1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6 #include <linux/slab.h>
7 #include <linux/blkdev.h>
8 #include <linux/writeback.h>
9 #include <linux/sched/mm.h>
10 #include "misc.h"
11 #include "ctree.h"
12 #include "transaction.h"
13 #include "btrfs_inode.h"
14 #include "extent_io.h"
15 #include "disk-io.h"
16 #include "compression.h"
17 #include "delalloc-space.h"
18 #include "qgroup.h"
19 #include "subpage.h"
20
21 static struct kmem_cache *btrfs_ordered_extent_cache;
22
entry_end(struct btrfs_ordered_extent * entry)23 static u64 entry_end(struct btrfs_ordered_extent *entry)
24 {
25 if (entry->file_offset + entry->num_bytes < entry->file_offset)
26 return (u64)-1;
27 return entry->file_offset + entry->num_bytes;
28 }
29
30 /* returns NULL if the insertion worked, or it returns the node it did find
31 * in the tree
32 */
tree_insert(struct rb_root * root,u64 file_offset,struct rb_node * node)33 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
34 struct rb_node *node)
35 {
36 struct rb_node **p = &root->rb_node;
37 struct rb_node *parent = NULL;
38 struct btrfs_ordered_extent *entry;
39
40 while (*p) {
41 parent = *p;
42 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
43
44 if (file_offset < entry->file_offset)
45 p = &(*p)->rb_left;
46 else if (file_offset >= entry_end(entry))
47 p = &(*p)->rb_right;
48 else
49 return parent;
50 }
51
52 rb_link_node(node, parent, p);
53 rb_insert_color(node, root);
54 return NULL;
55 }
56
57 /*
58 * look for a given offset in the tree, and if it can't be found return the
59 * first lesser offset
60 */
__tree_search(struct rb_root * root,u64 file_offset,struct rb_node ** prev_ret)61 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
62 struct rb_node **prev_ret)
63 {
64 struct rb_node *n = root->rb_node;
65 struct rb_node *prev = NULL;
66 struct rb_node *test;
67 struct btrfs_ordered_extent *entry;
68 struct btrfs_ordered_extent *prev_entry = NULL;
69
70 while (n) {
71 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
72 prev = n;
73 prev_entry = entry;
74
75 if (file_offset < entry->file_offset)
76 n = n->rb_left;
77 else if (file_offset >= entry_end(entry))
78 n = n->rb_right;
79 else
80 return n;
81 }
82 if (!prev_ret)
83 return NULL;
84
85 while (prev && file_offset >= entry_end(prev_entry)) {
86 test = rb_next(prev);
87 if (!test)
88 break;
89 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
90 rb_node);
91 if (file_offset < entry_end(prev_entry))
92 break;
93
94 prev = test;
95 }
96 if (prev)
97 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
98 rb_node);
99 while (prev && file_offset < entry_end(prev_entry)) {
100 test = rb_prev(prev);
101 if (!test)
102 break;
103 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
104 rb_node);
105 prev = test;
106 }
107 *prev_ret = prev;
108 return NULL;
109 }
110
range_overlaps(struct btrfs_ordered_extent * entry,u64 file_offset,u64 len)111 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
112 u64 len)
113 {
114 if (file_offset + len <= entry->file_offset ||
115 entry->file_offset + entry->num_bytes <= file_offset)
116 return 0;
117 return 1;
118 }
119
120 /*
121 * look find the first ordered struct that has this offset, otherwise
122 * the first one less than this offset
123 */
tree_search(struct btrfs_ordered_inode_tree * tree,u64 file_offset)124 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
125 u64 file_offset)
126 {
127 struct rb_root *root = &tree->tree;
128 struct rb_node *prev = NULL;
129 struct rb_node *ret;
130 struct btrfs_ordered_extent *entry;
131
132 if (tree->last) {
133 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
134 rb_node);
135 if (in_range(file_offset, entry->file_offset, entry->num_bytes))
136 return tree->last;
137 }
138 ret = __tree_search(root, file_offset, &prev);
139 if (!ret)
140 ret = prev;
141 if (ret)
142 tree->last = ret;
143 return ret;
144 }
145
146 /**
147 * Add an ordered extent to the per-inode tree.
