xref: /linux/fs/btrfs/ordered-data.c (revision 3eee9ac24cef892e6883b3669544c6101b70c91e)
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 
23 static struct kmem_cache *btrfs_ordered_extent_cache;
24 
25 static u64 entry_end(struct btrfs_ordered_extent *entry)
26 {
27 	if (entry->file_offset + entry->num_bytes < entry->file_offset)
28 		return (u64)-1;
29 	return entry->file_offset + entry->num_bytes;
30 }
31 
32 /* returns NULL if the insertion worked, or it returns the node it did find
33  * in the tree
34  */
35 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
36 				   struct rb_node *node)
37 {
38 	struct rb_node **p = &root->rb_node;
39 	struct rb_node *parent = NULL;
40 	struct btrfs_ordered_extent *entry;
41 
42 	while (*p) {
43 		parent = *p;
44 		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
45 
46 		if (file_offset < entry->file_offset)
47 			p = &(*p)->rb_left;
48 		else if (file_offset >= entry_end(entry))
49 			p = &(*p)->rb_right;
50 		else
51 			return parent;
52 	}
53 
54 	rb_link_node(node, parent, p);
55 	rb_insert_color(node, root);
56 	return NULL;
57 }
58 
59 /*
60  * look for a given offset in the tree, and if it can't be found return the
61  * first lesser offset
62  */
63 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
64 				     struct rb_node **prev_ret)
65 {
66 	struct rb_node *n = root->rb_node;
67 	struct rb_node *prev = NULL;
68 	struct rb_node *test;
69 	struct btrfs_ordered_extent *entry;
70 	struct btrfs_ordered_extent *prev_entry = NULL;
71 
72 	while (n) {
73 		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
74 		prev = n;
75 		prev_entry = entry;
76 
77 		if (file_offset < entry->file_offset)
78 			n = n->rb_left;
79 		else if (file_offset >= entry_end(entry))
80 			n = n->rb_right;
81 		else
82 			return n;
83 	}
84 	if (!prev_ret)
85 		return NULL;
86 
87 	while (prev && file_offset >= entry_end(prev_entry)) {
88 		test = rb_next(prev);
89 		if (!test)
90 			break;
91 		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
92 				      rb_node);
93 		if (file_offset < entry_end(prev_entry))
94 			break;
95 
96 		prev = test;
97 	}
98 	if (prev)
99 		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
100 				      rb_node);
101 	while (prev && file_offset < entry_end(prev_entry)) {
102 		test = rb_prev(prev);
103 		if (!test)
104 			break;
105 		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
106 				      rb_node);
107 		prev = test;
108 	}
109 	*prev_ret = prev;
110 	return NULL;
111 }
112 
113 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
114 			  u64 len)
115 {
116 	if (file_offset + len <= entry->file_offset ||
117 	    entry->file_offset + entry->num_bytes <= file_offset)
118 		return 0;
119 	return 1;
120 }
121 
122 /*
123  * look find the first ordered struct that has this offset, otherwise
124  * the first one less than this offset
125  */
126 static inline struct rb_node *ordered_tree_search(struct btrfs_inode *inode,
127 						  u64 file_offset)
128 {
129 	struct rb_node *prev = NULL;
130 	struct rb_node *ret;
131 	struct btrfs_ordered_extent *entry;
132 
133 	if (inode->ordered_tree_last) {
134 		entry = rb_entry(inode->ordered_tree_last, struct btrfs_ordered_extent,
135 				 rb_node);
136 		if (in_range(file_offset, entry->file_offset, entry->num_bytes))
137 			return inode->ordered_tree_last;
138 	}
139 	ret = __tree_search(&inode->ordered_tree, file_offset, &prev);
140 	if (!ret)
141 		ret = prev;
142 	if (ret)
143 		inode->ordered_tree_last = ret;
144 	return ret;
145 }
146 
147 static struct btrfs_ordered_extent *alloc_ordered_extent(
148 			struct btrfs_inode *inode, u64 file_offset, u64 num_bytes,
149 			u64 ram_bytes, u64 disk_bytenr, u64 disk_num_bytes,
150 			u64 offset, unsigned long flags, int compress_type)
151 {
152 	struct btrfs_ordered_extent *entry;
153 	int ret;
154 	u64 qgroup_rsv = 0;
155 
156 	if (flags &
157 	    ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
158 		/* For nocow write, we can release the qgroup rsv right now */
159 		ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes, &qgroup_rsv);
160 		if (ret < 0)
161 			return ERR_PTR(ret);
162 	} else {
163 		/*
164 		 * The ordered extent has reserved qgroup space, release now
165 		 * and pass the reserved number for qgroup_record to free.
166 		 */
167 		ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes, &qgroup_rsv);
168 		if (ret < 0)
169 			return ERR_PTR(ret);
170 	}
171 	entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
172 	if (!entry)
173 		return ERR_PTR(-ENOMEM);
174 
175 	entry->file_offset = file_offset;
176 	entry->num_bytes = num_bytes;
177 	entry->ram_bytes = ram_bytes;
178 	entry->disk_bytenr = disk_bytenr;
179 	entry->disk_num_bytes = disk_num_bytes;
180 	entry->offset = offset;
181 	entry->bytes_left = num_bytes;
182 	entry->inode = igrab(&inode->vfs_inode);
183 	entry->compress_type = compress_type;
184 	entry->truncated_len = (u64)-1;
185 	entry->qgroup_rsv = qgroup_rsv;
186 	entry->flags = flags;
187 	refcount_set(&entry->refs, 1);
188 	init_waitqueue_head(&entry->wait);
189 	INIT_LIST_HEAD(&entry->list);
190 	INIT_LIST_HEAD(&entry->log_list);
191 	INIT_LIST_HEAD(&entry->root_extent_list);
192 	INIT_LIST_HEAD(&entry->work_list);
193 	INIT_LIST_HEAD(&entry->bioc_list);
194 	init_completion(&entry->completion);
195 
196 	/*
197 	 * We don't need the count_max_extents here, we can assume that all of
198 	 * that work has been done at higher layers, so this is truly the
199 	 * smallest the extent is going to get.
