xref: /linux/fs/btrfs/transaction.c (revision f96a974170b749e3a56844e25b31d46a7233b6f6)
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 "fs.h"
27 #include "accessors.h"
28 #include "extent-tree.h"
29 #include "root-tree.h"
30 #include "dir-item.h"
31 #include "uuid-tree.h"
32 #include "ioctl.h"
33 #include "relocation.h"
34 #include "scrub.h"
35 
36 static struct kmem_cache *btrfs_trans_handle_cachep;
37 
38 /*
39  * Transaction states and transitions
40  *
41  * No running transaction (fs tree blocks are not modified)
42  * |
43  * | To next stage:
44  * |  Call start_transaction() variants. Except btrfs_join_transaction_nostart().
45  * V
46  * Transaction N [[TRANS_STATE_RUNNING]]
47  * |
48  * | New trans handles can be attached to transaction N by calling all
49  * | start_transaction() variants.
50  * |
51  * | To next stage:
52  * |  Call btrfs_commit_transaction() on any trans handle attached to
53  * |  transaction N
54  * V
55  * Transaction N [[TRANS_STATE_COMMIT_PREP]]
56  * |
57  * | If there are simultaneous calls to btrfs_commit_transaction() one will win
58  * | the race and the rest will wait for the winner to commit the transaction.
59  * |
60  * | The winner will wait for previous running transaction to completely finish
61  * | if there is one.
62  * |
63  * Transaction N [[TRANS_STATE_COMMIT_START]]
64  * |
65  * | Then one of the following happens:
66  * | - Wait for all other trans handle holders to release.
67  * |   The btrfs_commit_transaction() caller will do the commit work.
68  * | - Wait for current transaction to be committed by others.
69  * |   Other btrfs_commit_transaction() caller will do the commit work.
70  * |
71  * | At this stage, only btrfs_join_transaction*() variants can attach
72  * | to this running transaction.
73  * | All other variants will wait for current one to finish and attach to
74  * | transaction N+1.
75  * |
76  * | To next stage:
77  * |  Caller is chosen to commit transaction N, and all other trans handle
78  * |  haven been released.
79  * V
80  * Transaction N [[TRANS_STATE_COMMIT_DOING]]
81  * |
82  * | The heavy lifting transaction work is started.
83  * | From running delayed refs (modifying extent tree) to creating pending
84  * | snapshots, running qgroups.
85  * | In short, modify supporting trees to reflect modifications of subvolume
86  * | trees.
87  * |
88  * | At this stage, all start_transaction() calls will wait for this
89  * | transaction to finish and attach to transaction N+1.
90  * |
91  * | To next stage:
92  * |  Until all supporting trees are updated.
93  * V
94  * Transaction N [[TRANS_STATE_UNBLOCKED]]
95  * |						    Transaction N+1
96  * | All needed trees are modified, thus we only    [[TRANS_STATE_RUNNING]]
97  * | need to write them back to disk and update	    |
98  * | super blocks.				    |
99  * |						    |
100  * | At this stage, new transaction is allowed to   |
101  * | start.					    |
102  * | All new start_transaction() calls will be	    |
103  * | attached to transid N+1.			    |
104  * |						    |
105  * | To next stage:				    |
106  * |  Until all tree blocks are super blocks are    |
107  * |  written to block devices			    |
108  * V						    |
109  * Transaction N [[TRANS_STATE_COMPLETED]]	    V
110  *   All tree blocks and super blocks are written.  Transaction N+1
111  *   This transaction is finished and all its	    [[TRANS_STATE_COMMIT_START]]
112  *   data structures will be cleaned up.	    | Life goes on
113  */
114 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
115 	[TRANS_STATE_RUNNING]		= 0U,
116 	[TRANS_STATE_COMMIT_PREP]	= 0U,
117 	[TRANS_STATE_COMMIT_START]	= (__TRANS_START | __TRANS_ATTACH),
118 	[TRANS_STATE_COMMIT_DOING]	= (__TRANS_START |
119 					   __TRANS_ATTACH |
120 					   __TRANS_JOIN |
121 					   __TRANS_JOIN_NOSTART),
122 	[TRANS_STATE_UNBLOCKED]		= (__TRANS_START |
123 					   __TRANS_ATTACH |
124 					   __TRANS_JOIN |
125 					   __TRANS_JOIN_NOLOCK |
126 					   __TRANS_JOIN_NOSTART),
127 	[TRANS_STATE_SUPER_COMMITTED]	= (__TRANS_START |
128 					   __TRANS_ATTACH |
129 					   __TRANS_JOIN |
130 					   __TRANS_JOIN_NOLOCK |
131 					   __TRANS_JOIN_NOSTART),
132 	[TRANS_STATE_COMPLETED]		= (__TRANS_START |
133 					   __TRANS_ATTACH |
134 					   __TRANS_JOIN |
135 					   __TRANS_JOIN_NOLOCK |
136 					   __TRANS_JOIN_NOSTART),
137 };
138 
139 void btrfs_put_transaction(struct btrfs_transaction *transaction)
140 {
141 	WARN_ON(refcount_read(&transaction->use_count) == 0);
142 	if (refcount_dec_and_test(&transaction->use_count)) {
143 		BUG_ON(!list_empty(&transaction->list));
144 		WARN_ON(!xa_empty(&transaction->delayed_refs.head_refs));
145 		WARN_ON(!xa_empty(&transaction->delayed_refs.dirty_extents));
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 or TRANS_JOIN_NOSTART, we just want to catch the
296 	 * current transaction, and commit it. If there is no transaction, just
297 	 * return ENOENT.
298 	 */
299 	if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART)
300 		return -ENOENT;
301 
302 	/*
303 	 * JOIN_NOLOCK only happens during the transaction commit, so
304 	 * it is impossible that ->running_transaction is NULL
305 	 */
306 	BUG_ON(type == TRANS_JOIN_NOLOCK);
307 
308 	cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
309 	if (!cur_trans)
310 		return -ENOMEM;
311 
312 	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
313 	btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
314 
315 	spin_lock(&fs_info->trans_lock);
316 	if (fs_info->running_transaction) {
317 		/*
318 		 * someone started a transaction after we unlocked.  Make sure
319 		 * to redo the checks above
320 		 */
321 		btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
322 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
323 		kfree(cur_trans);
324 		goto loop;
325 	} else if (BTRFS_FS_ERROR(fs_info)) {
326 		spin_unlock(&fs_info->trans_lock);
327 		btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
328 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
329 		kfree(cur_trans);
330 		return -EROFS;
331 	}
332 
333 	cur_trans->fs_info = fs_info;
334 	atomic_set(&cur_trans->pending_ordered, 0);
335 	init_waitqueue_head(&cur_trans->pending_wait);
336 	atomic_set(&cur_trans->num_writers, 1);
337 	extwriter_counter_init(cur_trans, type);
338 	init_waitqueue_head(&cur_trans->writer_wait);
339 	init_waitqueue_head(&cur_trans->commit_wait);
340 	cur_trans->state = TRANS_STATE_RUNNING;
341 	/*
342 	 * One for this trans handle, one so it will live on until we
343 	 * commit the transaction.
344 	 */
345 	refcount_set(&cur_trans->use_count, 2);
346 	cur_trans->flags = 0;
347 	cur_trans->start_time = ktime_get_seconds();
348 
349 	memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
350 
351 	xa_init(&cur_trans->delayed_refs.head_refs);
352 	xa_init(&cur_trans->delayed_refs.dirty_extents);
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 	btrfs_set_fs_generation(fs_info, fs_info->generation + 1);
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 	    btrfs_get_root_last_trans(root) < 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 (btrfs_get_root_last_trans(root) == 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)btrfs_root_id(root),
427 				   BTRFS_ROOT_TRANS_TAG);
428 		spin_unlock(&fs_info->fs_roots_radix_lock);
429 		btrfs_set_root_last_trans(root, 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)btrfs_root_id(root),
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 (btrfs_get_root_last_trans(root) == 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 	    btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID ||
551 	    root->reloc_root)
552 		return false;
553 
554 	return true;
555 }
556 
557 static int btrfs_reserve_trans_metadata(struct btrfs_fs_info *fs_info,
558 					enum btrfs_reserve_flush_enum flush,
559 					u64 num_bytes,
560 					u64 *delayed_refs_bytes)
561 {
562 	struct btrfs_space_info *si = fs_info->trans_block_rsv.space_info;
563 	u64 bytes = num_bytes + *delayed_refs_bytes;
564 	int ret;
565 
566 	/*
567 	 * We want to reserve all the bytes we may need all at once, so we only
568 	 * do 1 enospc flushing cycle per transaction start.
569 	 */
570 	ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush);
571 
572 	/*
573 	 * If we are an emergency flush, which can steal from the global block
574 	 * reserve, then attempt to not reserve space for the delayed refs, as
575 	 * we will consume space for them from the global block reserve.
576 	 */
577 	if (ret && flush == BTRFS_RESERVE_FLUSH_ALL_STEAL) {
578 		bytes -= *delayed_refs_bytes;
579 		*delayed_refs_bytes = 0;
580 		ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush);
581 	}
582 
583 	return ret;
584 }
585 
586 static struct btrfs_trans_handle *
587 start_transaction(struct btrfs_root *root, unsigned int num_items,
588 		  unsigned int type, enum btrfs_reserve_flush_enum flush,
589 		  bool enforce_qgroups)
590 {
591 	struct btrfs_fs_info *fs_info = root->fs_info;
592 	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
593 	struct btrfs_block_rsv *trans_rsv = &fs_info->trans_block_rsv;
594 	struct btrfs_trans_handle *h;
595 	struct btrfs_transaction *cur_trans;
596 	u64 num_bytes = 0;
597 	u64 qgroup_reserved = 0;
598 	u64 delayed_refs_bytes = 0;
599 	bool reloc_reserved = false;
600 	bool do_chunk_alloc = false;
601 	int ret;
602 
603 	if (BTRFS_FS_ERROR(fs_info))
604 		return ERR_PTR(-EROFS);
605 
606 	if (current->journal_info) {
607 		WARN_ON(type & TRANS_EXTWRITERS);
608 		h = current->journal_info;
609 		refcount_inc(&h->use_count);
610 		WARN_ON(refcount_read(&h->use_count) > 2);
611 		h->orig_rsv = h->block_rsv;
612 		h->block_rsv = NULL;
613 		goto got_it;
614 	}
615 
616 	/*
617 	 * Do the reservation before we join the transaction so we can do all
618 	 * the appropriate flushing if need be.
