Loading...
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 "messages.h"
11#include "misc.h"
12#include "ctree.h"
13#include "transaction.h"
14#include "btrfs_inode.h"
15#include "extent_io.h"
16#include "disk-io.h"
17#include "compression.h"
18#include "delalloc-space.h"
19#include "qgroup.h"
20#include "subpage.h"
21#include "file.h"
22#include "super.h"
23
24static struct kmem_cache *btrfs_ordered_extent_cache;
25
26static u64 entry_end(struct btrfs_ordered_extent *entry)
27{
28 if (entry->file_offset + entry->num_bytes < entry->file_offset)
29 return (u64)-1;
30 return entry->file_offset + entry->num_bytes;
31}
32
33/* returns NULL if the insertion worked, or it returns the node it did find
34 * in the tree
35 */
36static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
37 struct rb_node *node)
38{
39 struct rb_node **p = &root->rb_node;
40 struct rb_node *parent = NULL;
41 struct btrfs_ordered_extent *entry;
42
43 while (*p) {
44 parent = *p;
45 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
46
47 if (file_offset < entry->file_offset)
48 p = &(*p)->rb_left;
49 else if (file_offset >= entry_end(entry))
50 p = &(*p)->rb_right;
51 else
52 return parent;
53 }
54
55 rb_link_node(node, parent, p);
56 rb_insert_color(node, root);
57 return NULL;
58}
59
60/*
61 * look for a given offset in the tree, and if it can't be found return the
62 * first lesser offset
63 */
64static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
65 struct rb_node **prev_ret)
66{
67 struct rb_node *n = root->rb_node;
68 struct rb_node *prev = NULL;
69 struct rb_node *test;
70 struct btrfs_ordered_extent *entry;
71 struct btrfs_ordered_extent *prev_entry = NULL;
72
73 while (n) {
74 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
75 prev = n;
76 prev_entry = entry;
77
78 if (file_offset < entry->file_offset)
79 n = n->rb_left;
80 else if (file_offset >= entry_end(entry))
81 n = n->rb_right;
82 else
83 return n;
84 }
85 if (!prev_ret)
86 return NULL;
87
88 while (prev && file_offset >= entry_end(prev_entry)) {
89 test = rb_next(prev);
90 if (!test)
91 break;
92 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
93 rb_node);
94 if (file_offset < entry_end(prev_entry))
95 break;
96
97 prev = test;
98 }
99 if (prev)
100 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
101 rb_node);
102 while (prev && file_offset < entry_end(prev_entry)) {
103 test = rb_prev(prev);
104 if (!test)
105 break;
106 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
107 rb_node);
108 prev = test;
109 }
110 *prev_ret = prev;
111 return NULL;
112}
113
114static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
115 u64 len)
116{
117 if (file_offset + len <= entry->file_offset ||
118 entry->file_offset + entry->num_bytes <= file_offset)
119 return 0;
120 return 1;
121}
122
123/*
124 * look find the first ordered struct that has this offset, otherwise
125 * the first one less than this offset
126 */
127static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
128 u64 file_offset)
129{
130 struct rb_root *root = &tree->tree;
131 struct rb_node *prev = NULL;
132 struct rb_node *ret;
133 struct btrfs_ordered_extent *entry;
134
135 if (tree->last) {
136 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
137 rb_node);
138 if (in_range(file_offset, entry->file_offset, entry->num_bytes))
139 return tree->last;
140 }
141 ret = __tree_search(root, file_offset, &prev);
142 if (!ret)
143 ret = prev;
144 if (ret)
145 tree->last = ret;
146 return ret;
147}
148
149/*
150 * Add an ordered extent to the per-inode tree.
151 *
152 * @inode: Inode that this extent is for.
153 * @file_offset: Logical offset in file where the extent starts.
154 * @num_bytes: Logical length of extent in file.
155 * @ram_bytes: Full length of unencoded data.
156 * @disk_bytenr: Offset of extent on disk.
157 * @disk_num_bytes: Size of extent on disk.
158 * @offset: Offset into unencoded data where file data starts.
159 * @flags: Flags specifying type of extent (1 << BTRFS_ORDERED_*).
160 * @compress_type: Compression algorithm used for data.
161 *
162 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
163 * tree is given a single reference on the ordered extent that was inserted.
164 *
165 * Return: 0 or -ENOMEM.
166 */
167int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
168 u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
169 u64 disk_num_bytes, u64 offset, unsigned flags,
170 int compress_type)
171{
172 struct btrfs_root *root = inode->root;
173 struct btrfs_fs_info *fs_info = root->fs_info;
174 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
175 struct rb_node *node;
176 struct btrfs_ordered_extent *entry;
177 int ret;
178
179 if (flags &
180 ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
181 /* For nocow write, we can release the qgroup rsv right now */
182 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
183 if (ret < 0)
184 return ret;
185 ret = 0;
186 } else {
187 /*
188 * The ordered extent has reserved qgroup space, release now
189 * and pass the reserved number for qgroup_record to free.
190 */
191 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
192 if (ret < 0)
193 return ret;
194 }
195 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
196 if (!entry)
197 return -ENOMEM;
198
199 entry->file_offset = file_offset;
200 entry->num_bytes = num_bytes;
201 entry->ram_bytes = ram_bytes;
202 entry->disk_bytenr = disk_bytenr;
203 entry->disk_num_bytes = disk_num_bytes;
204 entry->offset = offset;
205 entry->bytes_left = num_bytes;
206 entry->inode = igrab(&inode->vfs_inode);
207 entry->compress_type = compress_type;
208 entry->truncated_len = (u64)-1;
209 entry->qgroup_rsv = ret;
210 entry->physical = (u64)-1;
211
212 ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
213 entry->flags = flags;
214
215 percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
216 fs_info->delalloc_batch);
217
218 /* one ref for the tree */
219 refcount_set(&entry->refs, 1);
220 init_waitqueue_head(&entry->wait);
221 INIT_LIST_HEAD(&entry->list);
222 INIT_LIST_HEAD(&entry->log_list);
223 INIT_LIST_HEAD(&entry->root_extent_list);
224 INIT_LIST_HEAD(&entry->work_list);
225 init_completion(&entry->completion);
226
227 trace_btrfs_ordered_extent_add(inode, entry);
228
229 spin_lock_irq(&tree->lock);
230 node = tree_insert(&tree->tree, file_offset,
231 &entry->rb_node);
232 if (node)
233 btrfs_panic(fs_info, -EEXIST,
234 "inconsistency in ordered tree at offset %llu",
235 file_offset);
236 spin_unlock_irq(&tree->lock);
237
238 spin_lock(&root->ordered_extent_lock);
239 list_add_tail(&entry->root_extent_list,
240 &root->ordered_extents);
241 root->nr_ordered_extents++;
242 if (root->nr_ordered_extents == 1) {
243 spin_lock(&fs_info->ordered_root_lock);
244 BUG_ON(!list_empty(&root->ordered_root));
245 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
246 spin_unlock(&fs_info->ordered_root_lock);
247 }
248 spin_unlock(&root->ordered_extent_lock);
249
250 /*
251 * We don't need the count_max_extents here, we can assume that all of
252 * that work has been done at higher layers, so this is truly the
253 * smallest the extent is going to get.
254 */
255 spin_lock(&inode->lock);
256 btrfs_mod_outstanding_extents(inode, 1);
257 spin_unlock(&inode->lock);
258
259 return 0;
260}
261
262/*
263 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
264 * when an ordered extent is finished. If the list covers more than one
265 * ordered extent, it is split across multiples.
266 */
267void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
268 struct btrfs_ordered_sum *sum)
269{
270 struct btrfs_ordered_inode_tree *tree;
271
272 tree = &BTRFS_I(entry->inode)->ordered_tree;
273 spin_lock_irq(&tree->lock);
274 list_add_tail(&sum->list, &entry->list);
275 spin_unlock_irq(&tree->lock);
276}
277
278static void finish_ordered_fn(struct btrfs_work *work)
279{
280 struct btrfs_ordered_extent *ordered_extent;
281
282 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
283 btrfs_finish_ordered_io(ordered_extent);
284}
285
286/*
287 * Mark all ordered extents io inside the specified range finished.
288 *
289 * @page: The involved page for the operation.
290 * For uncompressed buffered IO, the page status also needs to be
291 * updated to indicate whether the pending ordered io is finished.
292 * Can be NULL for direct IO and compressed write.
293 * For these cases, callers are ensured they won't execute the
294 * endio function twice.
295 *
296 * This function is called for endio, thus the range must have ordered
297 * extent(s) covering it.
298 */
299void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
300 struct page *page, u64 file_offset,
301 u64 num_bytes, bool uptodate)
302{
303 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
304 struct btrfs_fs_info *fs_info = inode->root->fs_info;
305 struct btrfs_workqueue *wq;
306 struct rb_node *node;
307 struct btrfs_ordered_extent *entry = NULL;
308 unsigned long flags;
309 u64 cur = file_offset;
310
311 if (btrfs_is_free_space_inode(inode))
312 wq = fs_info->endio_freespace_worker;
313 else
314 wq = fs_info->endio_write_workers;
315
316 if (page)
317 ASSERT(page->mapping && page_offset(page) <= file_offset &&
318 file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
319
320 spin_lock_irqsave(&tree->lock, flags);
321 while (cur < file_offset + num_bytes) {
322 u64 entry_end;
323 u64 end;
324 u32 len;
325
326 node = tree_search(tree, cur);
327 /* No ordered extents at all */
328 if (!node)
329 break;
330
331 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
332 entry_end = entry->file_offset + entry->num_bytes;
333 /*
334 * |<-- OE --->| |
335 * cur
336 * Go to next OE.
