xref: /openbmc/linux/fs/btrfs/transaction.c (revision b755c25f)
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_START]]
60  * |
61  * | Will wait for previous running transaction to completely finish if there
62  * | is one
63  * |
64  * | Then one of the following happes:
65  * | - Wait for all other trans handle holders to release.
66  * |   The btrfs_commit_transaction() caller will do the commit work.
67  * | - Wait for current transaction to be committed by others.
68  * |   Other btrfs_commit_transaction() caller will do the commit work.
69  * |
70  * | At this stage, only btrfs_join_transaction*() variants can attach
71  * | to this running transaction.
72  * | All other variants will wait for current one to finish and attach to
73  * | transaction N+1.
74  * |
75  * | To next stage:
76  * |  Caller is chosen to commit transaction N, and all other trans handle
77  * |  haven been released.
78  * V
79  * Transaction N [[TRANS_STATE_COMMIT_DOING]]
80  * |
81  * | The heavy lifting transaction work is started.
82  * | From running delayed refs (modifying extent tree) to creating pending
83  * | snapshots, running qgroups.
84  * | In short, modify supporting trees to reflect modifications of subvolume
85  * | trees.
86  * |
87  * | At this stage, all start_transaction() calls will wait for this
88  * | transaction to finish and attach to transaction N+1.
89  * |
90  * | To next stage:
91  * |  Until all supporting trees are updated.
92  * V
93  * Transaction N [[TRANS_STATE_UNBLOCKED]]
94  * |						    Transaction N+1
95  * | All needed trees are modified, thus we only    [[TRANS_STATE_RUNNING]]
96  * | need to write them back to disk and update	    |
97  * | super blocks.				    |
98  * |						    |
99  * | At this stage, new transaction is allowed to   |
100  * | start.					    |
101  * | All new start_transaction() calls will be	    |
102  * | attached to transid N+1.			    |
103  * |						    |
104  * | To next stage:				    |
105  * |  Until all tree blocks are super blocks are    |
106  * |  written to block devices			    |
107  * V						    |
108  * Transaction N [[TRANS_STATE_COMPLETED]]	    V
109  *   All tree blocks and super blocks are written.  Transaction N+1
110  *   This transaction is finished and all its	    [[TRANS_STATE_COMMIT_START]]
111  *   data structures will be cleaned up.	    | Life goes on
112  */
113 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
114 	[TRANS_STATE_RUNNING]		= 0U,
115 	[TRANS_STATE_COMMIT_START]	= (__TRANS_START | __TRANS_ATTACH),
116 	[TRANS_STATE_COMMIT_DOING]	= (__TRANS_START |
117 					   __TRANS_ATTACH |
118 					   __TRANS_JOIN |
119 					   __TRANS_JOIN_NOSTART),
120 	[TRANS_STATE_UNBLOCKED]		= (__TRANS_START |
121 					   __TRANS_ATTACH |
122 					   __TRANS_JOIN |
123 					   __TRANS_JOIN_NOLOCK |
124 					   __TRANS_JOIN_NOSTART),
125 	[TRANS_STATE_SUPER_COMMITTED]	= (__TRANS_START |
126 					   __TRANS_ATTACH |
127 					   __TRANS_JOIN |
128 					   __TRANS_JOIN_NOLOCK |
129 					   __TRANS_JOIN_NOSTART),
130 	[TRANS_STATE_COMPLETED]		= (__TRANS_START |
131 					   __TRANS_ATTACH |
132 					   __TRANS_JOIN |
133 					   __TRANS_JOIN_NOLOCK |
134 					   __TRANS_JOIN_NOSTART),
135 };
136 
137 void btrfs_put_transaction(struct btrfs_transaction *transaction)
138 {
139 	WARN_ON(refcount_read(&transaction->use_count) == 0);
140 	if (refcount_dec_and_test(&transaction->use_count)) {
141 		BUG_ON(!list_empty(&transaction->list));
142 		WARN_ON(!RB_EMPTY_ROOT(
143 				&transaction->delayed_refs.href_root.rb_root));
144 		WARN_ON(!RB_EMPTY_ROOT(
145 				&transaction->delayed_refs.dirty_extent_root));
146 		if (transaction->delayed_refs.pending_csums)
147 			btrfs_err(transaction->fs_info,
148 				  "pending csums is %llu",
149 				  transaction->delayed_refs.pending_csums);
150 		/*
151 		 * If any block groups are found in ->deleted_bgs then it's
152 		 * because the transaction was aborted and a commit did not
153 		 * happen (things failed before writing the new superblock
154 		 * and calling btrfs_finish_extent_commit()), so we can not
155 		 * discard the physical locations of the block groups.
156 		 */
157 		while (!list_empty(&transaction->deleted_bgs)) {
158 			struct btrfs_block_group *cache;
159 
160 			cache = list_first_entry(&transaction->deleted_bgs,
161 						 struct btrfs_block_group,
162 						 bg_list);
163 			list_del_init(&cache->bg_list);
164 			btrfs_unfreeze_block_group(cache);
165 			btrfs_put_block_group(cache);
166 		}
167 		WARN_ON(!list_empty(&transaction->dev_update_list));
168 		kfree(transaction);
169 	}
170 }
171 
172 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
173 {
174 	struct btrfs_transaction *cur_trans = trans->transaction;
175 	struct btrfs_fs_info *fs_info = trans->fs_info;
176 	struct btrfs_root *root, *tmp;
177 
178 	/*
179 	 * At this point no one can be using this transaction to modify any tree
180 	 * and no one can start another transaction to modify any tree either.
181 	 */
182 	ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING);
183 
184 	down_write(&fs_info->commit_root_sem);
185 
186 	if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
187 		fs_info->last_reloc_trans = trans->transid;
188 
189 	list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
190 				 dirty_list) {
191 		list_del_init(&root->dirty_list);
192 		free_extent_buffer(root->commit_root);
193 		root->commit_root = btrfs_root_node(root);
194 		extent_io_tree_release(&root->dirty_log_pages);
195 		btrfs_qgroup_clean_swapped_blocks(root);
196 	}
197 
198 	/* We can free old roots now. */
199 	spin_lock(&cur_trans->dropped_roots_lock);
200 	while (!list_empty(&cur_trans->dropped_roots)) {
201 		root = list_first_entry(&cur_trans->dropped_roots,
202 					struct btrfs_root, root_list);
203 		list_del_init(&root->root_list);
204 		spin_unlock(&cur_trans->dropped_roots_lock);
205 		btrfs_free_log(trans, root);
206 		btrfs_drop_and_free_fs_root(fs_info, root);
207 		spin_lock(&cur_trans->dropped_roots_lock);
208 	}
209 	spin_unlock(&cur_trans->dropped_roots_lock);
210 
211 	up_write(&fs_info->commit_root_sem);
212 }
213 
214 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
215 					 unsigned int type)
216 {
217 	if (type & TRANS_EXTWRITERS)
218 		atomic_inc(&trans->num_extwriters);
219 }
220 
221 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
222 					 unsigned int type)
223 {
224 	if (type & TRANS_EXTWRITERS)
225 		atomic_dec(&trans->num_extwriters);
226 }
227 
228 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
229 					  unsigned int type)
230 {
231 	atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
232 }
233 
234 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
235 {
236 	return atomic_read(&trans->num_extwriters);
237 }
238 
239 /*
240  * To be called after doing the chunk btree updates right after allocating a new
241  * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
242  * chunk after all chunk btree updates and after finishing the second phase of
243  * chunk allocation (btrfs_create_pending_block_groups()) in case some block
244  * group had its chunk item insertion delayed to the second phase.
245  */
246 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
247 {
248 	struct btrfs_fs_info *fs_info = trans->fs_info;
249 
250 	if (!trans->chunk_bytes_reserved)
251 		return;
252 
253 	btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
254 				trans->chunk_bytes_reserved, NULL);
255 	trans->chunk_bytes_reserved = 0;
256 }
257 
258 /*
259  * either allocate a new transaction or hop into the existing one
260  */
261 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
262 				     unsigned int type)
263 {
264 	struct btrfs_transaction *cur_trans;
265 
266 	spin_lock(&fs_info->trans_lock);
267 loop:
268 	/* The file system has been taken offline. No new transactions. */
269 	if (BTRFS_FS_ERROR(fs_info)) {
270 		spin_unlock(&fs_info->trans_lock);
271 		return -EROFS;
272 	}
273 
274 	cur_trans = fs_info->running_transaction;
275 	if (cur_trans) {
276 		if (TRANS_ABORTED(cur_trans)) {
277 			spin_unlock(&fs_info->trans_lock);
278 			return cur_trans->aborted;
279 		}
280 		if (btrfs_blocked_trans_types[cur_trans->state] & type) {
281 			spin_unlock(&fs_info->trans_lock);
282 			return -EBUSY;
283 		}
284 		refcount_inc(&cur_trans->use_count);
285 		atomic_inc(&cur_trans->num_writers);
286 		extwriter_counter_inc(cur_trans, type);
287 		spin_unlock(&fs_info->trans_lock);
288 		btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
289 		btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
290 		return 0;
291 	}
292 	spin_unlock(&fs_info->trans_lock);
293 
294 	/*
295 	 * If we are ATTACH, we just want to catch the current transaction,
296 	 * and commit it. If there is no transaction, just return ENOENT.
297 	 */
298 	if (type == TRANS_ATTACH)
299 		return -ENOENT;
300 
301 	/*
302 	 * JOIN_NOLOCK only happens during the transaction commit, so
303 	 * it is impossible that ->running_transaction is NULL
304 	 */
305 	BUG_ON(type == TRANS_JOIN_NOLOCK);
306 
307 	cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
308 	if (!cur_trans)
309 		return -ENOMEM;
310 
311 	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
312 	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
313 
314 	spin_lock(&fs_info->trans_lock);
315 	if (fs_info->running_transaction) {
316 		/*
317 		 * someone started a transaction after we unlocked.  Make sure
318 		 * to redo the checks above
319 		 */
320 		btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
321 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
322 		kfree(cur_trans);
323 		goto loop;
324 	} else if (BTRFS_FS_ERROR(fs_info)) {
325 		spin_unlock(&fs_info->trans_lock);
326 		btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
327 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
328 		kfree(cur_trans);
329 		return -EROFS;
330 	}
331 
332 	cur_trans->fs_info = fs_info;
333 	atomic_set(&cur_trans->pending_ordered, 0);
334 	init_waitqueue_head(&cur_trans->pending_wait);
335 	atomic_set(&cur_trans->num_writers, 1);
336 	extwriter_counter_init(cur_trans, type);
337 	init_waitqueue_head(&cur_trans->writer_wait);
338 	init_waitqueue_head(&cur_trans->commit_wait);
339 	cur_trans->state = TRANS_STATE_RUNNING;
340 	/*
341 	 * One for this trans handle, one so it will live on until we
342 	 * commit the transaction.
343 	 */
344 	refcount_set(&cur_trans->use_count, 2);
345 	cur_trans->flags = 0;
346 	cur_trans->start_time = ktime_get_seconds();
347 
348 	memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
349 
350 	cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
351 	cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
352 	atomic_set(&cur_trans->delayed_refs.num_entries, 0);
353 
354 	/*
355 	 * although the tree mod log is per file system and not per transaction,
356 	 * the log must never go across transaction boundaries.
