 u64 found = 0;
1291 u64 total_bytes = 0;
1292
1293 spin_lock(&tree->lock);
1294
1295 /*
1296 * this search will find all the extents that end after
1297 * our range starts.
1298 */
1299 node = tree_search(tree, cur_start);
1300 if (!node) {
1301 if (!found)
1302 *end = (u64)-1;
1303 goto out;
1304 }
1305
1306 while (1) {
1307 state = rb_entry(node, struct extent_state, rb_node);
1308 if (found && (state->start != cur_start ||
1309 (state->state & EXTENT_BOUNDARY))) {
1310 goto out;
1311 }
1312 if (!(state->state & EXTENT_DELALLOC)) {
1313 if (!found)
1314 *end = state->end;
1315 goto out;
1316 }
1317 if (!found) {
1318 *start = state->start;
1319 *cached_state = state;
1320 atomic_inc(&state->refs);
1321 }
1322 found++;
1323 *end = state->end;
1324 cur_start = state->end + 1;
1325 node = rb_next(node);
1326 if (!node)
1327 break;
1328 total_bytes += state->end - state->start + 1;
1329 if (total_bytes >= max_bytes)
1330 break;
1331 }
1332 out:
1333 spin_unlock(&tree->lock);
1334 return found;
1335 }
1336
1337 static noinline int __unlock_for_delalloc(struct inode *inode,
1338 struct page *locked_page,
1339 u64 start, u64 end)
1340 {
1341 int ret;
1342 struct page *pages[16];
1343 unsigned long index = start >> PAGE_CACHE_SHIFT;
1344 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1345 unsigned long nr_pages = end_index - index + 1;
1346 int i;
1347
1348 if (index == locked_page->index && end_index == index)
1349 return 0;
1350
1351 while (nr_pages > 0) {
1352 ret = find_get_pages_contig(inode->i_mapping, index,
1353 min_t(unsigned long, nr_pages,
1354 ARRAY_SIZE(pages)), pages);
1355 for (i = 0; i < ret; i++) {
1356 if (pages[i] != locked_page)
1357 unlock_page(pages[i]);
1358 page_cache_release(pages[i]);
1359 }
1360 nr_pages -= ret;
1361 index += ret;
1362 cond_resched();
1363 }
1364 return 0;
1365 }
1366
1367 static noinline int lock_delalloc_pages(struct inode *inode,
1368 struct page *locked_page,
1369 u64 delalloc_start,
1370 u64 delalloc_end)
1371 {
1372 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1373 unsigned long start_index = index;
1374 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1375 unsigned long pages_locked = 0;
1376 struct page *pages[16];
1377 unsigned long nrpages;
1378 int ret;
1379 int i;
1380
1381 /* the caller is responsible for locking the start index */
1382 if (index == locked_page->index && index == end_index)
1383 return 0;
1384
1385 /* skip the page at the start index */
1386 nrpages = end_index - index + 1;
1387 while (nrpages > 0) {
1388 ret = find_get_pages_contig(inode->i_mapping, index,
1389 min_t(unsigned long,
1390 nrpages, ARRAY_SIZE(pages)), pages);
1391 if (ret == 0) {
1392 ret = -EAGAIN;
1393 goto done;
1394 }
1395 /* now we have an array of pages, lock them all */
1396 for (i = 0; i < ret; i++) {
1397 /*
1398 * the caller is taking responsibility for
1399 * locked_page
1400 */
1401 if (pages[i] != locked_page) {
1402 lock_page(pages[i]);
1403 if (!PageDirty(pages[i]) ||
1404 pages[i]->mapping != inode->i_mapping) {
1405 ret = -EAGAIN;
1406 unlock_page(pages[i]);
1407 page_cache_release(pages[i]);
1408 goto done;
1409 }
1410 }
1411 page_cache_release(pages[i]);
1412 pages_locked++;
1413 }
1414 nrpages -= ret;
1415 index += ret;
1416 cond_resched();
1417 }
1418 ret = 0;
1419 done:
1420 if (ret && pages_locked) {
1421 __unlock_for_delalloc(inode, locked_page,
1422 delalloc_start,
1423 ((u64)(start_index + pages_locked - 1)) <<
1424 PAGE_CACHE_SHIFT);
1425 }
1426 return ret;
1427 }
1428
1429 /*
1430 * find a contiguous range of bytes in the file marked as delalloc, not
1431 * more than 'max_bytes'. start and end are used to return the range,
1432 *
1433 * 1 is returned if we find something, 0 if nothing was in the tree
1434 */
1435 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1436 struct extent_io_tree *tree,
1437 struct page *locked_page,
1438 u64 *start, u64 *end,
1439 u64 max_bytes)
1440 {
1441 u64 delalloc_start;
1442 u64 delalloc_end;
1443 u64 found;
1444 struct extent_state *cached_state = NULL;
1445 int ret;
1446 int loops = 0;
1447
1448 again:
1449 /* step one, find a bunch of delalloc bytes starting at start */
1450 delalloc_start = *start;
1451 delalloc_end = 0;
1452 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1453 max_bytes, &cached_state);
1454 if (!found || delalloc_end <= *start) {
1455 *start = delalloc_start;
1456 *end = delalloc_end;
1457 free_extent_state(cached_state);
1458 return found;
1459 }
1460
1461 /*
1462 * start comes from the offset of locked_page. We have to lock
1463 * pages in order, so we can't process delalloc bytes before
1464 * locked_page
1465 */
1466 if (delalloc_start < *start)
1467 delalloc_start = *start;
1468
1469 /*
1470 * make sure to limit the number of pages we try to lock down
1471 * if we're looping.
1472 */
1473 if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1474 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1475
1476 /* step two, lock all the pages after the page that has start */
1477 ret = lock_delalloc_pages(inode, locked_page,
1478 delalloc_start, delalloc_end);
1479 if (ret == -EAGAIN) {
1480 /* some of the pages are gone, lets avoid looping by
1481 * shortening the size of the delalloc range we're searching
1482 */
1483 free_extent_state(cached_state);
1484 if (!loops) {
1485 unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1486 max_bytes = PAGE_CACHE_SIZE - offset;
1487 loops = 1;
1488 goto again;
1489 } else {
1490 found = 0;
1491 goto out_failed;
1492 }
1493 }
1494 BUG_ON(ret);
1495
1496 /* step three, lock the state bits for the whole range */
1497 lock_extent_bits(tree, delalloc_start, delalloc_end,
1498 0, &cached_state, GFP_NOFS);
1499
1500 /* then test to make sure it is all still delalloc */
1501 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1502 EXTENT_DELALLOC, 1, cached_state);
1503 if (!ret) {
1504 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1505 &cached_state, GFP_NOFS);
1506 __unlock_for_delalloc(inode, locked_page,
1507 delalloc_start, delalloc_end);
1508 cond_resched();
1509 goto again;
1510 }
1511 free_extent_state(cached_state);
1512 *start = delalloc_start;
1513 *end = delalloc_end;
1514 out_failed:
1515 return found;
1516 }
1517
1518 int extent_clear_unlock_delalloc(struct inode *inode,
1519 struct extent_io_tree *tree,
1520 u64 start, u64 end, struct page *locked_page,
1521 unsigned long op)
1522 {
1523 int ret;
1524 struct page *pages[16];
1525 unsigned long index = start >> PAGE_CACHE_SHIFT;
1526 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1527 unsigned long nr_pages = end_index - index + 1;
1528 int i;
1529 int clear_bits = 0;
1530
1531 if (op & EXTENT_CLEAR_UNLOCK)
1532 clear_bits |= EXTENT_LOCKED;
1533 if (op & EXTENT_CLEAR_DIRTY)
1534 clear_bits |= EXTENT_DIRTY;
1535
1536 if (op & EXTENT_CLEAR_DELALLOC)
1537 clear_bits |= EXTENT_DELALLOC;
1538
1539 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1540 if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1541 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1542 EXTENT_SET_PRIVATE2)))
1543 return 0;
1544
1545 while (nr_pages > 0) {
1546 ret = find_get_pages_contig(inode->i_mapping, index,
1547 min_t(unsigned long,
1548 nr_pages, ARRAY_SIZE(pages)), pages);
1549 for (i = 0; i < ret; i++) {
1550
1551 if (op & EXTENT_SET_PRIVATE2)
1552 SetPagePrivate2(pages[i]);
1553
1554 if (pages[i] == locked_page) {
1555 page_cache_release(pages[i]);
1556 continue;
1557 }
1558 if (op & EXTENT_CLEAR_DIRTY)
1559 clear_page_dirty_for_io(pages[i]);
1560 if (op & EXTENT_SET_WRITEBACK)
1561 set_page_writeback(pages[i]);
1562 if (op & EXTENT_END_WRITEBACK)
1563 end_page_writeback(pages[i]);
1564 if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1565 unlock_page(pages[i]);
1566 page_cache_release(pages[i]);
1567 }
1568 nr_pages -= ret;
1569 index += ret;
1570 cond_resched();
1571 }
1572 return 0;
1573 }
1574
1575 /*
1576 * count the number of bytes in the tree that have a given bit(s)
1577 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1578 * cached. The total number found is returned.
1579 */
1580 u64 count_range_bits(struct extent_io_tree *tree,
1581 u64 *start, u64 search_end, u64 max_bytes,
1582 unsigned long bits, int contig)
1583 {
1584 struct rb_node *node;
1585 struct extent_state *state;
1586 u64 cur_start = *start;
1587 u64 total_bytes = 0;
1588 u64 last = 0;
1589 int found = 0;
1590
1591 if (search_end <= cur_start) {
1592 WARN_ON(1);
1593 return 0;
1594 }
1595
1596 spin_lock(&tree->lock);
1597 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1598 total_bytes = tree->dirty_bytes;
1599 goto out;
1600 }
1601 /*
1602 * this search will find all the extents that end after
1603 * our range starts.
1604 */
1605 node = tree_search(tree, cur_start);
1606 if (!node)
1607 goto out;
1608
1609 while (1) {
1610 state = rb_entry(node, struct extent_state, rb_node);
1611 if (state->start > search_end)
1612 break;
1613 if (contig && found && state->start > last + 1)
1614 break;
1615 if (state->end >= cur_start && (state->state & bits) == bits) {
1616 total_bytes += min(search_end, state->end) + 1 -
1617 max(cur_start, state->start);
1618 if (total_bytes >= max_bytes)
1619 break;
1620 if (!found) {
1621 *start = max(cur_start, state->start);
1622 found = 1;
1623 }
1624 last = state->end;
1625 } else if (contig && found) {
1626 break;
1627 }
1628 node = rb_next(node);
1629 if (!node)
1630 break;
1631 }
1632 out:
1633 spin_unlock(&tree->lock);
1634 return total_bytes;
1635 }
1636
1637 /*
1638 * set the private field for a given byte offset in the tree. If there isn't
1639 * an extent_state there already, this does nothing.