337 */
338 if (cur >= entry_end) {
339 node = rb_next(node);
340 /* No more ordered extents, exit */
341 if (!node)
342 break;
343 entry = rb_entry(node, struct btrfs_ordered_extent,
344 rb_node);
345
346 /* Go to next ordered extent and continue */
347 cur = entry->file_offset;
348 continue;
349 }
350 /*
351 * | |<--- OE --->|
352 * cur
353 * Go to the start of OE.
354 */
355 if (cur < entry->file_offset) {
356 cur = entry->file_offset;
357 continue;
358 }
359
360 /*
361 * Now we are definitely inside one ordered extent.
362 *
363 * |<--- OE --->|
364 * |
365 * cur
366 */
367 end = min(entry->file_offset + entry->num_bytes,
368 file_offset + num_bytes) - 1;
369 ASSERT(end + 1 - cur < U32_MAX);
370 len = end + 1 - cur;
371
372 if (page) {
373 /*
374 * Ordered (Private2) bit indicates whether we still
375 * have pending io unfinished for the ordered extent.
376 *
377 * If there's no such bit, we need to skip to next range.
378 */
379 if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
380 cur += len;
381 continue;
382 }
383 btrfs_page_clear_ordered(fs_info, page, cur, len);
384 }
385
386 /* Now we're fine to update the accounting */
387 if (unlikely(len > entry->bytes_left)) {
388 WARN_ON(1);
389 btrfs_crit(fs_info,
390"bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
391 inode->root->root_key.objectid,
392 btrfs_ino(inode),
393 entry->file_offset,
394 entry->num_bytes,
395 len, entry->bytes_left);
396 entry->bytes_left = 0;
397 } else {
398 entry->bytes_left -= len;
399 }
400
401 if (!uptodate)
402 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
403
404 /*
405 * All the IO of the ordered extent is finished, we need to queue
406 * the finish_func to be executed.
407 */
408 if (entry->bytes_left == 0) {
409 set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
410 cond_wake_up(&entry->wait);
411 refcount_inc(&entry->refs);
412 trace_btrfs_ordered_extent_mark_finished(inode, entry);
413 spin_unlock_irqrestore(&tree->lock, flags);
414 btrfs_init_work(&entry->work, finish_ordered_fn, NULL, NULL);
415 btrfs_queue_work(wq, &entry->work);
416 spin_lock_irqsave(&tree->lock, flags);
417 }
418 cur += len;
419 }
420 spin_unlock_irqrestore(&tree->lock, flags);
421}
422
423/*
424 * Finish IO for one ordered extent across a given range. The range can only
425 * contain one ordered extent.
426 *
427 * @cached: The cached ordered extent. If not NULL, we can skip the tree
428 * search and use the ordered extent directly.
429 * Will be also used to store the finished ordered extent.
430 * @file_offset: File offset for the finished IO
431 * @io_size: Length of the finish IO range
432 *
433 * Return true if the ordered extent is finished in the range, and update
434 * @cached.
435 * Return false otherwise.
436 *
437 * NOTE: The range can NOT cross multiple ordered extents.
438 * Thus caller should ensure the range doesn't cross ordered extents.
439 */
440bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
441 struct btrfs_ordered_extent **cached,
442 u64 file_offset, u64 io_size)
443{
444 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
445 struct rb_node *node;
446 struct btrfs_ordered_extent *entry = NULL;
447 unsigned long flags;
448 bool finished = false;
449
450 spin_lock_irqsave(&tree->lock, flags);
451 if (cached && *cached) {
452 entry = *cached;
453 goto have_entry;
454 }
455
456 node = tree_search(tree, file_offset);
457 if (!node)
458 goto out;
459
460 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
461have_entry:
462 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
463 goto out;
464
465 if (io_size > entry->bytes_left)
466 btrfs_crit(inode->root->fs_info,
467 "bad ordered accounting left %llu size %llu",
468 entry->bytes_left, io_size);
469
470 entry->bytes_left -= io_size;
471
472 if (entry->bytes_left == 0) {
473 /*
474 * Ensure only one caller can set the flag and finished_ret
475 * accordingly
476 */
477 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
478 /* test_and_set_bit implies a barrier */
479 cond_wake_up_nomb(&entry->wait);
480 }
481out:
482 if (finished && cached && entry) {
483 *cached = entry;
484 refcount_inc(&entry->refs);
485 trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
486 }
487 spin_unlock_irqrestore(&tree->lock, flags);
488 return finished;
489}
490
491/*
492 * used to drop a reference on an ordered extent. This will free
493 * the extent if the last reference is dropped
494 */
495void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
496{
497 struct list_head *cur;
498 struct btrfs_ordered_sum *sum;
499
500 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
501
502 if (refcount_dec_and_test(&entry->refs)) {
503 ASSERT(list_empty(&entry->root_extent_list));
504 ASSERT(list_empty(&entry->log_list));
505 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
506 if (entry->inode)
507 btrfs_add_delayed_iput(BTRFS_I(entry->inode));
508 while (!list_empty(&entry->list)) {
509 cur = entry->list.next;
510 sum = list_entry(cur, struct btrfs_ordered_sum, list);
511 list_del(&sum->list);
512 kvfree(sum);
513 }
514 kmem_cache_free(btrfs_ordered_extent_cache, entry);
515 }
516}
517
518/*
519 * remove an ordered extent from the tree. No references are dropped
520 * and waiters are woken up.
521 */
522void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
523 struct btrfs_ordered_extent *entry)
524{
525 struct btrfs_ordered_inode_tree *tree;
526 struct btrfs_root *root = btrfs_inode->root;
527 struct btrfs_fs_info *fs_info = root->fs_info;
528 struct rb_node *node;
529 bool pending;
530 bool freespace_inode;
531
532 /*
533 * If this is a free space inode the thread has not acquired the ordered
534 * extents lockdep map.
535 */
536 freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
537
538 btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
539 /* This is paired with btrfs_add_ordered_extent. */
540 spin_lock(&btrfs_inode->lock);
541 btrfs_mod_outstanding_extents(btrfs_inode, -1);
542 spin_unlock(&btrfs_inode->lock);
543 if (root != fs_info->tree_root) {
544 u64 release;
545
546 if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
547 release = entry->disk_num_bytes;
548 else
549 release = entry->num_bytes;
550 btrfs_delalloc_release_metadata(btrfs_inode, release, false);
551 }
552
553 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
554 fs_info->delalloc_batch);
555
556 tree = &btrfs_inode->ordered_tree;
557 spin_lock_irq(&tree->lock);
558 node = &entry->rb_node;
559 rb_erase(node, &tree->tree);
560 RB_CLEAR_NODE(node);
561 if (tree->last == node)
562 tree->last = NULL;
563 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
564 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
565 spin_unlock_irq(&tree->lock);
566
567 /*
568 * The current running transaction is waiting on us, we need to let it
569 * know that we're complete and wake it up.
570 */
571 if (pending) {
572 struct btrfs_transaction *trans;
573
574 /*
575 * The checks for trans are just a formality, it should be set,
576 * but if it isn't we don't want to deref/assert under the spin
577 * lock, so be nice and check if trans is set, but ASSERT() so
578 * if it isn't set a developer will notice.
579 */
580 spin_lock(&fs_info->trans_lock);
581 trans = fs_info->running_transaction;
582 if (trans)
583 refcount_inc(&trans->use_count);
584 spin_unlock(&fs_info->trans_lock);
585
586 ASSERT(trans);
587 if (trans) {
588 if (atomic_dec_and_test(&trans->pending_ordered))
589 wake_up(&trans->pending_wait);
590 btrfs_put_transaction(trans);
591 }
592 }
593
594 btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered);
595
596 spin_lock(&root->ordered_extent_lock);
597 list_del_init(&entry->root_extent_list);
598 root->nr_ordered_extents--;
599
600 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
601
602 if (!root->nr_ordered_extents) {
603 spin_lock(&fs_info->ordered_root_lock);
604 BUG_ON(list_empty(&root->ordered_root));
605 list_del_init(&root->ordered_root);
606 spin_unlock(&fs_info->ordered_root_lock);
607 }
608 spin_unlock(&root->ordered_extent_lock);
609 wake_up(&entry->wait);
610 if (!freespace_inode)
611 btrfs_lockdep_release(fs_info, btrfs_ordered_extent);
612}
613
614static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
615{
616 struct btrfs_ordered_extent *ordered;
617
618 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
619 btrfs_start_ordered_extent(ordered, 1);
620 complete(&ordered->completion);
621}
622
623/*
624 * wait for all the ordered extents in a root. This is done when balancing
625 * space between drives.
