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