1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Copyright (C) 2007 Oracle. All rights reserved.
4	*/
5
6	#include <linux/fs.h>
7	#include <linux/slab.h>
8	#include <linux/sched.h>
9	#include <linux/sched/mm.h>
10	#include <linux/writeback.h>
11	#include <linux/pagemap.h>
12	#include <linux/blkdev.h>
13	#include <linux/uuid.h>
14	#include <linux/timekeeping.h>
15	#include "misc.h"
16	#include "ctree.h"
17	#include "disk-io.h"
18	#include "transaction.h"
19	#include "locking.h"
20	#include "tree-log.h"
21	#include "volumes.h"
22	#include "dev-replace.h"
23	#include "qgroup.h"
24	#include "block-group.h"
25	#include "space-info.h"
26	#include "zoned.h"
27	#include "fs.h"
28	#include "accessors.h"
29	#include "extent-tree.h"
30	#include "root-tree.h"
31	#include "defrag.h"
32	#include "dir-item.h"
33	#include "uuid-tree.h"
34	#include "ioctl.h"
35	#include "relocation.h"
36	#include "scrub.h"
37
38	static struct kmem_cache *btrfs_trans_handle_cachep;
39
40	#define BTRFS_ROOT_TRANS_TAG 0
41
42	/*
43	* Transaction states and transitions
44	*
45	* No running transaction (fs tree blocks are not modified)
46	* |
47	* | To next stage:
48	* | Call start_transaction() variants. Except btrfs_join_transaction_nostart().
49	* V
50	* Transaction N [[TRANS_STATE_RUNNING]]
51	* |
52	* | New trans handles can be attached to transaction N by calling all
53	* | start_transaction() variants.
54	* |
55	* | To next stage:
56	* | Call btrfs_commit_transaction() on any trans handle attached to
57	* | transaction N
58	* V
59	* Transaction N [[TRANS_STATE_COMMIT_PREP]]
60	* |
61	* | If there are simultaneous calls to btrfs_commit_transaction() one will win
62	* | the race and the rest will wait for the winner to commit the transaction.
63	* |
64	* | The winner will wait for previous running transaction to completely finish
65	* | if there is one.
66	* |
67	* Transaction N [[TRANS_STATE_COMMIT_START]]
68	* |
69	* | Then one of the following happens:
70	* | - Wait for all other trans handle holders to release.
71	* | The btrfs_commit_transaction() caller will do the commit work.
72	* | - Wait for current transaction to be committed by others.
73	* | Other btrfs_commit_transaction() caller will do the commit work.
74	* |
75	* | At this stage, only btrfs_join_transaction*() variants can attach
76	* | to this running transaction.
77	* | All other variants will wait for current one to finish and attach to
78	* | transaction N+1.
79	* |
80	* | To next stage:
81	* | Caller is chosen to commit transaction N, and all other trans handle
82	* | haven been released.
83	* V
84	* Transaction N [[TRANS_STATE_COMMIT_DOING]]
85	* |
86	* | The heavy lifting transaction work is started.
87	* | From running delayed refs (modifying extent tree) to creating pending
88	* | snapshots, running qgroups.
89	* | In short, modify supporting trees to reflect modifications of subvolume
90	* | trees.
91	* |
92	* | At this stage, all start_transaction() calls will wait for this
93	* | transaction to finish and attach to transaction N+1.
94	* |
95	* | To next stage:
96	* | Until all supporting trees are updated.
97	* V
98	* Transaction N [[TRANS_STATE_UNBLOCKED]]
99	* | Transaction N+1
100	* | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
101	* | need to write them back to disk and update |
102	* | super blocks. |
103	* | |
104	* | At this stage, new transaction is allowed to |
105	* | start. |
106	* | All new start_transaction() calls will be |
107	* | attached to transid N+1. |
108	* | |
109	* | To next stage: |
110	* | Until all tree blocks are super blocks are |
111	* | written to block devices |
112	* V |
113	* Transaction N [[TRANS_STATE_COMPLETED]] V
114	* All tree blocks and super blocks are written. Transaction N+1
115	* This transaction is finished and all its [[TRANS_STATE_COMMIT_START]]
116	* data structures will be cleaned up. | Life goes on
117	*/
118	static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
119	[TRANS_STATE_RUNNING] = 0U,
120	[TRANS_STATE_COMMIT_PREP] = 0U,
121	[TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH),
122	[TRANS_STATE_COMMIT_DOING] = (__TRANS_START |
123	__TRANS_ATTACH |
124	__TRANS_JOIN |
125	__TRANS_JOIN_NOSTART),
126	[TRANS_STATE_UNBLOCKED] = (__TRANS_START |
127	__TRANS_ATTACH |
128	__TRANS_JOIN |
129	__TRANS_JOIN_NOLOCK |
130	__TRANS_JOIN_NOSTART),
131	[TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START |
132	__TRANS_ATTACH |
133	__TRANS_JOIN |
134	__TRANS_JOIN_NOLOCK |
135	__TRANS_JOIN_NOSTART),
136	[TRANS_STATE_COMPLETED] = (__TRANS_START |
137	__TRANS_ATTACH |
138	__TRANS_JOIN |
139	__TRANS_JOIN_NOLOCK |
140	__TRANS_JOIN_NOSTART),
141	};
142
143	void btrfs_put_transaction(struct btrfs_transaction *transaction)
144	{
145	WARN_ON(refcount_read(&transaction->use_count) == 0);
146	if (refcount_dec_and_test(r: &transaction->use_count)) {
147	BUG_ON(!list_empty(&transaction->list));
148	WARN_ON(!RB_EMPTY_ROOT(
149	&transaction->delayed_refs.href_root.rb_root));
150	WARN_ON(!RB_EMPTY_ROOT(
151	&transaction->delayed_refs.dirty_extent_root));
152	if (transaction->delayed_refs.pending_csums)
153	btrfs_err(transaction->fs_info,
154	"pending csums is %llu",
155	transaction->delayed_refs.pending_csums);
156	/*
157	* If any block groups are found in ->deleted_bgs then it's
158	* because the transaction was aborted and a commit did not
159	* happen (things failed before writing the new superblock
160	* and calling btrfs_finish_extent_commit()), so we can not
161	* discard the physical locations of the block groups.
162	*/
163	while (!list_empty(head: &transaction->deleted_bgs)) {
164	struct btrfs_block_group *cache;
165
166	cache = list_first_entry(&transaction->deleted_bgs,
167	struct btrfs_block_group,
168	bg_list);
169	list_del_init(entry: &cache->bg_list);
170	btrfs_unfreeze_block_group(cache);
171	btrfs_put_block_group(cache);
172	}
173	WARN_ON(!list_empty(&transaction->dev_update_list));
174	kfree(objp: transaction);
175	}
176	}
177
178	static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
179	{
180	struct btrfs_transaction *cur_trans = trans->transaction;
181	struct btrfs_fs_info *fs_info = trans->fs_info;
182	struct btrfs_root *root, *tmp;
183
184	/*
185	* At this point no one can be using this transaction to modify any tree
186	* and no one can start another transaction to modify any tree either.
187	*/
188	ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING);
189
190	down_write(sem: &fs_info->commit_root_sem);
191
192	if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
193	fs_info->last_reloc_trans = trans->transid;
194
195	list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
196	dirty_list) {
197	list_del_init(entry: &root->dirty_list);
198	free_extent_buffer(eb: root->commit_root);
199	root->commit_root = btrfs_root_node(root);
200	extent_io_tree_release(tree: &root->dirty_log_pages);
201	btrfs_qgroup_clean_swapped_blocks(root);
202	}
203
204	/* We can free old roots now. */
205	spin_lock(lock: &cur_trans->dropped_roots_lock);
206	while (!list_empty(head: &cur_trans->dropped_roots)) {
207	root = list_first_entry(&cur_trans->dropped_roots,
208	struct btrfs_root, root_list);
209	list_del_init(entry: &root->root_list);
210	spin_unlock(lock: &cur_trans->dropped_roots_lock);
211	btrfs_free_log(trans, root);
212	btrfs_drop_and_free_fs_root(fs_info, root);
213	spin_lock(lock: &cur_trans->dropped_roots_lock);
214	}
215	spin_unlock(lock: &cur_trans->dropped_roots_lock);
216
217	up_write(sem: &fs_info->commit_root_sem);
218	}
219
220	static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
221	unsigned int type)
222	{
223	if (type & TRANS_EXTWRITERS)
224	atomic_inc(v: &trans->num_extwriters);
225	}
226
227	static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
228	unsigned int type)
229	{
230	if (type & TRANS_EXTWRITERS)
231	atomic_dec(v: &trans->num_extwriters);
232	}
233
234	static inline void extwriter_counter_init(struct btrfs_transaction *trans,
235	unsigned int type)
236	{
237	atomic_set(v: &trans->num_extwriters, i: ((type & TRANS_EXTWRITERS) ? 1 : 0));
238	}
239
240	static inline int extwriter_counter_read(struct btrfs_transaction *trans)
241	{
242	return atomic_read(v: &trans->num_extwriters);
243	}
244
245	/*
246	* To be called after doing the chunk btree updates right after allocating a new
247	* chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
248	* chunk after all chunk btree updates and after finishing the second phase of
249	* chunk allocation (btrfs_create_pending_block_groups()) in case some block
250	* group had its chunk item insertion delayed to the second phase.
251	*/
252	void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
253	{
254	struct btrfs_fs_info *fs_info = trans->fs_info;
255
256	if (!trans->chunk_bytes_reserved)
257	return;
258
259	btrfs_block_rsv_release(fs_info, block_rsv: &fs_info->chunk_block_rsv,
260	num_bytes: trans->chunk_bytes_reserved, NULL);
261	trans->chunk_bytes_reserved = 0;
262	}
263
264	/*
265	* either allocate a new transaction or hop into the existing one
266	*/
267	static noinline int join_transaction(struct btrfs_fs_info *fs_info,
268	unsigned int type)
269	{
270	struct btrfs_transaction *cur_trans;
271
272	spin_lock(lock: &fs_info->trans_lock);
273	loop:
274	/* The file system has been taken offline. No new transactions. */
275	if (BTRFS_FS_ERROR(fs_info)) {
276	spin_unlock(lock: &fs_info->trans_lock);
277	return -EROFS;
278	}
279
280	cur_trans = fs_info->running_transaction;
281	if (cur_trans) {
282	if (TRANS_ABORTED(cur_trans)) {
283	spin_unlock(lock: &fs_info->trans_lock);
284	return cur_trans->aborted;
285	}
286	if (btrfs_blocked_trans_types[cur_trans->state] & type) {
287	spin_unlock(lock: &fs_info->trans_lock);
288	return -EBUSY;
289	}
290	refcount_inc(r: &cur_trans->use_count);
291	atomic_inc(v: &cur_trans->num_writers);
292	extwriter_counter_inc(trans: cur_trans, type);
293	spin_unlock(lock: &fs_info->trans_lock);
294	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
295	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
296	return 0;
297	}
298	spin_unlock(lock: &fs_info->trans_lock);
299
300	/*
301	* If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the
302	* current transaction, and commit it. If there is no transaction, just
303	* return ENOENT.
304	*/
305	if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART)
306	return -ENOENT;
307
308	/*
309	* JOIN_NOLOCK only happens during the transaction commit, so
310	* it is impossible that ->running_transaction is NULL
311	*/
312	BUG_ON(type == TRANS_JOIN_NOLOCK);
313
314	cur_trans = kmalloc(size: sizeof(*cur_trans), GFP_NOFS);
315	if (!cur_trans)
316	return -ENOMEM;
317
318	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
319	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
320
321	spin_lock(lock: &fs_info->trans_lock);
322	if (fs_info->running_transaction) {
323	/*
324	* someone started a transaction after we unlocked. Make sure
325	* to redo the checks above
326	*/
327	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
328	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
329	kfree(objp: cur_trans);
330	goto loop;
331	} else if (BTRFS_FS_ERROR(fs_info)) {
332	spin_unlock(lock: &fs_info->trans_lock);
333	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
334	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
335	kfree(objp: cur_trans);
336	return -EROFS;
337	}
338
339	cur_trans->fs_info = fs_info;
340	atomic_set(v: &cur_trans->pending_ordered, i: 0);
341	init_waitqueue_head(&cur_trans->pending_wait);
342	atomic_set(v: &cur_trans->num_writers, i: 1);
343	extwriter_counter_init(trans: cur_trans, type);
344	init_waitqueue_head(&cur_trans->writer_wait);
345	init_waitqueue_head(&cur_trans->commit_wait);
346	cur_trans->state = TRANS_STATE_RUNNING;
347	/*
348	* One for this trans handle, one so it will live on until we
349	* commit the transaction.