357 	 */
358 	smp_mb();
359 	if (!list_empty(&fs_info->tree_mod_seq_list))
360 		WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
361 	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
362 		WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
363 	atomic64_set(&fs_info->tree_mod_seq, 0);
364 
365 	spin_lock_init(&cur_trans->delayed_refs.lock);
366 
367 	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
368 	INIT_LIST_HEAD(&cur_trans->dev_update_list);
369 	INIT_LIST_HEAD(&cur_trans->switch_commits);
370 	INIT_LIST_HEAD(&cur_trans->dirty_bgs);
371 	INIT_LIST_HEAD(&cur_trans->io_bgs);
372 	INIT_LIST_HEAD(&cur_trans->dropped_roots);
373 	mutex_init(&cur_trans->cache_write_mutex);
374 	spin_lock_init(&cur_trans->dirty_bgs_lock);
375 	INIT_LIST_HEAD(&cur_trans->deleted_bgs);
376 	spin_lock_init(&cur_trans->dropped_roots_lock);
377 	list_add_tail(&cur_trans->list, &fs_info->trans_list);
378 	extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
379 			IO_TREE_TRANS_DIRTY_PAGES);
380 	extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
381 			IO_TREE_FS_PINNED_EXTENTS);
382 	fs_info->generation++;
383 	cur_trans->transid = fs_info->generation;
384 	fs_info->running_transaction = cur_trans;
385 	cur_trans->aborted = 0;
386 	spin_unlock(&fs_info->trans_lock);
387 
388 	return 0;
389 }
390 
391 /*
392  * This does all the record keeping required to make sure that a shareable root
393  * is properly recorded in a given transaction.  This is required to make sure
394  * the old root from before we joined the transaction is deleted when the
395  * transaction commits.
396  */
397 static int record_root_in_trans(struct btrfs_trans_handle *trans,
398 			       struct btrfs_root *root,
399 			       int force)
400 {
401 	struct btrfs_fs_info *fs_info = root->fs_info;
402 	int ret = 0;
403 
404 	if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
405 	    root->last_trans < trans->transid) || force) {
406 		WARN_ON(!force && root->commit_root != root->node);
407 
408 		/*
409 		 * see below for IN_TRANS_SETUP usage rules
410 		 * we have the reloc mutex held now, so there
411 		 * is only one writer in this function
412 		 */
413 		set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
414 
415 		/* make sure readers find IN_TRANS_SETUP before
416 		 * they find our root->last_trans update
417 		 */
418 		smp_wmb();
419 
420 		spin_lock(&fs_info->fs_roots_radix_lock);
421 		if (root->last_trans == trans->transid && !force) {
422 			spin_unlock(&fs_info->fs_roots_radix_lock);
423 			return 0;
424 		}
425 		radix_tree_tag_set(&fs_info->fs_roots_radix,
426 				   (unsigned long)root->root_key.objectid,
427 				   BTRFS_ROOT_TRANS_TAG);
428 		spin_unlock(&fs_info->fs_roots_radix_lock);
429 		root->last_trans = trans->transid;
430 
431 		/* this is pretty tricky.  We don't want to
432 		 * take the relocation lock in btrfs_record_root_in_trans
433 		 * unless we're really doing the first setup for this root in
434 		 * this transaction.
435 		 *
436 		 * Normally we'd use root->last_trans as a flag to decide
437 		 * if we want to take the expensive mutex.
438 		 *
439 		 * But, we have to set root->last_trans before we
440 		 * init the relocation root, otherwise, we trip over warnings
441 		 * in ctree.c.  The solution used here is to flag ourselves
442 		 * with root IN_TRANS_SETUP.  When this is 1, we're still
443 		 * fixing up the reloc trees and everyone must wait.
444 		 *
445 		 * When this is zero, they can trust root->last_trans and fly
446 		 * through btrfs_record_root_in_trans without having to take the
447 		 * lock.  smp_wmb() makes sure that all the writes above are
448 		 * done before we pop in the zero below
449 		 */
450 		ret = btrfs_init_reloc_root(trans, root);
451 		smp_mb__before_atomic();
452 		clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
453 	}
454 	return ret;
455 }
456 
457 
458 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
459 			    struct btrfs_root *root)
460 {
461 	struct btrfs_fs_info *fs_info = root->fs_info;
462 	struct btrfs_transaction *cur_trans = trans->transaction;
463 
464 	/* Add ourselves to the transaction dropped list */
465 	spin_lock(&cur_trans->dropped_roots_lock);
466 	list_add_tail(&root->root_list, &cur_trans->dropped_roots);
467 	spin_unlock(&cur_trans->dropped_roots_lock);
468 
469 	/* Make sure we don't try to update the root at commit time */
470 	spin_lock(&fs_info->fs_roots_radix_lock);
471 	radix_tree_tag_clear(&fs_info->fs_roots_radix,
472 			     (unsigned long)root->root_key.objectid,
473 			     BTRFS_ROOT_TRANS_TAG);
474 	spin_unlock(&fs_info->fs_roots_radix_lock);
475 }
476 
477 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
478 			       struct btrfs_root *root)
479 {
480 	struct btrfs_fs_info *fs_info = root->fs_info;
481 	int ret;
482 
483 	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
484 		return 0;
485 
486 	/*
487 	 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
488 	 * and barriers
489 	 */
490 	smp_rmb();
491 	if (root->last_trans == trans->transid &&
492 	    !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
493 		return 0;
494 
495 	mutex_lock(&fs_info->reloc_mutex);
496 	ret = record_root_in_trans(trans, root, 0);
497 	mutex_unlock(&fs_info->reloc_mutex);
498 
499 	return ret;
500 }
501 
502 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
503 {
504 	return (trans->state >= TRANS_STATE_COMMIT_START &&
505 		trans->state < TRANS_STATE_UNBLOCKED &&
506 		!TRANS_ABORTED(trans));
507 }
508 
509 /* wait for commit against the current transaction to become unblocked
510  * when this is done, it is safe to start a new transaction, but the current
511  * transaction might not be fully on disk.
512  */
513 static void wait_current_trans(struct btrfs_fs_info *fs_info)
514 {
515 	struct btrfs_transaction *cur_trans;
516 
517 	spin_lock(&fs_info->trans_lock);
518 	cur_trans = fs_info->running_transaction;
519 	if (cur_trans && is_transaction_blocked(cur_trans)) {
520 		refcount_inc(&cur_trans->use_count);
521 		spin_unlock(&fs_info->trans_lock);
522 
523 		btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
524 		wait_event(fs_info->transaction_wait,
525 			   cur_trans->state >= TRANS_STATE_UNBLOCKED ||
526 			   TRANS_ABORTED(cur_trans));
527 		btrfs_put_transaction(cur_trans);
528 	} else {
529 		spin_unlock(&fs_info->trans_lock);
530 	}
531 }
532 
533 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
534 {
535 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
536 		return 0;
537 
538 	if (type == TRANS_START)
539 		return 1;
540 
541 	return 0;
542 }
543 
544 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
545 {
546 	struct btrfs_fs_info *fs_info = root->fs_info;
547 
548 	if (!fs_info->reloc_ctl ||
549 	    !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
550 	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
551 	    root->reloc_root)
552 		return false;
553 
554 	return true;
555 }
556 
557 static struct btrfs_trans_handle *
558 start_transaction(struct btrfs_root *root, unsigned int num_items,
559 		  unsigned int type, enum btrfs_reserve_flush_enum flush,
560 		  bool enforce_qgroups)
561 {
562 	struct btrfs_fs_info *fs_info = root->fs_info;
563 	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
564 	struct btrfs_trans_handle *h;
565 	struct btrfs_transaction *cur_trans;
566 	u64 num_bytes = 0;
567 	u64 qgroup_reserved = 0;
568 	bool reloc_reserved = false;
569 	bool do_chunk_alloc = false;
570 	int ret;
571 
572 	if (BTRFS_FS_ERROR(fs_info))
573 		return ERR_PTR(-EROFS);
574 
575 	if (current->journal_info) {
576 		WARN_ON(type & TRANS_EXTWRITERS);
577 		h = current->journal_info;
578 		refcount_inc(&h->use_count);
579 		WARN_ON(refcount_read(&h->use_count) > 2);
580 		h->orig_rsv = h->block_rsv;
581 		h->block_rsv = NULL;
582 		goto got_it;
583 	}
584 
585 	/*
586 	 * Do the reservation before we join the transaction so we can do all
587 	 * the appropriate flushing if need be.
588 	 */
589 	if (num_items && root != fs_info->chunk_root) {
590 		struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
591 		u64 delayed_refs_bytes = 0;
592 
593 		qgroup_reserved = num_items * fs_info->nodesize;
594 		ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
595 				enforce_qgroups);
596 		if (ret)
597 			return ERR_PTR(ret);
598 
599 		/*
600 		 * We want to reserve all the bytes we may need all at once, so
601 		 * we only do 1 enospc flushing cycle per transaction start.  We
602 		 * accomplish this by simply assuming we'll do num_items worth
603 		 * of delayed refs updates in this trans handle, and refill that
604 		 * amount for whatever is missing in the reserve.
605 		 */
606 		num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
607 		if (flush == BTRFS_RESERVE_FLUSH_ALL &&
608 		    !btrfs_block_rsv_full(delayed_refs_rsv)) {
609 			delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info,
610 									  num_items);
611 			num_bytes += delayed_refs_bytes;
612 		}
613 
614 		/*
615 		 * Do the reservation for the relocation root creation
616 		 */
617 		if (need_reserve_reloc_root(root)) {
618 			num_bytes += fs_info->nodesize;
619 			reloc_reserved = true;
620 		}
621 
622 		ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush);
623 		if (ret)
624 			goto reserve_fail;
625 		if (delayed_refs_bytes) {
626 			btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
627 							  delayed_refs_bytes);
628 			num_bytes -= delayed_refs_bytes;
629 		}
630 
631 		if (rsv->space_info->force_alloc)
632 			do_chunk_alloc = true;
633 	} else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
634 		   !btrfs_block_rsv_full(delayed_refs_rsv)) {
635 		/*
636 		 * Some people call with btrfs_start_transaction(root, 0)
637 		 * because they can be throttled, but have some other mechanism
638 		 * for reserving space.  We still want these guys to refill the
639 		 * delayed block_rsv so just add 1 items worth of reservation
640 		 * here.
641 		 */
642 		ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
643 		if (ret)
644 			goto reserve_fail;
645 	}
646 again:
647 	h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
648 	if (!h) {
649 		ret = -ENOMEM;
650 		goto alloc_fail;
651 	}
652 
653 	/*
654 	 * If we are JOIN_NOLOCK we're already committing a transaction and
655 	 * waiting on this guy, so we don't need to do the sb_start_intwrite
656 	 * because we're already holding a ref.  We need this because we could
657 	 * have raced in and did an fsync() on a file which can kick a commit
658 	 * and then we deadlock with somebody doing a freeze.
659 	 *
660 	 * If we are ATTACH, it means we just want to catch the current
661 	 * transaction and commit it, so we needn't do sb_start_intwrite().
