xref: /openbmc/linux/fs/btrfs/transaction.c (revision 6774def6)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/fs.h>
20 #include <linux/slab.h>
21 #include <linux/sched.h>
22 #include <linux/writeback.h>
23 #include <linux/pagemap.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
26 #include "ctree.h"
27 #include "disk-io.h"
28 #include "transaction.h"
29 #include "locking.h"
30 #include "tree-log.h"
31 #include "inode-map.h"
32 #include "volumes.h"
33 #include "dev-replace.h"
34 #include "qgroup.h"
35 
36 #define BTRFS_ROOT_TRANS_TAG 0
37 
38 static unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
39 	[TRANS_STATE_RUNNING]		= 0U,
40 	[TRANS_STATE_BLOCKED]		= (__TRANS_USERSPACE |
41 					   __TRANS_START),
42 	[TRANS_STATE_COMMIT_START]	= (__TRANS_USERSPACE |
43 					   __TRANS_START |
44 					   __TRANS_ATTACH),
45 	[TRANS_STATE_COMMIT_DOING]	= (__TRANS_USERSPACE |
46 					   __TRANS_START |
47 					   __TRANS_ATTACH |
48 					   __TRANS_JOIN),
49 	[TRANS_STATE_UNBLOCKED]		= (__TRANS_USERSPACE |
50 					   __TRANS_START |
51 					   __TRANS_ATTACH |
52 					   __TRANS_JOIN |
53 					   __TRANS_JOIN_NOLOCK),
54 	[TRANS_STATE_COMPLETED]		= (__TRANS_USERSPACE |
55 					   __TRANS_START |
56 					   __TRANS_ATTACH |
57 					   __TRANS_JOIN |
58 					   __TRANS_JOIN_NOLOCK),
59 };
60 
61 void btrfs_put_transaction(struct btrfs_transaction *transaction)
62 {
63 	WARN_ON(atomic_read(&transaction->use_count) == 0);
64 	if (atomic_dec_and_test(&transaction->use_count)) {
65 		BUG_ON(!list_empty(&transaction->list));
66 		WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
67 		while (!list_empty(&transaction->pending_chunks)) {
68 			struct extent_map *em;
69 
70 			em = list_first_entry(&transaction->pending_chunks,
71 					      struct extent_map, list);
72 			list_del_init(&em->list);
73 			free_extent_map(em);
74 		}
75 		kmem_cache_free(btrfs_transaction_cachep, transaction);
76 	}
77 }
78 
79 static noinline void switch_commit_roots(struct btrfs_transaction *trans,
80 					 struct btrfs_fs_info *fs_info)
81 {
82 	struct btrfs_root *root, *tmp;
83 
84 	down_write(&fs_info->commit_root_sem);
85 	list_for_each_entry_safe(root, tmp, &trans->switch_commits,
86 				 dirty_list) {
87 		list_del_init(&root->dirty_list);
88 		free_extent_buffer(root->commit_root);
89 		root->commit_root = btrfs_root_node(root);
90 		if (is_fstree(root->objectid))
91 			btrfs_unpin_free_ino(root);
92 	}
93 	up_write(&fs_info->commit_root_sem);
94 }
95 
96 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
97 					 unsigned int type)
98 {
99 	if (type & TRANS_EXTWRITERS)
100 		atomic_inc(&trans->num_extwriters);
101 }
102 
103 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
104 					 unsigned int type)
105 {
106 	if (type & TRANS_EXTWRITERS)
107 		atomic_dec(&trans->num_extwriters);
108 }
109 
110 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
111 					  unsigned int type)
112 {
113 	atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
114 }
115 
116 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
117 {
118 	return atomic_read(&trans->num_extwriters);
119 }
120 
121 /*
122  * either allocate a new transaction or hop into the existing one
123  */
124 static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
125 {
126 	struct btrfs_transaction *cur_trans;
127 	struct btrfs_fs_info *fs_info = root->fs_info;
128 
129 	spin_lock(&fs_info->trans_lock);
130 loop:
131 	/* The file system has been taken offline. No new transactions. */
132 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
133 		spin_unlock(&fs_info->trans_lock);
134 		return -EROFS;
135 	}
136 
137 	cur_trans = fs_info->running_transaction;
138 	if (cur_trans) {
139 		if (cur_trans->aborted) {
140 			spin_unlock(&fs_info->trans_lock);
141 			return cur_trans->aborted;
142 		}
143 		if (btrfs_blocked_trans_types[cur_trans->state] & type) {
144 			spin_unlock(&fs_info->trans_lock);
145 			return -EBUSY;
146 		}
147 		atomic_inc(&cur_trans->use_count);
148 		atomic_inc(&cur_trans->num_writers);
149 		extwriter_counter_inc(cur_trans, type);
150 		spin_unlock(&fs_info->trans_lock);
151 		return 0;
152 	}
153 	spin_unlock(&fs_info->trans_lock);
154 
155 	/*
156 	 * If we are ATTACH, we just want to catch the current transaction,
157 	 * and commit it. If there is no transaction, just return ENOENT.
158 	 */
159 	if (type == TRANS_ATTACH)
160 		return -ENOENT;
161 
162 	/*
163 	 * JOIN_NOLOCK only happens during the transaction commit, so
164 	 * it is impossible that ->running_transaction is NULL
165 	 */
166 	BUG_ON(type == TRANS_JOIN_NOLOCK);
167 
168 	cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
169 	if (!cur_trans)
170 		return -ENOMEM;
171 
172 	spin_lock(&fs_info->trans_lock);
173 	if (fs_info->running_transaction) {
174 		/*
175 		 * someone started a transaction after we unlocked.  Make sure
176 		 * to redo the checks above
177 		 */
178 		kmem_cache_free(btrfs_transaction_cachep, cur_trans);
179 		goto loop;
180 	} else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
181 		spin_unlock(&fs_info->trans_lock);
182 		kmem_cache_free(btrfs_transaction_cachep, cur_trans);
183 		return -EROFS;
184 	}
185 
186 	atomic_set(&cur_trans->num_writers, 1);
187 	extwriter_counter_init(cur_trans, type);
188 	init_waitqueue_head(&cur_trans->writer_wait);
189 	init_waitqueue_head(&cur_trans->commit_wait);
190 	cur_trans->state = TRANS_STATE_RUNNING;
191 	/*
192 	 * One for this trans handle, one so it will live on until we
193 	 * commit the transaction.
194 	 */
195 	atomic_set(&cur_trans->use_count, 2);
196 	cur_trans->start_time = get_seconds();
197 
198 	cur_trans->delayed_refs.href_root = RB_ROOT;
199 	atomic_set(&cur_trans->delayed_refs.num_entries, 0);
200 	cur_trans->delayed_refs.num_heads_ready = 0;
201 	cur_trans->delayed_refs.num_heads = 0;
202 	cur_trans->delayed_refs.flushing = 0;
203 	cur_trans->delayed_refs.run_delayed_start = 0;
204 
205 	/*
206 	 * although the tree mod log is per file system and not per transaction,
207 	 * the log must never go across transaction boundaries.
208 	 */
209 	smp_mb();
210 	if (!list_empty(&fs_info->tree_mod_seq_list))
211 		WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
212 			"creating a fresh transaction\n");
213 	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
214 		WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
215 			"creating a fresh transaction\n");
216 	atomic64_set(&fs_info->tree_mod_seq, 0);
217 
218 	spin_lock_init(&cur_trans->delayed_refs.lock);
219 
220 	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
221 	INIT_LIST_HEAD(&cur_trans->pending_chunks);
222 	INIT_LIST_HEAD(&cur_trans->switch_commits);
223 	list_add_tail(&cur_trans->list, &fs_info->trans_list);
224 	extent_io_tree_init(&cur_trans->dirty_pages,
225 			     fs_info->btree_inode->i_mapping);
226 	fs_info->generation++;
227 	cur_trans->transid = fs_info->generation;
228 	fs_info->running_transaction = cur_trans;
229 	cur_trans->aborted = 0;
230 	spin_unlock(&fs_info->trans_lock);
231 
232 	return 0;
233 }
234 
235 /*
236  * this does all the record keeping required to make sure that a reference
237  * counted root is properly recorded in a given transaction.  This is required
238  * to make sure the old root from before we joined the transaction is deleted
239  * when the transaction commits
240  */
241 static int record_root_in_trans(struct btrfs_trans_handle *trans,
242 			       struct btrfs_root *root)
243 {
244 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
245 	    root->last_trans < trans->transid) {
246 		WARN_ON(root == root->fs_info->extent_root);
247 		WARN_ON(root->commit_root != root->node);
248 
249 		/*
250 		 * see below for IN_TRANS_SETUP usage rules
251 		 * we have the reloc mutex held now, so there
252 		 * is only one writer in this function
253 		 */
254 		set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
255 
256 		/* make sure readers find IN_TRANS_SETUP before
257 		 * they find our root->last_trans update
258 		 */
259 		smp_wmb();
260 
261 		spin_lock(&root->fs_info->fs_roots_radix_lock);
262 		if (root->last_trans == trans->transid) {
263 			spin_unlock(&root->fs_info->fs_roots_radix_lock);
264 			return 0;
265 		}
266 		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
267 			   (unsigned long)root->root_key.objectid,
268 			   BTRFS_ROOT_TRANS_TAG);
269 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
270 		root->last_trans = trans->transid;
271 
272 		/* this is pretty tricky.  We don't want to
273 		 * take the relocation lock in btrfs_record_root_in_trans
274 		 * unless we're really doing the first setup for this root in
275 		 * this transaction.