626 */
627u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
628 const u64 range_start, const u64 range_len)
629{
630 struct btrfs_fs_info *fs_info = root->fs_info;
631 LIST_HEAD(splice);
632 LIST_HEAD(skipped);
633 LIST_HEAD(works);
634 struct btrfs_ordered_extent *ordered, *next;
635 u64 count = 0;
636 const u64 range_end = range_start + range_len;
637
638 mutex_lock(&root->ordered_extent_mutex);
639 spin_lock(&root->ordered_extent_lock);
640 list_splice_init(&root->ordered_extents, &splice);
641 while (!list_empty(&splice) && nr) {
642 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
643 root_extent_list);
644
645 if (range_end <= ordered->disk_bytenr ||
646 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
647 list_move_tail(&ordered->root_extent_list, &skipped);
648 cond_resched_lock(&root->ordered_extent_lock);
649 continue;
650 }
651
652 list_move_tail(&ordered->root_extent_list,
653 &root->ordered_extents);
654 refcount_inc(&ordered->refs);
655 spin_unlock(&root->ordered_extent_lock);
656
657 btrfs_init_work(&ordered->flush_work,
658 btrfs_run_ordered_extent_work, NULL, NULL);
659 list_add_tail(&ordered->work_list, &works);
660 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
661
662 cond_resched();
663 spin_lock(&root->ordered_extent_lock);
664 if (nr != U64_MAX)
665 nr--;
666 count++;
667 }
668 list_splice_tail(&skipped, &root->ordered_extents);
669 list_splice_tail(&splice, &root->ordered_extents);
670 spin_unlock(&root->ordered_extent_lock);
671
672 list_for_each_entry_safe(ordered, next, &works, work_list) {
673 list_del_init(&ordered->work_list);
674 wait_for_completion(&ordered->completion);
675 btrfs_put_ordered_extent(ordered);
676 cond_resched();
677 }
678 mutex_unlock(&root->ordered_extent_mutex);
679
680 return count;
681}
682
683void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
684 const u64 range_start, const u64 range_len)
685{
686 struct btrfs_root *root;
687 struct list_head splice;
688 u64 done;
689
690 INIT_LIST_HEAD(&splice);
691
692 mutex_lock(&fs_info->ordered_operations_mutex);
693 spin_lock(&fs_info->ordered_root_lock);
694 list_splice_init(&fs_info->ordered_roots, &splice);
695 while (!list_empty(&splice) && nr) {
696 root = list_first_entry(&splice, struct btrfs_root,
697 ordered_root);
698 root = btrfs_grab_root(root);
699 BUG_ON(!root);
700 list_move_tail(&root->ordered_root,
701 &fs_info->ordered_roots);
702 spin_unlock(&fs_info->ordered_root_lock);
703
704 done = btrfs_wait_ordered_extents(root, nr,
705 range_start, range_len);
706 btrfs_put_root(root);
707
708 spin_lock(&fs_info->ordered_root_lock);
709 if (nr != U64_MAX) {
710 nr -= done;
711 }
712 }
713 list_splice_tail(&splice, &fs_info->ordered_roots);
714 spin_unlock(&fs_info->ordered_root_lock);
715 mutex_unlock(&fs_info->ordered_operations_mutex);
716}
717
718/*
719 * Used to start IO or wait for a given ordered extent to finish.
720 *
721 * If wait is one, this effectively waits on page writeback for all the pages
722 * in the extent, and it waits on the io completion code to insert
723 * metadata into the btree corresponding to the extent
724 */
725void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry, int wait)
726{
727 u64 start = entry->file_offset;
728 u64 end = start + entry->num_bytes - 1;
729 struct btrfs_inode *inode = BTRFS_I(entry->inode);
730 bool freespace_inode;
731
732 trace_btrfs_ordered_extent_start(inode, entry);
733
734 /*
735 * If this is a free space inode do not take the ordered extents lockdep
736 * map.
737 */
738 freespace_inode = btrfs_is_free_space_inode(inode);
739
740 /*
741 * pages in the range can be dirty, clean or writeback. We
742 * start IO on any dirty ones so the wait doesn't stall waiting
743 * for the flusher thread to find them
744 */
745 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
746 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
747 if (wait) {
748 if (!freespace_inode)
749 btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent);
750 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
751 &entry->flags));
752 }
753}
754
755/*
756 * Used to wait on ordered extents across a large range of bytes.
757 */
758int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
759{
760 int ret = 0;
761 int ret_wb = 0;
762 u64 end;
763 u64 orig_end;
764 struct btrfs_ordered_extent *ordered;
765
766 if (start + len < start) {
767 orig_end = OFFSET_MAX;
768 } else {
769 orig_end = start + len - 1;
770 if (orig_end > OFFSET_MAX)
771 orig_end = OFFSET_MAX;
772 }
773
774 /* start IO across the range first to instantiate any delalloc
775 * extents
776 */
777 ret = btrfs_fdatawrite_range(inode, start, orig_end);
778 if (ret)
779 return ret;
780
781 /*
782 * If we have a writeback error don't return immediately. Wait first
783 * for any ordered extents that haven't completed yet. This is to make
784 * sure no one can dirty the same page ranges and call writepages()
785 * before the ordered extents complete - to avoid failures (-EEXIST)
786 * when adding the new ordered extents to the ordered tree.
787 */
788 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
789
790 end = orig_end;
791 while (1) {
792 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
793 if (!ordered)
794 break;
795 if (ordered->file_offset > orig_end) {
796 btrfs_put_ordered_extent(ordered);
797 break;
798 }
799 if (ordered->file_offset + ordered->num_bytes <= start) {
800 btrfs_put_ordered_extent(ordered);
801 break;
802 }
803 btrfs_start_ordered_extent(ordered, 1);
804 end = ordered->file_offset;
805 /*
806 * If the ordered extent had an error save the error but don't
807 * exit without waiting first for all other ordered extents in
808 * the range to complete.
809 */
810 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
811 ret = -EIO;
812 btrfs_put_ordered_extent(ordered);
813 if (end == 0 || end == start)
814 break;
815 end--;
816 }
817 return ret_wb ? ret_wb : ret;
818}
819
820/*
821 * find an ordered extent corresponding to file_offset. return NULL if
822 * nothing is found, otherwise take a reference on the extent and return it
823 */
824struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
825 u64 file_offset)
826{
827 struct btrfs_ordered_inode_tree *tree;
828 struct rb_node *node;
829 struct btrfs_ordered_extent *entry = NULL;
830 unsigned long flags;
831
832 tree = &inode->ordered_tree;
833 spin_lock_irqsave(&tree->lock, flags);
834 node = tree_search(tree, file_offset);
835 if (!node)
836 goto out;
837
838 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
839 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
840 entry = NULL;
841 if (entry) {
842 refcount_inc(&entry->refs);
843 trace_btrfs_ordered_extent_lookup(inode, entry);
844 }
845out:
846 spin_unlock_irqrestore(&tree->lock, flags);
847 return entry;
848}
849
850/* Since the DIO code tries to lock a wide area we need to look for any ordered
851 * extents that exist in the range, rather than just the start of the range.
852 */
853struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
854 struct btrfs_inode *inode, u64 file_offset, u64 len)
855{
856 struct btrfs_ordered_inode_tree *tree;
857 struct rb_node *node;
858 struct btrfs_ordered_extent *entry = NULL;
859
860 tree = &inode->ordered_tree;
861 spin_lock_irq(&tree->lock);
862 node = tree_search(tree, file_offset);
863 if (!node) {
864 node = tree_search(tree, file_offset + len);
865 if (!node)
866 goto out;
867 }
868
869 while (1) {
870 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
871 if (range_overlaps(entry, file_offset, len))
872 break;
873
874 if (entry->file_offset >= file_offset + len) {
875 entry = NULL;
876 break;
877 }
878 entry = NULL;
879 node = rb_next(node);
880 if (!node)
881 break;
882 }
883out:
884 if (entry) {
885 refcount_inc(&entry->refs);
886 trace_btrfs_ordered_extent_lookup_range(inode, entry);
887 }
888 spin_unlock_irq(&tree->lock);
889 return entry;
890}
891
892/*
893 * Adds all ordered extents to the given list. The list ends up sorted by the
894 * file_offset of the ordered extents.
895 */
896void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
897 struct list_head *list)
898{
899 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
900 struct rb_node *n;
901
902 ASSERT(inode_is_locked(&inode->vfs_inode));
903
904 spin_lock_irq(&tree->lock);
905 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
906 struct btrfs_ordered_extent *ordered;
907
908 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
909
910 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
911 continue;
912
913 ASSERT(list_empty(&ordered->log_list));
914 list_add_tail(&ordered->log_list, list);
915 refcount_inc(&ordered->refs);
916 trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered);
917 }
918 spin_unlock_irq(&tree->lock);
919}
920
921/*
922 * lookup and return any extent before 'file_offset'. NULL is returned
923 * if none is found
924 */
925struct btrfs_ordered_extent *
926btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
927{
928 struct btrfs_ordered_inode_tree *tree;
929 struct rb_node *node;
930 struct btrfs_ordered_extent *entry = NULL;
931
932 tree = &inode->ordered_tree;
933 spin_lock_irq(&tree->lock);
934 node = tree_search(tree, file_offset);
935 if (!node)
936 goto out;
937
938 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
939 refcount_inc(&entry->refs);
940 trace_btrfs_ordered_extent_lookup_first(inode, entry);
941out:
942 spin_unlock_irq(&tree->lock);
943 return entry;
944}
945
946/*
947 * Lookup the first ordered extent that overlaps the range
948 * [@file_offset, @file_offset + @len).