662 	 */
663 	if (type & __TRANS_FREEZABLE)
664 		sb_start_intwrite(fs_info->sb);
665 
666 	if (may_wait_transaction(fs_info, type))
667 		wait_current_trans(fs_info);
668 
669 	do {
670 		ret = join_transaction(fs_info, type);
671 		if (ret == -EBUSY) {
672 			wait_current_trans(fs_info);
673 			if (unlikely(type == TRANS_ATTACH ||
674 				     type == TRANS_JOIN_NOSTART))
675 				ret = -ENOENT;
676 		}
677 	} while (ret == -EBUSY);
678 
679 	if (ret < 0)
680 		goto join_fail;
681 
682 	cur_trans = fs_info->running_transaction;
683 
684 	h->transid = cur_trans->transid;
685 	h->transaction = cur_trans;
686 	refcount_set(&h->use_count, 1);
687 	h->fs_info = root->fs_info;
688 
689 	h->type = type;
690 	INIT_LIST_HEAD(&h->new_bgs);
691 
692 	smp_mb();
693 	if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
694 	    may_wait_transaction(fs_info, type)) {
695 		current->journal_info = h;
696 		btrfs_commit_transaction(h);
697 		goto again;
698 	}
699 
700 	if (num_bytes) {
701 		trace_btrfs_space_reservation(fs_info, "transaction",
702 					      h->transid, num_bytes, 1);
703 		h->block_rsv = &fs_info->trans_block_rsv;
704 		h->bytes_reserved = num_bytes;
705 		h->reloc_reserved = reloc_reserved;
706 	}
707 
708 got_it:
709 	if (!current->journal_info)
710 		current->journal_info = h;
711 
712 	/*
713 	 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
714 	 * ALLOC_FORCE the first run through, and then we won't allocate for
715 	 * anybody else who races in later.  We don't care about the return
716 	 * value here.
717 	 */
718 	if (do_chunk_alloc && num_bytes) {
719 		u64 flags = h->block_rsv->space_info->flags;
720 
721 		btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
722 				  CHUNK_ALLOC_NO_FORCE);
723 	}
724 
725 	/*
726 	 * btrfs_record_root_in_trans() needs to alloc new extents, and may
727 	 * call btrfs_join_transaction() while we're also starting a
728 	 * transaction.
729 	 *
730 	 * Thus it need to be called after current->journal_info initialized,
731 	 * or we can deadlock.
732 	 */
733 	ret = btrfs_record_root_in_trans(h, root);
734 	if (ret) {
735 		/*
736 		 * The transaction handle is fully initialized and linked with
737 		 * other structures so it needs to be ended in case of errors,
738 		 * not just freed.
739 		 */
740 		btrfs_end_transaction(h);
741 		return ERR_PTR(ret);
742 	}
743 
744 	return h;
745 
746 join_fail:
747 	if (type & __TRANS_FREEZABLE)
748 		sb_end_intwrite(fs_info->sb);
749 	kmem_cache_free(btrfs_trans_handle_cachep, h);
750 alloc_fail:
751 	if (num_bytes)
752 		btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
753 					num_bytes, NULL);
754 reserve_fail:
755 	btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
756 	return ERR_PTR(ret);
757 }
758 
759 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
760 						   unsigned int num_items)
761 {
762 	return start_transaction(root, num_items, TRANS_START,
763 				 BTRFS_RESERVE_FLUSH_ALL, true);
764 }
765 
766 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
767 					struct btrfs_root *root,
768 					unsigned int num_items)
769 {
770 	return start_transaction(root, num_items, TRANS_START,
771 				 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
772 }
773 
774 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
775 {
776 	return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
777 				 true);
778 }
779 
780 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
781 {
782 	return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
783 				 BTRFS_RESERVE_NO_FLUSH, true);
784 }
785 
786 /*
787  * Similar to regular join but it never starts a transaction when none is
788  * running or after waiting for the current one to finish.
789  */
790 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
791 {
792 	return start_transaction(root, 0, TRANS_JOIN_NOSTART,
793 				 BTRFS_RESERVE_NO_FLUSH, true);
794 }
795 
796 /*
797  * btrfs_attach_transaction() - catch the running transaction
798  *
799  * It is used when we want to commit the current the transaction, but
800  * don't want to start a new one.
801  *
802  * Note: If this function return -ENOENT, it just means there is no
803  * running transaction. But it is possible that the inactive transaction
804  * is still in the memory, not fully on disk. If you hope there is no
805  * inactive transaction in the fs when -ENOENT is returned, you should
806  * invoke
807  *     btrfs_attach_transaction_barrier()
808  */
809 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
810 {
811 	return start_transaction(root, 0, TRANS_ATTACH,
812 				 BTRFS_RESERVE_NO_FLUSH, true);
813 }
814 
815 /*
816  * btrfs_attach_transaction_barrier() - catch the running transaction
817  *
818  * It is similar to the above function, the difference is this one
819  * will wait for all the inactive transactions until they fully
820  * complete.
821  */
822 struct btrfs_trans_handle *
823 btrfs_attach_transaction_barrier(struct btrfs_root *root)
824 {
825 	struct btrfs_trans_handle *trans;
826 
827 	trans = start_transaction(root, 0, TRANS_ATTACH,
828 				  BTRFS_RESERVE_NO_FLUSH, true);
829 	if (trans == ERR_PTR(-ENOENT)) {
830 		int ret;
831 
832 		ret = btrfs_wait_for_commit(root->fs_info, 0);
833 		if (ret)
834 			return ERR_PTR(ret);
835 	}
836 
837 	return trans;
838 }
839 
840 /* Wait for a transaction commit to reach at least the given state. */
841 static noinline void wait_for_commit(struct btrfs_transaction *commit,
842 				     const enum btrfs_trans_state min_state)
843 {
844 	struct btrfs_fs_info *fs_info = commit->fs_info;
845 	u64 transid = commit->transid;
846 	bool put = false;
847 
848 	/*
849 	 * At the moment this function is called with min_state either being
850 	 * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED.
851 	 */
852 	if (min_state == TRANS_STATE_COMPLETED)
853 		btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
854 	else
855 		btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
856 
857 	while (1) {
858 		wait_event(commit->commit_wait, commit->state >= min_state);
859 		if (put)
860 			btrfs_put_transaction(commit);
861 
862 		if (min_state < TRANS_STATE_COMPLETED)
863 			break;
864 
865 		/*
866 		 * A transaction isn't really completed until all of the
867 		 * previous transactions are completed, but with fsync we can
868 		 * end up with SUPER_COMMITTED transactions before a COMPLETED
869 		 * transaction. Wait for those.
870 		 */
871 
872 		spin_lock(&fs_info->trans_lock);
873 		commit = list_first_entry_or_null(&fs_info->trans_list,
874 						  struct btrfs_transaction,
875 						  list);
876 		if (!commit || commit->transid > transid) {
877 			spin_unlock(&fs_info->trans_lock);
878 			break;
879 		}
880 		refcount_inc(&commit->use_count);
881 		put = true;
882 		spin_unlock(&fs_info->trans_lock);
883 	}
884 }
885 
886 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
887 {
888 	struct btrfs_transaction *cur_trans = NULL, *t;
889 	int ret = 0;
890 
891 	if (transid) {
892 		if (transid <= fs_info->last_trans_committed)
893 			goto out;
894 
895 		/* find specified transaction */
896 		spin_lock(&fs_info->trans_lock);
897 		list_for_each_entry(t, &fs_info->trans_list, list) {
898 			if (t->transid == transid) {
899 				cur_trans = t;
900 				refcount_inc(&cur_trans->use_count);
901 				ret = 0;
902 				break;
903 			}
904 			if (t->transid > transid) {
905 				ret = 0;
906 				break;
907 			}
908 		}
909 		spin_unlock(&fs_info->trans_lock);
910 
911 		/*
912 		 * The specified transaction doesn't exist, or we
913 		 * raced with btrfs_commit_transaction
914 		 */
915 		if (!cur_trans) {
916 			if (transid > fs_info->last_trans_committed)
917 				ret = -EINVAL;
918 			goto out;
919 		}
920 	} else {
921 		/* find newest transaction that is committing | committed */
922 		spin_lock(&fs_info->trans_lock);
923 		list_for_each_entry_reverse(t, &fs_info->trans_list,
924 					    list) {
925 			if (t->state >= TRANS_STATE_COMMIT_START) {
926 				if (t->state == TRANS_STATE_COMPLETED)
927 					break;
928 				cur_trans = t;
929 				refcount_inc(&cur_trans->use_count);
930 				break;
931 			}
932 		}
933 		spin_unlock(&fs_info->trans_lock);
934 		if (!cur_trans)
935 			goto out;  /* nothing committing|committed */
936 	}
937 
938 	wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
939 	ret = cur_trans->aborted;
940 	btrfs_put_transaction(cur_trans);
941 out:
942 	return ret;
943 }
944 
945 void btrfs_throttle(struct btrfs_fs_info *fs_info)
946 {
947 	wait_current_trans(fs_info);
948 }
949 
950 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
951 {
952 	struct btrfs_transaction *cur_trans = trans->transaction;
953 
954 	if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
955 	    test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
956 		return true;
957 
958 	if (btrfs_check_space_for_delayed_refs(trans->fs_info))
959 		return true;
960 
961 	return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50);
962 }
963 
964 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
965 
966 {
967 	struct btrfs_fs_info *fs_info = trans->fs_info;
968 
969 	if (!trans->block_rsv) {
970 		ASSERT(!trans->bytes_reserved);
971 		return;
972 	}
973 
974 	if (!trans->bytes_reserved)
975 		return;
976 
977 	ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
978 	trace_btrfs_space_reservation(fs_info, "transaction",
979 				      trans->transid, trans->bytes_reserved, 0);
980 	btrfs_block_rsv_release(fs_info, trans->block_rsv,
981 				trans->bytes_reserved, NULL);
982 	trans->bytes_reserved = 0;
983 }
984 
985 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
986 				   int throttle)
987 {
988 	struct btrfs_fs_info *info = trans->fs_info;
989 	struct btrfs_transaction *cur_trans = trans->transaction;
990 	int err = 0;
991 
992 	if (refcount_read(&trans->use_count) > 1) {
993 		refcount_dec(&trans->use_count);
994 		trans->block_rsv = trans->orig_rsv;
995 		return 0;
996 	}
997 
998 	btrfs_trans_release_metadata(trans);
999 	trans->block_rsv = NULL;
1000 
1001 	btrfs_create_pending_block_groups(trans);
1002 
1003 	btrfs_trans_release_chunk_metadata(trans);
1004 
1005 	if (trans->type & __TRANS_FREEZABLE)
1006 		sb_end_intwrite(info->sb);
1007 
1008 	WARN_ON(cur_trans != info->running_transaction);
1009 	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
1010 	atomic_dec(&cur_trans->num_writers);
1011 	extwriter_counter_dec(cur_trans, trans->type);
1012 
1013 	cond_wake_up(&cur_trans->writer_wait);
1014 
1015 	btrfs_lockdep_release(info, btrfs_trans_num_extwriters);
1016 	btrfs_lockdep_release(info, btrfs_trans_num_writers);
1017 
1018 	btrfs_put_transaction(cur_trans);
1019 
1020 	if (current->journal_info == trans)
1021 		current->journal_info = NULL;
1022 
1023 	if (throttle)
1024 		btrfs_run_delayed_iputs(info);
1025 
1026 	if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) {
1027 		wake_up_process(info->transaction_kthread);
1028 		if (TRANS_ABORTED(trans))
1029 			err = trans->aborted;
1030 		else
1031 			err = -EROFS;
1032 	}
1033 
1034 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
1035 	return err;
1036 }
1037 
1038 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1039 {
1040 	return __btrfs_end_transaction(trans, 0);
1041 }
1042 
1043 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1044 {
1045 	return __btrfs_end_transaction(trans, 1);
1046 }
1047 
1048 /*
1049  * when btree blocks are allocated, they have some corresponding bits set for
1050  * them in one of two extent_io trees.  This is used to make sure all of
1051  * those extents are sent to disk but does not wait on them
1052  */
1053 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1054 			       struct extent_io_tree *dirty_pages, int mark)
1055 {
1056 	int err = 0;
1057 	int werr = 0;
1058 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1059 	struct extent_state *cached_state = NULL;
1060 	u64 start = 0;
1061 	u64 end;
1062 
1063 	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1064 				      mark, &cached_state)) {
1065 		bool wait_writeback = false;
1066 
1067 		err = convert_extent_bit(dirty_pages, start, end,
1068 					 EXTENT_NEED_WAIT,
1069 					 mark, &cached_state);
1070 		/*
1071 		 * convert_extent_bit can return -ENOMEM, which is most of the
1072 		 * time a temporary error. So when it happens, ignore the error
1073 		 * and wait for writeback of this range to finish - because we
1074 		 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1075 		 * to __btrfs_wait_marked_extents() would not know that
1076 		 * writeback for this range started and therefore wouldn't
1077 		 * wait for it to finish - we don't want to commit a
1078 		 * superblock that points to btree nodes/leafs for which
1079 		 * writeback hasn't finished yet (and without errors).