276 		 *
277 		 * Normally we'd use root->last_trans as a flag to decide
278 		 * if we want to take the expensive mutex.
279 		 *
280 		 * But, we have to set root->last_trans before we
281 		 * init the relocation root, otherwise, we trip over warnings
282 		 * in ctree.c.  The solution used here is to flag ourselves
283 		 * with root IN_TRANS_SETUP.  When this is 1, we're still
284 		 * fixing up the reloc trees and everyone must wait.
285 		 *
286 		 * When this is zero, they can trust root->last_trans and fly
287 		 * through btrfs_record_root_in_trans without having to take the
288 		 * lock.  smp_wmb() makes sure that all the writes above are
289 		 * done before we pop in the zero below
290 		 */
291 		btrfs_init_reloc_root(trans, root);
292 		smp_mb__before_atomic();
293 		clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
294 	}
295 	return 0;
296 }
297 
298 
299 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
300 			       struct btrfs_root *root)
301 {
302 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
303 		return 0;
304 
305 	/*
306 	 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
307 	 * and barriers
308 	 */
309 	smp_rmb();
310 	if (root->last_trans == trans->transid &&
311 	    !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
312 		return 0;
313 
314 	mutex_lock(&root->fs_info->reloc_mutex);
315 	record_root_in_trans(trans, root);
316 	mutex_unlock(&root->fs_info->reloc_mutex);
317 
318 	return 0;
319 }
320 
321 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
322 {
323 	return (trans->state >= TRANS_STATE_BLOCKED &&
324 		trans->state < TRANS_STATE_UNBLOCKED &&
325 		!trans->aborted);
326 }
327 
328 /* wait for commit against the current transaction to become unblocked
329  * when this is done, it is safe to start a new transaction, but the current
330  * transaction might not be fully on disk.
331  */
332 static void wait_current_trans(struct btrfs_root *root)
333 {
334 	struct btrfs_transaction *cur_trans;
335 
336 	spin_lock(&root->fs_info->trans_lock);
337 	cur_trans = root->fs_info->running_transaction;
338 	if (cur_trans && is_transaction_blocked(cur_trans)) {
339 		atomic_inc(&cur_trans->use_count);
340 		spin_unlock(&root->fs_info->trans_lock);
341 
342 		wait_event(root->fs_info->transaction_wait,
343 			   cur_trans->state >= TRANS_STATE_UNBLOCKED ||
344 			   cur_trans->aborted);
345 		btrfs_put_transaction(cur_trans);
346 	} else {
347 		spin_unlock(&root->fs_info->trans_lock);
348 	}
349 }
350 
351 static int may_wait_transaction(struct btrfs_root *root, int type)
352 {
353 	if (root->fs_info->log_root_recovering)
354 		return 0;
355 
356 	if (type == TRANS_USERSPACE)
357 		return 1;
358 
359 	if (type == TRANS_START &&
360 	    !atomic_read(&root->fs_info->open_ioctl_trans))
361 		return 1;
362 
363 	return 0;
364 }
365 
366 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
367 {
368 	if (!root->fs_info->reloc_ctl ||
369 	    !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
370 	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
371 	    root->reloc_root)
372 		return false;
373 
374 	return true;
375 }
376 
377 static struct btrfs_trans_handle *
378 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
379 		  enum btrfs_reserve_flush_enum flush)
380 {
381 	struct btrfs_trans_handle *h;
382 	struct btrfs_transaction *cur_trans;
383 	u64 num_bytes = 0;
384 	u64 qgroup_reserved = 0;
385 	bool reloc_reserved = false;
386 	int ret;
387 
388 	/* Send isn't supposed to start transactions. */
389 	ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
390 
391 	if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
392 		return ERR_PTR(-EROFS);
393 
394 	if (current->journal_info) {
395 		WARN_ON(type & TRANS_EXTWRITERS);
396 		h = current->journal_info;
397 		h->use_count++;
398 		WARN_ON(h->use_count > 2);
399 		h->orig_rsv = h->block_rsv;
400 		h->block_rsv = NULL;
401 		goto got_it;
402 	}
403 
404 	/*
405 	 * Do the reservation before we join the transaction so we can do all
406 	 * the appropriate flushing if need be.
407 	 */
408 	if (num_items > 0 && root != root->fs_info->chunk_root) {
409 		if (root->fs_info->quota_enabled &&
410 		    is_fstree(root->root_key.objectid)) {
411 			qgroup_reserved = num_items * root->nodesize;
412 			ret = btrfs_qgroup_reserve(root, qgroup_reserved);
413 			if (ret)
414 				return ERR_PTR(ret);
415 		}
416 
417 		num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
418 		/*
419 		 * Do the reservation for the relocation root creation
420 		 */
421 		if (need_reserve_reloc_root(root)) {
422 			num_bytes += root->nodesize;
423 			reloc_reserved = true;
424 		}
425 
426 		ret = btrfs_block_rsv_add(root,
427 					  &root->fs_info->trans_block_rsv,
428 					  num_bytes, flush);
429 		if (ret)
430 			goto reserve_fail;
431 	}
432 again:
433 	h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
434 	if (!h) {
435 		ret = -ENOMEM;
436 		goto alloc_fail;
437 	}
438 
439 	/*
440 	 * If we are JOIN_NOLOCK we're already committing a transaction and
441 	 * waiting on this guy, so we don't need to do the sb_start_intwrite
442 	 * because we're already holding a ref.  We need this because we could
443 	 * have raced in and did an fsync() on a file which can kick a commit
444 	 * and then we deadlock with somebody doing a freeze.
445 	 *
446 	 * If we are ATTACH, it means we just want to catch the current
447 	 * transaction and commit it, so we needn't do sb_start_intwrite().
448 	 */
449 	if (type & __TRANS_FREEZABLE)
450 		sb_start_intwrite(root->fs_info->sb);
451 
452 	if (may_wait_transaction(root, type))
453 		wait_current_trans(root);
454 
455 	do {
456 		ret = join_transaction(root, type);
457 		if (ret == -EBUSY) {
458 			wait_current_trans(root);
459 			if (unlikely(type == TRANS_ATTACH))
460 				ret = -ENOENT;
461 		}
462 	} while (ret == -EBUSY);
463 
464 	if (ret < 0) {
465 		/* We must get the transaction if we are JOIN_NOLOCK. */
466 		BUG_ON(type == TRANS_JOIN_NOLOCK);
467 		goto join_fail;
468 	}
469 
470 	cur_trans = root->fs_info->running_transaction;
471 
472 	h->transid = cur_trans->transid;
473 	h->transaction = cur_trans;
474 	h->blocks_used = 0;
475 	h->bytes_reserved = 0;
476 	h->root = root;
477 	h->delayed_ref_updates = 0;
478 	h->use_count = 1;
479 	h->adding_csums = 0;
480 	h->block_rsv = NULL;
481 	h->orig_rsv = NULL;
482 	h->aborted = 0;
483 	h->qgroup_reserved = 0;
484 	h->delayed_ref_elem.seq = 0;
485 	h->type = type;
486 	h->allocating_chunk = false;
487 	h->reloc_reserved = false;
488 	h->sync = false;
489 	INIT_LIST_HEAD(&h->qgroup_ref_list);
490 	INIT_LIST_HEAD(&h->new_bgs);
491 
492 	smp_mb();
493 	if (cur_trans->state >= TRANS_STATE_BLOCKED &&
494 	    may_wait_transaction(root, type)) {
495 		current->journal_info = h;
496 		btrfs_commit_transaction(h, root);
497 		goto again;
498 	}
499 
500 	if (num_bytes) {
501 		trace_btrfs_space_reservation(root->fs_info, "transaction",
502 					      h->transid, num_bytes, 1);
503 		h->block_rsv = &root->fs_info->trans_block_rsv;
504 		h->bytes_reserved = num_bytes;
505 		h->reloc_reserved = reloc_reserved;
506 	}
507 	h->qgroup_reserved = qgroup_reserved;
508 
509 got_it:
510 	btrfs_record_root_in_trans(h, root);
511 
512 	if (!current->journal_info && type != TRANS_USERSPACE)
513 		current->journal_info = h;
514 	return h;
515 
516 join_fail:
517 	if (type & __TRANS_FREEZABLE)
518 		sb_end_intwrite(root->fs_info->sb);
519 	kmem_cache_free(btrfs_trans_handle_cachep, h);
520 alloc_fail:
521 	if (num_bytes)
522 		btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
523 					num_bytes);
524 reserve_fail:
525 	if (qgroup_reserved)
526 		btrfs_qgroup_free(root, qgroup_reserved);
527 	return ERR_PTR(ret);
528 }
529 
530 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
531 						   int num_items)
532 {
533 	return start_transaction(root, num_items, TRANS_START,
534 				 BTRFS_RESERVE_FLUSH_ALL);
535 }
536 
537 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
538 					struct btrfs_root *root, int num_items)
539 {
540 	return start_transaction(root, num_items, TRANS_START,
541 				 BTRFS_RESERVE_FLUSH_LIMIT);
542 }
543 
544 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
545 {
546 	return start_transaction(root, 0, TRANS_JOIN, 0);
547 }
548 
549 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
550 {
551 	return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
552 }
553 
554 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
555 {
556 	return start_transaction(root, 0, TRANS_USERSPACE, 0);
557 }
558 
559 /*
560  * btrfs_attach_transaction() - catch the running transaction
561  *
562  * It is used when we want to commit the current the transaction, but
563  * don't want to start a new one.
