]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/btrfs/ordered-data.c
Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux
[mirror_ubuntu-bionic-kernel.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,
71 "Inconsistency in ordered tree at offset %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_fs_info *fs_info = btrfs_sb(inode->i_sb);
190 struct btrfs_root *root = BTRFS_I(inode)->root;
191 struct btrfs_ordered_inode_tree *tree;
192 struct rb_node *node;
193 struct btrfs_ordered_extent *entry;
194
195 tree = &BTRFS_I(inode)->ordered_tree;
196 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
197 if (!entry)
198 return -ENOMEM;
199
200 entry->file_offset = file_offset;
201 entry->start = start;
202 entry->len = len;
203 entry->disk_len = disk_len;
204 entry->bytes_left = len;
205 entry->inode = igrab(inode);
206 entry->compress_type = compress_type;
207 entry->truncated_len = (u64)-1;
208 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
209 set_bit(type, &entry->flags);
210
211 if (dio)
212 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
213
214 /* one ref for the tree */
215 refcount_set(&entry->refs, 1);
216 init_waitqueue_head(&entry->wait);
217 INIT_LIST_HEAD(&entry->list);
218 INIT_LIST_HEAD(&entry->root_extent_list);
219 INIT_LIST_HEAD(&entry->work_list);
220 init_completion(&entry->completion);
221 INIT_LIST_HEAD(&entry->log_list);
222 INIT_LIST_HEAD(&entry->trans_list);
223
224 trace_btrfs_ordered_extent_add(inode, entry);
225
226 spin_lock_irq(&tree->lock);
227 node = tree_insert(&tree->tree, file_offset,
228 &entry->rb_node);
229 if (node)
230 ordered_data_tree_panic(inode, -EEXIST, file_offset);
231 spin_unlock_irq(&tree->lock);
232
233 spin_lock(&root->ordered_extent_lock);
234 list_add_tail(&entry->root_extent_list,
235 &root->ordered_extents);
236 root->nr_ordered_extents++;
237 if (root->nr_ordered_extents == 1) {
238 spin_lock(&fs_info->ordered_root_lock);
239 BUG_ON(!list_empty(&root->ordered_root));
240 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
241 spin_unlock(&fs_info->ordered_root_lock);
242 }
243 spin_unlock(&root->ordered_extent_lock);
244
245 /*
246 * We don't need the count_max_extents here, we can assume that all of
247 * that work has been done at higher layers, so this is truly the
248 * smallest the extent is going to get.
249 */
250 spin_lock(&BTRFS_I(inode)->lock);
251 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
252 spin_unlock(&BTRFS_I(inode)->lock);
253
254 return 0;
255 }
256
257 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
258 u64 start, u64 len, u64 disk_len, int type)
259 {
260 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
261 disk_len, type, 0,
262 BTRFS_COMPRESS_NONE);
263 }
264
265 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
266 u64 start, u64 len, u64 disk_len, int type)
267 {
268 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
269 disk_len, type, 1,
270 BTRFS_COMPRESS_NONE);
271 }
272
273 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
274 u64 start, u64 len, u64 disk_len,
275 int type, int compress_type)
276 {
277 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
278 disk_len, type, 0,
279 compress_type);
280 }
281
282 /*
283 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
284 * when an ordered extent is finished. If the list covers more than one
285 * ordered extent, it is split across multiples.
286 */
287 void btrfs_add_ordered_sum(struct inode *inode,
288 struct btrfs_ordered_extent *entry,
289 struct btrfs_ordered_sum *sum)
290 {
291 struct btrfs_ordered_inode_tree *tree;
292
293 tree = &BTRFS_I(inode)->ordered_tree;
294 spin_lock_irq(&tree->lock);
295 list_add_tail(&sum->list, &entry->list);
296 spin_unlock_irq(&tree->lock);
297 }
298
299 /*
300 * this is used to account for finished IO across a given range
301 * of the file. The IO may span ordered extents. If
302 * a given ordered_extent is completely done, 1 is returned, otherwise
303 * 0.