1080 		 * We cleanup any entries left in the io tree when committing
1081 		 * the transaction (through extent_io_tree_release()).
1082 		 */
1083 		if (err == -ENOMEM) {
1084 			err = 0;
1085 			wait_writeback = true;
1086 		}
1087 		if (!err)
1088 			err = filemap_fdatawrite_range(mapping, start, end);
1089 		if (err)
1090 			werr = err;
1091 		else if (wait_writeback)
1092 			werr = filemap_fdatawait_range(mapping, start, end);
1093 		free_extent_state(cached_state);
1094 		cached_state = NULL;
1095 		cond_resched();
1096 		start = end + 1;
1097 	}
1098 	return werr;
1099 }
1100 
1101 /*
1102  * when btree blocks are allocated, they have some corresponding bits set for
1103  * them in one of two extent_io trees.  This is used to make sure all of
1104  * those extents are on disk for transaction or log commit.  We wait
1105  * on all the pages and clear them from the dirty pages state tree
1106  */
1107 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1108 				       struct extent_io_tree *dirty_pages)
1109 {
1110 	int err = 0;
1111 	int werr = 0;
1112 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1113 	struct extent_state *cached_state = NULL;
1114 	u64 start = 0;
1115 	u64 end;
1116 
1117 	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1118 				      EXTENT_NEED_WAIT, &cached_state)) {
1119 		/*
1120 		 * Ignore -ENOMEM errors returned by clear_extent_bit().
1121 		 * When committing the transaction, we'll remove any entries
1122 		 * left in the io tree. For a log commit, we don't remove them
1123 		 * after committing the log because the tree can be accessed
1124 		 * concurrently - we do it only at transaction commit time when
1125 		 * it's safe to do it (through extent_io_tree_release()).
1126 		 */
1127 		err = clear_extent_bit(dirty_pages, start, end,
1128 				       EXTENT_NEED_WAIT, &cached_state);
1129 		if (err == -ENOMEM)
1130 			err = 0;
1131 		if (!err)
1132 			err = filemap_fdatawait_range(mapping, start, end);
1133 		if (err)
1134 			werr = err;
1135 		free_extent_state(cached_state);
1136 		cached_state = NULL;
1137 		cond_resched();
1138 		start = end + 1;
1139 	}
1140 	if (err)
1141 		werr = err;
1142 	return werr;
1143 }
1144 
1145 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1146 		       struct extent_io_tree *dirty_pages)
1147 {
1148 	bool errors = false;
1149 	int err;
1150 
1151 	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1152 	if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1153 		errors = true;
1154 
1155 	if (errors && !err)
1156 		err = -EIO;
1157 	return err;
1158 }
1159 
1160 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1161 {
1162 	struct btrfs_fs_info *fs_info = log_root->fs_info;
1163 	struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1164 	bool errors = false;
1165 	int err;
1166 
1167 	ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1168 
1169 	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1170 	if ((mark & EXTENT_DIRTY) &&
1171 	    test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1172 		errors = true;
1173 
1174 	if ((mark & EXTENT_NEW) &&
1175 	    test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1176 		errors = true;
1177 
1178 	if (errors && !err)
1179 		err = -EIO;
1180 	return err;
1181 }
1182 
1183 /*
1184  * When btree blocks are allocated the corresponding extents are marked dirty.
1185  * This function ensures such extents are persisted on disk for transaction or
1186  * log commit.
1187  *
1188  * @trans: transaction whose dirty pages we'd like to write
1189  */
1190 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1191 {
1192 	int ret;
1193 	int ret2;
1194 	struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1195 	struct btrfs_fs_info *fs_info = trans->fs_info;
1196 	struct blk_plug plug;
1197 
1198 	blk_start_plug(&plug);
1199 	ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1200 	blk_finish_plug(&plug);
1201 	ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1202 
1203 	extent_io_tree_release(&trans->transaction->dirty_pages);
1204 
1205 	if (ret)
1206 		return ret;
1207 	else if (ret2)
1208 		return ret2;
1209 	else
1210 		return 0;
1211 }
1212 
1213 /*
1214  * this is used to update the root pointer in the tree of tree roots.
1215  *
1216  * But, in the case of the extent allocation tree, updating the root
1217  * pointer may allocate blocks which may change the root of the extent
1218  * allocation tree.
1219  *
1220  * So, this loops and repeats and makes sure the cowonly root didn't
1221  * change while the root pointer was being updated in the metadata.
1222  */
1223 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1224 			       struct btrfs_root *root)
1225 {
1226 	int ret;
1227 	u64 old_root_bytenr;
1228 	u64 old_root_used;
1229 	struct btrfs_fs_info *fs_info = root->fs_info;
1230 	struct btrfs_root *tree_root = fs_info->tree_root;
1231 
1232 	old_root_used = btrfs_root_used(&root->root_item);
1233 
1234 	while (1) {
1235 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1236 		if (old_root_bytenr == root->node->start &&
1237 		    old_root_used == btrfs_root_used(&root->root_item))
1238 			break;
1239 
1240 		btrfs_set_root_node(&root->root_item, root->node);
1241 		ret = btrfs_update_root(trans, tree_root,
1242 					&root->root_key,
1243 					&root->root_item);
1244 		if (ret)
1245 			return ret;
1246 
1247 		old_root_used = btrfs_root_used(&root->root_item);
1248 	}
1249 
1250 	return 0;
1251 }
1252 
1253 /*
1254  * update all the cowonly tree roots on disk
1255  *
1256  * The error handling in this function may not be obvious. Any of the
1257  * failures will cause the file system to go offline. We still need
1258  * to clean up the delayed refs.
1259  */
1260 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1261 {
1262 	struct btrfs_fs_info *fs_info = trans->fs_info;
1263 	struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1264 	struct list_head *io_bgs = &trans->transaction->io_bgs;
1265 	struct list_head *next;
1266 	struct extent_buffer *eb;
1267 	int ret;
1268 
1269 	/*
1270 	 * At this point no one can be using this transaction to modify any tree
1271 	 * and no one can start another transaction to modify any tree either.
1272 	 */
1273 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1274 
1275 	eb = btrfs_lock_root_node(fs_info->tree_root);
1276 	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1277 			      0, &eb, BTRFS_NESTING_COW);
1278 	btrfs_tree_unlock(eb);
1279 	free_extent_buffer(eb);
1280 
1281 	if (ret)
1282 		return ret;
1283 
1284 	ret = btrfs_run_dev_stats(trans);
1285 	if (ret)
1286 		return ret;
1287 	ret = btrfs_run_dev_replace(trans);
1288 	if (ret)
1289 		return ret;
1290 	ret = btrfs_run_qgroups(trans);
1291 	if (ret)
1292 		return ret;
1293 
1294 	ret = btrfs_setup_space_cache(trans);
1295 	if (ret)
1296 		return ret;
1297 
1298 again:
1299 	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1300 		struct btrfs_root *root;
1301 		next = fs_info->dirty_cowonly_roots.next;
1302 		list_del_init(next);
1303 		root = list_entry(next, struct btrfs_root, dirty_list);
1304 		clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1305 
1306 		list_add_tail(&root->dirty_list,
1307 			      &trans->transaction->switch_commits);
1308 		ret = update_cowonly_root(trans, root);
1309 		if (ret)
1310 			return ret;
1311 	}
1312 
1313 	/* Now flush any delayed refs generated by updating all of the roots */
1314 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1315 	if (ret)
1316 		return ret;
1317 
1318 	while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1319 		ret = btrfs_write_dirty_block_groups(trans);
1320 		if (ret)
1321 			return ret;
1322 
1323 		/*
1324 		 * We're writing the dirty block groups, which could generate
1325 		 * delayed refs, which could generate more dirty block groups,
1326 		 * so we want to keep this flushing in this loop to make sure
1327 		 * everything gets run.
1328 		 */
1329 		ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1330 		if (ret)
1331 			return ret;
1332 	}
1333 
1334 	if (!list_empty(&fs_info->dirty_cowonly_roots))
1335 		goto again;
1336 
1337 	/* Update dev-replace pointer once everything is committed */
1338 	fs_info->dev_replace.committed_cursor_left =
1339 		fs_info->dev_replace.cursor_left_last_write_of_item;
1340 
1341 	return 0;
1342 }
1343 
1344 /*
1345  * If we had a pending drop we need to see if there are any others left in our
1346  * dead roots list, and if not clear our bit and wake any waiters.
1347  */
1348 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
1349 {
1350 	/*
1351 	 * We put the drop in progress roots at the front of the list, so if the
1352 	 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
1353 	 * up.
1354 	 */
1355 	spin_lock(&fs_info->trans_lock);
1356 	if (!list_empty(&fs_info->dead_roots)) {
1357 		struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
1358 							   struct btrfs_root,
1359 							   root_list);
1360 		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
1361 			spin_unlock(&fs_info->trans_lock);
1362 			return;
1363 		}
1364 	}
1365 	spin_unlock(&fs_info->trans_lock);
1366 
1367 	btrfs_wake_unfinished_drop(fs_info);
1368 }
1369 
1370 /*
1371  * dead roots are old snapshots that need to be deleted.  This allocates
1372  * a dirty root struct and adds it into the list of dead roots that need to
1373  * be deleted
1374  */
1375 void btrfs_add_dead_root(struct btrfs_root *root)
1376 {
1377 	struct btrfs_fs_info *fs_info = root->fs_info;
1378 
1379 	spin_lock(&fs_info->trans_lock);
1380 	if (list_empty(&root->root_list)) {
1381 		btrfs_grab_root(root);
1382 
1383 		/* We want to process the partially complete drops first. */
1384 		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
1385 			list_add(&root->root_list, &fs_info->dead_roots);
1386 		else
1387 			list_add_tail(&root->root_list, &fs_info->dead_roots);
1388 	}
1389 	spin_unlock(&fs_info->trans_lock);
1390 }
1391 
1392 /*
1393  * Update each subvolume root and its relocation root, if it exists, in the tree
1394  * of tree roots. Also free log roots if they exist.