350	*/
351	refcount_set(r: &cur_trans->use_count, n: 2);
352	cur_trans->flags = 0;
353	cur_trans->start_time = ktime_get_seconds();
354
355	memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
356
357	cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
358	cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
359	atomic_set(v: &cur_trans->delayed_refs.num_entries, i: 0);
360
361	/*
362	* although the tree mod log is per file system and not per transaction,
363	* the log must never go across transaction boundaries.
364	*/
365	smp_mb();
366	if (!list_empty(head: &fs_info->tree_mod_seq_list))
367	WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
368	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
369	WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
370	atomic64_set(v: &fs_info->tree_mod_seq, i: 0);
371
372	spin_lock_init(&cur_trans->delayed_refs.lock);
373
374	INIT_LIST_HEAD(list: &cur_trans->pending_snapshots);
375	INIT_LIST_HEAD(list: &cur_trans->dev_update_list);
376	INIT_LIST_HEAD(list: &cur_trans->switch_commits);
377	INIT_LIST_HEAD(list: &cur_trans->dirty_bgs);
378	INIT_LIST_HEAD(list: &cur_trans->io_bgs);
379	INIT_LIST_HEAD(list: &cur_trans->dropped_roots);
380	mutex_init(&cur_trans->cache_write_mutex);
381	spin_lock_init(&cur_trans->dirty_bgs_lock);
382	INIT_LIST_HEAD(list: &cur_trans->deleted_bgs);
383	spin_lock_init(&cur_trans->dropped_roots_lock);
384	list_add_tail(new: &cur_trans->list, head: &fs_info->trans_list);
385	extent_io_tree_init(fs_info, tree: &cur_trans->dirty_pages,
386	owner: IO_TREE_TRANS_DIRTY_PAGES);
387	extent_io_tree_init(fs_info, tree: &cur_trans->pinned_extents,
388	owner: IO_TREE_FS_PINNED_EXTENTS);
389	btrfs_set_fs_generation(fs_info, gen: fs_info->generation + 1);
390	cur_trans->transid = fs_info->generation;
391	fs_info->running_transaction = cur_trans;
392	cur_trans->aborted = 0;
393	spin_unlock(lock: &fs_info->trans_lock);
394
395	return 0;
396	}
397
398	/*
399	* This does all the record keeping required to make sure that a shareable root
400	* is properly recorded in a given transaction. This is required to make sure
401	* the old root from before we joined the transaction is deleted when the
402	* transaction commits.
403	*/
404	static int record_root_in_trans(struct btrfs_trans_handle *trans,
405	struct btrfs_root *root,
406	int force)
407	{
408	struct btrfs_fs_info *fs_info = root->fs_info;
409	int ret = 0;
410
411	if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
412	root->last_trans < trans->transid) || force) {
413	WARN_ON(!force && root->commit_root != root->node);
414
415	/*
416	* see below for IN_TRANS_SETUP usage rules
417	* we have the reloc mutex held now, so there
418	* is only one writer in this function
419	*/
420	set_bit(nr: BTRFS_ROOT_IN_TRANS_SETUP, addr: &root->state);
421
422	/* make sure readers find IN_TRANS_SETUP before
423	* they find our root->last_trans update
424	*/
425	smp_wmb();
426
427	spin_lock(lock: &fs_info->fs_roots_radix_lock);
428	if (root->last_trans == trans->transid && !force) {
429	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
430	return 0;
431	}
432	radix_tree_tag_set(&fs_info->fs_roots_radix,
433	index: (unsigned long)root->root_key.objectid,
434	BTRFS_ROOT_TRANS_TAG);
435	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
436	root->last_trans = trans->transid;
437
438	/* this is pretty tricky. We don't want to
439	* take the relocation lock in btrfs_record_root_in_trans
440	* unless we're really doing the first setup for this root in
441	* this transaction.
442	*
443	* Normally we'd use root->last_trans as a flag to decide
444	* if we want to take the expensive mutex.
445	*
446	* But, we have to set root->last_trans before we
447	* init the relocation root, otherwise, we trip over warnings
448	* in ctree.c. The solution used here is to flag ourselves
449	* with root IN_TRANS_SETUP. When this is 1, we're still
450	* fixing up the reloc trees and everyone must wait.
451	*
452	* When this is zero, they can trust root->last_trans and fly
453	* through btrfs_record_root_in_trans without having to take the
454	* lock. smp_wmb() makes sure that all the writes above are
455	* done before we pop in the zero below
456	*/
457	ret = btrfs_init_reloc_root(trans, root);
458	smp_mb__before_atomic();
459	clear_bit(nr: BTRFS_ROOT_IN_TRANS_SETUP, addr: &root->state);
460	}
461	return ret;
462	}
463
464
465	void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
466	struct btrfs_root *root)
467	{
468	struct btrfs_fs_info *fs_info = root->fs_info;
469	struct btrfs_transaction *cur_trans = trans->transaction;
470
471	/* Add ourselves to the transaction dropped list */
472	spin_lock(lock: &cur_trans->dropped_roots_lock);
473	list_add_tail(new: &root->root_list, head: &cur_trans->dropped_roots);
474	spin_unlock(lock: &cur_trans->dropped_roots_lock);
475
476	/* Make sure we don't try to update the root at commit time */
477	spin_lock(lock: &fs_info->fs_roots_radix_lock);
478	radix_tree_tag_clear(&fs_info->fs_roots_radix,
479	index: (unsigned long)root->root_key.objectid,
480	BTRFS_ROOT_TRANS_TAG);
481	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
482	}
483
484	int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
485	struct btrfs_root *root)
486	{
487	struct btrfs_fs_info *fs_info = root->fs_info;
488	int ret;
489
490	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
491	return 0;
492
493	/*
494	* see record_root_in_trans for comments about IN_TRANS_SETUP usage
495	* and barriers
496	*/
497	smp_rmb();
498	if (root->last_trans == trans->transid &&
499	!test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
500	return 0;
501
502	mutex_lock(&fs_info->reloc_mutex);
503	ret = record_root_in_trans(trans, root, force: 0);
504	mutex_unlock(lock: &fs_info->reloc_mutex);
505
506	return ret;
507	}
508
509	static inline int is_transaction_blocked(struct btrfs_transaction *trans)
510	{
511	return (trans->state >= TRANS_STATE_COMMIT_START &&
512	trans->state < TRANS_STATE_UNBLOCKED &&
513	!TRANS_ABORTED(trans));
514	}
515
516	/* wait for commit against the current transaction to become unblocked
517	* when this is done, it is safe to start a new transaction, but the current
518	* transaction might not be fully on disk.
519	*/
520	static void wait_current_trans(struct btrfs_fs_info *fs_info)
521	{
522	struct btrfs_transaction *cur_trans;
523
524	spin_lock(lock: &fs_info->trans_lock);
525	cur_trans = fs_info->running_transaction;
526	if (cur_trans && is_transaction_blocked(trans: cur_trans)) {
527	refcount_inc(r: &cur_trans->use_count);
528	spin_unlock(lock: &fs_info->trans_lock);
529
530	btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
531	wait_event(fs_info->transaction_wait,
532	cur_trans->state >= TRANS_STATE_UNBLOCKED ||
533	TRANS_ABORTED(cur_trans));
534	btrfs_put_transaction(transaction: cur_trans);
535	} else {
536	spin_unlock(lock: &fs_info->trans_lock);
537	}
538	}
539
540	static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
541	{
542	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
543	return 0;
544
545	if (type == TRANS_START)
546	return 1;
547
548	return 0;
549	}
550
551	static inline bool need_reserve_reloc_root(struct btrfs_root *root)
552	{
553	struct btrfs_fs_info *fs_info = root->fs_info;
554
555	if (!fs_info->reloc_ctl ||
556	!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
557	root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
558	root->reloc_root)
559	return false;
560
561	return true;
562	}
563
564	static int btrfs_reserve_trans_metadata(struct btrfs_fs_info *fs_info,
565	enum btrfs_reserve_flush_enum flush,
566	u64 num_bytes,
567	u64 *delayed_refs_bytes)
568	{
569	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
570	struct btrfs_space_info *si = fs_info->trans_block_rsv.space_info;
571	u64 extra_delayed_refs_bytes = 0;
572	u64 bytes;
573	int ret;
574
575	/*
576	* If there's a gap between the size of the delayed refs reserve and
577	* its reserved space, than some tasks have added delayed refs or bumped
578	* its size otherwise (due to block group creation or removal, or block
579	* group item update). Also try to allocate that gap in order to prevent
580	* using (and possibly abusing) the global reserve when committing the
581	* transaction.
582	*/
583	if (flush == BTRFS_RESERVE_FLUSH_ALL &&
584	!btrfs_block_rsv_full(rsv: delayed_refs_rsv)) {
585	spin_lock(lock: &delayed_refs_rsv->lock);
586	if (delayed_refs_rsv->size > delayed_refs_rsv->reserved)
587	extra_delayed_refs_bytes = delayed_refs_rsv->size -
588	delayed_refs_rsv->reserved;
589	spin_unlock(lock: &delayed_refs_rsv->lock);
590	}
591
592	bytes = num_bytes + *delayed_refs_bytes + extra_delayed_refs_bytes;
593
594	/*
595	* We want to reserve all the bytes we may need all at once, so we only
596	* do 1 enospc flushing cycle per transaction start.
597	*/
598	ret = btrfs_reserve_metadata_bytes(fs_info, space_info: si, orig_bytes: bytes, flush);
599	if (ret == 0) {
600	if (extra_delayed_refs_bytes > 0)
601	btrfs_migrate_to_delayed_refs_rsv(fs_info,
602	num_bytes: extra_delayed_refs_bytes);
603	return 0;
604	}
605
606	if (extra_delayed_refs_bytes > 0) {
607	bytes -= extra_delayed_refs_bytes;
608	ret = btrfs_reserve_metadata_bytes(fs_info, space_info: si, orig_bytes: bytes, flush);
609	if (ret == 0)
610	return 0;
611	}
612
613	/*
614	* If we are an emergency flush, which can steal from the global block
615	* reserve, then attempt to not reserve space for the delayed refs, as
616	* we will consume space for them from the global block reserve.
617	*/
618	if (flush == BTRFS_RESERVE_FLUSH_ALL_STEAL) {
619	bytes -= *delayed_refs_bytes;
620	*delayed_refs_bytes = 0;
621	ret = btrfs_reserve_metadata_bytes(fs_info, space_info: si, orig_bytes: bytes, flush);
622	}
623
624	return ret;
625	}
626
627	static struct btrfs_trans_handle *
628	start_transaction(struct btrfs_root *root, unsigned int num_items,
629	unsigned int type, enum btrfs_reserve_flush_enum flush,
630	bool enforce_qgroups)
631	{
632	struct btrfs_fs_info *fs_info = root->fs_info;
633	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
634	struct btrfs_block_rsv *trans_rsv = &fs_info->trans_block_rsv;
635	struct btrfs_trans_handle *h;
636	struct btrfs_transaction *cur_trans;
637	u64 num_bytes = 0;
638	u64 qgroup_reserved = 0;
639	u64 delayed_refs_bytes = 0;
640	bool reloc_reserved = false;
641	bool do_chunk_alloc = false;
642	int ret;
643
644	if (BTRFS_FS_ERROR(fs_info))
645	return ERR_PTR(error: -EROFS);
646
647	if (current->journal_info) {
648	WARN_ON(type & TRANS_EXTWRITERS);
649	h = current->journal_info;
650	refcount_inc(r: &h->use_count);
651	WARN_ON(refcount_read(&h->use_count) > 2);
652	h->orig_rsv = h->block_rsv;
653	h->block_rsv = NULL;
654	goto got_it;
655	}
656
657	/*
658	* Do the reservation before we join the transaction so we can do all
659	* the appropriate flushing if need be.
660	*/
661	if (num_items && root != fs_info->chunk_root) {
662	qgroup_reserved = num_items * fs_info->nodesize;
663	/*
664	* Use prealloc for now, as there might be a currently running
665	* transaction that could free this reserved space prematurely
666	* by committing.
667	*/
668	ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes: qgroup_reserved,
669	enforce: enforce_qgroups, noflush: false);
670	if (ret)
671	return ERR_PTR(error: ret);
672
673	num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
674	/*
675	* If we plan to insert/update/delete "num_items" from a btree,
676	* we will also generate delayed refs for extent buffers in the
677	* respective btree paths, so reserve space for the delayed refs
678	* that will be generated by the caller as it modifies btrees.
679	* Try to reserve them to avoid excessive use of the global
680	* block reserve.
