Merge remote-tracking branch 'media_tree/vsp1' into generic-zpos-v8
[deliverable/linux.git] / fs / btrfs / ordered-data.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/slab.h>
20 #include <linux/blkdev.h>
21 #include <linux/writeback.h>
22 #include <linux/pagevec.h>
23 #include "ctree.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "extent_io.h"
27 #include "disk-io.h"
28 #include "compression.h"
29
30 static struct kmem_cache *btrfs_ordered_extent_cache;
31
32 static u64 entry_end(struct btrfs_ordered_extent *entry)
33 {
34 if (entry->file_offset + entry->len < entry->file_offset)
35 return (u64)-1;
36 return entry->file_offset + entry->len;
37 }
38
39 /* returns NULL if the insertion worked, or it returns the node it did find
40 * in the tree
41 */
42 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
43 struct rb_node *node)
44 {
45 struct rb_node **p = &root->rb_node;
46 struct rb_node *parent = NULL;
47 struct btrfs_ordered_extent *entry;
48
49 while (*p) {
50 parent = *p;
51 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
52
53 if (file_offset < entry->file_offset)
54 p = &(*p)->rb_left;
55 else if (file_offset >= entry_end(entry))
56 p = &(*p)->rb_right;
57 else
58 return parent;
59 }
60
61 rb_link_node(node, parent, p);
62 rb_insert_color(node, root);
63 return NULL;
64 }
65
66 static void ordered_data_tree_panic(struct inode *inode, int errno,
67 u64 offset)
68 {
69 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
70 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
71 "%llu", offset);
72 }
73
74 /*
75 * look for a given offset in the tree, and if it can't be found return the
76 * first lesser offset
77 */
78 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
79 struct rb_node **prev_ret)
80 {
81 struct rb_node *n = root->rb_node;
82 struct rb_node *prev = NULL;
83 struct rb_node *test;
84 struct btrfs_ordered_extent *entry;
85 struct btrfs_ordered_extent *prev_entry = NULL;
86
87 while (n) {
88 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
89 prev = n;
90 prev_entry = entry;
91
92 if (file_offset < entry->file_offset)
93 n = n->rb_left;
94 else if (file_offset >= entry_end(entry))
95 n = n->rb_right;
96 else
97 return n;
98 }
99 if (!prev_ret)
100 return NULL;
101
102 while (prev && file_offset >= entry_end(prev_entry)) {
103 test = rb_next(prev);
104 if (!test)
105 break;
106 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
107 rb_node);
108 if (file_offset < entry_end(prev_entry))
109 break;
110
111 prev = test;
112 }
113 if (prev)
114 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
115 rb_node);
116 while (prev && file_offset < entry_end(prev_entry)) {
117 test = rb_prev(prev);
118 if (!test)
119 break;
120 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
121 rb_node);
122 prev = test;
123 }
124 *prev_ret = prev;
125 return NULL;
126 }
127
128 /*
129 * helper to check if a given offset is inside a given entry
130 */
131 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
132 {
133 if (file_offset < entry->file_offset ||
134 entry->file_offset + entry->len <= file_offset)
135 return 0;
136 return 1;
137 }
138
139 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
140 u64 len)
141 {
142 if (file_offset + len <= entry->file_offset ||
143 entry->file_offset + entry->len <= file_offset)
144 return 0;
145 return 1;
146 }
147
148 /*
149 * look find the first ordered struct that has this offset, otherwise
150 * the first one less than this offset
151 */
152 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
153 u64 file_offset)
154 {
155 struct rb_root *root = &tree->tree;
156 struct rb_node *prev = NULL;
157 struct rb_node *ret;
158 struct btrfs_ordered_extent *entry;
159
160 if (tree->last) {
161 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
162 rb_node);
163 if (offset_in_entry(entry, file_offset))
164 return tree->last;
165 }
166 ret = __tree_search(root, file_offset, &prev);
167 if (!ret)
168 ret = prev;
169 if (ret)
170 tree->last = ret;
171 return ret;
172 }
173
174 /* allocate and add a new ordered_extent into the per-inode tree.
175 * file_offset is the logical offset in the file
176 *
177 * start is the disk block number of an extent already reserved in the
178 * extent allocation tree
179 *
180 * len is the length of the extent
181 *
182 * The tree is given a single reference on the ordered extent that was
183 * inserted.