200 	 */
201 	spin_lock(&inode->lock);
202 	btrfs_mod_outstanding_extents(inode, 1);
203 	spin_unlock(&inode->lock);
204 
205 	return entry;
206 }
207 
208 static void insert_ordered_extent(struct btrfs_ordered_extent *entry)
209 {
210 	struct btrfs_inode *inode = BTRFS_I(entry->inode);
211 	struct btrfs_root *root = inode->root;
212 	struct btrfs_fs_info *fs_info = root->fs_info;
213 	struct rb_node *node;
214 
215 	trace_btrfs_ordered_extent_add(inode, entry);
216 
217 	percpu_counter_add_batch(&fs_info->ordered_bytes, entry->num_bytes,
218 				 fs_info->delalloc_batch);
219 
220 	/* One ref for the tree. */
221 	refcount_inc(&entry->refs);
222 
223 	spin_lock_irq(&inode->ordered_tree_lock);
224 	node = tree_insert(&inode->ordered_tree, entry->file_offset,
225 			   &entry->rb_node);
226 	if (node)
227 		btrfs_panic(fs_info, -EEXIST,
228 				"inconsistency in ordered tree at offset %llu",
229 				entry->file_offset);
230 	spin_unlock_irq(&inode->ordered_tree_lock);
231 
232 	spin_lock(&root->ordered_extent_lock);
233 	list_add_tail(&entry->root_extent_list,
234 		      &root->ordered_extents);
235 	root->nr_ordered_extents++;
236 	if (root->nr_ordered_extents == 1) {
237 		spin_lock(&fs_info->ordered_root_lock);
238 		BUG_ON(!list_empty(&root->ordered_root));
239 		list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
240 		spin_unlock(&fs_info->ordered_root_lock);
241 	}
242 	spin_unlock(&root->ordered_extent_lock);
243 }
244 
245 /*
246  * Add an ordered extent to the per-inode tree.
247  *
248  * @inode:           Inode that this extent is for.
249  * @file_offset:     Logical offset in file where the extent starts.
250  * @num_bytes:       Logical length of extent in file.
251  * @ram_bytes:       Full length of unencoded data.
252  * @disk_bytenr:     Offset of extent on disk.
253  * @disk_num_bytes:  Size of extent on disk.
254  * @offset:          Offset into unencoded data where file data starts.
255  * @flags:           Flags specifying type of extent (1 << BTRFS_ORDERED_*).
256  * @compress_type:   Compression algorithm used for data.
257  *
258  * Most of these parameters correspond to &struct btrfs_file_extent_item. The
259  * tree is given a single reference on the ordered extent that was inserted, and
260  * the returned pointer is given a second reference.
261  *
262  * Return: the new ordered extent or error pointer.
263  */
264 struct btrfs_ordered_extent *btrfs_alloc_ordered_extent(
265 			struct btrfs_inode *inode, u64 file_offset,
266 			u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
267 			u64 disk_num_bytes, u64 offset, unsigned long flags,
268 			int compress_type)
269 {
270 	struct btrfs_ordered_extent *entry;
271 
272 	ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
273 
274 	entry = alloc_ordered_extent(inode, file_offset, num_bytes, ram_bytes,
275 				     disk_bytenr, disk_num_bytes, offset, flags,
276 				     compress_type);
277 	if (!IS_ERR(entry))
278 		insert_ordered_extent(entry);
279 	return entry;
280 }
281 
282 /*
283  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
284  * when an ordered extent is finished.  If the list covers more than one
285  * ordered extent, it is split across multiples.
286  */
287 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
288 			   struct btrfs_ordered_sum *sum)
289 {
290 	struct btrfs_inode *inode = BTRFS_I(entry->inode);
291 
292 	spin_lock_irq(&inode->ordered_tree_lock);
293 	list_add_tail(&sum->list, &entry->list);
294 	spin_unlock_irq(&inode->ordered_tree_lock);
295 }
296 
297 static void finish_ordered_fn(struct btrfs_work *work)
298 {
299 	struct btrfs_ordered_extent *ordered_extent;
300 
301 	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
302 	btrfs_finish_ordered_io(ordered_extent);
303 }
304 
305 static bool can_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
306 				      struct page *page, u64 file_offset,
307 				      u64 len, bool uptodate)
308 {
309 	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
310 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
311 
312 	lockdep_assert_held(&inode->ordered_tree_lock);
313 
314 	if (page) {
315 		ASSERT(page->mapping);
316 		ASSERT(page_offset(page) <= file_offset);
317 		ASSERT(file_offset + len <= page_offset(page) + PAGE_SIZE);
318 
319 		/*
320 		 * Ordered (Private2) bit indicates whether we still have
321 		 * pending io unfinished for the ordered extent.