1395  */
1396 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1397 {
1398 	struct btrfs_fs_info *fs_info = trans->fs_info;
1399 	struct btrfs_root *gang[8];
1400 	int i;
1401 	int ret;
1402 
1403 	/*
1404 	 * At this point no one can be using this transaction to modify any tree
1405 	 * and no one can start another transaction to modify any tree either.
1406 	 */
1407 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1408 
1409 	spin_lock(&fs_info->fs_roots_radix_lock);
1410 	while (1) {
1411 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1412 						 (void **)gang, 0,
1413 						 ARRAY_SIZE(gang),
1414 						 BTRFS_ROOT_TRANS_TAG);
1415 		if (ret == 0)
1416 			break;
1417 		for (i = 0; i < ret; i++) {
1418 			struct btrfs_root *root = gang[i];
1419 			int ret2;
1420 
1421 			/*
1422 			 * At this point we can neither have tasks logging inodes
1423 			 * from a root nor trying to commit a log tree.
1424 			 */
1425 			ASSERT(atomic_read(&root->log_writers) == 0);
1426 			ASSERT(atomic_read(&root->log_commit[0]) == 0);
1427 			ASSERT(atomic_read(&root->log_commit[1]) == 0);
1428 
1429 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
1430 					(unsigned long)root->root_key.objectid,
1431 					BTRFS_ROOT_TRANS_TAG);
1432 			spin_unlock(&fs_info->fs_roots_radix_lock);
1433 
1434 			btrfs_free_log(trans, root);
1435 			ret2 = btrfs_update_reloc_root(trans, root);
1436 			if (ret2)
1437 				return ret2;
1438 
1439 			/* see comments in should_cow_block() */
1440 			clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1441 			smp_mb__after_atomic();
1442 
1443 			if (root->commit_root != root->node) {
1444 				list_add_tail(&root->dirty_list,
1445 					&trans->transaction->switch_commits);
1446 				btrfs_set_root_node(&root->root_item,
1447 						    root->node);
1448 			}
1449 
1450 			ret2 = btrfs_update_root(trans, fs_info->tree_root,
1451 						&root->root_key,
1452 						&root->root_item);
1453 			if (ret2)
1454 				return ret2;
1455 			spin_lock(&fs_info->fs_roots_radix_lock);
1456 			btrfs_qgroup_free_meta_all_pertrans(root);
1457 		}
1458 	}
1459 	spin_unlock(&fs_info->fs_roots_radix_lock);
1460 	return 0;
1461 }
1462 
1463 /*
1464  * defrag a given btree.
1465  * Every leaf in the btree is read and defragged.
1466  */
1467 int btrfs_defrag_root(struct btrfs_root *root)
1468 {
1469 	struct btrfs_fs_info *info = root->fs_info;
1470 	struct btrfs_trans_handle *trans;
1471 	int ret;
1472 
1473 	if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1474 		return 0;
1475 
1476 	while (1) {
1477 		trans = btrfs_start_transaction(root, 0);
1478 		if (IS_ERR(trans)) {
1479 			ret = PTR_ERR(trans);
1480 			break;
1481 		}
1482 
1483 		ret = btrfs_defrag_leaves(trans, root);
1484 
1485 		btrfs_end_transaction(trans);
1486 		btrfs_btree_balance_dirty(info);
1487 		cond_resched();
1488 
1489 		if (btrfs_fs_closing(info) || ret != -EAGAIN)
1490 			break;
1491 
1492 		if (btrfs_defrag_cancelled(info)) {
1493 			btrfs_debug(info, "defrag_root cancelled");
1494 			ret = -EAGAIN;
1495 			break;
1496 		}
1497 	}
1498 	clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1499 	return ret;
1500 }
1501 
1502 /*
1503  * Do all special snapshot related qgroup dirty hack.
1504  *
1505  * Will do all needed qgroup inherit and dirty hack like switch commit
1506  * roots inside one transaction and write all btree into disk, to make
1507  * qgroup works.
1508  */
1509 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1510 				   struct btrfs_root *src,
1511 				   struct btrfs_root *parent,
1512 				   struct btrfs_qgroup_inherit *inherit,
1513 				   u64 dst_objectid)
1514 {
1515 	struct btrfs_fs_info *fs_info = src->fs_info;
1516 	int ret;
1517 
1518 	/*
1519 	 * Save some performance in the case that qgroups are not
1520 	 * enabled. If this check races with the ioctl, rescan will
1521 	 * kick in anyway.
1522 	 */
1523 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1524 		return 0;
1525 
1526 	/*
1527 	 * Ensure dirty @src will be committed.  Or, after coming
1528 	 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1529 	 * recorded root will never be updated again, causing an outdated root
1530 	 * item.
1531 	 */
1532 	ret = record_root_in_trans(trans, src, 1);
1533 	if (ret)
1534 		return ret;
1535 
1536 	/*
1537 	 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1538 	 * src root, so we must run the delayed refs here.
1539 	 *
1540 	 * However this isn't particularly fool proof, because there's no
1541 	 * synchronization keeping us from changing the tree after this point
1542 	 * before we do the qgroup_inherit, or even from making changes while
1543 	 * we're doing the qgroup_inherit.  But that's a problem for the future,
1544 	 * for now flush the delayed refs to narrow the race window where the
1545 	 * qgroup counters could end up wrong.
1546 	 */
1547 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1548 	if (ret) {
1549 		btrfs_abort_transaction(trans, ret);
1550 		return ret;
1551 	}
1552 
1553 	ret = commit_fs_roots(trans);
1554 	if (ret)
1555 		goto out;
1556 	ret = btrfs_qgroup_account_extents(trans);
1557 	if (ret < 0)
1558 		goto out;
1559 
1560 	/* Now qgroup are all updated, we can inherit it to new qgroups */
1561 	ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1562 				   inherit);
1563 	if (ret < 0)
1564 		goto out;
1565 
1566 	/*
1567 	 * Now we do a simplified commit transaction, which will:
1568 	 * 1) commit all subvolume and extent tree
1569 	 *    To ensure all subvolume and extent tree have a valid
1570 	 *    commit_root to accounting later insert_dir_item()
1571 	 * 2) write all btree blocks onto disk
1572 	 *    This is to make sure later btree modification will be cowed
1573 	 *    Or commit_root can be populated and cause wrong qgroup numbers
1574 	 * In this simplified commit, we don't really care about other trees
1575 	 * like chunk and root tree, as they won't affect qgroup.
1576 	 * And we don't write super to avoid half committed status.
1577 	 */
1578 	ret = commit_cowonly_roots(trans);
1579 	if (ret)
1580 		goto out;
1581 	switch_commit_roots(trans);
1582 	ret = btrfs_write_and_wait_transaction(trans);
1583 	if (ret)
1584 		btrfs_handle_fs_error(fs_info, ret,
1585 			"Error while writing out transaction for qgroup");
1586 
1587 out:
1588 	/*
1589 	 * Force parent root to be updated, as we recorded it before so its
1590 	 * last_trans == cur_transid.
1591 	 * Or it won't be committed again onto disk after later
1592 	 * insert_dir_item()
1593 	 */
1594 	if (!ret)
1595 		ret = record_root_in_trans(trans, parent, 1);
1596 	return ret;
1597 }
1598 
1599 /*
1600  * new snapshots need to be created at a very specific time in the
1601  * transaction commit.  This does the actual creation.
1602  *
1603  * Note:
1604  * If the error which may affect the commitment of the current transaction
1605  * happens, we should return the error number. If the error which just affect
1606  * the creation of the pending snapshots, just return 0.
1607  */
1608 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1609 				   struct btrfs_pending_snapshot *pending)
1610 {
1611 
1612 	struct btrfs_fs_info *fs_info = trans->fs_info;
1613 	struct btrfs_key key;
1614 	struct btrfs_root_item *new_root_item;
1615 	struct btrfs_root *tree_root = fs_info->tree_root;
1616 	struct btrfs_root *root = pending->root;
1617 	struct btrfs_root *parent_root;
1618 	struct btrfs_block_rsv *rsv;
1619 	struct inode *parent_inode = pending->dir;
1620 	struct btrfs_path *path;
1621 	struct btrfs_dir_item *dir_item;
1622 	struct extent_buffer *tmp;
1623 	struct extent_buffer *old;
1624 	struct timespec64 cur_time;
1625 	int ret = 0;
1626 	u64 to_reserve = 0;
1627 	u64 index = 0;
1628 	u64 objectid;
1629 	u64 root_flags;
1630 	unsigned int nofs_flags;
1631 	struct fscrypt_name fname;
1632 
1633 	ASSERT(pending->path);
1634 	path = pending->path;
1635 
1636 	ASSERT(pending->root_item);
1637 	new_root_item = pending->root_item;
1638 
1639 	/*
1640 	 * We're inside a transaction and must make sure that any potential
1641 	 * allocations with GFP_KERNEL in fscrypt won't recurse back to
1642 	 * filesystem.
1643 	 */
1644 	nofs_flags = memalloc_nofs_save();
1645 	pending->error = fscrypt_setup_filename(parent_inode,
1646 						&pending->dentry->d_name, 0,
1647 						&fname);
1648 	memalloc_nofs_restore(nofs_flags);
1649 	if (pending->error)
1650 		goto free_pending;
1651 
1652 	pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1653 	if (pending->error)
1654 		goto free_fname;
1655 
1656 	/*
1657 	 * Make qgroup to skip current new snapshot's qgroupid, as it is
1658 	 * accounted by later btrfs_qgroup_inherit().
