Loading...
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
30static struct kmem_cache *btrfs_ordered_extent_cache;
31
32static 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 */
42static 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
66static 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 */
78static 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 */
131static 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
139static 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 */
152static 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 */
185static 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 atomic_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 return 0;
246}
247
248int 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
256int 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
264int 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 */
278void 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 */
302int 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_fs_info *fs_info = btrfs_sb(inode->i_sb);
307 struct btrfs_ordered_inode_tree *tree;
308 struct rb_node *node;
309 struct btrfs_ordered_extent *entry = NULL;
310 int ret;
311 unsigned long flags;
312 u64 dec_end;
313 u64 dec_start;
314 u64 to_dec;
315
316 tree = &BTRFS_I(inode)->ordered_tree;
317 spin_lock_irqsave(&tree->lock, flags);
318 node = tree_search(tree, *file_offset);
319 if (!node) {
320 ret = 1;
321 goto out;
322 }
323
324 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
325 if (!offset_in_entry(entry, *file_offset)) {
326 ret = 1;
327 goto out;
328 }
329
330 dec_start = max(*file_offset, entry->file_offset);
331 dec_end = min(*file_offset + io_size, entry->file_offset +
332 entry->len);
333 *file_offset = dec_end;
334 if (dec_start > dec_end) {
335 btrfs_crit(fs_info, "bad ordering dec_start %llu end %llu",
336 dec_start, dec_end);
337 }
338 to_dec = dec_end - dec_start;
339 if (to_dec > entry->bytes_left) {
340 btrfs_crit(fs_info,
341 "bad ordered accounting left %llu size %llu",
342 entry->bytes_left, to_dec);
343 }
344 entry->bytes_left -= to_dec;
345 if (!uptodate)
346 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
347
348 if (entry->bytes_left == 0) {
349 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
350 /*
351 * Implicit memory barrier after test_and_set_bit
352 */
353 if (waitqueue_active(&entry->wait))
354 wake_up(&entry->wait);
355 } else {
356 ret = 1;
357 }
358out:
359 if (!ret && cached && entry) {
360 *cached = entry;
361 atomic_inc(&entry->refs);
362 }
363 spin_unlock_irqrestore(&tree->lock, flags);
364 return ret == 0;
365}
366
367/*
368 * this is used to account for finished IO across a given range
369 * of the file. The IO should not span ordered extents. If
370 * a given ordered_extent is completely done, 1 is returned, otherwise
371 * 0.
372 *
373 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
374 * to make sure this function only returns 1 once for a given ordered extent.
375 */
376int btrfs_dec_test_ordered_pending(struct inode *inode,
377 struct btrfs_ordered_extent **cached,
378 u64 file_offset, u64 io_size, int uptodate)
379{
380 struct btrfs_ordered_inode_tree *tree;
381 struct rb_node *node;
382 struct btrfs_ordered_extent *entry = NULL;
383 unsigned long flags;
384 int ret;
385
386 tree = &BTRFS_I(inode)->ordered_tree;
387 spin_lock_irqsave(&tree->lock, flags);
388 if (cached && *cached) {
389 entry = *cached;
390 goto have_entry;
391 }
392
393 node = tree_search(tree, file_offset);
394 if (!node) {
395 ret = 1;
396 goto out;
397 }
398
399 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
400have_entry:
401 if (!offset_in_entry(entry, file_offset)) {
402 ret = 1;
403 goto out;
404 }
405
406 if (io_size > entry->bytes_left) {
407 btrfs_crit(BTRFS_I(inode)->root->fs_info,
408 "bad ordered accounting left %llu size %llu",
409 entry->bytes_left, io_size);
410 }
411 entry->bytes_left -= io_size;
412 if (!uptodate)
413 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
414
415 if (entry->bytes_left == 0) {
416 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
417 /*
418 * Implicit memory barrier after test_and_set_bit
419 */
420 if (waitqueue_active(&entry->wait))
421 wake_up(&entry->wait);
422 } else {
423 ret = 1;
424 }
425out:
426 if (!ret && cached && entry) {
427 *cached = entry;
428 atomic_inc(&entry->refs);
429 }
430 spin_unlock_irqrestore(&tree->lock, flags);
431 return ret == 0;
432}
433
434/* Needs to either be called under a log transaction or the log_mutex */
435void btrfs_get_logged_extents(struct inode *inode,
436 struct list_head *logged_list,
437 const loff_t start,
438 const loff_t end)
439{
440 struct btrfs_ordered_inode_tree *tree;
441 struct btrfs_ordered_extent *ordered;
442 struct rb_node *n;
443 struct rb_node *prev;
444
445 tree = &BTRFS_I(inode)->ordered_tree;
446 spin_lock_irq(&tree->lock);
447 n = __tree_search(&tree->tree, end, &prev);
448 if (!n)
449 n = prev;
450 for (; n; n = rb_prev(n)) {
451 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
452 if (ordered->file_offset > end)
453 continue;
454 if (entry_end(ordered) <= start)
455 break;
456 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
457 continue;
458 list_add(&ordered->log_list, logged_list);
459 atomic_inc(&ordered->refs);
460 }
461 spin_unlock_irq(&tree->lock);
462}
463
464void btrfs_put_logged_extents(struct list_head *logged_list)
465{
466 struct btrfs_ordered_extent *ordered;
467
468 while (!list_empty(logged_list)) {
469 ordered = list_first_entry(logged_list,
470 struct btrfs_ordered_extent,
471 log_list);
472 list_del_init(&ordered->log_list);
473 btrfs_put_ordered_extent(ordered);
474 }
475}
476
477void btrfs_submit_logged_extents(struct list_head *logged_list,
478 struct btrfs_root *log)
479{
480 int index = log->log_transid % 2;
481
482 spin_lock_irq(&log->log_extents_lock[index]);
483 list_splice_tail(logged_list, &log->logged_list[index]);
484 spin_unlock_irq(&log->log_extents_lock[index]);
485}
486
487void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
488 struct btrfs_root *log, u64 transid)
489{
490 struct btrfs_ordered_extent *ordered;
491 int index = transid % 2;
492
493 spin_lock_irq(&log->log_extents_lock[index]);
494 while (!list_empty(&log->logged_list[index])) {
495 struct inode *inode;
496 ordered = list_first_entry(&log->logged_list[index],
497 struct btrfs_ordered_extent,
498 log_list);
499 list_del_init(&ordered->log_list);
500 inode = ordered->inode;
501 spin_unlock_irq(&log->log_extents_lock[index]);
502
503 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
504 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
505 u64 start = ordered->file_offset;
506 u64 end = ordered->file_offset + ordered->len - 1;
507
508 WARN_ON(!inode);
509 filemap_fdatawrite_range(inode->i_mapping, start, end);
510 }
511 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
512 &ordered->flags));
513
514 /*
515 * In order to keep us from losing our ordered extent
516 * information when committing the transaction we have to make
517 * sure that any logged extents are completed when we go to
518 * commit the transaction. To do this we simply increase the
519 * current transactions pending_ordered counter and decrement it
520 * when the ordered extent completes.