322 		 *
323 		 * If there's no such bit, we need to skip to next range.
324 		 */
325 		if (!btrfs_folio_test_ordered(fs_info, page_folio(page),
326 					      file_offset, len))
327 			return false;
328 		btrfs_folio_clear_ordered(fs_info, page_folio(page), file_offset, len);
329 	}
330 
331 	/* Now we're fine to update the accounting. */
332 	if (WARN_ON_ONCE(len > ordered->bytes_left)) {
333 		btrfs_crit(fs_info,
334 "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%llu left=%llu",
335 			   inode->root->root_key.objectid, btrfs_ino(inode),
336 			   ordered->file_offset, ordered->num_bytes,
337 			   len, ordered->bytes_left);
338 		ordered->bytes_left = 0;
339 	} else {
340 		ordered->bytes_left -= len;
341 	}
342 
343 	if (!uptodate)
344 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
345 
346 	if (ordered->bytes_left)
347 		return false;
348 
349 	/*
350 	 * All the IO of the ordered extent is finished, we need to queue
351 	 * the finish_func to be executed.
352 	 */
353 	set_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags);
354 	cond_wake_up(&ordered->wait);
355 	refcount_inc(&ordered->refs);
356 	trace_btrfs_ordered_extent_mark_finished(inode, ordered);
357 	return true;
358 }
359 
360 static void btrfs_queue_ordered_fn(struct btrfs_ordered_extent *ordered)
361 {
362 	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
363 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
364 	struct btrfs_workqueue *wq = btrfs_is_free_space_inode(inode) ?
365 		fs_info->endio_freespace_worker : fs_info->endio_write_workers;
366 
367 	btrfs_init_work(&ordered->work, finish_ordered_fn, NULL);
368 	btrfs_queue_work(wq, &ordered->work);
369 }
370 
371 bool btrfs_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
372 				 struct page *page, u64 file_offset, u64 len,
373 				 bool uptodate)
374 {
375 	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
376 	unsigned long flags;
377 	bool ret;
378 
379 	trace_btrfs_finish_ordered_extent(inode, file_offset, len, uptodate);
380 
381 	spin_lock_irqsave(&inode->ordered_tree_lock, flags);
382 	ret = can_finish_ordered_extent(ordered, page, file_offset, len, uptodate);
383 	spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
384 
385 	if (ret)
386 		btrfs_queue_ordered_fn(ordered);
387 	return ret;
388 }
389 
390 /*
391  * Mark all ordered extents io inside the specified range finished.
392  *
393  * @page:	 The involved page for the operation.
394  *		 For uncompressed buffered IO, the page status also needs to be
395  *		 updated to indicate whether the pending ordered io is finished.
396  *		 Can be NULL for direct IO and compressed write.
397  *		 For these cases, callers are ensured they won't execute the
398  *		 endio function twice.
399  *
400  * This function is called for endio, thus the range must have ordered
401  * extent(s) covering it.
402  */
403 void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
404 				    struct page *page, u64 file_offset,
405 				    u64 num_bytes, bool uptodate)
406 {
407 	struct rb_node *node;
408 	struct btrfs_ordered_extent *entry = NULL;
409 	unsigned long flags;
410 	u64 cur = file_offset;
411 
412 	trace_btrfs_writepage_end_io_hook(inode, file_offset,
413 					  file_offset + num_bytes - 1,
414 					  uptodate);
415 
416 	spin_lock_irqsave(&inode->ordered_tree_lock, flags);
417 	while (cur < file_offset + num_bytes) {
418 		u64 entry_end;
419 		u64 end;
420 		u32 len;
421 
422 		node = ordered_tree_search(inode, cur);
423 		/* No ordered extents at all */
424 		if (!node)
425 			break;
426 
427 		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
428 		entry_end = entry->file_offset + entry->num_bytes;
429 		/*
430 		 * |<-- OE --->|  |
431 		 *		  cur
432 		 * Go to next OE.
433 		 */
434 		if (cur >= entry_end) {
435 			node = rb_next(node);
436 			/* No more ordered extents, exit */
437 			if (!node)
438 				break;
439 			entry = rb_entry(node, struct btrfs_ordered_extent,
440 					 rb_node);
441 
442 			/* Go to next ordered extent and continue */
443 			cur = entry->file_offset;
444 			continue;
445 		}
446 		/*
447 		 * |	|<--- OE --->|
448 		 * cur
449 		 * Go to the start of OE.
450 		 */
451 		if (cur < entry->file_offset) {
452 			cur = entry->file_offset;
453 			continue;
454 		}
455 
456 		/*
457 		 * Now we are definitely inside one ordered extent.
458 		 *
459 		 * |<--- OE --->|
460 		 *	|
461 		 *	cur
462 		 */
463 		end = min(entry->file_offset + entry->num_bytes,
464 			  file_offset + num_bytes) - 1;
465 		ASSERT(end + 1 - cur < U32_MAX);
466 		len = end + 1 - cur;
467 
468 		if (can_finish_ordered_extent(entry, page, cur, len, uptodate)) {
469 			spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
470 			btrfs_queue_ordered_fn(entry);
471 			spin_lock_irqsave(&inode->ordered_tree_lock, flags);
472 		}
473 		cur += len;
474 	}
475 	spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
476 }
477 
478 /*
479  * Finish IO for one ordered extent across a given range.  The range can only
480  * contain one ordered extent.