184 */
185 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
186 u64 start, u64 len, u64 disk_len,
187 int type, int dio, int compress_type)
188 {
189 struct btrfs_root *root = BTRFS_I(inode)->root;
190 struct btrfs_ordered_inode_tree *tree;
191 struct rb_node *node;
192 struct btrfs_ordered_extent *entry;
193
194 tree = &BTRFS_I(inode)->ordered_tree;
195 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
196 if (!entry)
197 return -ENOMEM;
198
199 entry->file_offset = file_offset;
200 entry->start = start;
201 entry->len = len;
202 entry->disk_len = disk_len;
203 entry->bytes_left = len;
204 entry->inode = igrab(inode);
205 entry->compress_type = compress_type;
206 entry->truncated_len = (u64)-1;
207 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
208 set_bit(type, &entry->flags);
209
210 if (dio)
211 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
212
213 /* one ref for the tree */
214 atomic_set(&entry->refs, 1);
215 init_waitqueue_head(&entry->wait);
216 INIT_LIST_HEAD(&entry->list);
217 INIT_LIST_HEAD(&entry->root_extent_list);
218 INIT_LIST_HEAD(&entry->work_list);
219 init_completion(&entry->completion);
220 INIT_LIST_HEAD(&entry->log_list);
221 INIT_LIST_HEAD(&entry->trans_list);
222
223 trace_btrfs_ordered_extent_add(inode, entry);
224
225 spin_lock_irq(&tree->lock);
226 node = tree_insert(&tree->tree, file_offset,
227 &entry->rb_node);
228 if (node)
229 ordered_data_tree_panic(inode, -EEXIST, file_offset);
230 spin_unlock_irq(&tree->lock);
231
232 spin_lock(&root->ordered_extent_lock);
233 list_add_tail(&entry->root_extent_list,
234 &root->ordered_extents);
235 root->nr_ordered_extents++;
236 if (root->nr_ordered_extents == 1) {
237 spin_lock(&root->fs_info->ordered_root_lock);
238 BUG_ON(!list_empty(&root->ordered_root));
239 list_add_tail(&root->ordered_root,
240 &root->fs_info->ordered_roots);
241 spin_unlock(&root->fs_info->ordered_root_lock);
242 }
243 spin_unlock(&root->ordered_extent_lock);
244
245 return 0;
246 }
247
248 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
249 u64 start, u64 len, u64 disk_len, int type)
250 {
251 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
252 disk_len, type, 0,
253 BTRFS_COMPRESS_NONE);
254 }
255
256 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
257 u64 start, u64 len, u64 disk_len, int type)
258 {
259 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
260 disk_len, type, 1,
261 BTRFS_COMPRESS_NONE);
262 }
263
264 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
265 u64 start, u64 len, u64 disk_len,
266 int type, int compress_type)
267 {
268 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
269 disk_len, type, 0,
270 compress_type);
271 }
272
273 /*
274 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
275 * when an ordered extent is finished. If the list covers more than one
276 * ordered extent, it is split across multiples.
277 */
278 void btrfs_add_ordered_sum(struct inode *inode,
279 struct btrfs_ordered_extent *entry,
280 struct btrfs_ordered_sum *sum)
281 {
282 struct btrfs_ordered_inode_tree *tree;
283
284 tree = &BTRFS_I(inode)->ordered_tree;
285 spin_lock_irq(&tree->lock);
286 list_add_tail(&sum->list, &entry->list);
287 spin_unlock_irq(&tree->lock);
288 }
289
290 /*
291 * this is used to account for finished IO across a given range
292 * of the file. The IO may span ordered extents. If
293 * a given ordered_extent is completely done, 1 is returned, otherwise
294 * 0.
295 *
296 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
297 * to make sure this function only returns 1 once for a given ordered extent.
298 *
299 * file_offset is updated to one byte past the range that is recorded as
300 * complete. This allows you to walk forward in the file.