681	*/
682	delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, num_delayed_refs: num_items);
683
684	/*
685	* Do the reservation for the relocation root creation
686	*/
687	if (need_reserve_reloc_root(root)) {
688	num_bytes += fs_info->nodesize;
689	reloc_reserved = true;
690	}
691
692	ret = btrfs_reserve_trans_metadata(fs_info, flush, num_bytes,
693	delayed_refs_bytes: &delayed_refs_bytes);
694	if (ret)
695	goto reserve_fail;
696
697	btrfs_block_rsv_add_bytes(block_rsv: trans_rsv, num_bytes, update_size: true);
698
699	if (trans_rsv->space_info->force_alloc)
700	do_chunk_alloc = true;
701	} else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
702	!btrfs_block_rsv_full(rsv: delayed_refs_rsv)) {
703	/*
704	* Some people call with btrfs_start_transaction(root, 0)
705	* because they can be throttled, but have some other mechanism
706	* for reserving space. We still want these guys to refill the
707	* delayed block_rsv so just add 1 items worth of reservation
708	* here.
709	*/
710	ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
711	if (ret)
712	goto reserve_fail;
713	}
714	again:
715	h = kmem_cache_zalloc(k: btrfs_trans_handle_cachep, GFP_NOFS);
716	if (!h) {
717	ret = -ENOMEM;
718	goto alloc_fail;
719	}
720
721	/*
722	* If we are JOIN_NOLOCK we're already committing a transaction and
723	* waiting on this guy, so we don't need to do the sb_start_intwrite
724	* because we're already holding a ref. We need this because we could
725	* have raced in and did an fsync() on a file which can kick a commit
726	* and then we deadlock with somebody doing a freeze.
727	*
728	* If we are ATTACH, it means we just want to catch the current
729	* transaction and commit it, so we needn't do sb_start_intwrite().
730	*/
731	if (type & __TRANS_FREEZABLE)
732	sb_start_intwrite(sb: fs_info->sb);
733
734	if (may_wait_transaction(fs_info, type))
735	wait_current_trans(fs_info);
736
737	do {
738	ret = join_transaction(fs_info, type);
739	if (ret == -EBUSY) {
740	wait_current_trans(fs_info);
741	if (unlikely(type == TRANS_ATTACH ||
742	type == TRANS_JOIN_NOSTART))
743	ret = -ENOENT;
744	}
745	} while (ret == -EBUSY);
746
747	if (ret < 0)
748	goto join_fail;
749
750	cur_trans = fs_info->running_transaction;
751
752	h->transid = cur_trans->transid;
753	h->transaction = cur_trans;
754	refcount_set(r: &h->use_count, n: 1);
755	h->fs_info = root->fs_info;
756
757	h->type = type;
758	INIT_LIST_HEAD(list: &h->new_bgs);
759	btrfs_init_metadata_block_rsv(fs_info, rsv: &h->delayed_rsv, type: BTRFS_BLOCK_RSV_DELOPS);
760
761	smp_mb();
762	if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
763	may_wait_transaction(fs_info, type)) {
764	current->journal_info = h;
765	btrfs_commit_transaction(trans: h);
766	goto again;
767	}
768
769	if (num_bytes) {
770	trace_btrfs_space_reservation(fs_info, type: "transaction",
771	val: h->transid, bytes: num_bytes, reserve: 1);
772	h->block_rsv = trans_rsv;
773	h->bytes_reserved = num_bytes;
774	if (delayed_refs_bytes > 0) {
775	trace_btrfs_space_reservation(fs_info,
776	type: "local_delayed_refs_rsv",
777	val: h->transid,
778	bytes: delayed_refs_bytes, reserve: 1);
779	h->delayed_refs_bytes_reserved = delayed_refs_bytes;
780	btrfs_block_rsv_add_bytes(block_rsv: &h->delayed_rsv, num_bytes: delayed_refs_bytes, update_size: true);
781	delayed_refs_bytes = 0;
782	}
783	h->reloc_reserved = reloc_reserved;
784	}
785
786	/*
787	* Now that we have found a transaction to be a part of, convert the
788	* qgroup reservation from prealloc to pertrans. A different transaction
789	* can't race in and free our pertrans out from under us.
790	*/
791	if (qgroup_reserved)
792	btrfs_qgroup_convert_reserved_meta(root, num_bytes: qgroup_reserved);
793
794	got_it:
795	if (!current->journal_info)
796	current->journal_info = h;
797
798	/*
799	* If the space_info is marked ALLOC_FORCE then we'll get upgraded to
800	* ALLOC_FORCE the first run through, and then we won't allocate for
801	* anybody else who races in later. We don't care about the return
802	* value here.
803	*/
804	if (do_chunk_alloc && num_bytes) {
805	u64 flags = h->block_rsv->space_info->flags;
806
807	btrfs_chunk_alloc(trans: h, flags: btrfs_get_alloc_profile(fs_info, orig_flags: flags),
808	force: CHUNK_ALLOC_NO_FORCE);
809	}
810
811	/*
812	* btrfs_record_root_in_trans() needs to alloc new extents, and may
813	* call btrfs_join_transaction() while we're also starting a
814	* transaction.
815	*
816	* Thus it need to be called after current->journal_info initialized,
817	* or we can deadlock.
818	*/
819	ret = btrfs_record_root_in_trans(trans: h, root);
820	if (ret) {
821	/*
822	* The transaction handle is fully initialized and linked with
823	* other structures so it needs to be ended in case of errors,
824	* not just freed.
825	*/
826	btrfs_end_transaction(trans: h);
827	return ERR_PTR(error: ret);
828	}
829
830	return h;
831
832	join_fail:
833	if (type & __TRANS_FREEZABLE)
834	sb_end_intwrite(sb: fs_info->sb);
835	kmem_cache_free(s: btrfs_trans_handle_cachep, objp: h);
836	alloc_fail:
837	if (num_bytes)
838	btrfs_block_rsv_release(fs_info, block_rsv: trans_rsv, num_bytes, NULL);
839	if (delayed_refs_bytes)
840	btrfs_space_info_free_bytes_may_use(fs_info, space_info: trans_rsv->space_info,
841	num_bytes: delayed_refs_bytes);
842	reserve_fail:
843	btrfs_qgroup_free_meta_prealloc(root, num_bytes: qgroup_reserved);
844	return ERR_PTR(error: ret);
845	}
846
847	struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
848	unsigned int num_items)
849	{
850	return start_transaction(root, num_items, TRANS_START,
851	flush: BTRFS_RESERVE_FLUSH_ALL, enforce_qgroups: true);
852	}
853
854	struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
855	struct btrfs_root *root,
856	unsigned int num_items)
857	{
858	return start_transaction(root, num_items, TRANS_START,
859	flush: BTRFS_RESERVE_FLUSH_ALL_STEAL, enforce_qgroups: false);
860	}
861
862	struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
863	{
864	return start_transaction(root, num_items: 0, TRANS_JOIN, flush: BTRFS_RESERVE_NO_FLUSH,
865	enforce_qgroups: true);
866	}
867
868	struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
869	{
870	return start_transaction(root, num_items: 0, TRANS_JOIN_NOLOCK,
871	flush: BTRFS_RESERVE_NO_FLUSH, enforce_qgroups: true);
872	}
873
874	/*
875	* Similar to regular join but it never starts a transaction when none is
876	* running or when there's a running one at a state >= TRANS_STATE_UNBLOCKED.
877	* This is similar to btrfs_attach_transaction() but it allows the join to
878	* happen if the transaction commit already started but it's not yet in the
879	* "doing" phase (the state is < TRANS_STATE_COMMIT_DOING).
880	*/
881	struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
882	{
883	return start_transaction(root, num_items: 0, TRANS_JOIN_NOSTART,
884	flush: BTRFS_RESERVE_NO_FLUSH, enforce_qgroups: true);
885	}
886
887	/*
888	* Catch the running transaction.
889	*
890	* It is used when we want to commit the current the transaction, but
891	* don't want to start a new one.
892	*
893	* Note: If this function return -ENOENT, it just means there is no
894	* running transaction. But it is possible that the inactive transaction
895	* is still in the memory, not fully on disk. If you hope there is no
896	* inactive transaction in the fs when -ENOENT is returned, you should
897	* invoke
898	* btrfs_attach_transaction_barrier()
899	*/
900	struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
901	{
902	return start_transaction(root, num_items: 0, TRANS_ATTACH,
903	flush: BTRFS_RESERVE_NO_FLUSH, enforce_qgroups: true);
904	}
905
906	/*
907	* Catch the running transaction.
908	*
909	* It is similar to the above function, the difference is this one
910	* will wait for all the inactive transactions until they fully
911	* complete.
912	*/
913	struct btrfs_trans_handle *
914	btrfs_attach_transaction_barrier(struct btrfs_root *root)
915	{
916	struct btrfs_trans_handle *trans;
917
918	trans = start_transaction(root, num_items: 0, TRANS_ATTACH,
919	flush: BTRFS_RESERVE_NO_FLUSH, enforce_qgroups: true);
920	if (trans == ERR_PTR(error: -ENOENT)) {
921	int ret;
922
923	ret = btrfs_wait_for_commit(fs_info: root->fs_info, transid: 0);
924	if (ret)
925	return ERR_PTR(error: ret);
926	}
927
928	return trans;
929	}
930
931	/* Wait for a transaction commit to reach at least the given state. */
932	static noinline void wait_for_commit(struct btrfs_transaction *commit,
933	const enum btrfs_trans_state min_state)
934	{
935	struct btrfs_fs_info *fs_info = commit->fs_info;
936	u64 transid = commit->transid;
937	bool put = false;
938
939	/*
940	* At the moment this function is called with min_state either being
941	* TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED.
942	*/
943	if (min_state == TRANS_STATE_COMPLETED)
944	btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
945	else
946	btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
947
948	while (1) {
949	wait_event(commit->commit_wait, commit->state >= min_state);
950	if (put)
951	btrfs_put_transaction(transaction: commit);
952
953	if (min_state < TRANS_STATE_COMPLETED)
954	break;
955
956	/*
957	* A transaction isn't really completed until all of the
958	* previous transactions are completed, but with fsync we can
959	* end up with SUPER_COMMITTED transactions before a COMPLETED
960	* transaction. Wait for those.