148 *
149 * @inode: Inode that this extent is for.
150 * @file_offset: Logical offset in file where the extent starts.
151 * @num_bytes: Logical length of extent in file.
152 * @ram_bytes: Full length of unencoded data.
153 * @disk_bytenr: Offset of extent on disk.
154 * @disk_num_bytes: Size of extent on disk.
155 * @offset: Offset into unencoded data where file data starts.
156 * @flags: Flags specifying type of extent (1 << BTRFS_ORDERED_*).
157 * @compress_type: Compression algorithm used for data.
158 *
159 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
160 * tree is given a single reference on the ordered extent that was inserted.
161 *
162 * Return: 0 or -ENOMEM.
163 */
btrfs_add_ordered_extent(struct btrfs_inode * inode,u64 file_offset,u64 num_bytes,u64 ram_bytes,u64 disk_bytenr,u64 disk_num_bytes,u64 offset,unsigned flags,int compress_type)164 int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
165 u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
166 u64 disk_num_bytes, u64 offset, unsigned flags,
167 int compress_type)
168 {
169 struct btrfs_root *root = inode->root;
170 struct btrfs_fs_info *fs_info = root->fs_info;
171 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
172 struct rb_node *node;
173 struct btrfs_ordered_extent *entry;
174 int ret;
175 u64 qgroup_rsv = 0;
176
177 if (flags &
178 ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
179 /* For nocow write, we can release the qgroup rsv right now */
180 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes, &qgroup_rsv);
181 if (ret < 0)
182 return ret;
183 ret = 0;
184 } else {
185 /*
186 * The ordered extent has reserved qgroup space, release now
187 * and pass the reserved number for qgroup_record to free.
188 */
189 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes, &qgroup_rsv);
190 if (ret < 0)
191 return ret;
192 }
193 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
194 if (!entry)
195 return -ENOMEM;
196
197 entry->file_offset = file_offset;
198 entry->num_bytes = num_bytes;
199 entry->ram_bytes = ram_bytes;
200 entry->disk_bytenr = disk_bytenr;
201 entry->disk_num_bytes = disk_num_bytes;
202 entry->offset = offset;
203 entry->bytes_left = num_bytes;
204 entry->inode = igrab(&inode->vfs_inode);
205 entry->compress_type = compress_type;
206 entry->truncated_len = (u64)-1;
207 entry->qgroup_rsv = qgroup_rsv;
208 entry->physical = (u64)-1;
209
210 ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
211 entry->flags = flags;
212
213 percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
214 fs_info->delalloc_batch);
215
216 /* one ref for the tree */
217 refcount_set(&entry->refs, 1);
218 init_waitqueue_head(&entry->wait);
219 INIT_LIST_HEAD(&entry->list);
220 INIT_LIST_HEAD(&entry->log_list);
221 INIT_LIST_HEAD(&entry->root_extent_list);
222 INIT_LIST_HEAD(&entry->work_list);
223 init_completion(&entry->completion);
224
225 trace_btrfs_ordered_extent_add(inode, entry);
226
227 spin_lock_irq(&tree->lock);
228 node = tree_insert(&tree->tree, file_offset,
229 &entry->rb_node);
230 if (node)
231 btrfs_panic(fs_info, -EEXIST,
232 "inconsistency in ordered tree at offset %llu",
233 file_offset);
234 spin_unlock_irq(&tree->lock);
235
236 spin_lock(&root->ordered_extent_lock);
237 list_add_tail(&entry->root_extent_list,
238 &root->ordered_extents);
239 root->nr_ordered_extents++;
240 if (root->nr_ordered_extents == 1) {
241 spin_lock(&fs_info->ordered_root_lock);
242 BUG_ON(!list_empty(&root->ordered_root));
243 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
244 spin_unlock(&fs_info->ordered_root_lock);
245 }
246 spin_unlock(&root->ordered_extent_lock);
247
248 /*
249 * We don't need the count_max_extents here, we can assume that all of
250 * that work has been done at higher layers, so this is truly the
251 * smallest the extent is going to get.