301 */
302 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
303 struct btrfs_ordered_extent **cached,
304 u64 *file_offset, u64 io_size, int uptodate)
305 {
306 struct btrfs_ordered_inode_tree *tree;
307 struct rb_node *node;
308 struct btrfs_ordered_extent *entry = NULL;
309 int ret;
310 unsigned long flags;
311 u64 dec_end;
312 u64 dec_start;
313 u64 to_dec;
314
315 tree = &BTRFS_I(inode)->ordered_tree;
316 spin_lock_irqsave(&tree->lock, flags);
317 node = tree_search(tree, *file_offset);
318 if (!node) {
319 ret = 1;
320 goto out;
321 }
322
323 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
324 if (!offset_in_entry(entry, *file_offset)) {
325 ret = 1;
326 goto out;
327 }
328
329 dec_start = max(*file_offset, entry->file_offset);
330 dec_end = min(*file_offset + io_size, entry->file_offset +
331 entry->len);
332 *file_offset = dec_end;
333 if (dec_start > dec_end) {
334 btrfs_crit(BTRFS_I(inode)->root->fs_info,
335 "bad ordering dec_start %llu end %llu", dec_start, dec_end);
336 }
337 to_dec = dec_end - dec_start;
338 if (to_dec > entry->bytes_left) {
339 btrfs_crit(BTRFS_I(inode)->root->fs_info,
340 "bad ordered accounting left %llu size %llu",
341 entry->bytes_left, to_dec);
342 }
343 entry->bytes_left -= to_dec;
344 if (!uptodate)
345 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
346
347 if (entry->bytes_left == 0) {
348 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
349 /*
350 * Implicit memory barrier after test_and_set_bit
351 */
352 if (waitqueue_active(&entry->wait))
353 wake_up(&entry->wait);
354 } else {
355 ret = 1;
356 }
357 out:
358 if (!ret && cached && entry) {
359 *cached = entry;
360 atomic_inc(&entry->refs);
361 }
362 spin_unlock_irqrestore(&tree->lock, flags);
363 return ret == 0;
364 }
365
366 /*
367 * this is used to account for finished IO across a given range
368 * of the file. The IO should not span ordered extents. If
369 * a given ordered_extent is completely done, 1 is returned, otherwise
370 * 0.
371 *
372 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
373 * to make sure this function only returns 1 once for a given ordered extent.
374 */
375 int btrfs_dec_test_ordered_pending(struct inode *inode,
376 struct btrfs_ordered_extent **cached,
377 u64 file_offset, u64 io_size, int uptodate)
378 {
379 struct btrfs_ordered_inode_tree *tree;
380 struct rb_node *node;
381 struct btrfs_ordered_extent *entry = NULL;
382 unsigned long flags;
383 int ret;
384
385 tree = &BTRFS_I(inode)->ordered_tree;
386 spin_lock_irqsave(&tree->lock, flags);
387 if (cached && *cached) {
388 entry = *cached;
389 goto have_entry;
390 }
391
392 node = tree_search(tree, file_offset);
393 if (!node) {
394 ret = 1;
395 goto out;
396 }
397
398 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
399 have_entry:
400 if (!offset_in_entry(entry, file_offset)) {
401 ret = 1;
402 goto out;
403 }
404
405 if (io_size > entry->bytes_left) {
406 btrfs_crit(BTRFS_I(inode)->root->fs_info,
407 "bad ordered accounting left %llu size %llu",
408 entry->bytes_left, io_size);
409 }
410 entry->bytes_left -= io_size;
411 if (!uptodate)
412 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
413
414 if (entry->bytes_left == 0) {
415 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
416 /*
417 * Implicit memory barrier after test_and_set_bit
418 */
419 if (waitqueue_active(&entry->wait))
420 wake_up(&entry->wait);
421 } else {
422 ret = 1;
423 }
424 out:
425 if (!ret && cached && entry) {
426 *cached = entry;
427 atomic_inc(&entry->refs);
428 }
429 spin_unlock_irqrestore(&tree->lock, flags);
430 return ret == 0;
431 }
432
433 /* Needs to either be called under a log transaction or the log_mutex */
434 void btrfs_get_logged_extents(struct inode *inode,
435 struct list_head *logged_list,
436 const loff_t start,
437 const loff_t end)
438 {
439 struct btrfs_ordered_inode_tree *tree;
440 struct btrfs_ordered_extent *ordered;
441 struct rb_node *n;
442 struct rb_node *prev;
443
444 tree = &BTRFS_I(inode)->ordered_tree;
445 spin_lock_irq(&tree->lock);
446 n = __tree_search(&tree->tree, end, &prev);
447 if (!n)
448 n = prev;
449 for (; n; n = rb_prev(n)) {
450 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
451 if (ordered->file_offset > end)
452 continue;
453 if (entry_end(ordered) <= start)
454 break;