961	*/
962
963	spin_lock(lock: &fs_info->trans_lock);
964	commit = list_first_entry_or_null(&fs_info->trans_list,
965	struct btrfs_transaction,
966	list);
967	if (!commit || commit->transid > transid) {
968	spin_unlock(lock: &fs_info->trans_lock);
969	break;
970	}
971	refcount_inc(r: &commit->use_count);
972	put = true;
973	spin_unlock(lock: &fs_info->trans_lock);
974	}
975	}
976
977	int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
978	{
979	struct btrfs_transaction *cur_trans = NULL, *t;
980	int ret = 0;
981
982	if (transid) {
983	if (transid <= btrfs_get_last_trans_committed(fs_info))
984	goto out;
985
986	/* find specified transaction */
987	spin_lock(lock: &fs_info->trans_lock);
988	list_for_each_entry(t, &fs_info->trans_list, list) {
989	if (t->transid == transid) {
990	cur_trans = t;
991	refcount_inc(r: &cur_trans->use_count);
992	ret = 0;
993	break;
994	}
995	if (t->transid > transid) {
996	ret = 0;
997	break;
998	}
999	}
1000	spin_unlock(lock: &fs_info->trans_lock);
1001
1002	/*
1003	* The specified transaction doesn't exist, or we
1004	* raced with btrfs_commit_transaction
1005	*/
1006	if (!cur_trans) {
1007	if (transid > btrfs_get_last_trans_committed(fs_info))
1008	ret = -EINVAL;
1009	goto out;
1010	}
1011	} else {
1012	/* find newest transaction that is committing | committed */
1013	spin_lock(lock: &fs_info->trans_lock);
1014	list_for_each_entry_reverse(t, &fs_info->trans_list,
1015	list) {
1016	if (t->state >= TRANS_STATE_COMMIT_START) {
1017	if (t->state == TRANS_STATE_COMPLETED)
1018	break;
1019	cur_trans = t;
1020	refcount_inc(r: &cur_trans->use_count);
1021	break;
1022	}
1023	}
1024	spin_unlock(lock: &fs_info->trans_lock);
1025	if (!cur_trans)
1026	goto out; /* nothing committing|committed */
1027	}
1028
1029	wait_for_commit(commit: cur_trans, min_state: TRANS_STATE_COMPLETED);
1030	ret = cur_trans->aborted;
1031	btrfs_put_transaction(transaction: cur_trans);
1032	out:
1033	return ret;
1034	}
1035
1036	void btrfs_throttle(struct btrfs_fs_info *fs_info)
1037	{
1038	wait_current_trans(fs_info);
1039	}
1040
1041	bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
1042	{
1043	struct btrfs_transaction *cur_trans = trans->transaction;
1044
1045	if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
1046	test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
1047	return true;
1048
1049	if (btrfs_check_space_for_delayed_refs(fs_info: trans->fs_info))
1050	return true;
1051
1052	return !!btrfs_block_rsv_check(block_rsv: &trans->fs_info->global_block_rsv, min_percent: 50);
1053	}
1054
1055	static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
1056
1057	{
1058	struct btrfs_fs_info *fs_info = trans->fs_info;
1059
1060	if (!trans->block_rsv) {
1061	ASSERT(!trans->bytes_reserved);
1062	ASSERT(!trans->delayed_refs_bytes_reserved);
1063	return;
1064	}
1065
1066	if (!trans->bytes_reserved) {
1067	ASSERT(!trans->delayed_refs_bytes_reserved);
1068	return;
1069	}
1070
1071	ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
1072	trace_btrfs_space_reservation(fs_info, type: "transaction",
1073	val: trans->transid, bytes: trans->bytes_reserved, reserve: 0);
1074	btrfs_block_rsv_release(fs_info, block_rsv: trans->block_rsv,
1075	num_bytes: trans->bytes_reserved, NULL);
1076	trans->bytes_reserved = 0;
1077
1078	if (!trans->delayed_refs_bytes_reserved)
1079	return;
1080
1081	trace_btrfs_space_reservation(fs_info, type: "local_delayed_refs_rsv",
1082	val: trans->transid,
1083	bytes: trans->delayed_refs_bytes_reserved, reserve: 0);
1084	btrfs_block_rsv_release(fs_info, block_rsv: &trans->delayed_rsv,
1085	num_bytes: trans->delayed_refs_bytes_reserved, NULL);
1086	trans->delayed_refs_bytes_reserved = 0;
1087	}
1088
1089	static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
1090	int throttle)
1091	{
1092	struct btrfs_fs_info *info = trans->fs_info;
1093	struct btrfs_transaction *cur_trans = trans->transaction;
1094	int err = 0;
1095
1096	if (refcount_read(r: &trans->use_count) > 1) {
1097	refcount_dec(r: &trans->use_count);
1098	trans->block_rsv = trans->orig_rsv;
1099	return 0;
1100	}
1101
1102	btrfs_trans_release_metadata(trans);
1103	trans->block_rsv = NULL;
1104
1105	btrfs_create_pending_block_groups(trans);
1106
1107	btrfs_trans_release_chunk_metadata(trans);
1108
1109	if (trans->type & __TRANS_FREEZABLE)
1110	sb_end_intwrite(sb: info->sb);
1111
1112	WARN_ON(cur_trans != info->running_transaction);
1113	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
1114	atomic_dec(v: &cur_trans->num_writers);
1115	extwriter_counter_dec(trans: cur_trans, type: trans->type);
1116
1117	cond_wake_up(wq: &cur_trans->writer_wait);
1118
1119	btrfs_lockdep_release(info, btrfs_trans_num_extwriters);
1120	btrfs_lockdep_release(info, btrfs_trans_num_writers);
1121
1122	btrfs_put_transaction(transaction: cur_trans);
1123
1124	if (current->journal_info == trans)
1125	current->journal_info = NULL;
1126
1127	if (throttle)
1128	btrfs_run_delayed_iputs(fs_info: info);
1129
1130	if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) {
1131	wake_up_process(tsk: info->transaction_kthread);
1132	if (TRANS_ABORTED(trans))
1133	err = trans->aborted;
1134	else
1135	err = -EROFS;
1136	}
1137
1138	kmem_cache_free(s: btrfs_trans_handle_cachep, objp: trans);
1139	return err;
1140	}
1141
1142	int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1143	{
1144	return __btrfs_end_transaction(trans, throttle: 0);
1145	}
1146
1147	int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1148	{
1149	return __btrfs_end_transaction(trans, throttle: 1);
1150	}
1151
1152	/*
1153	* when btree blocks are allocated, they have some corresponding bits set for
1154	* them in one of two extent_io trees. This is used to make sure all of
1155	* those extents are sent to disk but does not wait on them
1156	*/
1157	int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1158	struct extent_io_tree *dirty_pages, int mark)
1159	{
1160	int err = 0;
1161	int werr = 0;
1162	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1163	struct extent_state *cached_state = NULL;
1164	u64 start = 0;
1165	u64 end;
1166
1167	while (find_first_extent_bit(tree: dirty_pages, start, start_ret: &start, end_ret: &end,
1168	bits: mark, cached_state: &cached_state)) {
1169	bool wait_writeback = false;
1170
1171	err = convert_extent_bit(tree: dirty_pages, start, end,
1172	bits: EXTENT_NEED_WAIT,
1173	clear_bits: mark, cached_state: &cached_state);
1174	/*
1175	* convert_extent_bit can return -ENOMEM, which is most of the
1176	* time a temporary error. So when it happens, ignore the error
1177	* and wait for writeback of this range to finish - because we
1178	* failed to set the bit EXTENT_NEED_WAIT for the range, a call
1179	* to __btrfs_wait_marked_extents() would not know that
1180	* writeback for this range started and therefore wouldn't
1181	* wait for it to finish - we don't want to commit a
1182	* superblock that points to btree nodes/leafs for which
1183	* writeback hasn't finished yet (and without errors).
1184	* We cleanup any entries left in the io tree when committing
1185	* the transaction (through extent_io_tree_release()).
1186	*/
1187	if (err == -ENOMEM) {
1188	err = 0;
1189	wait_writeback = true;
1190	}
1191	if (!err)
1192	err = filemap_fdatawrite_range(mapping, start, end);
1193	if (err)
1194	werr = err;
1195	else if (wait_writeback)
1196	werr = filemap_fdatawait_range(mapping, lstart: start, lend: end);
1197	free_extent_state(state: cached_state);
1198	cached_state = NULL;
1199	cond_resched();
1200	start = end + 1;
1201	}
1202	return werr;
1203	}
1204
1205	/*
1206	* when btree blocks are allocated, they have some corresponding bits set for
1207	* them in one of two extent_io trees. This is used to make sure all of
1208	* those extents are on disk for transaction or log commit. We wait
1209	* on all the pages and clear them from the dirty pages state tree
1210	*/
1211	static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1212	struct extent_io_tree *dirty_pages)
1213	{
1214	int err = 0;
1215	int werr = 0;
1216	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1217	struct extent_state *cached_state = NULL;
1218	u64 start = 0;
1219	u64 end;
1220
1221	while (find_first_extent_bit(tree: dirty_pages, start, start_ret: &start, end_ret: &end,
1222	bits: EXTENT_NEED_WAIT, cached_state: &cached_state)) {
1223	/*
1224	* Ignore -ENOMEM errors returned by clear_extent_bit().
1225	* When committing the transaction, we'll remove any entries
1226	* left in the io tree. For a log commit, we don't remove them
1227	* after committing the log because the tree can be accessed
1228	* concurrently - we do it only at transaction commit time when
1229	* it's safe to do it (through extent_io_tree_release()).
1230	*/
1231	err = clear_extent_bit(tree: dirty_pages, start, end,
1232	bits: EXTENT_NEED_WAIT, cached: &cached_state);
1233	if (err == -ENOMEM)
1234	err = 0;
1235	if (!err)
1236	err = filemap_fdatawait_range(mapping, lstart: start, lend: end);
1237	if (err)
1238	werr = err;
1239	free_extent_state(state: cached_state);
1240	cached_state = NULL;
1241	cond_resched();
1242	start = end + 1;
1243	}
1244	if (err)
1245	werr = err;
1246	return werr;
1247	}
1248
1249	static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1250	struct extent_io_tree *dirty_pages)
1251	{
1252	bool errors = false;
1253	int err;
1254
1255	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1256	if (test_and_clear_bit(nr: BTRFS_FS_BTREE_ERR, addr: &fs_info->flags))
1257	errors = true;
1258
1259	if (errors && !err)
1260	err = -EIO;
1261	return err;
1262	}
1263
1264	int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1265	{
1266	struct btrfs_fs_info *fs_info = log_root->fs_info;
1267	struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1268	bool errors = false;
1269	int err;
1270
1271	ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1272
1273	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1274	if ((mark & EXTENT_DIRTY) &&
1275	test_and_clear_bit(nr: BTRFS_FS_LOG1_ERR, addr: &fs_info->flags))
1276	errors = true;
1277
1278	if ((mark & EXTENT_NEW) &&
1279	test_and_clear_bit(nr: BTRFS_FS_LOG2_ERR, addr: &fs_info->flags))
1280	errors = true;
1281
1282	if (errors && !err)
1283	err = -EIO;
1284	return err;
1285	}
1286
1287	/*
1288	* When btree blocks are allocated the corresponding extents are marked dirty.
1289	* This function ensures such extents are persisted on disk for transaction or
1290	* log commit.
1291	*
1292	* @trans: transaction whose dirty pages we'd like to write
1293	*/
1294	static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1295	{
1296	int ret;
1297	int ret2;
1298	struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1299	struct btrfs_fs_info *fs_info = trans->fs_info;
1300	struct blk_plug plug;
1301
1302	blk_start_plug(&plug);
1303	ret = btrfs_write_marked_extents(fs_info, dirty_pages, mark: EXTENT_DIRTY);
1304	blk_finish_plug(&plug);
1305	ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1306
1307	extent_io_tree_release(tree: &trans->transaction->dirty_pages);
1308
1309	if (ret)
1310	return ret;
1311	else if (ret2)
1312	return ret2;
1313	else
1314	return 0;
1315	}
1316
1317	/*
1318	* this is used to update the root pointer in the tree of tree roots.
1319	*
1320	* But, in the case of the extent allocation tree, updating the root
1321	* pointer may allocate blocks which may change the root of the extent
1322	* allocation tree.
1323	*
1324	* So, this loops and repeats and makes sure the cowonly root didn't
1325	* change while the root pointer was being updated in the metadata.
1326	*/
1327	static int update_cowonly_root(struct btrfs_trans_handle *trans,
1328	struct btrfs_root *root)
1329	{
1330	int ret;
1331	u64 old_root_bytenr;
1332	u64 old_root_used;
1333	struct btrfs_fs_info *fs_info = root->fs_info;
1334	struct btrfs_root *tree_root = fs_info->tree_root;
1335
1336	old_root_used = btrfs_root_used(s: &root->root_item);
1337
1338	while (1) {
1339	old_root_bytenr = btrfs_root_bytenr(s: &root->root_item);
1340	if (old_root_bytenr == root->node->start &&
1341	old_root_used == btrfs_root_used(s: &root->root_item))
1342	break;
1343
1344	btrfs_set_root_node(item: &root->root_item, node: root->node);
1345	ret = btrfs_update_root(trans, root: tree_root,
1346	key: &root->root_key,
1347	item: &root->root_item);
1348	if (ret)
1349	return ret;
1350
1351	old_root_used = btrfs_root_used(s: &root->root_item);
1352	}
1353
1354	return 0;
1355	}
1356
1357	/*
1358	* update all the cowonly tree roots on disk
1359	*
1360	* The error handling in this function may not be obvious. Any of the
1361	* failures will cause the file system to go offline. We still need
1362	* to clean up the delayed refs.
1363	*/
1364	static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1365	{
1366	struct btrfs_fs_info *fs_info = trans->fs_info;
1367	struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1368	struct list_head *io_bgs = &trans->transaction->io_bgs;
1369	struct list_head *next;
1370	struct extent_buffer *eb;
1371	int ret;
1372
1373	/*
1374	* At this point no one can be using this transaction to modify any tree
1375	* and no one can start another transaction to modify any tree either.