619 	 */
620 	if (num_items && root != fs_info->chunk_root) {
621 		qgroup_reserved = num_items * fs_info->nodesize;
622 		/*
623 		 * Use prealloc for now, as there might be a currently running
624 		 * transaction that could free this reserved space prematurely
625 		 * by committing.
626 		 */
627 		ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserved,
628 							 enforce_qgroups, false);
629 		if (ret)
630 			return ERR_PTR(ret);
631 
632 		num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
633 		/*
634 		 * If we plan to insert/update/delete "num_items" from a btree,
635 		 * we will also generate delayed refs for extent buffers in the
636 		 * respective btree paths, so reserve space for the delayed refs
637 		 * that will be generated by the caller as it modifies btrees.
638 		 * Try to reserve them to avoid excessive use of the global
639 		 * block reserve.
640 		 */
641 		delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, num_items);
642 
643 		/*
644 		 * Do the reservation for the relocation root creation
645 		 */
646 		if (need_reserve_reloc_root(root)) {
647 			num_bytes += fs_info->nodesize;
648 			reloc_reserved = true;
649 		}
650 
651 		ret = btrfs_reserve_trans_metadata(fs_info, flush, num_bytes,
652 						   &delayed_refs_bytes);
653 		if (ret)
654 			goto reserve_fail;
655 
656 		btrfs_block_rsv_add_bytes(trans_rsv, num_bytes, true);
657 
658 		if (trans_rsv->space_info->force_alloc)
659 			do_chunk_alloc = true;
660 	} else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
661 		   !btrfs_block_rsv_full(delayed_refs_rsv)) {
662 		/*
663 		 * Some people call with btrfs_start_transaction(root, 0)
664 		 * because they can be throttled, but have some other mechanism
665 		 * for reserving space.  We still want these guys to refill the
666 		 * delayed block_rsv so just add 1 items worth of reservation
667 		 * here.
668 		 */
669 		ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
670 		if (ret)
671 			goto reserve_fail;
672 	}
673 again:
674 	h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
675 	if (!h) {
676 		ret = -ENOMEM;
677 		goto alloc_fail;
678 	}
679 
680 	/*
681 	 * If we are JOIN_NOLOCK we're already committing a transaction and
682 	 * waiting on this guy, so we don't need to do the sb_start_intwrite
683 	 * because we're already holding a ref.  We need this because we could
684 	 * have raced in and did an fsync() on a file which can kick a commit
685 	 * and then we deadlock with somebody doing a freeze.
686 	 *
687 	 * If we are ATTACH, it means we just want to catch the current
688 	 * transaction and commit it, so we needn't do sb_start_intwrite().
689 	 */
690 	if (type & __TRANS_FREEZABLE)
691 		sb_start_intwrite(fs_info->sb);
692 
693 	if (may_wait_transaction(fs_info, type))
694 		wait_current_trans(fs_info);
695 
696 	do {
697 		ret = join_transaction(fs_info, type);
698 		if (ret == -EBUSY) {
699 			wait_current_trans(fs_info);
700 			if (unlikely(type == TRANS_ATTACH ||
701 				     type == TRANS_JOIN_NOSTART))
702 				ret = -ENOENT;
703 		}
704 	} while (ret == -EBUSY);
705 
706 	if (ret < 0)
707 		goto join_fail;
708 
709 	cur_trans = fs_info->running_transaction;
710 
711 	h->transid = cur_trans->transid;
712 	h->transaction = cur_trans;
713 	refcount_set(&h->use_count, 1);
714 	h->fs_info = root->fs_info;
715 
716 	h->type = type;
717 	INIT_LIST_HEAD(&h->new_bgs);
718 	btrfs_init_metadata_block_rsv(fs_info, &h->delayed_rsv, BTRFS_BLOCK_RSV_DELOPS);
719 
720 	smp_mb();
721 	if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
722 	    may_wait_transaction(fs_info, type)) {
723 		current->journal_info = h;
724 		btrfs_commit_transaction(h);
725 		goto again;
726 	}
727 
728 	if (num_bytes) {
729 		trace_btrfs_space_reservation(fs_info, "transaction",
730 					      h->transid, num_bytes, 1);
731 		h->block_rsv = trans_rsv;
732 		h->bytes_reserved = num_bytes;
733 		if (delayed_refs_bytes > 0) {
734 			trace_btrfs_space_reservation(fs_info,
735 						      "local_delayed_refs_rsv",
736 						      h->transid,
737 						      delayed_refs_bytes, 1);
738 			h->delayed_refs_bytes_reserved = delayed_refs_bytes;
739 			btrfs_block_rsv_add_bytes(&h->delayed_rsv, delayed_refs_bytes, true);
740 			delayed_refs_bytes = 0;
741 		}
742 		h->reloc_reserved = reloc_reserved;
743 	}
744 
745 got_it:
746 	if (!current->journal_info)
747 		current->journal_info = h;
748 
749 	/*
750 	 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
751 	 * ALLOC_FORCE the first run through, and then we won't allocate for
752 	 * anybody else who races in later.  We don't care about the return
753 	 * value here.
754 	 */
755 	if (do_chunk_alloc && num_bytes) {
756 		u64 flags = h->block_rsv->space_info->flags;
757 
758 		btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
759 				  CHUNK_ALLOC_NO_FORCE);
760 	}
761 
762 	/*
763 	 * btrfs_record_root_in_trans() needs to alloc new extents, and may
764 	 * call btrfs_join_transaction() while we're also starting a
765 	 * transaction.
766 	 *
767 	 * Thus it need to be called after current->journal_info initialized,
768 	 * or we can deadlock.
769 	 */
770 	ret = btrfs_record_root_in_trans(h, root);
771 	if (ret) {
772 		/*
773 		 * The transaction handle is fully initialized and linked with
774 		 * other structures so it needs to be ended in case of errors,
775 		 * not just freed.
776 		 */
777 		btrfs_end_transaction(h);
778 		goto reserve_fail;
779 	}
780 	/*
781 	 * Now that we have found a transaction to be a part of, convert the
782 	 * qgroup reservation from prealloc to pertrans. A different transaction
783 	 * can't race in and free our pertrans out from under us.
784 	 */
785 	if (qgroup_reserved)
786 		btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
787 
788 	return h;
789 
790 join_fail:
791 	if (type & __TRANS_FREEZABLE)
792 		sb_end_intwrite(fs_info->sb);
793 	kmem_cache_free(btrfs_trans_handle_cachep, h);
794 alloc_fail:
795 	if (num_bytes)
796 		btrfs_block_rsv_release(fs_info, trans_rsv, num_bytes, NULL);
797 	if (delayed_refs_bytes)
798 		btrfs_space_info_free_bytes_may_use(trans_rsv->space_info, delayed_refs_bytes);
799 reserve_fail:
800 	btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
801 	return ERR_PTR(ret);
802 }
803 
804 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
805 						   unsigned int num_items)
806 {
807 	return start_transaction(root, num_items, TRANS_START,
808 				 BTRFS_RESERVE_FLUSH_ALL, true);
809 }
810 
811 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
812 					struct btrfs_root *root,
813 					unsigned int num_items)
814 {
815 	return start_transaction(root, num_items, TRANS_START,
816 				 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
817 }
818 
819 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
820 {
821 	return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
822 				 true);
823 }
824 
825 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
826 {
827 	return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
828 				 BTRFS_RESERVE_NO_FLUSH, true);
829 }
830 
831 /*
832  * Similar to regular join but it never starts a transaction when none is
833  * running or when there's a running one at a state >= TRANS_STATE_UNBLOCKED.
834  * This is similar to btrfs_attach_transaction() but it allows the join to
835  * happen if the transaction commit already started but it's not yet in the
836  * "doing" phase (the state is < TRANS_STATE_COMMIT_DOING).
837  */
838 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
839 {
840 	return start_transaction(root, 0, TRANS_JOIN_NOSTART,
841 				 BTRFS_RESERVE_NO_FLUSH, true);
842 }
843 
844 /*
845  * Catch the running transaction.
846  *
847  * It is used when we want to commit the current the transaction, but
848  * don't want to start a new one.
849  *
850  * Note: If this function return -ENOENT, it just means there is no
851  * running transaction. But it is possible that the inactive transaction
852  * is still in the memory, not fully on disk. If you hope there is no
853  * inactive transaction in the fs when -ENOENT is returned, you should
854  * invoke
855  *     btrfs_attach_transaction_barrier()
856  */
857 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
858 {
859 	return start_transaction(root, 0, TRANS_ATTACH,
860 				 BTRFS_RESERVE_NO_FLUSH, true);
861 }
862 
863 /*
864  * Catch the running transaction.
865  *
866  * It is similar to the above function, the difference is this one
867  * will wait for all the inactive transactions until they fully
868  * complete.
869  */
870 struct btrfs_trans_handle *
871 btrfs_attach_transaction_barrier(struct btrfs_root *root)
872 {
873 	struct btrfs_trans_handle *trans;
874 
875 	trans = start_transaction(root, 0, TRANS_ATTACH,
876 				  BTRFS_RESERVE_NO_FLUSH, true);
877 	if (trans == ERR_PTR(-ENOENT)) {
878 		int ret;
879 
880 		ret = btrfs_wait_for_commit(root->fs_info, 0);
881 		if (ret)
882 			return ERR_PTR(ret);
883 	}
884 
885 	return trans;
886 }
887 
888 /* Wait for a transaction commit to reach at least the given state. */
889 static noinline void wait_for_commit(struct btrfs_transaction *commit,
890 				     const enum btrfs_trans_state min_state)
891 {
892 	struct btrfs_fs_info *fs_info = commit->fs_info;
893 	u64 transid = commit->transid;
894 	bool put = false;
895 
896 	/*
897 	 * At the moment this function is called with min_state either being
898 	 * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED.
899 	 */
900 	if (min_state == TRANS_STATE_COMPLETED)
901 		btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
902 	else
903 		btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
904 
905 	while (1) {
906 		wait_event(commit->commit_wait, commit->state >= min_state);
907 		if (put)
908 			btrfs_put_transaction(commit);
909 
910 		if (min_state < TRANS_STATE_COMPLETED)
911 			break;
912 
913 		/*
914 		 * A transaction isn't really completed until all of the
915 		 * previous transactions are completed, but with fsync we can
916 		 * end up with SUPER_COMMITTED transactions before a COMPLETED
917 		 * transaction. Wait for those.