481  *
482  * @cached:	 The cached ordered extent. If not NULL, we can skip the tree
483  *               search and use the ordered extent directly.
484  * 		 Will be also used to store the finished ordered extent.
485  * @file_offset: File offset for the finished IO
486  * @io_size:	 Length of the finish IO range
487  *
488  * Return true if the ordered extent is finished in the range, and update
489  * @cached.
490  * Return false otherwise.
491  *
492  * NOTE: The range can NOT cross multiple ordered extents.
493  * Thus caller should ensure the range doesn't cross ordered extents.
494  */
495 bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
496 				    struct btrfs_ordered_extent **cached,
497 				    u64 file_offset, u64 io_size)
498 {
499 	struct rb_node *node;
500 	struct btrfs_ordered_extent *entry = NULL;
501 	unsigned long flags;
502 	bool finished = false;
503 
504 	spin_lock_irqsave(&inode->ordered_tree_lock, flags);
505 	if (cached && *cached) {
506 		entry = *cached;
507 		goto have_entry;
508 	}
509 
510 	node = ordered_tree_search(inode, file_offset);
511 	if (!node)
512 		goto out;
513 
514 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
515 have_entry:
516 	if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
517 		goto out;
518 
519 	if (io_size > entry->bytes_left)
520 		btrfs_crit(inode->root->fs_info,
521 			   "bad ordered accounting left %llu size %llu",
522 		       entry->bytes_left, io_size);
523 
524 	entry->bytes_left -= io_size;
525 
526 	if (entry->bytes_left == 0) {
527 		/*
528 		 * Ensure only one caller can set the flag and finished_ret
529 		 * accordingly
530 		 */
531 		finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
532 		/* test_and_set_bit implies a barrier */
533 		cond_wake_up_nomb(&entry->wait);
534 	}
535 out:
536 	if (finished && cached && entry) {
537 		*cached = entry;
538 		refcount_inc(&entry->refs);
539 		trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
540 	}
541 	spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
542 	return finished;
543 }
544 
545 /*
546  * used to drop a reference on an ordered extent.  This will free
547  * the extent if the last reference is dropped
548  */
549 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
550 {
551 	struct list_head *cur;
552 	struct btrfs_ordered_sum *sum;
553 
554 	trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
555 
556 	if (refcount_dec_and_test(&entry->refs)) {
557 		ASSERT(list_empty(&entry->root_extent_list));
558 		ASSERT(list_empty(&entry->log_list));
559 		ASSERT(RB_EMPTY_NODE(&entry->rb_node));
560 		if (entry->inode)
561 			btrfs_add_delayed_iput(BTRFS_I(entry->inode));
562 		while (!list_empty(&entry->list)) {
563 			cur = entry->list.next;
564 			sum = list_entry(cur, struct btrfs_ordered_sum, list);
565 			list_del(&sum->list);
566 			kvfree(sum);
567 		}
568 		kmem_cache_free(btrfs_ordered_extent_cache, entry);
569 	}
570 }
571 
572 /*
573  * remove an ordered extent from the tree.  No references are dropped
574  * and waiters are woken up.
575  */
576 void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
577 				 struct btrfs_ordered_extent *entry)
578 {
579 	struct btrfs_root *root = btrfs_inode->root;
580 	struct btrfs_fs_info *fs_info = root->fs_info;
581 	struct rb_node *node;
582 	bool pending;
583 	bool freespace_inode;
584 
585 	/*
586 	 * If this is a free space inode the thread has not acquired the ordered
587 	 * extents lockdep map.
588 	 */
589 	freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
590 
591 	btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
592 	/* This is paired with btrfs_alloc_ordered_extent. */
593 	spin_lock(&btrfs_inode->lock);
594 	btrfs_mod_outstanding_extents(btrfs_inode, -1);
595 	spin_unlock(&btrfs_inode->lock);
596 	if (root != fs_info->tree_root) {
597 		u64 release;
598 
599 		if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
600 			release = entry->disk_num_bytes;
601 		else
602 			release = entry->num_bytes;
603 		btrfs_delalloc_release_metadata(btrfs_inode, release,
604 						test_bit(BTRFS_ORDERED_IOERR,
605 							 &entry->flags));
606 	}
607 
608 	percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
609 				 fs_info->delalloc_batch);
610 
611 	spin_lock_irq(&btrfs_inode->ordered_tree_lock);
612 	node = &entry->rb_node;
613 	rb_erase(node, &btrfs_inode->ordered_tree);
614 	RB_CLEAR_NODE(node);
615 	if (btrfs_inode->ordered_tree_last == node)
616 		btrfs_inode->ordered_tree_last = NULL;
617 	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
618 	pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
619 	spin_unlock_irq(&btrfs_inode->ordered_tree_lock);
620 
621 	/*
622 	 * The current running transaction is waiting on us, we need to let it
623 	 * know that we're complete and wake it up.
624 	 */
625 	if (pending) {
626 		struct btrfs_transaction *trans;
627 
628 		/*
629 		 * The checks for trans are just a formality, it should be set,
630 		 * but if it isn't we don't want to deref/assert under the spin
631 		 * lock, so be nice and check if trans is set, but ASSERT() so
632 		 * if it isn't set a developer will notice.