564  *
565  * Note: If this function return -ENOENT, it just means there is no
566  * running transaction. But it is possible that the inactive transaction
567  * is still in the memory, not fully on disk. If you hope there is no
568  * inactive transaction in the fs when -ENOENT is returned, you should
569  * invoke
570  *     btrfs_attach_transaction_barrier()
571  */
572 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
573 {
574 	return start_transaction(root, 0, TRANS_ATTACH, 0);
575 }
576 
577 /*
578  * btrfs_attach_transaction_barrier() - catch the running transaction
579  *
580  * It is similar to the above function, the differentia is this one
581  * will wait for all the inactive transactions until they fully
582  * complete.
583  */
584 struct btrfs_trans_handle *
585 btrfs_attach_transaction_barrier(struct btrfs_root *root)
586 {
587 	struct btrfs_trans_handle *trans;
588 
589 	trans = start_transaction(root, 0, TRANS_ATTACH, 0);
590 	if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
591 		btrfs_wait_for_commit(root, 0);
592 
593 	return trans;
594 }
595 
596 /* wait for a transaction commit to be fully complete */
597 static noinline void wait_for_commit(struct btrfs_root *root,
598 				    struct btrfs_transaction *commit)
599 {
600 	wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
601 }
602 
603 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
604 {
605 	struct btrfs_transaction *cur_trans = NULL, *t;
606 	int ret = 0;
607 
608 	if (transid) {
609 		if (transid <= root->fs_info->last_trans_committed)
610 			goto out;
611 
612 		/* find specified transaction */
613 		spin_lock(&root->fs_info->trans_lock);
614 		list_for_each_entry(t, &root->fs_info->trans_list, list) {
615 			if (t->transid == transid) {
616 				cur_trans = t;
617 				atomic_inc(&cur_trans->use_count);
618 				ret = 0;
619 				break;
620 			}
621 			if (t->transid > transid) {
622 				ret = 0;
623 				break;
624 			}
625 		}
626 		spin_unlock(&root->fs_info->trans_lock);
627 
628 		/*
629 		 * The specified transaction doesn't exist, or we
630 		 * raced with btrfs_commit_transaction
631 		 */
632 		if (!cur_trans) {
633 			if (transid > root->fs_info->last_trans_committed)
634 				ret = -EINVAL;
635 			goto out;
636 		}
637 	} else {
638 		/* find newest transaction that is committing | committed */
639 		spin_lock(&root->fs_info->trans_lock);
640 		list_for_each_entry_reverse(t, &root->fs_info->trans_list,
641 					    list) {
642 			if (t->state >= TRANS_STATE_COMMIT_START) {
643 				if (t->state == TRANS_STATE_COMPLETED)
644 					break;
645 				cur_trans = t;
646 				atomic_inc(&cur_trans->use_count);
647 				break;
648 			}
649 		}
650 		spin_unlock(&root->fs_info->trans_lock);
651 		if (!cur_trans)
652 			goto out;  /* nothing committing|committed */
653 	}
654 
655 	wait_for_commit(root, cur_trans);
656 	btrfs_put_transaction(cur_trans);
657 out:
658 	return ret;
659 }
660 
661 void btrfs_throttle(struct btrfs_root *root)
662 {
663 	if (!atomic_read(&root->fs_info->open_ioctl_trans))
664 		wait_current_trans(root);
665 }
666 
667 static int should_end_transaction(struct btrfs_trans_handle *trans,
668 				  struct btrfs_root *root)
669 {
670 	if (root->fs_info->global_block_rsv.space_info->full &&
671 	    btrfs_check_space_for_delayed_refs(trans, root))
672 		return 1;
673 
674 	return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
675 }
676 
677 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
678 				 struct btrfs_root *root)
679 {
680 	struct btrfs_transaction *cur_trans = trans->transaction;
681 	int updates;
682 	int err;
683 
684 	smp_mb();
685 	if (cur_trans->state >= TRANS_STATE_BLOCKED ||
686 	    cur_trans->delayed_refs.flushing)
687 		return 1;
688 
689 	updates = trans->delayed_ref_updates;
690 	trans->delayed_ref_updates = 0;
691 	if (updates) {
692 		err = btrfs_run_delayed_refs(trans, root, updates);
693 		if (err) /* Error code will also eval true */
694 			return err;
695 	}
696 
697 	return should_end_transaction(trans, root);
698 }
699 
700 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
701 			  struct btrfs_root *root, int throttle)
702 {
703 	struct btrfs_transaction *cur_trans = trans->transaction;
704 	struct btrfs_fs_info *info = root->fs_info;
705 	unsigned long cur = trans->delayed_ref_updates;
706 	int lock = (trans->type != TRANS_JOIN_NOLOCK);
707 	int err = 0;
708 	int must_run_delayed_refs = 0;
709 
710 	if (trans->use_count > 1) {
711 		trans->use_count--;
712 		trans->block_rsv = trans->orig_rsv;
713 		return 0;
714 	}
715 
716 	btrfs_trans_release_metadata(trans, root);
717 	trans->block_rsv = NULL;
718 
719 	if (!list_empty(&trans->new_bgs))
720 		btrfs_create_pending_block_groups(trans, root);
721 
722 	trans->delayed_ref_updates = 0;
723 	if (!trans->sync) {
724 		must_run_delayed_refs =
725 			btrfs_should_throttle_delayed_refs(trans, root);
726 		cur = max_t(unsigned long, cur, 32);
727 
728 		/*
729 		 * don't make the caller wait if they are from a NOLOCK
730 		 * or ATTACH transaction, it will deadlock with commit
731 		 */
732 		if (must_run_delayed_refs == 1 &&
733 		    (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
734 			must_run_delayed_refs = 2;
735 	}
736 
737 	if (trans->qgroup_reserved) {
738 		/*
739 		 * the same root has to be passed here between start_transaction
740 		 * and end_transaction. Subvolume quota depends on this.