918 		 */
919 
920 		spin_lock(&fs_info->trans_lock);
921 		commit = list_first_entry_or_null(&fs_info->trans_list,
922 						  struct btrfs_transaction,
923 						  list);
924 		if (!commit || commit->transid > transid) {
925 			spin_unlock(&fs_info->trans_lock);
926 			break;
927 		}
928 		refcount_inc(&commit->use_count);
929 		put = true;
930 		spin_unlock(&fs_info->trans_lock);
931 	}
932 }
933 
934 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
935 {
936 	struct btrfs_transaction *cur_trans = NULL, *t;
937 	int ret = 0;
938 
939 	if (transid) {
940 		if (transid <= btrfs_get_last_trans_committed(fs_info))
941 			goto out;
942 
943 		/* find specified transaction */
944 		spin_lock(&fs_info->trans_lock);
945 		list_for_each_entry(t, &fs_info->trans_list, list) {
946 			if (t->transid == transid) {
947 				cur_trans = t;
948 				refcount_inc(&cur_trans->use_count);
949 				ret = 0;
950 				break;
951 			}
952 			if (t->transid > transid) {
953 				ret = 0;
954 				break;
955 			}
956 		}
957 		spin_unlock(&fs_info->trans_lock);
958 
959 		/*
960 		 * The specified transaction doesn't exist, or we
961 		 * raced with btrfs_commit_transaction
962 		 */
963 		if (!cur_trans) {
964 			if (transid > btrfs_get_last_trans_committed(fs_info))
965 				ret = -EINVAL;
966 			goto out;
967 		}
968 	} else {
969 		/* find newest transaction that is committing | committed */
970 		spin_lock(&fs_info->trans_lock);
971 		list_for_each_entry_reverse(t, &fs_info->trans_list,
972 					    list) {
973 			if (t->state >= TRANS_STATE_COMMIT_START) {
974 				if (t->state == TRANS_STATE_COMPLETED)
975 					break;
976 				cur_trans = t;
977 				refcount_inc(&cur_trans->use_count);
978 				break;
979 			}
980 		}
981 		spin_unlock(&fs_info->trans_lock);
982 		if (!cur_trans)
983 			goto out;  /* nothing committing|committed */
984 	}
985 
986 	wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
987 	ret = cur_trans->aborted;
988 	btrfs_put_transaction(cur_trans);
989 out:
990 	return ret;
991 }
992 
993 void btrfs_throttle(struct btrfs_fs_info *fs_info)
994 {
995 	wait_current_trans(fs_info);
996 }
997 
998 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
999 {
1000 	struct btrfs_transaction *cur_trans = trans->transaction;
1001 
1002 	if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
1003 	    test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
1004 		return true;
1005 
1006 	if (btrfs_check_space_for_delayed_refs(trans->fs_info))
1007 		return true;
1008 
1009 	return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50);
1010 }
1011 
1012 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
1013 
1014 {
1015 	struct btrfs_fs_info *fs_info = trans->fs_info;
1016 
1017 	if (!trans->block_rsv) {
1018 		ASSERT(!trans->bytes_reserved);
1019 		ASSERT(!trans->delayed_refs_bytes_reserved);
1020 		return;
1021 	}
1022 
1023 	if (!trans->bytes_reserved) {
1024 		ASSERT(!trans->delayed_refs_bytes_reserved);
1025 		return;
1026 	}
1027 
1028 	ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
1029 	trace_btrfs_space_reservation(fs_info, "transaction",
1030 				      trans->transid, trans->bytes_reserved, 0);
1031 	btrfs_block_rsv_release(fs_info, trans->block_rsv,
1032 				trans->bytes_reserved, NULL);
1033 	trans->bytes_reserved = 0;
1034 
1035 	if (!trans->delayed_refs_bytes_reserved)
1036 		return;
1037 
1038 	trace_btrfs_space_reservation(fs_info, "local_delayed_refs_rsv",
1039 				      trans->transid,
1040 				      trans->delayed_refs_bytes_reserved, 0);
1041 	btrfs_block_rsv_release(fs_info, &trans->delayed_rsv,
1042 				trans->delayed_refs_bytes_reserved, NULL);
1043 	trans->delayed_refs_bytes_reserved = 0;
1044 }
1045 
1046 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
1047 				   int throttle)
1048 {
1049 	struct btrfs_fs_info *info = trans->fs_info;
1050 	struct btrfs_transaction *cur_trans = trans->transaction;
1051 	int ret = 0;
1052 
1053 	if (refcount_read(&trans->use_count) > 1) {
1054 		refcount_dec(&trans->use_count);
1055 		trans->block_rsv = trans->orig_rsv;
1056 		return 0;
1057 	}
1058 
1059 	btrfs_trans_release_metadata(trans);
1060 	trans->block_rsv = NULL;
1061 
1062 	btrfs_create_pending_block_groups(trans);
1063 
1064 	btrfs_trans_release_chunk_metadata(trans);
1065 
1066 	if (trans->type & __TRANS_FREEZABLE)
1067 		sb_end_intwrite(info->sb);
1068 
1069 	WARN_ON(cur_trans != info->running_transaction);
1070 	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
1071 	atomic_dec(&cur_trans->num_writers);
1072 	extwriter_counter_dec(cur_trans, trans->type);
1073 
1074 	cond_wake_up(&cur_trans->writer_wait);
1075 
1076 	btrfs_lockdep_release(info, btrfs_trans_num_extwriters);
1077 	btrfs_lockdep_release(info, btrfs_trans_num_writers);
1078 
1079 	btrfs_put_transaction(cur_trans);
1080 
1081 	if (current->journal_info == trans)
1082 		current->journal_info = NULL;
1083 
1084 	if (throttle)
1085 		btrfs_run_delayed_iputs(info);
1086 
1087 	if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) {
1088 		wake_up_process(info->transaction_kthread);
1089 		if (TRANS_ABORTED(trans))
1090 			ret = trans->aborted;
1091 		else
1092 			ret = -EROFS;
1093 	}
1094 
1095 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
1096 	return ret;
1097 }
1098 
1099 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1100 {
1101 	return __btrfs_end_transaction(trans, 0);
1102 }
1103 
1104 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1105 {
1106 	return __btrfs_end_transaction(trans, 1);
1107 }
1108 
1109 /*
1110  * when btree blocks are allocated, they have some corresponding bits set for
1111  * them in one of two extent_io trees.  This is used to make sure all of
1112  * those extents are sent to disk but does not wait on them
1113  */
1114 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1115 			       struct extent_io_tree *dirty_pages, int mark)
1116 {
1117 	int ret = 0;
1118 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1119 	struct extent_state *cached_state = NULL;
1120 	u64 start = 0;
1121 	u64 end;
1122 
1123 	while (find_first_extent_bit(dirty_pages, start, &start, &end,
1124 				     mark, &cached_state)) {
1125 		bool wait_writeback = false;
1126 
1127 		ret = convert_extent_bit(dirty_pages, start, end,
1128 					 EXTENT_NEED_WAIT,
1129 					 mark, &cached_state);
1130 		/*
1131 		 * convert_extent_bit can return -ENOMEM, which is most of the
1132 		 * time a temporary error. So when it happens, ignore the error
1133 		 * and wait for writeback of this range to finish - because we
1134 		 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1135 		 * to __btrfs_wait_marked_extents() would not know that
1136 		 * writeback for this range started and therefore wouldn't
1137 		 * wait for it to finish - we don't want to commit a
1138 		 * superblock that points to btree nodes/leafs for which
1139 		 * writeback hasn't finished yet (and without errors).
1140 		 * We cleanup any entries left in the io tree when committing
1141 		 * the transaction (through extent_io_tree_release()).
1142 		 */
1143 		if (ret == -ENOMEM) {
1144 			ret = 0;
1145 			wait_writeback = true;
1146 		}
1147 		if (!ret)
1148 			ret = filemap_fdatawrite_range(mapping, start, end);
1149 		if (!ret && wait_writeback)
1150 			ret = filemap_fdatawait_range(mapping, start, end);
1151 		free_extent_state(cached_state);
1152 		if (ret)
1153 			break;
1154 		cached_state = NULL;
1155 		cond_resched();
1156 		start = end + 1;
1157 	}
1158 	return ret;
1159 }
1160 
1161 /*
1162  * when btree blocks are allocated, they have some corresponding bits set for
1163  * them in one of two extent_io trees.  This is used to make sure all of
1164  * those extents are on disk for transaction or log commit.  We wait
1165  * on all the pages and clear them from the dirty pages state tree
1166  */
1167 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1168 				       struct extent_io_tree *dirty_pages)
1169 {
1170 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
1171 	struct extent_state *cached_state = NULL;
1172 	u64 start = 0;
1173 	u64 end;
1174 	int ret = 0;
1175 
1176 	while (find_first_extent_bit(dirty_pages, start, &start, &end,
1177 				     EXTENT_NEED_WAIT, &cached_state)) {
1178 		/*
1179 		 * Ignore -ENOMEM errors returned by clear_extent_bit().
1180 		 * When committing the transaction, we'll remove any entries
1181 		 * left in the io tree. For a log commit, we don't remove them
1182 		 * after committing the log because the tree can be accessed
1183 		 * concurrently - we do it only at transaction commit time when
1184 		 * it's safe to do it (through extent_io_tree_release()).
1185 		 */
1186 		ret = clear_extent_bit(dirty_pages, start, end,
1187 				       EXTENT_NEED_WAIT, &cached_state);
1188 		if (ret == -ENOMEM)
1189 			ret = 0;
1190 		if (!ret)
1191 			ret = filemap_fdatawait_range(mapping, start, end);
1192 		free_extent_state(cached_state);
1193 		if (ret)
1194 			break;
1195 		cached_state = NULL;
1196 		cond_resched();
1197 		start = end + 1;
1198 	}
1199 	return ret;
1200 }
1201 
1202 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1203 		       struct extent_io_tree *dirty_pages)
1204 {
1205 	bool errors = false;
1206 	int err;
1207 
1208 	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1209 	if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1210 		errors = true;
1211 
1212 	if (errors && !err)
1213 		err = -EIO;
1214 	return err;
1215 }
1216 
1217 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1218 {
1219 	struct btrfs_fs_info *fs_info = log_root->fs_info;
1220 	struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1221 	bool errors = false;
1222 	int err;
1223 
1224 	ASSERT(btrfs_root_id(log_root) == BTRFS_TREE_LOG_OBJECTID);
1225 
1226 	err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1227 	if ((mark & EXTENT_DIRTY) &&
1228 	    test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1229 		errors = true;
1230 
1231 	if ((mark & EXTENT_NEW) &&
1232 	    test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1233 		errors = true;
1234 
1235 	if (errors && !err)
1236 		err = -EIO;
1237 	return err;
1238 }
1239 
1240 /*
1241  * When btree blocks are allocated the corresponding extents are marked dirty.
1242  * This function ensures such extents are persisted on disk for transaction or
1243  * log commit.