455 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
456 continue;
457 list_add(&ordered->log_list, logged_list);
458 atomic_inc(&ordered->refs);
459 }
460 spin_unlock_irq(&tree->lock);
461 }
462
463 void btrfs_put_logged_extents(struct list_head *logged_list)
464 {
465 struct btrfs_ordered_extent *ordered;
466
467 while (!list_empty(logged_list)) {
468 ordered = list_first_entry(logged_list,
469 struct btrfs_ordered_extent,
470 log_list);
471 list_del_init(&ordered->log_list);
472 btrfs_put_ordered_extent(ordered);
473 }
474 }
475
476 void btrfs_submit_logged_extents(struct list_head *logged_list,
477 struct btrfs_root *log)
478 {
479 int index = log->log_transid % 2;
480
481 spin_lock_irq(&log->log_extents_lock[index]);
482 list_splice_tail(logged_list, &log->logged_list[index]);
483 spin_unlock_irq(&log->log_extents_lock[index]);
484 }
485
486 void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
487 struct btrfs_root *log, u64 transid)
488 {
489 struct btrfs_ordered_extent *ordered;
490 int index = transid % 2;
491
492 spin_lock_irq(&log->log_extents_lock[index]);
493 while (!list_empty(&log->logged_list[index])) {
494 struct inode *inode;
495 ordered = list_first_entry(&log->logged_list[index],
496 struct btrfs_ordered_extent,
497 log_list);
498 list_del_init(&ordered->log_list);
499 inode = ordered->inode;
500 spin_unlock_irq(&log->log_extents_lock[index]);
501
502 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
503 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
504 u64 start = ordered->file_offset;
505 u64 end = ordered->file_offset + ordered->len - 1;
506
507 WARN_ON(!inode);
508 filemap_fdatawrite_range(inode->i_mapping, start, end);
509 }
510 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
511 &ordered->flags));
512
513 /*
514 * In order to keep us from losing our ordered extent
515 * information when committing the transaction we have to make
516 * sure that any logged extents are completed when we go to
517 * commit the transaction. To do this we simply increase the
518 * current transactions pending_ordered counter and decrement it
519 * when the ordered extent completes.
520 */
521 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
522 struct btrfs_ordered_inode_tree *tree;
523
524 tree = &BTRFS_I(inode)->ordered_tree;
525 spin_lock_irq(&tree->lock);
526 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
527 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
528 atomic_inc(&trans->transaction->pending_ordered);
529 }
530 spin_unlock_irq(&tree->lock);
531 }
532 btrfs_put_ordered_extent(ordered);
533 spin_lock_irq(&log->log_extents_lock[index]);
534 }
535 spin_unlock_irq(&log->log_extents_lock[index]);
536 }
537
538 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
539 {
540 struct btrfs_ordered_extent *ordered;
541 int index = transid % 2;
542
543 spin_lock_irq(&log->log_extents_lock[index]);
544 while (!list_empty(&log->logged_list[index])) {
545 ordered = list_first_entry(&log->logged_list[index],
546 struct btrfs_ordered_extent,
547 log_list);
548 list_del_init(&ordered->log_list);
549 spin_unlock_irq(&log->log_extents_lock[index]);
550 btrfs_put_ordered_extent(ordered);
551 spin_lock_irq(&log->log_extents_lock[index]);
552 }
553 spin_unlock_irq(&log->log_extents_lock[index]);
554 }
555
556 /*
557 * used to drop a reference on an ordered extent. This will free
558 * the extent if the last reference is dropped
559 */
560 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
561 {
562 struct list_head *cur;
563 struct btrfs_ordered_sum *sum;
564
565 trace_btrfs_ordered_extent_put(entry->inode, entry);
566
567 if (atomic_dec_and_test(&entry->refs)) {
568 ASSERT(list_empty(&entry->log_list));
569 ASSERT(list_empty(&entry->trans_list));
570 ASSERT(list_empty(&entry->root_extent_list));
571 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
572 if (entry->inode)
573 btrfs_add_delayed_iput(entry->inode);
574 while (!list_empty(&entry->list)) {
575 cur = entry->list.next;
576 sum = list_entry(cur, struct btrfs_ordered_sum, list);
577 list_del(&sum->list);
578 kfree(sum);
579 }
580 kmem_cache_free(btrfs_ordered_extent_cache, entry);
581 }
582 }
583
584 /*
585 * remove an ordered extent from the tree. No references are dropped
586 * and waiters are woken up.