521 */
522 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
523 struct btrfs_ordered_inode_tree *tree;
524
525 tree = &BTRFS_I(inode)->ordered_tree;
526 spin_lock_irq(&tree->lock);
527 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
528 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
529 atomic_inc(&trans->transaction->pending_ordered);
530 }
531 spin_unlock_irq(&tree->lock);
532 }
533 btrfs_put_ordered_extent(ordered);
534 spin_lock_irq(&log->log_extents_lock[index]);
535 }
536 spin_unlock_irq(&log->log_extents_lock[index]);
537}
538
539void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
540{
541 struct btrfs_ordered_extent *ordered;
542 int index = transid % 2;
543
544 spin_lock_irq(&log->log_extents_lock[index]);
545 while (!list_empty(&log->logged_list[index])) {
546 ordered = list_first_entry(&log->logged_list[index],
547 struct btrfs_ordered_extent,
548 log_list);
549 list_del_init(&ordered->log_list);
550 spin_unlock_irq(&log->log_extents_lock[index]);
551 btrfs_put_ordered_extent(ordered);
552 spin_lock_irq(&log->log_extents_lock[index]);
553 }
554 spin_unlock_irq(&log->log_extents_lock[index]);
555}
556
557/*
558 * used to drop a reference on an ordered extent. This will free
559 * the extent if the last reference is dropped
560 */
561void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
562{
563 struct list_head *cur;
564 struct btrfs_ordered_sum *sum;
565
566 trace_btrfs_ordered_extent_put(entry->inode, entry);
567
568 if (atomic_dec_and_test(&entry->refs)) {
569 ASSERT(list_empty(&entry->log_list));
570 ASSERT(list_empty(&entry->trans_list));
571 ASSERT(list_empty(&entry->root_extent_list));
572 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
573 if (entry->inode)
574 btrfs_add_delayed_iput(entry->inode);
575 while (!list_empty(&entry->list)) {
576 cur = entry->list.next;
577 sum = list_entry(cur, struct btrfs_ordered_sum, list);
578 list_del(&sum->list);
579 kfree(sum);
580 }
581 kmem_cache_free(btrfs_ordered_extent_cache, entry);
582 }
583}
584
585/*
586 * remove an ordered extent from the tree. No references are dropped
587 * and waiters are woken up.
588 */
589void btrfs_remove_ordered_extent(struct inode *inode,
590 struct btrfs_ordered_extent *entry)
591{
592 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
593 struct btrfs_ordered_inode_tree *tree;
594 struct btrfs_root *root = BTRFS_I(inode)->root;
595 struct rb_node *node;
596 bool dec_pending_ordered = false;
597
598 tree = &BTRFS_I(inode)->ordered_tree;
599 spin_lock_irq(&tree->lock);
600 node = &entry->rb_node;
601 rb_erase(node, &tree->tree);
602 RB_CLEAR_NODE(node);
603 if (tree->last == node)
604 tree->last = NULL;
605 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
606 if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags))
607 dec_pending_ordered = true;
608 spin_unlock_irq(&tree->lock);
609
610 /*
611 * The current running transaction is waiting on us, we need to let it
612 * know that we're complete and wake it up.
613 */
614 if (dec_pending_ordered) {
615 struct btrfs_transaction *trans;
616
617 /*
618 * The checks for trans are just a formality, it should be set,
619 * but if it isn't we don't want to deref/assert under the spin
620 * lock, so be nice and check if trans is set, but ASSERT() so
621 * if it isn't set a developer will notice.
622 */
623 spin_lock(&fs_info->trans_lock);
624 trans = fs_info->running_transaction;
625 if (trans)
626 atomic_inc(&trans->use_count);
627 spin_unlock(&fs_info->trans_lock);
628
629 ASSERT(trans);
630 if (trans) {
631 if (atomic_dec_and_test(&trans->pending_ordered))
632 wake_up(&trans->pending_wait);
633 btrfs_put_transaction(trans);
634 }
635 }
636
637 spin_lock(&root->ordered_extent_lock);
638 list_del_init(&entry->root_extent_list);
639 root->nr_ordered_extents--;
640
641 trace_btrfs_ordered_extent_remove(inode, entry);
642
643 if (!root->nr_ordered_extents) {
644 spin_lock(&fs_info->ordered_root_lock);
645 BUG_ON(list_empty(&root->ordered_root));
646 list_del_init(&root->ordered_root);
647 spin_unlock(&fs_info->ordered_root_lock);
648 }
649 spin_unlock(&root->ordered_extent_lock);
650 wake_up(&entry->wait);
651}
652
653static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
654{
655 struct btrfs_ordered_extent *ordered;
656
657 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
658 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
659 complete(&ordered->completion);
660}
661
662/*
663 * wait for all the ordered extents in a root. This is done when balancing
664 * space between drives.
665 */
666int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr,
667 const u64 range_start, const u64 range_len)
668{
669 struct btrfs_fs_info *fs_info = root->fs_info;
670 LIST_HEAD(splice);
671 LIST_HEAD(skipped);
672 LIST_HEAD(works);
673 struct btrfs_ordered_extent *ordered, *next;
674 int count = 0;
675 const u64 range_end = range_start + range_len;
676
677 mutex_lock(&root->ordered_extent_mutex);
678 spin_lock(&root->ordered_extent_lock);
679 list_splice_init(&root->ordered_extents, &splice);
680 while (!list_empty(&splice) && nr) {
681 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
682 root_extent_list);
683
684 if (range_end <= ordered->start ||
685 ordered->start + ordered->disk_len <= range_start) {
686 list_move_tail(&ordered->root_extent_list, &skipped);
687 cond_resched_lock(&root->ordered_extent_lock);
688 continue;
689 }
690
691 list_move_tail(&ordered->root_extent_list,
692 &root->ordered_extents);
693 atomic_inc(&ordered->refs);
694 spin_unlock(&root->ordered_extent_lock);
695
696 btrfs_init_work(&ordered->flush_work,
697 btrfs_flush_delalloc_helper,
698 btrfs_run_ordered_extent_work, NULL, NULL);
699 list_add_tail(&ordered->work_list, &works);
700 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
701
702 cond_resched();
703 spin_lock(&root->ordered_extent_lock);
704 if (nr != -1)
705 nr--;
706 count++;
707 }
708 list_splice_tail(&skipped, &root->ordered_extents);
709 list_splice_tail(&splice, &root->ordered_extents);
710 spin_unlock(&root->ordered_extent_lock);
711
712 list_for_each_entry_safe(ordered, next, &works, work_list) {
713 list_del_init(&ordered->work_list);
714 wait_for_completion(&ordered->completion);
715 btrfs_put_ordered_extent(ordered);
716 cond_resched();
717 }
718 mutex_unlock(&root->ordered_extent_mutex);
719
720 return count;
721}
722
723int btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
724 const u64 range_start, const u64 range_len)
725{
726 struct btrfs_root *root;
727 struct list_head splice;
728 int done;
729 int total_done = 0;
730
731 INIT_LIST_HEAD(&splice);
732
733 mutex_lock(&fs_info->ordered_operations_mutex);
734 spin_lock(&fs_info->ordered_root_lock);
735 list_splice_init(&fs_info->ordered_roots, &splice);
736 while (!list_empty(&splice) && nr) {
737 root = list_first_entry(&splice, struct btrfs_root,
738 ordered_root);
739 root = btrfs_grab_fs_root(root);
740 BUG_ON(!root);
741 list_move_tail(&root->ordered_root,
742 &fs_info->ordered_roots);
743 spin_unlock(&fs_info->ordered_root_lock);
744
745 done = btrfs_wait_ordered_extents(root, nr,
746 range_start, range_len);
747 btrfs_put_fs_root(root);
748 total_done += done;
749
750 spin_lock(&fs_info->ordered_root_lock);
751 if (nr != -1) {
752 nr -= done;
753 WARN_ON(nr < 0);
754 }
755 }
756 list_splice_tail(&splice, &fs_info->ordered_roots);
757 spin_unlock(&fs_info->ordered_root_lock);
758 mutex_unlock(&fs_info->ordered_operations_mutex);
759
760 return total_done;
761}
762
763/*
764 * Used to start IO or wait for a given ordered extent to finish.