741 		 */
742 		btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
743 		trans->qgroup_reserved = 0;
744 	}
745 
746 	btrfs_trans_release_metadata(trans, root);
747 	trans->block_rsv = NULL;
748 
749 	if (!list_empty(&trans->new_bgs))
750 		btrfs_create_pending_block_groups(trans, root);
751 
752 	if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
753 	    should_end_transaction(trans, root) &&
754 	    ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
755 		spin_lock(&info->trans_lock);
756 		if (cur_trans->state == TRANS_STATE_RUNNING)
757 			cur_trans->state = TRANS_STATE_BLOCKED;
758 		spin_unlock(&info->trans_lock);
759 	}
760 
761 	if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
762 		if (throttle)
763 			return btrfs_commit_transaction(trans, root);
764 		else
765 			wake_up_process(info->transaction_kthread);
766 	}
767 
768 	if (trans->type & __TRANS_FREEZABLE)
769 		sb_end_intwrite(root->fs_info->sb);
770 
771 	WARN_ON(cur_trans != info->running_transaction);
772 	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
773 	atomic_dec(&cur_trans->num_writers);
774 	extwriter_counter_dec(cur_trans, trans->type);
775 
776 	smp_mb();
777 	if (waitqueue_active(&cur_trans->writer_wait))
778 		wake_up(&cur_trans->writer_wait);
779 	btrfs_put_transaction(cur_trans);
780 
781 	if (current->journal_info == trans)
782 		current->journal_info = NULL;
783 
784 	if (throttle)
785 		btrfs_run_delayed_iputs(root);
786 
787 	if (trans->aborted ||
788 	    test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
789 		wake_up_process(info->transaction_kthread);
790 		err = -EIO;
791 	}
792 	assert_qgroups_uptodate(trans);
793 
794 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
795 	if (must_run_delayed_refs) {
796 		btrfs_async_run_delayed_refs(root, cur,
797 					     must_run_delayed_refs == 1);
798 	}
799 	return err;
800 }
801 
802 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
803 			  struct btrfs_root *root)
804 {
805 	return __btrfs_end_transaction(trans, root, 0);
806 }
807 
808 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
809 				   struct btrfs_root *root)
810 {
811 	return __btrfs_end_transaction(trans, root, 1);
812 }
813 
814 /*
815  * when btree blocks are allocated, they have some corresponding bits set for
816  * them in one of two extent_io trees.  This is used to make sure all of
817  * those extents are sent to disk but does not wait on them
818  */
819 int btrfs_write_marked_extents(struct btrfs_root *root,
820 			       struct extent_io_tree *dirty_pages, int mark)
821 {
822 	int err = 0;
823 	int werr = 0;
824 	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
825 	struct extent_state *cached_state = NULL;
826 	u64 start = 0;
827 	u64 end;
828 
829 	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
830 				      mark, &cached_state)) {
831 		convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
832 				   mark, &cached_state, GFP_NOFS);
833 		cached_state = NULL;
834 		err = filemap_fdatawrite_range(mapping, start, end);
835 		if (err)
836 			werr = err;
837 		cond_resched();
838 		start = end + 1;
839 	}
840 	if (err)
841 		werr = err;
842 	return werr;
843 }
844 
845 /*
846  * when btree blocks are allocated, they have some corresponding bits set for
847  * them in one of two extent_io trees.  This is used to make sure all of
848  * those extents are on disk for transaction or log commit.  We wait
849  * on all the pages and clear them from the dirty pages state tree
850  */
851 int btrfs_wait_marked_extents(struct btrfs_root *root,
852 			      struct extent_io_tree *dirty_pages, int mark)
853 {
854 	int err = 0;
855 	int werr = 0;
856 	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
857 	struct extent_state *cached_state = NULL;
858 	u64 start = 0;
859 	u64 end;
860 	struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
861 	bool errors = false;
862 
863 	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
864 				      EXTENT_NEED_WAIT, &cached_state)) {
865 		clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
866 				 0, 0, &cached_state, GFP_NOFS);
867 		err = filemap_fdatawait_range(mapping, start, end);
868 		if (err)
869 			werr = err;
870 		cond_resched();
871 		start = end + 1;
872 	}
873 	if (err)
874 		werr = err;
875 
876 	if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
877 		if ((mark & EXTENT_DIRTY) &&
878 		    test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
879 				       &btree_ino->runtime_flags))
880 			errors = true;
881 
882 		if ((mark & EXTENT_NEW) &&
883 		    test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
884 				       &btree_ino->runtime_flags))
885 			errors = true;
886 	} else {
887 		if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
888 				       &btree_ino->runtime_flags))
889 			errors = true;
890 	}
891 
892 	if (errors && !werr)
893 		werr = -EIO;
894 
895 	return werr;
896 }
897 
898 /*
899  * when btree blocks are allocated, they have some corresponding bits set for
900  * them in one of two extent_io trees.  This is used to make sure all of
901  * those extents are on disk for transaction or log commit
902  */
903 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
904 				struct extent_io_tree *dirty_pages, int mark)
905 {
906 	int ret;
907 	int ret2;
908 	struct blk_plug plug;
909 
910 	blk_start_plug(&plug);
911 	ret = btrfs_write_marked_extents(root, dirty_pages, mark);
912 	blk_finish_plug(&plug);
913 	ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
914 
915 	if (ret)
916 		return ret;
917 	if (ret2)
918 		return ret2;
919 	return 0;
920 }
921 
922 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
923 				     struct btrfs_root *root)
924 {
925 	if (!trans || !trans->transaction) {
926 		struct inode *btree_inode;
927 		btree_inode = root->fs_info->btree_inode;
928 		return filemap_write_and_wait(btree_inode->i_mapping);
929 	}
930 	return btrfs_write_and_wait_marked_extents(root,
931 					   &trans->transaction->dirty_pages,
932 					   EXTENT_DIRTY);
933 }
934 
935 /*
936  * this is used to update the root pointer in the tree of tree roots.
937  *
938  * But, in the case of the extent allocation tree, updating the root
939  * pointer may allocate blocks which may change the root of the extent
940  * allocation tree.
941  *
942  * So, this loops and repeats and makes sure the cowonly root didn't
943  * change while the root pointer was being updated in the metadata.
944  */
945 static int update_cowonly_root(struct btrfs_trans_handle *trans,
946 			       struct btrfs_root *root)
947 {
948 	int ret;
949 	u64 old_root_bytenr;
950 	u64 old_root_used;
951 	struct btrfs_root *tree_root = root->fs_info->tree_root;
952 
953 	old_root_used = btrfs_root_used(&root->root_item);
954 	btrfs_write_dirty_block_groups(trans, root);
955 
956 	while (1) {
957 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
958 		if (old_root_bytenr == root->node->start &&
959 		    old_root_used == btrfs_root_used(&root->root_item))
960 			break;
961 
962 		btrfs_set_root_node(&root->root_item, root->node);
963 		ret = btrfs_update_root(trans, tree_root,
964 					&root->root_key,
965 					&root->root_item);
966 		if (ret)
967 			return ret;
968 
969 		old_root_used = btrfs_root_used(&root->root_item);
970 		ret = btrfs_write_dirty_block_groups(trans, root);
971 		if (ret)
972 			return ret;
973 	}
974 
975 	return 0;
976 }
977 
978 /*
979  * update all the cowonly tree roots on disk
980  *
981  * The error handling in this function may not be obvious. Any of the
982  * failures will cause the file system to go offline. We still need
983  * to clean up the delayed refs.
984  */
985 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
986 					 struct btrfs_root *root)
987 {
988 	struct btrfs_fs_info *fs_info = root->fs_info;
989 	struct list_head *next;
990 	struct extent_buffer *eb;
991 	int ret;
992 
993 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
994 	if (ret)
995 		return ret;
996 
997 	eb = btrfs_lock_root_node(fs_info->tree_root);
998 	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
999 			      0, &eb);
1000 	btrfs_tree_unlock(eb);
1001 	free_extent_buffer(eb);
1002 
1003 	if (ret)
1004 		return ret;
1005 
1006 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1007 	if (ret)
1008 		return ret;
1009 
1010 	ret = btrfs_run_dev_stats(trans, root->fs_info);
1011 	if (ret)
1012 		return ret;
1013 	ret = btrfs_run_dev_replace(trans, root->fs_info);
1014 	if (ret)
1015 		return ret;
1016 	ret = btrfs_run_qgroups(trans, root->fs_info);
1017 	if (ret)
1018 		return ret;
1019 
1020 	/* run_qgroups might have added some more refs */
1021 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1022 	if (ret)
1023 		return ret;
1024 
1025 	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1026 		next = fs_info->dirty_cowonly_roots.next;
1027 		list_del_init(next);
1028 		root = list_entry(next, struct btrfs_root, dirty_list);
1029 
1030 		if (root != fs_info->extent_root)
1031 			list_add_tail(&root->dirty_list,
1032 				      &trans->transaction->switch_commits);
1033 		ret = update_cowonly_root(trans, root);
1034 		if (ret)
1035 			return ret;
1036 	}
1037 
1038 	list_add_tail(&fs_info->extent_root->dirty_list,
1039 		      &trans->transaction->switch_commits);
1040 	btrfs_after_dev_replace_commit(fs_info);
1041 
1042 	return 0;
1043 }
1044 
1045 /*
1046  * dead roots are old snapshots that need to be deleted.  This allocates
1047  * a dirty root struct and adds it into the list of dead roots that need to
1048  * be deleted
1049  */
1050 void btrfs_add_dead_root(struct btrfs_root *root)
1051 {
1052 	spin_lock(&root->fs_info->trans_lock);
1053 	if (list_empty(&root->root_list))
1054 		list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1055 	spin_unlock(&root->fs_info->trans_lock);
1056 }
1057 
1058 /*
1059  * update all the cowonly tree roots on disk
1060  */
1061 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1062 				    struct btrfs_root *root)
1063 {
1064 	struct btrfs_root *gang[8];
1065 	struct btrfs_fs_info *fs_info = root->fs_info;
1066 	int i;
1067 	int ret;
1068 	int err = 0;
1069 
1070 	spin_lock(&fs_info->fs_roots_radix_lock);
1071 	while (1) {
1072 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1073 						 (void **)gang, 0,
1074 						 ARRAY_SIZE(gang),
1075 						 BTRFS_ROOT_TRANS_TAG);
1076 		if (ret == 0)
1077 			break;
1078 		for (i = 0; i < ret; i++) {
1079 			root = gang[i];
1080 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
1081 					(unsigned long)root->root_key.objectid,
1082 					BTRFS_ROOT_TRANS_TAG);
1083 			spin_unlock(&fs_info->fs_roots_radix_lock);
1084 
1085 			btrfs_free_log(trans, root);
1086 			btrfs_update_reloc_root(trans, root);
1087 			btrfs_orphan_commit_root(trans, root);
1088 
1089 			btrfs_save_ino_cache(root, trans);
1090 
1091 			/* see comments in should_cow_block() */
1092 			clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1093 			smp_mb__after_atomic();
1094 
1095 			if (root->commit_root != root->node) {
1096 				list_add_tail(&root->dirty_list,
1097 					&trans->transaction->switch_commits);
1098 				btrfs_set_root_node(&root->root_item,
1099 						    root->node);
1100 			}
1101 
1102 			err = btrfs_update_root(trans, fs_info->tree_root,
1103 						&root->root_key,
1104 						&root->root_item);
1105 			spin_lock(&fs_info->fs_roots_radix_lock);
1106 			if (err)
1107 				break;
1108 		}
1109 	}
1110 	spin_unlock(&fs_info->fs_roots_radix_lock);
1111 	return err;
1112 }
1113 
1114 /*
1115  * defrag a given btree.