587 */
588 void btrfs_remove_ordered_extent(struct inode *inode,
589 struct btrfs_ordered_extent *entry)
590 {
591 struct btrfs_ordered_inode_tree *tree;
592 struct btrfs_root *root = BTRFS_I(inode)->root;
593 struct rb_node *node;
594 bool dec_pending_ordered = false;
595
596 tree = &BTRFS_I(inode)->ordered_tree;
597 spin_lock_irq(&tree->lock);
598 node = &entry->rb_node;
599 rb_erase(node, &tree->tree);
600 RB_CLEAR_NODE(node);
601 if (tree->last == node)
602 tree->last = NULL;
603 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
604 if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags))
605 dec_pending_ordered = true;
606 spin_unlock_irq(&tree->lock);
607
608 /*
609 * The current running transaction is waiting on us, we need to let it
610 * know that we're complete and wake it up.
611 */
612 if (dec_pending_ordered) {
613 struct btrfs_transaction *trans;
614
615 /*
616 * The checks for trans are just a formality, it should be set,
617 * but if it isn't we don't want to deref/assert under the spin
618 * lock, so be nice and check if trans is set, but ASSERT() so
619 * if it isn't set a developer will notice.
620 */
621 spin_lock(&root->fs_info->trans_lock);
622 trans = root->fs_info->running_transaction;
623 if (trans)
624 atomic_inc(&trans->use_count);
625 spin_unlock(&root->fs_info->trans_lock);
626
627 ASSERT(trans);
628 if (trans) {
629 if (atomic_dec_and_test(&trans->pending_ordered))
630 wake_up(&trans->pending_wait);
631 btrfs_put_transaction(trans);
632 }
633 }
634
635 spin_lock(&root->ordered_extent_lock);
636 list_del_init(&entry->root_extent_list);
637 root->nr_ordered_extents--;
638
639 trace_btrfs_ordered_extent_remove(inode, entry);
640
641 if (!root->nr_ordered_extents) {
642 spin_lock(&root->fs_info->ordered_root_lock);
643 BUG_ON(list_empty(&root->ordered_root));
644 list_del_init(&root->ordered_root);
645 spin_unlock(&root->fs_info->ordered_root_lock);
646 }
647 spin_unlock(&root->ordered_extent_lock);
648 wake_up(&entry->wait);
649 }
650
651 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
652 {
653 struct btrfs_ordered_extent *ordered;
654
655 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
656 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
657 complete(&ordered->completion);
658 }
659
660 /*
661 * wait for all the ordered extents in a root. This is done when balancing
662 * space between drives.
663 */
664 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr,
665 const u64 range_start, const u64 range_len)
666 {
667 LIST_HEAD(splice);
668 LIST_HEAD(skipped);
669 LIST_HEAD(works);
670 struct btrfs_ordered_extent *ordered, *next;
671 int count = 0;
672 const u64 range_end = range_start + range_len;
673
674 mutex_lock(&root->ordered_extent_mutex);
675 spin_lock(&root->ordered_extent_lock);
676 list_splice_init(&root->ordered_extents, &splice);
677 while (!list_empty(&splice) && nr) {
678 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
679 root_extent_list);
680
681 if (range_end <= ordered->start ||
682 ordered->start + ordered->disk_len <= range_start) {
683 list_move_tail(&ordered->root_extent_list, &skipped);
684 cond_resched_lock(&root->ordered_extent_lock);
685 continue;
686 }
687
688 list_move_tail(&ordered->root_extent_list,
689 &root->ordered_extents);
690 atomic_inc(&ordered->refs);
691 spin_unlock(&root->ordered_extent_lock);
692
693 btrfs_init_work(&ordered->flush_work,
694 btrfs_flush_delalloc_helper,
695 btrfs_run_ordered_extent_work, NULL, NULL);
696 list_add_tail(&ordered->work_list, &works);
697 btrfs_queue_work(root->fs_info->flush_workers,
698 &ordered->flush_work);
699
700 cond_resched();