633 		 */
634 		spin_lock(&fs_info->trans_lock);
635 		trans = fs_info->running_transaction;
636 		if (trans)
637 			refcount_inc(&trans->use_count);
638 		spin_unlock(&fs_info->trans_lock);
639 
640 		ASSERT(trans || BTRFS_FS_ERROR(fs_info));
641 		if (trans) {
642 			if (atomic_dec_and_test(&trans->pending_ordered))
643 				wake_up(&trans->pending_wait);
644 			btrfs_put_transaction(trans);
645 		}
646 	}
647 
648 	btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered);
649 
650 	spin_lock(&root->ordered_extent_lock);
651 	list_del_init(&entry->root_extent_list);
652 	root->nr_ordered_extents--;
653 
654 	trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
655 
656 	if (!root->nr_ordered_extents) {
657 		spin_lock(&fs_info->ordered_root_lock);
658 		BUG_ON(list_empty(&root->ordered_root));
659 		list_del_init(&root->ordered_root);
660 		spin_unlock(&fs_info->ordered_root_lock);
661 	}
662 	spin_unlock(&root->ordered_extent_lock);
663 	wake_up(&entry->wait);
664 	if (!freespace_inode)
665 		btrfs_lockdep_release(fs_info, btrfs_ordered_extent);
666 }
667 
668 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
669 {
670 	struct btrfs_ordered_extent *ordered;
671 
672 	ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
673 	btrfs_start_ordered_extent(ordered);
674 	complete(&ordered->completion);
675 }
676 
677 /*
678  * wait for all the ordered extents in a root.  This is done when balancing
679  * space between drives.
680  */
681 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
682 			       const u64 range_start, const u64 range_len)
683 {
684 	struct btrfs_fs_info *fs_info = root->fs_info;
685 	LIST_HEAD(splice);
686 	LIST_HEAD(skipped);
687 	LIST_HEAD(works);
688 	struct btrfs_ordered_extent *ordered, *next;
689 	u64 count = 0;
690 	const u64 range_end = range_start + range_len;
691 
692 	mutex_lock(&root->ordered_extent_mutex);
693 	spin_lock(&root->ordered_extent_lock);
694 	list_splice_init(&root->ordered_extents, &splice);
695 	while (!list_empty(&splice) && nr) {
696 		ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
697 					   root_extent_list);
698 
699 		if (range_end <= ordered->disk_bytenr ||
700 		    ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
701 			list_move_tail(&ordered->root_extent_list, &skipped);
702 			cond_resched_lock(&root->ordered_extent_lock);
703 			continue;
704 		}
705 
706 		list_move_tail(&ordered->root_extent_list,
707 			       &root->ordered_extents);
708 		refcount_inc(&ordered->refs);
709 		spin_unlock(&root->ordered_extent_lock);
710 
711 		btrfs_init_work(&ordered->flush_work,
712 				btrfs_run_ordered_extent_work, NULL);
713 		list_add_tail(&ordered->work_list, &works);
714 		btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
715 
716 		cond_resched();
717 		spin_lock(&root->ordered_extent_lock);
718 		if (nr != U64_MAX)
719 			nr--;
720 		count++;
721 	}
722 	list_splice_tail(&skipped, &root->ordered_extents);
723 	list_splice_tail(&splice, &root->ordered_extents);
724 	spin_unlock(&root->ordered_extent_lock);
725 
726 	list_for_each_entry_safe(ordered, next, &works, work_list) {
727 		list_del_init(&ordered->work_list);
728 		wait_for_completion(&ordered->completion);
729 		btrfs_put_ordered_extent(ordered);
730 		cond_resched();
731 	}
732 	mutex_unlock(&root->ordered_extent_mutex);
733 
734 	return count;
735 }
736 
737 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
738 			     const u64 range_start, const u64 range_len)
739 {
740 	struct btrfs_root *root;
741 	LIST_HEAD(splice);
742 	u64 done;
743 
744 	mutex_lock(&fs_info->ordered_operations_mutex);
745 	spin_lock(&fs_info->ordered_root_lock);
746 	list_splice_init(&fs_info->ordered_roots, &splice);
747 	while (!list_empty(&splice) && nr) {
748 		root = list_first_entry(&splice, struct btrfs_root,
749 					ordered_root);
750 		root = btrfs_grab_root(root);
751 		BUG_ON(!root);
752 		list_move_tail(&root->ordered_root,
753 			       &fs_info->ordered_roots);
754 		spin_unlock(&fs_info->ordered_root_lock);
755 
756 		done = btrfs_wait_ordered_extents(root, nr,
757 						  range_start, range_len);
758 		btrfs_put_root(root);
759 
760 		spin_lock(&fs_info->ordered_root_lock);
761 		if (nr != U64_MAX) {
762 			nr -= done;
763 		}
764 	}
765 	list_splice_tail(&splice, &fs_info->ordered_roots);
766 	spin_unlock(&fs_info->ordered_root_lock);
767 	mutex_unlock(&fs_info->ordered_operations_mutex);
768 }
769 
770 /*
771  * Start IO and wait for a given ordered extent to finish.
772  *
773  * Wait on page writeback for all the pages in the extent and the IO completion
774  * code to insert metadata into the btree corresponding to the extent.
775  */
776 void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry)
777 {
778 	u64 start = entry->file_offset;
779 	u64 end = start + entry->num_bytes - 1;
780 	struct btrfs_inode *inode = BTRFS_I(entry->inode);
781 	bool freespace_inode;
782 
783 	trace_btrfs_ordered_extent_start(inode, entry);
784 
785 	/*
786 	 * If this is a free space inode do not take the ordered extents lockdep
787 	 * map.