1244  *
1245  * @trans: transaction whose dirty pages we'd like to write
1246  */
1247 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1248 {
1249 	int ret;
1250 	int ret2;
1251 	struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1252 	struct btrfs_fs_info *fs_info = trans->fs_info;
1253 	struct blk_plug plug;
1254 
1255 	blk_start_plug(&plug);
1256 	ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1257 	blk_finish_plug(&plug);
1258 	ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1259 
1260 	extent_io_tree_release(&trans->transaction->dirty_pages);
1261 
1262 	if (ret)
1263 		return ret;
1264 	else if (ret2)
1265 		return ret2;
1266 	else
1267 		return 0;
1268 }
1269 
1270 /*
1271  * this is used to update the root pointer in the tree of tree roots.
1272  *
1273  * But, in the case of the extent allocation tree, updating the root
1274  * pointer may allocate blocks which may change the root of the extent
1275  * allocation tree.
1276  *
1277  * So, this loops and repeats and makes sure the cowonly root didn't
1278  * change while the root pointer was being updated in the metadata.
1279  */
1280 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1281 			       struct btrfs_root *root)
1282 {
1283 	int ret;
1284 	u64 old_root_bytenr;
1285 	u64 old_root_used;
1286 	struct btrfs_fs_info *fs_info = root->fs_info;
1287 	struct btrfs_root *tree_root = fs_info->tree_root;
1288 
1289 	old_root_used = btrfs_root_used(&root->root_item);
1290 
1291 	while (1) {
1292 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1293 		if (old_root_bytenr == root->node->start &&
1294 		    old_root_used == btrfs_root_used(&root->root_item))
1295 			break;
1296 
1297 		btrfs_set_root_node(&root->root_item, root->node);
1298 		ret = btrfs_update_root(trans, tree_root,
1299 					&root->root_key,
1300 					&root->root_item);
1301 		if (ret)
1302 			return ret;
1303 
1304 		old_root_used = btrfs_root_used(&root->root_item);
1305 	}
1306 
1307 	return 0;
1308 }
1309 
1310 /*
1311  * update all the cowonly tree roots on disk
1312  *
1313  * The error handling in this function may not be obvious. Any of the
1314  * failures will cause the file system to go offline. We still need
1315  * to clean up the delayed refs.
1316  */
1317 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1318 {
1319 	struct btrfs_fs_info *fs_info = trans->fs_info;
1320 	struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1321 	struct list_head *io_bgs = &trans->transaction->io_bgs;
1322 	struct list_head *next;
1323 	struct extent_buffer *eb;
1324 	int ret;
1325 
1326 	/*
1327 	 * At this point no one can be using this transaction to modify any tree
1328 	 * and no one can start another transaction to modify any tree either.
1329 	 */
1330 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1331 
1332 	eb = btrfs_lock_root_node(fs_info->tree_root);
1333 	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1334 			      0, &eb, BTRFS_NESTING_COW);
1335 	btrfs_tree_unlock(eb);
1336 	free_extent_buffer(eb);
1337 
1338 	if (ret)
1339 		return ret;
1340 
1341 	ret = btrfs_run_dev_stats(trans);
1342 	if (ret)
1343 		return ret;
1344 	ret = btrfs_run_dev_replace(trans);
1345 	if (ret)
1346 		return ret;
1347 	ret = btrfs_run_qgroups(trans);
1348 	if (ret)
1349 		return ret;
1350 
1351 	ret = btrfs_setup_space_cache(trans);
1352 	if (ret)
1353 		return ret;
1354 
1355 again:
1356 	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1357 		struct btrfs_root *root;
1358 		next = fs_info->dirty_cowonly_roots.next;
1359 		list_del_init(next);
1360 		root = list_entry(next, struct btrfs_root, dirty_list);
1361 		clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1362 
1363 		list_add_tail(&root->dirty_list,
1364 			      &trans->transaction->switch_commits);
1365 		ret = update_cowonly_root(trans, root);
1366 		if (ret)
1367 			return ret;
1368 	}
1369 
1370 	/* Now flush any delayed refs generated by updating all of the roots */
1371 	ret = btrfs_run_delayed_refs(trans, U64_MAX);
1372 	if (ret)
1373 		return ret;
1374 
1375 	while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1376 		ret = btrfs_write_dirty_block_groups(trans);
1377 		if (ret)
1378 			return ret;
1379 
1380 		/*
1381 		 * We're writing the dirty block groups, which could generate
1382 		 * delayed refs, which could generate more dirty block groups,
1383 		 * so we want to keep this flushing in this loop to make sure
1384 		 * everything gets run.
1385 		 */
1386 		ret = btrfs_run_delayed_refs(trans, U64_MAX);
1387 		if (ret)
1388 			return ret;
1389 	}
1390 
1391 	if (!list_empty(&fs_info->dirty_cowonly_roots))
1392 		goto again;
1393 
1394 	/* Update dev-replace pointer once everything is committed */
1395 	fs_info->dev_replace.committed_cursor_left =
1396 		fs_info->dev_replace.cursor_left_last_write_of_item;
1397 
1398 	return 0;
1399 }
1400 
1401 /*
1402  * If we had a pending drop we need to see if there are any others left in our
1403  * dead roots list, and if not clear our bit and wake any waiters.
1404  */
1405 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
1406 {
1407 	/*
1408 	 * We put the drop in progress roots at the front of the list, so if the
1409 	 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
1410 	 * up.
1411 	 */
1412 	spin_lock(&fs_info->trans_lock);
1413 	if (!list_empty(&fs_info->dead_roots)) {
1414 		struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
1415 							   struct btrfs_root,
1416 							   root_list);
1417 		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
1418 			spin_unlock(&fs_info->trans_lock);
1419 			return;
1420 		}
1421 	}
1422 	spin_unlock(&fs_info->trans_lock);
1423 
1424 	btrfs_wake_unfinished_drop(fs_info);
1425 }
1426 
1427 /*
1428  * dead roots are old snapshots that need to be deleted.  This allocates
1429  * a dirty root struct and adds it into the list of dead roots that need to
1430  * be deleted
1431  */
1432 void btrfs_add_dead_root(struct btrfs_root *root)
1433 {
1434 	struct btrfs_fs_info *fs_info = root->fs_info;
1435 
1436 	spin_lock(&fs_info->trans_lock);
1437 	if (list_empty(&root->root_list)) {
1438 		btrfs_grab_root(root);
1439 
1440 		/* We want to process the partially complete drops first. */
1441 		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
1442 			list_add(&root->root_list, &fs_info->dead_roots);
1443 		else
1444 			list_add_tail(&root->root_list, &fs_info->dead_roots);
1445 	}
1446 	spin_unlock(&fs_info->trans_lock);
1447 }
1448 
1449 /*
1450  * Update each subvolume root and its relocation root, if it exists, in the tree
1451  * of tree roots. Also free log roots if they exist.
1452  */
1453 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1454 {
1455 	struct btrfs_fs_info *fs_info = trans->fs_info;
1456 	struct btrfs_root *gang[8];
1457 	int i;
1458 	int ret;
1459 
1460 	/*
1461 	 * At this point no one can be using this transaction to modify any tree
1462 	 * and no one can start another transaction to modify any tree either.
1463 	 */
1464 	ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1465 
1466 	spin_lock(&fs_info->fs_roots_radix_lock);
1467 	while (1) {
1468 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1469 						 (void **)gang, 0,
1470 						 ARRAY_SIZE(gang),
1471 						 BTRFS_ROOT_TRANS_TAG);
1472 		if (ret == 0)
1473 			break;
1474 		for (i = 0; i < ret; i++) {
1475 			struct btrfs_root *root = gang[i];
1476 			int ret2;
1477 
1478 			/*
1479 			 * At this point we can neither have tasks logging inodes
1480 			 * from a root nor trying to commit a log tree.
1481 			 */
1482 			ASSERT(atomic_read(&root->log_writers) == 0);
1483 			ASSERT(atomic_read(&root->log_commit[0]) == 0);
1484 			ASSERT(atomic_read(&root->log_commit[1]) == 0);
1485 
1486 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
1487 					(unsigned long)btrfs_root_id(root),
1488 					BTRFS_ROOT_TRANS_TAG);
1489 			btrfs_qgroup_free_meta_all_pertrans(root);
1490 			spin_unlock(&fs_info->fs_roots_radix_lock);
1491 
1492 			btrfs_free_log(trans, root);
1493 			ret2 = btrfs_update_reloc_root(trans, root);
1494 			if (ret2)
1495 				return ret2;
1496 
1497 			/* see comments in should_cow_block() */
1498 			clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1499 			smp_mb__after_atomic();
1500 
1501 			if (root->commit_root != root->node) {
1502 				list_add_tail(&root->dirty_list,
1503 					&trans->transaction->switch_commits);
1504 				btrfs_set_root_node(&root->root_item,
1505 						    root->node);
1506 			}
1507 
1508 			ret2 = btrfs_update_root(trans, fs_info->tree_root,
1509 						&root->root_key,
1510 						&root->root_item);
1511 			if (ret2)
1512 				return ret2;
1513 			spin_lock(&fs_info->fs_roots_radix_lock);
1514 		}
1515 	}
1516 	spin_unlock(&fs_info->fs_roots_radix_lock);
1517 	return 0;
1518 }
1519 
1520 /*
1521  * Do all special snapshot related qgroup dirty hack.
1522  *
1523  * Will do all needed qgroup inherit and dirty hack like switch commit
1524  * roots inside one transaction and write all btree into disk, to make
1525  * qgroup works.
1526  */
1527 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1528 				   struct btrfs_root *src,
1529 				   struct btrfs_root *parent,
1530 				   struct btrfs_qgroup_inherit *inherit,
1531 				   u64 dst_objectid)
1532 {
1533 	struct btrfs_fs_info *fs_info = src->fs_info;
1534 	int ret;
1535 
1536 	/*
1537 	 * Save some performance in the case that qgroups are not enabled. If
1538 	 * this check races with the ioctl, rescan will kick in anyway.
1539 	 */
1540 	if (!btrfs_qgroup_full_accounting(fs_info))
1541 		return 0;
1542 
1543 	/*
1544 	 * Ensure dirty @src will be committed.  Or, after coming
1545 	 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1546 	 * recorded root will never be updated again, causing an outdated root
1547 	 * item.
1548 	 */
1549 	ret = record_root_in_trans(trans, src, 1);
1550 	if (ret)
1551 		return ret;
1552 
1553 	/*
1554 	 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1555 	 * src root, so we must run the delayed refs here.