1376	*/
1377	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1378
1379	eb = btrfs_lock_root_node(root: fs_info->tree_root);
1380	ret = btrfs_cow_block(trans, root: fs_info->tree_root, buf: eb, NULL,
1381	parent_slot: 0, cow_ret: &eb, nest: BTRFS_NESTING_COW);
1382	btrfs_tree_unlock(eb);
1383	free_extent_buffer(eb);
1384
1385	if (ret)
1386	return ret;
1387
1388	ret = btrfs_run_dev_stats(trans);
1389	if (ret)
1390	return ret;
1391	ret = btrfs_run_dev_replace(trans);
1392	if (ret)
1393	return ret;
1394	ret = btrfs_run_qgroups(trans);
1395	if (ret)
1396	return ret;
1397
1398	ret = btrfs_setup_space_cache(trans);
1399	if (ret)
1400	return ret;
1401
1402	again:
1403	while (!list_empty(head: &fs_info->dirty_cowonly_roots)) {
1404	struct btrfs_root *root;
1405	next = fs_info->dirty_cowonly_roots.next;
1406	list_del_init(entry: next);
1407	root = list_entry(next, struct btrfs_root, dirty_list);
1408	clear_bit(nr: BTRFS_ROOT_DIRTY, addr: &root->state);
1409
1410	list_add_tail(new: &root->dirty_list,
1411	head: &trans->transaction->switch_commits);
1412	ret = update_cowonly_root(trans, root);
1413	if (ret)
1414	return ret;
1415	}
1416
1417	/* Now flush any delayed refs generated by updating all of the roots */
1418	ret = btrfs_run_delayed_refs(trans, U64_MAX);
1419	if (ret)
1420	return ret;
1421
1422	while (!list_empty(head: dirty_bgs) || !list_empty(head: io_bgs)) {
1423	ret = btrfs_write_dirty_block_groups(trans);
1424	if (ret)
1425	return ret;
1426
1427	/*
1428	* We're writing the dirty block groups, which could generate
1429	* delayed refs, which could generate more dirty block groups,
1430	* so we want to keep this flushing in this loop to make sure
1431	* everything gets run.
1432	*/
1433	ret = btrfs_run_delayed_refs(trans, U64_MAX);
1434	if (ret)
1435	return ret;
1436	}
1437
1438	if (!list_empty(head: &fs_info->dirty_cowonly_roots))
1439	goto again;
1440
1441	/* Update dev-replace pointer once everything is committed */
1442	fs_info->dev_replace.committed_cursor_left =
1443	fs_info->dev_replace.cursor_left_last_write_of_item;
1444
1445	return 0;
1446	}
1447
1448	/*
1449	* If we had a pending drop we need to see if there are any others left in our
1450	* dead roots list, and if not clear our bit and wake any waiters.
1451	*/
1452	void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
1453	{
1454	/*
1455	* We put the drop in progress roots at the front of the list, so if the
1456	* first entry doesn't have UNFINISHED_DROP set we can wake everybody
1457	* up.
1458	*/
1459	spin_lock(lock: &fs_info->trans_lock);
1460	if (!list_empty(head: &fs_info->dead_roots)) {
1461	struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
1462	struct btrfs_root,
1463	root_list);
1464	if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
1465	spin_unlock(lock: &fs_info->trans_lock);
1466	return;
1467	}
1468	}
1469	spin_unlock(lock: &fs_info->trans_lock);
1470
1471	btrfs_wake_unfinished_drop(fs_info);
1472	}
1473
1474	/*
1475	* dead roots are old snapshots that need to be deleted. This allocates
1476	* a dirty root struct and adds it into the list of dead roots that need to
1477	* be deleted
1478	*/
1479	void btrfs_add_dead_root(struct btrfs_root *root)
1480	{
1481	struct btrfs_fs_info *fs_info = root->fs_info;
1482
1483	spin_lock(lock: &fs_info->trans_lock);
1484	if (list_empty(head: &root->root_list)) {
1485	btrfs_grab_root(root);
1486
1487	/* We want to process the partially complete drops first. */
1488	if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
1489	list_add(new: &root->root_list, head: &fs_info->dead_roots);
1490	else
1491	list_add_tail(new: &root->root_list, head: &fs_info->dead_roots);
1492	}
1493	spin_unlock(lock: &fs_info->trans_lock);
1494	}
1495
1496	/*
1497	* Update each subvolume root and its relocation root, if it exists, in the tree
1498	* of tree roots. Also free log roots if they exist.
1499	*/
1500	static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1501	{
1502	struct btrfs_fs_info *fs_info = trans->fs_info;
1503	struct btrfs_root *gang[8];
1504	int i;
1505	int ret;
1506
1507	/*
1508	* At this point no one can be using this transaction to modify any tree
1509	* and no one can start another transaction to modify any tree either.
1510	*/
1511	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1512
1513	spin_lock(lock: &fs_info->fs_roots_radix_lock);
1514	while (1) {
1515	ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1516	results: (void **)gang, first_index: 0,
1517	ARRAY_SIZE(gang),
1518	BTRFS_ROOT_TRANS_TAG);
1519	if (ret == 0)
1520	break;
1521	for (i = 0; i < ret; i++) {
1522	struct btrfs_root *root = gang[i];
1523	int ret2;
1524
1525	/*
1526	* At this point we can neither have tasks logging inodes
1527	* from a root nor trying to commit a log tree.
1528	*/
1529	ASSERT(atomic_read(&root->log_writers) == 0);
1530	ASSERT(atomic_read(&root->log_commit[0]) == 0);
1531	ASSERT(atomic_read(&root->log_commit[1]) == 0);
1532
1533	radix_tree_tag_clear(&fs_info->fs_roots_radix,
1534	index: (unsigned long)root->root_key.objectid,
1535	BTRFS_ROOT_TRANS_TAG);
1536	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
1537
1538	btrfs_free_log(trans, root);
1539	ret2 = btrfs_update_reloc_root(trans, root);
1540	if (ret2)
1541	return ret2;
1542
1543	/* see comments in should_cow_block() */
1544	clear_bit(nr: BTRFS_ROOT_FORCE_COW, addr: &root->state);
1545	smp_mb__after_atomic();
1546
1547	if (root->commit_root != root->node) {
1548	list_add_tail(new: &root->dirty_list,
1549	head: &trans->transaction->switch_commits);
1550	btrfs_set_root_node(item: &root->root_item,
1551	node: root->node);
1552	}
1553
1554	ret2 = btrfs_update_root(trans, root: fs_info->tree_root,
1555	key: &root->root_key,
1556	item: &root->root_item);
1557	if (ret2)
1558	return ret2;
1559	spin_lock(lock: &fs_info->fs_roots_radix_lock);
1560	btrfs_qgroup_free_meta_all_pertrans(root);
1561	}
1562	}
1563	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
1564	return 0;
1565	}
1566
1567	/*
1568	* Do all special snapshot related qgroup dirty hack.
1569	*
1570	* Will do all needed qgroup inherit and dirty hack like switch commit
1571	* roots inside one transaction and write all btree into disk, to make
1572	* qgroup works.
1573	*/
1574	static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1575	struct btrfs_root *src,
1576	struct btrfs_root *parent,
1577	struct btrfs_qgroup_inherit *inherit,
1578	u64 dst_objectid)
1579	{
1580	struct btrfs_fs_info *fs_info = src->fs_info;
1581	int ret;
1582
1583	/*
1584	* Save some performance in the case that qgroups are not enabled. If
1585	* this check races with the ioctl, rescan will kick in anyway.
1586	*/
1587	if (!btrfs_qgroup_full_accounting(fs_info))
1588	return 0;
1589
1590	/*
1591	* Ensure dirty @src will be committed. Or, after coming
1592	* commit_fs_roots() and switch_commit_roots(), any dirty but not
1593	* recorded root will never be updated again, causing an outdated root
1594	* item.
1595	*/
1596	ret = record_root_in_trans(trans, root: src, force: 1);
1597	if (ret)
1598	return ret;
1599
1600	/*
1601	* btrfs_qgroup_inherit relies on a consistent view of the usage for the
1602	* src root, so we must run the delayed refs here.
1603	*
1604	* However this isn't particularly fool proof, because there's no
1605	* synchronization keeping us from changing the tree after this point
1606	* before we do the qgroup_inherit, or even from making changes while
1607	* we're doing the qgroup_inherit. But that's a problem for the future,
1608	* for now flush the delayed refs to narrow the race window where the
1609	* qgroup counters could end up wrong.
1610	*/
1611	ret = btrfs_run_delayed_refs(trans, U64_MAX);
1612	if (ret) {
1613	btrfs_abort_transaction(trans, ret);
1614	return ret;
1615	}
1616
1617	ret = commit_fs_roots(trans);
1618	if (ret)
1619	goto out;
1620	ret = btrfs_qgroup_account_extents(trans);
1621	if (ret < 0)
1622	goto out;
1623
1624	/* Now qgroup are all updated, we can inherit it to new qgroups */
1625	ret = btrfs_qgroup_inherit(trans, srcid: src->root_key.objectid, objectid: dst_objectid,
1626	inode_rootid: parent->root_key.objectid, inherit);
1627	if (ret < 0)
1628	goto out;
1629
1630	/*
1631	* Now we do a simplified commit transaction, which will:
1632	* 1) commit all subvolume and extent tree
1633	* To ensure all subvolume and extent tree have a valid
1634	* commit_root to accounting later insert_dir_item()
1635	* 2) write all btree blocks onto disk
1636	* This is to make sure later btree modification will be cowed
1637	* Or commit_root can be populated and cause wrong qgroup numbers
1638	* In this simplified commit, we don't really care about other trees
1639	* like chunk and root tree, as they won't affect qgroup.
1640	* And we don't write super to avoid half committed status.
1641	*/
1642	ret = commit_cowonly_roots(trans);
1643	if (ret)
1644	goto out;
1645	switch_commit_roots(trans);
1646	ret = btrfs_write_and_wait_transaction(trans);
1647	if (ret)
1648	btrfs_handle_fs_error(fs_info, ret,
1649	"Error while writing out transaction for qgroup");
1650
1651	out:
1652	/*
1653	* Force parent root to be updated, as we recorded it before so its
1654	* last_trans == cur_transid.
1655	* Or it won't be committed again onto disk after later
1656	* insert_dir_item()
1657	*/
1658	if (!ret)
1659	ret = record_root_in_trans(trans, root: parent, force: 1);
1660	return ret;
1661	}
1662
1663	/*
1664	* new snapshots need to be created at a very specific time in the
1665	* transaction commit. This does the actual creation.
1666	*
1667	* Note:
1668	* If the error which may affect the commitment of the current transaction
1669	* happens, we should return the error number. If the error which just affect
1670	* the creation of the pending snapshots, just return 0.
1671	*/
1672	static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1673	struct btrfs_pending_snapshot *pending)
1674	{
1675
1676	struct btrfs_fs_info *fs_info = trans->fs_info;
1677	struct btrfs_key key;
1678	struct btrfs_root_item *new_root_item;
1679	struct btrfs_root *tree_root = fs_info->tree_root;
1680	struct btrfs_root *root = pending->root;
1681	struct btrfs_root *parent_root;
1682	struct btrfs_block_rsv *rsv;
1683	struct inode *parent_inode = pending->dir;
1684	struct btrfs_path *path;
1685	struct btrfs_dir_item *dir_item;
1686	struct extent_buffer *tmp;
1687	struct extent_buffer *old;
1688	struct timespec64 cur_time;
1689	int ret = 0;
1690	u64 to_reserve = 0;
1691	u64 index = 0;
1692	u64 objectid;
1693	u64 root_flags;
1694	unsigned int nofs_flags;
1695	struct fscrypt_name fname;
1696
1697	ASSERT(pending->path);
1698	path = pending->path;
1699
1700	ASSERT(pending->root_item);
1701	new_root_item = pending->root_item;
1702
1703	/*
1704	* We're inside a transaction and must make sure that any potential
1705	* allocations with GFP_KERNEL in fscrypt won't recurse back to
1706	* filesystem.
1707	*/
1708	nofs_flags = memalloc_nofs_save();
1709	pending->error = fscrypt_setup_filename(inode: parent_inode,
1710	iname: &pending->dentry->d_name, lookup: 0,
1711	fname: &fname);
1712	memalloc_nofs_restore(flags: nofs_flags);
1713	if (pending->error)
1714	goto free_pending;
1715
1716	pending->error = btrfs_get_free_objectid(root: tree_root, objectid: &objectid);
1717	if (pending->error)
1718	goto free_fname;
1719
1720	/*
1721	* Make qgroup to skip current new snapshot's qgroupid, as it is
1722	* accounted by later btrfs_qgroup_inherit().