949 *
950 * The difference between this and btrfs_lookup_first_ordered_extent() is
951 * that this one won't return any ordered extent that does not overlap the range.
952 * And the difference against btrfs_lookup_ordered_extent() is, this function
953 * ensures the first ordered extent gets returned.
954 */
955struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
956 struct btrfs_inode *inode, u64 file_offset, u64 len)
957{
958 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
959 struct rb_node *node;
960 struct rb_node *cur;
961 struct rb_node *prev;
962 struct rb_node *next;
963 struct btrfs_ordered_extent *entry = NULL;
964
965 spin_lock_irq(&tree->lock);
966 node = tree->tree.rb_node;
967 /*
968 * Here we don't want to use tree_search() which will use tree->last
969 * and screw up the search order.
970 * And __tree_search() can't return the adjacent ordered extents
971 * either, thus here we do our own search.
972 */
973 while (node) {
974 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
975
976 if (file_offset < entry->file_offset) {
977 node = node->rb_left;
978 } else if (file_offset >= entry_end(entry)) {
979 node = node->rb_right;
980 } else {
981 /*
982 * Direct hit, got an ordered extent that starts at
983 * @file_offset
984 */
985 goto out;
986 }
987 }
988 if (!entry) {
989 /* Empty tree */
990 goto out;
991 }
992
993 cur = &entry->rb_node;
994 /* We got an entry around @file_offset, check adjacent entries */
995 if (entry->file_offset < file_offset) {
996 prev = cur;
997 next = rb_next(cur);
998 } else {
999 prev = rb_prev(cur);
1000 next = cur;
1001 }
1002 if (prev) {
1003 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
1004 if (range_overlaps(entry, file_offset, len))
1005 goto out;
1006 }
1007 if (next) {
1008 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
1009 if (range_overlaps(entry, file_offset, len))
1010 goto out;
1011 }
1012 /* No ordered extent in the range */
1013 entry = NULL;
1014out:
1015 if (entry) {
1016 refcount_inc(&entry->refs);
1017 trace_btrfs_ordered_extent_lookup_first_range(inode, entry);
1018 }
1019
1020 spin_unlock_irq(&tree->lock);
1021 return entry;
1022}
1023
1024/*
1025 * Lock the passed range and ensures all pending ordered extents in it are run
1026 * to completion.
1027 *
1028 * @inode: Inode whose ordered tree is to be searched
1029 * @start: Beginning of range to flush
1030 * @end: Last byte of range to lock
1031 * @cached_state: If passed, will return the extent state responsible for the
1032 * locked range. It's the caller's responsibility to free the
1033 * cached state.
1034 *
1035 * Always return with the given range locked, ensuring after it's called no
1036 * order extent can be pending.
1037 */
1038void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1039 u64 end,
1040 struct extent_state **cached_state)
1041{
1042 struct btrfs_ordered_extent *ordered;
1043 struct extent_state *cache = NULL;
1044 struct extent_state **cachedp = &cache;
1045
1046 if (cached_state)
1047 cachedp = cached_state;
1048
1049 while (1) {
1050 lock_extent(&inode->io_tree, start, end, cachedp);
1051 ordered = btrfs_lookup_ordered_range(inode, start,
1052 end - start + 1);
1053 if (!ordered) {
1054 /*
1055 * If no external cached_state has been passed then
1056 * decrement the extra ref taken for cachedp since we
1057 * aren't exposing it outside of this function
1058 */
1059 if (!cached_state)
1060 refcount_dec(&cache->refs);
1061 break;
1062 }
1063 unlock_extent(&inode->io_tree, start, end, cachedp);
1064 btrfs_start_ordered_extent(ordered, 1);
1065 btrfs_put_ordered_extent(ordered);
1066 }
1067}
1068
1069/*
1070 * Lock the passed range and ensure all pending ordered extents in it are run
1071 * to completion in nowait mode.
1072 *
1073 * Return true if btrfs_lock_ordered_range does not return any extents,
1074 * otherwise false.
1075 */
1076bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end,
1077 struct extent_state **cached_state)
1078{
1079 struct btrfs_ordered_extent *ordered;
1080
1081 if (!try_lock_extent(&inode->io_tree, start, end, cached_state))
1082 return false;
1083
1084 ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1085 if (!ordered)
1086 return true;
1087
1088 btrfs_put_ordered_extent(ordered);
1089 unlock_extent(&inode->io_tree, start, end, cached_state);
1090
1091 return false;
1092}
1093
1094
1095static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
1096 u64 len)
1097{
1098 struct inode *inode = ordered->inode;
1099 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1100 u64 file_offset = ordered->file_offset + pos;
1101 u64 disk_bytenr = ordered->disk_bytenr + pos;
1102 unsigned long flags = ordered->flags & BTRFS_ORDERED_TYPE_FLAGS;
1103
1104 /*
1105 * The splitting extent is already counted and will be added again in
1106 * btrfs_add_ordered_extent_*(). Subtract len to avoid double counting.
1107 */
1108 percpu_counter_add_batch(&fs_info->ordered_bytes, -len,
1109 fs_info->delalloc_batch);
1110 WARN_ON_ONCE(flags & (1 << BTRFS_ORDERED_COMPRESSED));
1111 return btrfs_add_ordered_extent(BTRFS_I(inode), file_offset, len, len,
1112 disk_bytenr, len, 0, flags,
1113 ordered->compress_type);
1114}
1115
1116int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
1117 u64 post)
1118{
1119 struct inode *inode = ordered->inode;
1120 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1121 struct rb_node *node;
1122 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1123 int ret = 0;
1124
1125 trace_btrfs_ordered_extent_split(BTRFS_I(inode), ordered);
1126
1127 spin_lock_irq(&tree->lock);
1128 /* Remove from tree once */
1129 node = &ordered->rb_node;
1130 rb_erase(node, &tree->tree);
1131 RB_CLEAR_NODE(node);
1132 if (tree->last == node)
1133 tree->last = NULL;
1134
1135 ordered->file_offset += pre;
1136 ordered->disk_bytenr += pre;
1137 ordered->num_bytes -= (pre + post);
1138 ordered->disk_num_bytes -= (pre + post);
1139 ordered->bytes_left -= (pre + post);
1140
1141 /* Re-insert the node */
1142 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1143 if (node)
1144 btrfs_panic(fs_info, -EEXIST,
1145 "zoned: inconsistency in ordered tree at offset %llu",
1146 ordered->file_offset);
1147
1148 spin_unlock_irq(&tree->lock);
1149
1150 if (pre)
1151 ret = clone_ordered_extent(ordered, 0, pre);
1152 if (ret == 0 && post)
1153 ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
1154 post);
1155
1156 return ret;
1157}
1158
1159int __init ordered_data_init(void)
1160{
1161 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1162 sizeof(struct btrfs_ordered_extent), 0,
1163 SLAB_MEM_SPREAD,
1164 NULL);
1165 if (!btrfs_ordered_extent_cache)
1166 return -ENOMEM;
1167
1168 return 0;
1169}
1170
1171void __cold ordered_data_exit(void)
1172{
1173 kmem_cache_destroy(btrfs_ordered_extent_cache);
1174}
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 "messages.h"
11#include "misc.h"
12#include "ctree.h"
13#include "transaction.h"
14#include "btrfs_inode.h"
15#include "extent_io.h"
16#include "disk-io.h"
17#include "compression.h"
18#include "delalloc-space.h"
19#include "qgroup.h"
20#include "subpage.h"
21#include "file.h"
22
23static struct kmem_cache *btrfs_ordered_extent_cache;
24
25static u64 entry_end(struct btrfs_ordered_extent *entry)
26{
27 if (entry->file_offset + entry->num_bytes < entry->file_offset)
28 return (u64)-1;
29 return entry->file_offset + entry->num_bytes;
30}
31
32/* returns NULL if the insertion worked, or it returns the node it did find
33 * in the tree
34 */
35static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
36 struct rb_node *node)
37{
38 struct rb_node **p = &root->rb_node;
39 struct rb_node *parent = NULL;
40 struct btrfs_ordered_extent *entry;
41
42 while (*p) {
43 parent = *p;
44 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
45
46 if (file_offset < entry->file_offset)
47 p = &(*p)->rb_left;
48 else if (file_offset >= entry_end(entry))
49 p = &(*p)->rb_right;
50 else
51 return parent;
52 }
53
54 rb_link_node(node, parent, p);
55 rb_insert_color(node, root);
56 return NULL;
57}
58
59/*
60 * look for a given offset in the tree, and if it can't be found return the
61 * first lesser offset
62 */
63static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
64 struct rb_node **prev_ret)
65{
66 struct rb_node *n = root->rb_node;
67 struct rb_node *prev = NULL;
68 struct rb_node *test;
69 struct btrfs_ordered_extent *entry;
70 struct btrfs_ordered_extent *prev_entry = NULL;
71
72 while (n) {
73 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
74 prev = n;
75 prev_entry = entry;
76
77 if (file_offset < entry->file_offset)
78 n = n->rb_left;
79 else if (file_offset >= entry_end(entry))
80 n = n->rb_right;
81 else
82 return n;
83 }
84 if (!prev_ret)
85 return NULL;
86
87 while (prev && file_offset >= entry_end(prev_entry)) {
88 test = rb_next(prev);
89 if (!test)
90 break;
91 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
92 rb_node);
93 if (file_offset < entry_end(prev_entry))
94 break;
95
96 prev = test;
97 }
98 if (prev)
99 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
100 rb_node);
101 while (prev && file_offset < entry_end(prev_entry)) {
102 test = rb_prev(prev);
103 if (!test)
104 break;
105 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
106 rb_node);
107 prev = test;
108 }
109 *prev_ret = prev;
110 return NULL;
111}
112
113static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
114 u64 len)
115{
116 if (file_offset + len <= entry->file_offset ||
117 entry->file_offset + entry->num_bytes <= file_offset)
118 return 0;
119 return 1;
120}
121
122/*
123 * look find the first ordered struct that has this offset, otherwise
124 * the first one less than this offset
125 */
126static inline struct rb_node *ordered_tree_search(struct btrfs_inode *inode,
127 u64 file_offset)
128{
129 struct rb_node *prev = NULL;
130 struct rb_node *ret;
131 struct btrfs_ordered_extent *entry;
132
133 if (inode->ordered_tree_last) {
134 entry = rb_entry(inode->ordered_tree_last, struct btrfs_ordered_extent,
135 rb_node);
136 if (in_range(file_offset, entry->file_offset, entry->num_bytes))
137 return inode->ordered_tree_last;
138 }
139 ret = __tree_search(&inode->ordered_tree, file_offset, &prev);
140 if (!ret)
141 ret = prev;
142 if (ret)
143 inode->ordered_tree_last = ret;
144 return ret;
145}
146
147static struct btrfs_ordered_extent *alloc_ordered_extent(
148 struct btrfs_inode *inode, u64 file_offset, u64 num_bytes,
149 u64 ram_bytes, u64 disk_bytenr, u64 disk_num_bytes,
150 u64 offset, unsigned long flags, int compress_type)
151{
152 struct btrfs_ordered_extent *entry;
153 int ret;
154 u64 qgroup_rsv = 0;
155
156 if (flags &
157 ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
158 /* For nocow write, we can release the qgroup rsv right now */
159 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes, &qgroup_rsv);
160 if (ret < 0)
161 return ERR_PTR(ret);
162 } else {
163 /*
164 * The ordered extent has reserved qgroup space, release now
165 * and pass the reserved number for qgroup_record to free.