765 *
766 * If wait is one, this effectively waits on page writeback for all the pages
767 * in the extent, and it waits on the io completion code to insert
768 * metadata into the btree corresponding to the extent
769 */
770void btrfs_start_ordered_extent(struct inode *inode,
771 struct btrfs_ordered_extent *entry,
772 int wait)
773{
774 u64 start = entry->file_offset;
775 u64 end = start + entry->len - 1;
776
777 trace_btrfs_ordered_extent_start(inode, entry);
778
779 /*
780 * pages in the range can be dirty, clean or writeback. We
781 * start IO on any dirty ones so the wait doesn't stall waiting
782 * for the flusher thread to find them
783 */
784 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
785 filemap_fdatawrite_range(inode->i_mapping, start, end);
786 if (wait) {
787 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
788 &entry->flags));
789 }
790}
791
792/*
793 * Used to wait on ordered extents across a large range of bytes.
794 */
795int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
796{
797 int ret = 0;
798 int ret_wb = 0;
799 u64 end;
800 u64 orig_end;
801 struct btrfs_ordered_extent *ordered;
802
803 if (start + len < start) {
804 orig_end = INT_LIMIT(loff_t);
805 } else {
806 orig_end = start + len - 1;
807 if (orig_end > INT_LIMIT(loff_t))
808 orig_end = INT_LIMIT(loff_t);
809 }
810
811 /* start IO across the range first to instantiate any delalloc
812 * extents
813 */
814 ret = btrfs_fdatawrite_range(inode, start, orig_end);
815 if (ret)
816 return ret;
817
818 /*
819 * If we have a writeback error don't return immediately. Wait first
820 * for any ordered extents that haven't completed yet. This is to make
821 * sure no one can dirty the same page ranges and call writepages()
822 * before the ordered extents complete - to avoid failures (-EEXIST)
823 * when adding the new ordered extents to the ordered tree.
824 */
825 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
826
827 end = orig_end;
828 while (1) {
829 ordered = btrfs_lookup_first_ordered_extent(inode, end);
830 if (!ordered)
831 break;
832 if (ordered->file_offset > orig_end) {
833 btrfs_put_ordered_extent(ordered);
834 break;
835 }
836 if (ordered->file_offset + ordered->len <= start) {
837 btrfs_put_ordered_extent(ordered);
838 break;
839 }
840 btrfs_start_ordered_extent(inode, ordered, 1);
841 end = ordered->file_offset;
842 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
843 ret = -EIO;
844 btrfs_put_ordered_extent(ordered);
845 if (ret || end == 0 || end == start)
846 break;
847 end--;
848 }
849 return ret_wb ? ret_wb : ret;
850}
851
852/*
853 * find an ordered extent corresponding to file_offset. return NULL if
854 * nothing is found, otherwise take a reference on the extent and return it
855 */
856struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
857 u64 file_offset)
858{
859 struct btrfs_ordered_inode_tree *tree;
860 struct rb_node *node;
861 struct btrfs_ordered_extent *entry = NULL;
862
863 tree = &BTRFS_I(inode)->ordered_tree;
864 spin_lock_irq(&tree->lock);
865 node = tree_search(tree, file_offset);
866 if (!node)
867 goto out;
868
869 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
870 if (!offset_in_entry(entry, file_offset))
871 entry = NULL;
872 if (entry)
873 atomic_inc(&entry->refs);
874out:
875 spin_unlock_irq(&tree->lock);
876 return entry;
877}
878
879/* Since the DIO code tries to lock a wide area we need to look for any ordered
880 * extents that exist in the range, rather than just the start of the range.
881 */
882struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
883 u64 file_offset,
884 u64 len)
885{
886 struct btrfs_ordered_inode_tree *tree;
887 struct rb_node *node;
888 struct btrfs_ordered_extent *entry = NULL;
889
890 tree = &BTRFS_I(inode)->ordered_tree;
891 spin_lock_irq(&tree->lock);
892 node = tree_search(tree, file_offset);
893 if (!node) {
894 node = tree_search(tree, file_offset + len);
895 if (!node)
896 goto out;
897 }
898
899 while (1) {
900 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
901 if (range_overlaps(entry, file_offset, len))
902 break;
903
904 if (entry->file_offset >= file_offset + len) {
905 entry = NULL;
906 break;
907 }
908 entry = NULL;
909 node = rb_next(node);
910 if (!node)
911 break;
912 }
913out:
914 if (entry)
915 atomic_inc(&entry->refs);
916 spin_unlock_irq(&tree->lock);
917 return entry;
918}
919
920bool btrfs_have_ordered_extents_in_range(struct inode *inode,
921 u64 file_offset,
922 u64 len)
923{
924 struct btrfs_ordered_extent *oe;
925
926 oe = btrfs_lookup_ordered_range(inode, file_offset, len);
927 if (oe) {
928 btrfs_put_ordered_extent(oe);
929 return true;
930 }
931 return false;
932}
933
934/*
935 * lookup and return any extent before 'file_offset'. NULL is returned
936 * if none is found
937 */
938struct btrfs_ordered_extent *
939btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
940{
941 struct btrfs_ordered_inode_tree *tree;
942 struct rb_node *node;
943 struct btrfs_ordered_extent *entry = NULL;
944
945 tree = &BTRFS_I(inode)->ordered_tree;
946 spin_lock_irq(&tree->lock);
947 node = tree_search(tree, file_offset);
948 if (!node)
949 goto out;
950
951 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
952 atomic_inc(&entry->refs);
953out:
954 spin_unlock_irq(&tree->lock);
955 return entry;
956}
957
958/*
959 * After an extent is done, call this to conditionally update the on disk
960 * i_size. i_size is updated to cover any fully written part of the file.
961 */
962int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
963 struct btrfs_ordered_extent *ordered)
964{
965 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
966 u64 disk_i_size;
967 u64 new_i_size;
968 u64 i_size = i_size_read(inode);
969 struct rb_node *node;
970 struct rb_node *prev = NULL;
971 struct btrfs_ordered_extent *test;
972 int ret = 1;
973 u64 orig_offset = offset;
974
975 spin_lock_irq(&tree->lock);
976 if (ordered) {
977 offset = entry_end(ordered);
978 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
979 offset = min(offset,
980 ordered->file_offset +
981 ordered->truncated_len);
982 } else {
983 offset = ALIGN(offset, btrfs_inode_sectorsize(inode));
984 }
985 disk_i_size = BTRFS_I(inode)->disk_i_size;
986
987 /* truncate file */
988 if (disk_i_size > i_size) {
989 BTRFS_I(inode)->disk_i_size = orig_offset;
990 ret = 0;
991 goto out;
992 }
993
994 /*
995 * if the disk i_size is already at the inode->i_size, or
996 * this ordered extent is inside the disk i_size, we're done
997 */
998 if (disk_i_size == i_size)
999 goto out;
1000
1001 /*
1002 * We still need to update disk_i_size if outstanding_isize is greater
1003 * than disk_i_size.
1004 */
1005 if (offset <= disk_i_size &&
1006 (!ordered || ordered->outstanding_isize <= disk_i_size))
1007 goto out;
1008
1009 /*
1010 * walk backward from this ordered extent to disk_i_size.
1011 * if we find an ordered extent then we can't update disk i_size
1012 * yet
1013 */
1014 if (ordered) {
1015 node = rb_prev(&ordered->rb_node);
1016 } else {
1017 prev = tree_search(tree, offset);
1018 /*
1019 * we insert file extents without involving ordered struct,
1020 * so there should be no ordered struct cover this offset
1021 */
1022 if (prev) {
1023 test = rb_entry(prev, struct btrfs_ordered_extent,
1024 rb_node);
1025 BUG_ON(offset_in_entry(test, offset));
1026 }
1027 node = prev;
1028 }
1029 for (; node; node = rb_prev(node)) {
1030 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1031
1032 /* We treat this entry as if it doesn't exist */
1033 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1034 continue;
1035 if (test->file_offset + test->len <= disk_i_size)
1036 break;
1037 if (test->file_offset >= i_size)
1038 break;
1039 if (entry_end(test) > disk_i_size) {
1040 /*
1041 * we don't update disk_i_size now, so record this
1042 * undealt i_size. Or we will not know the real
1043 * i_size.