1640 */
1641 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1642 {
1643 struct rb_node *node;
1644 struct extent_state *state;
1645 int ret = 0;
1646
1647 spin_lock(&tree->lock);
1648 /*
1649 * this search will find all the extents that end after
1650 * our range starts.
1651 */
1652 node = tree_search(tree, start);
1653 if (!node) {
1654 ret = -ENOENT;
1655 goto out;
1656 }
1657 state = rb_entry(node, struct extent_state, rb_node);
1658 if (state->start != start) {
1659 ret = -ENOENT;
1660 goto out;
1661 }
1662 state->private = private;
1663 out:
1664 spin_unlock(&tree->lock);
1665 return ret;
1666 }
1667
1668 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1669 {
1670 struct rb_node *node;
1671 struct extent_state *state;
1672 int ret = 0;
1673
1674 spin_lock(&tree->lock);
1675 /*
1676 * this search will find all the extents that end after
1677 * our range starts.
1678 */
1679 node = tree_search(tree, start);
1680 if (!node) {
1681 ret = -ENOENT;
1682 goto out;
1683 }
1684 state = rb_entry(node, struct extent_state, rb_node);
1685 if (state->start != start) {
1686 ret = -ENOENT;
1687 goto out;
1688 }
1689 *private = state->private;
1690 out:
1691 spin_unlock(&tree->lock);
1692 return ret;
1693 }
1694
1695 /*
1696 * searches a range in the state tree for a given mask.
1697 * If 'filled' == 1, this returns 1 only if every extent in the tree
1698 * has the bits set. Otherwise, 1 is returned if any bit in the
1699 * range is found set.
1700 */
1701 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1702 int bits, int filled, struct extent_state *cached)
1703 {
1704 struct extent_state *state = NULL;
1705 struct rb_node *node;
1706 int bitset = 0;
1707
1708 spin_lock(&tree->lock);
1709 if (cached && cached->tree && cached->start <= start &&
1710 cached->end > start)
1711 node = &cached->rb_node;
1712 else
1713 node = tree_search(tree, start);
1714 while (node && start <= end) {
1715 state = rb_entry(node, struct extent_state, rb_node);
1716
1717 if (filled && state->start > start) {
1718 bitset = 0;
1719 break;
1720 }
1721
1722 if (state->start > end)
1723 break;
1724
1725 if (state->state & bits) {
1726 bitset = 1;
1727 if (!filled)
1728 break;
1729 } else if (filled) {
1730 bitset = 0;
1731 break;
1732 }
1733
1734 if (state->end == (u64)-1)
1735 break;
1736
1737 start = state->end + 1;
1738 if (start > end)
1739 break;
1740 node = rb_next(node);
1741 if (!node) {
1742 if (filled)
1743 bitset = 0;
1744 break;
1745 }
1746 }
1747 spin_unlock(&tree->lock);
1748 return bitset;
1749 }
1750
1751 /*
1752 * helper function to set a given page up to date if all the
1753 * extents in the tree for that page are up to date
1754 */
1755 static int check_page_uptodate(struct extent_io_tree *tree,
1756 struct page *page)
1757 {
1758 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1759 u64 end = start + PAGE_CACHE_SIZE - 1;
1760 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1761 SetPageUptodate(page);
1762 return 0;
1763 }
1764
1765 /*
1766 * helper function to unlock a page if all the extents in the tree
1767 * for that page are unlocked
1768 */
1769 static int check_page_locked(struct extent_io_tree *tree,
1770 struct page *page)
1771 {
1772 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
1773 u64 end = start + PAGE_CACHE_SIZE - 1;
1774 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL))
1775 unlock_page(page);
1776 return 0;
1777 }
1778
1779 /*
1780 * helper function to end page writeback if all the extents
1781 * in the tree for that page are done with writeback
1782 */
1783 static int check_page_writeback(struct extent_io_tree *tree,
1784 struct page *page)
1785 {
1786 end_page_writeback(page);
1787 return 0;
1788 }
1789
1790 /* lots and lots of room for performance fixes in the end_bio funcs */
1791
1792 /*
1793 * after a writepage IO is done, we need to:
1794 * clear the uptodate bits on error
1795 * clear the writeback bits in the extent tree for this IO
1796 * end_page_writeback if the page has no more pending IO
1797 *
1798 * Scheduling is not allowed, so the extent state tree is expected
1799 * to have one and only one object corresponding to this IO.
1800 */
1801 static void end_bio_extent_writepage(struct bio *bio, int err)
1802 {
1803 int uptodate = err == 0;
1804 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1805 struct extent_io_tree *tree;
1806 u64 start;
1807 u64 end;
1808 int whole_page;
1809 int ret;
1810
1811 do {
1812 struct page *page = bvec->bv_page;
1813 tree = &BTRFS_I(page->mapping->host)->io_tree;
1814
1815 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1816 bvec->bv_offset;
1817 end = start + bvec->bv_len - 1;
1818
1819 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1820 whole_page = 1;
1821 else
1822 whole_page = 0;
1823
1824 if (--bvec >= bio->bi_io_vec)
1825 prefetchw(&bvec->bv_page->flags);
1826 if (tree->ops && tree->ops->writepage_end_io_hook) {
1827 ret = tree->ops->writepage_end_io_hook(page, start,
1828 end, NULL, uptodate);
1829 if (ret)
1830 uptodate = 0;
1831 }
1832
1833 if (!uptodate && tree->ops &&
1834 tree->ops->writepage_io_failed_hook) {
1835 ret = tree->ops->writepage_io_failed_hook(bio, page,
1836 start, end, NULL);
1837 if (ret == 0) {
1838 uptodate = (err == 0);
1839 continue;
1840 }
1841 }
1842
1843 if (!uptodate) {
1844 clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS);
1845 ClearPageUptodate(page);
1846 SetPageError(page);
1847 }
1848
1849 if (whole_page)
1850 end_page_writeback(page);
1851 else
1852 check_page_writeback(tree, page);
1853 } while (bvec >= bio->bi_io_vec);
1854
1855 bio_put(bio);
1856 }
1857
1858 /*
1859 * after a readpage IO is done, we need to:
1860 * clear the uptodate bits on error
1861 * set the uptodate bits if things worked
1862 * set the page up to date if all extents in the tree are uptodate
1863 * clear the lock bit in the extent tree
1864 * unlock the page if there are no other extents locked for it
1865 *
1866 * Scheduling is not allowed, so the extent state tree is expected
1867 * to have one and only one object corresponding to this IO.
1868 */
1869 static void end_bio_extent_readpage(struct bio *bio, int err)
1870 {
1871 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1872 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
1873 struct bio_vec *bvec = bio->bi_io_vec;
1874 struct extent_io_tree *tree;
1875 u64 start;
1876 u64 end;
1877 int whole_page;
1878 int ret;
1879
1880 if (err)
1881 uptodate = 0;
1882
1883 do {
1884 struct page *page = bvec->bv_page;
1885 struct extent_state *cached = NULL;
1886 struct extent_state *state;
1887
1888 tree = &BTRFS_I(page->mapping->host)->io_tree;
1889
1890 start = ((u64)page->index << PAGE_CACHE_SHIFT) +
1891 bvec->bv_offset;
1892 end = start + bvec->bv_len - 1;
1893
1894 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1895 whole_page = 1;
1896 else
1897 whole_page = 0;
1898
1899 if (++bvec <= bvec_end)
1900 prefetchw(&bvec->bv_page->flags);
1901
1902 spin_lock(&tree->lock);
1903 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
1904 if (state && state->start == start) {
1905 /*
1906 * take a reference on the state, unlock will drop
1907 * the ref
1908 */
1909 cache_state(state, &cached);
1910 }
1911 spin_unlock(&tree->lock);
1912
1913 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
1914 ret = tree->ops->readpage_end_io_hook(page, start, end,
1915 state);
1916 if (ret)
1917 uptodate = 0;
1918 }
1919 if (!uptodate && tree->ops &&
1920 tree->ops->readpage_io_failed_hook) {
1921 ret = tree->ops->readpage_io_failed_hook(bio, page,
1922 start, end, NULL);
1923 if (ret == 0) {
1924 uptodate =
1925 test_bit(BIO_UPTODATE, &bio->bi_flags);
1926 if (err)
1927 uptodate = 0;
1928 uncache_state(&cached);
1929 continue;
1930 }
1931 }
1932
1933 if (uptodate) {
1934 set_extent_uptodate(tree, start, end, &cached,
1935 GFP_ATOMIC);
1936 }
1937 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
1938
1939 if (whole_page) {
1940 if (uptodate) {
1941 SetPageUptodate(page);
1942 } else {
1943 ClearPageUptodate(page);
1944 SetPageError(page);
1945 }
1946 unlock_page(page);
1947 } else {
1948 if (uptodate) {
1949 check_page_uptodate(tree, page);
1950 } else {
1951 ClearPageUptodate(page);
1952 SetPageError(page);
1953 }
1954 check_page_locked(tree, page);
1955 }
1956 } while (bvec <= bvec_end);
1957
1958 bio_put(bio);
1959 }
1960
1961 struct bio *
1962 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
1963 gfp_t gfp_flags)
1964 {
1965 struct bio *bio;
1966
1967 bio = bio_alloc(gfp_flags, nr_vecs);
1968
1969 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
1970 while (!bio && (nr_vecs /= 2))
1971 bio = bio_alloc(gfp_flags, nr_vecs);
1972 }
1973
1974 if (bio) {
1975 bio->bi_size = 0;
1976 bio->bi_bdev = bdev;
1977 bio->bi_sector = first_sector;
1978 }
1979 return bio;
1980 }
1981
1982 static int submit_one_bio(int rw, struct bio *bio, int mirror_num,
1983 unsigned long bio_flags)
1984 {
1985 int ret = 0;
1986 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1987 struct page *page = bvec->bv_page;
1988 struct extent_io_tree *tree = bio->bi_private;
1989 u64 start;
1990
1991 start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1992
1993 bio->bi_private = NULL;
1994
1995 bio_get(bio);
1996
1997 if (tree->ops && tree->ops->submit_bio_hook)
1998 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
1999 mirror_num, bio_flags, start);
2000 else
2001 submit_bio(rw, bio);
2002 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2003 ret = -EOPNOTSUPP;
2004 bio_put(bio);
2005 return ret;
2006 }
2007
2008 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2009 struct page *page, sector_t sector,
2010 size_t size, unsigned long offset,
2011 struct block_device *bdev,
2012 struct bio **bio_ret,
2013 unsigned long max_pages,
2014 bio_end_io_t end_io_func,
2015 int mirror_num,
2016 unsigned long prev_bio_flags,
2017 unsigned long bio_flags)
2018 {
2019 int ret = 0;
2020 struct bio *bio;
2021 int nr;
2022 int contig = 0;