304 *
305 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
306 * to make sure this function only returns 1 once for a given ordered extent.
307 *
308 * file_offset is updated to one byte past the range that is recorded as
309 * complete. This allows you to walk forward in the file.
310 */
311 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
312 struct btrfs_ordered_extent **cached,
313 u64 *file_offset, u64 io_size, int uptodate)
314 {
315 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
316 struct btrfs_ordered_inode_tree *tree;
317 struct rb_node *node;
318 struct btrfs_ordered_extent *entry = NULL;
319 int ret;
320 unsigned long flags;
321 u64 dec_end;
322 u64 dec_start;
323 u64 to_dec;
324
325 tree = &BTRFS_I(inode)->ordered_tree;
326 spin_lock_irqsave(&tree->lock, flags);
327 node = tree_search(tree, *file_offset);
328 if (!node) {
329 ret = 1;
330 goto out;
331 }
332
333 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
334 if (!offset_in_entry(entry, *file_offset)) {
335 ret = 1;
336 goto out;
337 }
338
339 dec_start = max(*file_offset, entry->file_offset);
340 dec_end = min(*file_offset + io_size, entry->file_offset +
341 entry->len);
342 *file_offset = dec_end;
343 if (dec_start > dec_end) {
344 btrfs_crit(fs_info, "bad ordering dec_start %llu end %llu",
345 dec_start, dec_end);
346 }
347 to_dec = dec_end - dec_start;
348 if (to_dec > entry->bytes_left) {
349 btrfs_crit(fs_info,
350 "bad ordered accounting left %llu size %llu",
351 entry->bytes_left, to_dec);
352 }
353 entry->bytes_left -= to_dec;
354 if (!uptodate)
355 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
356
357 if (entry->bytes_left == 0) {
358 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
359 /*
360 * Implicit memory barrier after test_and_set_bit
361 */
362 if (waitqueue_active(&entry->wait))
363 wake_up(&entry->wait);
364 } else {
365 ret = 1;
366 }
367 out:
368 if (!ret && cached && entry) {
369 *cached = entry;
370 refcount_inc(&entry->refs);
371 }
372 spin_unlock_irqrestore(&tree->lock, flags);
373 return ret == 0;
374 }
375
376 /*
377 * this is used to account for finished IO across a given range
378 * of the file. The IO should not span ordered extents. If
379 * a given ordered_extent is completely done, 1 is returned, otherwise
380 * 0.
381 *
382 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
383 * to make sure this function only returns 1 once for a given ordered extent.
384 */
385 int btrfs_dec_test_ordered_pending(struct inode *inode,
386 struct btrfs_ordered_extent **cached,
387 u64 file_offset, u64 io_size, int uptodate)
388 {
389 struct btrfs_ordered_inode_tree *tree;
390 struct rb_node *node;
391 struct btrfs_ordered_extent *entry = NULL;
392 unsigned long flags;
393 int ret;
394
395 tree = &BTRFS_I(inode)->ordered_tree;
396 spin_lock_irqsave(&tree->lock, flags);
397 if (cached && *cached) {
398 entry = *cached;
399 goto have_entry;
400 }
401
402 node = tree_search(tree, file_offset);
403 if (!node) {
404 ret = 1;
405 goto out;
406 }
407
408 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
409 have_entry:
410 if (!offset_in_entry(entry, file_offset)) {
411 ret = 1;
412 goto out;
413 }
414
415 if (io_size > entry->bytes_left) {
416 btrfs_crit(BTRFS_I(inode)->root->fs_info,
417 "bad ordered accounting left %llu size %llu",
418 entry->bytes_left, io_size);
419 }
420 entry->bytes_left -= io_size;