252 */
253 spin_lock(&inode->lock);
254 btrfs_mod_outstanding_extents(inode, 1);
255 spin_unlock(&inode->lock);
256
257 return 0;
258 }
259
260 /*
261 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
262 * when an ordered extent is finished. If the list covers more than one
263 * ordered extent, it is split across multiples.
264 */
btrfs_add_ordered_sum(struct btrfs_ordered_extent * entry,struct btrfs_ordered_sum * sum)265 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
266 struct btrfs_ordered_sum *sum)
267 {
268 struct btrfs_ordered_inode_tree *tree;
269
270 tree = &BTRFS_I(entry->inode)->ordered_tree;
271 spin_lock_irq(&tree->lock);
272 list_add_tail(&sum->list, &entry->list);
273 spin_unlock_irq(&tree->lock);
274 }
275
finish_ordered_fn(struct btrfs_work * work)276 static void finish_ordered_fn(struct btrfs_work *work)
277 {
278 struct btrfs_ordered_extent *ordered_extent;
279
280 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
281 btrfs_finish_ordered_io(ordered_extent);
282 }
283
284 /*
285 * Mark all ordered extents io inside the specified range finished.
286 *
287 * @page: The involved page for the operation.
288 * For uncompressed buffered IO, the page status also needs to be
289 * updated to indicate whether the pending ordered io is finished.
290 * Can be NULL for direct IO and compressed write.
291 * For these cases, callers are ensured they won't execute the
292 * endio function twice.
293 *
294 * This function is called for endio, thus the range must have ordered
295 * extent(s) covering it.
296 */
btrfs_mark_ordered_io_finished(struct btrfs_inode * inode,struct page * page,u64 file_offset,u64 num_bytes,bool uptodate)297 void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
298 struct page *page, u64 file_offset,
299 u64 num_bytes, bool uptodate)
300 {
301 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
302 struct btrfs_fs_info *fs_info = inode->root->fs_info;
303 struct btrfs_workqueue *wq;
304 struct rb_node *node;
305 struct btrfs_ordered_extent *entry = NULL;
306 unsigned long flags;
307 u64 cur = file_offset;
308
309 if (btrfs_is_free_space_inode(inode))
310 wq = fs_info->endio_freespace_worker;
311 else
312 wq = fs_info->endio_write_workers;
313
314 if (page)
315 ASSERT(page->mapping && page_offset(page) <= file_offset &&
316 file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
317
318 spin_lock_irqsave(&tree->lock, flags);
319 while (cur < file_offset + num_bytes) {
320 u64 entry_end;
321 u64 end;
322 u32 len;
323
324 node = tree_search(tree, cur);
325 /* No ordered extents at all */
326 if (!node)
327 break;
328
329 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
330 entry_end = entry->file_offset + entry->num_bytes;
331 /*
332 * |<-- OE --->| |
333 * cur
334 * Go to next OE.
335 */
336 if (cur >= entry_end) {
337 node = rb_next(node);
338 /* No more ordered extents, exit */
339 if (!node)
340 break;
341 entry = rb_entry(node, struct btrfs_ordered_extent,
342 rb_node);
343
344 /* Go to next ordered extent and continue */
345 cur = entry->file_offset;
346 continue;
347 }
348 /*
349 * | |<--- OE --->|
350 * cur
351 * Go to the start of OE.
352 */
353 if (cur < entry->file_offset) {
354 cur = entry->file_offset;
355 continue;
356 }
357
358 /*
359 * Now we are definitely inside one ordered extent.
360 *
361 * |<--- OE --->|
362 * |
363 * cur
364 */
365 end = min(entry->file_offset + entry->num_bytes,
366 file_offset + num_bytes) - 1;
367 ASSERT(end + 1 - cur < U32_MAX);
368 len = end + 1 - cur;
369
370 if (page) {
371 /*
372 * Ordered (Private2) bit indicates whether we still
373 * have pending io unfinished for the ordered extent.
374 *
375 * If there's no such bit, we need to skip to next range.