1723	*/
1724	btrfs_set_skip_qgroup(trans, qgroupid: objectid);
1725
1726	btrfs_reloc_pre_snapshot(pending, bytes_to_reserve: &to_reserve);
1727
1728	if (to_reserve > 0) {
1729	pending->error = btrfs_block_rsv_add(fs_info,
1730	block_rsv: &pending->block_rsv,
1731	num_bytes: to_reserve,
1732	flush: BTRFS_RESERVE_NO_FLUSH);
1733	if (pending->error)
1734	goto clear_skip_qgroup;
1735	}
1736
1737	key.objectid = objectid;
1738	key.offset = (u64)-1;
1739	key.type = BTRFS_ROOT_ITEM_KEY;
1740
1741	rsv = trans->block_rsv;
1742	trans->block_rsv = &pending->block_rsv;
1743	trans->bytes_reserved = trans->block_rsv->reserved;
1744	trace_btrfs_space_reservation(fs_info, type: "transaction",
1745	val: trans->transid,
1746	bytes: trans->bytes_reserved, reserve: 1);
1747	parent_root = BTRFS_I(inode: parent_inode)->root;
1748	ret = record_root_in_trans(trans, root: parent_root, force: 0);
1749	if (ret)
1750	goto fail;
1751	cur_time = current_time(inode: parent_inode);
1752
1753	/*
1754	* insert the directory item
1755	*/
1756	ret = btrfs_set_inode_index(dir: BTRFS_I(inode: parent_inode), index: &index);
1757	if (ret) {
1758	btrfs_abort_transaction(trans, ret);
1759	goto fail;
1760	}
1761
1762	/* check if there is a file/dir which has the same name. */
1763	dir_item = btrfs_lookup_dir_item(NULL, root: parent_root, path,
1764	dir: btrfs_ino(inode: BTRFS_I(inode: parent_inode)),
1765	name: &fname.disk_name, mod: 0);
1766	if (dir_item != NULL && !IS_ERR(ptr: dir_item)) {
1767	pending->error = -EEXIST;
1768	goto dir_item_existed;
1769	} else if (IS_ERR(ptr: dir_item)) {
1770	ret = PTR_ERR(ptr: dir_item);
1771	btrfs_abort_transaction(trans, ret);
1772	goto fail;
1773	}
1774	btrfs_release_path(p: path);
1775
1776	ret = btrfs_create_qgroup(trans, qgroupid: objectid);
1777	if (ret) {
1778	btrfs_abort_transaction(trans, ret);
1779	goto fail;
1780	}
1781
1782	/*
1783	* pull in the delayed directory update
1784	* and the delayed inode item
1785	* otherwise we corrupt the FS during
1786	* snapshot
1787	*/
1788	ret = btrfs_run_delayed_items(trans);
1789	if (ret) { /* Transaction aborted */
1790	btrfs_abort_transaction(trans, ret);
1791	goto fail;
1792	}
1793
1794	ret = record_root_in_trans(trans, root, force: 0);
1795	if (ret) {
1796	btrfs_abort_transaction(trans, ret);
1797	goto fail;
1798	}
1799	btrfs_set_root_last_snapshot(s: &root->root_item, val: trans->transid);
1800	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1801	btrfs_check_and_init_root_item(item: new_root_item);
1802
1803	root_flags = btrfs_root_flags(s: new_root_item);
1804	if (pending->readonly)
1805	root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1806	else
1807	root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1808	btrfs_set_root_flags(s: new_root_item, val: root_flags);
1809
1810	btrfs_set_root_generation_v2(s: new_root_item,
1811	val: trans->transid);
1812	generate_random_guid(guid: new_root_item->uuid);
1813	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1814	BTRFS_UUID_SIZE);
1815	if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1816	memset(new_root_item->received_uuid, 0,
1817	sizeof(new_root_item->received_uuid));
1818	memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1819	memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1820	btrfs_set_root_stransid(s: new_root_item, val: 0);
1821	btrfs_set_root_rtransid(s: new_root_item, val: 0);
1822	}
1823	btrfs_set_stack_timespec_sec(s: &new_root_item->otime, val: cur_time.tv_sec);
1824	btrfs_set_stack_timespec_nsec(s: &new_root_item->otime, val: cur_time.tv_nsec);
1825	btrfs_set_root_otransid(s: new_root_item, val: trans->transid);
1826
1827	old = btrfs_lock_root_node(root);
1828	ret = btrfs_cow_block(trans, root, buf: old, NULL, parent_slot: 0, cow_ret: &old,
1829	nest: BTRFS_NESTING_COW);
1830	if (ret) {
1831	btrfs_tree_unlock(eb: old);
1832	free_extent_buffer(eb: old);
1833	btrfs_abort_transaction(trans, ret);
1834	goto fail;
1835	}
1836
1837	ret = btrfs_copy_root(trans, root, buf: old, cow_ret: &tmp, new_root_objectid: objectid);
1838	/* clean up in any case */
1839	btrfs_tree_unlock(eb: old);
1840	free_extent_buffer(eb: old);
1841	if (ret) {
1842	btrfs_abort_transaction(trans, ret);
1843	goto fail;
1844	}
1845	/* see comments in should_cow_block() */
1846	set_bit(nr: BTRFS_ROOT_FORCE_COW, addr: &root->state);
1847	smp_wmb();
1848
1849	btrfs_set_root_node(item: new_root_item, node: tmp);
1850	/* record when the snapshot was created in key.offset */
1851	key.offset = trans->transid;
1852	ret = btrfs_insert_root(trans, root: tree_root, key: &key, item: new_root_item);
1853	btrfs_tree_unlock(eb: tmp);
1854	free_extent_buffer(eb: tmp);
1855	if (ret) {
1856	btrfs_abort_transaction(trans, ret);
1857	goto fail;
1858	}
1859
1860	/*
1861	* insert root back/forward references
1862	*/
1863	ret = btrfs_add_root_ref(trans, root_id: objectid,
1864	ref_id: parent_root->root_key.objectid,
1865	dirid: btrfs_ino(inode: BTRFS_I(inode: parent_inode)), sequence: index,
1866	name: &fname.disk_name);
1867	if (ret) {
1868	btrfs_abort_transaction(trans, ret);
1869	goto fail;
1870	}
1871
1872	key.offset = (u64)-1;
1873	pending->snap = btrfs_get_new_fs_root(fs_info, objectid, anon_dev: pending->anon_dev);
1874	if (IS_ERR(ptr: pending->snap)) {
1875	ret = PTR_ERR(ptr: pending->snap);
1876	pending->snap = NULL;
1877	btrfs_abort_transaction(trans, ret);
1878	goto fail;
1879	}
1880
1881	ret = btrfs_reloc_post_snapshot(trans, pending);
1882	if (ret) {
1883	btrfs_abort_transaction(trans, ret);
1884	goto fail;
1885	}
1886
1887	/*
1888	* Do special qgroup accounting for snapshot, as we do some qgroup
1889	* snapshot hack to do fast snapshot.
1890	* To co-operate with that hack, we do hack again.
1891	* Or snapshot will be greatly slowed down by a subtree qgroup rescan
1892	*/
1893	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL)
1894	ret = qgroup_account_snapshot(trans, src: root, parent: parent_root,
1895	inherit: pending->inherit, dst_objectid: objectid);
1896	else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
1897	ret = btrfs_qgroup_inherit(trans, srcid: root->root_key.objectid, objectid,
1898	inode_rootid: parent_root->root_key.objectid, inherit: pending->inherit);
1899	if (ret < 0)
1900	goto fail;
1901
1902	ret = btrfs_insert_dir_item(trans, name: &fname.disk_name,
1903	dir: BTRFS_I(inode: parent_inode), location: &key, BTRFS_FT_DIR,
1904	index);
1905	/* We have check then name at the beginning, so it is impossible. */
1906	BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1907	if (ret) {
1908	btrfs_abort_transaction(trans, ret);
1909	goto fail;
1910	}
1911
1912	btrfs_i_size_write(inode: BTRFS_I(inode: parent_inode), size: parent_inode->i_size +
1913	fname.disk_name.len * 2);
1914	inode_set_mtime_to_ts(inode: parent_inode,
1915	ts: inode_set_ctime_current(inode: parent_inode));
1916	ret = btrfs_update_inode_fallback(trans, inode: BTRFS_I(inode: parent_inode));
1917	if (ret) {
1918	btrfs_abort_transaction(trans, ret);
1919	goto fail;
1920	}
1921	ret = btrfs_uuid_tree_add(trans, uuid: new_root_item->uuid,
1922	BTRFS_UUID_KEY_SUBVOL,
1923	subid: objectid);
1924	if (ret) {
1925	btrfs_abort_transaction(trans, ret);
1926	goto fail;
1927	}
1928	if (!btrfs_is_empty_uuid(uuid: new_root_item->received_uuid)) {
1929	ret = btrfs_uuid_tree_add(trans, uuid: new_root_item->received_uuid,
1930	BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1931	subid: objectid);
1932	if (ret && ret != -EEXIST) {
1933	btrfs_abort_transaction(trans, ret);
1934	goto fail;
1935	}
1936	}
1937
1938	fail:
1939	pending->error = ret;
1940	dir_item_existed:
1941	trans->block_rsv = rsv;
1942	trans->bytes_reserved = 0;
1943	clear_skip_qgroup:
1944	btrfs_clear_skip_qgroup(trans);
1945	free_fname:
1946	fscrypt_free_filename(fname: &fname);
1947	free_pending:
1948	kfree(objp: new_root_item);
1949	pending->root_item = NULL;
1950	btrfs_free_path(p: path);
1951	pending->path = NULL;
1952
1953	return ret;
1954	}
1955
1956	/*
1957	* create all the snapshots we've scheduled for creation
1958	*/
1959	static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1960	{
1961	struct btrfs_pending_snapshot *pending, *next;
1962	struct list_head *head = &trans->transaction->pending_snapshots;
1963	int ret = 0;
1964
1965	list_for_each_entry_safe(pending, next, head, list) {
1966	list_del(entry: &pending->list);
1967	ret = create_pending_snapshot(trans, pending);
1968	if (ret)
1969	break;
1970	}
1971	return ret;
1972	}
1973
1974	static void update_super_roots(struct btrfs_fs_info *fs_info)
1975	{
1976	struct btrfs_root_item *root_item;
1977	struct btrfs_super_block *super;
1978
1979	super = fs_info->super_copy;
1980
1981	root_item = &fs_info->chunk_root->root_item;
1982	super->chunk_root = root_item->bytenr;
1983	super->chunk_root_generation = root_item->generation;
1984	super->chunk_root_level = root_item->level;
1985
1986	root_item = &fs_info->tree_root->root_item;
1987	super->root = root_item->bytenr;
1988	super->generation = root_item->generation;
1989	super->root_level = root_item->level;
1990	if (btrfs_test_opt(fs_info, SPACE_CACHE))
1991	super->cache_generation = root_item->generation;
1992	else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1993	super->cache_generation = 0;
1994	if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1995	super->uuid_tree_generation = root_item->generation;
1996	}
1997
1998	int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1999	{
2000	struct btrfs_transaction *trans;
2001	int ret = 0;
2002
2003	spin_lock(lock: &info->trans_lock);
2004	trans = info->running_transaction;
2005	if (trans)
2006	ret = (trans->state >= TRANS_STATE_COMMIT_START);
2007	spin_unlock(lock: &info->trans_lock);
2008	return ret;
2009	}
2010
2011	int btrfs_transaction_blocked(struct btrfs_fs_info *info)
2012	{
2013	struct btrfs_transaction *trans;
2014	int ret = 0;
2015
2016	spin_lock(lock: &info->trans_lock);
2017	trans = info->running_transaction;
2018	if (trans)
2019	ret = is_transaction_blocked(trans);
2020	spin_unlock(lock: &info->trans_lock);
2021	return ret;
2022	}
2023
2024	void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
2025	{
2026	struct btrfs_fs_info *fs_info = trans->fs_info;
2027	struct btrfs_transaction *cur_trans;
2028
2029	/* Kick the transaction kthread. */
2030	set_bit(nr: BTRFS_FS_COMMIT_TRANS, addr: &fs_info->flags);
2031	wake_up_process(tsk: fs_info->transaction_kthread);
2032
2033	/* take transaction reference */
2034	cur_trans = trans->transaction;
2035	refcount_inc(r: &cur_trans->use_count);
2036
2037	btrfs_end_transaction(trans);
2038
2039	/*
2040	* Wait for the current transaction commit to start and block
2041	* subsequent transaction joins
2042	*/
2043	btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2044	wait_event(fs_info->transaction_blocked_wait,
2045	cur_trans->state >= TRANS_STATE_COMMIT_START ||
2046	TRANS_ABORTED(cur_trans));
2047	btrfs_put_transaction(transaction: cur_trans);
2048	}
2049
2050	static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
2051	{
2052	struct btrfs_fs_info *fs_info = trans->fs_info;
2053	struct btrfs_transaction *cur_trans = trans->transaction;
2054
2055	WARN_ON(refcount_read(&trans->use_count) > 1);
2056
2057	btrfs_abort_transaction(trans, err);
2058
2059	spin_lock(lock: &fs_info->trans_lock);
2060
2061	/*
2062	* If the transaction is removed from the list, it means this
2063	* transaction has been committed successfully, so it is impossible
2064	* to call the cleanup function.