166 */
167 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes, &qgroup_rsv);
168 if (ret < 0)
169 return ERR_PTR(ret);
170 }
171 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
172 if (!entry)
173 return ERR_PTR(-ENOMEM);
174
175 entry->file_offset = file_offset;
176 entry->num_bytes = num_bytes;
177 entry->ram_bytes = ram_bytes;
178 entry->disk_bytenr = disk_bytenr;
179 entry->disk_num_bytes = disk_num_bytes;
180 entry->offset = offset;
181 entry->bytes_left = num_bytes;
182 entry->inode = igrab(&inode->vfs_inode);
183 entry->compress_type = compress_type;
184 entry->truncated_len = (u64)-1;
185 entry->qgroup_rsv = qgroup_rsv;
186 entry->flags = flags;
187 refcount_set(&entry->refs, 1);
188 init_waitqueue_head(&entry->wait);
189 INIT_LIST_HEAD(&entry->list);
190 INIT_LIST_HEAD(&entry->log_list);
191 INIT_LIST_HEAD(&entry->root_extent_list);
192 INIT_LIST_HEAD(&entry->work_list);
193 INIT_LIST_HEAD(&entry->bioc_list);
194 init_completion(&entry->completion);
195
196 /*
197 * We don't need the count_max_extents here, we can assume that all of
198 * that work has been done at higher layers, so this is truly the
199 * smallest the extent is going to get.
200 */
201 spin_lock(&inode->lock);
202 btrfs_mod_outstanding_extents(inode, 1);
203 spin_unlock(&inode->lock);
204
205 return entry;
206}
207
208static void insert_ordered_extent(struct btrfs_ordered_extent *entry)
209{
210 struct btrfs_inode *inode = BTRFS_I(entry->inode);
211 struct btrfs_root *root = inode->root;
212 struct btrfs_fs_info *fs_info = root->fs_info;
213 struct rb_node *node;
214
215 trace_btrfs_ordered_extent_add(inode, entry);
216
217 percpu_counter_add_batch(&fs_info->ordered_bytes, entry->num_bytes,
218 fs_info->delalloc_batch);
219
220 /* One ref for the tree. */
221 refcount_inc(&entry->refs);
222
223 spin_lock_irq(&inode->ordered_tree_lock);
224 node = tree_insert(&inode->ordered_tree, entry->file_offset,
225 &entry->rb_node);
226 if (node)
227 btrfs_panic(fs_info, -EEXIST,
228 "inconsistency in ordered tree at offset %llu",
229 entry->file_offset);
230 spin_unlock_irq(&inode->ordered_tree_lock);
231
232 spin_lock(&root->ordered_extent_lock);
233 list_add_tail(&entry->root_extent_list,
234 &root->ordered_extents);
235 root->nr_ordered_extents++;
236 if (root->nr_ordered_extents == 1) {
237 spin_lock(&fs_info->ordered_root_lock);
238 BUG_ON(!list_empty(&root->ordered_root));
239 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
240 spin_unlock(&fs_info->ordered_root_lock);
241 }
242 spin_unlock(&root->ordered_extent_lock);
243}
244
245/*
246 * Add an ordered extent to the per-inode tree.
247 *
248 * @inode: Inode that this extent is for.
249 * @file_offset: Logical offset in file where the extent starts.
250 * @num_bytes: Logical length of extent in file.
251 * @ram_bytes: Full length of unencoded data.
252 * @disk_bytenr: Offset of extent on disk.
253 * @disk_num_bytes: Size of extent on disk.
254 * @offset: Offset into unencoded data where file data starts.
255 * @flags: Flags specifying type of extent (1 << BTRFS_ORDERED_*).
256 * @compress_type: Compression algorithm used for data.
257 *
258 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
259 * tree is given a single reference on the ordered extent that was inserted, and
260 * the returned pointer is given a second reference.
261 *
262 * Return: the new ordered extent or error pointer.
263 */
264struct btrfs_ordered_extent *btrfs_alloc_ordered_extent(
265 struct btrfs_inode *inode, u64 file_offset,
266 u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
267 u64 disk_num_bytes, u64 offset, unsigned long flags,
268 int compress_type)
269{
270 struct btrfs_ordered_extent *entry;
271
272 ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
273
274 entry = alloc_ordered_extent(inode, file_offset, num_bytes, ram_bytes,
275 disk_bytenr, disk_num_bytes, offset, flags,
276 compress_type);
277 if (!IS_ERR(entry))
278 insert_ordered_extent(entry);
279 return entry;
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 */
287void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
288 struct btrfs_ordered_sum *sum)
289{
290 struct btrfs_inode *inode = BTRFS_I(entry->inode);
291
292 spin_lock_irq(&inode->ordered_tree_lock);
293 list_add_tail(&sum->list, &entry->list);
294 spin_unlock_irq(&inode->ordered_tree_lock);
295}
296
297static void finish_ordered_fn(struct btrfs_work *work)
298{
299 struct btrfs_ordered_extent *ordered_extent;
300
301 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
302 btrfs_finish_ordered_io(ordered_extent);
303}
304
305static bool can_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
306 struct page *page, u64 file_offset,
307 u64 len, bool uptodate)
308{
309 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
310 struct btrfs_fs_info *fs_info = inode->root->fs_info;
311
312 lockdep_assert_held(&inode->ordered_tree_lock);
313
314 if (page) {
315 ASSERT(page->mapping);
316 ASSERT(page_offset(page) <= file_offset);
317 ASSERT(file_offset + len <= page_offset(page) + PAGE_SIZE);
318
319 /*
320 * Ordered (Private2) bit indicates whether we still have
321 * pending io unfinished for the ordered extent.
322 *
323 * If there's no such bit, we need to skip to next range.
324 */
325 if (!btrfs_folio_test_ordered(fs_info, page_folio(page),
326 file_offset, len))
327 return false;
328 btrfs_folio_clear_ordered(fs_info, page_folio(page), file_offset, len);
329 }
330
331 /* Now we're fine to update the accounting. */
332 if (WARN_ON_ONCE(len > ordered->bytes_left)) {
333 btrfs_crit(fs_info,
334"bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%llu left=%llu",
335 inode->root->root_key.objectid, btrfs_ino(inode),
336 ordered->file_offset, ordered->num_bytes,
337 len, ordered->bytes_left);
338 ordered->bytes_left = 0;
339 } else {
340 ordered->bytes_left -= len;
341 }
342
343 if (!uptodate)
344 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
345
346 if (ordered->bytes_left)
347 return false;
348
349 /*
350 * All the IO of the ordered extent is finished, we need to queue
351 * the finish_func to be executed.
352 */
353 set_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags);
354 cond_wake_up(&ordered->wait);
355 refcount_inc(&ordered->refs);
356 trace_btrfs_ordered_extent_mark_finished(inode, ordered);
357 return true;
358}
359
360static void btrfs_queue_ordered_fn(struct btrfs_ordered_extent *ordered)
361{
362 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
363 struct btrfs_fs_info *fs_info = inode->root->fs_info;
364 struct btrfs_workqueue *wq = btrfs_is_free_space_inode(inode) ?