1659 	 */
1660 	btrfs_set_skip_qgroup(trans, objectid);
1661 
1662 	btrfs_reloc_pre_snapshot(pending, &to_reserve);
1663 
1664 	if (to_reserve > 0) {
1665 		pending->error = btrfs_block_rsv_add(fs_info,
1666 						     &pending->block_rsv,
1667 						     to_reserve,
1668 						     BTRFS_RESERVE_NO_FLUSH);
1669 		if (pending->error)
1670 			goto clear_skip_qgroup;
1671 	}
1672 
1673 	key.objectid = objectid;
1674 	key.offset = (u64)-1;
1675 	key.type = BTRFS_ROOT_ITEM_KEY;
1676 
1677 	rsv = trans->block_rsv;
1678 	trans->block_rsv = &pending->block_rsv;
1679 	trans->bytes_reserved = trans->block_rsv->reserved;
1680 	trace_btrfs_space_reservation(fs_info, "transaction",
1681 				      trans->transid,
1682 				      trans->bytes_reserved, 1);
1683 	parent_root = BTRFS_I(parent_inode)->root;
1684 	ret = record_root_in_trans(trans, parent_root, 0);
1685 	if (ret)
1686 		goto fail;
1687 	cur_time = current_time(parent_inode);
1688 
1689 	/*
1690 	 * insert the directory item
1691 	 */
1692 	ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1693 	if (ret) {
1694 		btrfs_abort_transaction(trans, ret);
1695 		goto fail;
1696 	}
1697 
1698 	/* check if there is a file/dir which has the same name. */
1699 	dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1700 					 btrfs_ino(BTRFS_I(parent_inode)),
1701 					 &fname.disk_name, 0);
1702 	if (dir_item != NULL && !IS_ERR(dir_item)) {
1703 		pending->error = -EEXIST;
1704 		goto dir_item_existed;
1705 	} else if (IS_ERR(dir_item)) {
1706 		ret = PTR_ERR(dir_item);
1707 		btrfs_abort_transaction(trans, ret);
1708 		goto fail;
1709 	}
1710 	btrfs_release_path(path);
1711 
1712 	/*
1713 	 * pull in the delayed directory update
1714 	 * and the delayed inode item
1715 	 * otherwise we corrupt the FS during
1716 	 * snapshot
1717 	 */
1718 	ret = btrfs_run_delayed_items(trans);
1719 	if (ret) {	/* Transaction aborted */
1720 		btrfs_abort_transaction(trans, ret);
1721 		goto fail;
1722 	}
1723 
1724 	ret = record_root_in_trans(trans, root, 0);
1725 	if (ret) {
1726 		btrfs_abort_transaction(trans, ret);
1727 		goto fail;
1728 	}
1729 	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1730 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1731 	btrfs_check_and_init_root_item(new_root_item);
1732 
1733 	root_flags = btrfs_root_flags(new_root_item);
1734 	if (pending->readonly)
1735 		root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1736 	else
1737 		root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1738 	btrfs_set_root_flags(new_root_item, root_flags);
1739 
1740 	btrfs_set_root_generation_v2(new_root_item,
1741 			trans->transid);
1742 	generate_random_guid(new_root_item->uuid);
1743 	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1744 			BTRFS_UUID_SIZE);
1745 	if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1746 		memset(new_root_item->received_uuid, 0,
1747 		       sizeof(new_root_item->received_uuid));
1748 		memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1749 		memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1750 		btrfs_set_root_stransid(new_root_item, 0);
1751 		btrfs_set_root_rtransid(new_root_item, 0);
1752 	}
1753 	btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1754 	btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1755 	btrfs_set_root_otransid(new_root_item, trans->transid);
1756 
1757 	old = btrfs_lock_root_node(root);
1758 	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1759 			      BTRFS_NESTING_COW);
1760 	if (ret) {
1761 		btrfs_tree_unlock(old);
1762 		free_extent_buffer(old);
1763 		btrfs_abort_transaction(trans, ret);
1764 		goto fail;
1765 	}
1766 
1767 	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1768 	/* clean up in any case */
1769 	btrfs_tree_unlock(old);
1770 	free_extent_buffer(old);
1771 	if (ret) {
1772 		btrfs_abort_transaction(trans, ret);
1773 		goto fail;
1774 	}
1775 	/* see comments in should_cow_block() */
1776 	set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1777 	smp_wmb();
1778 
1779 	btrfs_set_root_node(new_root_item, tmp);
1780 	/* record when the snapshot was created in key.offset */
1781 	key.offset = trans->transid;
1782 	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1783 	btrfs_tree_unlock(tmp);
1784 	free_extent_buffer(tmp);
1785 	if (ret) {
1786 		btrfs_abort_transaction(trans, ret);
1787 		goto fail;
1788 	}
1789 
1790 	/*
1791 	 * insert root back/forward references
1792 	 */
1793 	ret = btrfs_add_root_ref(trans, objectid,
1794 				 parent_root->root_key.objectid,
1795 				 btrfs_ino(BTRFS_I(parent_inode)), index,
1796 				 &fname.disk_name);
1797 	if (ret) {
1798 		btrfs_abort_transaction(trans, ret);
1799 		goto fail;
1800 	}
1801 
1802 	key.offset = (u64)-1;
1803 	pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1804 	if (IS_ERR(pending->snap)) {
1805 		ret = PTR_ERR(pending->snap);
1806 		pending->snap = NULL;
1807 		btrfs_abort_transaction(trans, ret);
1808 		goto fail;
1809 	}
1810 
1811 	ret = btrfs_reloc_post_snapshot(trans, pending);
1812 	if (ret) {
1813 		btrfs_abort_transaction(trans, ret);
1814 		goto fail;
1815 	}
1816 
1817 	/*
1818 	 * Do special qgroup accounting for snapshot, as we do some qgroup
1819 	 * snapshot hack to do fast snapshot.
1820 	 * To co-operate with that hack, we do hack again.
1821 	 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1822 	 */
1823 	ret = qgroup_account_snapshot(trans, root, parent_root,
1824 				      pending->inherit, objectid);
1825 	if (ret < 0)
1826 		goto fail;
1827 
1828 	ret = btrfs_insert_dir_item(trans, &fname.disk_name,
1829 				    BTRFS_I(parent_inode), &key, BTRFS_FT_DIR,
1830 				    index);
1831 	/* We have check then name at the beginning, so it is impossible. */
1832 	BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1833 	if (ret) {
1834 		btrfs_abort_transaction(trans, ret);
1835 		goto fail;
1836 	}
1837 
1838 	btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1839 						  fname.disk_name.len * 2);
1840 	parent_inode->i_mtime = current_time(parent_inode);
1841 	parent_inode->i_ctime = parent_inode->i_mtime;
1842 	ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1843 	if (ret) {
1844 		btrfs_abort_transaction(trans, ret);
1845 		goto fail;
1846 	}
1847 	ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1848 				  BTRFS_UUID_KEY_SUBVOL,
1849 				  objectid);
1850 	if (ret) {
1851 		btrfs_abort_transaction(trans, ret);
1852 		goto fail;
1853 	}
1854 	if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1855 		ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1856 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1857 					  objectid);
1858 		if (ret && ret != -EEXIST) {
1859 			btrfs_abort_transaction(trans, ret);
1860 			goto fail;
1861 		}
1862 	}
1863 
1864 fail:
1865 	pending->error = ret;
1866 dir_item_existed:
1867 	trans->block_rsv = rsv;
1868 	trans->bytes_reserved = 0;
1869 clear_skip_qgroup:
1870 	btrfs_clear_skip_qgroup(trans);
1871 free_fname:
1872 	fscrypt_free_filename(&fname);
1873 free_pending:
1874 	kfree(new_root_item);
1875 	pending->root_item = NULL;
1876 	btrfs_free_path(path);
1877 	pending->path = NULL;
1878 
1879 	return ret;
1880 }
1881 
1882 /*
1883  * create all the snapshots we've scheduled for creation
1884  */
1885 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1886 {
1887 	struct btrfs_pending_snapshot *pending, *next;
1888 	struct list_head *head = &trans->transaction->pending_snapshots;
1889 	int ret = 0;
1890 
1891 	list_for_each_entry_safe(pending, next, head, list) {
1892 		list_del(&pending->list);
1893 		ret = create_pending_snapshot(trans, pending);
1894 		if (ret)
1895 			break;
1896 	}
1897 	return ret;
1898 }
1899 
1900 static void update_super_roots(struct btrfs_fs_info *fs_info)
1901 {
1902 	struct btrfs_root_item *root_item;
1903 	struct btrfs_super_block *super;
1904 
1905 	super = fs_info->super_copy;
1906 
1907 	root_item = &fs_info->chunk_root->root_item;
1908 	super->chunk_root = root_item->bytenr;
1909 	super->chunk_root_generation = root_item->generation;
1910 	super->chunk_root_level = root_item->level;
1911 
1912 	root_item = &fs_info->tree_root->root_item;
1913 	super->root = root_item->bytenr;
1914 	super->generation = root_item->generation;
1915 	super->root_level = root_item->level;
1916 	if (btrfs_test_opt(fs_info, SPACE_CACHE))
1917 		super->cache_generation = root_item->generation;
1918 	else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1919 		super->cache_generation = 0;
1920 	if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1921 		super->uuid_tree_generation = root_item->generation;
1922 }
1923 
1924 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1925 {
1926 	struct btrfs_transaction *trans;
1927 	int ret = 0;
1928 
1929 	spin_lock(&info->trans_lock);
1930 	trans = info->running_transaction;
1931 	if (trans)
1932 		ret = (trans->state >= TRANS_STATE_COMMIT_START);
1933 	spin_unlock(&info->trans_lock);
1934 	return ret;
1935 }
1936 
1937 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1938 {
1939 	struct btrfs_transaction *trans;
1940 	int ret = 0;
1941 
1942 	spin_lock(&info->trans_lock);
1943 	trans = info->running_transaction;
1944 	if (trans)
1945 		ret = is_transaction_blocked(trans);
1946 	spin_unlock(&info->trans_lock);
1947 	return ret;
1948 }
1949 
1950 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1951 {
1952 	struct btrfs_fs_info *fs_info = trans->fs_info;
1953 	struct btrfs_transaction *cur_trans;
1954 
1955 	/* Kick the transaction kthread. */
1956 	set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
1957 	wake_up_process(fs_info->transaction_kthread);
1958 
1959 	/* take transaction reference */
1960 	cur_trans = trans->transaction;
1961 	refcount_inc(&cur_trans->use_count);
1962 
1963 	btrfs_end_transaction(trans);
1964 
1965 	/*
1966 	 * Wait for the current transaction commit to start and block
1967 	 * subsequent transaction joins
1968 	 */
1969 	btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
1970 	wait_event(fs_info->transaction_blocked_wait,
1971 		   cur_trans->state >= TRANS_STATE_COMMIT_START ||
1972 		   TRANS_ABORTED(cur_trans));
1973 	btrfs_put_transaction(cur_trans);
1974 }
1975 
1976 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1977 {
1978 	struct btrfs_fs_info *fs_info = trans->fs_info;
1979 	struct btrfs_transaction *cur_trans = trans->transaction;
1980 
1981 	WARN_ON(refcount_read(&trans->use_count) > 1);
1982 
1983 	btrfs_abort_transaction(trans, err);
1984 
1985 	spin_lock(&fs_info->trans_lock);
1986 
1987 	/*
1988 	 * If the transaction is removed from the list, it means this
1989 	 * transaction has been committed successfully, so it is impossible
1990 	 * to call the cleanup function.
1991 	 */
1992 	BUG_ON(list_empty(&cur_trans->list));
1993 
1994 	if (cur_trans == fs_info->running_transaction) {
1995 		cur_trans->state = TRANS_STATE_COMMIT_DOING;
1996 		spin_unlock(&fs_info->trans_lock);
1997 
1998 		/*
1999 		 * The thread has already released the lockdep map as reader
2000 		 * already in btrfs_commit_transaction().
2001 		 */
2002 		btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2003 		wait_event(cur_trans->writer_wait,
2004 			   atomic_read(&cur_trans->num_writers) == 1);
2005 
2006 		spin_lock(&fs_info->trans_lock);
2007 	}
2008 
2009 	/*
2010 	 * Now that we know no one else is still using the transaction we can
2011 	 * remove the transaction from the list of transactions. This avoids
2012 	 * the transaction kthread from cleaning up the transaction while some
2013 	 * other task is still using it, which could result in a use-after-free
2014 	 * on things like log trees, as it forces the transaction kthread to
2015 	 * wait for this transaction to be cleaned up by us.
2016 	 */
2017 	list_del_init(&cur_trans->list);
2018 
2019 	spin_unlock(&fs_info->trans_lock);
2020 
2021 	btrfs_cleanup_one_transaction(trans->transaction, fs_info);
2022 
2023 	spin_lock(&fs_info->trans_lock);
2024 	if (cur_trans == fs_info->running_transaction)
2025 		fs_info->running_transaction = NULL;
2026 	spin_unlock(&fs_info->trans_lock);
2027 
2028 	if (trans->type & __TRANS_FREEZABLE)
2029 		sb_end_intwrite(fs_info->sb);
2030 	btrfs_put_transaction(cur_trans);
2031 	btrfs_put_transaction(cur_trans);
2032 
2033 	trace_btrfs_transaction_commit(fs_info);
2034 
2035 	if (current->journal_info == trans)
2036 		current->journal_info = NULL;
2037 
2038 	/*
2039 	 * If relocation is running, we can't cancel scrub because that will
2040 	 * result in a deadlock. Before relocating a block group, relocation
2041 	 * pauses scrub, then starts and commits a transaction before unpausing
2042 	 * scrub. If the transaction commit is being done by the relocation
2043 	 * task or triggered by another task and the relocation task is waiting
2044 	 * for the commit, and we end up here due to an error in the commit
2045 	 * path, then calling btrfs_scrub_cancel() will deadlock, as we are
2046 	 * asking for scrub to stop while having it asked to be paused higher
2047 	 * above in relocation code.