788 	 */
789 	freespace_inode = btrfs_is_free_space_inode(inode);
790 
791 	/*
792 	 * pages in the range can be dirty, clean or writeback.  We
793 	 * start IO on any dirty ones so the wait doesn't stall waiting
794 	 * for the flusher thread to find them
795 	 */
796 	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
797 		filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
798 
799 	if (!freespace_inode)
800 		btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent);
801 	wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, &entry->flags));
802 }
803 
804 /*
805  * Used to wait on ordered extents across a large range of bytes.
806  */
807 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
808 {
809 	int ret = 0;
810 	int ret_wb = 0;
811 	u64 end;
812 	u64 orig_end;
813 	struct btrfs_ordered_extent *ordered;
814 
815 	if (start + len < start) {
816 		orig_end = OFFSET_MAX;
817 	} else {
818 		orig_end = start + len - 1;
819 		if (orig_end > OFFSET_MAX)
820 			orig_end = OFFSET_MAX;
821 	}
822 
823 	/* start IO across the range first to instantiate any delalloc
824 	 * extents
825 	 */
826 	ret = btrfs_fdatawrite_range(inode, start, orig_end);
827 	if (ret)
828 		return ret;
829 
830 	/*
831 	 * If we have a writeback error don't return immediately. Wait first
832 	 * for any ordered extents that haven't completed yet. This is to make
833 	 * sure no one can dirty the same page ranges and call writepages()
834 	 * before the ordered extents complete - to avoid failures (-EEXIST)
835 	 * when adding the new ordered extents to the ordered tree.
836 	 */
837 	ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
838 
839 	end = orig_end;
840 	while (1) {
841 		ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
842 		if (!ordered)
843 			break;
844 		if (ordered->file_offset > orig_end) {
845 			btrfs_put_ordered_extent(ordered);
846 			break;
847 		}
848 		if (ordered->file_offset + ordered->num_bytes <= start) {
849 			btrfs_put_ordered_extent(ordered);
850 			break;
851 		}
852 		btrfs_start_ordered_extent(ordered);
853 		end = ordered->file_offset;
854 		/*
855 		 * If the ordered extent had an error save the error but don't
856 		 * exit without waiting first for all other ordered extents in
857 		 * the range to complete.
858 		 */
859 		if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
860 			ret = -EIO;
861 		btrfs_put_ordered_extent(ordered);
862 		if (end == 0 || end == start)
863 			break;
864 		end--;
865 	}
866 	return ret_wb ? ret_wb : ret;
867 }
868 
869 /*
870  * find an ordered extent corresponding to file_offset.  return NULL if
871  * nothing is found, otherwise take a reference on the extent and return it
872  */
873 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
874 							 u64 file_offset)
875 {
876 	struct rb_node *node;
877 	struct btrfs_ordered_extent *entry = NULL;
878 	unsigned long flags;
879 
880 	spin_lock_irqsave(&inode->ordered_tree_lock, flags);
881 	node = ordered_tree_search(inode, file_offset);
882 	if (!node)
883 		goto out;
884 
885 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
886 	if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
887 		entry = NULL;
888 	if (entry) {
889 		refcount_inc(&entry->refs);
890 		trace_btrfs_ordered_extent_lookup(inode, entry);
891 	}
892 out:
893 	spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
894 	return entry;
895 }
896 
897 /* Since the DIO code tries to lock a wide area we need to look for any ordered
898  * extents that exist in the range, rather than just the start of the range.
899  */
900 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
901 		struct btrfs_inode *inode, u64 file_offset, u64 len)
902 {
903 	struct rb_node *node;
904 	struct btrfs_ordered_extent *entry = NULL;
905 
906 	spin_lock_irq(&inode->ordered_tree_lock);
907 	node = ordered_tree_search(inode, file_offset);
908 	if (!node) {
909 		node = ordered_tree_search(inode, file_offset + len);
910 		if (!node)
911 			goto out;
912 	}
913 
914 	while (1) {
915 		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
916 		if (range_overlaps(entry, file_offset, len))
917 			break;
918 
919 		if (entry->file_offset >= file_offset + len) {
920 			entry = NULL;
921 			break;
922 		}
923 		entry = NULL;
924 		node = rb_next(node);
925 		if (!node)
926 			break;
927 	}
928 out:
929 	if (entry) {
930 		refcount_inc(&entry->refs);
931 		trace_btrfs_ordered_extent_lookup_range(inode, entry);
932 	}
933 	spin_unlock_irq(&inode->ordered_tree_lock);
934 	return entry;
935 }
936 
937 /*
938  * Adds all ordered extents to the given list. The list ends up sorted by the
939  * file_offset of the ordered extents.