701 spin_lock(&root->ordered_extent_lock);
702 if (nr != -1)
703 nr--;
704 count++;
705 }
706 list_splice_tail(&skipped, &root->ordered_extents);
707 list_splice_tail(&splice, &root->ordered_extents);
708 spin_unlock(&root->ordered_extent_lock);
709
710 list_for_each_entry_safe(ordered, next, &works, work_list) {
711 list_del_init(&ordered->work_list);
712 wait_for_completion(&ordered->completion);
713 btrfs_put_ordered_extent(ordered);
714 cond_resched();
715 }
716 mutex_unlock(&root->ordered_extent_mutex);
717
718 return count;
719 }
720
721 int btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
722 const u64 range_start, const u64 range_len)
723 {
724 struct btrfs_root *root;
725 struct list_head splice;
726 int done;
727 int total_done = 0;
728
729 INIT_LIST_HEAD(&splice);
730
731 mutex_lock(&fs_info->ordered_operations_mutex);
732 spin_lock(&fs_info->ordered_root_lock);
733 list_splice_init(&fs_info->ordered_roots, &splice);
734 while (!list_empty(&splice) && nr) {
735 root = list_first_entry(&splice, struct btrfs_root,
736 ordered_root);
737 root = btrfs_grab_fs_root(root);
738 BUG_ON(!root);
739 list_move_tail(&root->ordered_root,
740 &fs_info->ordered_roots);
741 spin_unlock(&fs_info->ordered_root_lock);
742
743 done = btrfs_wait_ordered_extents(root, nr,
744 range_start, range_len);
745 btrfs_put_fs_root(root);
746 total_done += done;
747
748 spin_lock(&fs_info->ordered_root_lock);
749 if (nr != -1) {
750 nr -= done;
751 WARN_ON(nr < 0);
752 }
753 }
754 list_splice_tail(&splice, &fs_info->ordered_roots);
755 spin_unlock(&fs_info->ordered_root_lock);
756 mutex_unlock(&fs_info->ordered_operations_mutex);
757
758 return total_done;
759 }
760
761 /*
762 * Used to start IO or wait for a given ordered extent to finish.
763 *
764 * If wait is one, this effectively waits on page writeback for all the pages
765 * in the extent, and it waits on the io completion code to insert
766 * metadata into the btree corresponding to the extent
767 */
768 void btrfs_start_ordered_extent(struct inode *inode,
769 struct btrfs_ordered_extent *entry,
770 int wait)
771 {
772 u64 start = entry->file_offset;
773 u64 end = start + entry->len - 1;
774
775 trace_btrfs_ordered_extent_start(inode, entry);
776
777 /*
778 * pages in the range can be dirty, clean or writeback. We
779 * start IO on any dirty ones so the wait doesn't stall waiting
780 * for the flusher thread to find them
781 */
782 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
783 filemap_fdatawrite_range(inode->i_mapping, start, end);
784 if (wait) {
785 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
786 &entry->flags));
787 }
788 }
789
790 /*
791 * Used to wait on ordered extents across a large range of bytes.
792 */
793 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
794 {
795 int ret = 0;
796 int ret_wb = 0;
797 u64 end;
798 u64 orig_end;
799 struct btrfs_ordered_extent *ordered;
800
801 if (start + len < start) {
802 orig_end = INT_LIMIT(loff_t);
803 } else {
804 orig_end = start + len - 1;
805 if (orig_end > INT_LIMIT(loff_t))
806 orig_end = INT_LIMIT(loff_t);
807 }
808
809 /* start IO across the range first to instantiate any delalloc
810 * extents
811 */
812 ret = btrfs_fdatawrite_range(inode, start, orig_end);
813 if (ret)
814 return ret;
815
816 /*
817 * If we have a writeback error don't return immediately. Wait first
818 * for any ordered extents that haven't completed yet. This is to make
819 * sure no one can dirty the same page ranges and call writepages()
820 * before the ordered extents complete - to avoid failures (-EEXIST)
821 * when adding the new ordered extents to the ordered tree.