2023 int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2024 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2025 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2026
2027 if (bio_ret && *bio_ret) {
2028 bio = *bio_ret;
2029 if (old_compressed)
2030 contig = bio->bi_sector == sector;
2031 else
2032 contig = bio->bi_sector + (bio->bi_size >> 9) ==
2033 sector;
2034
2035 if (prev_bio_flags != bio_flags || !contig ||
2036 (tree->ops && tree->ops->merge_bio_hook &&
2037 tree->ops->merge_bio_hook(page, offset, page_size, bio,
2038 bio_flags)) ||
2039 bio_add_page(bio, page, page_size, offset) < page_size) {
2040 ret = submit_one_bio(rw, bio, mirror_num,
2041 prev_bio_flags);
2042 bio = NULL;
2043 } else {
2044 return 0;
2045 }
2046 }
2047 if (this_compressed)
2048 nr = BIO_MAX_PAGES;
2049 else
2050 nr = bio_get_nr_vecs(bdev);
2051
2052 bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2053 if (!bio)
2054 return -ENOMEM;
2055
2056 bio_add_page(bio, page, page_size, offset);
2057 bio->bi_end_io = end_io_func;
2058 bio->bi_private = tree;
2059
2060 if (bio_ret)
2061 *bio_ret = bio;
2062 else
2063 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2064
2065 return ret;
2066 }
2067
2068 void set_page_extent_mapped(struct page *page)
2069 {
2070 if (!PagePrivate(page)) {
2071 SetPagePrivate(page);
2072 page_cache_get(page);
2073 set_page_private(page, EXTENT_PAGE_PRIVATE);
2074 }
2075 }
2076
2077 static void set_page_extent_head(struct page *page, unsigned long len)
2078 {
2079 WARN_ON(!PagePrivate(page));
2080 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2);
2081 }
2082
2083 /*
2084 * basic readpage implementation. Locked extent state structs are inserted
2085 * into the tree that are removed when the IO is done (by the end_io
2086 * handlers)
2087 */
2088 static int __extent_read_full_page(struct extent_io_tree *tree,
2089 struct page *page,
2090 get_extent_t *get_extent,
2091 struct bio **bio, int mirror_num,
2092 unsigned long *bio_flags)
2093 {
2094 struct inode *inode = page->mapping->host;
2095 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2096 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2097 u64 end;
2098 u64 cur = start;
2099 u64 extent_offset;
2100 u64 last_byte = i_size_read(inode);
2101 u64 block_start;
2102 u64 cur_end;
2103 sector_t sector;
2104 struct extent_map *em;
2105 struct block_device *bdev;
2106 struct btrfs_ordered_extent *ordered;
2107 int ret;
2108 int nr = 0;
2109 size_t pg_offset = 0;
2110 size_t iosize;
2111 size_t disk_io_size;
2112 size_t blocksize = inode->i_sb->s_blocksize;
2113 unsigned long this_bio_flag = 0;
2114
2115 set_page_extent_mapped(page);
2116
2117 if (!PageUptodate(page)) {
2118 if (cleancache_get_page(page) == 0) {
2119 BUG_ON(blocksize != PAGE_SIZE);
2120 goto out;
2121 }
2122 }
2123
2124 end = page_end;
2125 while (1) {
2126 lock_extent(tree, start, end, GFP_NOFS);
2127 ordered = btrfs_lookup_ordered_extent(inode, start);
2128 if (!ordered)
2129 break;
2130 unlock_extent(tree, start, end, GFP_NOFS);
2131 btrfs_start_ordered_extent(inode, ordered, 1);
2132 btrfs_put_ordered_extent(ordered);
2133 }
2134
2135 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2136 char *userpage;
2137 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2138
2139 if (zero_offset) {
2140 iosize = PAGE_CACHE_SIZE - zero_offset;
2141 userpage = kmap_atomic(page, KM_USER0);
2142 memset(userpage + zero_offset, 0, iosize);
2143 flush_dcache_page(page);
2144 kunmap_atomic(userpage, KM_USER0);
2145 }
2146 }
2147 while (cur <= end) {
2148 if (cur >= last_byte) {
2149 char *userpage;
2150 struct extent_state *cached = NULL;
2151
2152 iosize = PAGE_CACHE_SIZE - pg_offset;
2153 userpage = kmap_atomic(page, KM_USER0);
2154 memset(userpage + pg_offset, 0, iosize);
2155 flush_dcache_page(page);
2156 kunmap_atomic(userpage, KM_USER0);
2157 set_extent_uptodate(tree, cur, cur + iosize - 1,
2158 &cached, GFP_NOFS);
2159 unlock_extent_cached(tree, cur, cur + iosize - 1,
2160 &cached, GFP_NOFS);
2161 break;
2162 }
2163 em = get_extent(inode, page, pg_offset, cur,
2164 end - cur + 1, 0);
2165 if (IS_ERR_OR_NULL(em)) {
2166 SetPageError(page);
2167 unlock_extent(tree, cur, end, GFP_NOFS);
2168 break;
2169 }
2170 extent_offset = cur - em->start;
2171 BUG_ON(extent_map_end(em) <= cur);
2172 BUG_ON(end < cur);
2173
2174 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2175 this_bio_flag = EXTENT_BIO_COMPRESSED;
2176 extent_set_compress_type(&this_bio_flag,
2177 em->compress_type);
2178 }
2179
2180 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2181 cur_end = min(extent_map_end(em) - 1, end);
2182 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2183 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2184 disk_io_size = em->block_len;
2185 sector = em->block_start >> 9;
2186 } else {
2187 sector = (em->block_start + extent_offset) >> 9;
2188 disk_io_size = iosize;
2189 }
2190 bdev = em->bdev;
2191 block_start = em->block_start;
2192 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2193 block_start = EXTENT_MAP_HOLE;
2194 free_extent_map(em);
2195 em = NULL;
2196
2197 /* we've found a hole, just zero and go on */
2198 if (block_start == EXTENT_MAP_HOLE) {
2199 char *userpage;
2200 struct extent_state *cached = NULL;
2201
2202 userpage = kmap_atomic(page, KM_USER0);
2203 memset(userpage + pg_offset, 0, iosize);
2204 flush_dcache_page(page);
2205 kunmap_atomic(userpage, KM_USER0);
2206
2207 set_extent_uptodate(tree, cur, cur + iosize - 1,
2208 &cached, GFP_NOFS);
2209 unlock_extent_cached(tree, cur, cur + iosize - 1,
2210 &cached, GFP_NOFS);
2211 cur = cur + iosize;
2212 pg_offset += iosize;
2213 continue;
2214 }
2215 /* the get_extent function already copied into the page */
2216 if (test_range_bit(tree, cur, cur_end,
2217 EXTENT_UPTODATE, 1, NULL)) {
2218 check_page_uptodate(tree, page);
2219 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2220 cur = cur + iosize;
2221 pg_offset += iosize;
2222 continue;
2223 }
2224 /* we have an inline extent but it didn't get marked up
2225 * to date. Error out
2226 */
2227 if (block_start == EXTENT_MAP_INLINE) {
2228 SetPageError(page);
2229 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
2230 cur = cur + iosize;
2231 pg_offset += iosize;
2232 continue;
2233 }
2234
2235 ret = 0;
2236 if (tree->ops && tree->ops->readpage_io_hook) {
2237 ret = tree->ops->readpage_io_hook(page, cur,
2238 cur + iosize - 1);
2239 }
2240 if (!ret) {
2241 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2242 pnr -= page->index;
2243 ret = submit_extent_page(READ, tree, page,
2244 sector, disk_io_size, pg_offset,
2245 bdev, bio, pnr,
2246 end_bio_extent_readpage, mirror_num,
2247 *bio_flags,
2248 this_bio_flag);
2249 nr++;
2250 *bio_flags = this_bio_flag;
2251 }
2252 if (ret)
2253 SetPageError(page);
2254 cur = cur + iosize;
2255 pg_offset += iosize;
2256 }
2257 out:
2258 if (!nr) {
2259 if (!PageError(page))
2260 SetPageUptodate(page);
2261 unlock_page(page);
2262 }
2263 return 0;
2264 }
2265
2266 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2267 get_extent_t *get_extent)
2268 {
2269 struct bio *bio = NULL;
2270 unsigned long bio_flags = 0;
2271 int ret;
2272
2273 ret = __extent_read_full_page(tree, page, get_extent, &bio, 0,
2274 &bio_flags);
2275 if (bio)
2276 ret = submit_one_bio(READ, bio, 0, bio_flags);
2277 return ret;
2278 }
2279
2280 static noinline void update_nr_written(struct page *page,
2281 struct writeback_control *wbc,
2282 unsigned long nr_written)
2283 {
2284 wbc->nr_to_write -= nr_written;
2285 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2286 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2287 page->mapping->writeback_index = page->index + nr_written;
2288 }
2289
2290 /*
2291 * the writepage semantics are similar to regular writepage. extent
2292 * records are inserted to lock ranges in the tree, and as dirty areas
2293 * are found, they are marked writeback. Then the lock bits are removed
2294 * and the end_io handler clears the writeback ranges
2295 */
2296 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2297 void *data)
2298 {
2299 struct inode *inode = page->mapping->host;
2300 struct extent_page_data *epd = data;
2301 struct extent_io_tree *tree = epd->tree;
2302 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2303 u64 delalloc_start;
2304 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2305 u64 end;
2306 u64 cur = start;
2307 u64 extent_offset;
2308 u64 last_byte = i_size_read(inode);
2309 u64 block_start;
2310 u64 iosize;
2311 sector_t sector;
2312 struct extent_state *cached_state = NULL;
2313 struct extent_map *em;
2314 struct block_device *bdev;
2315 int ret;
2316 int nr = 0;
2317 size_t pg_offset = 0;
2318 size_t blocksize;
2319 loff_t i_size = i_size_read(inode);
2320 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2321 u64 nr_delalloc;
2322 u64 delalloc_end;
2323 int page_started;
2324 int compressed;
2325 int write_flags;
2326 unsigned long nr_written = 0;
2327 bool fill_delalloc = true;
2328
2329 if (wbc->sync_mode == WB_SYNC_ALL)
2330 write_flags = WRITE_SYNC;
2331 else
2332 write_flags = WRITE;
2333
2334 trace___extent_writepage(page, inode, wbc);
2335
2336 WARN_ON(!PageLocked(page));
2337 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2338 if (page->index > end_index ||
2339 (page->index == end_index && !pg_offset)) {
2340 page->mapping->a_ops->invalidatepage(page, 0);
2341 unlock_page(page);
2342 return 0;
2343 }
2344
2345 if (page->index == end_index) {
2346 char *userpage;
2347
2348 userpage = kmap_atomic(page, KM_USER0);
2349 memset(userpage + pg_offset, 0,
2350 PAGE_CACHE_SIZE - pg_offset);
2351 kunmap_atomic(userpage, KM_USER0);
2352 flush_dcache_page(page);
2353 }
2354 pg_offset = 0;
2355
2356 set_page_extent_mapped(page);
2357
2358 if (!tree->ops || !tree->ops->fill_delalloc)
2359 fill_delalloc = false;
2360
2361 delalloc_start = start;
2362 delalloc_end = 0;
2363 page_started = 0;
2364 if (!epd->extent_locked && fill_delalloc) {
2365 u64 delalloc_to_write = 0;
2366 /*
2367 * make sure the wbc mapping index is at least updated
2368 * to this page.