1116  * Every leaf in the btree is read and defragged.
1117  */
1118 int btrfs_defrag_root(struct btrfs_root *root)
1119 {
1120 	struct btrfs_fs_info *info = root->fs_info;
1121 	struct btrfs_trans_handle *trans;
1122 	int ret;
1123 
1124 	if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1125 		return 0;
1126 
1127 	while (1) {
1128 		trans = btrfs_start_transaction(root, 0);
1129 		if (IS_ERR(trans))
1130 			return PTR_ERR(trans);
1131 
1132 		ret = btrfs_defrag_leaves(trans, root);
1133 
1134 		btrfs_end_transaction(trans, root);
1135 		btrfs_btree_balance_dirty(info->tree_root);
1136 		cond_resched();
1137 
1138 		if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1139 			break;
1140 
1141 		if (btrfs_defrag_cancelled(root->fs_info)) {
1142 			pr_debug("BTRFS: defrag_root cancelled\n");
1143 			ret = -EAGAIN;
1144 			break;
1145 		}
1146 	}
1147 	clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1148 	return ret;
1149 }
1150 
1151 /*
1152  * new snapshots need to be created at a very specific time in the
1153  * transaction commit.  This does the actual creation.
1154  *
1155  * Note:
1156  * If the error which may affect the commitment of the current transaction
1157  * happens, we should return the error number. If the error which just affect
1158  * the creation of the pending snapshots, just return 0.
1159  */
1160 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1161 				   struct btrfs_fs_info *fs_info,
1162 				   struct btrfs_pending_snapshot *pending)
1163 {
1164 	struct btrfs_key key;
1165 	struct btrfs_root_item *new_root_item;
1166 	struct btrfs_root *tree_root = fs_info->tree_root;
1167 	struct btrfs_root *root = pending->root;
1168 	struct btrfs_root *parent_root;
1169 	struct btrfs_block_rsv *rsv;
1170 	struct inode *parent_inode;
1171 	struct btrfs_path *path;
1172 	struct btrfs_dir_item *dir_item;
1173 	struct dentry *dentry;
1174 	struct extent_buffer *tmp;
1175 	struct extent_buffer *old;
1176 	struct timespec cur_time = CURRENT_TIME;
1177 	int ret = 0;
1178 	u64 to_reserve = 0;
1179 	u64 index = 0;
1180 	u64 objectid;
1181 	u64 root_flags;
1182 	uuid_le new_uuid;
1183 
1184 	path = btrfs_alloc_path();
1185 	if (!path) {
1186 		pending->error = -ENOMEM;
1187 		return 0;
1188 	}
1189 
1190 	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1191 	if (!new_root_item) {
1192 		pending->error = -ENOMEM;
1193 		goto root_item_alloc_fail;
1194 	}
1195 
1196 	pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1197 	if (pending->error)
1198 		goto no_free_objectid;
1199 
1200 	btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1201 
1202 	if (to_reserve > 0) {
1203 		pending->error = btrfs_block_rsv_add(root,
1204 						     &pending->block_rsv,
1205 						     to_reserve,
1206 						     BTRFS_RESERVE_NO_FLUSH);
1207 		if (pending->error)
1208 			goto no_free_objectid;
1209 	}
1210 
1211 	key.objectid = objectid;
1212 	key.offset = (u64)-1;
1213 	key.type = BTRFS_ROOT_ITEM_KEY;
1214 
1215 	rsv = trans->block_rsv;
1216 	trans->block_rsv = &pending->block_rsv;
1217 	trans->bytes_reserved = trans->block_rsv->reserved;
1218 
1219 	dentry = pending->dentry;
1220 	parent_inode = pending->dir;
1221 	parent_root = BTRFS_I(parent_inode)->root;
1222 	record_root_in_trans(trans, parent_root);
1223 
1224 	/*
1225 	 * insert the directory item
1226 	 */
1227 	ret = btrfs_set_inode_index(parent_inode, &index);
1228 	BUG_ON(ret); /* -ENOMEM */
1229 
1230 	/* check if there is a file/dir which has the same name. */
1231 	dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1232 					 btrfs_ino(parent_inode),
1233 					 dentry->d_name.name,
1234 					 dentry->d_name.len, 0);
1235 	if (dir_item != NULL && !IS_ERR(dir_item)) {
1236 		pending->error = -EEXIST;
1237 		goto dir_item_existed;
1238 	} else if (IS_ERR(dir_item)) {
1239 		ret = PTR_ERR(dir_item);
1240 		btrfs_abort_transaction(trans, root, ret);
1241 		goto fail;
1242 	}
1243 	btrfs_release_path(path);
1244 
1245 	/*
1246 	 * pull in the delayed directory update
1247 	 * and the delayed inode item
1248 	 * otherwise we corrupt the FS during
1249 	 * snapshot
1250 	 */
1251 	ret = btrfs_run_delayed_items(trans, root);
1252 	if (ret) {	/* Transaction aborted */
1253 		btrfs_abort_transaction(trans, root, ret);
1254 		goto fail;
1255 	}
1256 
1257 	record_root_in_trans(trans, root);
1258 	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1259 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1260 	btrfs_check_and_init_root_item(new_root_item);
1261 
1262 	root_flags = btrfs_root_flags(new_root_item);
1263 	if (pending->readonly)
1264 		root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1265 	else
1266 		root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1267 	btrfs_set_root_flags(new_root_item, root_flags);
1268 
1269 	btrfs_set_root_generation_v2(new_root_item,
1270 			trans->transid);
1271 	uuid_le_gen(&new_uuid);
1272 	memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1273 	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1274 			BTRFS_UUID_SIZE);
1275 	if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1276 		memset(new_root_item->received_uuid, 0,
1277 		       sizeof(new_root_item->received_uuid));
1278 		memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1279 		memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1280 		btrfs_set_root_stransid(new_root_item, 0);
1281 		btrfs_set_root_rtransid(new_root_item, 0);
1282 	}
1283 	btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1284 	btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1285 	btrfs_set_root_otransid(new_root_item, trans->transid);
1286 
1287 	old = btrfs_lock_root_node(root);
1288 	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1289 	if (ret) {
1290 		btrfs_tree_unlock(old);
1291 		free_extent_buffer(old);
1292 		btrfs_abort_transaction(trans, root, ret);
1293 		goto fail;
1294 	}
1295 
1296 	btrfs_set_lock_blocking(old);
1297 
1298 	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1299 	/* clean up in any case */
1300 	btrfs_tree_unlock(old);
1301 	free_extent_buffer(old);
1302 	if (ret) {
1303 		btrfs_abort_transaction(trans, root, ret);
1304 		goto fail;
1305 	}
1306 
1307 	/*
1308 	 * We need to flush delayed refs in order to make sure all of our quota
1309 	 * operations have been done before we call btrfs_qgroup_inherit.