376 */
377 if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
378 cur += len;
379 continue;
380 }
381 btrfs_page_clear_ordered(fs_info, page, cur, len);
382 }
383
384 /* Now we're fine to update the accounting */
385 if (unlikely(len > entry->bytes_left)) {
386 WARN_ON(1);
387 btrfs_crit(fs_info,
388 "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
389 inode->root->root_key.objectid,
390 btrfs_ino(inode),
391 entry->file_offset,
392 entry->num_bytes,
393 len, entry->bytes_left);
394 entry->bytes_left = 0;
395 } else {
396 entry->bytes_left -= len;
397 }
398
399 if (!uptodate)
400 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
401
402 /*
403 * All the IO of the ordered extent is finished, we need to queue
404 * the finish_func to be executed.
405 */
406 if (entry->bytes_left == 0) {
407 set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
408 cond_wake_up(&entry->wait);
409 refcount_inc(&entry->refs);
410 trace_btrfs_ordered_extent_mark_finished(inode, entry);
411 spin_unlock_irqrestore(&tree->lock, flags);
412 btrfs_init_work(&entry->work, finish_ordered_fn, NULL, NULL);
413 btrfs_queue_work(wq, &entry->work);
414 spin_lock_irqsave(&tree->lock, flags);
415 }
416 cur += len;
417 }
418 spin_unlock_irqrestore(&tree->lock, flags);
419 }
420
421 /*
422 * Finish IO for one ordered extent across a given range. The range can only
423 * contain one ordered extent.
424 *
425 * @cached: The cached ordered extent. If not NULL, we can skip the tree
426 * search and use the ordered extent directly.
427 * Will be also used to store the finished ordered extent.
428 * @file_offset: File offset for the finished IO
429 * @io_size: Length of the finish IO range
430 *
431 * Return true if the ordered extent is finished in the range, and update
432 * @cached.
433 * Return false otherwise.
434 *
435 * NOTE: The range can NOT cross multiple ordered extents.
436 * Thus caller should ensure the range doesn't cross ordered extents.
437 */
btrfs_dec_test_ordered_pending(struct btrfs_inode * inode,struct btrfs_ordered_extent ** cached,u64 file_offset,u64 io_size)438 bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
439 struct btrfs_ordered_extent **cached,
440 u64 file_offset, u64 io_size)
441 {
442 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
443 struct rb_node *node;
444 struct btrfs_ordered_extent *entry = NULL;
445 unsigned long flags;
446 bool finished = false;
447
448 spin_lock_irqsave(&tree->lock, flags);
449 if (cached && *cached) {
450 entry = *cached;
451 goto have_entry;
452 }
453
454 node = tree_search(tree, file_offset);
455 if (!node)
456 goto out;
457
458 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
459 have_entry:
460 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
461 goto out;
462
463 if (io_size > entry->bytes_left)
464 btrfs_crit(inode->root->fs_info,
465 "bad ordered accounting left %llu size %llu",
466 entry->bytes_left, io_size);
467
468 entry->bytes_left -= io_size;
469
470 if (entry->bytes_left == 0) {
471 /*
472 * Ensure only one caller can set the flag and finished_ret
473 * accordingly
474 */
475 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
476 /* test_and_set_bit implies a barrier */
477 cond_wake_up_nomb(&entry->wait);
478 }
479 out:
480 if (finished && cached && entry) {
481 *cached = entry;
482 refcount_inc(&entry->refs);
483 trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
484 }
485 spin_unlock_irqrestore(&tree->lock, flags);
486 return finished;
487 }
488
489 /*
490 * used to drop a reference on an ordered extent. This will free
491 * the extent if the last reference is dropped
492 */
btrfs_put_ordered_extent(struct btrfs_ordered_extent * entry)493 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
494 {
495 struct list_head *cur;
496 struct btrfs_ordered_sum *sum;
497
498 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
499
500 if (refcount_dec_and_test(&entry->refs)) {
501 ASSERT(list_empty(&entry->root_extent_list));
502 ASSERT(list_empty(&entry->log_list));
503 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
504 if (entry->inode)
505 btrfs_add_delayed_iput(entry->inode);
506 while (!list_empty(&entry->list)) {
507 cur = entry->list.next;
508 sum = list_entry(cur, struct btrfs_ordered_sum, list);
509 list_del(&sum->list);
510 kvfree(sum);
511 }
512 kmem_cache_free(btrfs_ordered_extent_cache, entry);
513 }
514 }
515
516 /*
517 * remove an ordered extent from the tree. No references are dropped
518 * and waiters are woken up.