421 if (!uptodate)
422 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
423
424 if (entry->bytes_left == 0) {
425 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
426 /*
427 * Implicit memory barrier after test_and_set_bit
428 */
429 if (waitqueue_active(&entry->wait))
430 wake_up(&entry->wait);
431 } else {
432 ret = 1;
433 }
434 out:
435 if (!ret && cached && entry) {
436 *cached = entry;
437 refcount_inc(&entry->refs);
438 }
439 spin_unlock_irqrestore(&tree->lock, flags);
440 return ret == 0;
441 }
442
443 /* Needs to either be called under a log transaction or the log_mutex */
444 void btrfs_get_logged_extents(struct btrfs_inode *inode,
445 struct list_head *logged_list,
446 const loff_t start,
447 const loff_t end)
448 {
449 struct btrfs_ordered_inode_tree *tree;
450 struct btrfs_ordered_extent *ordered;
451 struct rb_node *n;
452 struct rb_node *prev;
453
454 tree = &inode->ordered_tree;
455 spin_lock_irq(&tree->lock);
456 n = __tree_search(&tree->tree, end, &prev);
457 if (!n)
458 n = prev;
459 for (; n; n = rb_prev(n)) {
460 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
461 if (ordered->file_offset > end)
462 continue;
463 if (entry_end(ordered) <= start)
464 break;
465 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
466 continue;
467 list_add(&ordered->log_list, logged_list);
468 refcount_inc(&ordered->refs);
469 }
470 spin_unlock_irq(&tree->lock);
471 }
472
473 void btrfs_put_logged_extents(struct list_head *logged_list)
474 {
475 struct btrfs_ordered_extent *ordered;
476
477 while (!list_empty(logged_list)) {
478 ordered = list_first_entry(logged_list,
479 struct btrfs_ordered_extent,
480 log_list);
481 list_del_init(&ordered->log_list);
482 btrfs_put_ordered_extent(ordered);
483 }
484 }
485
486 void btrfs_submit_logged_extents(struct list_head *logged_list,
487 struct btrfs_root *log)
488 {
489 int index = log->log_transid % 2;
490
491 spin_lock_irq(&log->log_extents_lock[index]);
492 list_splice_tail(logged_list, &log->logged_list[index]);
493 spin_unlock_irq(&log->log_extents_lock[index]);
494 }
495
496 void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
497 struct btrfs_root *log, u64 transid)
498 {
499 struct btrfs_ordered_extent *ordered;
500 int index = transid % 2;
501
502 spin_lock_irq(&log->log_extents_lock[index]);
503 while (!list_empty(&log->logged_list[index])) {
504 struct inode *inode;
505 ordered = list_first_entry(&log->logged_list[index],
506 struct btrfs_ordered_extent,
507 log_list);
508 list_del_init(&ordered->log_list);
509 inode = ordered->inode;
510 spin_unlock_irq(&log->log_extents_lock[index]);
511
512 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
513 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
514 u64 start = ordered->file_offset;
515 u64 end = ordered->file_offset + ordered->len - 1;
516
517 WARN_ON(!inode);
518 filemap_fdatawrite_range(inode->i_mapping, start, end);
519 }
520 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
521 &ordered->flags));
522
523 /*
524 * In order to keep us from losing our ordered extent
525 * information when committing the transaction we have to make
526 * sure that any logged extents are completed when we go to
527 * commit the transaction. To do this we simply increase the
528 * current transactions pending_ordered counter and decrement it
529 * when the ordered extent completes.