940  */
941 void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
942 					   struct list_head *list)
943 {
944 	struct rb_node *n;
945 
946 	ASSERT(inode_is_locked(&inode->vfs_inode));
947 
948 	spin_lock_irq(&inode->ordered_tree_lock);
949 	for (n = rb_first(&inode->ordered_tree); n; n = rb_next(n)) {
950 		struct btrfs_ordered_extent *ordered;
951 
952 		ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
953 
954 		if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
955 			continue;
956 
957 		ASSERT(list_empty(&ordered->log_list));
958 		list_add_tail(&ordered->log_list, list);
959 		refcount_inc(&ordered->refs);
960 		trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered);
961 	}
962 	spin_unlock_irq(&inode->ordered_tree_lock);
963 }
964 
965 /*
966  * lookup and return any extent before 'file_offset'.  NULL is returned
967  * if none is found
968  */
969 struct btrfs_ordered_extent *
970 btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
971 {
972 	struct rb_node *node;
973 	struct btrfs_ordered_extent *entry = NULL;
974 
975 	spin_lock_irq(&inode->ordered_tree_lock);
976 	node = ordered_tree_search(inode, file_offset);
977 	if (!node)
978 		goto out;
979 
980 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
981 	refcount_inc(&entry->refs);
982 	trace_btrfs_ordered_extent_lookup_first(inode, entry);
983 out:
984 	spin_unlock_irq(&inode->ordered_tree_lock);
985 	return entry;
986 }
987 
988 /*
989  * Lookup the first ordered extent that overlaps the range
990  * [@file_offset, @file_offset + @len).
991  *
992  * The difference between this and btrfs_lookup_first_ordered_extent() is
993  * that this one won't return any ordered extent that does not overlap the range.
994  * And the difference against btrfs_lookup_ordered_extent() is, this function
995  * ensures the first ordered extent gets returned.
996  */
997 struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
998 			struct btrfs_inode *inode, u64 file_offset, u64 len)
999 {
1000 	struct rb_node *node;
1001 	struct rb_node *cur;
1002 	struct rb_node *prev;
1003 	struct rb_node *next;
1004 	struct btrfs_ordered_extent *entry = NULL;
1005 
1006 	spin_lock_irq(&inode->ordered_tree_lock);
1007 	node = inode->ordered_tree.rb_node;
1008 	/*
1009 	 * Here we don't want to use tree_search() which will use tree->last
1010 	 * and screw up the search order.
1011 	 * And __tree_search() can't return the adjacent ordered extents
1012 	 * either, thus here we do our own search.
1013 	 */
1014 	while (node) {
1015 		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1016 
1017 		if (file_offset < entry->file_offset) {
1018 			node = node->rb_left;
1019 		} else if (file_offset >= entry_end(entry)) {
1020 			node = node->rb_right;
1021 		} else {
1022 			/*
1023 			 * Direct hit, got an ordered extent that starts at
1024 			 * @file_offset
1025 			 */
1026 			goto out;
1027 		}
1028 	}
1029 	if (!entry) {
1030 		/* Empty tree */
1031 		goto out;
1032 	}
1033 
1034 	cur = &entry->rb_node;
1035 	/* We got an entry around @file_offset, check adjacent entries */
1036 	if (entry->file_offset < file_offset) {
1037 		prev = cur;
1038 		next = rb_next(cur);
1039 	} else {
1040 		prev = rb_prev(cur);
1041 		next = cur;
1042 	}
1043 	if (prev) {
1044 		entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
1045 		if (range_overlaps(entry, file_offset, len))
1046 			goto out;
1047 	}
1048 	if (next) {
1049 		entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
1050 		if (range_overlaps(entry, file_offset, len))
1051 			goto out;
1052 	}
1053 	/* No ordered extent in the range */
1054 	entry = NULL;
1055 out:
1056 	if (entry) {
1057 		refcount_inc(&entry->refs);
1058 		trace_btrfs_ordered_extent_lookup_first_range(inode, entry);
1059 	}
1060 
1061 	spin_unlock_irq(&inode->ordered_tree_lock);
1062 	return entry;
1063 }
1064 
1065 /*
1066  * Lock the passed range and ensures all pending ordered extents in it are run
1067  * to completion.
1068  *
1069  * @inode:        Inode whose ordered tree is to be searched
1070  * @start:        Beginning of range to flush
1071  * @end:          Last byte of range to lock
1072  * @cached_state: If passed, will return the extent state responsible for the
1073  *                locked range. It's the caller's responsibility to free the
1074  *                cached state.
1075  *
1076  * Always return with the given range locked, ensuring after it's called no
1077  * order extent can be pending.
1078  */
1079 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1080 					u64 end,
1081 					struct extent_state **cached_state)
1082 {
1083 	struct btrfs_ordered_extent *ordered;
1084 	struct extent_state *cache = NULL;
1085 	struct extent_state **cachedp = &cache;
1086 
1087 	if (cached_state)
1088 		cachedp = cached_state;
1089 
1090 	while (1) {
1091 		lock_extent(&inode->io_tree, start, end, cachedp);
1092 		ordered = btrfs_lookup_ordered_range(inode, start,
1093 						     end - start + 1);
1094 		if (!ordered) {
1095 			/*
1096 			 * If no external cached_state has been passed then
1097 			 * decrement the extra ref taken for cachedp since we
1098 			 * aren't exposing it outside of this function
1099 			 */
1100 			if (!cached_state)
1101 				refcount_dec(&cache->refs);
1102 			break;
1103 		}
1104 		unlock_extent(&inode->io_tree, start, end, cachedp);
1105 		btrfs_start_ordered_extent(ordered);
1106 		btrfs_put_ordered_extent(ordered);
1107 	}
1108 }
1109 
1110 /*
1111  * Lock the passed range and ensure all pending ordered extents in it are run
1112  * to completion in nowait mode.
1113  *
1114  * Return true if btrfs_lock_ordered_range does not return any extents,
1115  * otherwise false.