2369 */
2370 update_nr_written(page, wbc, 0);
2371
2372 while (delalloc_end < page_end) {
2373 nr_delalloc = find_lock_delalloc_range(inode, tree,
2374 page,
2375 &delalloc_start,
2376 &delalloc_end,
2377 128 * 1024 * 1024);
2378 if (nr_delalloc == 0) {
2379 delalloc_start = delalloc_end + 1;
2380 continue;
2381 }
2382 tree->ops->fill_delalloc(inode, page, delalloc_start,
2383 delalloc_end, &page_started,
2384 &nr_written);
2385 /*
2386 * delalloc_end is already one less than the total
2387 * length, so we don't subtract one from
2388 * PAGE_CACHE_SIZE
2389 */
2390 delalloc_to_write += (delalloc_end - delalloc_start +
2391 PAGE_CACHE_SIZE) >>
2392 PAGE_CACHE_SHIFT;
2393 delalloc_start = delalloc_end + 1;
2394 }
2395 if (wbc->nr_to_write < delalloc_to_write) {
2396 int thresh = 8192;
2397
2398 if (delalloc_to_write < thresh * 2)
2399 thresh = delalloc_to_write;
2400 wbc->nr_to_write = min_t(u64, delalloc_to_write,
2401 thresh);
2402 }
2403
2404 /* did the fill delalloc function already unlock and start
2405 * the IO?
2406 */
2407 if (page_started) {
2408 ret = 0;
2409 /*
2410 * we've unlocked the page, so we can't update
2411 * the mapping's writeback index, just update
2412 * nr_to_write.
2413 */
2414 wbc->nr_to_write -= nr_written;
2415 goto done_unlocked;
2416 }
2417 }
2418 if (tree->ops && tree->ops->writepage_start_hook) {
2419 ret = tree->ops->writepage_start_hook(page, start,
2420 page_end);
2421 if (ret == -EAGAIN) {
2422 redirty_page_for_writepage(wbc, page);
2423 update_nr_written(page, wbc, nr_written);
2424 unlock_page(page);
2425 ret = 0;
2426 goto done_unlocked;
2427 }
2428 }
2429
2430 /*
2431 * we don't want to touch the inode after unlocking the page,
2432 * so we update the mapping writeback index now
2433 */
2434 update_nr_written(page, wbc, nr_written + 1);
2435
2436 end = page_end;
2437 if (last_byte <= start) {
2438 if (tree->ops && tree->ops->writepage_end_io_hook)
2439 tree->ops->writepage_end_io_hook(page, start,
2440 page_end, NULL, 1);
2441 goto done;
2442 }
2443
2444 blocksize = inode->i_sb->s_blocksize;
2445
2446 while (cur <= end) {
2447 if (cur >= last_byte) {
2448 if (tree->ops && tree->ops->writepage_end_io_hook)
2449 tree->ops->writepage_end_io_hook(page, cur,
2450 page_end, NULL, 1);
2451 break;
2452 }
2453 em = epd->get_extent(inode, page, pg_offset, cur,
2454 end - cur + 1, 1);
2455 if (IS_ERR_OR_NULL(em)) {
2456 SetPageError(page);
2457 break;
2458 }
2459
2460 extent_offset = cur - em->start;
2461 BUG_ON(extent_map_end(em) <= cur);
2462 BUG_ON(end < cur);
2463 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2464 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
2465 sector = (em->block_start + extent_offset) >> 9;
2466 bdev = em->bdev;
2467 block_start = em->block_start;
2468 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2469 free_extent_map(em);
2470 em = NULL;
2471
2472 /*
2473 * compressed and inline extents are written through other
2474 * paths in the FS
2475 */
2476 if (compressed || block_start == EXTENT_MAP_HOLE ||
2477 block_start == EXTENT_MAP_INLINE) {
2478 /*
2479 * end_io notification does not happen here for
2480 * compressed extents
2481 */
2482 if (!compressed && tree->ops &&
2483 tree->ops->writepage_end_io_hook)
2484 tree->ops->writepage_end_io_hook(page, cur,
2485 cur + iosize - 1,
2486 NULL, 1);
2487 else if (compressed) {
2488 /* we don't want to end_page_writeback on
2489 * a compressed extent. this happens
2490 * elsewhere
2491 */
2492 nr++;
2493 }
2494
2495 cur += iosize;
2496 pg_offset += iosize;
2497 continue;
2498 }
2499 /* leave this out until we have a page_mkwrite call */
2500 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
2501 EXTENT_DIRTY, 0, NULL)) {
2502 cur = cur + iosize;
2503 pg_offset += iosize;
2504 continue;
2505 }
2506
2507 if (tree->ops && tree->ops->writepage_io_hook) {
2508 ret = tree->ops->writepage_io_hook(page, cur,
2509 cur + iosize - 1);
2510 } else {
2511 ret = 0;
2512 }
2513 if (ret) {
2514 SetPageError(page);
2515 } else {
2516 unsigned long max_nr = end_index + 1;
2517
2518 set_range_writeback(tree, cur, cur + iosize - 1);
2519 if (!PageWriteback(page)) {
2520 printk(KERN_ERR "btrfs warning page %lu not "
2521 "writeback, cur %llu end %llu\n",
2522 page->index, (unsigned long long)cur,
2523 (unsigned long long)end);
2524 }
2525
2526 ret = submit_extent_page(write_flags, tree, page,
2527 sector, iosize, pg_offset,
2528 bdev, &epd->bio, max_nr,
2529 end_bio_extent_writepage,
2530 0, 0, 0);
2531 if (ret)
2532 SetPageError(page);
2533 }
2534 cur = cur + iosize;
2535 pg_offset += iosize;
2536 nr++;
2537 }
2538 done:
2539 if (nr == 0) {
2540 /* make sure the mapping tag for page dirty gets cleared */
2541 set_page_writeback(page);
2542 end_page_writeback(page);
2543 }
2544 unlock_page(page);
2545
2546 done_unlocked:
2547
2548 /* drop our reference on any cached states */
2549 free_extent_state(cached_state);
2550 return 0;
2551 }
2552
2553 /**
2554 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
2555 * @mapping: address space structure to write
2556 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2557 * @writepage: function called for each page
2558 * @data: data passed to writepage function
2559 *
2560 * If a page is already under I/O, write_cache_pages() skips it, even
2561 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2562 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2563 * and msync() need to guarantee that all the data which was dirty at the time
2564 * the call was made get new I/O started against them. If wbc->sync_mode is
2565 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2566 * existing IO to complete.
2567 */
2568 static int extent_write_cache_pages(struct extent_io_tree *tree,
2569 struct address_space *mapping,
2570 struct writeback_control *wbc,
2571 writepage_t writepage, void *data,
2572 void (*flush_fn)(void *))
2573 {
2574 int ret = 0;
2575 int done = 0;
2576 int nr_to_write_done = 0;
2577 struct pagevec pvec;
2578 int nr_pages;
2579 pgoff_t index;
2580 pgoff_t end; /* Inclusive */
2581 int scanned = 0;
2582 int tag;
2583
2584 pagevec_init(&pvec, 0);
2585 if (wbc->range_cyclic) {
2586 index = mapping->writeback_index; /* Start from prev offset */
2587 end = -1;
2588 } else {
2589 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2590 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2591 scanned = 1;
2592 }
2593 if (wbc->sync_mode == WB_SYNC_ALL)
2594 tag = PAGECACHE_TAG_TOWRITE;
2595 else
2596 tag = PAGECACHE_TAG_DIRTY;
2597 retry:
2598 if (wbc->sync_mode == WB_SYNC_ALL)
2599 tag_pages_for_writeback(mapping, index, end);
2600 while (!done && !nr_to_write_done && (index <= end) &&
2601 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2602 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
2603 unsigned i;
2604
2605 scanned = 1;
2606 for (i = 0; i < nr_pages; i++) {
2607 struct page *page = pvec.pages[i];
2608
2609 /*
2610 * At this point we hold neither mapping->tree_lock nor
2611 * lock on the page itself: the page may be truncated or
2612 * invalidated (changing page->mapping to NULL), or even
2613 * swizzled back from swapper_space to tmpfs file
2614 * mapping
2615 */
2616 if (tree->ops &&
2617 tree->ops->write_cache_pages_lock_hook) {
2618 tree->ops->write_cache_pages_lock_hook(page,
2619 data, flush_fn);
2620 } else {
2621 if (!trylock_page(page)) {
2622 flush_fn(data);
2623 lock_page(page);
2624 }
2625 }
2626
2627 if (unlikely(page->mapping != mapping)) {
2628 unlock_page(page);
2629 continue;
2630 }
2631
2632 if (!wbc->range_cyclic && page->index > end) {
2633 done = 1;
2634 unlock_page(page);
2635 continue;
2636 }
2637
2638 if (wbc->sync_mode != WB_SYNC_NONE) {
2639 if (PageWriteback(page))
2640 flush_fn(data);
2641 wait_on_page_writeback(page);
2642 }
2643
2644 if (PageWriteback(page) ||
2645 !clear_page_dirty_for_io(page)) {
2646 unlock_page(page);
2647 continue;
2648 }
2649
2650 ret = (*writepage)(page, wbc, data);
2651
2652 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
2653 unlock_page(page);
2654 ret = 0;
2655 }
2656 if (ret)
2657 done = 1;
2658
2659 /*
2660 * the filesystem may choose to bump up nr_to_write.