1310 	 */
1311 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1312 	if (ret) {
1313 		btrfs_abort_transaction(trans, root, ret);
1314 		goto fail;
1315 	}
1316 
1317 	ret = btrfs_qgroup_inherit(trans, fs_info,
1318 				   root->root_key.objectid,
1319 				   objectid, pending->inherit);
1320 	if (ret) {
1321 		btrfs_abort_transaction(trans, root, ret);
1322 		goto fail;
1323 	}
1324 
1325 	/* see comments in should_cow_block() */
1326 	set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1327 	smp_wmb();
1328 
1329 	btrfs_set_root_node(new_root_item, tmp);
1330 	/* record when the snapshot was created in key.offset */
1331 	key.offset = trans->transid;
1332 	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1333 	btrfs_tree_unlock(tmp);
1334 	free_extent_buffer(tmp);
1335 	if (ret) {
1336 		btrfs_abort_transaction(trans, root, ret);
1337 		goto fail;
1338 	}
1339 
1340 	/*
1341 	 * insert root back/forward references
1342 	 */
1343 	ret = btrfs_add_root_ref(trans, tree_root, objectid,
1344 				 parent_root->root_key.objectid,
1345 				 btrfs_ino(parent_inode), index,
1346 				 dentry->d_name.name, dentry->d_name.len);
1347 	if (ret) {
1348 		btrfs_abort_transaction(trans, root, ret);
1349 		goto fail;
1350 	}
1351 
1352 	key.offset = (u64)-1;
1353 	pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1354 	if (IS_ERR(pending->snap)) {
1355 		ret = PTR_ERR(pending->snap);
1356 		btrfs_abort_transaction(trans, root, ret);
1357 		goto fail;
1358 	}
1359 
1360 	ret = btrfs_reloc_post_snapshot(trans, pending);
1361 	if (ret) {
1362 		btrfs_abort_transaction(trans, root, ret);
1363 		goto fail;
1364 	}
1365 
1366 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1367 	if (ret) {
1368 		btrfs_abort_transaction(trans, root, ret);
1369 		goto fail;
1370 	}
1371 
1372 	ret = btrfs_insert_dir_item(trans, parent_root,
1373 				    dentry->d_name.name, dentry->d_name.len,
1374 				    parent_inode, &key,
1375 				    BTRFS_FT_DIR, index);
1376 	/* We have check then name at the beginning, so it is impossible. */
1377 	BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1378 	if (ret) {
1379 		btrfs_abort_transaction(trans, root, ret);
1380 		goto fail;
1381 	}
1382 
1383 	btrfs_i_size_write(parent_inode, parent_inode->i_size +
1384 					 dentry->d_name.len * 2);
1385 	parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1386 	ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1387 	if (ret) {
1388 		btrfs_abort_transaction(trans, root, ret);
1389 		goto fail;
1390 	}
1391 	ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1392 				  BTRFS_UUID_KEY_SUBVOL, objectid);
1393 	if (ret) {
1394 		btrfs_abort_transaction(trans, root, ret);
1395 		goto fail;
1396 	}
1397 	if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1398 		ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1399 					  new_root_item->received_uuid,
1400 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1401 					  objectid);
1402 		if (ret && ret != -EEXIST) {
1403 			btrfs_abort_transaction(trans, root, ret);
1404 			goto fail;
1405 		}
1406 	}
1407 fail:
1408 	pending->error = ret;
1409 dir_item_existed:
1410 	trans->block_rsv = rsv;
1411 	trans->bytes_reserved = 0;
1412 no_free_objectid:
1413 	kfree(new_root_item);
1414 root_item_alloc_fail:
1415 	btrfs_free_path(path);
1416 	return ret;
1417 }
1418 
1419 /*
1420  * create all the snapshots we've scheduled for creation
1421  */
1422 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1423 					     struct btrfs_fs_info *fs_info)
1424 {
1425 	struct btrfs_pending_snapshot *pending, *next;
1426 	struct list_head *head = &trans->transaction->pending_snapshots;
1427 	int ret = 0;
1428 
1429 	list_for_each_entry_safe(pending, next, head, list) {
1430 		list_del(&pending->list);
1431 		ret = create_pending_snapshot(trans, fs_info, pending);
1432 		if (ret)
1433 			break;
1434 	}
1435 	return ret;
1436 }
1437 
1438 static void update_super_roots(struct btrfs_root *root)
1439 {
1440 	struct btrfs_root_item *root_item;
1441 	struct btrfs_super_block *super;
1442 
1443 	super = root->fs_info->super_copy;
1444 
1445 	root_item = &root->fs_info->chunk_root->root_item;
1446 	super->chunk_root = root_item->bytenr;
1447 	super->chunk_root_generation = root_item->generation;
1448 	super->chunk_root_level = root_item->level;
1449 
1450 	root_item = &root->fs_info->tree_root->root_item;
1451 	super->root = root_item->bytenr;
1452 	super->generation = root_item->generation;
1453 	super->root_level = root_item->level;
1454 	if (btrfs_test_opt(root, SPACE_CACHE))
1455 		super->cache_generation = root_item->generation;
1456 	if (root->fs_info->update_uuid_tree_gen)
1457 		super->uuid_tree_generation = root_item->generation;
1458 }
1459 
1460 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1461 {
1462 	struct btrfs_transaction *trans;
1463 	int ret = 0;
1464 
1465 	spin_lock(&info->trans_lock);
1466 	trans = info->running_transaction;
1467 	if (trans)
1468 		ret = (trans->state >= TRANS_STATE_COMMIT_START);
1469 	spin_unlock(&info->trans_lock);
1470 	return ret;
1471 }
1472 
1473 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1474 {
1475 	struct btrfs_transaction *trans;
1476 	int ret = 0;
1477 
1478 	spin_lock(&info->trans_lock);
1479 	trans = info->running_transaction;
1480 	if (trans)
1481 		ret = is_transaction_blocked(trans);
1482 	spin_unlock(&info->trans_lock);
1483 	return ret;
1484 }
1485 
1486 /*
1487  * wait for the current transaction commit to start and block subsequent
1488  * transaction joins
1489  */
1490 static void wait_current_trans_commit_start(struct btrfs_root *root,
1491 					    struct btrfs_transaction *trans)
1492 {
1493 	wait_event(root->fs_info->transaction_blocked_wait,
1494 		   trans->state >= TRANS_STATE_COMMIT_START ||
1495 		   trans->aborted);
1496 }
1497 
1498 /*
1499  * wait for the current transaction to start and then become unblocked.
1500  * caller holds ref.
1501  */
1502 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1503 					 struct btrfs_transaction *trans)
1504 {
1505 	wait_event(root->fs_info->transaction_wait,
1506 		   trans->state >= TRANS_STATE_UNBLOCKED ||
1507 		   trans->aborted);
1508 }
1509 
1510 /*
1511  * commit transactions asynchronously. once btrfs_commit_transaction_async
1512  * returns, any subsequent transaction will not be allowed to join.
1513  */
1514 struct btrfs_async_commit {
1515 	struct btrfs_trans_handle *newtrans;
1516 	struct btrfs_root *root;
1517 	struct work_struct work;
1518 };
1519 
1520 static void do_async_commit(struct work_struct *work)
1521 {
1522 	struct btrfs_async_commit *ac =
1523 		container_of(work, struct btrfs_async_commit, work);
1524 
1525 	/*
1526 	 * We've got freeze protection passed with the transaction.
1527 	 * Tell lockdep about it.
1528 	 */
1529 	if (ac->newtrans->type & __TRANS_FREEZABLE)
1530 		rwsem_acquire_read(
1531 		     &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1532 		     0, 1, _THIS_IP_);
1533 
1534 	current->journal_info = ac->newtrans;
1535 
1536 	btrfs_commit_transaction(ac->newtrans, ac->root);
1537 	kfree(ac);
1538 }
1539 
1540 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1541 				   struct btrfs_root *root,
1542 				   int wait_for_unblock)
1543 {
1544 	struct btrfs_async_commit *ac;
1545 	struct btrfs_transaction *cur_trans;
1546 
1547 	ac = kmalloc(sizeof(*ac), GFP_NOFS);
1548 	if (!ac)
1549 		return -ENOMEM;
1550 
1551 	INIT_WORK(&ac->work, do_async_commit);
1552 	ac->root = root;
1553 	ac->newtrans = btrfs_join_transaction(root);
1554 	if (IS_ERR(ac->newtrans)) {
1555 		int err = PTR_ERR(ac->newtrans);
1556 		kfree(ac);
1557 		return err;
1558 	}
1559 
1560 	/* take transaction reference */
1561 	cur_trans = trans->transaction;
1562 	atomic_inc(&cur_trans->use_count);
1563 
1564 	btrfs_end_transaction(trans, root);
1565 
1566 	/*
1567 	 * Tell lockdep we've released the freeze rwsem, since the
1568 	 * async commit thread will be the one to unlock it.