1044 */
1045 if (test->outstanding_isize < offset)
1046 test->outstanding_isize = offset;
1047 if (ordered &&
1048 ordered->outstanding_isize >
1049 test->outstanding_isize)
1050 test->outstanding_isize =
1051 ordered->outstanding_isize;
1052 goto out;
1053 }
1054 }
1055 new_i_size = min_t(u64, offset, i_size);
1056
1057 /*
1058 * Some ordered extents may completed before the current one, and
1059 * we hold the real i_size in ->outstanding_isize.
1060 */
1061 if (ordered && ordered->outstanding_isize > new_i_size)
1062 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
1063 BTRFS_I(inode)->disk_i_size = new_i_size;
1064 ret = 0;
1065out:
1066 /*
1067 * We need to do this because we can't remove ordered extents until
1068 * after the i_disk_size has been updated and then the inode has been
1069 * updated to reflect the change, so we need to tell anybody who finds
1070 * this ordered extent that we've already done all the real work, we
1071 * just haven't completed all the other work.
1072 */
1073 if (ordered)
1074 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1075 spin_unlock_irq(&tree->lock);
1076 return ret;
1077}
1078
1079/*
1080 * search the ordered extents for one corresponding to 'offset' and
1081 * try to find a checksum. This is used because we allow pages to
1082 * be reclaimed before their checksum is actually put into the btree
1083 */
1084int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1085 u32 *sum, int len)
1086{
1087 struct btrfs_ordered_sum *ordered_sum;
1088 struct btrfs_ordered_extent *ordered;
1089 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1090 unsigned long num_sectors;
1091 unsigned long i;
1092 u32 sectorsize = btrfs_inode_sectorsize(inode);
1093 int index = 0;
1094
1095 ordered = btrfs_lookup_ordered_extent(inode, offset);
1096 if (!ordered)
1097 return 0;
1098
1099 spin_lock_irq(&tree->lock);
1100 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1101 if (disk_bytenr >= ordered_sum->bytenr &&
1102 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1103 i = (disk_bytenr - ordered_sum->bytenr) >>
1104 inode->i_sb->s_blocksize_bits;
1105 num_sectors = ordered_sum->len >>
1106 inode->i_sb->s_blocksize_bits;
1107 num_sectors = min_t(int, len - index, num_sectors - i);
1108 memcpy(sum + index, ordered_sum->sums + i,
1109 num_sectors);
1110
1111 index += (int)num_sectors;
1112 if (index == len)
1113 goto out;
1114 disk_bytenr += num_sectors * sectorsize;
1115 }
1116 }
1117out:
1118 spin_unlock_irq(&tree->lock);
1119 btrfs_put_ordered_extent(ordered);
1120 return index;
1121}
1122
1123int __init ordered_data_init(void)
1124{
1125 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1126 sizeof(struct btrfs_ordered_extent), 0,
1127 SLAB_MEM_SPREAD,
1128 NULL);
1129 if (!btrfs_ordered_extent_cache)
1130 return -ENOMEM;
1131
1132 return 0;
1133}
1134
1135void ordered_data_exit(void)
1136{
1137 kmem_cache_destroy(btrfs_ordered_extent_cache);
1138}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/slab.h>
7#include <linux/blkdev.h>
8#include <linux/writeback.h>
9#include <linux/sched/mm.h>
10#include "misc.h"
11#include "ctree.h"
12#include "transaction.h"
13#include "btrfs_inode.h"
14#include "extent_io.h"
15#include "disk-io.h"
16#include "compression.h"
17#include "delalloc-space.h"
18#include "qgroup.h"
19#include "subpage.h"
20
21static struct kmem_cache *btrfs_ordered_extent_cache;
22
23static u64 entry_end(struct btrfs_ordered_extent *entry)
24{
25 if (entry->file_offset + entry->num_bytes < entry->file_offset)
26 return (u64)-1;
27 return entry->file_offset + entry->num_bytes;
28}
29
30/* returns NULL if the insertion worked, or it returns the node it did find
31 * in the tree
32 */
33static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
34 struct rb_node *node)
35{
36 struct rb_node **p = &root->rb_node;
37 struct rb_node *parent = NULL;
38 struct btrfs_ordered_extent *entry;
39
40 while (*p) {
41 parent = *p;
42 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
43
44 if (file_offset < entry->file_offset)
45 p = &(*p)->rb_left;
46 else if (file_offset >= entry_end(entry))
47 p = &(*p)->rb_right;
48 else
49 return parent;
50 }
51
52 rb_link_node(node, parent, p);
53 rb_insert_color(node, root);
54 return NULL;
55}
56
57/*
58 * look for a given offset in the tree, and if it can't be found return the
59 * first lesser offset
60 */
61static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
62 struct rb_node **prev_ret)
63{
64 struct rb_node *n = root->rb_node;
65 struct rb_node *prev = NULL;
66 struct rb_node *test;
67 struct btrfs_ordered_extent *entry;
68 struct btrfs_ordered_extent *prev_entry = NULL;
69
70 while (n) {
71 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
72 prev = n;
73 prev_entry = entry;
74
75 if (file_offset < entry->file_offset)
76 n = n->rb_left;
77 else if (file_offset >= entry_end(entry))
78 n = n->rb_right;
79 else
80 return n;
81 }
82 if (!prev_ret)
83 return NULL;
84
85 while (prev && file_offset >= entry_end(prev_entry)) {
86 test = rb_next(prev);
87 if (!test)
88 break;
89 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
90 rb_node);
91 if (file_offset < entry_end(prev_entry))
92 break;
93
94 prev = test;
95 }
96 if (prev)
97 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
98 rb_node);
99 while (prev && file_offset < entry_end(prev_entry)) {
100 test = rb_prev(prev);
101 if (!test)
102 break;
103 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
104 rb_node);
105 prev = test;
106 }
107 *prev_ret = prev;
108 return NULL;
109}
110
111static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
112 u64 len)
113{
114 if (file_offset + len <= entry->file_offset ||
115 entry->file_offset + entry->num_bytes <= file_offset)
116 return 0;
117 return 1;
118}
119
120/*
121 * look find the first ordered struct that has this offset, otherwise
122 * the first one less than this offset
123 */
124static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
125 u64 file_offset)
126{
127 struct rb_root *root = &tree->tree;
128 struct rb_node *prev = NULL;
129 struct rb_node *ret;
130 struct btrfs_ordered_extent *entry;
131
132 if (tree->last) {
133 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
134 rb_node);
135 if (in_range(file_offset, entry->file_offset, entry->num_bytes))
136 return tree->last;
137 }
138 ret = __tree_search(root, file_offset, &prev);
139 if (!ret)
140 ret = prev;
141 if (ret)
142 tree->last = ret;
143 return ret;
144}
145
146/*
147 * Allocate and add a new ordered_extent into the per-inode tree.
148 *
149 * The tree is given a single reference on the ordered extent that was
150 * inserted.
151 */
152static int __btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
153 u64 disk_bytenr, u64 num_bytes,
154 u64 disk_num_bytes, int type, int dio,
155 int compress_type)
156{
157 struct btrfs_root *root = inode->root;
158 struct btrfs_fs_info *fs_info = root->fs_info;
159 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
160 struct rb_node *node;
161 struct btrfs_ordered_extent *entry;
162 int ret;
163
164 if (type == BTRFS_ORDERED_NOCOW || type == BTRFS_ORDERED_PREALLOC) {
165 /* For nocow write, we can release the qgroup rsv right now */
166 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
167 if (ret < 0)
168 return ret;
169 ret = 0;
170 } else {
171 /*
172 * The ordered extent has reserved qgroup space, release now
173 * and pass the reserved number for qgroup_record to free.