2661 * We have to make sure to honor the new nr_to_write
2662 * at any time
2663 */
2664 nr_to_write_done = wbc->nr_to_write <= 0;
2665 }
2666 pagevec_release(&pvec);
2667 cond_resched();
2668 }
2669 if (!scanned && !done) {
2670 /*
2671 * We hit the last page and there is more work to be done: wrap
2672 * back to the start of the file
2673 */
2674 scanned = 1;
2675 index = 0;
2676 goto retry;
2677 }
2678 return ret;
2679 }
2680
2681 static void flush_epd_write_bio(struct extent_page_data *epd)
2682 {
2683 if (epd->bio) {
2684 if (epd->sync_io)
2685 submit_one_bio(WRITE_SYNC, epd->bio, 0, 0);
2686 else
2687 submit_one_bio(WRITE, epd->bio, 0, 0);
2688 epd->bio = NULL;
2689 }
2690 }
2691
2692 static noinline void flush_write_bio(void *data)
2693 {
2694 struct extent_page_data *epd = data;
2695 flush_epd_write_bio(epd);
2696 }
2697
2698 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
2699 get_extent_t *get_extent,
2700 struct writeback_control *wbc)
2701 {
2702 int ret;
2703 struct extent_page_data epd = {
2704 .bio = NULL,
2705 .tree = tree,
2706 .get_extent = get_extent,
2707 .extent_locked = 0,
2708 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2709 };
2710
2711 ret = __extent_writepage(page, wbc, &epd);
2712
2713 flush_epd_write_bio(&epd);
2714 return ret;
2715 }
2716
2717 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
2718 u64 start, u64 end, get_extent_t *get_extent,
2719 int mode)
2720 {
2721 int ret = 0;
2722 struct address_space *mapping = inode->i_mapping;
2723 struct page *page;
2724 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
2725 PAGE_CACHE_SHIFT;
2726
2727 struct extent_page_data epd = {
2728 .bio = NULL,
2729 .tree = tree,
2730 .get_extent = get_extent,
2731 .extent_locked = 1,
2732 .sync_io = mode == WB_SYNC_ALL,
2733 };
2734 struct writeback_control wbc_writepages = {
2735 .sync_mode = mode,
2736 .nr_to_write = nr_pages * 2,
2737 .range_start = start,
2738 .range_end = end + 1,
2739 };
2740
2741 while (start <= end) {
2742 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
2743 if (clear_page_dirty_for_io(page))
2744 ret = __extent_writepage(page, &wbc_writepages, &epd);
2745 else {
2746 if (tree->ops && tree->ops->writepage_end_io_hook)
2747 tree->ops->writepage_end_io_hook(page, start,
2748 start + PAGE_CACHE_SIZE - 1,
2749 NULL, 1);
2750 unlock_page(page);
2751 }
2752 page_cache_release(page);
2753 start += PAGE_CACHE_SIZE;
2754 }
2755
2756 flush_epd_write_bio(&epd);
2757 return ret;
2758 }
2759
2760 int extent_writepages(struct extent_io_tree *tree,
2761 struct address_space *mapping,
2762 get_extent_t *get_extent,
2763 struct writeback_control *wbc)
2764 {
2765 int ret = 0;
2766 struct extent_page_data epd = {
2767 .bio = NULL,
2768 .tree = tree,
2769 .get_extent = get_extent,
2770 .extent_locked = 0,
2771 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
2772 };
2773
2774 ret = extent_write_cache_pages(tree, mapping, wbc,
2775 __extent_writepage, &epd,
2776 flush_write_bio);
2777 flush_epd_write_bio(&epd);
2778 return ret;
2779 }
2780
2781 int extent_readpages(struct extent_io_tree *tree,
2782 struct address_space *mapping,
2783 struct list_head *pages, unsigned nr_pages,
2784 get_extent_t get_extent)
2785 {
2786 struct bio *bio = NULL;
2787 unsigned page_idx;
2788 unsigned long bio_flags = 0;
2789
2790 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
2791 struct page *page = list_entry(pages->prev, struct page, lru);
2792
2793 prefetchw(&page->flags);
2794 list_del(&page->lru);
2795 if (!add_to_page_cache_lru(page, mapping,
2796 page->index, GFP_NOFS)) {
2797 __extent_read_full_page(tree, page, get_extent,
2798 &bio, 0, &bio_flags);
2799 }
2800 page_cache_release(page);
2801 }
2802 BUG_ON(!list_empty(pages));
2803 if (bio)
2804 submit_one_bio(READ, bio, 0, bio_flags);
2805 return 0;
2806 }
2807
2808 /*
2809 * basic invalidatepage code, this waits on any locked or writeback
2810 * ranges corresponding to the page, and then deletes any extent state
2811 * records from the tree
2812 */
2813 int extent_invalidatepage(struct extent_io_tree *tree,
2814 struct page *page, unsigned long offset)
2815 {
2816 struct extent_state *cached_state = NULL;
2817 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT);
2818 u64 end = start + PAGE_CACHE_SIZE - 1;
2819 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
2820
2821 start += (offset + blocksize - 1) & ~(blocksize - 1);
2822 if (start > end)
2823 return 0;
2824
2825 lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS);
2826 wait_on_page_writeback(page);
2827 clear_extent_bit(tree, start, end,
2828 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2829 EXTENT_DO_ACCOUNTING,
2830 1, 1, &cached_state, GFP_NOFS);
2831 return 0;
2832 }
2833
2834 /*
2835 * a helper for releasepage, this tests for areas of the page that
2836 * are locked or under IO and drops the related state bits if it is safe
2837 * to drop the page.
2838 */
2839 int try_release_extent_state(struct extent_map_tree *map,
2840 struct extent_io_tree *tree, struct page *page,
2841 gfp_t mask)
2842 {
2843 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2844 u64 end = start + PAGE_CACHE_SIZE - 1;
2845 int ret = 1;
2846
2847 if (test_range_bit(tree, start, end,
2848 EXTENT_IOBITS, 0, NULL))
2849 ret = 0;
2850 else {
2851 if ((mask & GFP_NOFS) == GFP_NOFS)
2852 mask = GFP_NOFS;
2853 /*
2854 * at this point we can safely clear everything except the
2855 * locked bit and the nodatasum bit
2856 */
2857 ret = clear_extent_bit(tree, start, end,
2858 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
2859 0, 0, NULL, mask);
2860
2861 /* if clear_extent_bit failed for enomem reasons,
2862 * we can't allow the release to continue.
2863 */
2864 if (ret < 0)
2865 ret = 0;
2866 else
2867 ret = 1;
2868 }
2869 return ret;
2870 }
2871
2872 /*
2873 * a helper for releasepage. As long as there are no locked extents
2874 * in the range corresponding to the page, both state records and extent
2875 * map records are removed
2876 */
2877 int try_release_extent_mapping(struct extent_map_tree *map,
2878 struct extent_io_tree *tree, struct page *page,
2879 gfp_t mask)
2880 {
2881 struct extent_map *em;
2882 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
2883 u64 end = start + PAGE_CACHE_SIZE - 1;
2884
2885 if ((mask & __GFP_WAIT) &&
2886 page->mapping->host->i_size > 16 * 1024 * 1024) {
2887 u64 len;
2888 while (start <= end) {
2889 len = end - start + 1;
2890 write_lock(&map->lock);
2891 em = lookup_extent_mapping(map, start, len);
2892 if (IS_ERR_OR_NULL(em)) {
2893 write_unlock(&map->lock);
2894 break;
2895 }
2896 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2897 em->start != start) {
2898 write_unlock(&map->lock);
2899 free_extent_map(em);
2900 break;
2901 }
2902 if (!test_range_bit(tree, em->start,
2903 extent_map_end(em) - 1,
2904 EXTENT_LOCKED | EXTENT_WRITEBACK,
2905 0, NULL)) {
2906 remove_extent_mapping(map, em);
2907 /* once for the rb tree */
2908 free_extent_map(em);
2909 }
2910 start = extent_map_end(em);
2911 write_unlock(&map->lock);
2912
2913 /* once for us */
2914 free_extent_map(em);
2915 }
2916 }
2917 return try_release_extent_state(map, tree, page, mask);
2918 }
2919
2920 /*
2921 * helper function for fiemap, which doesn't want to see any holes.