2048 	 */
2049 	if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
2050 		btrfs_scrub_cancel(fs_info);
2051 
2052 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
2053 }
2054 
2055 /*
2056  * Release reserved delayed ref space of all pending block groups of the
2057  * transaction and remove them from the list
2058  */
2059 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2060 {
2061        struct btrfs_fs_info *fs_info = trans->fs_info;
2062        struct btrfs_block_group *block_group, *tmp;
2063 
2064        list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2065                btrfs_delayed_refs_rsv_release(fs_info, 1);
2066                list_del_init(&block_group->bg_list);
2067        }
2068 }
2069 
2070 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2071 {
2072 	/*
2073 	 * We use try_to_writeback_inodes_sb() here because if we used
2074 	 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2075 	 * Currently are holding the fs freeze lock, if we do an async flush
2076 	 * we'll do btrfs_join_transaction() and deadlock because we need to
2077 	 * wait for the fs freeze lock.  Using the direct flushing we benefit
2078 	 * from already being in a transaction and our join_transaction doesn't
2079 	 * have to re-take the fs freeze lock.
2080 	 *
2081 	 * Note that try_to_writeback_inodes_sb() will only trigger writeback
2082 	 * if it can read lock sb->s_umount. It will always be able to lock it,
2083 	 * except when the filesystem is being unmounted or being frozen, but in
2084 	 * those cases sync_filesystem() is called, which results in calling
2085 	 * writeback_inodes_sb() while holding a write lock on sb->s_umount.
2086 	 * Note that we don't call writeback_inodes_sb() directly, because it
2087 	 * will emit a warning if sb->s_umount is not locked.
2088 	 */
2089 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2090 		try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2091 	return 0;
2092 }
2093 
2094 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2095 {
2096 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2097 		btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2098 }
2099 
2100 /*
2101  * Add a pending snapshot associated with the given transaction handle to the
2102  * respective handle. This must be called after the transaction commit started
2103  * and while holding fs_info->trans_lock.
2104  * This serves to guarantee a caller of btrfs_commit_transaction() that it can
2105  * safely free the pending snapshot pointer in case btrfs_commit_transaction()
2106  * returns an error.
2107  */
2108 static void add_pending_snapshot(struct btrfs_trans_handle *trans)
2109 {
2110 	struct btrfs_transaction *cur_trans = trans->transaction;
2111 
2112 	if (!trans->pending_snapshot)
2113 		return;
2114 
2115 	lockdep_assert_held(&trans->fs_info->trans_lock);
2116 	ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START);
2117 
2118 	list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots);
2119 }
2120 
2121 static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval)
2122 {
2123 	fs_info->commit_stats.commit_count++;
2124 	fs_info->commit_stats.last_commit_dur = interval;
2125 	fs_info->commit_stats.max_commit_dur =
2126 			max_t(u64, fs_info->commit_stats.max_commit_dur, interval);
2127 	fs_info->commit_stats.total_commit_dur += interval;
2128 }
2129 
2130 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2131 {
2132 	struct btrfs_fs_info *fs_info = trans->fs_info;
2133 	struct btrfs_transaction *cur_trans = trans->transaction;
2134 	struct btrfs_transaction *prev_trans = NULL;
2135 	int ret;
2136 	ktime_t start_time;
2137 	ktime_t interval;
2138 
2139 	ASSERT(refcount_read(&trans->use_count) == 1);
2140 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2141 
2142 	clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags);
2143 
2144 	/* Stop the commit early if ->aborted is set */
2145 	if (TRANS_ABORTED(cur_trans)) {
2146 		ret = cur_trans->aborted;
2147 		goto lockdep_trans_commit_start_release;
2148 	}
2149 
2150 	btrfs_trans_release_metadata(trans);
2151 	trans->block_rsv = NULL;
2152 
2153 	/*
2154 	 * We only want one transaction commit doing the flushing so we do not
2155 	 * waste a bunch of time on lock contention on the extent root node.
2156 	 */
2157 	if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2158 			      &cur_trans->delayed_refs.flags)) {
2159 		/*
2160 		 * Make a pass through all the delayed refs we have so far.
2161 		 * Any running threads may add more while we are here.
2162 		 */
2163 		ret = btrfs_run_delayed_refs(trans, 0);
2164 		if (ret)
2165 			goto lockdep_trans_commit_start_release;
2166 	}
2167 
2168 	btrfs_create_pending_block_groups(trans);
2169 
2170 	if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2171 		int run_it = 0;
2172 
2173 		/* this mutex is also taken before trying to set
2174 		 * block groups readonly.  We need to make sure
2175 		 * that nobody has set a block group readonly
2176 		 * after a extents from that block group have been
2177 		 * allocated for cache files.  btrfs_set_block_group_ro
2178 		 * will wait for the transaction to commit if it
2179 		 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2180 		 *
2181 		 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2182 		 * only one process starts all the block group IO.  It wouldn't
2183 		 * hurt to have more than one go through, but there's no
2184 		 * real advantage to it either.
2185 		 */
2186 		mutex_lock(&fs_info->ro_block_group_mutex);
2187 		if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2188 				      &cur_trans->flags))
2189 			run_it = 1;
2190 		mutex_unlock(&fs_info->ro_block_group_mutex);
2191 
2192 		if (run_it) {
2193 			ret = btrfs_start_dirty_block_groups(trans);
2194 			if (ret)
2195 				goto lockdep_trans_commit_start_release;
2196 		}
2197 	}
2198 
2199 	spin_lock(&fs_info->trans_lock);
2200 	if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2201 		enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2202 
2203 		add_pending_snapshot(trans);
2204 
2205 		spin_unlock(&fs_info->trans_lock);
2206 		refcount_inc(&cur_trans->use_count);
2207 
2208 		if (trans->in_fsync)
2209 			want_state = TRANS_STATE_SUPER_COMMITTED;
2210 
2211 		btrfs_trans_state_lockdep_release(fs_info,
2212 						  BTRFS_LOCKDEP_TRANS_COMMIT_START);
2213 		ret = btrfs_end_transaction(trans);
2214 		wait_for_commit(cur_trans, want_state);
2215 
2216 		if (TRANS_ABORTED(cur_trans))
2217 			ret = cur_trans->aborted;
2218 
2219 		btrfs_put_transaction(cur_trans);
2220 
2221 		return ret;
2222 	}
2223 
2224 	cur_trans->state = TRANS_STATE_COMMIT_START;
2225 	wake_up(&fs_info->transaction_blocked_wait);
2226 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2227 
2228 	if (cur_trans->list.prev != &fs_info->trans_list) {
2229 		enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2230 
2231 		if (trans->in_fsync)
2232 			want_state = TRANS_STATE_SUPER_COMMITTED;
2233 
2234 		prev_trans = list_entry(cur_trans->list.prev,
2235 					struct btrfs_transaction, list);
2236 		if (prev_trans->state < want_state) {
2237 			refcount_inc(&prev_trans->use_count);
2238 			spin_unlock(&fs_info->trans_lock);
2239 
2240 			wait_for_commit(prev_trans, want_state);
2241 
2242 			ret = READ_ONCE(prev_trans->aborted);
2243 
2244 			btrfs_put_transaction(prev_trans);
2245 			if (ret)
2246 				goto lockdep_release;
2247 		} else {
2248 			spin_unlock(&fs_info->trans_lock);
2249 		}
2250 	} else {
2251 		spin_unlock(&fs_info->trans_lock);
2252 		/*
2253 		 * The previous transaction was aborted and was already removed
2254 		 * from the list of transactions at fs_info->trans_list. So we
2255 		 * abort to prevent writing a new superblock that reflects a
2256 		 * corrupt state (pointing to trees with unwritten nodes/leafs).
2257 		 */
2258 		if (BTRFS_FS_ERROR(fs_info)) {
2259 			ret = -EROFS;
2260 			goto lockdep_release;
2261 		}
2262 	}
2263 
2264 	/*
2265 	 * Get the time spent on the work done by the commit thread and not
2266 	 * the time spent waiting on a previous commit
2267 	 */
2268 	start_time = ktime_get_ns();
2269 
2270 	extwriter_counter_dec(cur_trans, trans->type);
2271 
2272 	ret = btrfs_start_delalloc_flush(fs_info);
2273 	if (ret)
2274 		goto lockdep_release;
2275 
2276 	ret = btrfs_run_delayed_items(trans);
2277 	if (ret)
2278 		goto lockdep_release;
2279 
2280 	/*
2281 	 * The thread has started/joined the transaction thus it holds the
2282 	 * lockdep map as a reader. It has to release it before acquiring the
2283 	 * lockdep map as a writer.
2284 	 */
2285 	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2286 	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters);
2287 	wait_event(cur_trans->writer_wait,
2288 		   extwriter_counter_read(cur_trans) == 0);
2289 
2290 	/* some pending stuffs might be added after the previous flush. */
2291 	ret = btrfs_run_delayed_items(trans);
2292 	if (ret) {
2293 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2294 		goto cleanup_transaction;
2295 	}
2296 
2297 	btrfs_wait_delalloc_flush(fs_info);
2298 
2299 	/*
2300 	 * Wait for all ordered extents started by a fast fsync that joined this
2301 	 * transaction. Otherwise if this transaction commits before the ordered
2302 	 * extents complete we lose logged data after a power failure.
2303 	 */
2304 	btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered);
2305 	wait_event(cur_trans->pending_wait,
2306 		   atomic_read(&cur_trans->pending_ordered) == 0);
2307 
2308 	btrfs_scrub_pause(fs_info);
2309 	/*
2310 	 * Ok now we need to make sure to block out any other joins while we
2311 	 * commit the transaction.  We could have started a join before setting
2312 	 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2313 	 */
2314 	spin_lock(&fs_info->trans_lock);
2315 	add_pending_snapshot(trans);
2316 	cur_trans->state = TRANS_STATE_COMMIT_DOING;
2317 	spin_unlock(&fs_info->trans_lock);
2318 
2319 	/*
2320 	 * The thread has started/joined the transaction thus it holds the
2321 	 * lockdep map as a reader. It has to release it before acquiring the
2322 	 * lockdep map as a writer.
2323 	 */
2324 	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2325 	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2326 	wait_event(cur_trans->writer_wait,
2327 		   atomic_read(&cur_trans->num_writers) == 1);
2328 
2329 	/*
2330 	 * Make lockdep happy by acquiring the state locks after
2331 	 * btrfs_trans_num_writers is released. If we acquired the state locks
2332 	 * before releasing the btrfs_trans_num_writers lock then lockdep would
2333 	 * complain because we did not follow the reverse order unlocking rule.
2334 	 */
2335 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2336 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2337 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2338 
2339 	/*
2340 	 * We've started the commit, clear the flag in case we were triggered to
2341 	 * do an async commit but somebody else started before the transaction
2342 	 * kthread could do the work.