365 fs_info->endio_freespace_worker : fs_info->endio_write_workers;
366
367 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL);
368 btrfs_queue_work(wq, &ordered->work);
369}
370
371bool btrfs_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
372 struct page *page, u64 file_offset, u64 len,
373 bool uptodate)
374{
375 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
376 unsigned long flags;
377 bool ret;
378
379 trace_btrfs_finish_ordered_extent(inode, file_offset, len, uptodate);
380
381 spin_lock_irqsave(&inode->ordered_tree_lock, flags);
382 ret = can_finish_ordered_extent(ordered, page, file_offset, len, uptodate);
383 spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
384
385 if (ret)
386 btrfs_queue_ordered_fn(ordered);
387 return ret;
388}
389
390/*
391 * Mark all ordered extents io inside the specified range finished.
392 *
393 * @page: The involved page for the operation.
394 * For uncompressed buffered IO, the page status also needs to be
395 * updated to indicate whether the pending ordered io is finished.
396 * Can be NULL for direct IO and compressed write.
397 * For these cases, callers are ensured they won't execute the
398 * endio function twice.
399 *
400 * This function is called for endio, thus the range must have ordered
401 * extent(s) covering it.
402 */
403void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
404 struct page *page, u64 file_offset,
405 u64 num_bytes, bool uptodate)
406{
407 struct rb_node *node;
408 struct btrfs_ordered_extent *entry = NULL;
409 unsigned long flags;
410 u64 cur = file_offset;
411
412 trace_btrfs_writepage_end_io_hook(inode, file_offset,
413 file_offset + num_bytes - 1,
414 uptodate);
415
416 spin_lock_irqsave(&inode->ordered_tree_lock, flags);
417 while (cur < file_offset + num_bytes) {
418 u64 entry_end;
419 u64 end;
420 u32 len;
421
422 node = ordered_tree_search(inode, cur);
423 /* No ordered extents at all */
424 if (!node)
425 break;
426
427 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
428 entry_end = entry->file_offset + entry->num_bytes;
429 /*
430 * |<-- OE --->| |
431 * cur
432 * Go to next OE.
433 */
434 if (cur >= entry_end) {
435 node = rb_next(node);
436 /* No more ordered extents, exit */
437 if (!node)
438 break;
439 entry = rb_entry(node, struct btrfs_ordered_extent,
440 rb_node);
441
442 /* Go to next ordered extent and continue */
443 cur = entry->file_offset;
444 continue;
445 }
446 /*
447 * | |<--- OE --->|
448 * cur
449 * Go to the start of OE.
450 */
451 if (cur < entry->file_offset) {
452 cur = entry->file_offset;
453 continue;
454 }
455
456 /*
457 * Now we are definitely inside one ordered extent.
458 *
459 * |<--- OE --->|
460 * |
461 * cur
462 */
463 end = min(entry->file_offset + entry->num_bytes,
464 file_offset + num_bytes) - 1;
465 ASSERT(end + 1 - cur < U32_MAX);
466 len = end + 1 - cur;
467
468 if (can_finish_ordered_extent(entry, page, cur, len, uptodate)) {
469 spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
470 btrfs_queue_ordered_fn(entry);
471 spin_lock_irqsave(&inode->ordered_tree_lock, flags);
472 }
473 cur += len;
474 }
475 spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
476}
477
478/*
479 * Finish IO for one ordered extent across a given range. The range can only
480 * contain one ordered extent.
481 *
482 * @cached: The cached ordered extent. If not NULL, we can skip the tree
483 * search and use the ordered extent directly.
484 * Will be also used to store the finished ordered extent.
485 * @file_offset: File offset for the finished IO
486 * @io_size: Length of the finish IO range
487 *
488 * Return true if the ordered extent is finished in the range, and update
489 * @cached.
490 * Return false otherwise.
491 *
492 * NOTE: The range can NOT cross multiple ordered extents.
493 * Thus caller should ensure the range doesn't cross ordered extents.
494 */
495bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
496 struct btrfs_ordered_extent **cached,
497 u64 file_offset, u64 io_size)
498{
499 struct rb_node *node;
500 struct btrfs_ordered_extent *entry = NULL;
501 unsigned long flags;
502 bool finished = false;
503
504 spin_lock_irqsave(&inode->ordered_tree_lock, flags);
505 if (cached && *cached) {
506 entry = *cached;
507 goto have_entry;
508 }
509
510 node = ordered_tree_search(inode, file_offset);
511 if (!node)
512 goto out;
513
514 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
515have_entry:
516 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
517 goto out;
518
519 if (io_size > entry->bytes_left)
520 btrfs_crit(inode->root->fs_info,
521 "bad ordered accounting left %llu size %llu",
522 entry->bytes_left, io_size);
523
524 entry->bytes_left -= io_size;
525
526 if (entry->bytes_left == 0) {
527 /*
528 * Ensure only one caller can set the flag and finished_ret
529 * accordingly
530 */
531 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
532 /* test_and_set_bit implies a barrier */
533 cond_wake_up_nomb(&entry->wait);
534 }
535out:
536 if (finished && cached && entry) {
537 *cached = entry;
538 refcount_inc(&entry->refs);
539 trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
540 }
541 spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
542 return finished;
543}
544
545/*
546 * used to drop a reference on an ordered extent. This will free
547 * the extent if the last reference is dropped
548 */
549void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
550{
551 struct list_head *cur;
552 struct btrfs_ordered_sum *sum;
553
554 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
555
556 if (refcount_dec_and_test(&entry->refs)) {
557 ASSERT(list_empty(&entry->root_extent_list));
558 ASSERT(list_empty(&entry->log_list));
559 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
560 if (entry->inode)
561 btrfs_add_delayed_iput(BTRFS_I(entry->inode));
562 while (!list_empty(&entry->list)) {
563 cur = entry->list.next;
564 sum = list_entry(cur, struct btrfs_ordered_sum, list);
565 list_del(&sum->list);
566 kvfree(sum);
567 }
568 kmem_cache_free(btrfs_ordered_extent_cache, entry);
569 }
570}
571
572/*
573 * remove an ordered extent from the tree. No references are dropped
574 * and waiters are woken up.
575 */
576void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
577 struct btrfs_ordered_extent *entry)
578{
579 struct btrfs_root *root = btrfs_inode->root;
580 struct btrfs_fs_info *fs_info = root->fs_info;
581 struct rb_node *node;
582 bool pending;
583 bool freespace_inode;
584
585 /*
586 * If this is a free space inode the thread has not acquired the ordered
587 * extents lockdep map.
588 */
589 freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
590
591 btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
592 /* This is paired with btrfs_alloc_ordered_extent. */
593 spin_lock(&btrfs_inode->lock);
594 btrfs_mod_outstanding_extents(btrfs_inode, -1);
595 spin_unlock(&btrfs_inode->lock);
596 if (root != fs_info->tree_root) {
597 u64 release;
598
599 if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
600 release = entry->disk_num_bytes;
601 else
602 release = entry->num_bytes;
603 btrfs_delalloc_release_metadata(btrfs_inode, release,
604 test_bit(BTRFS_ORDERED_IOERR,
605 &entry->flags));
606 }
607
608 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
609 fs_info->delalloc_batch);
610
611 spin_lock_irq(&btrfs_inode->ordered_tree_lock);
612 node = &entry->rb_node;
613 rb_erase(node, &btrfs_inode->ordered_tree);
614 RB_CLEAR_NODE(node);
615 if (btrfs_inode->ordered_tree_last == node)
616 btrfs_inode->ordered_tree_last = NULL;
617 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
618 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
619 spin_unlock_irq(&btrfs_inode->ordered_tree_lock);
620
621 /*
622 * The current running transaction is waiting on us, we need to let it
623 * know that we're complete and wake it up.
624 */
625 if (pending) {
626 struct btrfs_transaction *trans;
627
628 /*
629 * The checks for trans are just a formality, it should be set,
630 * but if it isn't we don't want to deref/assert under the spin
631 * lock, so be nice and check if trans is set, but ASSERT() so
632 * if it isn't set a developer will notice.
633 */
634 spin_lock(&fs_info->trans_lock);
635 trans = fs_info->running_transaction;
636 if (trans)
637 refcount_inc(&trans->use_count);
638 spin_unlock(&fs_info->trans_lock);
639
640 ASSERT(trans || BTRFS_FS_ERROR(fs_info));
641 if (trans) {
642 if (atomic_dec_and_test(&trans->pending_ordered))
643 wake_up(&trans->pending_wait);
644 btrfs_put_transaction(trans);
645 }
646 }
647
648 btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered);
649
650 spin_lock(&root->ordered_extent_lock);
651 list_del_init(&entry->root_extent_list);
652 root->nr_ordered_extents--;
653
654 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
655
656 if (!root->nr_ordered_extents) {
657 spin_lock(&fs_info->ordered_root_lock);
658 BUG_ON(list_empty(&root->ordered_root));
659 list_del_init(&root->ordered_root);
660 spin_unlock(&fs_info->ordered_root_lock);
661 }
662 spin_unlock(&root->ordered_extent_lock);
663 wake_up(&entry->wait);
664 if (!freespace_inode)
665 btrfs_lockdep_release(fs_info, btrfs_ordered_extent);
666}
667
668static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
669{
670 struct btrfs_ordered_extent *ordered;
671
672 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
673 btrfs_start_ordered_extent(ordered);
674 complete(&ordered->completion);
675}
676
677/*
678 * wait for all the ordered extents in a root. This is done when balancing
679 * space between drives.