2922 * This maps until we find something past 'last'
2923 */
2924 static struct extent_map *get_extent_skip_holes(struct inode *inode,
2925 u64 offset,
2926 u64 last,
2927 get_extent_t *get_extent)
2928 {
2929 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
2930 struct extent_map *em;
2931 u64 len;
2932
2933 if (offset >= last)
2934 return NULL;
2935
2936 while(1) {
2937 len = last - offset;
2938 if (len == 0)
2939 break;
2940 len = (len + sectorsize - 1) & ~(sectorsize - 1);
2941 em = get_extent(inode, NULL, 0, offset, len, 0);
2942 if (IS_ERR_OR_NULL(em))
2943 return em;
2944
2945 /* if this isn't a hole return it */
2946 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
2947 em->block_start != EXTENT_MAP_HOLE) {
2948 return em;
2949 }
2950
2951 /* this is a hole, advance to the next extent */
2952 offset = extent_map_end(em);
2953 free_extent_map(em);
2954 if (offset >= last)
2955 break;
2956 }
2957 return NULL;
2958 }
2959
2960 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2961 __u64 start, __u64 len, get_extent_t *get_extent)
2962 {
2963 int ret = 0;
2964 u64 off = start;
2965 u64 max = start + len;
2966 u32 flags = 0;
2967 u32 found_type;
2968 u64 last;
2969 u64 last_for_get_extent = 0;
2970 u64 disko = 0;
2971 u64 isize = i_size_read(inode);
2972 struct btrfs_key found_key;
2973 struct extent_map *em = NULL;
2974 struct extent_state *cached_state = NULL;
2975 struct btrfs_path *path;
2976 struct btrfs_file_extent_item *item;
2977 int end = 0;
2978 u64 em_start = 0;
2979 u64 em_len = 0;
2980 u64 em_end = 0;
2981 unsigned long emflags;
2982
2983 if (len == 0)
2984 return -EINVAL;
2985
2986 path = btrfs_alloc_path();
2987 if (!path)
2988 return -ENOMEM;
2989 path->leave_spinning = 1;
2990
2991 /*
2992 * lookup the last file extent. We're not using i_size here
2993 * because there might be preallocation past i_size
2994 */
2995 ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
2996 path, btrfs_ino(inode), -1, 0);
2997 if (ret < 0) {
2998 btrfs_free_path(path);
2999 return ret;
3000 }
3001 WARN_ON(!ret);
3002 path->slots[0]--;
3003 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3004 struct btrfs_file_extent_item);
3005 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3006 found_type = btrfs_key_type(&found_key);
3007
3008 /* No extents, but there might be delalloc bits */
3009 if (found_key.objectid != btrfs_ino(inode) ||
3010 found_type != BTRFS_EXTENT_DATA_KEY) {
3011 /* have to trust i_size as the end */
3012 last = (u64)-1;
3013 last_for_get_extent = isize;
3014 } else {
3015 /*
3016 * remember the start of the last extent. There are a
3017 * bunch of different factors that go into the length of the
3018 * extent, so its much less complex to remember where it started
3019 */
3020 last = found_key.offset;
3021 last_for_get_extent = last + 1;
3022 }
3023 btrfs_free_path(path);
3024
3025 /*
3026 * we might have some extents allocated but more delalloc past those
3027 * extents. so, we trust isize unless the start of the last extent is
3028 * beyond isize
3029 */
3030 if (last < isize) {
3031 last = (u64)-1;
3032 last_for_get_extent = isize;
3033 }
3034
3035 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0,
3036 &cached_state, GFP_NOFS);
3037
3038 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3039 get_extent);
3040 if (!em)
3041 goto out;
3042 if (IS_ERR(em)) {
3043 ret = PTR_ERR(em);
3044 goto out;
3045 }
3046
3047 while (!end) {
3048 u64 offset_in_extent;
3049
3050 /* break if the extent we found is outside the range */
3051 if (em->start >= max || extent_map_end(em) < off)
3052 break;
3053
3054 /*
3055 * get_extent may return an extent that starts before our
3056 * requested range. We have to make sure the ranges
3057 * we return to fiemap always move forward and don't
3058 * overlap, so adjust the offsets here
3059 */
3060 em_start = max(em->start, off);
3061
3062 /*
3063 * record the offset from the start of the extent
3064 * for adjusting the disk offset below
3065 */
3066 offset_in_extent = em_start - em->start;
3067 em_end = extent_map_end(em);
3068 em_len = em_end - em_start;
3069 emflags = em->flags;
3070 disko = 0;
3071 flags = 0;
3072
3073 /*
3074 * bump off for our next call to get_extent
3075 */
3076 off = extent_map_end(em);
3077 if (off >= max)
3078 end = 1;
3079
3080 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
3081 end = 1;
3082 flags |= FIEMAP_EXTENT_LAST;
3083 } else if (em->block_start == EXTENT_MAP_INLINE) {
3084 flags |= (FIEMAP_EXTENT_DATA_INLINE |
3085 FIEMAP_EXTENT_NOT_ALIGNED);
3086 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
3087 flags |= (FIEMAP_EXTENT_DELALLOC |
3088 FIEMAP_EXTENT_UNKNOWN);
3089 } else {
3090 disko = em->block_start + offset_in_extent;
3091 }
3092 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
3093 flags |= FIEMAP_EXTENT_ENCODED;
3094
3095 free_extent_map(em);
3096 em = NULL;
3097 if ((em_start >= last) || em_len == (u64)-1 ||
3098 (last == (u64)-1 && isize <= em_end)) {
3099 flags |= FIEMAP_EXTENT_LAST;
3100 end = 1;
3101 }
3102
3103 /* now scan forward to see if this is really the last extent. */
3104 em = get_extent_skip_holes(inode, off, last_for_get_extent,
3105 get_extent);
3106 if (IS_ERR(em)) {
3107 ret = PTR_ERR(em);
3108 goto out;
3109 }
3110 if (!em) {
3111 flags |= FIEMAP_EXTENT_LAST;
3112 end = 1;
3113 }
3114 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
3115 em_len, flags);
3116 if (ret)
3117 goto out_free;
3118 }
3119 out_free:
3120 free_extent_map(em);
3121 out:
3122 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len,
3123 &cached_state, GFP_NOFS);
3124 return ret;
3125 }
3126
3127 static inline struct page *extent_buffer_page(struct extent_buffer *eb,
3128 unsigned long i)
3129 {
3130 struct page *p;
3131 struct address_space *mapping;
3132
3133 if (i == 0)
3134 return eb->first_page;
3135 i += eb->start >> PAGE_CACHE_SHIFT;
3136 mapping = eb->first_page->mapping;
3137 if (!mapping)
3138 return NULL;
3139
3140 /*
3141 * extent_buffer_page is only called after pinning the page
3142 * by increasing the reference count. So we know the page must
3143 * be in the radix tree.
3144 */
3145 rcu_read_lock();
3146 p = radix_tree_lookup(&mapping->page_tree, i);
3147 rcu_read_unlock();
3148
3149 return p;
3150 }
3151
3152 static inline unsigned long num_extent_pages(u64 start, u64 len)
3153 {
3154 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
3155 (start >> PAGE_CACHE_SHIFT);
3156 }
3157
3158 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
3159 u64 start,
3160 unsigned long len,
3161 gfp_t mask)
3162 {
3163 struct extent_buffer *eb = NULL;
3164 #if LEAK_DEBUG
3165 unsigned long flags;
3166 #endif
3167
3168 eb = kmem_cache_zalloc(extent_buffer_cache, mask);
3169 if (eb == NULL)
3170 return NULL;
3171 eb->start = start;
3172 eb->len = len;
3173 rwlock_init(&eb->lock);
3174 atomic_set(&eb->write_locks, 0);
3175 atomic_set(&eb->read_locks, 0);
3176 atomic_set(&eb->blocking_readers, 0);
3177 atomic_set(&eb->blocking_writers, 0);
3178 atomic_set(&eb->spinning_readers, 0);
3179 atomic_set(&eb->spinning_writers, 0);
3180 init_waitqueue_head(&eb->write_lock_wq);
3181 init_waitqueue_head(&eb->read_lock_wq);
3182
3183 #if LEAK_DEBUG
3184 spin_lock_irqsave(&leak_lock, flags);
3185 list_add(&eb->leak_list, &buffers);
3186 spin_unlock_irqrestore(&leak_lock, flags);
3187 #endif
3188 atomic_set(&eb->refs, 1);
3189
3190 return eb;
3191 }
3192
3193 static void __free_extent_buffer(struct extent_buffer *eb)
3194 {
3195 #if LEAK_DEBUG
3196 unsigned long flags;
3197 spin_lock_irqsave(&leak_lock, flags);
3198 list_del(&eb->leak_list);
3199 spin_unlock_irqrestore(&leak_lock, flags);
3200 #endif
3201 kmem_cache_free(extent_buffer_cache, eb);
3202 }
3203
3204 /*
3205 * Helper for releasing extent buffer page.
3206 */
3207 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
3208 unsigned long start_idx)
3209 {
3210 unsigned long index;
3211 struct page *page;
3212
3213 if (!eb->first_page)
3214 return;
3215
3216 index = num_extent_pages(eb->start, eb->len);
3217 if (start_idx >= index)
3218 return;
3219
3220 do {
3221 index--;
3222 page = extent_buffer_page(eb, index);
3223 if (page)
3224 page_cache_release(page);
3225 } while (index != start_idx);
3226 }
3227
3228 /*
3229 * Helper for releasing the extent buffer.
3230 */
3231 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3232 {
3233 btrfs_release_extent_buffer_page(eb, 0);
3234 __free_extent_buffer(eb);
3235 }
3236
3237 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
3238 u64 start, unsigned long len,
3239 struct page *page0)
3240 {
3241 unsigned long num_pages = num_extent_pages(start, len);
3242 unsigned long i;
3243 unsigned long index = start >> PAGE_CACHE_SHIFT;
3244 struct extent_buffer *eb;
3245 struct extent_buffer *exists = NULL;
3246 struct page *p;
3247 struct address_space *mapping = tree->mapping;
3248 int uptodate = 1;
3249 int ret;
3250
3251 rcu_read_lock();
3252 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3253 if (eb && atomic_inc_not_zero(&eb->refs)) {
3254 rcu_read_unlock();
3255 mark_page_accessed(eb->first_page);
3256 return eb;
3257 }
3258 rcu_read_unlock();
3259
3260 eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
3261 if (!eb)
3262 return NULL;
3263
3264 if (page0) {
3265 eb->first_page = page0;
3266 i = 1;
3267 index++;
3268 page_cache_get(page0);
3269 mark_page_accessed(page0);
3270 set_page_extent_mapped(page0);
3271 set_page_extent_head(page0, len);
3272 uptodate = PageUptodate(page0);
3273 } else {
3274 i = 0;
3275 }
3276 for (; i < num_pages; i++, index++) {
3277 p = find_or_create_page(mapping, index, GFP_NOFS);
3278 if (!p) {
3279 WARN_ON(1);
3280 goto free_eb;
3281 }
3282 set_page_extent_mapped(p);
3283 mark_page_accessed(p);
3284 if (i == 0) {
3285 eb->first_page = p;
3286 set_page_extent_head(p, len);
3287 } else {
3288 set_page_private(p, EXTENT_PAGE_PRIVATE);
3289 }
3290 if (!PageUptodate(p))
3291 uptodate = 0;
3292
3293 /*
3294 * see below about how we avoid a nasty race with release page
3295 * and why we unlock later
3296 */
3297 if (i != 0)
3298 unlock_page(p);
3299 }
3300 if (uptodate)
3301 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3302
3303 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
3304 if (ret)
3305 goto free_eb;
3306
3307 spin_lock(&tree->buffer_lock);
3308 ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
3309 if (ret == -EEXIST) {
3310 exists = radix_tree_lookup(&tree->buffer,
3311 start >> PAGE_CACHE_SHIFT);
3312 /* add one reference for the caller */
3313 atomic_inc(&exists->refs);
3314 spin_unlock(&tree->buffer_lock);
3315 radix_tree_preload_end();
3316 goto free_eb;
3317 }
3318 /* add one reference for the tree */
3319 atomic_inc(&eb->refs);
3320 spin_unlock(&tree->buffer_lock);
3321 radix_tree_preload_end();
3322
3323 /*
3324 * there is a race where release page may have
3325 * tried to find this extent buffer in the radix
3326 * but failed. It will tell the VM it is safe to
3327 * reclaim the, and it will clear the page private bit.