174 */
175 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
176 if (ret < 0)
177 return ret;
178 }
179 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
180 if (!entry)
181 return -ENOMEM;
182
183 entry->file_offset = file_offset;
184 entry->disk_bytenr = disk_bytenr;
185 entry->num_bytes = num_bytes;
186 entry->disk_num_bytes = disk_num_bytes;
187 entry->bytes_left = num_bytes;
188 entry->inode = igrab(&inode->vfs_inode);
189 entry->compress_type = compress_type;
190 entry->truncated_len = (u64)-1;
191 entry->qgroup_rsv = ret;
192 entry->physical = (u64)-1;
193
194 ASSERT(type == BTRFS_ORDERED_REGULAR ||
195 type == BTRFS_ORDERED_NOCOW ||
196 type == BTRFS_ORDERED_PREALLOC ||
197 type == BTRFS_ORDERED_COMPRESSED);
198 set_bit(type, &entry->flags);
199
200 percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
201 fs_info->delalloc_batch);
202
203 if (dio)
204 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
205
206 /* one ref for the tree */
207 refcount_set(&entry->refs, 1);
208 init_waitqueue_head(&entry->wait);
209 INIT_LIST_HEAD(&entry->list);
210 INIT_LIST_HEAD(&entry->log_list);
211 INIT_LIST_HEAD(&entry->root_extent_list);
212 INIT_LIST_HEAD(&entry->work_list);
213 init_completion(&entry->completion);
214
215 trace_btrfs_ordered_extent_add(inode, entry);
216
217 spin_lock_irq(&tree->lock);
218 node = tree_insert(&tree->tree, file_offset,
219 &entry->rb_node);
220 if (node)
221 btrfs_panic(fs_info, -EEXIST,
222 "inconsistency in ordered tree at offset %llu",
223 file_offset);
224 spin_unlock_irq(&tree->lock);
225
226 spin_lock(&root->ordered_extent_lock);
227 list_add_tail(&entry->root_extent_list,
228 &root->ordered_extents);
229 root->nr_ordered_extents++;
230 if (root->nr_ordered_extents == 1) {
231 spin_lock(&fs_info->ordered_root_lock);
232 BUG_ON(!list_empty(&root->ordered_root));
233 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
234 spin_unlock(&fs_info->ordered_root_lock);
235 }
236 spin_unlock(&root->ordered_extent_lock);
237
238 /*
239 * We don't need the count_max_extents here, we can assume that all of
240 * that work has been done at higher layers, so this is truly the
241 * smallest the extent is going to get.
242 */
243 spin_lock(&inode->lock);
244 btrfs_mod_outstanding_extents(inode, 1);
245 spin_unlock(&inode->lock);
246
247 return 0;
248}
249
250int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
251 u64 disk_bytenr, u64 num_bytes, u64 disk_num_bytes,
252 int type)
253{
254 ASSERT(type == BTRFS_ORDERED_REGULAR ||
255 type == BTRFS_ORDERED_NOCOW ||
256 type == BTRFS_ORDERED_PREALLOC);
257 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
258 num_bytes, disk_num_bytes, type, 0,
259 BTRFS_COMPRESS_NONE);
260}
261
262int btrfs_add_ordered_extent_dio(struct btrfs_inode *inode, u64 file_offset,
263 u64 disk_bytenr, u64 num_bytes,
264 u64 disk_num_bytes, int type)
265{
266 ASSERT(type == BTRFS_ORDERED_REGULAR ||
267 type == BTRFS_ORDERED_NOCOW ||
268 type == BTRFS_ORDERED_PREALLOC);
269 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
270 num_bytes, disk_num_bytes, type, 1,
271 BTRFS_COMPRESS_NONE);
272}
273
274int btrfs_add_ordered_extent_compress(struct btrfs_inode *inode, u64 file_offset,
275 u64 disk_bytenr, u64 num_bytes,
276 u64 disk_num_bytes, int compress_type)
277{
278 ASSERT(compress_type != BTRFS_COMPRESS_NONE);
279 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
280 num_bytes, disk_num_bytes,
281 BTRFS_ORDERED_COMPRESSED, 0,
282 compress_type);
283}
284
285/*
286 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
287 * when an ordered extent is finished. If the list covers more than one
288 * ordered extent, it is split across multiples.
289 */
290void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
291 struct btrfs_ordered_sum *sum)
292{
293 struct btrfs_ordered_inode_tree *tree;
294
295 tree = &BTRFS_I(entry->inode)->ordered_tree;
296 spin_lock_irq(&tree->lock);
297 list_add_tail(&sum->list, &entry->list);
298 spin_unlock_irq(&tree->lock);
299}
300
301/*
302 * Mark all ordered extents io inside the specified range finished.
303 *
304 * @page: The invovled page for the opeartion.
305 * For uncompressed buffered IO, the page status also needs to be
306 * updated to indicate whether the pending ordered io is finished.
307 * Can be NULL for direct IO and compressed write.
308 * For these cases, callers are ensured they won't execute the
309 * endio function twice.
310 * @finish_func: The function to be executed when all the IO of an ordered
311 * extent are finished.
312 *
313 * This function is called for endio, thus the range must have ordered
314 * extent(s) coveri it.
315 */
316void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
317 struct page *page, u64 file_offset,
318 u64 num_bytes, btrfs_func_t finish_func,
319 bool uptodate)
320{
321 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
322 struct btrfs_fs_info *fs_info = inode->root->fs_info;
323 struct btrfs_workqueue *wq;
324 struct rb_node *node;
325 struct btrfs_ordered_extent *entry = NULL;
326 unsigned long flags;
327 u64 cur = file_offset;
328
329 if (btrfs_is_free_space_inode(inode))
330 wq = fs_info->endio_freespace_worker;
331 else
332 wq = fs_info->endio_write_workers;
333
334 if (page)
335 ASSERT(page->mapping && page_offset(page) <= file_offset &&
336 file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
337
338 spin_lock_irqsave(&tree->lock, flags);
339 while (cur < file_offset + num_bytes) {
340 u64 entry_end;
341 u64 end;
342 u32 len;
343
344 node = tree_search(tree, cur);
345 /* No ordered extents at all */
346 if (!node)
347 break;
348
349 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
350 entry_end = entry->file_offset + entry->num_bytes;
351 /*
352 * |<-- OE --->| |
353 * cur
354 * Go to next OE.
355 */
356 if (cur >= entry_end) {
357 node = rb_next(node);
358 /* No more ordered extents, exit */
359 if (!node)
360 break;
361 entry = rb_entry(node, struct btrfs_ordered_extent,
362 rb_node);
363
364 /* Go to next ordered extent and continue */
365 cur = entry->file_offset;
366 continue;
367 }
368 /*
369 * | |<--- OE --->|
370 * cur
371 * Go to the start of OE.
372 */
373 if (cur < entry->file_offset) {
374 cur = entry->file_offset;
375 continue;
376 }
377
378 /*
379 * Now we are definitely inside one ordered extent.
380 *
381 * |<--- OE --->|
382 * |
383 * cur
384 */
385 end = min(entry->file_offset + entry->num_bytes,
386 file_offset + num_bytes) - 1;
387 ASSERT(end + 1 - cur < U32_MAX);
388 len = end + 1 - cur;
389
390 if (page) {
391 /*
392 * Ordered (Private2) bit indicates whether we still
393 * have pending io unfinished for the ordered extent.
394 *
395 * If there's no such bit, we need to skip to next range.