530 */
531 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
532 struct btrfs_ordered_inode_tree *tree;
533
534 tree = &BTRFS_I(inode)->ordered_tree;
535 spin_lock_irq(&tree->lock);
536 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
537 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
538 atomic_inc(&trans->transaction->pending_ordered);
539 }
540 spin_unlock_irq(&tree->lock);
541 }
542 btrfs_put_ordered_extent(ordered);
543 spin_lock_irq(&log->log_extents_lock[index]);
544 }
545 spin_unlock_irq(&log->log_extents_lock[index]);
546 }
547
548 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
549 {
550 struct btrfs_ordered_extent *ordered;
551 int index = transid % 2;
552
553 spin_lock_irq(&log->log_extents_lock[index]);
554 while (!list_empty(&log->logged_list[index])) {
555 ordered = list_first_entry(&log->logged_list[index],
556 struct btrfs_ordered_extent,
557 log_list);
558 list_del_init(&ordered->log_list);
559 spin_unlock_irq(&log->log_extents_lock[index]);
560 btrfs_put_ordered_extent(ordered);
561 spin_lock_irq(&log->log_extents_lock[index]);
562 }
563 spin_unlock_irq(&log->log_extents_lock[index]);
564 }
565
566 /*
567 * used to drop a reference on an ordered extent. This will free
568 * the extent if the last reference is dropped
569 */
570 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
571 {
572 struct list_head *cur;
573 struct btrfs_ordered_sum *sum;
574
575 trace_btrfs_ordered_extent_put(entry->inode, entry);
576
577 if (refcount_dec_and_test(&entry->refs)) {
578 ASSERT(list_empty(&entry->log_list));
579 ASSERT(list_empty(&entry->trans_list));
580 ASSERT(list_empty(&entry->root_extent_list));
581 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
582 if (entry->inode)
583 btrfs_add_delayed_iput(entry->inode);
584 while (!list_empty(&entry->list)) {
585 cur = entry->list.next;
586 sum = list_entry(cur, struct btrfs_ordered_sum, list);
587 list_del(&sum->list);
588 kfree(sum);
589 }
590 kmem_cache_free(btrfs_ordered_extent_cache, entry);
591 }
592 }
593
594 /*
595 * remove an ordered extent from the tree. No references are dropped
596 * and waiters are woken up.
597 */
598 void btrfs_remove_ordered_extent(struct inode *inode,
599 struct btrfs_ordered_extent *entry)
600 {
601 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
602 struct btrfs_ordered_inode_tree *tree;
603 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
604 struct btrfs_root *root = btrfs_inode->root;
605 struct rb_node *node;
606 bool dec_pending_ordered = false;
607
608 /* This is paired with btrfs_add_ordered_extent. */
609 spin_lock(&btrfs_inode->lock);
610 btrfs_mod_outstanding_extents(btrfs_inode, -1);
611 spin_unlock(&btrfs_inode->lock);
612 if (root != fs_info->tree_root)
613 btrfs_delalloc_release_metadata(btrfs_inode, entry->len);
614
615 tree = &btrfs_inode->ordered_tree;
616 spin_lock_irq(&tree->lock);
617 node = &entry->rb_node;
618 rb_erase(node, &tree->tree);
619 RB_CLEAR_NODE(node);
620 if (tree->last == node)
621 tree->last = NULL;
622 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
623 if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags))
624 dec_pending_ordered = true;
625 spin_unlock_irq(&tree->lock);
626
627 /*
628 * The current running transaction is waiting on us, we need to let it
629 * know that we're complete and wake it up.
630 */
631 if (dec_pending_ordered) {
632 struct btrfs_transaction *trans;
633
634 /*
635 * The checks for trans are just a formality, it should be set,
636 * but if it isn't we don't want to deref/assert under the spin
637 * lock, so be nice and check if trans is set, but ASSERT() so
638 * if it isn't set a developer will notice.
639 */
640 spin_lock(&fs_info->trans_lock);
641 trans = fs_info->running_transaction;
642 if (trans)
643 refcount_inc(&trans->use_count);
644 spin_unlock(&fs_info->trans_lock);
645
646 ASSERT(trans);
647 if (trans) {
648 if (atomic_dec_and_test(&trans->pending_ordered))
649 wake_up(&trans->pending_wait);
650 btrfs_put_transaction(trans);
651 }
652 }
653
654 spin_lock(&root->ordered_extent_lock);
655 list_del_init(&entry->root_extent_list);
656 root->nr_ordered_extents--;
657
658 trace_btrfs_ordered_extent_remove(inode, entry);
659
660 if (!root->nr_ordered_extents) {
661 spin_lock(&fs_info->ordered_root_lock);
662 BUG_ON(list_empty(&root->ordered_root));
663 list_del_init(&root->ordered_root);
664 spin_unlock(&fs_info->ordered_root_lock);
665 }
666 spin_unlock(&root->ordered_extent_lock);
667 wake_up(&entry->wait);
668 }
669
670 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
671 {
672 struct btrfs_ordered_extent *ordered;
673
674 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
675 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
676 complete(&ordered->completion);
677 }
678
679 /*
680 * wait for all the ordered extents in a root. This is done when balancing
681 * space between drives.