2065	*/
2066	BUG_ON(list_empty(&cur_trans->list));
2067
2068	if (cur_trans == fs_info->running_transaction) {
2069	cur_trans->state = TRANS_STATE_COMMIT_DOING;
2070	spin_unlock(lock: &fs_info->trans_lock);
2071
2072	/*
2073	* The thread has already released the lockdep map as reader
2074	* already in btrfs_commit_transaction().
2075	*/
2076	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2077	wait_event(cur_trans->writer_wait,
2078	atomic_read(&cur_trans->num_writers) == 1);
2079
2080	spin_lock(lock: &fs_info->trans_lock);
2081	}
2082
2083	/*
2084	* Now that we know no one else is still using the transaction we can
2085	* remove the transaction from the list of transactions. This avoids
2086	* the transaction kthread from cleaning up the transaction while some
2087	* other task is still using it, which could result in a use-after-free
2088	* on things like log trees, as it forces the transaction kthread to
2089	* wait for this transaction to be cleaned up by us.
2090	*/
2091	list_del_init(entry: &cur_trans->list);
2092
2093	spin_unlock(lock: &fs_info->trans_lock);
2094
2095	btrfs_cleanup_one_transaction(trans: trans->transaction, fs_info);
2096
2097	spin_lock(lock: &fs_info->trans_lock);
2098	if (cur_trans == fs_info->running_transaction)
2099	fs_info->running_transaction = NULL;
2100	spin_unlock(lock: &fs_info->trans_lock);
2101
2102	if (trans->type & __TRANS_FREEZABLE)
2103	sb_end_intwrite(sb: fs_info->sb);
2104	btrfs_put_transaction(transaction: cur_trans);
2105	btrfs_put_transaction(transaction: cur_trans);
2106
2107	trace_btrfs_transaction_commit(fs_info);
2108
2109	if (current->journal_info == trans)
2110	current->journal_info = NULL;
2111
2112	/*
2113	* If relocation is running, we can't cancel scrub because that will
2114	* result in a deadlock. Before relocating a block group, relocation
2115	* pauses scrub, then starts and commits a transaction before unpausing
2116	* scrub. If the transaction commit is being done by the relocation
2117	* task or triggered by another task and the relocation task is waiting
2118	* for the commit, and we end up here due to an error in the commit
2119	* path, then calling btrfs_scrub_cancel() will deadlock, as we are
2120	* asking for scrub to stop while having it asked to be paused higher
2121	* above in relocation code.
2122	*/
2123	if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
2124	btrfs_scrub_cancel(info: fs_info);
2125
2126	kmem_cache_free(s: btrfs_trans_handle_cachep, objp: trans);
2127	}
2128
2129	/*
2130	* Release reserved delayed ref space of all pending block groups of the
2131	* transaction and remove them from the list
2132	*/
2133	static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2134	{
2135	struct btrfs_fs_info *fs_info = trans->fs_info;
2136	struct btrfs_block_group *block_group, *tmp;
2137
2138	list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2139	btrfs_dec_delayed_refs_rsv_bg_inserts(fs_info);
2140	list_del_init(entry: &block_group->bg_list);
2141	}
2142	}
2143
2144	static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2145	{
2146	/*
2147	* We use try_to_writeback_inodes_sb() here because if we used
2148	* btrfs_start_delalloc_roots we would deadlock with fs freeze.
2149	* Currently are holding the fs freeze lock, if we do an async flush
2150	* we'll do btrfs_join_transaction() and deadlock because we need to
2151	* wait for the fs freeze lock. Using the direct flushing we benefit
2152	* from already being in a transaction and our join_transaction doesn't
2153	* have to re-take the fs freeze lock.
2154	*
2155	* Note that try_to_writeback_inodes_sb() will only trigger writeback
2156	* if it can read lock sb->s_umount. It will always be able to lock it,
2157	* except when the filesystem is being unmounted or being frozen, but in
2158	* those cases sync_filesystem() is called, which results in calling
2159	* writeback_inodes_sb() while holding a write lock on sb->s_umount.
2160	* Note that we don't call writeback_inodes_sb() directly, because it
2161	* will emit a warning if sb->s_umount is not locked.
2162	*/
2163	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2164	try_to_writeback_inodes_sb(sb: fs_info->sb, reason: WB_REASON_SYNC);
2165	return 0;
2166	}
2167
2168	static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2169	{
2170	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2171	btrfs_wait_ordered_roots(fs_info, U64_MAX, range_start: 0, range_len: (u64)-1);
2172	}
2173
2174	/*
2175	* Add a pending snapshot associated with the given transaction handle to the
2176	* respective handle. This must be called after the transaction commit started
2177	* and while holding fs_info->trans_lock.
2178	* This serves to guarantee a caller of btrfs_commit_transaction() that it can
2179	* safely free the pending snapshot pointer in case btrfs_commit_transaction()
2180	* returns an error.
2181	*/
2182	static void add_pending_snapshot(struct btrfs_trans_handle *trans)
2183	{
2184	struct btrfs_transaction *cur_trans = trans->transaction;
2185
2186	if (!trans->pending_snapshot)
2187	return;
2188
2189	lockdep_assert_held(&trans->fs_info->trans_lock);
2190	ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_PREP);
2191
2192	list_add(new: &trans->pending_snapshot->list, head: &cur_trans->pending_snapshots);
2193	}
2194
2195	static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval)
2196	{
2197	fs_info->commit_stats.commit_count++;
2198	fs_info->commit_stats.last_commit_dur = interval;
2199	fs_info->commit_stats.max_commit_dur =
2200	max_t(u64, fs_info->commit_stats.max_commit_dur, interval);
2201	fs_info->commit_stats.total_commit_dur += interval;
2202	}
2203
2204	int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2205	{
2206	struct btrfs_fs_info *fs_info = trans->fs_info;
2207	struct btrfs_transaction *cur_trans = trans->transaction;
2208	struct btrfs_transaction *prev_trans = NULL;
2209	int ret;
2210	ktime_t start_time;
2211	ktime_t interval;
2212
2213	ASSERT(refcount_read(&trans->use_count) == 1);
2214	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2215
2216	clear_bit(nr: BTRFS_FS_NEED_TRANS_COMMIT, addr: &fs_info->flags);
2217
2218	/* Stop the commit early if ->aborted is set */
2219	if (TRANS_ABORTED(cur_trans)) {
2220	ret = cur_trans->aborted;
2221	goto lockdep_trans_commit_start_release;
2222	}
2223
2224	btrfs_trans_release_metadata(trans);
2225	trans->block_rsv = NULL;
2226
2227	/*
2228	* We only want one transaction commit doing the flushing so we do not
2229	* waste a bunch of time on lock contention on the extent root node.
2230	*/
2231	if (!test_and_set_bit(nr: BTRFS_DELAYED_REFS_FLUSHING,
2232	addr: &cur_trans->delayed_refs.flags)) {
2233	/*
2234	* Make a pass through all the delayed refs we have so far.
2235	* Any running threads may add more while we are here.
2236	*/
2237	ret = btrfs_run_delayed_refs(trans, min_bytes: 0);
2238	if (ret)
2239	goto lockdep_trans_commit_start_release;
2240	}
2241
2242	btrfs_create_pending_block_groups(trans);
2243
2244	if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2245	int run_it = 0;
2246
2247	/* this mutex is also taken before trying to set
2248	* block groups readonly. We need to make sure
2249	* that nobody has set a block group readonly
2250	* after a extents from that block group have been
2251	* allocated for cache files. btrfs_set_block_group_ro
2252	* will wait for the transaction to commit if it
2253	* finds BTRFS_TRANS_DIRTY_BG_RUN set.
2254	*
2255	* The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2256	* only one process starts all the block group IO. It wouldn't
2257	* hurt to have more than one go through, but there's no
2258	* real advantage to it either.
2259	*/
2260	mutex_lock(&fs_info->ro_block_group_mutex);
2261	if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2262	addr: &cur_trans->flags))
2263	run_it = 1;
2264	mutex_unlock(lock: &fs_info->ro_block_group_mutex);
2265
2266	if (run_it) {
2267	ret = btrfs_start_dirty_block_groups(trans);
2268	if (ret)
2269	goto lockdep_trans_commit_start_release;
2270	}
2271	}
2272
2273	spin_lock(lock: &fs_info->trans_lock);
2274	if (cur_trans->state >= TRANS_STATE_COMMIT_PREP) {
2275	enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2276
2277	add_pending_snapshot(trans);
2278
2279	spin_unlock(lock: &fs_info->trans_lock);
2280	refcount_inc(r: &cur_trans->use_count);
2281
2282	if (trans->in_fsync)
2283	want_state = TRANS_STATE_SUPER_COMMITTED;
2284
2285	btrfs_trans_state_lockdep_release(fs_info,
2286	BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2287	ret = btrfs_end_transaction(trans);
2288	wait_for_commit(commit: cur_trans, min_state: want_state);
2289
2290	if (TRANS_ABORTED(cur_trans))
2291	ret = cur_trans->aborted;
2292
2293	btrfs_put_transaction(transaction: cur_trans);
2294
2295	return ret;
2296	}
2297
2298	cur_trans->state = TRANS_STATE_COMMIT_PREP;
2299	wake_up(&fs_info->transaction_blocked_wait);
2300	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2301
2302	if (cur_trans->list.prev != &fs_info->trans_list) {
2303	enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2304
2305	if (trans->in_fsync)
2306	want_state = TRANS_STATE_SUPER_COMMITTED;
2307
2308	prev_trans = list_entry(cur_trans->list.prev,
2309	struct btrfs_transaction, list);
2310	if (prev_trans->state < want_state) {
2311	refcount_inc(r: &prev_trans->use_count);
2312	spin_unlock(lock: &fs_info->trans_lock);
2313
2314	wait_for_commit(commit: prev_trans, min_state: want_state);
2315
2316	ret = READ_ONCE(prev_trans->aborted);
2317
2318	btrfs_put_transaction(transaction: prev_trans);
2319	if (ret)
2320	goto lockdep_release;
2321	spin_lock(lock: &fs_info->trans_lock);
2322	}
2323	} else {
2324	/*
2325	* The previous transaction was aborted and was already removed
2326	* from the list of transactions at fs_info->trans_list. So we
2327	* abort to prevent writing a new superblock that reflects a
2328	* corrupt state (pointing to trees with unwritten nodes/leafs).
2329	*/
2330	if (BTRFS_FS_ERROR(fs_info)) {
2331	spin_unlock(lock: &fs_info->trans_lock);
2332	ret = -EROFS;
2333	goto lockdep_release;
2334	}
2335	}
2336
2337	cur_trans->state = TRANS_STATE_COMMIT_START;
2338	wake_up(&fs_info->transaction_blocked_wait);
2339	spin_unlock(lock: &fs_info->trans_lock);
2340
2341	/*
2342	* Get the time spent on the work done by the commit thread and not
2343	* the time spent waiting on a previous commit
2344	*/
2345	start_time = ktime_get_ns();
2346
2347	extwriter_counter_dec(trans: cur_trans, type: trans->type);
2348
2349	ret = btrfs_start_delalloc_flush(fs_info);
2350	if (ret)
2351	goto lockdep_release;
2352
2353	ret = btrfs_run_delayed_items(trans);
2354	if (ret)
2355	goto lockdep_release;
2356
2357	/*
2358	* The thread has started/joined the transaction thus it holds the
2359	* lockdep map as a reader. It has to release it before acquiring the
2360	* lockdep map as a writer.
2361	*/
2362	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2363	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters);
2364	wait_event(cur_trans->writer_wait,
2365	extwriter_counter_read(cur_trans) == 0);
2366
2367	/* some pending stuffs might be added after the previous flush. */
2368	ret = btrfs_run_delayed_items(trans);
2369	if (ret) {
2370	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2371	goto cleanup_transaction;
2372	}
2373
2374	btrfs_wait_delalloc_flush(fs_info);
2375
2376	/*
2377	* Wait for all ordered extents started by a fast fsync that joined this
2378	* transaction. Otherwise if this transaction commits before the ordered
2379	* extents complete we lose logged data after a power failure.
2380	*/
2381	btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered);
2382	wait_event(cur_trans->pending_wait,
2383	atomic_read(&cur_trans->pending_ordered) == 0);
2384
2385	btrfs_scrub_pause(fs_info);
2386	/*
2387	* Ok now we need to make sure to block out any other joins while we
2388	* commit the transaction. We could have started a join before setting
2389	* COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2390	*/
2391	spin_lock(lock: &fs_info->trans_lock);
2392	add_pending_snapshot(trans);
2393	cur_trans->state = TRANS_STATE_COMMIT_DOING;
2394	spin_unlock(lock: &fs_info->trans_lock);
2395
2396	/*
2397	* The thread has started/joined the transaction thus it holds the
2398	* lockdep map as a reader. It has to release it before acquiring the
2399	* lockdep map as a writer.