680 */
681u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
682 const u64 range_start, const u64 range_len)
683{
684 struct btrfs_fs_info *fs_info = root->fs_info;
685 LIST_HEAD(splice);
686 LIST_HEAD(skipped);
687 LIST_HEAD(works);
688 struct btrfs_ordered_extent *ordered, *next;
689 u64 count = 0;
690 const u64 range_end = range_start + range_len;
691
692 mutex_lock(&root->ordered_extent_mutex);
693 spin_lock(&root->ordered_extent_lock);
694 list_splice_init(&root->ordered_extents, &splice);
695 while (!list_empty(&splice) && nr) {
696 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
697 root_extent_list);
698
699 if (range_end <= ordered->disk_bytenr ||
700 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
701 list_move_tail(&ordered->root_extent_list, &skipped);
702 cond_resched_lock(&root->ordered_extent_lock);
703 continue;
704 }
705
706 list_move_tail(&ordered->root_extent_list,
707 &root->ordered_extents);
708 refcount_inc(&ordered->refs);
709 spin_unlock(&root->ordered_extent_lock);
710
711 btrfs_init_work(&ordered->flush_work,
712 btrfs_run_ordered_extent_work, NULL);
713 list_add_tail(&ordered->work_list, &works);
714 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
715
716 cond_resched();
717 spin_lock(&root->ordered_extent_lock);
718 if (nr != U64_MAX)
719 nr--;
720 count++;
721 }
722 list_splice_tail(&skipped, &root->ordered_extents);
723 list_splice_tail(&splice, &root->ordered_extents);
724 spin_unlock(&root->ordered_extent_lock);
725
726 list_for_each_entry_safe(ordered, next, &works, work_list) {
727 list_del_init(&ordered->work_list);
728 wait_for_completion(&ordered->completion);
729 btrfs_put_ordered_extent(ordered);
730 cond_resched();
731 }
732 mutex_unlock(&root->ordered_extent_mutex);
733
734 return count;
735}
736
737void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
738 const u64 range_start, const u64 range_len)
739{
740 struct btrfs_root *root;
741 LIST_HEAD(splice);
742 u64 done;
743
744 mutex_lock(&fs_info->ordered_operations_mutex);
745 spin_lock(&fs_info->ordered_root_lock);
746 list_splice_init(&fs_info->ordered_roots, &splice);
747 while (!list_empty(&splice) && nr) {
748 root = list_first_entry(&splice, struct btrfs_root,
749 ordered_root);
750 root = btrfs_grab_root(root);
751 BUG_ON(!root);
752 list_move_tail(&root->ordered_root,
753 &fs_info->ordered_roots);
754 spin_unlock(&fs_info->ordered_root_lock);
755
756 done = btrfs_wait_ordered_extents(root, nr,
757 range_start, range_len);
758 btrfs_put_root(root);
759
760 spin_lock(&fs_info->ordered_root_lock);
761 if (nr != U64_MAX) {
762 nr -= done;
763 }
764 }
765 list_splice_tail(&splice, &fs_info->ordered_roots);
766 spin_unlock(&fs_info->ordered_root_lock);
767 mutex_unlock(&fs_info->ordered_operations_mutex);
768}
769
770/*
771 * Start IO and wait for a given ordered extent to finish.
772 *
773 * Wait on page writeback for all the pages in the extent and the IO completion
774 * code to insert metadata into the btree corresponding to the extent.
775 */
776void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry)
777{
778 u64 start = entry->file_offset;
779 u64 end = start + entry->num_bytes - 1;
780 struct btrfs_inode *inode = BTRFS_I(entry->inode);
781 bool freespace_inode;
782
783 trace_btrfs_ordered_extent_start(inode, entry);
784
785 /*
786 * If this is a free space inode do not take the ordered extents lockdep
787 * map.
788 */
789 freespace_inode = btrfs_is_free_space_inode(inode);
790
791 /*
792 * pages in the range can be dirty, clean or writeback. We
793 * start IO on any dirty ones so the wait doesn't stall waiting
794 * for the flusher thread to find them
795 */
796 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
797 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
798
799 if (!freespace_inode)
800 btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent);
801 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, &entry->flags));
802}
803
804/*
805 * Used to wait on ordered extents across a large range of bytes.
806 */
807int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
808{
809 int ret = 0;
810 int ret_wb = 0;
811 u64 end;
812 u64 orig_end;
813 struct btrfs_ordered_extent *ordered;
814
815 if (start + len < start) {
816 orig_end = OFFSET_MAX;
817 } else {
818 orig_end = start + len - 1;
819 if (orig_end > OFFSET_MAX)
820 orig_end = OFFSET_MAX;
821 }
822
823 /* start IO across the range first to instantiate any delalloc
824 * extents
825 */
826 ret = btrfs_fdatawrite_range(inode, start, orig_end);
827 if (ret)
828 return ret;
829
830 /*
831 * If we have a writeback error don't return immediately. Wait first
832 * for any ordered extents that haven't completed yet. This is to make
833 * sure no one can dirty the same page ranges and call writepages()
834 * before the ordered extents complete - to avoid failures (-EEXIST)
835 * when adding the new ordered extents to the ordered tree.
836 */
837 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
838
839 end = orig_end;
840 while (1) {
841 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
842 if (!ordered)
843 break;
844 if (ordered->file_offset > orig_end) {
845 btrfs_put_ordered_extent(ordered);
846 break;
847 }
848 if (ordered->file_offset + ordered->num_bytes <= start) {
849 btrfs_put_ordered_extent(ordered);
850 break;
851 }
852 btrfs_start_ordered_extent(ordered);
853 end = ordered->file_offset;
854 /*
855 * If the ordered extent had an error save the error but don't
856 * exit without waiting first for all other ordered extents in
857 * the range to complete.
858 */
859 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
860 ret = -EIO;
861 btrfs_put_ordered_extent(ordered);
862 if (end == 0 || end == start)
863 break;
864 end--;
865 }
866 return ret_wb ? ret_wb : ret;
867}
868
869/*
870 * find an ordered extent corresponding to file_offset. return NULL if
871 * nothing is found, otherwise take a reference on the extent and return it
872 */
873struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
874 u64 file_offset)
875{
876 struct rb_node *node;
877 struct btrfs_ordered_extent *entry = NULL;
878 unsigned long flags;
879
880 spin_lock_irqsave(&inode->ordered_tree_lock, flags);
881 node = ordered_tree_search(inode, file_offset);
882 if (!node)
883 goto out;
884
885 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
886 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
887 entry = NULL;
888 if (entry) {
889 refcount_inc(&entry->refs);
890 trace_btrfs_ordered_extent_lookup(inode, entry);
891 }
892out:
893 spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
894 return entry;
895}
896
897/* Since the DIO code tries to lock a wide area we need to look for any ordered
898 * extents that exist in the range, rather than just the start of the range.
899 */
900struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
901 struct btrfs_inode *inode, u64 file_offset, u64 len)
902{
903 struct rb_node *node;
904 struct btrfs_ordered_extent *entry = NULL;
905
906 spin_lock_irq(&inode->ordered_tree_lock);
907 node = ordered_tree_search(inode, file_offset);
908 if (!node) {
909 node = ordered_tree_search(inode, 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 }
928out:
929 if (entry) {
930 refcount_inc(&entry->refs);
931 trace_btrfs_ordered_extent_lookup_range(inode, entry);
932 }
933 spin_unlock_irq(&inode->ordered_tree_lock);
934 return entry;
935}
936
937/*
938 * Adds all ordered extents to the given list. The list ends up sorted by the
939 * file_offset of the ordered extents.
940 */
941void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
942 struct list_head *list)
943{
944 struct rb_node *n;
945
946 ASSERT(inode_is_locked(&inode->vfs_inode));
947
948 spin_lock_irq(&inode->ordered_tree_lock);
949 for (n = rb_first(&inode->ordered_tree); n; n = rb_next(n)) {
950 struct btrfs_ordered_extent *ordered;
951
952 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
953
954 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
955 continue;
956
957 ASSERT(list_empty(&ordered->log_list));
958 list_add_tail(&ordered->log_list, list);
959 refcount_inc(&ordered->refs);
960 trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered);
961 }
962 spin_unlock_irq(&inode->ordered_tree_lock);
963}
964
965/*
966 * lookup and return any extent before 'file_offset'. NULL is returned
967 * if none is found
968 */
969struct btrfs_ordered_extent *
970btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
971{
972 struct rb_node *node;
973 struct btrfs_ordered_extent *entry = NULL;
974
975 spin_lock_irq(&inode->ordered_tree_lock);
976 node = ordered_tree_search(inode, file_offset);
977 if (!node)
978 goto out;
979
980 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
981 refcount_inc(&entry->refs);
982 trace_btrfs_ordered_extent_lookup_first(inode, entry);
983out:
984 spin_unlock_irq(&inode->ordered_tree_lock);
985 return entry;
986}
987
988/*
989 * Lookup the first ordered extent that overlaps the range
990 * [@file_offset, @file_offset + @len).