519 */
btrfs_remove_ordered_extent(struct btrfs_inode * btrfs_inode,struct btrfs_ordered_extent * entry)520 void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
521 struct btrfs_ordered_extent *entry)
522 {
523 struct btrfs_ordered_inode_tree *tree;
524 struct btrfs_root *root = btrfs_inode->root;
525 struct btrfs_fs_info *fs_info = root->fs_info;
526 struct rb_node *node;
527 bool pending;
528 bool freespace_inode;
529
530 /*
531 * If this is a free space inode the thread has not acquired the ordered
532 * extents lockdep map.
533 */
534 freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
535
536 btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
537 /* This is paired with btrfs_add_ordered_extent. */
538 spin_lock(&btrfs_inode->lock);
539 btrfs_mod_outstanding_extents(btrfs_inode, -1);
540 spin_unlock(&btrfs_inode->lock);
541 if (root != fs_info->tree_root) {
542 u64 release;
543
544 if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
545 release = entry->disk_num_bytes;
546 else
547 release = entry->num_bytes;
548 btrfs_delalloc_release_metadata(btrfs_inode, release,
549 test_bit(BTRFS_ORDERED_IOERR,
550 &entry->flags));
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 || BTRFS_FS_ERROR(fs_info));
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
btrfs_run_ordered_extent_work(struct btrfs_work * work)614 static 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 */
btrfs_wait_ordered_extents(struct btrfs_root * root,u64 nr,const u64 range_start,const u64 range_len)627 u64 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
btrfs_wait_ordered_roots(struct btrfs_fs_info * fs_info,u64 nr,const u64 range_start,const u64 range_len)683 void 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 */
btrfs_start_ordered_extent(struct btrfs_ordered_extent * entry,int wait)725 void 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 */
btrfs_wait_ordered_range(struct inode * inode,u64 start,u64 len)758 int 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 = INT_LIMIT(loff_t);
768 } else {
769 orig_end = start + len - 1;
770 if (orig_end > INT_LIMIT(loff_t))
771 orig_end = INT_LIMIT(loff_t);
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 */
btrfs_lookup_ordered_extent(struct btrfs_inode * inode,u64 file_offset)824 struct 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 }
845 out:
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 */
btrfs_lookup_ordered_range(struct btrfs_inode * inode,u64 file_offset,u64 len)853 struct 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 }
883 out:
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 */
btrfs_get_ordered_extents_for_logging(struct btrfs_inode * inode,struct list_head * list)896 void 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 */
925 struct btrfs_ordered_extent *
btrfs_lookup_first_ordered_extent(struct btrfs_inode * inode,u64 file_offset)926 btrfs_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);
941 out:
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 */
btrfs_lookup_first_ordered_range(struct btrfs_inode * inode,u64 file_offset,u64 len)955 struct 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;
1014 out:
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 * btrfs_flush_ordered_range - Lock the passed range and ensures all pending
1026 * ordered extents in it are run 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 cached state.
1033 *
1034 * This function always returns with the given range locked, ensuring after it's
1035 * called no order extent can be pending.
1036 */
btrfs_lock_and_flush_ordered_range(struct btrfs_inode * inode,u64 start,u64 end,struct extent_state ** cached_state)1037 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1038 u64 end,
1039 struct extent_state **cached_state)
1040 {
1041 struct btrfs_ordered_extent *ordered;
1042 struct extent_state *cache = NULL;
1043 struct extent_state **cachedp = &cache;
1044
1045 if (cached_state)
1046 cachedp = cached_state;
1047
1048 while (1) {
1049 lock_extent(&inode->io_tree, start, end, cachedp);
1050 ordered = btrfs_lookup_ordered_range(inode, start,
1051 end - start + 1);
1052 if (!ordered) {
1053 /*
1054 * If no external cached_state has been passed then
1055 * decrement the extra ref taken for cachedp since we
1056 * aren't exposing it outside of this function
1057 */
1058 if (!cached_state)
1059 refcount_dec(&cache->refs);
1060 break;
1061 }
1062 unlock_extent(&inode->io_tree, start, end, cachedp);
1063 btrfs_start_ordered_extent(ordered, 1);
1064 btrfs_put_ordered_extent(ordered);
1065 }
1066 }
1067
1068 /*
1069 * Lock the passed range and ensure all pending ordered extents in it are run
1070 * to completion in nowait mode.