2400	*/
2401	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2402	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2403	wait_event(cur_trans->writer_wait,
2404	atomic_read(&cur_trans->num_writers) == 1);
2405
2406	/*
2407	* Make lockdep happy by acquiring the state locks after
2408	* btrfs_trans_num_writers is released. If we acquired the state locks
2409	* before releasing the btrfs_trans_num_writers lock then lockdep would
2410	* complain because we did not follow the reverse order unlocking rule.
2411	*/
2412	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2413	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2414	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2415
2416	/*
2417	* We've started the commit, clear the flag in case we were triggered to
2418	* do an async commit but somebody else started before the transaction
2419	* kthread could do the work.
2420	*/
2421	clear_bit(nr: BTRFS_FS_COMMIT_TRANS, addr: &fs_info->flags);
2422
2423	if (TRANS_ABORTED(cur_trans)) {
2424	ret = cur_trans->aborted;
2425	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2426	goto scrub_continue;
2427	}
2428	/*
2429	* the reloc mutex makes sure that we stop
2430	* the balancing code from coming in and moving
2431	* extents around in the middle of the commit
2432	*/
2433	mutex_lock(&fs_info->reloc_mutex);
2434
2435	/*
2436	* We needn't worry about the delayed items because we will
2437	* deal with them in create_pending_snapshot(), which is the
2438	* core function of the snapshot creation.
2439	*/
2440	ret = create_pending_snapshots(trans);
2441	if (ret)
2442	goto unlock_reloc;
2443
2444	/*
2445	* We insert the dir indexes of the snapshots and update the inode
2446	* of the snapshots' parents after the snapshot creation, so there
2447	* are some delayed items which are not dealt with. Now deal with
2448	* them.
2449	*
2450	* We needn't worry that this operation will corrupt the snapshots,
2451	* because all the tree which are snapshoted will be forced to COW
2452	* the nodes and leaves.
2453	*/
2454	ret = btrfs_run_delayed_items(trans);
2455	if (ret)
2456	goto unlock_reloc;
2457
2458	ret = btrfs_run_delayed_refs(trans, U64_MAX);
2459	if (ret)
2460	goto unlock_reloc;
2461
2462	/*
2463	* make sure none of the code above managed to slip in a
2464	* delayed item
2465	*/
2466	btrfs_assert_delayed_root_empty(fs_info);
2467
2468	WARN_ON(cur_trans != trans->transaction);
2469
2470	ret = commit_fs_roots(trans);
2471	if (ret)
2472	goto unlock_reloc;
2473
2474	/* commit_fs_roots gets rid of all the tree log roots, it is now
2475	* safe to free the root of tree log roots
2476	*/
2477	btrfs_free_log_root_tree(trans, fs_info);
2478
2479	/*
2480	* Since fs roots are all committed, we can get a quite accurate
2481	* new_roots. So let's do quota accounting.
2482	*/
2483	ret = btrfs_qgroup_account_extents(trans);
2484	if (ret < 0)
2485	goto unlock_reloc;
2486
2487	ret = commit_cowonly_roots(trans);
2488	if (ret)
2489	goto unlock_reloc;
2490
2491	/*
2492	* The tasks which save the space cache and inode cache may also
2493	* update ->aborted, check it.
2494	*/
2495	if (TRANS_ABORTED(cur_trans)) {
2496	ret = cur_trans->aborted;
2497	goto unlock_reloc;
2498	}
2499
2500	cur_trans = fs_info->running_transaction;
2501
2502	btrfs_set_root_node(item: &fs_info->tree_root->root_item,
2503	node: fs_info->tree_root->node);
2504	list_add_tail(new: &fs_info->tree_root->dirty_list,
2505	head: &cur_trans->switch_commits);
2506
2507	btrfs_set_root_node(item: &fs_info->chunk_root->root_item,
2508	node: fs_info->chunk_root->node);
2509	list_add_tail(new: &fs_info->chunk_root->dirty_list,
2510	head: &cur_trans->switch_commits);
2511
2512	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2513	btrfs_set_root_node(item: &fs_info->block_group_root->root_item,
2514	node: fs_info->block_group_root->node);
2515	list_add_tail(new: &fs_info->block_group_root->dirty_list,
2516	head: &cur_trans->switch_commits);
2517	}
2518
2519	switch_commit_roots(trans);
2520
2521	ASSERT(list_empty(&cur_trans->dirty_bgs));
2522	ASSERT(list_empty(&cur_trans->io_bgs));
2523	update_super_roots(fs_info);
2524
2525	btrfs_set_super_log_root(s: fs_info->super_copy, val: 0);
2526	btrfs_set_super_log_root_level(s: fs_info->super_copy, val: 0);
2527	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2528	sizeof(*fs_info->super_copy));
2529
2530	btrfs_commit_device_sizes(trans: cur_trans);
2531
2532	clear_bit(nr: BTRFS_FS_LOG1_ERR, addr: &fs_info->flags);
2533	clear_bit(nr: BTRFS_FS_LOG2_ERR, addr: &fs_info->flags);
2534
2535	btrfs_trans_release_chunk_metadata(trans);
2536
2537	/*
2538	* Before changing the transaction state to TRANS_STATE_UNBLOCKED and
2539	* setting fs_info->running_transaction to NULL, lock tree_log_mutex to
2540	* make sure that before we commit our superblock, no other task can
2541	* start a new transaction and commit a log tree before we commit our
2542	* superblock. Anyone trying to commit a log tree locks this mutex before
2543	* writing its superblock.
2544	*/
2545	mutex_lock(&fs_info->tree_log_mutex);
2546
2547	spin_lock(lock: &fs_info->trans_lock);
2548	cur_trans->state = TRANS_STATE_UNBLOCKED;
2549	fs_info->running_transaction = NULL;
2550	spin_unlock(lock: &fs_info->trans_lock);
2551	mutex_unlock(lock: &fs_info->reloc_mutex);
2552
2553	wake_up(&fs_info->transaction_wait);
2554	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2555
2556	/* If we have features changed, wake up the cleaner to update sysfs. */
2557	if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) &&
2558	fs_info->cleaner_kthread)
2559	wake_up_process(tsk: fs_info->cleaner_kthread);
2560
2561	ret = btrfs_write_and_wait_transaction(trans);
2562	if (ret) {
2563	btrfs_handle_fs_error(fs_info, ret,
2564	"Error while writing out transaction");
2565	mutex_unlock(lock: &fs_info->tree_log_mutex);
2566	goto scrub_continue;
2567	}
2568
2569	ret = write_all_supers(fs_info, max_mirrors: 0);
2570	/*
2571	* the super is written, we can safely allow the tree-loggers
2572	* to go about their business
2573	*/
2574	mutex_unlock(lock: &fs_info->tree_log_mutex);
2575	if (ret)
2576	goto scrub_continue;
2577
2578	/*
2579	* We needn't acquire the lock here because there is no other task
2580	* which can change it.
2581	*/
2582	cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2583	wake_up(&cur_trans->commit_wait);
2584	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2585
2586	btrfs_finish_extent_commit(trans);
2587
2588	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2589	btrfs_clear_space_info_full(info: fs_info);
2590
2591	btrfs_set_last_trans_committed(fs_info, gen: cur_trans->transid);
2592	/*
2593	* We needn't acquire the lock here because there is no other task
2594	* which can change it.
2595	*/
2596	cur_trans->state = TRANS_STATE_COMPLETED;
2597	wake_up(&cur_trans->commit_wait);
2598	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2599
2600	spin_lock(lock: &fs_info->trans_lock);
2601	list_del_init(entry: &cur_trans->list);
2602	spin_unlock(lock: &fs_info->trans_lock);
2603
2604	btrfs_put_transaction(transaction: cur_trans);
2605	btrfs_put_transaction(transaction: cur_trans);
2606
2607	if (trans->type & __TRANS_FREEZABLE)
2608	sb_end_intwrite(sb: fs_info->sb);
2609
2610	trace_btrfs_transaction_commit(fs_info);
2611
2612	interval = ktime_get_ns() - start_time;
2613
2614	btrfs_scrub_continue(fs_info);
2615
2616	if (current->journal_info == trans)
2617	current->journal_info = NULL;
2618
2619	kmem_cache_free(s: btrfs_trans_handle_cachep, objp: trans);
2620
2621	update_commit_stats(fs_info, interval);
2622
2623	return ret;
2624
2625	unlock_reloc:
2626	mutex_unlock(lock: &fs_info->reloc_mutex);
2627	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2628	scrub_continue:
2629	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2630	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2631	btrfs_scrub_continue(fs_info);
2632	cleanup_transaction:
2633	btrfs_trans_release_metadata(trans);
2634	btrfs_cleanup_pending_block_groups(trans);
2635	btrfs_trans_release_chunk_metadata(trans);
2636	trans->block_rsv = NULL;
2637	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2638	if (current->journal_info == trans)
2639	current->journal_info = NULL;
2640	cleanup_transaction(trans, err: ret);
2641
2642	return ret;
2643
2644	lockdep_release:
2645	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2646	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2647	goto cleanup_transaction;
2648
2649	lockdep_trans_commit_start_release:
2650	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2651	btrfs_end_transaction(trans);
2652	return ret;
2653	}
2654
2655	/*
2656	* return < 0 if error
2657	* 0 if there are no more dead_roots at the time of call
2658	* 1 there are more to be processed, call me again
2659	*
2660	* The return value indicates there are certainly more snapshots to delete, but
2661	* if there comes a new one during processing, it may return 0. We don't mind,
2662	* because btrfs_commit_super will poke cleaner thread and it will process it a
2663	* few seconds later.
2664	*/
2665	int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
2666	{
2667	struct btrfs_root *root;
2668	int ret;
2669
2670	spin_lock(lock: &fs_info->trans_lock);
2671	if (list_empty(head: &fs_info->dead_roots)) {
2672	spin_unlock(lock: &fs_info->trans_lock);
2673	return 0;
2674	}
2675	root = list_first_entry(&fs_info->dead_roots,
2676	struct btrfs_root, root_list);
2677	list_del_init(entry: &root->root_list);
2678	spin_unlock(lock: &fs_info->trans_lock);
2679
2680	btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2681
2682	btrfs_kill_all_delayed_nodes(root);
2683
2684	if (btrfs_header_backref_rev(eb: root->node) <
2685	BTRFS_MIXED_BACKREF_REV)
2686	ret = btrfs_drop_snapshot(root, update_ref: 0, for_reloc: 0);
2687	else
2688	ret = btrfs_drop_snapshot(root, update_ref: 1, for_reloc: 0);
2689
2690	btrfs_put_root(root);
2691	return (ret < 0) ? 0 : 1;
2692	}
2693
2694	/*
2695	* We only mark the transaction aborted and then set the file system read-only.
2696	* This will prevent new transactions from starting or trying to join this
2697	* one.
2698	*
2699	* This means that error recovery at the call site is limited to freeing
2700	* any local memory allocations and passing the error code up without
2701	* further cleanup. The transaction should complete as it normally would
2702	* in the call path but will return -EIO.
2703	*
2704	* We'll complete the cleanup in btrfs_end_transaction and
2705	* btrfs_commit_transaction.
2706	*/
2707	void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
2708	const char *function,
2709	unsigned int line, int error, bool first_hit)
2710	{
2711	struct btrfs_fs_info *fs_info = trans->fs_info;
2712
2713	WRITE_ONCE(trans->aborted, error);
2714	WRITE_ONCE(trans->transaction->aborted, error);
2715	if (first_hit && error == -ENOSPC)
2716	btrfs_dump_space_info_for_trans_abort(fs_info);
2717	/* Wake up anybody who may be waiting on this transaction */
2718	wake_up(&fs_info->transaction_wait);
2719	wake_up(&fs_info->transaction_blocked_wait);
2720	__btrfs_handle_fs_error(fs_info, function, line, error, NULL);
2721	}
2722
2723	int __init btrfs_transaction_init(void)
2724	{
2725	btrfs_trans_handle_cachep = kmem_cache_create(name: "btrfs_trans_handle",
2726	size: sizeof(struct btrfs_trans_handle), align: 0,
2727	SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
2728	if (!btrfs_trans_handle_cachep)
2729	return -ENOMEM;
2730	return 0;
2731	}
2732
2733	void __cold btrfs_transaction_exit(void)
2734	{
2735	kmem_cache_destroy(s: btrfs_trans_handle_cachep);
2736	}
2737