991 *
992 * The difference between this and btrfs_lookup_first_ordered_extent() is
993 * that this one won't return any ordered extent that does not overlap the range.
994 * And the difference against btrfs_lookup_ordered_extent() is, this function
995 * ensures the first ordered extent gets returned.
996 */
997struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
998 struct btrfs_inode *inode, u64 file_offset, u64 len)
999{
1000 struct rb_node *node;
1001 struct rb_node *cur;
1002 struct rb_node *prev;
1003 struct rb_node *next;
1004 struct btrfs_ordered_extent *entry = NULL;
1005
1006 spin_lock_irq(&inode->ordered_tree_lock);
1007 node = inode->ordered_tree.rb_node;
1008 /*
1009 * Here we don't want to use tree_search() which will use tree->last
1010 * and screw up the search order.
1011 * And __tree_search() can't return the adjacent ordered extents
1012 * either, thus here we do our own search.
1013 */
1014 while (node) {
1015 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1016
1017 if (file_offset < entry->file_offset) {
1018 node = node->rb_left;
1019 } else if (file_offset >= entry_end(entry)) {
1020 node = node->rb_right;
1021 } else {
1022 /*
1023 * Direct hit, got an ordered extent that starts at
1024 * @file_offset
1025 */
1026 goto out;
1027 }
1028 }
1029 if (!entry) {
1030 /* Empty tree */
1031 goto out;
1032 }
1033
1034 cur = &entry->rb_node;
1035 /* We got an entry around @file_offset, check adjacent entries */
1036 if (entry->file_offset < file_offset) {
1037 prev = cur;
1038 next = rb_next(cur);
1039 } else {
1040 prev = rb_prev(cur);
1041 next = cur;
1042 }
1043 if (prev) {
1044 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
1045 if (range_overlaps(entry, file_offset, len))
1046 goto out;
1047 }
1048 if (next) {
1049 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
1050 if (range_overlaps(entry, file_offset, len))
1051 goto out;
1052 }
1053 /* No ordered extent in the range */
1054 entry = NULL;
1055out:
1056 if (entry) {
1057 refcount_inc(&entry->refs);
1058 trace_btrfs_ordered_extent_lookup_first_range(inode, entry);
1059 }
1060
1061 spin_unlock_irq(&inode->ordered_tree_lock);
1062 return entry;
1063}
1064
1065/*
1066 * Lock the passed range and ensures all pending ordered extents in it are run
1067 * to completion.
1068 *
1069 * @inode: Inode whose ordered tree is to be searched
1070 * @start: Beginning of range to flush
1071 * @end: Last byte of range to lock
1072 * @cached_state: If passed, will return the extent state responsible for the
1073 * locked range. It's the caller's responsibility to free the
1074 * cached state.
1075 *
1076 * Always return with the given range locked, ensuring after it's called no
1077 * order extent can be pending.
1078 */
1079void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1080 u64 end,
1081 struct extent_state **cached_state)
1082{
1083 struct btrfs_ordered_extent *ordered;
1084 struct extent_state *cache = NULL;
1085 struct extent_state **cachedp = &cache;
1086
1087 if (cached_state)
1088 cachedp = cached_state;
1089
1090 while (1) {
1091 lock_extent(&inode->io_tree, start, end, cachedp);
1092 ordered = btrfs_lookup_ordered_range(inode, start,
1093 end - start + 1);
1094 if (!ordered) {
1095 /*
1096 * If no external cached_state has been passed then
1097 * decrement the extra ref taken for cachedp since we
1098 * aren't exposing it outside of this function
1099 */
1100 if (!cached_state)
1101 refcount_dec(&cache->refs);
1102 break;
1103 }
1104 unlock_extent(&inode->io_tree, start, end, cachedp);
1105 btrfs_start_ordered_extent(ordered);
1106 btrfs_put_ordered_extent(ordered);
1107 }
1108}
1109
1110/*
1111 * Lock the passed range and ensure all pending ordered extents in it are run
1112 * to completion in nowait mode.
1113 *
1114 * Return true if btrfs_lock_ordered_range does not return any extents,
1115 * otherwise false.
1116 */
1117bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end,
1118 struct extent_state **cached_state)
1119{
1120 struct btrfs_ordered_extent *ordered;
1121
1122 if (!try_lock_extent(&inode->io_tree, start, end, cached_state))
1123 return false;
1124
1125 ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1126 if (!ordered)
1127 return true;
1128
1129 btrfs_put_ordered_extent(ordered);
1130 unlock_extent(&inode->io_tree, start, end, cached_state);
1131
1132 return false;
1133}
1134
1135/* Split out a new ordered extent for this first @len bytes of @ordered. */
1136struct btrfs_ordered_extent *btrfs_split_ordered_extent(
1137 struct btrfs_ordered_extent *ordered, u64 len)
1138{
1139 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1140 struct btrfs_root *root = inode->root;
1141 struct btrfs_fs_info *fs_info = root->fs_info;
1142 u64 file_offset = ordered->file_offset;
1143 u64 disk_bytenr = ordered->disk_bytenr;
1144 unsigned long flags = ordered->flags;
1145 struct btrfs_ordered_sum *sum, *tmpsum;
1146 struct btrfs_ordered_extent *new;
1147 struct rb_node *node;
1148 u64 offset = 0;
1149
1150 trace_btrfs_ordered_extent_split(inode, ordered);
1151
1152 ASSERT(!(flags & (1U << BTRFS_ORDERED_COMPRESSED)));
1153
1154 /*
1155 * The entire bio must be covered by the ordered extent, but we can't
1156 * reduce the original extent to a zero length either.
1157 */
1158 if (WARN_ON_ONCE(len >= ordered->num_bytes))
1159 return ERR_PTR(-EINVAL);
1160 /* We cannot split partially completed ordered extents. */
1161 if (ordered->bytes_left) {
1162 ASSERT(!(flags & ~BTRFS_ORDERED_TYPE_FLAGS));
1163 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes))
1164 return ERR_PTR(-EINVAL);
1165 }
1166 /* We cannot split a compressed ordered extent. */
1167 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes))
1168 return ERR_PTR(-EINVAL);
1169
1170 new = alloc_ordered_extent(inode, file_offset, len, len, disk_bytenr,
1171 len, 0, flags, ordered->compress_type);
1172 if (IS_ERR(new))
1173 return new;
1174
1175 /* One ref for the tree. */
1176 refcount_inc(&new->refs);
1177
1178 spin_lock_irq(&root->ordered_extent_lock);
1179 spin_lock(&inode->ordered_tree_lock);
1180 /* Remove from tree once */
1181 node = &ordered->rb_node;
1182 rb_erase(node, &inode->ordered_tree);
1183 RB_CLEAR_NODE(node);
1184 if (inode->ordered_tree_last == node)
1185 inode->ordered_tree_last = NULL;
1186
1187 ordered->file_offset += len;
1188 ordered->disk_bytenr += len;
1189 ordered->num_bytes -= len;
1190 ordered->disk_num_bytes -= len;
1191 ordered->ram_bytes -= len;
1192
1193 if (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags)) {
1194 ASSERT(ordered->bytes_left == 0);
1195 new->bytes_left = 0;
1196 } else {
1197 ordered->bytes_left -= len;
1198 }
1199
1200 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) {
1201 if (ordered->truncated_len > len) {
1202 ordered->truncated_len -= len;
1203 } else {
1204 new->truncated_len = ordered->truncated_len;
1205 ordered->truncated_len = 0;
1206 }
1207 }
1208
1209 list_for_each_entry_safe(sum, tmpsum, &ordered->list, list) {
1210 if (offset == len)
1211 break;
1212 list_move_tail(&sum->list, &new->list);
1213 offset += sum->len;
1214 }
1215
1216 /* Re-insert the node */
1217 node = tree_insert(&inode->ordered_tree, ordered->file_offset,
1218 &ordered->rb_node);
1219 if (node)
1220 btrfs_panic(fs_info, -EEXIST,
1221 "zoned: inconsistency in ordered tree at offset %llu",
1222 ordered->file_offset);
1223
1224 node = tree_insert(&inode->ordered_tree, new->file_offset, &new->rb_node);
1225 if (node)
1226 btrfs_panic(fs_info, -EEXIST,
1227 "zoned: inconsistency in ordered tree at offset %llu",
1228 new->file_offset);
1229 spin_unlock(&inode->ordered_tree_lock);
1230
1231 list_add_tail(&new->root_extent_list, &root->ordered_extents);
1232 root->nr_ordered_extents++;
1233 spin_unlock_irq(&root->ordered_extent_lock);
1234 return new;
1235}
1236
1237int __init ordered_data_init(void)
1238{
1239 btrfs_ordered_extent_cache = KMEM_CACHE(btrfs_ordered_extent, 0);
1240 if (!btrfs_ordered_extent_cache)
1241 return -ENOMEM;
1242
1243 return 0;
1244}
1245
1246void __cold ordered_data_exit(void)
1247{
1248 kmem_cache_destroy(btrfs_ordered_extent_cache);
1249}