1556 	 *
1557 	 * However this isn't particularly fool proof, because there's no
1558 	 * synchronization keeping us from changing the tree after this point
1559 	 * before we do the qgroup_inherit, or even from making changes while
1560 	 * we're doing the qgroup_inherit.  But that's a problem for the future,
1561 	 * for now flush the delayed refs to narrow the race window where the
1562 	 * qgroup counters could end up wrong.
1563 	 */
1564 	ret = btrfs_run_delayed_refs(trans, U64_MAX);
1565 	if (ret) {
1566 		btrfs_abort_transaction(trans, ret);
1567 		return ret;
1568 	}
1569 
1570 	ret = commit_fs_roots(trans);
1571 	if (ret)
1572 		goto out;
1573 	ret = btrfs_qgroup_account_extents(trans);
1574 	if (ret < 0)
1575 		goto out;
1576 
1577 	/* Now qgroup are all updated, we can inherit it to new qgroups */
1578 	ret = btrfs_qgroup_inherit(trans, btrfs_root_id(src), dst_objectid,
1579 				   btrfs_root_id(parent), inherit);
1580 	if (ret < 0)
1581 		goto out;
1582 
1583 	/*
1584 	 * Now we do a simplified commit transaction, which will:
1585 	 * 1) commit all subvolume and extent tree
1586 	 *    To ensure all subvolume and extent tree have a valid
1587 	 *    commit_root to accounting later insert_dir_item()
1588 	 * 2) write all btree blocks onto disk
1589 	 *    This is to make sure later btree modification will be cowed
1590 	 *    Or commit_root can be populated and cause wrong qgroup numbers
1591 	 * In this simplified commit, we don't really care about other trees
1592 	 * like chunk and root tree, as they won't affect qgroup.
1593 	 * And we don't write super to avoid half committed status.
1594 	 */
1595 	ret = commit_cowonly_roots(trans);
1596 	if (ret)
1597 		goto out;
1598 	switch_commit_roots(trans);
1599 	ret = btrfs_write_and_wait_transaction(trans);
1600 	if (ret)
1601 		btrfs_handle_fs_error(fs_info, ret,
1602 			"Error while writing out transaction for qgroup");
1603 
1604 out:
1605 	/*
1606 	 * Force parent root to be updated, as we recorded it before so its
1607 	 * last_trans == cur_transid.
1608 	 * Or it won't be committed again onto disk after later
1609 	 * insert_dir_item()
1610 	 */
1611 	if (!ret)
1612 		ret = record_root_in_trans(trans, parent, 1);
1613 	return ret;
1614 }
1615 
1616 /*
1617  * new snapshots need to be created at a very specific time in the
1618  * transaction commit.  This does the actual creation.
1619  *
1620  * Note:
1621  * If the error which may affect the commitment of the current transaction
1622  * happens, we should return the error number. If the error which just affect
1623  * the creation of the pending snapshots, just return 0.
1624  */
1625 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1626 				   struct btrfs_pending_snapshot *pending)
1627 {
1628 
1629 	struct btrfs_fs_info *fs_info = trans->fs_info;
1630 	struct btrfs_key key;
1631 	struct btrfs_root_item *new_root_item;
1632 	struct btrfs_root *tree_root = fs_info->tree_root;
1633 	struct btrfs_root *root = pending->root;
1634 	struct btrfs_root *parent_root;
1635 	struct btrfs_block_rsv *rsv;
1636 	struct inode *parent_inode = &pending->dir->vfs_inode;
1637 	struct btrfs_path *path;
1638 	struct btrfs_dir_item *dir_item;
1639 	struct extent_buffer *tmp;
1640 	struct extent_buffer *old;
1641 	struct timespec64 cur_time;
1642 	int ret = 0;
1643 	u64 to_reserve = 0;
1644 	u64 index = 0;
1645 	u64 objectid;
1646 	u64 root_flags;
1647 	unsigned int nofs_flags;
1648 	struct fscrypt_name fname;
1649 
1650 	ASSERT(pending->path);
1651 	path = pending->path;
1652 
1653 	ASSERT(pending->root_item);
1654 	new_root_item = pending->root_item;
1655 
1656 	/*
1657 	 * We're inside a transaction and must make sure that any potential
1658 	 * allocations with GFP_KERNEL in fscrypt won't recurse back to
1659 	 * filesystem.
1660 	 */
1661 	nofs_flags = memalloc_nofs_save();
1662 	pending->error = fscrypt_setup_filename(parent_inode,
1663 						&pending->dentry->d_name, 0,
1664 						&fname);
1665 	memalloc_nofs_restore(nofs_flags);
1666 	if (pending->error)
1667 		goto free_pending;
1668 
1669 	pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1670 	if (pending->error)
1671 		goto free_fname;
1672 
1673 	/*
1674 	 * Make qgroup to skip current new snapshot's qgroupid, as it is
1675 	 * accounted by later btrfs_qgroup_inherit().
1676 	 */
1677 	btrfs_set_skip_qgroup(trans, objectid);
1678 
1679 	btrfs_reloc_pre_snapshot(pending, &to_reserve);
1680 
1681 	if (to_reserve > 0) {
1682 		pending->error = btrfs_block_rsv_add(fs_info,
1683 						     &pending->block_rsv,
1684 						     to_reserve,
1685 						     BTRFS_RESERVE_NO_FLUSH);
1686 		if (pending->error)
1687 			goto clear_skip_qgroup;
1688 	}
1689 
1690 	key.objectid = objectid;
1691 	key.offset = (u64)-1;
1692 	key.type = BTRFS_ROOT_ITEM_KEY;
1693 
1694 	rsv = trans->block_rsv;
1695 	trans->block_rsv = &pending->block_rsv;
1696 	trans->bytes_reserved = trans->block_rsv->reserved;
1697 	trace_btrfs_space_reservation(fs_info, "transaction",
1698 				      trans->transid,
1699 				      trans->bytes_reserved, 1);
1700 	parent_root = BTRFS_I(parent_inode)->root;
1701 	ret = record_root_in_trans(trans, parent_root, 0);
1702 	if (ret)
1703 		goto fail;
1704 	cur_time = current_time(parent_inode);
1705 
1706 	/*
1707 	 * insert the directory item
1708 	 */
1709 	ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1710 	if (ret) {
1711 		btrfs_abort_transaction(trans, ret);
1712 		goto fail;
1713 	}
1714 
1715 	/* check if there is a file/dir which has the same name. */
1716 	dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1717 					 btrfs_ino(BTRFS_I(parent_inode)),
1718 					 &fname.disk_name, 0);
1719 	if (dir_item != NULL && !IS_ERR(dir_item)) {
1720 		pending->error = -EEXIST;
1721 		goto dir_item_existed;
1722 	} else if (IS_ERR(dir_item)) {
1723 		ret = PTR_ERR(dir_item);
1724 		btrfs_abort_transaction(trans, ret);
1725 		goto fail;
1726 	}
1727 	btrfs_release_path(path);
1728 
1729 	ret = btrfs_create_qgroup(trans, objectid);
1730 	if (ret && ret != -EEXIST) {
1731 		btrfs_abort_transaction(trans, ret);
1732 		goto fail;
1733 	}
1734 
1735 	/*
1736 	 * pull in the delayed directory update
1737 	 * and the delayed inode item
1738 	 * otherwise we corrupt the FS during
1739 	 * snapshot
1740 	 */
1741 	ret = btrfs_run_delayed_items(trans);
1742 	if (ret) {	/* Transaction aborted */
1743 		btrfs_abort_transaction(trans, ret);
1744 		goto fail;
1745 	}
1746 
1747 	ret = record_root_in_trans(trans, root, 0);
1748 	if (ret) {
1749 		btrfs_abort_transaction(trans, ret);
1750 		goto fail;
1751 	}
1752 	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1753 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1754 	btrfs_check_and_init_root_item(new_root_item);
1755 
1756 	root_flags = btrfs_root_flags(new_root_item);
1757 	if (pending->readonly)
1758 		root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1759 	else
1760 		root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1761 	btrfs_set_root_flags(new_root_item, root_flags);
1762 
1763 	btrfs_set_root_generation_v2(new_root_item,
1764 			trans->transid);
1765 	generate_random_guid(new_root_item->uuid);
1766 	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1767 			BTRFS_UUID_SIZE);
1768 	if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1769 		memset(new_root_item->received_uuid, 0,
1770 		       sizeof(new_root_item->received_uuid));
1771 		memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1772 		memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1773 		btrfs_set_root_stransid(new_root_item, 0);
1774 		btrfs_set_root_rtransid(new_root_item, 0);
1775 	}
1776 	btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1777 	btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1778 	btrfs_set_root_otransid(new_root_item, trans->transid);
1779 
1780 	old = btrfs_lock_root_node(root);
1781 	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1782 			      BTRFS_NESTING_COW);
1783 	if (ret) {
1784 		btrfs_tree_unlock(old);
1785 		free_extent_buffer(old);
1786 		btrfs_abort_transaction(trans, ret);
1787 		goto fail;
1788 	}
1789 
1790 	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1791 	/* clean up in any case */
1792 	btrfs_tree_unlock(old);
1793 	free_extent_buffer(old);
1794 	if (ret) {
1795 		btrfs_abort_transaction(trans, ret);
1796 		goto fail;
1797 	}
1798 	/* see comments in should_cow_block() */
1799 	set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1800 	smp_wmb();
1801 
1802 	btrfs_set_root_node(new_root_item, tmp);
1803 	/* record when the snapshot was created in key.offset */
1804 	key.offset = trans->transid;
1805 	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1806 	btrfs_tree_unlock(tmp);
1807 	free_extent_buffer(tmp);
1808 	if (ret) {
1809 		btrfs_abort_transaction(trans, ret);
1810 		goto fail;
1811 	}
1812 
1813 	/*
1814 	 * insert root back/forward references
1815 	 */
1816 	ret = btrfs_add_root_ref(trans, objectid,
1817 				 btrfs_root_id(parent_root),
1818 				 btrfs_ino(BTRFS_I(parent_inode)), index,
1819 				 &fname.disk_name);
1820 	if (ret) {
1821 		btrfs_abort_transaction(trans, ret);
1822 		goto fail;
1823 	}
1824 
1825 	key.offset = (u64)-1;
1826 	pending->snap = btrfs_get_new_fs_root(fs_info, objectid, &pending->anon_dev);
1827 	if (IS_ERR(pending->snap)) {
1828 		ret = PTR_ERR(pending->snap);
1829 		pending->snap = NULL;
1830 		btrfs_abort_transaction(trans, ret);
1831 		goto fail;
1832 	}
1833 
1834 	ret = btrfs_reloc_post_snapshot(trans, pending);
1835 	if (ret) {
1836 		btrfs_abort_transaction(trans, ret);
1837 		goto fail;
1838 	}
1839 
1840 	/*
1841 	 * Do special qgroup accounting for snapshot, as we do some qgroup
1842 	 * snapshot hack to do fast snapshot.