396 */
397 if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
398 cur += len;
399 continue;
400 }
401 btrfs_page_clear_ordered(fs_info, page, cur, len);
402 }
403
404 /* Now we're fine to update the accounting */
405 if (unlikely(len > entry->bytes_left)) {
406 WARN_ON(1);
407 btrfs_crit(fs_info,
408"bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
409 inode->root->root_key.objectid,
410 btrfs_ino(inode),
411 entry->file_offset,
412 entry->num_bytes,
413 len, entry->bytes_left);
414 entry->bytes_left = 0;
415 } else {
416 entry->bytes_left -= len;
417 }
418
419 if (!uptodate)
420 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
421
422 /*
423 * All the IO of the ordered extent is finished, we need to queue
424 * the finish_func to be executed.
425 */
426 if (entry->bytes_left == 0) {
427 set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
428 cond_wake_up(&entry->wait);
429 refcount_inc(&entry->refs);
430 spin_unlock_irqrestore(&tree->lock, flags);
431 btrfs_init_work(&entry->work, finish_func, NULL, NULL);
432 btrfs_queue_work(wq, &entry->work);
433 spin_lock_irqsave(&tree->lock, flags);
434 }
435 cur += len;
436 }
437 spin_unlock_irqrestore(&tree->lock, flags);
438}
439
440/*
441 * Finish IO for one ordered extent across a given range. The range can only
442 * contain one ordered extent.
443 *
444 * @cached: The cached ordered extent. If not NULL, we can skip the tree
445 * search and use the ordered extent directly.
446 * Will be also used to store the finished ordered extent.
447 * @file_offset: File offset for the finished IO
448 * @io_size: Length of the finish IO range
449 * @uptodate: If the IO finishes without problem
450 *
451 * Return true if the ordered extent is finished in the range, and update
452 * @cached.
453 * Return false otherwise.
454 *
455 * NOTE: The range can NOT cross multiple ordered extents.
456 * Thus caller should ensure the range doesn't cross ordered extents.
457 */
458bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
459 struct btrfs_ordered_extent **cached,
460 u64 file_offset, u64 io_size, int uptodate)
461{
462 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
463 struct rb_node *node;
464 struct btrfs_ordered_extent *entry = NULL;
465 unsigned long flags;
466 bool finished = false;
467
468 spin_lock_irqsave(&tree->lock, flags);
469 if (cached && *cached) {
470 entry = *cached;
471 goto have_entry;
472 }
473
474 node = tree_search(tree, file_offset);
475 if (!node)
476 goto out;
477
478 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
479have_entry:
480 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
481 goto out;
482
483 if (io_size > entry->bytes_left)
484 btrfs_crit(inode->root->fs_info,
485 "bad ordered accounting left %llu size %llu",
486 entry->bytes_left, io_size);
487
488 entry->bytes_left -= io_size;
489 if (!uptodate)
490 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
491
492 if (entry->bytes_left == 0) {
493 /*
494 * Ensure only one caller can set the flag and finished_ret
495 * accordingly
496 */
497 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
498 /* test_and_set_bit implies a barrier */
499 cond_wake_up_nomb(&entry->wait);
500 }
501out:
502 if (finished && cached && entry) {
503 *cached = entry;
504 refcount_inc(&entry->refs);
505 }
506 spin_unlock_irqrestore(&tree->lock, flags);
507 return finished;
508}
509
510/*
511 * used to drop a reference on an ordered extent. This will free
512 * the extent if the last reference is dropped
513 */
514void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
515{
516 struct list_head *cur;
517 struct btrfs_ordered_sum *sum;
518
519 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
520
521 if (refcount_dec_and_test(&entry->refs)) {
522 ASSERT(list_empty(&entry->root_extent_list));
523 ASSERT(list_empty(&entry->log_list));
524 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
525 if (entry->inode)
526 btrfs_add_delayed_iput(entry->inode);
527 while (!list_empty(&entry->list)) {
528 cur = entry->list.next;
529 sum = list_entry(cur, struct btrfs_ordered_sum, list);
530 list_del(&sum->list);
531 kvfree(sum);
532 }
533 kmem_cache_free(btrfs_ordered_extent_cache, entry);
534 }
535}
536
537/*
538 * remove an ordered extent from the tree. No references are dropped
539 * and waiters are woken up.
540 */
541void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
542 struct btrfs_ordered_extent *entry)
543{
544 struct btrfs_ordered_inode_tree *tree;
545 struct btrfs_root *root = btrfs_inode->root;
546 struct btrfs_fs_info *fs_info = root->fs_info;
547 struct rb_node *node;
548 bool pending;
549
550 /* This is paired with btrfs_add_ordered_extent. */
551 spin_lock(&btrfs_inode->lock);
552 btrfs_mod_outstanding_extents(btrfs_inode, -1);
553 spin_unlock(&btrfs_inode->lock);
554 if (root != fs_info->tree_root)
555 btrfs_delalloc_release_metadata(btrfs_inode, entry->num_bytes,
556 false);
557
558 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
559 fs_info->delalloc_batch);
560
561 tree = &btrfs_inode->ordered_tree;
562 spin_lock_irq(&tree->lock);
563 node = &entry->rb_node;
564 rb_erase(node, &tree->tree);
565 RB_CLEAR_NODE(node);
566 if (tree->last == node)
567 tree->last = NULL;
568 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
569 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
570 spin_unlock_irq(&tree->lock);
571
572 /*
573 * The current running transaction is waiting on us, we need to let it
574 * know that we're complete and wake it up.
575 */
576 if (pending) {
577 struct btrfs_transaction *trans;
578
579 /*
580 * The checks for trans are just a formality, it should be set,
581 * but if it isn't we don't want to deref/assert under the spin
582 * lock, so be nice and check if trans is set, but ASSERT() so
583 * if it isn't set a developer will notice.
584 */
585 spin_lock(&fs_info->trans_lock);
586 trans = fs_info->running_transaction;
587 if (trans)
588 refcount_inc(&trans->use_count);
589 spin_unlock(&fs_info->trans_lock);
590
591 ASSERT(trans);
592 if (trans) {
593 if (atomic_dec_and_test(&trans->pending_ordered))
594 wake_up(&trans->pending_wait);
595 btrfs_put_transaction(trans);
596 }
597 }
598
599 spin_lock(&root->ordered_extent_lock);
600 list_del_init(&entry->root_extent_list);
601 root->nr_ordered_extents--;
602
603 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
604
605 if (!root->nr_ordered_extents) {
606 spin_lock(&fs_info->ordered_root_lock);
607 BUG_ON(list_empty(&root->ordered_root));
608 list_del_init(&root->ordered_root);
609 spin_unlock(&fs_info->ordered_root_lock);
610 }
611 spin_unlock(&root->ordered_extent_lock);
612 wake_up(&entry->wait);
613}
614
615static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
616{
617 struct btrfs_ordered_extent *ordered;
618
619 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
620 btrfs_start_ordered_extent(ordered, 1);
621 complete(&ordered->completion);
622}
623
624/*
625 * wait for all the ordered extents in a root. This is done when balancing
626 * space between drives.