1071 *
1072 * Return true if btrfs_lock_ordered_range does not return any extents,
1073 * otherwise false.
1074 */
btrfs_try_lock_ordered_range(struct btrfs_inode * inode,u64 start,u64 end)1075 bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end)
1076 {
1077 struct btrfs_ordered_extent *ordered;
1078
1079 if (!try_lock_extent(&inode->io_tree, start, end))
1080 return false;
1081
1082 ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1083 if (!ordered)
1084 return true;
1085
1086 btrfs_put_ordered_extent(ordered);
1087 unlock_extent(&inode->io_tree, start, end, NULL);
1088
1089 return false;
1090 }
1091
1092
clone_ordered_extent(struct btrfs_ordered_extent * ordered,u64 pos,u64 len)1093 static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
1094 u64 len)
1095 {
1096 struct inode *inode = ordered->inode;
1097 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1098 u64 file_offset = ordered->file_offset + pos;
1099 u64 disk_bytenr = ordered->disk_bytenr + pos;
1100 unsigned long flags = ordered->flags & BTRFS_ORDERED_TYPE_FLAGS;
1101
1102 /*
1103 * The splitting extent is already counted and will be added again in
1104 * btrfs_add_ordered_extent_*(). Subtract len to avoid double counting.
1105 */
1106 percpu_counter_add_batch(&fs_info->ordered_bytes, -len,
1107 fs_info->delalloc_batch);
1108 WARN_ON_ONCE(flags & (1 << BTRFS_ORDERED_COMPRESSED));
1109 return btrfs_add_ordered_extent(BTRFS_I(inode), file_offset, len, len,
1110 disk_bytenr, len, 0, flags,
1111 ordered->compress_type);
1112 }
1113
btrfs_split_ordered_extent(struct btrfs_ordered_extent * ordered,u64 pre,u64 post)1114 int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
1115 u64 post)
1116 {
1117 struct inode *inode = ordered->inode;
1118 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1119 struct rb_node *node;
1120 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1121 int ret = 0;
1122
1123 trace_btrfs_ordered_extent_split(BTRFS_I(inode), ordered);
1124
1125 spin_lock_irq(&tree->lock);
1126 /* Remove from tree once */
1127 node = &ordered->rb_node;
1128 rb_erase(node, &tree->tree);
1129 RB_CLEAR_NODE(node);
1130 if (tree->last == node)
1131 tree->last = NULL;
1132
1133 ordered->file_offset += pre;
1134 ordered->disk_bytenr += pre;
1135 ordered->num_bytes -= (pre + post);
1136 ordered->disk_num_bytes -= (pre + post);
1137 ordered->bytes_left -= (pre + post);
1138
1139 /* Re-insert the node */
1140 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1141 if (node)
1142 btrfs_panic(fs_info, -EEXIST,
1143 "zoned: inconsistency in ordered tree at offset %llu",
1144 ordered->file_offset);
1145
1146 spin_unlock_irq(&tree->lock);
1147
1148 if (pre)
1149 ret = clone_ordered_extent(ordered, 0, pre);
1150 if (ret == 0 && post)
1151 ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
1152 post);
1153
1154 return ret;
1155 }
1156
ordered_data_init(void)1157 int __init ordered_data_init(void)
1158 {
1159 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1160 sizeof(struct btrfs_ordered_extent), 0,
1161 SLAB_MEM_SPREAD,
1162 NULL);
1163 if (!btrfs_ordered_extent_cache)
1164 return -ENOMEM;
1165
1166 return 0;
1167 }
1168
ordered_data_exit(void)1169 void __cold ordered_data_exit(void)
1170 {
1171 kmem_cache_destroy(btrfs_ordered_extent_cache);
1172 }
1173