822 */
823 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
824
825 end = orig_end;
826 while (1) {
827 ordered = btrfs_lookup_first_ordered_extent(inode, end);
828 if (!ordered)
829 break;
830 if (ordered->file_offset > orig_end) {
831 btrfs_put_ordered_extent(ordered);
832 break;
833 }
834 if (ordered->file_offset + ordered->len <= start) {
835 btrfs_put_ordered_extent(ordered);
836 break;
837 }
838 btrfs_start_ordered_extent(inode, ordered, 1);
839 end = ordered->file_offset;
840 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
841 ret = -EIO;
842 btrfs_put_ordered_extent(ordered);
843 if (ret || end == 0 || end == start)
844 break;
845 end--;
846 }
847 return ret_wb ? ret_wb : ret;
848 }
849
850 /*
851 * find an ordered extent corresponding to file_offset. return NULL if
852 * nothing is found, otherwise take a reference on the extent and return it
853 */
854 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
855 u64 file_offset)
856 {
857 struct btrfs_ordered_inode_tree *tree;
858 struct rb_node *node;
859 struct btrfs_ordered_extent *entry = NULL;
860
861 tree = &BTRFS_I(inode)->ordered_tree;
862 spin_lock_irq(&tree->lock);
863 node = tree_search(tree, file_offset);
864 if (!node)
865 goto out;
866
867 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
868 if (!offset_in_entry(entry, file_offset))
869 entry = NULL;
870 if (entry)
871 atomic_inc(&entry->refs);
872 out:
873 spin_unlock_irq(&tree->lock);
874 return entry;
875 }
876
877 /* Since the DIO code tries to lock a wide area we need to look for any ordered
878 * extents that exist in the range, rather than just the start of the range.
879 */
880 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
881 u64 file_offset,
882 u64 len)
883 {
884 struct btrfs_ordered_inode_tree *tree;
885 struct rb_node *node;
886 struct btrfs_ordered_extent *entry = NULL;
887
888 tree = &BTRFS_I(inode)->ordered_tree;
889 spin_lock_irq(&tree->lock);
890 node = tree_search(tree, file_offset);
891 if (!node) {
892 node = tree_search(tree, file_offset + len);
893 if (!node)
894 goto out;
895 }
896
897 while (1) {
898 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
899 if (range_overlaps(entry, file_offset, len))
900 break;
901
902 if (entry->file_offset >= file_offset + len) {
903 entry = NULL;
904 break;
905 }
906 entry = NULL;
907 node = rb_next(node);
908 if (!node)
909 break;
910 }
911 out:
912 if (entry)
913 atomic_inc(&entry->refs);
914 spin_unlock_irq(&tree->lock);
915 return entry;
916 }
917
918 bool btrfs_have_ordered_extents_in_range(struct inode *inode,
919 u64 file_offset,
920 u64 len)
921 {
922 struct btrfs_ordered_extent *oe;
923
924 oe = btrfs_lookup_ordered_range(inode, file_offset, len);
925 if (oe) {
926 btrfs_put_ordered_extent(oe);
927 return true;
928 }
929 return false;
930 }
931
932 /*
933 * lookup and return any extent before 'file_offset'. NULL is returned
934 * if none is found
935 */
936 struct btrfs_ordered_extent *
937 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
938 {
939 struct btrfs_ordered_inode_tree *tree;
940 struct rb_node *node;
941 struct btrfs_ordered_extent *entry = NULL;
942
943 tree = &BTRFS_I(inode)->ordered_tree;
944 spin_lock_irq(&tree->lock);
945 node = tree_search(tree, file_offset);
946 if (!node)
947 goto out;
948
949 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
950 atomic_inc(&entry->refs);
951 out:
952 spin_unlock_irq(&tree->lock);
953 return entry;
954 }
955
956 /*
957 * After an extent is done, call this to conditionally update the on disk
958 * i_size. i_size is updated to cover any fully written part of the file.
959 */
960 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
961 struct btrfs_ordered_extent *ordered)
962 {
963 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
964 u64 disk_i_size;
965 u64 new_i_size;
966 u64 i_size = i_size_read(inode);
967 struct rb_node *node;
968 struct rb_node *prev = NULL;
969 struct btrfs_ordered_extent *test;
970 int ret = 1;
971 u64 orig_offset = offset;
972
973 spin_lock_irq(&tree->lock);
974 if (ordered) {
975 offset = entry_end(ordered);
976 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
977 offset = min(offset,
978 ordered->file_offset +
979 ordered->truncated_len);
980 } else {
981 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
982 }
983 disk_i_size = BTRFS_I(inode)->disk_i_size;
984
985 /* truncate file */
986 if (disk_i_size > i_size) {
987 BTRFS_I(inode)->disk_i_size = orig_offset;
988 ret = 0;
989 goto out;
990 }
991
992 /*
993 * if the disk i_size is already at the inode->i_size, or
994 * this ordered extent is inside the disk i_size, we're done
995 */
996 if (disk_i_size == i_size)
997 goto out;
998
999 /*
1000 * We still need to update disk_i_size if outstanding_isize is greater
1001 * than disk_i_size.