627 */
628u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
629 const u64 range_start, const u64 range_len)
630{
631 struct btrfs_fs_info *fs_info = root->fs_info;
632 LIST_HEAD(splice);
633 LIST_HEAD(skipped);
634 LIST_HEAD(works);
635 struct btrfs_ordered_extent *ordered, *next;
636 u64 count = 0;
637 const u64 range_end = range_start + range_len;
638
639 mutex_lock(&root->ordered_extent_mutex);
640 spin_lock(&root->ordered_extent_lock);
641 list_splice_init(&root->ordered_extents, &splice);
642 while (!list_empty(&splice) && nr) {
643 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
644 root_extent_list);
645
646 if (range_end <= ordered->disk_bytenr ||
647 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
648 list_move_tail(&ordered->root_extent_list, &skipped);
649 cond_resched_lock(&root->ordered_extent_lock);
650 continue;
651 }
652
653 list_move_tail(&ordered->root_extent_list,
654 &root->ordered_extents);
655 refcount_inc(&ordered->refs);
656 spin_unlock(&root->ordered_extent_lock);
657
658 btrfs_init_work(&ordered->flush_work,
659 btrfs_run_ordered_extent_work, NULL, NULL);
660 list_add_tail(&ordered->work_list, &works);
661 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
662
663 cond_resched();
664 spin_lock(&root->ordered_extent_lock);
665 if (nr != U64_MAX)
666 nr--;
667 count++;
668 }
669 list_splice_tail(&skipped, &root->ordered_extents);
670 list_splice_tail(&splice, &root->ordered_extents);
671 spin_unlock(&root->ordered_extent_lock);
672
673 list_for_each_entry_safe(ordered, next, &works, work_list) {
674 list_del_init(&ordered->work_list);
675 wait_for_completion(&ordered->completion);
676 btrfs_put_ordered_extent(ordered);
677 cond_resched();
678 }
679 mutex_unlock(&root->ordered_extent_mutex);
680
681 return count;
682}
683
684void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
685 const u64 range_start, const u64 range_len)
686{
687 struct btrfs_root *root;
688 struct list_head splice;
689 u64 done;
690
691 INIT_LIST_HEAD(&splice);
692
693 mutex_lock(&fs_info->ordered_operations_mutex);
694 spin_lock(&fs_info->ordered_root_lock);
695 list_splice_init(&fs_info->ordered_roots, &splice);
696 while (!list_empty(&splice) && nr) {
697 root = list_first_entry(&splice, struct btrfs_root,
698 ordered_root);
699 root = btrfs_grab_root(root);
700 BUG_ON(!root);
701 list_move_tail(&root->ordered_root,
702 &fs_info->ordered_roots);
703 spin_unlock(&fs_info->ordered_root_lock);
704
705 done = btrfs_wait_ordered_extents(root, nr,
706 range_start, range_len);
707 btrfs_put_root(root);
708
709 spin_lock(&fs_info->ordered_root_lock);
710 if (nr != U64_MAX) {
711 nr -= done;
712 }
713 }
714 list_splice_tail(&splice, &fs_info->ordered_roots);
715 spin_unlock(&fs_info->ordered_root_lock);
716 mutex_unlock(&fs_info->ordered_operations_mutex);
717}
718
719/*
720 * Used to start IO or wait for a given ordered extent to finish.
721 *
722 * If wait is one, this effectively waits on page writeback for all the pages
723 * in the extent, and it waits on the io completion code to insert
724 * metadata into the btree corresponding to the extent
725 */
726void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry, int wait)
727{
728 u64 start = entry->file_offset;
729 u64 end = start + entry->num_bytes - 1;
730 struct btrfs_inode *inode = BTRFS_I(entry->inode);
731
732 trace_btrfs_ordered_extent_start(inode, entry);
733
734 /*
735 * pages in the range can be dirty, clean or writeback. We
736 * start IO on any dirty ones so the wait doesn't stall waiting
737 * for the flusher thread to find them
738 */
739 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
740 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
741 if (wait) {
742 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
743 &entry->flags));
744 }
745}
746
747/*
748 * Used to wait on ordered extents across a large range of bytes.
749 */
750int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
751{
752 int ret = 0;
753 int ret_wb = 0;
754 u64 end;
755 u64 orig_end;
756 struct btrfs_ordered_extent *ordered;
757
758 if (start + len < start) {
759 orig_end = INT_LIMIT(loff_t);
760 } else {
761 orig_end = start + len - 1;
762 if (orig_end > INT_LIMIT(loff_t))
763 orig_end = INT_LIMIT(loff_t);
764 }
765
766 /* start IO across the range first to instantiate any delalloc
767 * extents
768 */
769 ret = btrfs_fdatawrite_range(inode, start, orig_end);
770 if (ret)
771 return ret;
772
773 /*
774 * If we have a writeback error don't return immediately. Wait first
775 * for any ordered extents that haven't completed yet. This is to make
776 * sure no one can dirty the same page ranges and call writepages()
777 * before the ordered extents complete - to avoid failures (-EEXIST)
778 * when adding the new ordered extents to the ordered tree.
779 */
780 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
781
782 end = orig_end;
783 while (1) {
784 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
785 if (!ordered)
786 break;
787 if (ordered->file_offset > orig_end) {
788 btrfs_put_ordered_extent(ordered);
789 break;
790 }
791 if (ordered->file_offset + ordered->num_bytes <= start) {
792 btrfs_put_ordered_extent(ordered);
793 break;
794 }
795 btrfs_start_ordered_extent(ordered, 1);
796 end = ordered->file_offset;
797 /*
798 * If the ordered extent had an error save the error but don't
799 * exit without waiting first for all other ordered extents in
800 * the range to complete.
801 */
802 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
803 ret = -EIO;
804 btrfs_put_ordered_extent(ordered);
805 if (end == 0 || end == start)
806 break;
807 end--;
808 }
809 return ret_wb ? ret_wb : ret;
810}
811
812/*
813 * find an ordered extent corresponding to file_offset. return NULL if
814 * nothing is found, otherwise take a reference on the extent and return it
815 */
816struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
817 u64 file_offset)
818{
819 struct btrfs_ordered_inode_tree *tree;
820 struct rb_node *node;
821 struct btrfs_ordered_extent *entry = NULL;
822 unsigned long flags;
823
824 tree = &inode->ordered_tree;
825 spin_lock_irqsave(&tree->lock, flags);
826 node = tree_search(tree, file_offset);
827 if (!node)
828 goto out;
829
830 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
831 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
832 entry = NULL;
833 if (entry)
834 refcount_inc(&entry->refs);
835out:
836 spin_unlock_irqrestore(&tree->lock, flags);
837 return entry;
838}
839
840/* Since the DIO code tries to lock a wide area we need to look for any ordered
841 * extents that exist in the range, rather than just the start of the range.
842 */
843struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
844 struct btrfs_inode *inode, u64 file_offset, u64 len)
845{
846 struct btrfs_ordered_inode_tree *tree;
847 struct rb_node *node;
848 struct btrfs_ordered_extent *entry = NULL;
849
850 tree = &inode->ordered_tree;
851 spin_lock_irq(&tree->lock);
852 node = tree_search(tree, file_offset);
853 if (!node) {
854 node = tree_search(tree, file_offset + len);
855 if (!node)
856 goto out;
857 }
858
859 while (1) {
860 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
861 if (range_overlaps(entry, file_offset, len))
862 break;
863
864 if (entry->file_offset >= file_offset + len) {
865 entry = NULL;
866 break;
867 }
868 entry = NULL;
869 node = rb_next(node);
870 if (!node)
871 break;
872 }
873out:
874 if (entry)
875 refcount_inc(&entry->refs);
876 spin_unlock_irq(&tree->lock);
877 return entry;
878}
879
880/*
881 * Adds all ordered extents to the given list. The list ends up sorted by the
882 * file_offset of the ordered extents.