1116  */
1117 bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end,
1118 				  struct extent_state **cached_state)
1119 {
1120 	struct btrfs_ordered_extent *ordered;
1121 
1122 	if (!try_lock_extent(&inode->io_tree, start, end, cached_state))
1123 		return false;
1124 
1125 	ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1126 	if (!ordered)
1127 		return true;
1128 
1129 	btrfs_put_ordered_extent(ordered);
1130 	unlock_extent(&inode->io_tree, start, end, cached_state);
1131 
1132 	return false;
1133 }
1134 
1135 /* Split out a new ordered extent for this first @len bytes of @ordered. */
1136 struct btrfs_ordered_extent *btrfs_split_ordered_extent(
1137 			struct btrfs_ordered_extent *ordered, u64 len)
1138 {
1139 	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1140 	struct btrfs_root *root = inode->root;
1141 	struct btrfs_fs_info *fs_info = root->fs_info;
1142 	u64 file_offset = ordered->file_offset;
1143 	u64 disk_bytenr = ordered->disk_bytenr;
1144 	unsigned long flags = ordered->flags;
1145 	struct btrfs_ordered_sum *sum, *tmpsum;
1146 	struct btrfs_ordered_extent *new;
1147 	struct rb_node *node;
1148 	u64 offset = 0;
1149 
1150 	trace_btrfs_ordered_extent_split(inode, ordered);
1151 
1152 	ASSERT(!(flags & (1U << BTRFS_ORDERED_COMPRESSED)));
1153 
1154 	/*
1155 	 * The entire bio must be covered by the ordered extent, but we can't
1156 	 * reduce the original extent to a zero length either.
1157 	 */
1158 	if (WARN_ON_ONCE(len >= ordered->num_bytes))
1159 		return ERR_PTR(-EINVAL);
1160 	/* We cannot split partially completed ordered extents. */
1161 	if (ordered->bytes_left) {
1162 		ASSERT(!(flags & ~BTRFS_ORDERED_TYPE_FLAGS));
1163 		if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes))
1164 			return ERR_PTR(-EINVAL);
1165 	}
1166 	/* We cannot split a compressed ordered extent. */
1167 	if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes))
1168 		return ERR_PTR(-EINVAL);
1169 
1170 	new = alloc_ordered_extent(inode, file_offset, len, len, disk_bytenr,
1171 				   len, 0, flags, ordered->compress_type);
1172 	if (IS_ERR(new))
1173 		return new;
1174 
1175 	/* One ref for the tree. */
1176 	refcount_inc(&new->refs);
1177 
1178 	spin_lock_irq(&root->ordered_extent_lock);
1179 	spin_lock(&inode->ordered_tree_lock);
1180 	/* Remove from tree once */
1181 	node = &ordered->rb_node;
1182 	rb_erase(node, &inode->ordered_tree);
1183 	RB_CLEAR_NODE(node);
1184 	if (inode->ordered_tree_last == node)
1185 		inode->ordered_tree_last = NULL;
1186 
1187 	ordered->file_offset += len;
1188 	ordered->disk_bytenr += len;
1189 	ordered->num_bytes -= len;
1190 	ordered->disk_num_bytes -= len;
1191 
1192 	if (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags)) {
1193 		ASSERT(ordered->bytes_left == 0);
1194 		new->bytes_left = 0;
1195 	} else {
1196 		ordered->bytes_left -= len;
1197 	}
1198 
1199 	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) {
1200 		if (ordered->truncated_len > len) {
1201 			ordered->truncated_len -= len;
1202 		} else {
1203 			new->truncated_len = ordered->truncated_len;
1204 			ordered->truncated_len = 0;
1205 		}
1206 	}
1207 
1208 	list_for_each_entry_safe(sum, tmpsum, &ordered->list, list) {
1209 		if (offset == len)
1210 			break;
1211 		list_move_tail(&sum->list, &new->list);
1212 		offset += sum->len;
1213 	}
1214 
1215 	/* Re-insert the node */
1216 	node = tree_insert(&inode->ordered_tree, ordered->file_offset,
1217 			   &ordered->rb_node);
1218 	if (node)
1219 		btrfs_panic(fs_info, -EEXIST,
1220 			"zoned: inconsistency in ordered tree at offset %llu",
1221 			ordered->file_offset);
1222 
1223 	node = tree_insert(&inode->ordered_tree, new->file_offset, &new->rb_node);
1224 	if (node)
1225 		btrfs_panic(fs_info, -EEXIST,
1226 			"zoned: inconsistency in ordered tree at offset %llu",
1227 			new->file_offset);
1228 	spin_unlock(&inode->ordered_tree_lock);
1229 
1230 	list_add_tail(&new->root_extent_list, &root->ordered_extents);
1231 	root->nr_ordered_extents++;
1232 	spin_unlock_irq(&root->ordered_extent_lock);
1233 	return new;
1234 }
1235 
1236 int __init ordered_data_init(void)
1237 {
1238 	btrfs_ordered_extent_cache = KMEM_CACHE(btrfs_ordered_extent, 0);
1239 	if (!btrfs_ordered_extent_cache)
1240 		return -ENOMEM;
1241 
1242 	return 0;
1243 }
1244 
1245 void __cold ordered_data_exit(void)
1246 {
1247 	kmem_cache_destroy(btrfs_ordered_extent_cache);
1248 }
1249