1002 */
1003 if (offset <= disk_i_size &&
1004 (!ordered || ordered->outstanding_isize <= disk_i_size))
1005 goto out;
1006
1007 /*
1008 * walk backward from this ordered extent to disk_i_size.
1009 * if we find an ordered extent then we can't update disk i_size
1010 * yet
1011 */
1012 if (ordered) {
1013 node = rb_prev(&ordered->rb_node);
1014 } else {
1015 prev = tree_search(tree, offset);
1016 /*
1017 * we insert file extents without involving ordered struct,
1018 * so there should be no ordered struct cover this offset
1019 */
1020 if (prev) {
1021 test = rb_entry(prev, struct btrfs_ordered_extent,
1022 rb_node);
1023 BUG_ON(offset_in_entry(test, offset));
1024 }
1025 node = prev;
1026 }
1027 for (; node; node = rb_prev(node)) {
1028 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1029
1030 /* We treat this entry as if it doesn't exist */
1031 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1032 continue;
1033 if (test->file_offset + test->len <= disk_i_size)
1034 break;
1035 if (test->file_offset >= i_size)
1036 break;
1037 if (entry_end(test) > disk_i_size) {
1038 /*
1039 * we don't update disk_i_size now, so record this
1040 * undealt i_size. Or we will not know the real
1041 * i_size.
1042 */
1043 if (test->outstanding_isize < offset)
1044 test->outstanding_isize = offset;
1045 if (ordered &&
1046 ordered->outstanding_isize >
1047 test->outstanding_isize)
1048 test->outstanding_isize =
1049 ordered->outstanding_isize;
1050 goto out;
1051 }
1052 }
1053 new_i_size = min_t(u64, offset, i_size);
1054
1055 /*
1056 * Some ordered extents may completed before the current one, and
1057 * we hold the real i_size in ->outstanding_isize.
1058 */
1059 if (ordered && ordered->outstanding_isize > new_i_size)
1060 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
1061 BTRFS_I(inode)->disk_i_size = new_i_size;
1062 ret = 0;
1063 out:
1064 /*
1065 * We need to do this because we can't remove ordered extents until
1066 * after the i_disk_size has been updated and then the inode has been
1067 * updated to reflect the change, so we need to tell anybody who finds
1068 * this ordered extent that we've already done all the real work, we
1069 * just haven't completed all the other work.
1070 */
1071 if (ordered)
1072 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1073 spin_unlock_irq(&tree->lock);
1074 return ret;
1075 }
1076
1077 /*
1078 * search the ordered extents for one corresponding to 'offset' and
1079 * try to find a checksum. This is used because we allow pages to
1080 * be reclaimed before their checksum is actually put into the btree
1081 */
1082 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1083 u32 *sum, int len)
1084 {
1085 struct btrfs_ordered_sum *ordered_sum;
1086 struct btrfs_ordered_extent *ordered;
1087 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1088 unsigned long num_sectors;
1089 unsigned long i;
1090 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1091 int index = 0;
1092
1093 ordered = btrfs_lookup_ordered_extent(inode, offset);
1094 if (!ordered)
1095 return 0;
1096
1097 spin_lock_irq(&tree->lock);
1098 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1099 if (disk_bytenr >= ordered_sum->bytenr &&
1100 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1101 i = (disk_bytenr - ordered_sum->bytenr) >>
1102 inode->i_sb->s_blocksize_bits;
1103 num_sectors = ordered_sum->len >>
1104 inode->i_sb->s_blocksize_bits;
1105 num_sectors = min_t(int, len - index, num_sectors - i);
1106 memcpy(sum + index, ordered_sum->sums + i,
1107 num_sectors);
1108
1109 index += (int)num_sectors;
1110 if (index == len)
1111 goto out;
1112 disk_bytenr += num_sectors * sectorsize;
1113 }
1114 }
1115 out:
1116 spin_unlock_irq(&tree->lock);
1117 btrfs_put_ordered_extent(ordered);
1118 return index;
1119 }
1120
1121 int __init ordered_data_init(void)
1122 {
1123 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1124 sizeof(struct btrfs_ordered_extent), 0,
1125 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1126 NULL);
1127 if (!btrfs_ordered_extent_cache)
1128 return -ENOMEM;
1129
1130 return 0;
1131 }
1132
1133 void ordered_data_exit(void)
1134 {
1135 kmem_cache_destroy(btrfs_ordered_extent_cache);
1136 }
This page took 0.053102 seconds and 6 git commands to generate.