1569 	 */
1570 	if (ac->newtrans->type & __TRANS_FREEZABLE)
1571 		rwsem_release(
1572 			&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1573 			1, _THIS_IP_);
1574 
1575 	schedule_work(&ac->work);
1576 
1577 	/* wait for transaction to start and unblock */
1578 	if (wait_for_unblock)
1579 		wait_current_trans_commit_start_and_unblock(root, cur_trans);
1580 	else
1581 		wait_current_trans_commit_start(root, cur_trans);
1582 
1583 	if (current->journal_info == trans)
1584 		current->journal_info = NULL;
1585 
1586 	btrfs_put_transaction(cur_trans);
1587 	return 0;
1588 }
1589 
1590 
1591 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1592 				struct btrfs_root *root, int err)
1593 {
1594 	struct btrfs_transaction *cur_trans = trans->transaction;
1595 	DEFINE_WAIT(wait);
1596 
1597 	WARN_ON(trans->use_count > 1);
1598 
1599 	btrfs_abort_transaction(trans, root, err);
1600 
1601 	spin_lock(&root->fs_info->trans_lock);
1602 
1603 	/*
1604 	 * If the transaction is removed from the list, it means this
1605 	 * transaction has been committed successfully, so it is impossible
1606 	 * to call the cleanup function.
1607 	 */
1608 	BUG_ON(list_empty(&cur_trans->list));
1609 
1610 	list_del_init(&cur_trans->list);
1611 	if (cur_trans == root->fs_info->running_transaction) {
1612 		cur_trans->state = TRANS_STATE_COMMIT_DOING;
1613 		spin_unlock(&root->fs_info->trans_lock);
1614 		wait_event(cur_trans->writer_wait,
1615 			   atomic_read(&cur_trans->num_writers) == 1);
1616 
1617 		spin_lock(&root->fs_info->trans_lock);
1618 	}
1619 	spin_unlock(&root->fs_info->trans_lock);
1620 
1621 	btrfs_cleanup_one_transaction(trans->transaction, root);
1622 
1623 	spin_lock(&root->fs_info->trans_lock);
1624 	if (cur_trans == root->fs_info->running_transaction)
1625 		root->fs_info->running_transaction = NULL;
1626 	spin_unlock(&root->fs_info->trans_lock);
1627 
1628 	if (trans->type & __TRANS_FREEZABLE)
1629 		sb_end_intwrite(root->fs_info->sb);
1630 	btrfs_put_transaction(cur_trans);
1631 	btrfs_put_transaction(cur_trans);
1632 
1633 	trace_btrfs_transaction_commit(root);
1634 
1635 	if (current->journal_info == trans)
1636 		current->journal_info = NULL;
1637 	btrfs_scrub_cancel(root->fs_info);
1638 
1639 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
1640 }
1641 
1642 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1643 {
1644 	if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1645 		return btrfs_start_delalloc_roots(fs_info, 1, -1);
1646 	return 0;
1647 }
1648 
1649 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1650 {
1651 	if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1652 		btrfs_wait_ordered_roots(fs_info, -1);
1653 }
1654 
1655 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1656 			     struct btrfs_root *root)
1657 {
1658 	struct btrfs_transaction *cur_trans = trans->transaction;
1659 	struct btrfs_transaction *prev_trans = NULL;
1660 	struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
1661 	int ret;
1662 
1663 	/* Stop the commit early if ->aborted is set */
1664 	if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1665 		ret = cur_trans->aborted;
1666 		btrfs_end_transaction(trans, root);
1667 		return ret;
1668 	}
1669 
1670 	/* make a pass through all the delayed refs we have so far
1671 	 * any runnings procs may add more while we are here
1672 	 */
1673 	ret = btrfs_run_delayed_refs(trans, root, 0);
1674 	if (ret) {
1675 		btrfs_end_transaction(trans, root);
1676 		return ret;
1677 	}
1678 
1679 	btrfs_trans_release_metadata(trans, root);
1680 	trans->block_rsv = NULL;
1681 	if (trans->qgroup_reserved) {
1682 		btrfs_qgroup_free(root, trans->qgroup_reserved);
1683 		trans->qgroup_reserved = 0;
1684 	}
1685 
1686 	cur_trans = trans->transaction;
1687 
1688 	/*
1689 	 * set the flushing flag so procs in this transaction have to
1690 	 * start sending their work down.
1691 	 */
1692 	cur_trans->delayed_refs.flushing = 1;
1693 	smp_wmb();
1694 
1695 	if (!list_empty(&trans->new_bgs))
1696 		btrfs_create_pending_block_groups(trans, root);
1697 
1698 	ret = btrfs_run_delayed_refs(trans, root, 0);
1699 	if (ret) {
1700 		btrfs_end_transaction(trans, root);
1701 		return ret;
1702 	}
1703 
1704 	spin_lock(&root->fs_info->trans_lock);
1705 	if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1706 		spin_unlock(&root->fs_info->trans_lock);
1707 		atomic_inc(&cur_trans->use_count);
1708 		ret = btrfs_end_transaction(trans, root);
1709 
1710 		wait_for_commit(root, cur_trans);
1711 
1712 		btrfs_put_transaction(cur_trans);
1713 
1714 		return ret;
1715 	}
1716 
1717 	cur_trans->state = TRANS_STATE_COMMIT_START;
1718 	wake_up(&root->fs_info->transaction_blocked_wait);
1719 
1720 	if (cur_trans->list.prev != &root->fs_info->trans_list) {
1721 		prev_trans = list_entry(cur_trans->list.prev,
1722 					struct btrfs_transaction, list);
1723 		if (prev_trans->state != TRANS_STATE_COMPLETED) {
1724 			atomic_inc(&prev_trans->use_count);
1725 			spin_unlock(&root->fs_info->trans_lock);
1726 
1727 			wait_for_commit(root, prev_trans);
1728 
1729 			btrfs_put_transaction(prev_trans);
1730 		} else {
1731 			spin_unlock(&root->fs_info->trans_lock);
1732 		}
1733 	} else {
1734 		spin_unlock(&root->fs_info->trans_lock);
1735 	}
1736 
1737 	extwriter_counter_dec(cur_trans, trans->type);
1738 
1739 	ret = btrfs_start_delalloc_flush(root->fs_info);
1740 	if (ret)
1741 		goto cleanup_transaction;
1742 
1743 	ret = btrfs_run_delayed_items(trans, root);
1744 	if (ret)
1745 		goto cleanup_transaction;
1746 
1747 	wait_event(cur_trans->writer_wait,
1748 		   extwriter_counter_read(cur_trans) == 0);
1749 
1750 	/* some pending stuffs might be added after the previous flush. */
1751 	ret = btrfs_run_delayed_items(trans, root);
1752 	if (ret)
1753 		goto cleanup_transaction;
1754 
1755 	btrfs_wait_delalloc_flush(root->fs_info);
1756 
1757 	btrfs_scrub_pause(root);
1758 	/*
1759 	 * Ok now we need to make sure to block out any other joins while we
1760 	 * commit the transaction.  We could have started a join before setting
1761 	 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1762 	 */
1763 	spin_lock(&root->fs_info->trans_lock);
1764 	cur_trans->state = TRANS_STATE_COMMIT_DOING;
1765 	spin_unlock(&root->fs_info->trans_lock);
1766 	wait_event(cur_trans->writer_wait,
1767 		   atomic_read(&cur_trans->num_writers) == 1);
1768 
1769 	/* ->aborted might be set after the previous check, so check it */
1770 	if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1771 		ret = cur_trans->aborted;
1772 		goto scrub_continue;
1773 	}
1774 	/*
1775 	 * the reloc mutex makes sure that we stop
1776 	 * the balancing code from coming in and moving
1777 	 * extents around in the middle of the commit
1778 	 */
1779 	mutex_lock(&root->fs_info->reloc_mutex);
1780 
1781 	/*
1782 	 * We needn't worry about the delayed items because we will
1783 	 * deal with them in create_pending_snapshot(), which is the
1784 	 * core function of the snapshot creation.
1785 	 */
1786 	ret = create_pending_snapshots(trans, root->fs_info);
1787 	if (ret) {
1788 		mutex_unlock(&root->fs_info->reloc_mutex);
1789 		goto scrub_continue;
1790 	}
1791 
1792 	/*
1793 	 * We insert the dir indexes of the snapshots and update the inode
1794 	 * of the snapshots' parents after the snapshot creation, so there
1795 	 * are some delayed items which are not dealt with. Now deal with
1796 	 * them.