1843 	 * To co-operate with that hack, we do hack again.
1844 	 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1845 	 */
1846 	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL)
1847 		ret = qgroup_account_snapshot(trans, root, parent_root,
1848 					      pending->inherit, objectid);
1849 	else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
1850 		ret = btrfs_qgroup_inherit(trans, btrfs_root_id(root), objectid,
1851 					   btrfs_root_id(parent_root), pending->inherit);
1852 	if (ret < 0)
1853 		goto fail;
1854 
1855 	ret = btrfs_insert_dir_item(trans, &fname.disk_name,
1856 				    BTRFS_I(parent_inode), &key, BTRFS_FT_DIR,
1857 				    index);
1858 	if (ret) {
1859 		btrfs_abort_transaction(trans, ret);
1860 		goto fail;
1861 	}
1862 
1863 	btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1864 						  fname.disk_name.len * 2);
1865 	inode_set_mtime_to_ts(parent_inode,
1866 			      inode_set_ctime_current(parent_inode));
1867 	ret = btrfs_update_inode_fallback(trans, BTRFS_I(parent_inode));
1868 	if (ret) {
1869 		btrfs_abort_transaction(trans, ret);
1870 		goto fail;
1871 	}
1872 	ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1873 				  BTRFS_UUID_KEY_SUBVOL,
1874 				  objectid);
1875 	if (ret) {
1876 		btrfs_abort_transaction(trans, ret);
1877 		goto fail;
1878 	}
1879 	if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1880 		ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1881 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1882 					  objectid);
1883 		if (ret && ret != -EEXIST) {
1884 			btrfs_abort_transaction(trans, ret);
1885 			goto fail;
1886 		}
1887 	}
1888 
1889 fail:
1890 	pending->error = ret;
1891 dir_item_existed:
1892 	trans->block_rsv = rsv;
1893 	trans->bytes_reserved = 0;
1894 clear_skip_qgroup:
1895 	btrfs_clear_skip_qgroup(trans);
1896 free_fname:
1897 	fscrypt_free_filename(&fname);
1898 free_pending:
1899 	kfree(new_root_item);
1900 	pending->root_item = NULL;
1901 	btrfs_free_path(path);
1902 	pending->path = NULL;
1903 
1904 	return ret;
1905 }
1906 
1907 /*
1908  * create all the snapshots we've scheduled for creation
1909  */
1910 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1911 {
1912 	struct btrfs_pending_snapshot *pending, *next;
1913 	struct list_head *head = &trans->transaction->pending_snapshots;
1914 	int ret = 0;
1915 
1916 	list_for_each_entry_safe(pending, next, head, list) {
1917 		list_del(&pending->list);
1918 		ret = create_pending_snapshot(trans, pending);
1919 		if (ret)
1920 			break;
1921 	}
1922 	return ret;
1923 }
1924 
1925 static void update_super_roots(struct btrfs_fs_info *fs_info)
1926 {
1927 	struct btrfs_root_item *root_item;
1928 	struct btrfs_super_block *super;
1929 
1930 	super = fs_info->super_copy;
1931 
1932 	root_item = &fs_info->chunk_root->root_item;
1933 	super->chunk_root = root_item->bytenr;
1934 	super->chunk_root_generation = root_item->generation;
1935 	super->chunk_root_level = root_item->level;
1936 
1937 	root_item = &fs_info->tree_root->root_item;
1938 	super->root = root_item->bytenr;
1939 	super->generation = root_item->generation;
1940 	super->root_level = root_item->level;
1941 	if (btrfs_test_opt(fs_info, SPACE_CACHE))
1942 		super->cache_generation = root_item->generation;
1943 	else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1944 		super->cache_generation = 0;
1945 	if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1946 		super->uuid_tree_generation = root_item->generation;
1947 }
1948 
1949 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1950 {
1951 	struct btrfs_transaction *trans;
1952 	int ret = 0;
1953 
1954 	spin_lock(&info->trans_lock);
1955 	trans = info->running_transaction;
1956 	if (trans)
1957 		ret = is_transaction_blocked(trans);
1958 	spin_unlock(&info->trans_lock);
1959 	return ret;
1960 }
1961 
1962 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1963 {
1964 	struct btrfs_fs_info *fs_info = trans->fs_info;
1965 	struct btrfs_transaction *cur_trans;
1966 
1967 	/* Kick the transaction kthread. */
1968 	set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
1969 	wake_up_process(fs_info->transaction_kthread);
1970 
1971 	/* take transaction reference */
1972 	cur_trans = trans->transaction;
1973 	refcount_inc(&cur_trans->use_count);
1974 
1975 	btrfs_end_transaction(trans);
1976 
1977 	/*
1978 	 * Wait for the current transaction commit to start and block
1979 	 * subsequent transaction joins
1980 	 */
1981 	btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
1982 	wait_event(fs_info->transaction_blocked_wait,
1983 		   cur_trans->state >= TRANS_STATE_COMMIT_START ||
1984 		   TRANS_ABORTED(cur_trans));
1985 	btrfs_put_transaction(cur_trans);
1986 }
1987 
1988 /*
1989  * If there is a running transaction commit it or if it's already committing,
1990  * wait for its commit to complete. Does not start and commit a new transaction
1991  * if there isn't any running.
1992  */
1993 int btrfs_commit_current_transaction(struct btrfs_root *root)
1994 {
1995 	struct btrfs_trans_handle *trans;
1996 
1997 	trans = btrfs_attach_transaction_barrier(root);
1998 	if (IS_ERR(trans)) {
1999 		int ret = PTR_ERR(trans);
2000 
2001 		return (ret == -ENOENT) ? 0 : ret;
2002 	}
2003 
2004 	return btrfs_commit_transaction(trans);
2005 }
2006 
2007 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
2008 {
2009 	struct btrfs_fs_info *fs_info = trans->fs_info;
2010 	struct btrfs_transaction *cur_trans = trans->transaction;
2011 
2012 	WARN_ON(refcount_read(&trans->use_count) > 1);
2013 
2014 	btrfs_abort_transaction(trans, err);
2015 
2016 	spin_lock(&fs_info->trans_lock);
2017 
2018 	/*
2019 	 * If the transaction is removed from the list, it means this
2020 	 * transaction has been committed successfully, so it is impossible
2021 	 * to call the cleanup function.
2022 	 */
2023 	BUG_ON(list_empty(&cur_trans->list));
2024 
2025 	if (cur_trans == fs_info->running_transaction) {
2026 		cur_trans->state = TRANS_STATE_COMMIT_DOING;
2027 		spin_unlock(&fs_info->trans_lock);
2028 
2029 		/*
2030 		 * The thread has already released the lockdep map as reader
2031 		 * already in btrfs_commit_transaction().
2032 		 */
2033 		btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2034 		wait_event(cur_trans->writer_wait,
2035 			   atomic_read(&cur_trans->num_writers) == 1);
2036 
2037 		spin_lock(&fs_info->trans_lock);
2038 	}
2039 
2040 	/*
2041 	 * Now that we know no one else is still using the transaction we can
2042 	 * remove the transaction from the list of transactions. This avoids
2043 	 * the transaction kthread from cleaning up the transaction while some
2044 	 * other task is still using it, which could result in a use-after-free
2045 	 * on things like log trees, as it forces the transaction kthread to
2046 	 * wait for this transaction to be cleaned up by us.
2047 	 */
2048 	list_del_init(&cur_trans->list);
2049 
2050 	spin_unlock(&fs_info->trans_lock);
2051 
2052 	btrfs_cleanup_one_transaction(trans->transaction);
2053 
2054 	spin_lock(&fs_info->trans_lock);
2055 	if (cur_trans == fs_info->running_transaction)
2056 		fs_info->running_transaction = NULL;
2057 	spin_unlock(&fs_info->trans_lock);
2058 
2059 	if (trans->type & __TRANS_FREEZABLE)
2060 		sb_end_intwrite(fs_info->sb);
2061 	btrfs_put_transaction(cur_trans);
2062 	btrfs_put_transaction(cur_trans);
2063 
2064 	trace_btrfs_transaction_commit(fs_info);
2065 
2066 	if (current->journal_info == trans)
2067 		current->journal_info = NULL;
2068 
2069 	/*
2070 	 * If relocation is running, we can't cancel scrub because that will
2071 	 * result in a deadlock. Before relocating a block group, relocation
2072 	 * pauses scrub, then starts and commits a transaction before unpausing
2073 	 * scrub. If the transaction commit is being done by the relocation
2074 	 * task or triggered by another task and the relocation task is waiting
2075 	 * for the commit, and we end up here due to an error in the commit
2076 	 * path, then calling btrfs_scrub_cancel() will deadlock, as we are
2077 	 * asking for scrub to stop while having it asked to be paused higher
2078 	 * above in relocation code.
2079 	 */
2080 	if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
2081 		btrfs_scrub_cancel(fs_info);
2082 
2083 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
2084 }
2085 
2086 /*
2087  * Release reserved delayed ref space of all pending block groups of the
2088  * transaction and remove them from the list
2089  */
2090 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2091 {
2092        struct btrfs_fs_info *fs_info = trans->fs_info;
2093        struct btrfs_block_group *block_group, *tmp;
2094 
2095        list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2096                btrfs_dec_delayed_refs_rsv_bg_inserts(fs_info);
2097                list_del_init(&block_group->bg_list);
2098        }
2099 }
2100 
2101 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2102 {
2103 	/*
2104 	 * We use try_to_writeback_inodes_sb() here because if we used
2105 	 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2106 	 * Currently are holding the fs freeze lock, if we do an async flush
2107 	 * we'll do btrfs_join_transaction() and deadlock because we need to
2108 	 * wait for the fs freeze lock.  Using the direct flushing we benefit
2109 	 * from already being in a transaction and our join_transaction doesn't
2110 	 * have to re-take the fs freeze lock.
2111 	 *
2112 	 * Note that try_to_writeback_inodes_sb() will only trigger writeback
2113 	 * if it can read lock sb->s_umount. It will always be able to lock it,
2114 	 * except when the filesystem is being unmounted or being frozen, but in
2115 	 * those cases sync_filesystem() is called, which results in calling
2116 	 * writeback_inodes_sb() while holding a write lock on sb->s_umount.
2117 	 * Note that we don't call writeback_inodes_sb() directly, because it
2118 	 * will emit a warning if sb->s_umount is not locked.