2343 	 */
2344 	clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
2345 
2346 	if (TRANS_ABORTED(cur_trans)) {
2347 		ret = cur_trans->aborted;
2348 		btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2349 		goto scrub_continue;
2350 	}
2351 	/*
2352 	 * the reloc mutex makes sure that we stop
2353 	 * the balancing code from coming in and moving
2354 	 * extents around in the middle of the commit
2355 	 */
2356 	mutex_lock(&fs_info->reloc_mutex);
2357 
2358 	/*
2359 	 * We needn't worry about the delayed items because we will
2360 	 * deal with them in create_pending_snapshot(), which is the
2361 	 * core function of the snapshot creation.
2362 	 */
2363 	ret = create_pending_snapshots(trans);
2364 	if (ret)
2365 		goto unlock_reloc;
2366 
2367 	/*
2368 	 * We insert the dir indexes of the snapshots and update the inode
2369 	 * of the snapshots' parents after the snapshot creation, so there
2370 	 * are some delayed items which are not dealt with. Now deal with
2371 	 * them.
2372 	 *
2373 	 * We needn't worry that this operation will corrupt the snapshots,
2374 	 * because all the tree which are snapshoted will be forced to COW
2375 	 * the nodes and leaves.
2376 	 */
2377 	ret = btrfs_run_delayed_items(trans);
2378 	if (ret)
2379 		goto unlock_reloc;
2380 
2381 	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2382 	if (ret)
2383 		goto unlock_reloc;
2384 
2385 	/*
2386 	 * make sure none of the code above managed to slip in a
2387 	 * delayed item
2388 	 */
2389 	btrfs_assert_delayed_root_empty(fs_info);
2390 
2391 	WARN_ON(cur_trans != trans->transaction);
2392 
2393 	ret = commit_fs_roots(trans);
2394 	if (ret)
2395 		goto unlock_reloc;
2396 
2397 	/* commit_fs_roots gets rid of all the tree log roots, it is now
2398 	 * safe to free the root of tree log roots
2399 	 */
2400 	btrfs_free_log_root_tree(trans, fs_info);
2401 
2402 	/*
2403 	 * Since fs roots are all committed, we can get a quite accurate
2404 	 * new_roots. So let's do quota accounting.
2405 	 */
2406 	ret = btrfs_qgroup_account_extents(trans);
2407 	if (ret < 0)
2408 		goto unlock_reloc;
2409 
2410 	ret = commit_cowonly_roots(trans);
2411 	if (ret)
2412 		goto unlock_reloc;
2413 
2414 	/*
2415 	 * The tasks which save the space cache and inode cache may also
2416 	 * update ->aborted, check it.
2417 	 */
2418 	if (TRANS_ABORTED(cur_trans)) {
2419 		ret = cur_trans->aborted;
2420 		goto unlock_reloc;
2421 	}
2422 
2423 	cur_trans = fs_info->running_transaction;
2424 
2425 	btrfs_set_root_node(&fs_info->tree_root->root_item,
2426 			    fs_info->tree_root->node);
2427 	list_add_tail(&fs_info->tree_root->dirty_list,
2428 		      &cur_trans->switch_commits);
2429 
2430 	btrfs_set_root_node(&fs_info->chunk_root->root_item,
2431 			    fs_info->chunk_root->node);
2432 	list_add_tail(&fs_info->chunk_root->dirty_list,
2433 		      &cur_trans->switch_commits);
2434 
2435 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2436 		btrfs_set_root_node(&fs_info->block_group_root->root_item,
2437 				    fs_info->block_group_root->node);
2438 		list_add_tail(&fs_info->block_group_root->dirty_list,
2439 			      &cur_trans->switch_commits);
2440 	}
2441 
2442 	switch_commit_roots(trans);
2443 
2444 	ASSERT(list_empty(&cur_trans->dirty_bgs));
2445 	ASSERT(list_empty(&cur_trans->io_bgs));
2446 	update_super_roots(fs_info);
2447 
2448 	btrfs_set_super_log_root(fs_info->super_copy, 0);
2449 	btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2450 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2451 	       sizeof(*fs_info->super_copy));
2452 
2453 	btrfs_commit_device_sizes(cur_trans);
2454 
2455 	clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2456 	clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2457 
2458 	btrfs_trans_release_chunk_metadata(trans);
2459 
2460 	/*
2461 	 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and
2462 	 * setting fs_info->running_transaction to NULL, lock tree_log_mutex to
2463 	 * make sure that before we commit our superblock, no other task can
2464 	 * start a new transaction and commit a log tree before we commit our
2465 	 * superblock. Anyone trying to commit a log tree locks this mutex before
2466 	 * writing its superblock.
2467 	 */
2468 	mutex_lock(&fs_info->tree_log_mutex);
2469 
2470 	spin_lock(&fs_info->trans_lock);
2471 	cur_trans->state = TRANS_STATE_UNBLOCKED;
2472 	fs_info->running_transaction = NULL;
2473 	spin_unlock(&fs_info->trans_lock);
2474 	mutex_unlock(&fs_info->reloc_mutex);
2475 
2476 	wake_up(&fs_info->transaction_wait);
2477 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2478 
2479 	/* If we have features changed, wake up the cleaner to update sysfs. */
2480 	if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) &&
2481 	    fs_info->cleaner_kthread)
2482 		wake_up_process(fs_info->cleaner_kthread);
2483 
2484 	ret = btrfs_write_and_wait_transaction(trans);
2485 	if (ret) {
2486 		btrfs_handle_fs_error(fs_info, ret,
2487 				      "Error while writing out transaction");
2488 		mutex_unlock(&fs_info->tree_log_mutex);
2489 		goto scrub_continue;
2490 	}
2491 
2492 	ret = write_all_supers(fs_info, 0);
2493 	/*
2494 	 * the super is written, we can safely allow the tree-loggers
2495 	 * to go about their business
2496 	 */
2497 	mutex_unlock(&fs_info->tree_log_mutex);
2498 	if (ret)
2499 		goto scrub_continue;
2500 
2501 	/*
2502 	 * We needn't acquire the lock here because there is no other task
2503 	 * which can change it.
2504 	 */
2505 	cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2506 	wake_up(&cur_trans->commit_wait);
2507 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2508 
2509 	btrfs_finish_extent_commit(trans);
2510 
2511 	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2512 		btrfs_clear_space_info_full(fs_info);
2513 
2514 	fs_info->last_trans_committed = cur_trans->transid;
2515 	/*
2516 	 * We needn't acquire the lock here because there is no other task
2517 	 * which can change it.
2518 	 */
2519 	cur_trans->state = TRANS_STATE_COMPLETED;
2520 	wake_up(&cur_trans->commit_wait);
2521 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2522 
2523 	spin_lock(&fs_info->trans_lock);
2524 	list_del_init(&cur_trans->list);
2525 	spin_unlock(&fs_info->trans_lock);
2526 
2527 	btrfs_put_transaction(cur_trans);
2528 	btrfs_put_transaction(cur_trans);
2529 
2530 	if (trans->type & __TRANS_FREEZABLE)
2531 		sb_end_intwrite(fs_info->sb);
2532 
2533 	trace_btrfs_transaction_commit(fs_info);
2534 
2535 	interval = ktime_get_ns() - start_time;
2536 
2537 	btrfs_scrub_continue(fs_info);
2538 
2539 	if (current->journal_info == trans)
2540 		current->journal_info = NULL;
2541 
2542 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
2543 
2544 	update_commit_stats(fs_info, interval);
2545 
2546 	return ret;
2547 
2548 unlock_reloc:
2549 	mutex_unlock(&fs_info->reloc_mutex);
2550 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2551 scrub_continue:
2552 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2553 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2554 	btrfs_scrub_continue(fs_info);
2555 cleanup_transaction:
2556 	btrfs_trans_release_metadata(trans);
2557 	btrfs_cleanup_pending_block_groups(trans);
2558 	btrfs_trans_release_chunk_metadata(trans);
2559 	trans->block_rsv = NULL;
2560 	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2561 	if (current->journal_info == trans)
2562 		current->journal_info = NULL;
2563 	cleanup_transaction(trans, ret);
2564 
2565 	return ret;
2566 
2567 lockdep_release:
2568 	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2569 	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2570 	goto cleanup_transaction;
2571 
2572 lockdep_trans_commit_start_release:
2573 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2574 	btrfs_end_transaction(trans);
2575 	return ret;
2576 }
2577 
2578 /*
2579  * return < 0 if error
2580  * 0 if there are no more dead_roots at the time of call
2581  * 1 there are more to be processed, call me again
2582  *
2583  * The return value indicates there are certainly more snapshots to delete, but
2584  * if there comes a new one during processing, it may return 0. We don't mind,
2585  * because btrfs_commit_super will poke cleaner thread and it will process it a
2586  * few seconds later.
2587  */
2588 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
2589 {
2590 	struct btrfs_root *root;
2591 	int ret;
2592 
2593 	spin_lock(&fs_info->trans_lock);
2594 	if (list_empty(&fs_info->dead_roots)) {
2595 		spin_unlock(&fs_info->trans_lock);
2596 		return 0;
2597 	}
2598 	root = list_first_entry(&fs_info->dead_roots,
2599 			struct btrfs_root, root_list);
2600 	list_del_init(&root->root_list);
2601 	spin_unlock(&fs_info->trans_lock);
2602 
2603 	btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2604 
2605 	btrfs_kill_all_delayed_nodes(root);
2606 
2607 	if (btrfs_header_backref_rev(root->node) <
2608 			BTRFS_MIXED_BACKREF_REV)
2609 		ret = btrfs_drop_snapshot(root, 0, 0);
2610 	else
2611 		ret = btrfs_drop_snapshot(root, 1, 0);
2612 
2613 	btrfs_put_root(root);
2614 	return (ret < 0) ? 0 : 1;
2615 }
2616 
2617 /*
2618  * We only mark the transaction aborted and then set the file system read-only.
2619  * This will prevent new transactions from starting or trying to join this
2620  * one.
2621  *
2622  * This means that error recovery at the call site is limited to freeing
2623  * any local memory allocations and passing the error code up without
2624  * further cleanup. The transaction should complete as it normally would
2625  * in the call path but will return -EIO.
2626  *
2627  * We'll complete the cleanup in btrfs_end_transaction and
2628  * btrfs_commit_transaction.
2629  */
2630 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
2631 				      const char *function,
2632 				      unsigned int line, int errno, bool first_hit)
2633 {
2634 	struct btrfs_fs_info *fs_info = trans->fs_info;
2635 
2636 	WRITE_ONCE(trans->aborted, errno);
2637 	WRITE_ONCE(trans->transaction->aborted, errno);
2638 	if (first_hit && errno == -ENOSPC)
2639 		btrfs_dump_space_info_for_trans_abort(fs_info);
2640 	/* Wake up anybody who may be waiting on this transaction */
2641 	wake_up(&fs_info->transaction_wait);
2642 	wake_up(&fs_info->transaction_blocked_wait);
2643 	__btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
2644 }
2645 
2646 int __init btrfs_transaction_init(void)
2647 {
2648 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
2649 			sizeof(struct btrfs_trans_handle), 0,
2650 			SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
2651 	if (!btrfs_trans_handle_cachep)
2652 		return -ENOMEM;
2653 	return 0;
2654 }
2655 
2656 void __cold btrfs_transaction_exit(void)
2657 {
2658 	kmem_cache_destroy(btrfs_trans_handle_cachep);
2659 }
2660