1797 	 *
1798 	 * We needn't worry that this operation will corrupt the snapshots,
1799 	 * because all the tree which are snapshoted will be forced to COW
1800 	 * the nodes and leaves.
1801 	 */
1802 	ret = btrfs_run_delayed_items(trans, root);
1803 	if (ret) {
1804 		mutex_unlock(&root->fs_info->reloc_mutex);
1805 		goto scrub_continue;
1806 	}
1807 
1808 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1809 	if (ret) {
1810 		mutex_unlock(&root->fs_info->reloc_mutex);
1811 		goto scrub_continue;
1812 	}
1813 
1814 	/*
1815 	 * make sure none of the code above managed to slip in a
1816 	 * delayed item
1817 	 */
1818 	btrfs_assert_delayed_root_empty(root);
1819 
1820 	WARN_ON(cur_trans != trans->transaction);
1821 
1822 	/* btrfs_commit_tree_roots is responsible for getting the
1823 	 * various roots consistent with each other.  Every pointer
1824 	 * in the tree of tree roots has to point to the most up to date
1825 	 * root for every subvolume and other tree.  So, we have to keep
1826 	 * the tree logging code from jumping in and changing any
1827 	 * of the trees.
1828 	 *
1829 	 * At this point in the commit, there can't be any tree-log
1830 	 * writers, but a little lower down we drop the trans mutex
1831 	 * and let new people in.  By holding the tree_log_mutex
1832 	 * from now until after the super is written, we avoid races
1833 	 * with the tree-log code.
1834 	 */
1835 	mutex_lock(&root->fs_info->tree_log_mutex);
1836 
1837 	ret = commit_fs_roots(trans, root);
1838 	if (ret) {
1839 		mutex_unlock(&root->fs_info->tree_log_mutex);
1840 		mutex_unlock(&root->fs_info->reloc_mutex);
1841 		goto scrub_continue;
1842 	}
1843 
1844 	/*
1845 	 * Since the transaction is done, we should set the inode map cache flag
1846 	 * before any other comming transaction.
1847 	 */
1848 	if (btrfs_test_opt(root, CHANGE_INODE_CACHE))
1849 		btrfs_set_opt(root->fs_info->mount_opt, INODE_MAP_CACHE);
1850 	else
1851 		btrfs_clear_opt(root->fs_info->mount_opt, INODE_MAP_CACHE);
1852 
1853 	/* commit_fs_roots gets rid of all the tree log roots, it is now
1854 	 * safe to free the root of tree log roots
1855 	 */
1856 	btrfs_free_log_root_tree(trans, root->fs_info);
1857 
1858 	ret = commit_cowonly_roots(trans, root);
1859 	if (ret) {
1860 		mutex_unlock(&root->fs_info->tree_log_mutex);
1861 		mutex_unlock(&root->fs_info->reloc_mutex);
1862 		goto scrub_continue;
1863 	}
1864 
1865 	/*
1866 	 * The tasks which save the space cache and inode cache may also
1867 	 * update ->aborted, check it.
1868 	 */
1869 	if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1870 		ret = cur_trans->aborted;
1871 		mutex_unlock(&root->fs_info->tree_log_mutex);
1872 		mutex_unlock(&root->fs_info->reloc_mutex);
1873 		goto scrub_continue;
1874 	}
1875 
1876 	btrfs_prepare_extent_commit(trans, root);
1877 
1878 	cur_trans = root->fs_info->running_transaction;
1879 
1880 	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1881 			    root->fs_info->tree_root->node);
1882 	list_add_tail(&root->fs_info->tree_root->dirty_list,
1883 		      &cur_trans->switch_commits);
1884 
1885 	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1886 			    root->fs_info->chunk_root->node);
1887 	list_add_tail(&root->fs_info->chunk_root->dirty_list,
1888 		      &cur_trans->switch_commits);
1889 
1890 	switch_commit_roots(cur_trans, root->fs_info);
1891 
1892 	assert_qgroups_uptodate(trans);
1893 	update_super_roots(root);
1894 
1895 	btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1896 	btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1897 	memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1898 	       sizeof(*root->fs_info->super_copy));
1899 
1900 	btrfs_update_commit_device_size(root->fs_info);
1901 	btrfs_update_commit_device_bytes_used(root, cur_trans);
1902 
1903 	clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
1904 	clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
1905 
1906 	spin_lock(&root->fs_info->trans_lock);
1907 	cur_trans->state = TRANS_STATE_UNBLOCKED;
1908 	root->fs_info->running_transaction = NULL;
1909 	spin_unlock(&root->fs_info->trans_lock);
1910 	mutex_unlock(&root->fs_info->reloc_mutex);
1911 
1912 	wake_up(&root->fs_info->transaction_wait);
1913 
1914 	ret = btrfs_write_and_wait_transaction(trans, root);
1915 	if (ret) {
1916 		btrfs_error(root->fs_info, ret,
1917 			    "Error while writing out transaction");
1918 		mutex_unlock(&root->fs_info->tree_log_mutex);
1919 		goto scrub_continue;
1920 	}
1921 
1922 	ret = write_ctree_super(trans, root, 0);
1923 	if (ret) {
1924 		mutex_unlock(&root->fs_info->tree_log_mutex);
1925 		goto scrub_continue;
1926 	}
1927 
1928 	/*
1929 	 * the super is written, we can safely allow the tree-loggers
1930 	 * to go about their business
1931 	 */
1932 	mutex_unlock(&root->fs_info->tree_log_mutex);
1933 
1934 	btrfs_finish_extent_commit(trans, root);
1935 
1936 	root->fs_info->last_trans_committed = cur_trans->transid;
1937 	/*
1938 	 * We needn't acquire the lock here because there is no other task
1939 	 * which can change it.
1940 	 */
1941 	cur_trans->state = TRANS_STATE_COMPLETED;
1942 	wake_up(&cur_trans->commit_wait);
1943 
1944 	spin_lock(&root->fs_info->trans_lock);
1945 	list_del_init(&cur_trans->list);
1946 	spin_unlock(&root->fs_info->trans_lock);
1947 
1948 	btrfs_put_transaction(cur_trans);
1949 	btrfs_put_transaction(cur_trans);
1950 
1951 	if (trans->type & __TRANS_FREEZABLE)
1952 		sb_end_intwrite(root->fs_info->sb);
1953 
1954 	trace_btrfs_transaction_commit(root);
1955 
1956 	btrfs_scrub_continue(root);
1957 
1958 	if (current->journal_info == trans)
1959 		current->journal_info = NULL;
1960 
1961 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
1962 
1963 	if (current != root->fs_info->transaction_kthread)
1964 		btrfs_run_delayed_iputs(root);
1965 
1966 	return ret;
1967 
1968 scrub_continue:
1969 	btrfs_scrub_continue(root);
1970 cleanup_transaction:
1971 	btrfs_trans_release_metadata(trans, root);
1972 	trans->block_rsv = NULL;
1973 	if (trans->qgroup_reserved) {
1974 		btrfs_qgroup_free(root, trans->qgroup_reserved);
1975 		trans->qgroup_reserved = 0;
1976 	}
1977 	btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
1978 	if (current->journal_info == trans)
1979 		current->journal_info = NULL;
1980 	cleanup_transaction(trans, root, ret);
1981 
1982 	return ret;
1983 }
1984 
1985 /*
1986  * return < 0 if error
1987  * 0 if there are no more dead_roots at the time of call
1988  * 1 there are more to be processed, call me again
1989  *
1990  * The return value indicates there are certainly more snapshots to delete, but
1991  * if there comes a new one during processing, it may return 0. We don't mind,
1992  * because btrfs_commit_super will poke cleaner thread and it will process it a
1993  * few seconds later.
1994  */
1995 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
1996 {
1997 	int ret;
1998 	struct btrfs_fs_info *fs_info = root->fs_info;
1999 
2000 	spin_lock(&fs_info->trans_lock);
2001 	if (list_empty(&fs_info->dead_roots)) {
2002 		spin_unlock(&fs_info->trans_lock);
2003 		return 0;
2004 	}
2005 	root = list_first_entry(&fs_info->dead_roots,
2006 			struct btrfs_root, root_list);
2007 	list_del_init(&root->root_list);
2008 	spin_unlock(&fs_info->trans_lock);
2009 
2010 	pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2011 
2012 	btrfs_kill_all_delayed_nodes(root);
2013 
2014 	if (btrfs_header_backref_rev(root->node) <
2015 			BTRFS_MIXED_BACKREF_REV)
2016 		ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2017 	else
2018 		ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2019 
2020 	return (ret < 0) ? 0 : 1;
2021 }
2022