2119 	 */
2120 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2121 		try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2122 	return 0;
2123 }
2124 
2125 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2126 {
2127 	if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2128 		btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
2129 }
2130 
2131 /*
2132  * Add a pending snapshot associated with the given transaction handle to the
2133  * respective handle. This must be called after the transaction commit started
2134  * and while holding fs_info->trans_lock.
2135  * This serves to guarantee a caller of btrfs_commit_transaction() that it can
2136  * safely free the pending snapshot pointer in case btrfs_commit_transaction()
2137  * returns an error.
2138  */
2139 static void add_pending_snapshot(struct btrfs_trans_handle *trans)
2140 {
2141 	struct btrfs_transaction *cur_trans = trans->transaction;
2142 
2143 	if (!trans->pending_snapshot)
2144 		return;
2145 
2146 	lockdep_assert_held(&trans->fs_info->trans_lock);
2147 	ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_PREP);
2148 
2149 	list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots);
2150 }
2151 
2152 static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval)
2153 {
2154 	fs_info->commit_stats.commit_count++;
2155 	fs_info->commit_stats.last_commit_dur = interval;
2156 	fs_info->commit_stats.max_commit_dur =
2157 			max_t(u64, fs_info->commit_stats.max_commit_dur, interval);
2158 	fs_info->commit_stats.total_commit_dur += interval;
2159 }
2160 
2161 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2162 {
2163 	struct btrfs_fs_info *fs_info = trans->fs_info;
2164 	struct btrfs_transaction *cur_trans = trans->transaction;
2165 	struct btrfs_transaction *prev_trans = NULL;
2166 	int ret;
2167 	ktime_t start_time;
2168 	ktime_t interval;
2169 
2170 	ASSERT(refcount_read(&trans->use_count) == 1);
2171 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2172 
2173 	clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags);
2174 
2175 	/* Stop the commit early if ->aborted is set */
2176 	if (TRANS_ABORTED(cur_trans)) {
2177 		ret = cur_trans->aborted;
2178 		goto lockdep_trans_commit_start_release;
2179 	}
2180 
2181 	btrfs_trans_release_metadata(trans);
2182 	trans->block_rsv = NULL;
2183 
2184 	/*
2185 	 * We only want one transaction commit doing the flushing so we do not
2186 	 * waste a bunch of time on lock contention on the extent root node.
2187 	 */
2188 	if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2189 			      &cur_trans->delayed_refs.flags)) {
2190 		/*
2191 		 * Make a pass through all the delayed refs we have so far.
2192 		 * Any running threads may add more while we are here.
2193 		 */
2194 		ret = btrfs_run_delayed_refs(trans, 0);
2195 		if (ret)
2196 			goto lockdep_trans_commit_start_release;
2197 	}
2198 
2199 	btrfs_create_pending_block_groups(trans);
2200 
2201 	if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2202 		int run_it = 0;
2203 
2204 		/* this mutex is also taken before trying to set
2205 		 * block groups readonly.  We need to make sure
2206 		 * that nobody has set a block group readonly
2207 		 * after a extents from that block group have been
2208 		 * allocated for cache files.  btrfs_set_block_group_ro
2209 		 * will wait for the transaction to commit if it
2210 		 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2211 		 *
2212 		 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2213 		 * only one process starts all the block group IO.  It wouldn't
2214 		 * hurt to have more than one go through, but there's no
2215 		 * real advantage to it either.
2216 		 */
2217 		mutex_lock(&fs_info->ro_block_group_mutex);
2218 		if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2219 				      &cur_trans->flags))
2220 			run_it = 1;
2221 		mutex_unlock(&fs_info->ro_block_group_mutex);
2222 
2223 		if (run_it) {
2224 			ret = btrfs_start_dirty_block_groups(trans);
2225 			if (ret)
2226 				goto lockdep_trans_commit_start_release;
2227 		}
2228 	}
2229 
2230 	spin_lock(&fs_info->trans_lock);
2231 	if (cur_trans->state >= TRANS_STATE_COMMIT_PREP) {
2232 		enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2233 
2234 		add_pending_snapshot(trans);
2235 
2236 		spin_unlock(&fs_info->trans_lock);
2237 		refcount_inc(&cur_trans->use_count);
2238 
2239 		if (trans->in_fsync)
2240 			want_state = TRANS_STATE_SUPER_COMMITTED;
2241 
2242 		btrfs_trans_state_lockdep_release(fs_info,
2243 						  BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2244 		ret = btrfs_end_transaction(trans);
2245 		wait_for_commit(cur_trans, want_state);
2246 
2247 		if (TRANS_ABORTED(cur_trans))
2248 			ret = cur_trans->aborted;
2249 
2250 		btrfs_put_transaction(cur_trans);
2251 
2252 		return ret;
2253 	}
2254 
2255 	cur_trans->state = TRANS_STATE_COMMIT_PREP;
2256 	wake_up(&fs_info->transaction_blocked_wait);
2257 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2258 
2259 	if (cur_trans->list.prev != &fs_info->trans_list) {
2260 		enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2261 
2262 		if (trans->in_fsync)
2263 			want_state = TRANS_STATE_SUPER_COMMITTED;
2264 
2265 		prev_trans = list_entry(cur_trans->list.prev,
2266 					struct btrfs_transaction, list);
2267 		if (prev_trans->state < want_state) {
2268 			refcount_inc(&prev_trans->use_count);
2269 			spin_unlock(&fs_info->trans_lock);
2270 
2271 			wait_for_commit(prev_trans, want_state);
2272 
2273 			ret = READ_ONCE(prev_trans->aborted);
2274 
2275 			btrfs_put_transaction(prev_trans);
2276 			if (ret)
2277 				goto lockdep_release;
2278 			spin_lock(&fs_info->trans_lock);
2279 		}
2280 	} else {
2281 		/*
2282 		 * The previous transaction was aborted and was already removed
2283 		 * from the list of transactions at fs_info->trans_list. So we
2284 		 * abort to prevent writing a new superblock that reflects a
2285 		 * corrupt state (pointing to trees with unwritten nodes/leafs).
2286 		 */
2287 		if (BTRFS_FS_ERROR(fs_info)) {
2288 			spin_unlock(&fs_info->trans_lock);
2289 			ret = -EROFS;
2290 			goto lockdep_release;
2291 		}
2292 	}
2293 
2294 	cur_trans->state = TRANS_STATE_COMMIT_START;
2295 	wake_up(&fs_info->transaction_blocked_wait);
2296 	spin_unlock(&fs_info->trans_lock);
2297 
2298 	/*
2299 	 * Get the time spent on the work done by the commit thread and not
2300 	 * the time spent waiting on a previous commit
2301 	 */
2302 	start_time = ktime_get_ns();
2303 
2304 	extwriter_counter_dec(cur_trans, trans->type);
2305 
2306 	ret = btrfs_start_delalloc_flush(fs_info);
2307 	if (ret)
2308 		goto lockdep_release;
2309 
2310 	ret = btrfs_run_delayed_items(trans);
2311 	if (ret)
2312 		goto lockdep_release;
2313 
2314 	/*
2315 	 * The thread has started/joined the transaction thus it holds the
2316 	 * lockdep map as a reader. It has to release it before acquiring the
2317 	 * lockdep map as a writer.
2318 	 */
2319 	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2320 	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters);
2321 	wait_event(cur_trans->writer_wait,
2322 		   extwriter_counter_read(cur_trans) == 0);
2323 
2324 	/* some pending stuffs might be added after the previous flush. */
2325 	ret = btrfs_run_delayed_items(trans);
2326 	if (ret) {
2327 		btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2328 		goto cleanup_transaction;
2329 	}
2330 
2331 	btrfs_wait_delalloc_flush(fs_info);
2332 
2333 	/*
2334 	 * Wait for all ordered extents started by a fast fsync that joined this
2335 	 * transaction. Otherwise if this transaction commits before the ordered
2336 	 * extents complete we lose logged data after a power failure.
2337 	 */
2338 	btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered);
2339 	wait_event(cur_trans->pending_wait,
2340 		   atomic_read(&cur_trans->pending_ordered) == 0);
2341 
2342 	btrfs_scrub_pause(fs_info);
2343 	/*
2344 	 * Ok now we need to make sure to block out any other joins while we
2345 	 * commit the transaction.  We could have started a join before setting
2346 	 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2347 	 */
2348 	spin_lock(&fs_info->trans_lock);
2349 	add_pending_snapshot(trans);
2350 	cur_trans->state = TRANS_STATE_COMMIT_DOING;
2351 	spin_unlock(&fs_info->trans_lock);
2352 
2353 	/*
2354 	 * The thread has started/joined the transaction thus it holds the
2355 	 * lockdep map as a reader. It has to release it before acquiring the
2356 	 * lockdep map as a writer.
2357 	 */
2358 	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2359 	btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2360 	wait_event(cur_trans->writer_wait,
2361 		   atomic_read(&cur_trans->num_writers) == 1);
2362 
2363 	/*
2364 	 * Make lockdep happy by acquiring the state locks after
2365 	 * btrfs_trans_num_writers is released. If we acquired the state locks
2366 	 * before releasing the btrfs_trans_num_writers lock then lockdep would
2367 	 * complain because we did not follow the reverse order unlocking rule.
2368 	 */
2369 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2370 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2371 	btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2372 
2373 	/*
2374 	 * We've started the commit, clear the flag in case we were triggered to
2375 	 * do an async commit but somebody else started before the transaction
2376 	 * kthread could do the work.
2377 	 */
2378 	clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
2379 
2380 	if (TRANS_ABORTED(cur_trans)) {
2381 		ret = cur_trans->aborted;
2382 		btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2383 		goto scrub_continue;
2384 	}
2385 	/*
2386 	 * the reloc mutex makes sure that we stop
2387 	 * the balancing code from coming in and moving
2388 	 * extents around in the middle of the commit
2389 	 */
2390 	mutex_lock(&fs_info->reloc_mutex);
2391 
2392 	/*
2393 	 * We needn't worry about the delayed items because we will
2394 	 * deal with them in create_pending_snapshot(), which is the
2395 	 * core function of the snapshot creation.
2396 	 */
2397 	ret = create_pending_snapshots(trans);
2398 	if (ret)
2399 		goto unlock_reloc;
2400 
2401 	/*
2402 	 * We insert the dir indexes of the snapshots and update the inode
2403 	 * of the snapshots' parents after the snapshot creation, so there
2404 	 * are some delayed items which are not dealt with. Now deal with
2405 	 * them.
2406 	 *
2407 	 * We needn't worry that this operation will corrupt the snapshots,
2408 	 * because all the tree which are snapshoted will be forced to COW
2409 	 * the nodes and leaves.