3328 * We must make sure to set the page private bit properly
3329 * after the extent buffer is in the radix tree so
3330 * it doesn't get lost
3331 */
3332 set_page_extent_mapped(eb->first_page);
3333 set_page_extent_head(eb->first_page, eb->len);
3334 if (!page0)
3335 unlock_page(eb->first_page);
3336 return eb;
3337
3338 free_eb:
3339 if (eb->first_page && !page0)
3340 unlock_page(eb->first_page);
3341
3342 if (!atomic_dec_and_test(&eb->refs))
3343 return exists;
3344 btrfs_release_extent_buffer(eb);
3345 return exists;
3346 }
3347
3348 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
3349 u64 start, unsigned long len)
3350 {
3351 struct extent_buffer *eb;
3352
3353 rcu_read_lock();
3354 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
3355 if (eb && atomic_inc_not_zero(&eb->refs)) {
3356 rcu_read_unlock();
3357 mark_page_accessed(eb->first_page);
3358 return eb;
3359 }
3360 rcu_read_unlock();
3361
3362 return NULL;
3363 }
3364
3365 void free_extent_buffer(struct extent_buffer *eb)
3366 {
3367 if (!eb)
3368 return;
3369
3370 if (!atomic_dec_and_test(&eb->refs))
3371 return;
3372
3373 WARN_ON(1);
3374 }
3375
3376 int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3377 struct extent_buffer *eb)
3378 {
3379 unsigned long i;
3380 unsigned long num_pages;
3381 struct page *page;
3382
3383 num_pages = num_extent_pages(eb->start, eb->len);
3384
3385 for (i = 0; i < num_pages; i++) {
3386 page = extent_buffer_page(eb, i);
3387 if (!PageDirty(page))
3388 continue;
3389
3390 lock_page(page);
3391 WARN_ON(!PagePrivate(page));
3392
3393 set_page_extent_mapped(page);
3394 if (i == 0)
3395 set_page_extent_head(page, eb->len);
3396
3397 clear_page_dirty_for_io(page);
3398 spin_lock_irq(&page->mapping->tree_lock);
3399 if (!PageDirty(page)) {
3400 radix_tree_tag_clear(&page->mapping->page_tree,
3401 page_index(page),
3402 PAGECACHE_TAG_DIRTY);
3403 }
3404 spin_unlock_irq(&page->mapping->tree_lock);
3405 unlock_page(page);
3406 }
3407 return 0;
3408 }
3409
3410 int set_extent_buffer_dirty(struct extent_io_tree *tree,
3411 struct extent_buffer *eb)
3412 {
3413 unsigned long i;
3414 unsigned long num_pages;
3415 int was_dirty = 0;
3416
3417 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3418 num_pages = num_extent_pages(eb->start, eb->len);
3419 for (i = 0; i < num_pages; i++)
3420 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3421 return was_dirty;
3422 }
3423
3424 static int __eb_straddles_pages(u64 start, u64 len)
3425 {
3426 if (len < PAGE_CACHE_SIZE)
3427 return 1;
3428 if (start & (PAGE_CACHE_SIZE - 1))
3429 return 1;
3430 if ((start + len) & (PAGE_CACHE_SIZE - 1))
3431 return 1;
3432 return 0;
3433 }
3434
3435 static int eb_straddles_pages(struct extent_buffer *eb)
3436 {
3437 return __eb_straddles_pages(eb->start, eb->len);
3438 }
3439
3440 int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
3441 struct extent_buffer *eb,
3442 struct extent_state **cached_state)
3443 {
3444 unsigned long i;
3445 struct page *page;
3446 unsigned long num_pages;
3447
3448 num_pages = num_extent_pages(eb->start, eb->len);
3449 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3450
3451 if (eb_straddles_pages(eb)) {
3452 clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3453 cached_state, GFP_NOFS);
3454 }
3455 for (i = 0; i < num_pages; i++) {
3456 page = extent_buffer_page(eb, i);
3457 if (page)
3458 ClearPageUptodate(page);
3459 }
3460 return 0;
3461 }
3462
3463 int set_extent_buffer_uptodate(struct extent_io_tree *tree,
3464 struct extent_buffer *eb)
3465 {
3466 unsigned long i;
3467 struct page *page;
3468 unsigned long num_pages;
3469
3470 num_pages = num_extent_pages(eb->start, eb->len);
3471
3472 if (eb_straddles_pages(eb)) {
3473 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1,
3474 NULL, GFP_NOFS);
3475 }
3476 for (i = 0; i < num_pages; i++) {
3477 page = extent_buffer_page(eb, i);
3478 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3479 ((i == num_pages - 1) &&
3480 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3481 check_page_uptodate(tree, page);
3482 continue;
3483 }
3484 SetPageUptodate(page);
3485 }
3486 return 0;
3487 }
3488
3489 int extent_range_uptodate(struct extent_io_tree *tree,
3490 u64 start, u64 end)
3491 {
3492 struct page *page;
3493 int ret;
3494 int pg_uptodate = 1;
3495 int uptodate;
3496 unsigned long index;
3497
3498 if (__eb_straddles_pages(start, end - start + 1)) {
3499 ret = test_range_bit(tree, start, end,
3500 EXTENT_UPTODATE, 1, NULL);
3501 if (ret)
3502 return 1;
3503 }
3504 while (start <= end) {
3505 index = start >> PAGE_CACHE_SHIFT;
3506 page = find_get_page(tree->mapping, index);
3507 uptodate = PageUptodate(page);
3508 page_cache_release(page);
3509 if (!uptodate) {
3510 pg_uptodate = 0;
3511 break;
3512 }
3513 start += PAGE_CACHE_SIZE;
3514 }
3515 return pg_uptodate;
3516 }
3517
3518 int extent_buffer_uptodate(struct extent_io_tree *tree,
3519 struct extent_buffer *eb,
3520 struct extent_state *cached_state)
3521 {
3522 int ret = 0;
3523 unsigned long num_pages;
3524 unsigned long i;
3525 struct page *page;
3526 int pg_uptodate = 1;
3527
3528 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3529 return 1;
3530
3531 if (eb_straddles_pages(eb)) {
3532 ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3533 EXTENT_UPTODATE, 1, cached_state);
3534 if (ret)
3535 return ret;
3536 }
3537
3538 num_pages = num_extent_pages(eb->start, eb->len);
3539 for (i = 0; i < num_pages; i++) {
3540 page = extent_buffer_page(eb, i);
3541 if (!PageUptodate(page)) {
3542 pg_uptodate = 0;
3543 break;
3544 }
3545 }
3546 return pg_uptodate;
3547 }
3548
3549 int read_extent_buffer_pages(struct extent_io_tree *tree,
3550 struct extent_buffer *eb,
3551 u64 start, int wait,
3552 get_extent_t *get_extent, int mirror_num)
3553 {
3554 unsigned long i;
3555 unsigned long start_i;
3556 struct page *page;
3557 int err;
3558 int ret = 0;
3559 int locked_pages = 0;
3560 int all_uptodate = 1;
3561 int inc_all_pages = 0;
3562 unsigned long num_pages;
3563 struct bio *bio = NULL;
3564 unsigned long bio_flags = 0;
3565
3566 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3567 return 0;
3568
3569 if (eb_straddles_pages(eb)) {
3570 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
3571 EXTENT_UPTODATE, 1, NULL)) {
3572 return 0;
3573 }
3574 }
3575
3576 if (start) {
3577 WARN_ON(start < eb->start);
3578 start_i = (start >> PAGE_CACHE_SHIFT) -
3579 (eb->start >> PAGE_CACHE_SHIFT);
3580 } else {
3581 start_i = 0;
3582 }
3583
3584 num_pages = num_extent_pages(eb->start, eb->len);
3585 for (i = start_i; i < num_pages; i++) {
3586 page = extent_buffer_page(eb, i);
3587 if (!wait) {
3588 if (!trylock_page(page))
3589 goto unlock_exit;
3590 } else {
3591 lock_page(page);
3592 }
3593 locked_pages++;
3594 if (!PageUptodate(page))
3595 all_uptodate = 0;
3596 }
3597 if (all_uptodate) {
3598 if (start_i == 0)
3599 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3600 goto unlock_exit;
3601 }
3602
3603 for (i = start_i; i < num_pages; i++) {
3604 page = extent_buffer_page(eb, i);
3605
3606 WARN_ON(!PagePrivate(page));
3607
3608 set_page_extent_mapped(page);
3609 if (i == 0)
3610 set_page_extent_head(page, eb->len);
3611
3612 if (inc_all_pages)
3613 page_cache_get(page);
3614 if (!PageUptodate(page)) {
3615 if (start_i == 0)
3616 inc_all_pages = 1;
3617 ClearPageError(page);
3618 err = __extent_read_full_page(tree, page,
3619 get_extent, &bio,
3620 mirror_num, &bio_flags);
3621 if (err)
3622 ret = err;
3623 } else {
3624 unlock_page(page);
3625 }
3626 }
3627
3628 if (bio)
3629 submit_one_bio(READ, bio, mirror_num, bio_flags);
3630
3631 if (ret || !wait)
3632 return ret;
3633
3634 for (i = start_i; i < num_pages; i++) {
3635 page = extent_buffer_page(eb, i);
3636 wait_on_page_locked(page);
3637 if (!PageUptodate(page))
3638 ret = -EIO;
3639 }
3640
3641 if (!ret)
3642 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3643 return ret;
3644
3645 unlock_exit:
3646 i = start_i;
3647 while (locked_pages > 0) {
3648 page = extent_buffer_page(eb, i);
3649 i++;
3650 unlock_page(page);
3651 locked_pages--;
3652 }
3653 return ret;
3654 }
3655
3656 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3657 unsigned long start,
3658 unsigned long len)
3659 {
3660 size_t cur;
3661 size_t offset;
3662 struct page *page;
3663 char *kaddr;
3664 char *dst = (char *)dstv;
3665 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3666 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3667
3668 WARN_ON(start > eb->len);
3669 WARN_ON(start + len > eb->start + eb->len);
3670
3671 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3672
3673 while (len > 0) {
3674 page = extent_buffer_page(eb, i);
3675
3676 cur = min(len, (PAGE_CACHE_SIZE - offset));
3677 kaddr = page_address(page);
3678 memcpy(dst, kaddr + offset, cur);
3679
3680 dst += cur;
3681 len -= cur;
3682 offset = 0;
3683 i++;
3684 }
3685 }
3686
3687 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
3688 unsigned long min_len, char **map,
3689 unsigned long *map_start,
3690 unsigned long *map_len)
3691 {
3692 size_t offset = start & (PAGE_CACHE_SIZE - 1);
3693 char *kaddr;
3694 struct page *p;
3695 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3696 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3697 unsigned long end_i = (start_offset + start + min_len - 1) >>