883 */
884void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
885 struct list_head *list)
886{
887 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
888 struct rb_node *n;
889
890 ASSERT(inode_is_locked(&inode->vfs_inode));
891
892 spin_lock_irq(&tree->lock);
893 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
894 struct btrfs_ordered_extent *ordered;
895
896 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
897
898 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
899 continue;
900
901 ASSERT(list_empty(&ordered->log_list));
902 list_add_tail(&ordered->log_list, list);
903 refcount_inc(&ordered->refs);
904 }
905 spin_unlock_irq(&tree->lock);
906}
907
908/*
909 * lookup and return any extent before 'file_offset'. NULL is returned
910 * if none is found
911 */
912struct btrfs_ordered_extent *
913btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
914{
915 struct btrfs_ordered_inode_tree *tree;
916 struct rb_node *node;
917 struct btrfs_ordered_extent *entry = NULL;
918
919 tree = &inode->ordered_tree;
920 spin_lock_irq(&tree->lock);
921 node = tree_search(tree, file_offset);
922 if (!node)
923 goto out;
924
925 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
926 refcount_inc(&entry->refs);
927out:
928 spin_unlock_irq(&tree->lock);
929 return entry;
930}
931
932/*
933 * Lookup the first ordered extent that overlaps the range
934 * [@file_offset, @file_offset + @len).
935 *
936 * The difference between this and btrfs_lookup_first_ordered_extent() is
937 * that this one won't return any ordered extent that does not overlap the range.
938 * And the difference against btrfs_lookup_ordered_extent() is, this function
939 * ensures the first ordered extent gets returned.
940 */
941struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
942 struct btrfs_inode *inode, u64 file_offset, u64 len)
943{
944 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
945 struct rb_node *node;
946 struct rb_node *cur;
947 struct rb_node *prev;
948 struct rb_node *next;
949 struct btrfs_ordered_extent *entry = NULL;
950
951 spin_lock_irq(&tree->lock);
952 node = tree->tree.rb_node;
953 /*
954 * Here we don't want to use tree_search() which will use tree->last
955 * and screw up the search order.
956 * And __tree_search() can't return the adjacent ordered extents
957 * either, thus here we do our own search.
958 */
959 while (node) {
960 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
961
962 if (file_offset < entry->file_offset) {
963 node = node->rb_left;
964 } else if (file_offset >= entry_end(entry)) {
965 node = node->rb_right;
966 } else {
967 /*
968 * Direct hit, got an ordered extent that starts at
969 * @file_offset
970 */
971 goto out;
972 }
973 }
974 if (!entry) {
975 /* Empty tree */
976 goto out;
977 }
978
979 cur = &entry->rb_node;
980 /* We got an entry around @file_offset, check adjacent entries */
981 if (entry->file_offset < file_offset) {
982 prev = cur;
983 next = rb_next(cur);
984 } else {
985 prev = rb_prev(cur);
986 next = cur;
987 }
988 if (prev) {
989 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
990 if (range_overlaps(entry, file_offset, len))
991 goto out;
992 }
993 if (next) {
994 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
995 if (range_overlaps(entry, file_offset, len))
996 goto out;
997 }
998 /* No ordered extent in the range */
999 entry = NULL;
1000out:
1001 if (entry)
1002 refcount_inc(&entry->refs);
1003 spin_unlock_irq(&tree->lock);
1004 return entry;
1005}
1006
1007/*
1008 * btrfs_flush_ordered_range - Lock the passed range and ensures all pending
1009 * ordered extents in it are run to completion.
1010 *
1011 * @inode: Inode whose ordered tree is to be searched
1012 * @start: Beginning of range to flush
1013 * @end: Last byte of range to lock
1014 * @cached_state: If passed, will return the extent state responsible for the
1015 * locked range. It's the caller's responsibility to free the cached state.
1016 *
1017 * This function always returns with the given range locked, ensuring after it's
1018 * called no order extent can be pending.
1019 */
1020void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1021 u64 end,
1022 struct extent_state **cached_state)
1023{
1024 struct btrfs_ordered_extent *ordered;
1025 struct extent_state *cache = NULL;
1026 struct extent_state **cachedp = &cache;
1027
1028 if (cached_state)
1029 cachedp = cached_state;
1030
1031 while (1) {
1032 lock_extent_bits(&inode->io_tree, start, end, cachedp);
1033 ordered = btrfs_lookup_ordered_range(inode, start,
1034 end - start + 1);
1035 if (!ordered) {
1036 /*
1037 * If no external cached_state has been passed then
1038 * decrement the extra ref taken for cachedp since we
1039 * aren't exposing it outside of this function
1040 */
1041 if (!cached_state)
1042 refcount_dec(&cache->refs);
1043 break;
1044 }
1045 unlock_extent_cached(&inode->io_tree, start, end, cachedp);
1046 btrfs_start_ordered_extent(ordered, 1);
1047 btrfs_put_ordered_extent(ordered);
1048 }
1049}
1050
1051static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
1052 u64 len)
1053{
1054 struct inode *inode = ordered->inode;
1055 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1056 u64 file_offset = ordered->file_offset + pos;
1057 u64 disk_bytenr = ordered->disk_bytenr + pos;
1058 u64 num_bytes = len;
1059 u64 disk_num_bytes = len;
1060 int type;
1061 unsigned long flags_masked = ordered->flags & ~(1 << BTRFS_ORDERED_DIRECT);
1062 int compress_type = ordered->compress_type;
1063 unsigned long weight;
1064 int ret;
1065
1066 weight = hweight_long(flags_masked);
1067 WARN_ON_ONCE(weight > 1);
1068 if (!weight)
1069 type = 0;
1070 else
1071 type = __ffs(flags_masked);
1072
1073 /*
1074 * The splitting extent is already counted and will be added again
1075 * in btrfs_add_ordered_extent_*(). Subtract num_bytes to avoid
1076 * double counting.
1077 */
1078 percpu_counter_add_batch(&fs_info->ordered_bytes, -num_bytes,
1079 fs_info->delalloc_batch);
1080 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered->flags)) {
1081 WARN_ON_ONCE(1);
1082 ret = btrfs_add_ordered_extent_compress(BTRFS_I(inode),
1083 file_offset, disk_bytenr, num_bytes,
1084 disk_num_bytes, compress_type);
1085 } else if (test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
1086 ret = btrfs_add_ordered_extent_dio(BTRFS_I(inode), file_offset,
1087 disk_bytenr, num_bytes, disk_num_bytes, type);
1088 } else {
1089 ret = btrfs_add_ordered_extent(BTRFS_I(inode), file_offset,
1090 disk_bytenr, num_bytes, disk_num_bytes, type);
1091 }
1092
1093 return ret;
1094}
1095
1096int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
1097 u64 post)
1098{
1099 struct inode *inode = ordered->inode;
1100 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1101 struct rb_node *node;
1102 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1103 int ret = 0;
1104
1105 spin_lock_irq(&tree->lock);
1106 /* Remove from tree once */
1107 node = &ordered->rb_node;
1108 rb_erase(node, &tree->tree);
1109 RB_CLEAR_NODE(node);
1110 if (tree->last == node)
1111 tree->last = NULL;
1112
1113 ordered->file_offset += pre;
1114 ordered->disk_bytenr += pre;
1115 ordered->num_bytes -= (pre + post);
1116 ordered->disk_num_bytes -= (pre + post);
1117 ordered->bytes_left -= (pre + post);
1118
1119 /* Re-insert the node */
1120 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1121 if (node)
1122 btrfs_panic(fs_info, -EEXIST,
1123 "zoned: inconsistency in ordered tree at offset %llu",
1124 ordered->file_offset);
1125
1126 spin_unlock_irq(&tree->lock);
1127
1128 if (pre)
1129 ret = clone_ordered_extent(ordered, 0, pre);
1130 if (ret == 0 && post)
1131 ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
1132 post);
1133
1134 return ret;
1135}
1136
1137int __init ordered_data_init(void)
1138{
1139 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1140 sizeof(struct btrfs_ordered_extent), 0,
1141 SLAB_MEM_SPREAD,
1142 NULL);
1143 if (!btrfs_ordered_extent_cache)
1144 return -ENOMEM;
1145
1146 return 0;
1147}
1148
1149void __cold ordered_data_exit(void)
1150{
1151 kmem_cache_destroy(btrfs_ordered_extent_cache);
1152}