2410 	 */
2411 	ret = btrfs_run_delayed_items(trans);
2412 	if (ret)
2413 		goto unlock_reloc;
2414 
2415 	ret = btrfs_run_delayed_refs(trans, U64_MAX);
2416 	if (ret)
2417 		goto unlock_reloc;
2418 
2419 	/*
2420 	 * make sure none of the code above managed to slip in a
2421 	 * delayed item
2422 	 */
2423 	btrfs_assert_delayed_root_empty(fs_info);
2424 
2425 	WARN_ON(cur_trans != trans->transaction);
2426 
2427 	ret = commit_fs_roots(trans);
2428 	if (ret)
2429 		goto unlock_reloc;
2430 
2431 	/* commit_fs_roots gets rid of all the tree log roots, it is now
2432 	 * safe to free the root of tree log roots
2433 	 */
2434 	btrfs_free_log_root_tree(trans, fs_info);
2435 
2436 	/*
2437 	 * Since fs roots are all committed, we can get a quite accurate
2438 	 * new_roots. So let's do quota accounting.
2439 	 */
2440 	ret = btrfs_qgroup_account_extents(trans);
2441 	if (ret < 0)
2442 		goto unlock_reloc;
2443 
2444 	ret = commit_cowonly_roots(trans);
2445 	if (ret)
2446 		goto unlock_reloc;
2447 
2448 	/*
2449 	 * The tasks which save the space cache and inode cache may also
2450 	 * update ->aborted, check it.
2451 	 */
2452 	if (TRANS_ABORTED(cur_trans)) {
2453 		ret = cur_trans->aborted;
2454 		goto unlock_reloc;
2455 	}
2456 
2457 	cur_trans = fs_info->running_transaction;
2458 
2459 	btrfs_set_root_node(&fs_info->tree_root->root_item,
2460 			    fs_info->tree_root->node);
2461 	list_add_tail(&fs_info->tree_root->dirty_list,
2462 		      &cur_trans->switch_commits);
2463 
2464 	btrfs_set_root_node(&fs_info->chunk_root->root_item,
2465 			    fs_info->chunk_root->node);
2466 	list_add_tail(&fs_info->chunk_root->dirty_list,
2467 		      &cur_trans->switch_commits);
2468 
2469 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2470 		btrfs_set_root_node(&fs_info->block_group_root->root_item,
2471 				    fs_info->block_group_root->node);
2472 		list_add_tail(&fs_info->block_group_root->dirty_list,
2473 			      &cur_trans->switch_commits);
2474 	}
2475 
2476 	switch_commit_roots(trans);
2477 
2478 	ASSERT(list_empty(&cur_trans->dirty_bgs));
2479 	ASSERT(list_empty(&cur_trans->io_bgs));
2480 	update_super_roots(fs_info);
2481 
2482 	btrfs_set_super_log_root(fs_info->super_copy, 0);
2483 	btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2484 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2485 	       sizeof(*fs_info->super_copy));
2486 
2487 	btrfs_commit_device_sizes(cur_trans);
2488 
2489 	clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2490 	clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2491 
2492 	btrfs_trans_release_chunk_metadata(trans);
2493 
2494 	/*
2495 	 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and
2496 	 * setting fs_info->running_transaction to NULL, lock tree_log_mutex to
2497 	 * make sure that before we commit our superblock, no other task can
2498 	 * start a new transaction and commit a log tree before we commit our
2499 	 * superblock. Anyone trying to commit a log tree locks this mutex before
2500 	 * writing its superblock.
2501 	 */
2502 	mutex_lock(&fs_info->tree_log_mutex);
2503 
2504 	spin_lock(&fs_info->trans_lock);
2505 	cur_trans->state = TRANS_STATE_UNBLOCKED;
2506 	fs_info->running_transaction = NULL;
2507 	spin_unlock(&fs_info->trans_lock);
2508 	mutex_unlock(&fs_info->reloc_mutex);
2509 
2510 	wake_up(&fs_info->transaction_wait);
2511 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2512 
2513 	/* If we have features changed, wake up the cleaner to update sysfs. */
2514 	if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) &&
2515 	    fs_info->cleaner_kthread)
2516 		wake_up_process(fs_info->cleaner_kthread);
2517 
2518 	ret = btrfs_write_and_wait_transaction(trans);
2519 	if (ret) {
2520 		btrfs_handle_fs_error(fs_info, ret,
2521 				      "Error while writing out transaction");
2522 		mutex_unlock(&fs_info->tree_log_mutex);
2523 		goto scrub_continue;
2524 	}
2525 
2526 	ret = write_all_supers(fs_info, 0);
2527 	/*
2528 	 * the super is written, we can safely allow the tree-loggers
2529 	 * to go about their business
2530 	 */
2531 	mutex_unlock(&fs_info->tree_log_mutex);
2532 	if (ret)
2533 		goto scrub_continue;
2534 
2535 	/*
2536 	 * We needn't acquire the lock here because there is no other task
2537 	 * which can change it.
2538 	 */
2539 	cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2540 	wake_up(&cur_trans->commit_wait);
2541 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2542 
2543 	btrfs_finish_extent_commit(trans);
2544 
2545 	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2546 		btrfs_clear_space_info_full(fs_info);
2547 
2548 	btrfs_set_last_trans_committed(fs_info, cur_trans->transid);
2549 	/*
2550 	 * We needn't acquire the lock here because there is no other task
2551 	 * which can change it.
2552 	 */
2553 	cur_trans->state = TRANS_STATE_COMPLETED;
2554 	wake_up(&cur_trans->commit_wait);
2555 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2556 
2557 	spin_lock(&fs_info->trans_lock);
2558 	list_del_init(&cur_trans->list);
2559 	spin_unlock(&fs_info->trans_lock);
2560 
2561 	btrfs_put_transaction(cur_trans);
2562 	btrfs_put_transaction(cur_trans);
2563 
2564 	if (trans->type & __TRANS_FREEZABLE)
2565 		sb_end_intwrite(fs_info->sb);
2566 
2567 	trace_btrfs_transaction_commit(fs_info);
2568 
2569 	interval = ktime_get_ns() - start_time;
2570 
2571 	btrfs_scrub_continue(fs_info);
2572 
2573 	if (current->journal_info == trans)
2574 		current->journal_info = NULL;
2575 
2576 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
2577 
2578 	update_commit_stats(fs_info, interval);
2579 
2580 	return ret;
2581 
2582 unlock_reloc:
2583 	mutex_unlock(&fs_info->reloc_mutex);
2584 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2585 scrub_continue:
2586 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2587 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2588 	btrfs_scrub_continue(fs_info);
2589 cleanup_transaction:
2590 	btrfs_trans_release_metadata(trans);
2591 	btrfs_cleanup_pending_block_groups(trans);
2592 	btrfs_trans_release_chunk_metadata(trans);
2593 	trans->block_rsv = NULL;
2594 	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2595 	if (current->journal_info == trans)
2596 		current->journal_info = NULL;
2597 	cleanup_transaction(trans, ret);
2598 
2599 	return ret;
2600 
2601 lockdep_release:
2602 	btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2603 	btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2604 	goto cleanup_transaction;
2605 
2606 lockdep_trans_commit_start_release:
2607 	btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2608 	btrfs_end_transaction(trans);
2609 	return ret;
2610 }
2611 
2612 /*
2613  * return < 0 if error
2614  * 0 if there are no more dead_roots at the time of call
2615  * 1 there are more to be processed, call me again
2616  *
2617  * The return value indicates there are certainly more snapshots to delete, but
2618  * if there comes a new one during processing, it may return 0. We don't mind,
2619  * because btrfs_commit_super will poke cleaner thread and it will process it a
2620  * few seconds later.
2621  */
2622 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
2623 {
2624 	struct btrfs_root *root;
2625 	int ret;
2626 
2627 	spin_lock(&fs_info->trans_lock);
2628 	if (list_empty(&fs_info->dead_roots)) {
2629 		spin_unlock(&fs_info->trans_lock);
2630 		return 0;
2631 	}
2632 	root = list_first_entry(&fs_info->dead_roots,
2633 			struct btrfs_root, root_list);
2634 	list_del_init(&root->root_list);
2635 	spin_unlock(&fs_info->trans_lock);
2636 
2637 	btrfs_debug(fs_info, "cleaner removing %llu", btrfs_root_id(root));
2638 
2639 	btrfs_kill_all_delayed_nodes(root);
2640 
2641 	if (btrfs_header_backref_rev(root->node) <
2642 			BTRFS_MIXED_BACKREF_REV)
2643 		ret = btrfs_drop_snapshot(root, 0, 0);
2644 	else
2645 		ret = btrfs_drop_snapshot(root, 1, 0);
2646 
2647 	btrfs_put_root(root);
2648 	return (ret < 0) ? 0 : 1;
2649 }
2650 
2651 /*
2652  * We only mark the transaction aborted and then set the file system read-only.
2653  * This will prevent new transactions from starting or trying to join this
2654  * one.
2655  *
2656  * This means that error recovery at the call site is limited to freeing
2657  * any local memory allocations and passing the error code up without
2658  * further cleanup. The transaction should complete as it normally would
2659  * in the call path but will return -EIO.
2660  *
2661  * We'll complete the cleanup in btrfs_end_transaction and
2662  * btrfs_commit_transaction.
2663  */
2664 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
2665 				      const char *function,
2666 				      unsigned int line, int error, bool first_hit)
2667 {
2668 	struct btrfs_fs_info *fs_info = trans->fs_info;
2669 
2670 	WRITE_ONCE(trans->aborted, error);
2671 	WRITE_ONCE(trans->transaction->aborted, error);
2672 	if (first_hit && error == -ENOSPC)
2673 		btrfs_dump_space_info_for_trans_abort(fs_info);
2674 	/* Wake up anybody who may be waiting on this transaction */
2675 	wake_up(&fs_info->transaction_wait);
2676 	wake_up(&fs_info->transaction_blocked_wait);
2677 	__btrfs_handle_fs_error(fs_info, function, line, error, NULL);
2678 }
2679 
2680 int __init btrfs_transaction_init(void)
2681 {
2682 	btrfs_trans_handle_cachep = KMEM_CACHE(btrfs_trans_handle, SLAB_TEMPORARY);
2683 	if (!btrfs_trans_handle_cachep)
2684 		return -ENOMEM;
2685 	return 0;
2686 }
2687 
2688 void __cold btrfs_transaction_exit(void)
2689 {
2690 	kmem_cache_destroy(btrfs_trans_handle_cachep);
2691 }
2692