682 */
683 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
684 const u64 range_start, const u64 range_len)
685 {
686 struct btrfs_fs_info *fs_info = root->fs_info;
687 LIST_HEAD(splice);
688 LIST_HEAD(skipped);
689 LIST_HEAD(works);
690 struct btrfs_ordered_extent *ordered, *next;
691 u64 count = 0;
692 const u64 range_end = range_start + range_len;
693
694 mutex_lock(&root->ordered_extent_mutex);
695 spin_lock(&root->ordered_extent_lock);
696 list_splice_init(&root->ordered_extents, &splice);
697 while (!list_empty(&splice) && nr) {
698 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
699 root_extent_list);
700
701 if (range_end <= ordered->start ||
702 ordered->start + ordered->disk_len <= range_start) {
703 list_move_tail(&ordered->root_extent_list, &skipped);
704 cond_resched_lock(&root->ordered_extent_lock);
705 continue;
706 }
707
708 list_move_tail(&ordered->root_extent_list,
709 &root->ordered_extents);
710 refcount_inc(&ordered->refs);
711 spin_unlock(&root->ordered_extent_lock);
712
713 btrfs_init_work(&ordered->flush_work,
714 btrfs_flush_delalloc_helper,
715 btrfs_run_ordered_extent_work, NULL, NULL);
716 list_add_tail(&ordered->work_list, &works);
717 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
718
719 cond_resched();
720 spin_lock(&root->ordered_extent_lock);
721 if (nr != U64_MAX)
722 nr--;
723 count++;
724 }
725 list_splice_tail(&skipped, &root->ordered_extents);
726 list_splice_tail(&splice, &root->ordered_extents);
727 spin_unlock(&root->ordered_extent_lock);
728
729 list_for_each_entry_safe(ordered, next, &works, work_list) {
730 list_del_init(&ordered->work_list);
731 wait_for_completion(&ordered->completion);
732 btrfs_put_ordered_extent(ordered);
733 cond_resched();
734 }
735 mutex_unlock(&root->ordered_extent_mutex);
736
737 return count;
738 }
739
740 u64 btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
741 const u64 range_start, const u64 range_len)
742 {
743 struct btrfs_root *root;
744 struct list_head splice;
745 u64 total_done = 0;
746 u64 done;
747
748 INIT_LIST_HEAD(&splice);
749
750 mutex_lock(&fs_info->ordered_operations_mutex);
751 spin_lock(&fs_info->ordered_root_lock);
752 list_splice_init(&fs_info->ordered_roots, &splice);
753 while (!list_empty(&splice) && nr) {
754 root = list_first_entry(&splice, struct btrfs_root,
755 ordered_root);
756 root = btrfs_grab_fs_root(root);
757 BUG_ON(!root);
758 list_move_tail(&root->ordered_root,
759 &fs_info->ordered_roots);
760 spin_unlock(&fs_info->ordered_root_lock);
761
762 done = btrfs_wait_ordered_extents(root, nr,
763 range_start, range_len);
764 btrfs_put_fs_root(root);
765 total_done += done;
766
767 spin_lock(&fs_info->ordered_root_lock);
768 if (nr != U64_MAX) {
769 nr -= done;
770 }
771 }
772 list_splice_tail(&splice, &fs_info->ordered_roots);
773 spin_unlock(&fs_info->ordered_root_lock);
774 mutex_unlock(&fs_info->ordered_operations_mutex);
775
776 return total_done;
777 }
778
779 /*
780 * Used to start IO or wait for a given ordered extent to finish.
781 *
782 * If wait is one, this effectively waits on page writeback for all the pages
783 * in the extent, and it waits on the io completion code to insert
784 * metadata into the btree corresponding to the extent
785 */
786 void btrfs_start_ordered_extent(struct inode *inode,
787 struct btrfs_ordered_extent *entry,
788 int wait)
789 {
790 u64 start = entry->file_offset;
791 u64 end = start + entry->len - 1;
792
793 trace_btrfs_ordered_extent_start(inode, entry);
794
795 /*
796 * pages in the range can be dirty, clean or writeback. We
797 * start IO on any dirty ones so the wait doesn't stall waiting
798 * for the flusher thread to find them
799 */
800 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
801 filemap_fdatawrite_range(inode->i_mapping, start, end);
802 if (wait) {
803 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
804 &entry->flags));
805 }
806 }
807
808 /*
809 * Used to wait on ordered extents across a large range of bytes.
810 */
811 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
812 {
813 int ret = 0;
814 int ret_wb = 0;
815 u64 end;
816 u64 orig_end;
817 struct btrfs_ordered_extent *ordered;
818
819 if (start + len < start) {
820 orig_end = INT_LIMIT(loff_t);
821 } else {
822 orig_end = start + len - 1;
823 if (orig_end > INT_LIMIT(loff_t))
824 orig_end = INT_LIMIT(loff_t);
825 }
826
827 /* start IO across the range first to instantiate any delalloc
828 * extents
829 */
830 ret = btrfs_fdatawrite_range(inode, start, orig_end);
831 if (ret)
832 return ret;
833
834 /*
835 * If we have a writeback error don't return immediately. Wait first
836 * for any ordered extents that haven't completed yet. This is to make
837 * sure no one can dirty the same page ranges and call writepages()
838 * before the ordered extents complete - to avoid failures (-EEXIST)
839 * when adding the new ordered extents to the ordered tree.
840 */
841 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
842
843 end = orig_end;
844 while (1) {
845 ordered = btrfs_lookup_first_ordered_extent(inode, end);
846 if (!ordered)
847 break;
848 if (ordered->file_offset > orig_end) {
849 btrfs_put_ordered_extent(ordered);
850 break;
851 }
852 if (ordered->file_offset + ordered->len <= start) {
853 btrfs_put_ordered_extent(ordered);
854 break;
855 }
856 btrfs_start_ordered_extent(inode, ordered, 1);
857 end = ordered->file_offset;
858 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
859 ret = -EIO;
860 btrfs_put_ordered_extent(ordered);
861 if (ret || end == 0 || end == start)
862 break;
863 end--;
864 }
865 return ret_wb ? ret_wb : ret;
866 }
867
868 /*
869 * find an ordered extent corresponding to file_offset. return NULL if
870 * nothing is found, otherwise take a reference on the extent and return it
871 */
872 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
873 u64 file_offset)
874 {
875 struct btrfs_ordered_inode_tree *tree;
876 struct rb_node *node;
877 struct btrfs_ordered_extent *entry = NULL;
878
879 tree = &BTRFS_I(inode)->ordered_tree;
880 spin_lock_irq(&tree->lock);
881 node = tree_search(tree, file_offset);
882 if (!node)
883 goto out;
884
885 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
886 if (!offset_in_entry(entry, file_offset))
887 entry = NULL;
888 if (entry)
889 refcount_inc(&entry->refs);
890 out:
891 spin_unlock_irq(&tree->lock);
892 return entry;
893 }
894
895 /* Since the DIO code tries to lock a wide area we need to look for any ordered
896 * extents that exist in the range, rather than just the start of the range.
897 */
898 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
899 struct btrfs_inode *inode, u64 file_offset, u64 len)
900 {
901 struct btrfs_ordered_inode_tree *tree;
902 struct rb_node *node;
903 struct btrfs_ordered_extent *entry = NULL;
904
905 tree = &inode->ordered_tree;
906 spin_lock_irq(&tree->lock);
907 node = tree_search(tree, file_offset);
908 if (!node) {
909 node = tree_search(tree, file_offset + len);
910 if (!node)
911 goto out;
912 }
913
914 while (1) {
915 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
916 if (range_overlaps(entry, file_offset, len))
917 break;
918
919 if (entry->file_offset >= file_offset + len) {
920 entry = NULL;
921 break;
922 }
923 entry = NULL;
924 node = rb_next(node);
925 if (!node)
926 break;
927 }
928 out:
929 if (entry)
930 refcount_inc(&entry->refs);
931 spin_unlock_irq(&tree->lock);
932 return entry;
933 }
934
935 bool btrfs_have_ordered_extents_in_range(struct inode *inode,
936 u64 file_offset,
937 u64 len)
938 {
939 struct btrfs_ordered_extent *oe;
940
941 oe = btrfs_lookup_ordered_range(BTRFS_I(inode), file_offset, len);
942 if (oe) {
943 btrfs_put_ordered_extent(oe);
944 return true;
945 }
946 return false;
947 }
948
949 /*
950 * lookup and return any extent before 'file_offset'. NULL is returned
951 * if none is found
952 */
953 struct btrfs_ordered_extent *
954 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
955 {
956 struct btrfs_ordered_inode_tree *tree;
957 struct rb_node *node;
958 struct btrfs_ordered_extent *entry = NULL;
959
960 tree = &BTRFS_I(inode)->ordered_tree;
961 spin_lock_irq(&tree->lock);
962 node = tree_search(tree, file_offset);
963 if (!node)
964 goto out;
965
966 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
967 refcount_inc(&entry->refs);
968 out:
969 spin_unlock_irq(&tree->lock);
970 return entry;
971 }
972
973 /*
974 * After an extent is done, call this to conditionally update the on disk
975 * i_size. i_size is updated to cover any fully written part of the file.
976 */
977 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
978 struct btrfs_ordered_extent *ordered)
979 {
980 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
981 u64 disk_i_size;
982 u64 new_i_size;
983 u64 i_size = i_size_read(inode);
984 struct rb_node *node;
985 struct rb_node *prev = NULL;
986 struct btrfs_ordered_extent *test;
987 int ret = 1;
988 u64 orig_offset = offset;
989
990 spin_lock_irq(&tree->lock);
991 if (ordered) {
992 offset = entry_end(ordered);
993 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
994 offset = min(offset,
995 ordered->file_offset +
996 ordered->truncated_len);
997 } else {
998 offset = ALIGN(offset, btrfs_inode_sectorsize(inode));
999 }
1000 disk_i_size = BTRFS_I(inode)->disk_i_size;
1001
1002 /*
1003 * truncate file.
1004 * If ordered is not NULL, then this is called from endio and
1005 * disk_i_size will be updated by either truncate itself or any
1006 * in-flight IOs which are inside the disk_i_size.
1007 *
1008 * Because btrfs_setsize() may set i_size with disk_i_size if truncate
1009 * fails somehow, we need to make sure we have a precise disk_i_size by
1010 * updating it as usual.
1011 *
1012 */
1013 if (!ordered && disk_i_size > i_size) {
1014 BTRFS_I(inode)->disk_i_size = orig_offset;
1015 ret = 0;
1016 goto out;
1017 }
1018
1019 /*
1020 * if the disk i_size is already at the inode->i_size, or
1021 * this ordered extent is inside the disk i_size, we're done
1022 */
1023 if (disk_i_size == i_size)
1024 goto out;
1025
1026 /*
1027 * We still need to update disk_i_size if outstanding_isize is greater
1028 * than disk_i_size.
1029 */
1030 if (offset <= disk_i_size &&
1031 (!ordered || ordered->outstanding_isize <= disk_i_size))
1032 goto out;
1033
1034 /*
1035 * walk backward from this ordered extent to disk_i_size.
1036 * if we find an ordered extent then we can't update disk i_size
1037 * yet
1038 */
1039 if (ordered) {
1040 node = rb_prev(&ordered->rb_node);
1041 } else {
1042 prev = tree_search(tree, offset);
1043 /*
1044 * we insert file extents without involving ordered struct,
1045 * so there should be no ordered struct cover this offset
1046 */
1047 if (prev) {
1048 test = rb_entry(prev, struct btrfs_ordered_extent,
1049 rb_node);
1050 BUG_ON(offset_in_entry(test, offset));
1051 }
1052 node = prev;
1053 }
1054 for (; node; node = rb_prev(node)) {
1055 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1056
1057 /* We treat this entry as if it doesn't exist */
1058 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1059 continue;
1060
1061 if (entry_end(test) <= disk_i_size)
1062 break;
1063 if (test->file_offset >= i_size)
1064 break;
1065
1066 /*
1067 * We don't update disk_i_size now, so record this undealt
1068 * i_size. Or we will not know the real i_size.
1069 */
1070 if (test->outstanding_isize < offset)
1071 test->outstanding_isize = offset;
1072 if (ordered &&
1073 ordered->outstanding_isize > test->outstanding_isize)
1074 test->outstanding_isize = ordered->outstanding_isize;
1075 goto out;
1076 }
1077 new_i_size = min_t(u64, offset, i_size);
1078
1079 /*
1080 * Some ordered extents may completed before the current one, and
1081 * we hold the real i_size in ->outstanding_isize.
1082 */
1083 if (ordered && ordered->outstanding_isize > new_i_size)
1084 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
1085 BTRFS_I(inode)->disk_i_size = new_i_size;
1086 ret = 0;
1087 out:
1088 /*
1089 * We need to do this because we can't remove ordered extents until
1090 * after the i_disk_size has been updated and then the inode has been
1091 * updated to reflect the change, so we need to tell anybody who finds
1092 * this ordered extent that we've already done all the real work, we
1093 * just haven't completed all the other work.
1094 */
1095 if (ordered)
1096 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1097 spin_unlock_irq(&tree->lock);
1098 return ret;
1099 }
1100
1101 /*
1102 * search the ordered extents for one corresponding to 'offset' and
1103 * try to find a checksum. This is used because we allow pages to
1104 * be reclaimed before their checksum is actually put into the btree
1105 */
1106 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1107 u32 *sum, int len)
1108 {
1109 struct btrfs_ordered_sum *ordered_sum;
1110 struct btrfs_ordered_extent *ordered;
1111 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1112 unsigned long num_sectors;
1113 unsigned long i;
1114 u32 sectorsize = btrfs_inode_sectorsize(inode);
1115 int index = 0;
1116
1117 ordered = btrfs_lookup_ordered_extent(inode, offset);
1118 if (!ordered)
1119 return 0;
1120
1121 spin_lock_irq(&tree->lock);
1122 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1123 if (disk_bytenr >= ordered_sum->bytenr &&
1124 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1125 i = (disk_bytenr - ordered_sum->bytenr) >>
1126 inode->i_sb->s_blocksize_bits;
1127 num_sectors = ordered_sum->len >>
1128 inode->i_sb->s_blocksize_bits;
1129 num_sectors = min_t(int, len - index, num_sectors - i);
1130 memcpy(sum + index, ordered_sum->sums + i,
1131 num_sectors);
1132
1133 index += (int)num_sectors;
1134 if (index == len)
1135 goto out;
1136 disk_bytenr += num_sectors * sectorsize;
1137 }
1138 }
1139 out:
1140 spin_unlock_irq(&tree->lock);
1141 btrfs_put_ordered_extent(ordered);
1142 return index;
1143 }
1144
1145 int __init ordered_data_init(void)
1146 {
1147 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1148 sizeof(struct btrfs_ordered_extent), 0,
1149 SLAB_MEM_SPREAD,
1150 NULL);
1151 if (!btrfs_ordered_extent_cache)
1152 return -ENOMEM;
1153
1154 return 0;
1155 }
1156
1157 void ordered_data_exit(void)
1158 {
1159 kmem_cache_destroy(btrfs_ordered_extent_cache);
1160 }