3698 PAGE_CACHE_SHIFT;
3699
3700 if (i != end_i)
3701 return -EINVAL;
3702
3703 if (i == 0) {
3704 offset = start_offset;
3705 *map_start = 0;
3706 } else {
3707 offset = 0;
3708 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
3709 }
3710
3711 if (start + min_len > eb->len) {
3712 printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
3713 "wanted %lu %lu\n", (unsigned long long)eb->start,
3714 eb->len, start, min_len);
3715 WARN_ON(1);
3716 return -EINVAL;
3717 }
3718
3719 p = extent_buffer_page(eb, i);
3720 kaddr = page_address(p);
3721 *map = kaddr + offset;
3722 *map_len = PAGE_CACHE_SIZE - offset;
3723 return 0;
3724 }
3725
3726 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3727 unsigned long start,
3728 unsigned long len)
3729 {
3730 size_t cur;
3731 size_t offset;
3732 struct page *page;
3733 char *kaddr;
3734 char *ptr = (char *)ptrv;
3735 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3736 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3737 int ret = 0;
3738
3739 WARN_ON(start > eb->len);
3740 WARN_ON(start + len > eb->start + eb->len);
3741
3742 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3743
3744 while (len > 0) {
3745 page = extent_buffer_page(eb, i);
3746
3747 cur = min(len, (PAGE_CACHE_SIZE - offset));
3748
3749 kaddr = page_address(page);
3750 ret = memcmp(ptr, kaddr + offset, cur);
3751 if (ret)
3752 break;
3753
3754 ptr += cur;
3755 len -= cur;
3756 offset = 0;
3757 i++;
3758 }
3759 return ret;
3760 }
3761
3762 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
3763 unsigned long start, unsigned long len)
3764 {
3765 size_t cur;
3766 size_t offset;
3767 struct page *page;
3768 char *kaddr;
3769 char *src = (char *)srcv;
3770 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3771 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3772
3773 WARN_ON(start > eb->len);
3774 WARN_ON(start + len > eb->start + eb->len);
3775
3776 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3777
3778 while (len > 0) {
3779 page = extent_buffer_page(eb, i);
3780 WARN_ON(!PageUptodate(page));
3781
3782 cur = min(len, PAGE_CACHE_SIZE - offset);
3783 kaddr = page_address(page);
3784 memcpy(kaddr + offset, src, cur);
3785
3786 src += cur;
3787 len -= cur;
3788 offset = 0;
3789 i++;
3790 }
3791 }
3792
3793 void memset_extent_buffer(struct extent_buffer *eb, char c,
3794 unsigned long start, unsigned long len)
3795 {
3796 size_t cur;
3797 size_t offset;
3798 struct page *page;
3799 char *kaddr;
3800 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3801 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3802
3803 WARN_ON(start > eb->len);
3804 WARN_ON(start + len > eb->start + eb->len);
3805
3806 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
3807
3808 while (len > 0) {
3809 page = extent_buffer_page(eb, i);
3810 WARN_ON(!PageUptodate(page));
3811
3812 cur = min(len, PAGE_CACHE_SIZE - offset);
3813 kaddr = page_address(page);
3814 memset(kaddr + offset, c, cur);
3815
3816 len -= cur;
3817 offset = 0;
3818 i++;
3819 }
3820 }
3821
3822 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
3823 unsigned long dst_offset, unsigned long src_offset,
3824 unsigned long len)
3825 {
3826 u64 dst_len = dst->len;
3827 size_t cur;
3828 size_t offset;
3829 struct page *page;
3830 char *kaddr;
3831 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3832 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3833
3834 WARN_ON(src->len != dst_len);
3835
3836 offset = (start_offset + dst_offset) &
3837 ((unsigned long)PAGE_CACHE_SIZE - 1);
3838
3839 while (len > 0) {
3840 page = extent_buffer_page(dst, i);
3841 WARN_ON(!PageUptodate(page));
3842
3843 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
3844
3845 kaddr = page_address(page);
3846 read_extent_buffer(src, kaddr + offset, src_offset, cur);
3847
3848 src_offset += cur;
3849 len -= cur;
3850 offset = 0;
3851 i++;
3852 }
3853 }
3854
3855 static void move_pages(struct page *dst_page, struct page *src_page,
3856 unsigned long dst_off, unsigned long src_off,
3857 unsigned long len)
3858 {
3859 char *dst_kaddr = page_address(dst_page);
3860 if (dst_page == src_page) {
3861 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
3862 } else {
3863 char *src_kaddr = page_address(src_page);
3864 char *p = dst_kaddr + dst_off + len;
3865 char *s = src_kaddr + src_off + len;
3866
3867 while (len--)
3868 *--p = *--s;
3869 }
3870 }
3871
3872 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
3873 {
3874 unsigned long distance = (src > dst) ? src - dst : dst - src;
3875 return distance < len;
3876 }
3877
3878 static void copy_pages(struct page *dst_page, struct page *src_page,
3879 unsigned long dst_off, unsigned long src_off,
3880 unsigned long len)
3881 {
3882 char *dst_kaddr = page_address(dst_page);
3883 char *src_kaddr;
3884
3885 if (dst_page != src_page) {
3886 src_kaddr = page_address(src_page);
3887 } else {
3888 src_kaddr = dst_kaddr;
3889 BUG_ON(areas_overlap(src_off, dst_off, len));
3890 }
3891
3892 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
3893 }
3894
3895 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3896 unsigned long src_offset, unsigned long len)
3897 {
3898 size_t cur;
3899 size_t dst_off_in_page;
3900 size_t src_off_in_page;
3901 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3902 unsigned long dst_i;
3903 unsigned long src_i;
3904
3905 if (src_offset + len > dst->len) {
3906 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3907 "len %lu dst len %lu\n", src_offset, len, dst->len);
3908 BUG_ON(1);
3909 }
3910 if (dst_offset + len > dst->len) {
3911 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3912 "len %lu dst len %lu\n", dst_offset, len, dst->len);
3913 BUG_ON(1);
3914 }
3915
3916 while (len > 0) {
3917 dst_off_in_page = (start_offset + dst_offset) &
3918 ((unsigned long)PAGE_CACHE_SIZE - 1);
3919 src_off_in_page = (start_offset + src_offset) &
3920 ((unsigned long)PAGE_CACHE_SIZE - 1);
3921
3922 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
3923 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
3924
3925 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
3926 src_off_in_page));
3927 cur = min_t(unsigned long, cur,
3928 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
3929
3930 copy_pages(extent_buffer_page(dst, dst_i),
3931 extent_buffer_page(dst, src_i),
3932 dst_off_in_page, src_off_in_page, cur);
3933
3934 src_offset += cur;
3935 dst_offset += cur;
3936 len -= cur;
3937 }
3938 }
3939
3940 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
3941 unsigned long src_offset, unsigned long len)
3942 {
3943 size_t cur;
3944 size_t dst_off_in_page;
3945 size_t src_off_in_page;
3946 unsigned long dst_end = dst_offset + len - 1;
3947 unsigned long src_end = src_offset + len - 1;
3948 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
3949 unsigned long dst_i;
3950 unsigned long src_i;
3951
3952 if (src_offset + len > dst->len) {
3953 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
3954 "len %lu len %lu\n", src_offset, len, dst->len);
3955 BUG_ON(1);
3956 }
3957 if (dst_offset + len > dst->len) {
3958 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
3959 "len %lu len %lu\n", dst_offset, len, dst->len);
3960 BUG_ON(1);
3961 }
3962 if (!areas_overlap(src_offset, dst_offset, len)) {
3963 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
3964 return;
3965 }
3966 while (len > 0) {
3967 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
3968 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
3969
3970 dst_off_in_page = (start_offset + dst_end) &
3971 ((unsigned long)PAGE_CACHE_SIZE - 1);
3972 src_off_in_page = (start_offset + src_end) &
3973 ((unsigned long)PAGE_CACHE_SIZE - 1);
3974
3975 cur = min_t(unsigned long, len, src_off_in_page + 1);
3976 cur = min(cur, dst_off_in_page + 1);
3977 move_pages(extent_buffer_page(dst, dst_i),
3978 extent_buffer_page(dst, src_i),
3979 dst_off_in_page - cur + 1,
3980 src_off_in_page - cur + 1, cur);
3981
3982 dst_end -= cur;
3983 src_end -= cur;
3984 len -= cur;
3985 }
3986 }
3987
3988 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3989 {
3990 struct extent_buffer *eb =
3991 container_of(head, struct extent_buffer, rcu_head);
3992
3993 btrfs_release_extent_buffer(eb);
3994 }
3995
3996 int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3997 {
3998 u64 start = page_offset(page);
3999 struct extent_buffer *eb;
4000 int ret = 1;
4001
4002 spin_lock(&tree->buffer_lock);
4003 eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4004 if (!eb) {
4005 spin_unlock(&tree->buffer_lock);
4006 return ret;
4007 }
4008
4009 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4010 ret = 0;
4011 goto out;
4012 }
4013
4014 /*
4015 * set @eb->refs to 0 if it is already 1, and then release the @eb.
4016 * Or go back.
4017 */
4018 if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) {
4019 ret = 0;
4020 goto out;
4021 }
4022
4023 radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4024 out:
4025 spin_unlock(&tree->buffer_lock);
4026
4027 /* at this point we can safely release the extent buffer */
4028 if (atomic_read(&eb->refs) == 0)
4029 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4030 return ret;
4031 }
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree