xref: /openbmc/linux/fs/btrfs/transaction.c (revision 354aa0fb)
16cbd5570SChris Mason /*
26cbd5570SChris Mason  * Copyright (C) 2007 Oracle.  All rights reserved.
36cbd5570SChris Mason  *
46cbd5570SChris Mason  * This program is free software; you can redistribute it and/or
56cbd5570SChris Mason  * modify it under the terms of the GNU General Public
66cbd5570SChris Mason  * License v2 as published by the Free Software Foundation.
76cbd5570SChris Mason  *
86cbd5570SChris Mason  * This program is distributed in the hope that it will be useful,
96cbd5570SChris Mason  * but WITHOUT ANY WARRANTY; without even the implied warranty of
106cbd5570SChris Mason  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
116cbd5570SChris Mason  * General Public License for more details.
126cbd5570SChris Mason  *
136cbd5570SChris Mason  * You should have received a copy of the GNU General Public
146cbd5570SChris Mason  * License along with this program; if not, write to the
156cbd5570SChris Mason  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
166cbd5570SChris Mason  * Boston, MA 021110-1307, USA.
176cbd5570SChris Mason  */
186cbd5570SChris Mason 
1979154b1bSChris Mason #include <linux/fs.h>
205a0e3ad6STejun Heo #include <linux/slab.h>
2134088780SChris Mason #include <linux/sched.h>
22d3c2fdcfSChris Mason #include <linux/writeback.h>
235f39d397SChris Mason #include <linux/pagemap.h>
245f2cc086SChris Mason #include <linux/blkdev.h>
258ea05e3aSAlexander Block #include <linux/uuid.h>
2679154b1bSChris Mason #include "ctree.h"
2779154b1bSChris Mason #include "disk-io.h"
2879154b1bSChris Mason #include "transaction.h"
29925baeddSChris Mason #include "locking.h"
30e02119d5SChris Mason #include "tree-log.h"
31581bb050SLi Zefan #include "inode-map.h"
32733f4fbbSStefan Behrens #include "volumes.h"
3379154b1bSChris Mason 
340f7d52f4SChris Mason #define BTRFS_ROOT_TRANS_TAG 0
350f7d52f4SChris Mason 
3649b25e05SJeff Mahoney void put_transaction(struct btrfs_transaction *transaction)
3779154b1bSChris Mason {
3813c5a93eSJosef Bacik 	WARN_ON(atomic_read(&transaction->use_count) == 0);
3913c5a93eSJosef Bacik 	if (atomic_dec_and_test(&transaction->use_count)) {
40a4abeea4SJosef Bacik 		BUG_ON(!list_empty(&transaction->list));
4100f04b88SArne Jansen 		WARN_ON(transaction->delayed_refs.root.rb_node);
422c90e5d6SChris Mason 		memset(transaction, 0, sizeof(*transaction));
432c90e5d6SChris Mason 		kmem_cache_free(btrfs_transaction_cachep, transaction);
4479154b1bSChris Mason 	}
4578fae27eSChris Mason }
4679154b1bSChris Mason 
47817d52f8SJosef Bacik static noinline void switch_commit_root(struct btrfs_root *root)
48817d52f8SJosef Bacik {
49817d52f8SJosef Bacik 	free_extent_buffer(root->commit_root);
50817d52f8SJosef Bacik 	root->commit_root = btrfs_root_node(root);
51817d52f8SJosef Bacik }
52817d52f8SJosef Bacik 
53d352ac68SChris Mason /*
54d352ac68SChris Mason  * either allocate a new transaction or hop into the existing one
55d352ac68SChris Mason  */
56354aa0fbSMiao Xie static noinline int join_transaction(struct btrfs_root *root, int type)
5779154b1bSChris Mason {
5879154b1bSChris Mason 	struct btrfs_transaction *cur_trans;
5919ae4e81SJan Schmidt 	struct btrfs_fs_info *fs_info = root->fs_info;
60a4abeea4SJosef Bacik 
6119ae4e81SJan Schmidt 	spin_lock(&fs_info->trans_lock);
62d43317dcSChris Mason loop:
6349b25e05SJeff Mahoney 	/* The file system has been taken offline. No new transactions. */
6419ae4e81SJan Schmidt 	if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
6519ae4e81SJan Schmidt 		spin_unlock(&fs_info->trans_lock);
6649b25e05SJeff Mahoney 		return -EROFS;
6749b25e05SJeff Mahoney 	}
6849b25e05SJeff Mahoney 
6919ae4e81SJan Schmidt 	if (fs_info->trans_no_join) {
70354aa0fbSMiao Xie 		/*
71354aa0fbSMiao Xie 		 * If we are JOIN_NOLOCK we're already committing a current
72354aa0fbSMiao Xie 		 * transaction, we just need a handle to deal with something
73354aa0fbSMiao Xie 		 * when committing the transaction, such as inode cache and
74354aa0fbSMiao Xie 		 * space cache. It is a special case.
75354aa0fbSMiao Xie 		 */
76354aa0fbSMiao Xie 		if (type != TRANS_JOIN_NOLOCK) {
7719ae4e81SJan Schmidt 			spin_unlock(&fs_info->trans_lock);
78a4abeea4SJosef Bacik 			return -EBUSY;
79a4abeea4SJosef Bacik 		}
80a4abeea4SJosef Bacik 	}
81a4abeea4SJosef Bacik 
8219ae4e81SJan Schmidt 	cur_trans = fs_info->running_transaction;
83a4abeea4SJosef Bacik 	if (cur_trans) {
84871383beSDavid Sterba 		if (cur_trans->aborted) {
8519ae4e81SJan Schmidt 			spin_unlock(&fs_info->trans_lock);
8649b25e05SJeff Mahoney 			return cur_trans->aborted;
87871383beSDavid Sterba 		}
88a4abeea4SJosef Bacik 		atomic_inc(&cur_trans->use_count);
89a4abeea4SJosef Bacik 		atomic_inc(&cur_trans->num_writers);
90a4abeea4SJosef Bacik 		cur_trans->num_joined++;
9119ae4e81SJan Schmidt 		spin_unlock(&fs_info->trans_lock);
92a4abeea4SJosef Bacik 		return 0;
93a4abeea4SJosef Bacik 	}
9419ae4e81SJan Schmidt 	spin_unlock(&fs_info->trans_lock);
95a4abeea4SJosef Bacik 
96354aa0fbSMiao Xie 	/*
97354aa0fbSMiao Xie 	 * If we are ATTACH, we just want to catch the current transaction,
98354aa0fbSMiao Xie 	 * and commit it. If there is no transaction, just return ENOENT.
99354aa0fbSMiao Xie 	 */
100354aa0fbSMiao Xie 	if (type == TRANS_ATTACH)
101354aa0fbSMiao Xie 		return -ENOENT;
102354aa0fbSMiao Xie 
103a4abeea4SJosef Bacik 	cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
104db5b493aSTsutomu Itoh 	if (!cur_trans)
105db5b493aSTsutomu Itoh 		return -ENOMEM;
106d43317dcSChris Mason 
10719ae4e81SJan Schmidt 	spin_lock(&fs_info->trans_lock);
10819ae4e81SJan Schmidt 	if (fs_info->running_transaction) {
109d43317dcSChris Mason 		/*
110d43317dcSChris Mason 		 * someone started a transaction after we unlocked.  Make sure
111d43317dcSChris Mason 		 * to redo the trans_no_join checks above
112d43317dcSChris Mason 		 */
113a4abeea4SJosef Bacik 		kmem_cache_free(btrfs_transaction_cachep, cur_trans);
11419ae4e81SJan Schmidt 		cur_trans = fs_info->running_transaction;
115d43317dcSChris Mason 		goto loop;
116e4b50e14SDan Carpenter 	} else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
117e4b50e14SDan Carpenter 		spin_unlock(&fs_info->trans_lock);
1187b8b92afSJosef Bacik 		kmem_cache_free(btrfs_transaction_cachep, cur_trans);
1197b8b92afSJosef Bacik 		return -EROFS;
120a4abeea4SJosef Bacik 	}
121d43317dcSChris Mason 
12213c5a93eSJosef Bacik 	atomic_set(&cur_trans->num_writers, 1);
12315ee9bc7SJosef Bacik 	cur_trans->num_joined = 0;
12479154b1bSChris Mason 	init_waitqueue_head(&cur_trans->writer_wait);
12579154b1bSChris Mason 	init_waitqueue_head(&cur_trans->commit_wait);
12679154b1bSChris Mason 	cur_trans->in_commit = 0;
127f9295749SChris Mason 	cur_trans->blocked = 0;
128a4abeea4SJosef Bacik 	/*
129a4abeea4SJosef Bacik 	 * One for this trans handle, one so it will live on until we
130a4abeea4SJosef Bacik 	 * commit the transaction.
131a4abeea4SJosef Bacik 	 */
132a4abeea4SJosef Bacik 	atomic_set(&cur_trans->use_count, 2);
13379154b1bSChris Mason 	cur_trans->commit_done = 0;
13408607c1bSChris Mason 	cur_trans->start_time = get_seconds();
13556bec294SChris Mason 
1366bef4d31SEric Paris 	cur_trans->delayed_refs.root = RB_ROOT;
13756bec294SChris Mason 	cur_trans->delayed_refs.num_entries = 0;
138c3e69d58SChris Mason 	cur_trans->delayed_refs.num_heads_ready = 0;
139c3e69d58SChris Mason 	cur_trans->delayed_refs.num_heads = 0;
14056bec294SChris Mason 	cur_trans->delayed_refs.flushing = 0;
141c3e69d58SChris Mason 	cur_trans->delayed_refs.run_delayed_start = 0;
14220b297d6SJan Schmidt 
14320b297d6SJan Schmidt 	/*
14420b297d6SJan Schmidt 	 * although the tree mod log is per file system and not per transaction,
14520b297d6SJan Schmidt 	 * the log must never go across transaction boundaries.
14620b297d6SJan Schmidt 	 */
14720b297d6SJan Schmidt 	smp_mb();
14820b297d6SJan Schmidt 	if (!list_empty(&fs_info->tree_mod_seq_list)) {
14920b297d6SJan Schmidt 		printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when "
15020b297d6SJan Schmidt 			"creating a fresh transaction\n");
15120b297d6SJan Schmidt 		WARN_ON(1);
15220b297d6SJan Schmidt 	}
15320b297d6SJan Schmidt 	if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) {
15420b297d6SJan Schmidt 		printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
15520b297d6SJan Schmidt 			"creating a fresh transaction\n");
15620b297d6SJan Schmidt 		WARN_ON(1);
15720b297d6SJan Schmidt 	}
15820b297d6SJan Schmidt 	atomic_set(&fs_info->tree_mod_seq, 0);
15920b297d6SJan Schmidt 
160a4abeea4SJosef Bacik 	spin_lock_init(&cur_trans->commit_lock);
16156bec294SChris Mason 	spin_lock_init(&cur_trans->delayed_refs.lock);
16256bec294SChris Mason 
1633063d29fSChris Mason 	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
16419ae4e81SJan Schmidt 	list_add_tail(&cur_trans->list, &fs_info->trans_list);
165d1310b2eSChris Mason 	extent_io_tree_init(&cur_trans->dirty_pages,
16619ae4e81SJan Schmidt 			     fs_info->btree_inode->i_mapping);
16719ae4e81SJan Schmidt 	fs_info->generation++;
16819ae4e81SJan Schmidt 	cur_trans->transid = fs_info->generation;
16919ae4e81SJan Schmidt 	fs_info->running_transaction = cur_trans;
17049b25e05SJeff Mahoney 	cur_trans->aborted = 0;
17119ae4e81SJan Schmidt 	spin_unlock(&fs_info->trans_lock);
17215ee9bc7SJosef Bacik 
17379154b1bSChris Mason 	return 0;
17479154b1bSChris Mason }
17579154b1bSChris Mason 
176d352ac68SChris Mason /*
177d397712bSChris Mason  * this does all the record keeping required to make sure that a reference
178d397712bSChris Mason  * counted root is properly recorded in a given transaction.  This is required
179d397712bSChris Mason  * to make sure the old root from before we joined the transaction is deleted
180d397712bSChris Mason  * when the transaction commits
181d352ac68SChris Mason  */
1827585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans,
1835d4f98a2SYan Zheng 			       struct btrfs_root *root)
1846702ed49SChris Mason {
1855d4f98a2SYan Zheng 	if (root->ref_cows && root->last_trans < trans->transid) {
1866702ed49SChris Mason 		WARN_ON(root == root->fs_info->extent_root);
1875d4f98a2SYan Zheng 		WARN_ON(root->commit_root != root->node);
1885d4f98a2SYan Zheng 
1897585717fSChris Mason 		/*
1907585717fSChris Mason 		 * see below for in_trans_setup usage rules
1917585717fSChris Mason 		 * we have the reloc mutex held now, so there
1927585717fSChris Mason 		 * is only one writer in this function
1937585717fSChris Mason 		 */
1947585717fSChris Mason 		root->in_trans_setup = 1;
1957585717fSChris Mason 
1967585717fSChris Mason 		/* make sure readers find in_trans_setup before
1977585717fSChris Mason 		 * they find our root->last_trans update
1987585717fSChris Mason 		 */
1997585717fSChris Mason 		smp_wmb();
2007585717fSChris Mason 
201a4abeea4SJosef Bacik 		spin_lock(&root->fs_info->fs_roots_radix_lock);
202a4abeea4SJosef Bacik 		if (root->last_trans == trans->transid) {
203a4abeea4SJosef Bacik 			spin_unlock(&root->fs_info->fs_roots_radix_lock);
204a4abeea4SJosef Bacik 			return 0;
205a4abeea4SJosef Bacik 		}
2066702ed49SChris Mason 		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
2076702ed49SChris Mason 			   (unsigned long)root->root_key.objectid,
2086702ed49SChris Mason 			   BTRFS_ROOT_TRANS_TAG);
209a4abeea4SJosef Bacik 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
2107585717fSChris Mason 		root->last_trans = trans->transid;
2117585717fSChris Mason 
2127585717fSChris Mason 		/* this is pretty tricky.  We don't want to
2137585717fSChris Mason 		 * take the relocation lock in btrfs_record_root_in_trans
2147585717fSChris Mason 		 * unless we're really doing the first setup for this root in
2157585717fSChris Mason 		 * this transaction.
2167585717fSChris Mason 		 *
2177585717fSChris Mason 		 * Normally we'd use root->last_trans as a flag to decide
2187585717fSChris Mason 		 * if we want to take the expensive mutex.
2197585717fSChris Mason 		 *
2207585717fSChris Mason 		 * But, we have to set root->last_trans before we
2217585717fSChris Mason 		 * init the relocation root, otherwise, we trip over warnings
2227585717fSChris Mason 		 * in ctree.c.  The solution used here is to flag ourselves
2237585717fSChris Mason 		 * with root->in_trans_setup.  When this is 1, we're still
2247585717fSChris Mason 		 * fixing up the reloc trees and everyone must wait.
2257585717fSChris Mason 		 *
2267585717fSChris Mason 		 * When this is zero, they can trust root->last_trans and fly
2277585717fSChris Mason 		 * through btrfs_record_root_in_trans without having to take the
2287585717fSChris Mason 		 * lock.  smp_wmb() makes sure that all the writes above are
2297585717fSChris Mason 		 * done before we pop in the zero below
2307585717fSChris Mason 		 */
2315d4f98a2SYan Zheng 		btrfs_init_reloc_root(trans, root);
2327585717fSChris Mason 		smp_wmb();
2337585717fSChris Mason 		root->in_trans_setup = 0;
2346702ed49SChris Mason 	}
2355d4f98a2SYan Zheng 	return 0;
2366702ed49SChris Mason }
2375d4f98a2SYan Zheng 
2387585717fSChris Mason 
2397585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
2407585717fSChris Mason 			       struct btrfs_root *root)
2417585717fSChris Mason {
2427585717fSChris Mason 	if (!root->ref_cows)
2437585717fSChris Mason 		return 0;
2447585717fSChris Mason 
2457585717fSChris Mason 	/*
2467585717fSChris Mason 	 * see record_root_in_trans for comments about in_trans_setup usage
2477585717fSChris Mason 	 * and barriers
2487585717fSChris Mason 	 */
2497585717fSChris Mason 	smp_rmb();
2507585717fSChris Mason 	if (root->last_trans == trans->transid &&
2517585717fSChris Mason 	    !root->in_trans_setup)
2527585717fSChris Mason 		return 0;
2537585717fSChris Mason 
2547585717fSChris Mason 	mutex_lock(&root->fs_info->reloc_mutex);
2557585717fSChris Mason 	record_root_in_trans(trans, root);
2567585717fSChris Mason 	mutex_unlock(&root->fs_info->reloc_mutex);
2577585717fSChris Mason 
2587585717fSChris Mason 	return 0;
2597585717fSChris Mason }
2607585717fSChris Mason 
261d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked
262d352ac68SChris Mason  * when this is done, it is safe to start a new transaction, but the current
263d352ac68SChris Mason  * transaction might not be fully on disk.
264d352ac68SChris Mason  */
26537d1aeeeSChris Mason static void wait_current_trans(struct btrfs_root *root)
26679154b1bSChris Mason {
267f9295749SChris Mason 	struct btrfs_transaction *cur_trans;
26879154b1bSChris Mason 
269a4abeea4SJosef Bacik 	spin_lock(&root->fs_info->trans_lock);
270f9295749SChris Mason 	cur_trans = root->fs_info->running_transaction;
27137d1aeeeSChris Mason 	if (cur_trans && cur_trans->blocked) {
27213c5a93eSJosef Bacik 		atomic_inc(&cur_trans->use_count);
273a4abeea4SJosef Bacik 		spin_unlock(&root->fs_info->trans_lock);
27472d63ed6SLi Zefan 
27572d63ed6SLi Zefan 		wait_event(root->fs_info->transaction_wait,
27672d63ed6SLi Zefan 			   !cur_trans->blocked);
277f9295749SChris Mason 		put_transaction(cur_trans);
278a4abeea4SJosef Bacik 	} else {
279a4abeea4SJosef Bacik 		spin_unlock(&root->fs_info->trans_lock);
280f9295749SChris Mason 	}
28137d1aeeeSChris Mason }
28237d1aeeeSChris Mason 
283a22285a6SYan, Zheng static int may_wait_transaction(struct btrfs_root *root, int type)
28437d1aeeeSChris Mason {
285a4abeea4SJosef Bacik 	if (root->fs_info->log_root_recovering)
286a4abeea4SJosef Bacik 		return 0;
287a4abeea4SJosef Bacik 
288a4abeea4SJosef Bacik 	if (type == TRANS_USERSPACE)
289a22285a6SYan, Zheng 		return 1;
290a4abeea4SJosef Bacik 
291a4abeea4SJosef Bacik 	if (type == TRANS_START &&
292a4abeea4SJosef Bacik 	    !atomic_read(&root->fs_info->open_ioctl_trans))
293a4abeea4SJosef Bacik 		return 1;
294a4abeea4SJosef Bacik 
295a22285a6SYan, Zheng 	return 0;
296a22285a6SYan, Zheng }
297a22285a6SYan, Zheng 
298a22285a6SYan, Zheng static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
2998407aa46SMiao Xie 						    u64 num_items, int type,
3008407aa46SMiao Xie 						    int noflush)
301a22285a6SYan, Zheng {
302a22285a6SYan, Zheng 	struct btrfs_trans_handle *h;
303a22285a6SYan, Zheng 	struct btrfs_transaction *cur_trans;
304b5009945SJosef Bacik 	u64 num_bytes = 0;
305a22285a6SYan, Zheng 	int ret;
306c5567237SArne Jansen 	u64 qgroup_reserved = 0;
307acce952bSliubo 
308acce952bSliubo 	if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
309acce952bSliubo 		return ERR_PTR(-EROFS);
3102a1eb461SJosef Bacik 
3112a1eb461SJosef Bacik 	if (current->journal_info) {
3122a1eb461SJosef Bacik 		WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
3132a1eb461SJosef Bacik 		h = current->journal_info;
3142a1eb461SJosef Bacik 		h->use_count++;
3152a1eb461SJosef Bacik 		h->orig_rsv = h->block_rsv;
3162a1eb461SJosef Bacik 		h->block_rsv = NULL;
3172a1eb461SJosef Bacik 		goto got_it;
3182a1eb461SJosef Bacik 	}
319b5009945SJosef Bacik 
320b5009945SJosef Bacik 	/*
321b5009945SJosef Bacik 	 * Do the reservation before we join the transaction so we can do all
322b5009945SJosef Bacik 	 * the appropriate flushing if need be.
323b5009945SJosef Bacik 	 */
324b5009945SJosef Bacik 	if (num_items > 0 && root != root->fs_info->chunk_root) {
325c5567237SArne Jansen 		if (root->fs_info->quota_enabled &&
326c5567237SArne Jansen 		    is_fstree(root->root_key.objectid)) {
327c5567237SArne Jansen 			qgroup_reserved = num_items * root->leafsize;
328c5567237SArne Jansen 			ret = btrfs_qgroup_reserve(root, qgroup_reserved);
329c5567237SArne Jansen 			if (ret)
330c5567237SArne Jansen 				return ERR_PTR(ret);
331c5567237SArne Jansen 		}
332c5567237SArne Jansen 
333b5009945SJosef Bacik 		num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
3348407aa46SMiao Xie 		if (noflush)
3358407aa46SMiao Xie 			ret = btrfs_block_rsv_add_noflush(root,
3368407aa46SMiao Xie 						&root->fs_info->trans_block_rsv,
3378407aa46SMiao Xie 						num_bytes);
3388407aa46SMiao Xie 		else
3394a92b1b8SJosef Bacik 			ret = btrfs_block_rsv_add(root,
340b5009945SJosef Bacik 						&root->fs_info->trans_block_rsv,
341b5009945SJosef Bacik 						num_bytes);
342b5009945SJosef Bacik 		if (ret)
343b5009945SJosef Bacik 			return ERR_PTR(ret);
344b5009945SJosef Bacik 	}
345a22285a6SYan, Zheng again:
346a22285a6SYan, Zheng 	h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
347a22285a6SYan, Zheng 	if (!h)
348a22285a6SYan, Zheng 		return ERR_PTR(-ENOMEM);
349a22285a6SYan, Zheng 
35098114659SJosef Bacik 	/*
35198114659SJosef Bacik 	 * If we are JOIN_NOLOCK we're already committing a transaction and
35298114659SJosef Bacik 	 * waiting on this guy, so we don't need to do the sb_start_intwrite
35398114659SJosef Bacik 	 * because we're already holding a ref.  We need this because we could
35498114659SJosef Bacik 	 * have raced in and did an fsync() on a file which can kick a commit
35598114659SJosef Bacik 	 * and then we deadlock with somebody doing a freeze.
356354aa0fbSMiao Xie 	 *
357354aa0fbSMiao Xie 	 * If we are ATTACH, it means we just want to catch the current
358354aa0fbSMiao Xie 	 * transaction and commit it, so we needn't do sb_start_intwrite().
35998114659SJosef Bacik 	 */
360354aa0fbSMiao Xie 	if (type < TRANS_JOIN_NOLOCK)
361b2b5ef5cSJan Kara 		sb_start_intwrite(root->fs_info->sb);
362b2b5ef5cSJan Kara 
363a22285a6SYan, Zheng 	if (may_wait_transaction(root, type))
36437d1aeeeSChris Mason 		wait_current_trans(root);
365a22285a6SYan, Zheng 
366a4abeea4SJosef Bacik 	do {
367354aa0fbSMiao Xie 		ret = join_transaction(root, type);
368a4abeea4SJosef Bacik 		if (ret == -EBUSY)
369a4abeea4SJosef Bacik 			wait_current_trans(root);
370a4abeea4SJosef Bacik 	} while (ret == -EBUSY);
371a4abeea4SJosef Bacik 
372db5b493aSTsutomu Itoh 	if (ret < 0) {
373354aa0fbSMiao Xie 		/* We must get the transaction if we are JOIN_NOLOCK. */
374354aa0fbSMiao Xie 		BUG_ON(type == TRANS_JOIN_NOLOCK);
375354aa0fbSMiao Xie 
376354aa0fbSMiao Xie 		if (type < TRANS_JOIN_NOLOCK)
377b2b5ef5cSJan Kara 			sb_end_intwrite(root->fs_info->sb);
3786e8df2aeSYoshinori Sano 		kmem_cache_free(btrfs_trans_handle_cachep, h);
379db5b493aSTsutomu Itoh 		return ERR_PTR(ret);
380db5b493aSTsutomu Itoh 	}
3810f7d52f4SChris Mason 
382a22285a6SYan, Zheng 	cur_trans = root->fs_info->running_transaction;
383a22285a6SYan, Zheng 
384a22285a6SYan, Zheng 	h->transid = cur_trans->transid;
385a22285a6SYan, Zheng 	h->transaction = cur_trans;
38679154b1bSChris Mason 	h->blocks_used = 0;
387a22285a6SYan, Zheng 	h->bytes_reserved = 0;
388d13603efSArne Jansen 	h->root = root;
38956bec294SChris Mason 	h->delayed_ref_updates = 0;
3902a1eb461SJosef Bacik 	h->use_count = 1;
3910e721106SJosef Bacik 	h->adding_csums = 0;
392f0486c68SYan, Zheng 	h->block_rsv = NULL;
3932a1eb461SJosef Bacik 	h->orig_rsv = NULL;
39449b25e05SJeff Mahoney 	h->aborted = 0;
395c5567237SArne Jansen 	h->qgroup_reserved = qgroup_reserved;
396bed92eaeSArne Jansen 	h->delayed_ref_elem.seq = 0;
397a698d075SMiao Xie 	h->type = type;
398bed92eaeSArne Jansen 	INIT_LIST_HEAD(&h->qgroup_ref_list);
399ea658badSJosef Bacik 	INIT_LIST_HEAD(&h->new_bgs);
400b7ec40d7SChris Mason 
401a22285a6SYan, Zheng 	smp_mb();
402a22285a6SYan, Zheng 	if (cur_trans->blocked && may_wait_transaction(root, type)) {
403a22285a6SYan, Zheng 		btrfs_commit_transaction(h, root);
404a22285a6SYan, Zheng 		goto again;
405a22285a6SYan, Zheng 	}
4069ed74f2dSJosef Bacik 
407b5009945SJosef Bacik 	if (num_bytes) {
4088c2a3ca2SJosef Bacik 		trace_btrfs_space_reservation(root->fs_info, "transaction",
4092bcc0328SLiu Bo 					      h->transid, num_bytes, 1);
410b5009945SJosef Bacik 		h->block_rsv = &root->fs_info->trans_block_rsv;
411b5009945SJosef Bacik 		h->bytes_reserved = num_bytes;
412a22285a6SYan, Zheng 	}
413a22285a6SYan, Zheng 
4142a1eb461SJosef Bacik got_it:
415a4abeea4SJosef Bacik 	btrfs_record_root_in_trans(h, root);
416a22285a6SYan, Zheng 
417a22285a6SYan, Zheng 	if (!current->journal_info && type != TRANS_USERSPACE)
418a22285a6SYan, Zheng 		current->journal_info = h;
41979154b1bSChris Mason 	return h;
42079154b1bSChris Mason }
42179154b1bSChris Mason 
422f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
423a22285a6SYan, Zheng 						   int num_items)
424f9295749SChris Mason {
4258407aa46SMiao Xie 	return start_transaction(root, num_items, TRANS_START, 0);
426f9295749SChris Mason }
4278407aa46SMiao Xie 
4288407aa46SMiao Xie struct btrfs_trans_handle *btrfs_start_transaction_noflush(
4298407aa46SMiao Xie 					struct btrfs_root *root, int num_items)
4308407aa46SMiao Xie {
4318407aa46SMiao Xie 	return start_transaction(root, num_items, TRANS_START, 1);
4328407aa46SMiao Xie }
4338407aa46SMiao Xie 
4347a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
435f9295749SChris Mason {
4368407aa46SMiao Xie 	return start_transaction(root, 0, TRANS_JOIN, 0);
437f9295749SChris Mason }
438f9295749SChris Mason 
4397a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
4400af3d00bSJosef Bacik {
4418407aa46SMiao Xie 	return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
4420af3d00bSJosef Bacik }
4430af3d00bSJosef Bacik 
4447a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
4459ca9ee09SSage Weil {
4468407aa46SMiao Xie 	return start_transaction(root, 0, TRANS_USERSPACE, 0);
4479ca9ee09SSage Weil }
4489ca9ee09SSage Weil 
449354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
45060376ce4SJosef Bacik {
451354aa0fbSMiao Xie 	return start_transaction(root, 0, TRANS_ATTACH, 0);
45260376ce4SJosef Bacik }
45360376ce4SJosef Bacik 
454d352ac68SChris Mason /* wait for a transaction commit to be fully complete */
455b9c8300cSLi Zefan static noinline void wait_for_commit(struct btrfs_root *root,
45689ce8a63SChris Mason 				    struct btrfs_transaction *commit)
45789ce8a63SChris Mason {
45872d63ed6SLi Zefan 	wait_event(commit->commit_wait, commit->commit_done);
45989ce8a63SChris Mason }
46089ce8a63SChris Mason 
46146204592SSage Weil int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
46246204592SSage Weil {
46346204592SSage Weil 	struct btrfs_transaction *cur_trans = NULL, *t;
46446204592SSage Weil 	int ret;
46546204592SSage Weil 
46646204592SSage Weil 	ret = 0;
46746204592SSage Weil 	if (transid) {
46846204592SSage Weil 		if (transid <= root->fs_info->last_trans_committed)
469a4abeea4SJosef Bacik 			goto out;
47046204592SSage Weil 
47146204592SSage Weil 		/* find specified transaction */
472a4abeea4SJosef Bacik 		spin_lock(&root->fs_info->trans_lock);
47346204592SSage Weil 		list_for_each_entry(t, &root->fs_info->trans_list, list) {
47446204592SSage Weil 			if (t->transid == transid) {
47546204592SSage Weil 				cur_trans = t;
476a4abeea4SJosef Bacik 				atomic_inc(&cur_trans->use_count);
47746204592SSage Weil 				break;
47846204592SSage Weil 			}
47946204592SSage Weil 			if (t->transid > transid)
48046204592SSage Weil 				break;
48146204592SSage Weil 		}
482a4abeea4SJosef Bacik 		spin_unlock(&root->fs_info->trans_lock);
48346204592SSage Weil 		ret = -EINVAL;
48446204592SSage Weil 		if (!cur_trans)
485a4abeea4SJosef Bacik 			goto out;  /* bad transid */
48646204592SSage Weil 	} else {
48746204592SSage Weil 		/* find newest transaction that is committing | committed */
488a4abeea4SJosef Bacik 		spin_lock(&root->fs_info->trans_lock);
48946204592SSage Weil 		list_for_each_entry_reverse(t, &root->fs_info->trans_list,
49046204592SSage Weil 					    list) {
49146204592SSage Weil 			if (t->in_commit) {
49246204592SSage Weil 				if (t->commit_done)
4933473f3c0SJosef Bacik 					break;
49446204592SSage Weil 				cur_trans = t;
495a4abeea4SJosef Bacik 				atomic_inc(&cur_trans->use_count);
49646204592SSage Weil 				break;
49746204592SSage Weil 			}
49846204592SSage Weil 		}
499a4abeea4SJosef Bacik 		spin_unlock(&root->fs_info->trans_lock);
50046204592SSage Weil 		if (!cur_trans)
501a4abeea4SJosef Bacik 			goto out;  /* nothing committing|committed */
50246204592SSage Weil 	}
50346204592SSage Weil 
50446204592SSage Weil 	wait_for_commit(root, cur_trans);
50546204592SSage Weil 
50646204592SSage Weil 	put_transaction(cur_trans);
50746204592SSage Weil 	ret = 0;
508a4abeea4SJosef Bacik out:
50946204592SSage Weil 	return ret;
51046204592SSage Weil }
51146204592SSage Weil 
51237d1aeeeSChris Mason void btrfs_throttle(struct btrfs_root *root)
51337d1aeeeSChris Mason {
514a4abeea4SJosef Bacik 	if (!atomic_read(&root->fs_info->open_ioctl_trans))
51537d1aeeeSChris Mason 		wait_current_trans(root);
51637d1aeeeSChris Mason }
51737d1aeeeSChris Mason 
5188929ecfaSYan, Zheng static int should_end_transaction(struct btrfs_trans_handle *trans,
5198929ecfaSYan, Zheng 				  struct btrfs_root *root)
5208929ecfaSYan, Zheng {
5218929ecfaSYan, Zheng 	int ret;
52236ba022aSJosef Bacik 
52336ba022aSJosef Bacik 	ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
5248929ecfaSYan, Zheng 	return ret ? 1 : 0;
5258929ecfaSYan, Zheng }
5268929ecfaSYan, Zheng 
5278929ecfaSYan, Zheng int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
5288929ecfaSYan, Zheng 				 struct btrfs_root *root)
5298929ecfaSYan, Zheng {
5308929ecfaSYan, Zheng 	struct btrfs_transaction *cur_trans = trans->transaction;
5318929ecfaSYan, Zheng 	int updates;
53249b25e05SJeff Mahoney 	int err;
5338929ecfaSYan, Zheng 
534a4abeea4SJosef Bacik 	smp_mb();
5358929ecfaSYan, Zheng 	if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
5368929ecfaSYan, Zheng 		return 1;
5378929ecfaSYan, Zheng 
5388929ecfaSYan, Zheng 	updates = trans->delayed_ref_updates;
5398929ecfaSYan, Zheng 	trans->delayed_ref_updates = 0;
54049b25e05SJeff Mahoney 	if (updates) {
54149b25e05SJeff Mahoney 		err = btrfs_run_delayed_refs(trans, root, updates);
54249b25e05SJeff Mahoney 		if (err) /* Error code will also eval true */
54349b25e05SJeff Mahoney 			return err;
54449b25e05SJeff Mahoney 	}
5458929ecfaSYan, Zheng 
5468929ecfaSYan, Zheng 	return should_end_transaction(trans, root);
5478929ecfaSYan, Zheng }
5488929ecfaSYan, Zheng 
54989ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
550a698d075SMiao Xie 			  struct btrfs_root *root, int throttle)
55179154b1bSChris Mason {
5528929ecfaSYan, Zheng 	struct btrfs_transaction *cur_trans = trans->transaction;
553ab78c84dSChris Mason 	struct btrfs_fs_info *info = root->fs_info;
554c3e69d58SChris Mason 	int count = 0;
555a698d075SMiao Xie 	int lock = (trans->type != TRANS_JOIN_NOLOCK);
5564edc2ca3SDave Jones 	int err = 0;
557d6e4a428SChris Mason 
5582a1eb461SJosef Bacik 	if (--trans->use_count) {
5592a1eb461SJosef Bacik 		trans->block_rsv = trans->orig_rsv;
5602a1eb461SJosef Bacik 		return 0;
5612a1eb461SJosef Bacik 	}
5622a1eb461SJosef Bacik 
563edf39272SJan Schmidt 	/*
564edf39272SJan Schmidt 	 * do the qgroup accounting as early as possible
565edf39272SJan Schmidt 	 */
566edf39272SJan Schmidt 	err = btrfs_delayed_refs_qgroup_accounting(trans, info);
567edf39272SJan Schmidt 
568b24e03dbSJosef Bacik 	btrfs_trans_release_metadata(trans, root);
5694c13d758SJosef Bacik 	trans->block_rsv = NULL;
570d13603efSArne Jansen 	/*
571d13603efSArne Jansen 	 * the same root has to be passed to start_transaction and
572d13603efSArne Jansen 	 * end_transaction. Subvolume quota depends on this.
573d13603efSArne Jansen 	 */
574d13603efSArne Jansen 	WARN_ON(trans->root != root);
575c5567237SArne Jansen 
576c5567237SArne Jansen 	if (trans->qgroup_reserved) {
577c5567237SArne Jansen 		btrfs_qgroup_free(root, trans->qgroup_reserved);
578c5567237SArne Jansen 		trans->qgroup_reserved = 0;
579c5567237SArne Jansen 	}
580c5567237SArne Jansen 
581ea658badSJosef Bacik 	if (!list_empty(&trans->new_bgs))
582ea658badSJosef Bacik 		btrfs_create_pending_block_groups(trans, root);
583ea658badSJosef Bacik 
584203bf287SChris Mason 	while (count < 2) {
585c3e69d58SChris Mason 		unsigned long cur = trans->delayed_ref_updates;
586c3e69d58SChris Mason 		trans->delayed_ref_updates = 0;
587c3e69d58SChris Mason 		if (cur &&
588c3e69d58SChris Mason 		    trans->transaction->delayed_refs.num_heads_ready > 64) {
589c3e69d58SChris Mason 			trans->delayed_ref_updates = 0;
590c3e69d58SChris Mason 			btrfs_run_delayed_refs(trans, root, cur);
591c3e69d58SChris Mason 		} else {
592c3e69d58SChris Mason 			break;
593c3e69d58SChris Mason 		}
594c3e69d58SChris Mason 		count++;
59556bec294SChris Mason 	}
5960e721106SJosef Bacik 	btrfs_trans_release_metadata(trans, root);
5970e721106SJosef Bacik 	trans->block_rsv = NULL;
59856bec294SChris Mason 
599ea658badSJosef Bacik 	if (!list_empty(&trans->new_bgs))
600ea658badSJosef Bacik 		btrfs_create_pending_block_groups(trans, root);
601ea658badSJosef Bacik 
602a4abeea4SJosef Bacik 	if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
603a4abeea4SJosef Bacik 	    should_end_transaction(trans, root)) {
6048929ecfaSYan, Zheng 		trans->transaction->blocked = 1;
605a4abeea4SJosef Bacik 		smp_wmb();
606a4abeea4SJosef Bacik 	}
6078929ecfaSYan, Zheng 
6080af3d00bSJosef Bacik 	if (lock && cur_trans->blocked && !cur_trans->in_commit) {
60981317fdeSJosef Bacik 		if (throttle) {
61081317fdeSJosef Bacik 			/*
61181317fdeSJosef Bacik 			 * We may race with somebody else here so end up having
61281317fdeSJosef Bacik 			 * to call end_transaction on ourselves again, so inc
61381317fdeSJosef Bacik 			 * our use_count.
61481317fdeSJosef Bacik 			 */
61581317fdeSJosef Bacik 			trans->use_count++;
6168929ecfaSYan, Zheng 			return btrfs_commit_transaction(trans, root);
61781317fdeSJosef Bacik 		} else {
6188929ecfaSYan, Zheng 			wake_up_process(info->transaction_kthread);
6198929ecfaSYan, Zheng 		}
62081317fdeSJosef Bacik 	}
6218929ecfaSYan, Zheng 
622354aa0fbSMiao Xie 	if (trans->type < TRANS_JOIN_NOLOCK)
6236df7881aSJosef Bacik 		sb_end_intwrite(root->fs_info->sb);
6246df7881aSJosef Bacik 
6258929ecfaSYan, Zheng 	WARN_ON(cur_trans != info->running_transaction);
62613c5a93eSJosef Bacik 	WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
62713c5a93eSJosef Bacik 	atomic_dec(&cur_trans->num_writers);
62889ce8a63SChris Mason 
62999d16cbcSSage Weil 	smp_mb();
63079154b1bSChris Mason 	if (waitqueue_active(&cur_trans->writer_wait))
63179154b1bSChris Mason 		wake_up(&cur_trans->writer_wait);
63279154b1bSChris Mason 	put_transaction(cur_trans);
6339ed74f2dSJosef Bacik 
6349ed74f2dSJosef Bacik 	if (current->journal_info == trans)
6359ed74f2dSJosef Bacik 		current->journal_info = NULL;
636ab78c84dSChris Mason 
63724bbcf04SYan, Zheng 	if (throttle)
63824bbcf04SYan, Zheng 		btrfs_run_delayed_iputs(root);
63924bbcf04SYan, Zheng 
64049b25e05SJeff Mahoney 	if (trans->aborted ||
64149b25e05SJeff Mahoney 	    root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
6424edc2ca3SDave Jones 		err = -EIO;
64349b25e05SJeff Mahoney 	}
644edf39272SJan Schmidt 	assert_qgroups_uptodate(trans);
64549b25e05SJeff Mahoney 
6464edc2ca3SDave Jones 	memset(trans, 0, sizeof(*trans));
6474edc2ca3SDave Jones 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
6484edc2ca3SDave Jones 	return err;
64979154b1bSChris Mason }
65079154b1bSChris Mason 
65189ce8a63SChris Mason int btrfs_end_transaction(struct btrfs_trans_handle *trans,
65289ce8a63SChris Mason 			  struct btrfs_root *root)
65389ce8a63SChris Mason {
65416cdcec7SMiao Xie 	int ret;
65516cdcec7SMiao Xie 
656a698d075SMiao Xie 	ret = __btrfs_end_transaction(trans, root, 0);
65716cdcec7SMiao Xie 	if (ret)
65816cdcec7SMiao Xie 		return ret;
65916cdcec7SMiao Xie 	return 0;
66089ce8a63SChris Mason }
66189ce8a63SChris Mason 
66289ce8a63SChris Mason int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
66389ce8a63SChris Mason 				   struct btrfs_root *root)
66489ce8a63SChris Mason {
66516cdcec7SMiao Xie 	int ret;
66616cdcec7SMiao Xie 
667a698d075SMiao Xie 	ret = __btrfs_end_transaction(trans, root, 1);
66816cdcec7SMiao Xie 	if (ret)
66916cdcec7SMiao Xie 		return ret;
67016cdcec7SMiao Xie 	return 0;
67116cdcec7SMiao Xie }
67216cdcec7SMiao Xie 
67316cdcec7SMiao Xie int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
67416cdcec7SMiao Xie 				struct btrfs_root *root)
67516cdcec7SMiao Xie {
676a698d075SMiao Xie 	return __btrfs_end_transaction(trans, root, 1);
67789ce8a63SChris Mason }
67889ce8a63SChris Mason 
679d352ac68SChris Mason /*
680d352ac68SChris Mason  * when btree blocks are allocated, they have some corresponding bits set for
681d352ac68SChris Mason  * them in one of two extent_io trees.  This is used to make sure all of
682690587d1SChris Mason  * those extents are sent to disk but does not wait on them
683d352ac68SChris Mason  */
684690587d1SChris Mason int btrfs_write_marked_extents(struct btrfs_root *root,
6858cef4e16SYan, Zheng 			       struct extent_io_tree *dirty_pages, int mark)
68679154b1bSChris Mason {
687777e6bd7SChris Mason 	int err = 0;
6887c4452b9SChris Mason 	int werr = 0;
6891728366eSJosef Bacik 	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
690777e6bd7SChris Mason 	u64 start = 0;
6915f39d397SChris Mason 	u64 end;
6927c4452b9SChris Mason 
6931728366eSJosef Bacik 	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
6941728366eSJosef Bacik 				      mark)) {
6951728366eSJosef Bacik 		convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
6961728366eSJosef Bacik 				   GFP_NOFS);
6971728366eSJosef Bacik 		err = filemap_fdatawrite_range(mapping, start, end);
6987c4452b9SChris Mason 		if (err)
6997c4452b9SChris Mason 			werr = err;
7001728366eSJosef Bacik 		cond_resched();
7011728366eSJosef Bacik 		start = end + 1;
7027c4452b9SChris Mason 	}
703690587d1SChris Mason 	if (err)
704690587d1SChris Mason 		werr = err;
705690587d1SChris Mason 	return werr;
706690587d1SChris Mason }
707690587d1SChris Mason 
708690587d1SChris Mason /*
709690587d1SChris Mason  * when btree blocks are allocated, they have some corresponding bits set for
710690587d1SChris Mason  * them in one of two extent_io trees.  This is used to make sure all of
711690587d1SChris Mason  * those extents are on disk for transaction or log commit.  We wait
712690587d1SChris Mason  * on all the pages and clear them from the dirty pages state tree
713690587d1SChris Mason  */
714690587d1SChris Mason int btrfs_wait_marked_extents(struct btrfs_root *root,
7158cef4e16SYan, Zheng 			      struct extent_io_tree *dirty_pages, int mark)
716690587d1SChris Mason {
717690587d1SChris Mason 	int err = 0;
718690587d1SChris Mason 	int werr = 0;
7191728366eSJosef Bacik 	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
720690587d1SChris Mason 	u64 start = 0;
721690587d1SChris Mason 	u64 end;
722690587d1SChris Mason 
7231728366eSJosef Bacik 	while (!find_first_extent_bit(dirty_pages, start, &start, &end,
7241728366eSJosef Bacik 				      EXTENT_NEED_WAIT)) {
7251728366eSJosef Bacik 		clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
7261728366eSJosef Bacik 		err = filemap_fdatawait_range(mapping, start, end);
727777e6bd7SChris Mason 		if (err)
728777e6bd7SChris Mason 			werr = err;
729777e6bd7SChris Mason 		cond_resched();
7301728366eSJosef Bacik 		start = end + 1;
731777e6bd7SChris Mason 	}
7327c4452b9SChris Mason 	if (err)
7337c4452b9SChris Mason 		werr = err;
7347c4452b9SChris Mason 	return werr;
73579154b1bSChris Mason }
73679154b1bSChris Mason 
737690587d1SChris Mason /*
738690587d1SChris Mason  * when btree blocks are allocated, they have some corresponding bits set for
739690587d1SChris Mason  * them in one of two extent_io trees.  This is used to make sure all of
740690587d1SChris Mason  * those extents are on disk for transaction or log commit
741690587d1SChris Mason  */
742690587d1SChris Mason int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
7438cef4e16SYan, Zheng 				struct extent_io_tree *dirty_pages, int mark)
744690587d1SChris Mason {
745690587d1SChris Mason 	int ret;
746690587d1SChris Mason 	int ret2;
747690587d1SChris Mason 
7488cef4e16SYan, Zheng 	ret = btrfs_write_marked_extents(root, dirty_pages, mark);
7498cef4e16SYan, Zheng 	ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
750bf0da8c1SChris Mason 
751bf0da8c1SChris Mason 	if (ret)
752bf0da8c1SChris Mason 		return ret;
753bf0da8c1SChris Mason 	if (ret2)
754bf0da8c1SChris Mason 		return ret2;
755bf0da8c1SChris Mason 	return 0;
756690587d1SChris Mason }
757690587d1SChris Mason 
758d0c803c4SChris Mason int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
759d0c803c4SChris Mason 				     struct btrfs_root *root)
760d0c803c4SChris Mason {
761d0c803c4SChris Mason 	if (!trans || !trans->transaction) {
762d0c803c4SChris Mason 		struct inode *btree_inode;
763d0c803c4SChris Mason 		btree_inode = root->fs_info->btree_inode;
764d0c803c4SChris Mason 		return filemap_write_and_wait(btree_inode->i_mapping);
765d0c803c4SChris Mason 	}
766d0c803c4SChris Mason 	return btrfs_write_and_wait_marked_extents(root,
7678cef4e16SYan, Zheng 					   &trans->transaction->dirty_pages,
7688cef4e16SYan, Zheng 					   EXTENT_DIRTY);
769d0c803c4SChris Mason }
770d0c803c4SChris Mason 
771d352ac68SChris Mason /*
772d352ac68SChris Mason  * this is used to update the root pointer in the tree of tree roots.
773d352ac68SChris Mason  *
774d352ac68SChris Mason  * But, in the case of the extent allocation tree, updating the root
775d352ac68SChris Mason  * pointer may allocate blocks which may change the root of the extent
776d352ac68SChris Mason  * allocation tree.
777d352ac68SChris Mason  *
778d352ac68SChris Mason  * So, this loops and repeats and makes sure the cowonly root didn't
779d352ac68SChris Mason  * change while the root pointer was being updated in the metadata.
780d352ac68SChris Mason  */
7810b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans,
78279154b1bSChris Mason 			       struct btrfs_root *root)
78379154b1bSChris Mason {
78479154b1bSChris Mason 	int ret;
7850b86a832SChris Mason 	u64 old_root_bytenr;
78686b9f2ecSYan, Zheng 	u64 old_root_used;
7870b86a832SChris Mason 	struct btrfs_root *tree_root = root->fs_info->tree_root;
78879154b1bSChris Mason 
78986b9f2ecSYan, Zheng 	old_root_used = btrfs_root_used(&root->root_item);
7900b86a832SChris Mason 	btrfs_write_dirty_block_groups(trans, root);
79156bec294SChris Mason 
79279154b1bSChris Mason 	while (1) {
7930b86a832SChris Mason 		old_root_bytenr = btrfs_root_bytenr(&root->root_item);
79486b9f2ecSYan, Zheng 		if (old_root_bytenr == root->node->start &&
79586b9f2ecSYan, Zheng 		    old_root_used == btrfs_root_used(&root->root_item))
79679154b1bSChris Mason 			break;
79787ef2bb4SChris Mason 
7985d4f98a2SYan Zheng 		btrfs_set_root_node(&root->root_item, root->node);
79979154b1bSChris Mason 		ret = btrfs_update_root(trans, tree_root,
8000b86a832SChris Mason 					&root->root_key,
8010b86a832SChris Mason 					&root->root_item);
80249b25e05SJeff Mahoney 		if (ret)
80349b25e05SJeff Mahoney 			return ret;
80456bec294SChris Mason 
80586b9f2ecSYan, Zheng 		old_root_used = btrfs_root_used(&root->root_item);
8064a8c9a62SYan Zheng 		ret = btrfs_write_dirty_block_groups(trans, root);
80749b25e05SJeff Mahoney 		if (ret)
80849b25e05SJeff Mahoney 			return ret;
8090b86a832SChris Mason 	}
810276e680dSYan Zheng 
811276e680dSYan Zheng 	if (root != root->fs_info->extent_root)
812817d52f8SJosef Bacik 		switch_commit_root(root);
813276e680dSYan Zheng 
8140b86a832SChris Mason 	return 0;
8150b86a832SChris Mason }
8160b86a832SChris Mason 
817d352ac68SChris Mason /*
818d352ac68SChris Mason  * update all the cowonly tree roots on disk
81949b25e05SJeff Mahoney  *
82049b25e05SJeff Mahoney  * The error handling in this function may not be obvious. Any of the
82149b25e05SJeff Mahoney  * failures will cause the file system to go offline. We still need
82249b25e05SJeff Mahoney  * to clean up the delayed refs.
823d352ac68SChris Mason  */
8245d4f98a2SYan Zheng static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
8250b86a832SChris Mason 					 struct btrfs_root *root)
8260b86a832SChris Mason {
8270b86a832SChris Mason 	struct btrfs_fs_info *fs_info = root->fs_info;
8280b86a832SChris Mason 	struct list_head *next;
82984234f3aSYan Zheng 	struct extent_buffer *eb;
83056bec294SChris Mason 	int ret;
83184234f3aSYan Zheng 
83256bec294SChris Mason 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
83349b25e05SJeff Mahoney 	if (ret)
83449b25e05SJeff Mahoney 		return ret;
83587ef2bb4SChris Mason 
83684234f3aSYan Zheng 	eb = btrfs_lock_root_node(fs_info->tree_root);
83749b25e05SJeff Mahoney 	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
83849b25e05SJeff Mahoney 			      0, &eb);
83984234f3aSYan Zheng 	btrfs_tree_unlock(eb);
84084234f3aSYan Zheng 	free_extent_buffer(eb);
8410b86a832SChris Mason 
84249b25e05SJeff Mahoney 	if (ret)
84349b25e05SJeff Mahoney 		return ret;
84449b25e05SJeff Mahoney 
84556bec294SChris Mason 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
84649b25e05SJeff Mahoney 	if (ret)
84749b25e05SJeff Mahoney 		return ret;
84887ef2bb4SChris Mason 
849733f4fbbSStefan Behrens 	ret = btrfs_run_dev_stats(trans, root->fs_info);
850733f4fbbSStefan Behrens 	BUG_ON(ret);
851733f4fbbSStefan Behrens 
852546adb0dSJan Schmidt 	ret = btrfs_run_qgroups(trans, root->fs_info);
853546adb0dSJan Schmidt 	BUG_ON(ret);
854546adb0dSJan Schmidt 
855546adb0dSJan Schmidt 	/* run_qgroups might have added some more refs */
856546adb0dSJan Schmidt 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
857546adb0dSJan Schmidt 	BUG_ON(ret);
858546adb0dSJan Schmidt 
8590b86a832SChris Mason 	while (!list_empty(&fs_info->dirty_cowonly_roots)) {
8600b86a832SChris Mason 		next = fs_info->dirty_cowonly_roots.next;
8610b86a832SChris Mason 		list_del_init(next);
8620b86a832SChris Mason 		root = list_entry(next, struct btrfs_root, dirty_list);
86387ef2bb4SChris Mason 
86449b25e05SJeff Mahoney 		ret = update_cowonly_root(trans, root);
86549b25e05SJeff Mahoney 		if (ret)
86649b25e05SJeff Mahoney 			return ret;
86779154b1bSChris Mason 	}
868276e680dSYan Zheng 
869276e680dSYan Zheng 	down_write(&fs_info->extent_commit_sem);
870276e680dSYan Zheng 	switch_commit_root(fs_info->extent_root);
871276e680dSYan Zheng 	up_write(&fs_info->extent_commit_sem);
872276e680dSYan Zheng 
87379154b1bSChris Mason 	return 0;
87479154b1bSChris Mason }
87579154b1bSChris Mason 
876d352ac68SChris Mason /*
877d352ac68SChris Mason  * dead roots are old snapshots that need to be deleted.  This allocates
878d352ac68SChris Mason  * a dirty root struct and adds it into the list of dead roots that need to
879d352ac68SChris Mason  * be deleted
880d352ac68SChris Mason  */
8815d4f98a2SYan Zheng int btrfs_add_dead_root(struct btrfs_root *root)
8825eda7b5eSChris Mason {
883a4abeea4SJosef Bacik 	spin_lock(&root->fs_info->trans_lock);
8845d4f98a2SYan Zheng 	list_add(&root->root_list, &root->fs_info->dead_roots);
885a4abeea4SJosef Bacik 	spin_unlock(&root->fs_info->trans_lock);
8865eda7b5eSChris Mason 	return 0;
8875eda7b5eSChris Mason }
8885eda7b5eSChris Mason 
889d352ac68SChris Mason /*
8905d4f98a2SYan Zheng  * update all the cowonly tree roots on disk
891d352ac68SChris Mason  */
8925d4f98a2SYan Zheng static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
8935d4f98a2SYan Zheng 				    struct btrfs_root *root)
8940f7d52f4SChris Mason {
8950f7d52f4SChris Mason 	struct btrfs_root *gang[8];
8965d4f98a2SYan Zheng 	struct btrfs_fs_info *fs_info = root->fs_info;
8970f7d52f4SChris Mason 	int i;
8980f7d52f4SChris Mason 	int ret;
89954aa1f4dSChris Mason 	int err = 0;
90054aa1f4dSChris Mason 
901a4abeea4SJosef Bacik 	spin_lock(&fs_info->fs_roots_radix_lock);
9020f7d52f4SChris Mason 	while (1) {
9035d4f98a2SYan Zheng 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
9045d4f98a2SYan Zheng 						 (void **)gang, 0,
9050f7d52f4SChris Mason 						 ARRAY_SIZE(gang),
9060f7d52f4SChris Mason 						 BTRFS_ROOT_TRANS_TAG);
9070f7d52f4SChris Mason 		if (ret == 0)
9080f7d52f4SChris Mason 			break;
9090f7d52f4SChris Mason 		for (i = 0; i < ret; i++) {
9100f7d52f4SChris Mason 			root = gang[i];
9115d4f98a2SYan Zheng 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
9122619ba1fSChris Mason 					(unsigned long)root->root_key.objectid,
9130f7d52f4SChris Mason 					BTRFS_ROOT_TRANS_TAG);
914a4abeea4SJosef Bacik 			spin_unlock(&fs_info->fs_roots_radix_lock);
91531153d81SYan Zheng 
916e02119d5SChris Mason 			btrfs_free_log(trans, root);
9175d4f98a2SYan Zheng 			btrfs_update_reloc_root(trans, root);
918d68fc57bSYan, Zheng 			btrfs_orphan_commit_root(trans, root);
919e02119d5SChris Mason 
92082d5902dSLi Zefan 			btrfs_save_ino_cache(root, trans);
92182d5902dSLi Zefan 
922f1ebcc74SLiu Bo 			/* see comments in should_cow_block() */
923f1ebcc74SLiu Bo 			root->force_cow = 0;
924f1ebcc74SLiu Bo 			smp_wmb();
925f1ebcc74SLiu Bo 
926978d910dSYan Zheng 			if (root->commit_root != root->node) {
927581bb050SLi Zefan 				mutex_lock(&root->fs_commit_mutex);
928817d52f8SJosef Bacik 				switch_commit_root(root);
929581bb050SLi Zefan 				btrfs_unpin_free_ino(root);
930581bb050SLi Zefan 				mutex_unlock(&root->fs_commit_mutex);
931581bb050SLi Zefan 
932978d910dSYan Zheng 				btrfs_set_root_node(&root->root_item,
933978d910dSYan Zheng 						    root->node);
934978d910dSYan Zheng 			}
93531153d81SYan Zheng 
9365d4f98a2SYan Zheng 			err = btrfs_update_root(trans, fs_info->tree_root,
9370f7d52f4SChris Mason 						&root->root_key,
9380f7d52f4SChris Mason 						&root->root_item);
939a4abeea4SJosef Bacik 			spin_lock(&fs_info->fs_roots_radix_lock);
94054aa1f4dSChris Mason 			if (err)
94154aa1f4dSChris Mason 				break;
9420f7d52f4SChris Mason 		}
9439f3a7427SChris Mason 	}
944a4abeea4SJosef Bacik 	spin_unlock(&fs_info->fs_roots_radix_lock);
94554aa1f4dSChris Mason 	return err;
9460f7d52f4SChris Mason }
9470f7d52f4SChris Mason 
948d352ac68SChris Mason /*
949d352ac68SChris Mason  * defrag a given btree.  If cacheonly == 1, this won't read from the disk,
950d352ac68SChris Mason  * otherwise every leaf in the btree is read and defragged.
951d352ac68SChris Mason  */
952e9d0b13bSChris Mason int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
953e9d0b13bSChris Mason {
954e9d0b13bSChris Mason 	struct btrfs_fs_info *info = root->fs_info;
955e9d0b13bSChris Mason 	struct btrfs_trans_handle *trans;
9568929ecfaSYan, Zheng 	int ret;
957d3c2fdcfSChris Mason 	unsigned long nr;
958e9d0b13bSChris Mason 
9598929ecfaSYan, Zheng 	if (xchg(&root->defrag_running, 1))
960e9d0b13bSChris Mason 		return 0;
9618929ecfaSYan, Zheng 
9626b80053dSChris Mason 	while (1) {
9638929ecfaSYan, Zheng 		trans = btrfs_start_transaction(root, 0);
9648929ecfaSYan, Zheng 		if (IS_ERR(trans))
9658929ecfaSYan, Zheng 			return PTR_ERR(trans);
9668929ecfaSYan, Zheng 
967e9d0b13bSChris Mason 		ret = btrfs_defrag_leaves(trans, root, cacheonly);
9688929ecfaSYan, Zheng 
969d3c2fdcfSChris Mason 		nr = trans->blocks_used;
970e9d0b13bSChris Mason 		btrfs_end_transaction(trans, root);
971d3c2fdcfSChris Mason 		btrfs_btree_balance_dirty(info->tree_root, nr);
972e9d0b13bSChris Mason 		cond_resched();
973e9d0b13bSChris Mason 
9747841cb28SDavid Sterba 		if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
975e9d0b13bSChris Mason 			break;
976e9d0b13bSChris Mason 	}
977e9d0b13bSChris Mason 	root->defrag_running = 0;
9788929ecfaSYan, Zheng 	return ret;
979e9d0b13bSChris Mason }
980e9d0b13bSChris Mason 
981d352ac68SChris Mason /*
982d352ac68SChris Mason  * new snapshots need to be created at a very specific time in the
983d352ac68SChris Mason  * transaction commit.  This does the actual creation
984d352ac68SChris Mason  */
98580b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
9863063d29fSChris Mason 				   struct btrfs_fs_info *fs_info,
9873063d29fSChris Mason 				   struct btrfs_pending_snapshot *pending)
9883063d29fSChris Mason {
9893063d29fSChris Mason 	struct btrfs_key key;
99080b6794dSChris Mason 	struct btrfs_root_item *new_root_item;
9913063d29fSChris Mason 	struct btrfs_root *tree_root = fs_info->tree_root;
9923063d29fSChris Mason 	struct btrfs_root *root = pending->root;
9936bdb72deSSage Weil 	struct btrfs_root *parent_root;
99498c9942aSLiu Bo 	struct btrfs_block_rsv *rsv;
9956bdb72deSSage Weil 	struct inode *parent_inode;
99642874b3dSMiao Xie 	struct btrfs_path *path;
99742874b3dSMiao Xie 	struct btrfs_dir_item *dir_item;
9986a912213SJosef Bacik 	struct dentry *parent;
999a22285a6SYan, Zheng 	struct dentry *dentry;
10003063d29fSChris Mason 	struct extent_buffer *tmp;
1001925baeddSChris Mason 	struct extent_buffer *old;
10028ea05e3aSAlexander Block 	struct timespec cur_time = CURRENT_TIME;
10033063d29fSChris Mason 	int ret;
1004d68fc57bSYan, Zheng 	u64 to_reserve = 0;
10056bdb72deSSage Weil 	u64 index = 0;
1006a22285a6SYan, Zheng 	u64 objectid;
1007b83cc969SLi Zefan 	u64 root_flags;
10088ea05e3aSAlexander Block 	uuid_le new_uuid;
10093063d29fSChris Mason 
101042874b3dSMiao Xie 	path = btrfs_alloc_path();
101142874b3dSMiao Xie 	if (!path) {
101242874b3dSMiao Xie 		ret = pending->error = -ENOMEM;
101342874b3dSMiao Xie 		goto path_alloc_fail;
101442874b3dSMiao Xie 	}
101542874b3dSMiao Xie 
101680b6794dSChris Mason 	new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
101780b6794dSChris Mason 	if (!new_root_item) {
101849b25e05SJeff Mahoney 		ret = pending->error = -ENOMEM;
10196fa9700eSMiao Xie 		goto root_item_alloc_fail;
102080b6794dSChris Mason 	}
1021a22285a6SYan, Zheng 
1022581bb050SLi Zefan 	ret = btrfs_find_free_objectid(tree_root, &objectid);
1023a22285a6SYan, Zheng 	if (ret) {
1024a22285a6SYan, Zheng 		pending->error = ret;
10256fa9700eSMiao Xie 		goto no_free_objectid;
1026a22285a6SYan, Zheng 	}
10273063d29fSChris Mason 
10283fd0a558SYan, Zheng 	btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1029d68fc57bSYan, Zheng 
1030d68fc57bSYan, Zheng 	if (to_reserve > 0) {
103162f30c54SMiao Xie 		ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
10328bb8ab2eSJosef Bacik 						  to_reserve);
1033d68fc57bSYan, Zheng 		if (ret) {
1034d68fc57bSYan, Zheng 			pending->error = ret;
10356fa9700eSMiao Xie 			goto no_free_objectid;
1036d68fc57bSYan, Zheng 		}
1037d68fc57bSYan, Zheng 	}
1038d68fc57bSYan, Zheng 
10396f72c7e2SArne Jansen 	ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
10406f72c7e2SArne Jansen 				   objectid, pending->inherit);
10416f72c7e2SArne Jansen 	if (ret) {
10426f72c7e2SArne Jansen 		pending->error = ret;
10436fa9700eSMiao Xie 		goto no_free_objectid;
10446f72c7e2SArne Jansen 	}
10456f72c7e2SArne Jansen 
10463063d29fSChris Mason 	key.objectid = objectid;
1047a22285a6SYan, Zheng 	key.offset = (u64)-1;
1048a22285a6SYan, Zheng 	key.type = BTRFS_ROOT_ITEM_KEY;
10493063d29fSChris Mason 
10506fa9700eSMiao Xie 	rsv = trans->block_rsv;
1051a22285a6SYan, Zheng 	trans->block_rsv = &pending->block_rsv;
10526bdb72deSSage Weil 
1053a22285a6SYan, Zheng 	dentry = pending->dentry;
10546a912213SJosef Bacik 	parent = dget_parent(dentry);
10556a912213SJosef Bacik 	parent_inode = parent->d_inode;
1056a22285a6SYan, Zheng 	parent_root = BTRFS_I(parent_inode)->root;
10577585717fSChris Mason 	record_root_in_trans(trans, parent_root);
1058a22285a6SYan, Zheng 
10596bdb72deSSage Weil 	/*
10606bdb72deSSage Weil 	 * insert the directory item
10616bdb72deSSage Weil 	 */
10626bdb72deSSage Weil 	ret = btrfs_set_inode_index(parent_inode, &index);
106349b25e05SJeff Mahoney 	BUG_ON(ret); /* -ENOMEM */
106442874b3dSMiao Xie 
106542874b3dSMiao Xie 	/* check if there is a file/dir which has the same name. */
106642874b3dSMiao Xie 	dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
106742874b3dSMiao Xie 					 btrfs_ino(parent_inode),
106842874b3dSMiao Xie 					 dentry->d_name.name,
106942874b3dSMiao Xie 					 dentry->d_name.len, 0);
107042874b3dSMiao Xie 	if (dir_item != NULL && !IS_ERR(dir_item)) {
1071fe66a05aSChris Mason 		pending->error = -EEXIST;
1072fe66a05aSChris Mason 		goto fail;
107342874b3dSMiao Xie 	} else if (IS_ERR(dir_item)) {
107442874b3dSMiao Xie 		ret = PTR_ERR(dir_item);
10758732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
10768732d44fSMiao Xie 		goto fail;
107779787eaaSJeff Mahoney 	}
107842874b3dSMiao Xie 	btrfs_release_path(path);
10796bdb72deSSage Weil 
1080e999376fSChris Mason 	/*
1081e999376fSChris Mason 	 * pull in the delayed directory update
1082e999376fSChris Mason 	 * and the delayed inode item
1083e999376fSChris Mason 	 * otherwise we corrupt the FS during
1084e999376fSChris Mason 	 * snapshot
1085e999376fSChris Mason 	 */
1086e999376fSChris Mason 	ret = btrfs_run_delayed_items(trans, root);
10878732d44fSMiao Xie 	if (ret) {	/* Transaction aborted */
10888732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
10898732d44fSMiao Xie 		goto fail;
10908732d44fSMiao Xie 	}
1091e999376fSChris Mason 
10927585717fSChris Mason 	record_root_in_trans(trans, root);
10936bdb72deSSage Weil 	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
10946bdb72deSSage Weil 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
109508fe4db1SLi Zefan 	btrfs_check_and_init_root_item(new_root_item);
10966bdb72deSSage Weil 
1097b83cc969SLi Zefan 	root_flags = btrfs_root_flags(new_root_item);
1098b83cc969SLi Zefan 	if (pending->readonly)
1099b83cc969SLi Zefan 		root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1100b83cc969SLi Zefan 	else
1101b83cc969SLi Zefan 		root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1102b83cc969SLi Zefan 	btrfs_set_root_flags(new_root_item, root_flags);
1103b83cc969SLi Zefan 
11048ea05e3aSAlexander Block 	btrfs_set_root_generation_v2(new_root_item,
11058ea05e3aSAlexander Block 			trans->transid);
11068ea05e3aSAlexander Block 	uuid_le_gen(&new_uuid);
11078ea05e3aSAlexander Block 	memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
11088ea05e3aSAlexander Block 	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
11098ea05e3aSAlexander Block 			BTRFS_UUID_SIZE);
11108ea05e3aSAlexander Block 	new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1111dadd1105SDan Carpenter 	new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
11128ea05e3aSAlexander Block 	btrfs_set_root_otransid(new_root_item, trans->transid);
11138ea05e3aSAlexander Block 	memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
11148ea05e3aSAlexander Block 	memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
11158ea05e3aSAlexander Block 	btrfs_set_root_stransid(new_root_item, 0);
11168ea05e3aSAlexander Block 	btrfs_set_root_rtransid(new_root_item, 0);
11178ea05e3aSAlexander Block 
1118925baeddSChris Mason 	old = btrfs_lock_root_node(root);
111949b25e05SJeff Mahoney 	ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
112079787eaaSJeff Mahoney 	if (ret) {
112179787eaaSJeff Mahoney 		btrfs_tree_unlock(old);
112279787eaaSJeff Mahoney 		free_extent_buffer(old);
11238732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
11248732d44fSMiao Xie 		goto fail;
112579787eaaSJeff Mahoney 	}
112649b25e05SJeff Mahoney 
11275d4f98a2SYan Zheng 	btrfs_set_lock_blocking(old);
11283063d29fSChris Mason 
112949b25e05SJeff Mahoney 	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
113079787eaaSJeff Mahoney 	/* clean up in any case */
1131925baeddSChris Mason 	btrfs_tree_unlock(old);
1132925baeddSChris Mason 	free_extent_buffer(old);
11338732d44fSMiao Xie 	if (ret) {
11348732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
11358732d44fSMiao Xie 		goto fail;
11368732d44fSMiao Xie 	}
11373063d29fSChris Mason 
1138f1ebcc74SLiu Bo 	/* see comments in should_cow_block() */
1139f1ebcc74SLiu Bo 	root->force_cow = 1;
1140f1ebcc74SLiu Bo 	smp_wmb();
1141f1ebcc74SLiu Bo 
11425d4f98a2SYan Zheng 	btrfs_set_root_node(new_root_item, tmp);
1143a22285a6SYan, Zheng 	/* record when the snapshot was created in key.offset */
1144a22285a6SYan, Zheng 	key.offset = trans->transid;
1145a22285a6SYan, Zheng 	ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1146925baeddSChris Mason 	btrfs_tree_unlock(tmp);
11473063d29fSChris Mason 	free_extent_buffer(tmp);
11488732d44fSMiao Xie 	if (ret) {
11498732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
11508732d44fSMiao Xie 		goto fail;
11518732d44fSMiao Xie 	}
11520660b5afSChris Mason 
1153a22285a6SYan, Zheng 	/*
1154a22285a6SYan, Zheng 	 * insert root back/forward references
1155a22285a6SYan, Zheng 	 */
1156a22285a6SYan, Zheng 	ret = btrfs_add_root_ref(trans, tree_root, objectid,
1157a22285a6SYan, Zheng 				 parent_root->root_key.objectid,
115833345d01SLi Zefan 				 btrfs_ino(parent_inode), index,
1159a22285a6SYan, Zheng 				 dentry->d_name.name, dentry->d_name.len);
11608732d44fSMiao Xie 	if (ret) {
11618732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
11628732d44fSMiao Xie 		goto fail;
11638732d44fSMiao Xie 	}
1164a22285a6SYan, Zheng 
1165a22285a6SYan, Zheng 	key.offset = (u64)-1;
1166a22285a6SYan, Zheng 	pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
116779787eaaSJeff Mahoney 	if (IS_ERR(pending->snap)) {
116879787eaaSJeff Mahoney 		ret = PTR_ERR(pending->snap);
11698732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
11708732d44fSMiao Xie 		goto fail;
117179787eaaSJeff Mahoney 	}
1172d68fc57bSYan, Zheng 
117349b25e05SJeff Mahoney 	ret = btrfs_reloc_post_snapshot(trans, pending);
11748732d44fSMiao Xie 	if (ret) {
11758732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
11768732d44fSMiao Xie 		goto fail;
11778732d44fSMiao Xie 	}
1178361048f5SMiao Xie 
1179361048f5SMiao Xie 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
11808732d44fSMiao Xie 	if (ret) {
11818732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
11828732d44fSMiao Xie 		goto fail;
11838732d44fSMiao Xie 	}
118442874b3dSMiao Xie 
118542874b3dSMiao Xie 	ret = btrfs_insert_dir_item(trans, parent_root,
118642874b3dSMiao Xie 				    dentry->d_name.name, dentry->d_name.len,
118742874b3dSMiao Xie 				    parent_inode, &key,
118842874b3dSMiao Xie 				    BTRFS_FT_DIR, index);
118942874b3dSMiao Xie 	/* We have check then name at the beginning, so it is impossible. */
119042874b3dSMiao Xie 	BUG_ON(ret == -EEXIST);
11918732d44fSMiao Xie 	if (ret) {
11928732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
11938732d44fSMiao Xie 		goto fail;
11948732d44fSMiao Xie 	}
119542874b3dSMiao Xie 
119642874b3dSMiao Xie 	btrfs_i_size_write(parent_inode, parent_inode->i_size +
119742874b3dSMiao Xie 					 dentry->d_name.len * 2);
119842874b3dSMiao Xie 	parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
119942874b3dSMiao Xie 	ret = btrfs_update_inode(trans, parent_root, parent_inode);
120042874b3dSMiao Xie 	if (ret)
12018732d44fSMiao Xie 		btrfs_abort_transaction(trans, root, ret);
12023063d29fSChris Mason fail:
12036fa9700eSMiao Xie 	dput(parent);
120498c9942aSLiu Bo 	trans->block_rsv = rsv;
12056fa9700eSMiao Xie no_free_objectid:
12066fa9700eSMiao Xie 	kfree(new_root_item);
12076fa9700eSMiao Xie root_item_alloc_fail:
120842874b3dSMiao Xie 	btrfs_free_path(path);
120942874b3dSMiao Xie path_alloc_fail:
1210a22285a6SYan, Zheng 	btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
121149b25e05SJeff Mahoney 	return ret;
12123063d29fSChris Mason }
12133063d29fSChris Mason 
1214d352ac68SChris Mason /*
1215d352ac68SChris Mason  * create all the snapshots we've scheduled for creation
1216d352ac68SChris Mason  */
121780b6794dSChris Mason static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
12183063d29fSChris Mason 					     struct btrfs_fs_info *fs_info)
12193063d29fSChris Mason {
12203063d29fSChris Mason 	struct btrfs_pending_snapshot *pending;
12213063d29fSChris Mason 	struct list_head *head = &trans->transaction->pending_snapshots;
12223de4586cSChris Mason 
1223fe66a05aSChris Mason 	list_for_each_entry(pending, head, list)
1224fe66a05aSChris Mason 		create_pending_snapshot(trans, fs_info, pending);
12253de4586cSChris Mason 	return 0;
12263de4586cSChris Mason }
12273de4586cSChris Mason 
12285d4f98a2SYan Zheng static void update_super_roots(struct btrfs_root *root)
12295d4f98a2SYan Zheng {
12305d4f98a2SYan Zheng 	struct btrfs_root_item *root_item;
12315d4f98a2SYan Zheng 	struct btrfs_super_block *super;
12325d4f98a2SYan Zheng 
12336c41761fSDavid Sterba 	super = root->fs_info->super_copy;
12345d4f98a2SYan Zheng 
12355d4f98a2SYan Zheng 	root_item = &root->fs_info->chunk_root->root_item;
12365d4f98a2SYan Zheng 	super->chunk_root = root_item->bytenr;
12375d4f98a2SYan Zheng 	super->chunk_root_generation = root_item->generation;
12385d4f98a2SYan Zheng 	super->chunk_root_level = root_item->level;
12395d4f98a2SYan Zheng 
12405d4f98a2SYan Zheng 	root_item = &root->fs_info->tree_root->root_item;
12415d4f98a2SYan Zheng 	super->root = root_item->bytenr;
12425d4f98a2SYan Zheng 	super->generation = root_item->generation;
12435d4f98a2SYan Zheng 	super->root_level = root_item->level;
124473bc1876SJosef Bacik 	if (btrfs_test_opt(root, SPACE_CACHE))
12450af3d00bSJosef Bacik 		super->cache_generation = root_item->generation;
12465d4f98a2SYan Zheng }
12475d4f98a2SYan Zheng 
1248f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1249f36f3042SChris Mason {
1250f36f3042SChris Mason 	int ret = 0;
1251a4abeea4SJosef Bacik 	spin_lock(&info->trans_lock);
1252f36f3042SChris Mason 	if (info->running_transaction)
1253f36f3042SChris Mason 		ret = info->running_transaction->in_commit;
1254a4abeea4SJosef Bacik 	spin_unlock(&info->trans_lock);
1255f36f3042SChris Mason 	return ret;
1256f36f3042SChris Mason }
1257f36f3042SChris Mason 
12588929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info)
12598929ecfaSYan, Zheng {
12608929ecfaSYan, Zheng 	int ret = 0;
1261a4abeea4SJosef Bacik 	spin_lock(&info->trans_lock);
12628929ecfaSYan, Zheng 	if (info->running_transaction)
12638929ecfaSYan, Zheng 		ret = info->running_transaction->blocked;
1264a4abeea4SJosef Bacik 	spin_unlock(&info->trans_lock);
12658929ecfaSYan, Zheng 	return ret;
12668929ecfaSYan, Zheng }
12678929ecfaSYan, Zheng 
1268bb9c12c9SSage Weil /*
1269bb9c12c9SSage Weil  * wait for the current transaction commit to start and block subsequent
1270bb9c12c9SSage Weil  * transaction joins
1271bb9c12c9SSage Weil  */
1272bb9c12c9SSage Weil static void wait_current_trans_commit_start(struct btrfs_root *root,
1273bb9c12c9SSage Weil 					    struct btrfs_transaction *trans)
1274bb9c12c9SSage Weil {
127572d63ed6SLi Zefan 	wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1276bb9c12c9SSage Weil }
1277bb9c12c9SSage Weil 
1278bb9c12c9SSage Weil /*
1279bb9c12c9SSage Weil  * wait for the current transaction to start and then become unblocked.
1280bb9c12c9SSage Weil  * caller holds ref.
1281bb9c12c9SSage Weil  */
1282bb9c12c9SSage Weil static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1283bb9c12c9SSage Weil 					 struct btrfs_transaction *trans)
1284bb9c12c9SSage Weil {
128572d63ed6SLi Zefan 	wait_event(root->fs_info->transaction_wait,
128672d63ed6SLi Zefan 		   trans->commit_done || (trans->in_commit && !trans->blocked));
1287bb9c12c9SSage Weil }
1288bb9c12c9SSage Weil 
1289bb9c12c9SSage Weil /*
1290bb9c12c9SSage Weil  * commit transactions asynchronously. once btrfs_commit_transaction_async
1291bb9c12c9SSage Weil  * returns, any subsequent transaction will not be allowed to join.
1292bb9c12c9SSage Weil  */
1293bb9c12c9SSage Weil struct btrfs_async_commit {
1294bb9c12c9SSage Weil 	struct btrfs_trans_handle *newtrans;
1295bb9c12c9SSage Weil 	struct btrfs_root *root;
1296bb9c12c9SSage Weil 	struct delayed_work work;
1297bb9c12c9SSage Weil };
1298bb9c12c9SSage Weil 
1299bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work)
1300bb9c12c9SSage Weil {
1301bb9c12c9SSage Weil 	struct btrfs_async_commit *ac =
1302bb9c12c9SSage Weil 		container_of(work, struct btrfs_async_commit, work.work);
1303bb9c12c9SSage Weil 
13046fc4e354SSage Weil 	/*
13056fc4e354SSage Weil 	 * We've got freeze protection passed with the transaction.
13066fc4e354SSage Weil 	 * Tell lockdep about it.
13076fc4e354SSage Weil 	 */
13086fc4e354SSage Weil 	rwsem_acquire_read(
13096fc4e354SSage Weil 		&ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
13106fc4e354SSage Weil 		0, 1, _THIS_IP_);
13116fc4e354SSage Weil 
1312e209db7aSSage Weil 	current->journal_info = ac->newtrans;
1313e209db7aSSage Weil 
1314bb9c12c9SSage Weil 	btrfs_commit_transaction(ac->newtrans, ac->root);
1315bb9c12c9SSage Weil 	kfree(ac);
1316bb9c12c9SSage Weil }
1317bb9c12c9SSage Weil 
1318bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1319bb9c12c9SSage Weil 				   struct btrfs_root *root,
1320bb9c12c9SSage Weil 				   int wait_for_unblock)
1321bb9c12c9SSage Weil {
1322bb9c12c9SSage Weil 	struct btrfs_async_commit *ac;
1323bb9c12c9SSage Weil 	struct btrfs_transaction *cur_trans;
1324bb9c12c9SSage Weil 
1325bb9c12c9SSage Weil 	ac = kmalloc(sizeof(*ac), GFP_NOFS);
1326db5b493aSTsutomu Itoh 	if (!ac)
1327db5b493aSTsutomu Itoh 		return -ENOMEM;
1328bb9c12c9SSage Weil 
1329bb9c12c9SSage Weil 	INIT_DELAYED_WORK(&ac->work, do_async_commit);
1330bb9c12c9SSage Weil 	ac->root = root;
13317a7eaa40SJosef Bacik 	ac->newtrans = btrfs_join_transaction(root);
13323612b495STsutomu Itoh 	if (IS_ERR(ac->newtrans)) {
13333612b495STsutomu Itoh 		int err = PTR_ERR(ac->newtrans);
13343612b495STsutomu Itoh 		kfree(ac);
13353612b495STsutomu Itoh 		return err;
13363612b495STsutomu Itoh 	}
1337bb9c12c9SSage Weil 
1338bb9c12c9SSage Weil 	/* take transaction reference */
1339bb9c12c9SSage Weil 	cur_trans = trans->transaction;
134013c5a93eSJosef Bacik 	atomic_inc(&cur_trans->use_count);
1341bb9c12c9SSage Weil 
1342bb9c12c9SSage Weil 	btrfs_end_transaction(trans, root);
13436fc4e354SSage Weil 
13446fc4e354SSage Weil 	/*
13456fc4e354SSage Weil 	 * Tell lockdep we've released the freeze rwsem, since the
13466fc4e354SSage Weil 	 * async commit thread will be the one to unlock it.
13476fc4e354SSage Weil 	 */
13486fc4e354SSage Weil 	rwsem_release(&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
13496fc4e354SSage Weil 		      1, _THIS_IP_);
13506fc4e354SSage Weil 
1351bb9c12c9SSage Weil 	schedule_delayed_work(&ac->work, 0);
1352bb9c12c9SSage Weil 
1353bb9c12c9SSage Weil 	/* wait for transaction to start and unblock */
1354bb9c12c9SSage Weil 	if (wait_for_unblock)
1355bb9c12c9SSage Weil 		wait_current_trans_commit_start_and_unblock(root, cur_trans);
1356bb9c12c9SSage Weil 	else
1357bb9c12c9SSage Weil 		wait_current_trans_commit_start(root, cur_trans);
1358bb9c12c9SSage Weil 
135938e88054SSage Weil 	if (current->journal_info == trans)
136038e88054SSage Weil 		current->journal_info = NULL;
136138e88054SSage Weil 
136238e88054SSage Weil 	put_transaction(cur_trans);
1363bb9c12c9SSage Weil 	return 0;
1364bb9c12c9SSage Weil }
1365bb9c12c9SSage Weil 
136649b25e05SJeff Mahoney 
136749b25e05SJeff Mahoney static void cleanup_transaction(struct btrfs_trans_handle *trans,
13687b8b92afSJosef Bacik 				struct btrfs_root *root, int err)
136949b25e05SJeff Mahoney {
137049b25e05SJeff Mahoney 	struct btrfs_transaction *cur_trans = trans->transaction;
137149b25e05SJeff Mahoney 
137249b25e05SJeff Mahoney 	WARN_ON(trans->use_count > 1);
137349b25e05SJeff Mahoney 
13747b8b92afSJosef Bacik 	btrfs_abort_transaction(trans, root, err);
13757b8b92afSJosef Bacik 
137649b25e05SJeff Mahoney 	spin_lock(&root->fs_info->trans_lock);
137749b25e05SJeff Mahoney 	list_del_init(&cur_trans->list);
1378d7096fc3SJosef Bacik 	if (cur_trans == root->fs_info->running_transaction) {
1379d7096fc3SJosef Bacik 		root->fs_info->running_transaction = NULL;
1380d7096fc3SJosef Bacik 		root->fs_info->trans_no_join = 0;
1381d7096fc3SJosef Bacik 	}
138249b25e05SJeff Mahoney 	spin_unlock(&root->fs_info->trans_lock);
138349b25e05SJeff Mahoney 
138449b25e05SJeff Mahoney 	btrfs_cleanup_one_transaction(trans->transaction, root);
138549b25e05SJeff Mahoney 
138649b25e05SJeff Mahoney 	put_transaction(cur_trans);
138749b25e05SJeff Mahoney 	put_transaction(cur_trans);
138849b25e05SJeff Mahoney 
138949b25e05SJeff Mahoney 	trace_btrfs_transaction_commit(root);
139049b25e05SJeff Mahoney 
139149b25e05SJeff Mahoney 	btrfs_scrub_continue(root);
139249b25e05SJeff Mahoney 
139349b25e05SJeff Mahoney 	if (current->journal_info == trans)
139449b25e05SJeff Mahoney 		current->journal_info = NULL;
139549b25e05SJeff Mahoney 
139649b25e05SJeff Mahoney 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
139749b25e05SJeff Mahoney }
139849b25e05SJeff Mahoney 
1399bb9c12c9SSage Weil /*
1400bb9c12c9SSage Weil  * btrfs_transaction state sequence:
1401bb9c12c9SSage Weil  *    in_commit = 0, blocked = 0  (initial)
1402bb9c12c9SSage Weil  *    in_commit = 1, blocked = 1
1403bb9c12c9SSage Weil  *    blocked = 0
1404bb9c12c9SSage Weil  *    commit_done = 1
1405bb9c12c9SSage Weil  */
140679154b1bSChris Mason int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
140779154b1bSChris Mason 			     struct btrfs_root *root)
140879154b1bSChris Mason {
140915ee9bc7SJosef Bacik 	unsigned long joined = 0;
141049b25e05SJeff Mahoney 	struct btrfs_transaction *cur_trans = trans->transaction;
14118fd17795SChris Mason 	struct btrfs_transaction *prev_trans = NULL;
141279154b1bSChris Mason 	DEFINE_WAIT(wait);
141349b25e05SJeff Mahoney 	int ret = -EIO;
141489573b9cSChris Mason 	int should_grow = 0;
141589573b9cSChris Mason 	unsigned long now = get_seconds();
1416dccae999SSage Weil 	int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
141779154b1bSChris Mason 
14185a3f23d5SChris Mason 	btrfs_run_ordered_operations(root, 0);
14195a3f23d5SChris Mason 
142049b25e05SJeff Mahoney 	if (cur_trans->aborted)
142149b25e05SJeff Mahoney 		goto cleanup_transaction;
142249b25e05SJeff Mahoney 
142356bec294SChris Mason 	/* make a pass through all the delayed refs we have so far
142456bec294SChris Mason 	 * any runnings procs may add more while we are here
142556bec294SChris Mason 	 */
142656bec294SChris Mason 	ret = btrfs_run_delayed_refs(trans, root, 0);
142749b25e05SJeff Mahoney 	if (ret)
142849b25e05SJeff Mahoney 		goto cleanup_transaction;
142956bec294SChris Mason 
14300e721106SJosef Bacik 	btrfs_trans_release_metadata(trans, root);
14310e721106SJosef Bacik 	trans->block_rsv = NULL;
14320e721106SJosef Bacik 
1433b7ec40d7SChris Mason 	cur_trans = trans->transaction;
143449b25e05SJeff Mahoney 
143556bec294SChris Mason 	/*
143656bec294SChris Mason 	 * set the flushing flag so procs in this transaction have to
143756bec294SChris Mason 	 * start sending their work down.
143856bec294SChris Mason 	 */
1439b7ec40d7SChris Mason 	cur_trans->delayed_refs.flushing = 1;
144056bec294SChris Mason 
1441ea658badSJosef Bacik 	if (!list_empty(&trans->new_bgs))
1442ea658badSJosef Bacik 		btrfs_create_pending_block_groups(trans, root);
1443ea658badSJosef Bacik 
1444c3e69d58SChris Mason 	ret = btrfs_run_delayed_refs(trans, root, 0);
144549b25e05SJeff Mahoney 	if (ret)
144649b25e05SJeff Mahoney 		goto cleanup_transaction;
144756bec294SChris Mason 
1448a4abeea4SJosef Bacik 	spin_lock(&cur_trans->commit_lock);
1449b7ec40d7SChris Mason 	if (cur_trans->in_commit) {
1450a4abeea4SJosef Bacik 		spin_unlock(&cur_trans->commit_lock);
145113c5a93eSJosef Bacik 		atomic_inc(&cur_trans->use_count);
145249b25e05SJeff Mahoney 		ret = btrfs_end_transaction(trans, root);
1453ccd467d6SChris Mason 
1454b9c8300cSLi Zefan 		wait_for_commit(root, cur_trans);
145515ee9bc7SJosef Bacik 
145679154b1bSChris Mason 		put_transaction(cur_trans);
145715ee9bc7SJosef Bacik 
145849b25e05SJeff Mahoney 		return ret;
145979154b1bSChris Mason 	}
14604313b399SChris Mason 
14612c90e5d6SChris Mason 	trans->transaction->in_commit = 1;
1462f9295749SChris Mason 	trans->transaction->blocked = 1;
1463a4abeea4SJosef Bacik 	spin_unlock(&cur_trans->commit_lock);
1464bb9c12c9SSage Weil 	wake_up(&root->fs_info->transaction_blocked_wait);
1465bb9c12c9SSage Weil 
1466a4abeea4SJosef Bacik 	spin_lock(&root->fs_info->trans_lock);
1467ccd467d6SChris Mason 	if (cur_trans->list.prev != &root->fs_info->trans_list) {
1468ccd467d6SChris Mason 		prev_trans = list_entry(cur_trans->list.prev,
1469ccd467d6SChris Mason 					struct btrfs_transaction, list);
1470ccd467d6SChris Mason 		if (!prev_trans->commit_done) {
147113c5a93eSJosef Bacik 			atomic_inc(&prev_trans->use_count);
1472a4abeea4SJosef Bacik 			spin_unlock(&root->fs_info->trans_lock);
1473ccd467d6SChris Mason 
1474ccd467d6SChris Mason 			wait_for_commit(root, prev_trans);
1475ccd467d6SChris Mason 
147615ee9bc7SJosef Bacik 			put_transaction(prev_trans);
1477a4abeea4SJosef Bacik 		} else {
1478a4abeea4SJosef Bacik 			spin_unlock(&root->fs_info->trans_lock);
1479ccd467d6SChris Mason 		}
1480a4abeea4SJosef Bacik 	} else {
1481a4abeea4SJosef Bacik 		spin_unlock(&root->fs_info->trans_lock);
1482ccd467d6SChris Mason 	}
148315ee9bc7SJosef Bacik 
1484e39e64acSChris Mason 	if (!btrfs_test_opt(root, SSD) &&
1485e39e64acSChris Mason 	    (now < cur_trans->start_time || now - cur_trans->start_time < 1))
148689573b9cSChris Mason 		should_grow = 1;
148789573b9cSChris Mason 
148815ee9bc7SJosef Bacik 	do {
14897ea394f1SYan Zheng 		int snap_pending = 0;
1490a4abeea4SJosef Bacik 
149115ee9bc7SJosef Bacik 		joined = cur_trans->num_joined;
14927ea394f1SYan Zheng 		if (!list_empty(&trans->transaction->pending_snapshots))
14937ea394f1SYan Zheng 			snap_pending = 1;
14947ea394f1SYan Zheng 
14952c90e5d6SChris Mason 		WARN_ON(cur_trans != trans->transaction);
149615ee9bc7SJosef Bacik 
14970bdb1db2SSage Weil 		if (flush_on_commit || snap_pending) {
149824bbcf04SYan, Zheng 			btrfs_start_delalloc_inodes(root, 1);
14996bbe3a9cSLiu Bo 			btrfs_wait_ordered_extents(root, 1);
15007ea394f1SYan Zheng 		}
15017ea394f1SYan Zheng 
150216cdcec7SMiao Xie 		ret = btrfs_run_delayed_items(trans, root);
150349b25e05SJeff Mahoney 		if (ret)
150449b25e05SJeff Mahoney 			goto cleanup_transaction;
150516cdcec7SMiao Xie 
15065a3f23d5SChris Mason 		/*
1507edf39272SJan Schmidt 		 * running the delayed items may have added new refs. account
1508edf39272SJan Schmidt 		 * them now so that they hinder processing of more delayed refs
1509edf39272SJan Schmidt 		 * as little as possible.
1510edf39272SJan Schmidt 		 */
1511edf39272SJan Schmidt 		btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1512edf39272SJan Schmidt 
1513edf39272SJan Schmidt 		/*
15145a3f23d5SChris Mason 		 * rename don't use btrfs_join_transaction, so, once we
15155a3f23d5SChris Mason 		 * set the transaction to blocked above, we aren't going
15165a3f23d5SChris Mason 		 * to get any new ordered operations.  We can safely run
15175a3f23d5SChris Mason 		 * it here and no for sure that nothing new will be added
15185a3f23d5SChris Mason 		 * to the list
15195a3f23d5SChris Mason 		 */
15205a3f23d5SChris Mason 		btrfs_run_ordered_operations(root, 1);
15215a3f23d5SChris Mason 
1522ed3b3d31SChris Mason 		prepare_to_wait(&cur_trans->writer_wait, &wait,
1523ed3b3d31SChris Mason 				TASK_UNINTERRUPTIBLE);
1524ed3b3d31SChris Mason 
152513c5a93eSJosef Bacik 		if (atomic_read(&cur_trans->num_writers) > 1)
152699d16cbcSSage Weil 			schedule_timeout(MAX_SCHEDULE_TIMEOUT);
152799d16cbcSSage Weil 		else if (should_grow)
152899d16cbcSSage Weil 			schedule_timeout(1);
152915ee9bc7SJosef Bacik 
153015ee9bc7SJosef Bacik 		finish_wait(&cur_trans->writer_wait, &wait);
1531ed0ca140SJosef Bacik 	} while (atomic_read(&cur_trans->num_writers) > 1 ||
1532ed0ca140SJosef Bacik 		 (should_grow && cur_trans->num_joined != joined));
1533ed0ca140SJosef Bacik 
1534ed0ca140SJosef Bacik 	/*
1535ed0ca140SJosef Bacik 	 * Ok now we need to make sure to block out any other joins while we
1536ed0ca140SJosef Bacik 	 * commit the transaction.  We could have started a join before setting
1537ed0ca140SJosef Bacik 	 * no_join so make sure to wait for num_writers to == 1 again.
1538ed0ca140SJosef Bacik 	 */
1539a4abeea4SJosef Bacik 	spin_lock(&root->fs_info->trans_lock);
1540a4abeea4SJosef Bacik 	root->fs_info->trans_no_join = 1;
1541a4abeea4SJosef Bacik 	spin_unlock(&root->fs_info->trans_lock);
1542ed0ca140SJosef Bacik 	wait_event(cur_trans->writer_wait,
1543ed0ca140SJosef Bacik 		   atomic_read(&cur_trans->num_writers) == 1);
154415ee9bc7SJosef Bacik 
15457585717fSChris Mason 	/*
15467585717fSChris Mason 	 * the reloc mutex makes sure that we stop
15477585717fSChris Mason 	 * the balancing code from coming in and moving
15487585717fSChris Mason 	 * extents around in the middle of the commit
15497585717fSChris Mason 	 */
15507585717fSChris Mason 	mutex_lock(&root->fs_info->reloc_mutex);
15517585717fSChris Mason 
155242874b3dSMiao Xie 	/*
155342874b3dSMiao Xie 	 * We needn't worry about the delayed items because we will
155442874b3dSMiao Xie 	 * deal with them in create_pending_snapshot(), which is the
155542874b3dSMiao Xie 	 * core function of the snapshot creation.
155642874b3dSMiao Xie 	 */
155742874b3dSMiao Xie 	ret = create_pending_snapshots(trans, root->fs_info);
155849b25e05SJeff Mahoney 	if (ret) {
155949b25e05SJeff Mahoney 		mutex_unlock(&root->fs_info->reloc_mutex);
156049b25e05SJeff Mahoney 		goto cleanup_transaction;
156149b25e05SJeff Mahoney 	}
15623063d29fSChris Mason 
156342874b3dSMiao Xie 	/*
156442874b3dSMiao Xie 	 * We insert the dir indexes of the snapshots and update the inode
156542874b3dSMiao Xie 	 * of the snapshots' parents after the snapshot creation, so there
156642874b3dSMiao Xie 	 * are some delayed items which are not dealt with. Now deal with
156742874b3dSMiao Xie 	 * them.
156842874b3dSMiao Xie 	 *
156942874b3dSMiao Xie 	 * We needn't worry that this operation will corrupt the snapshots,
157042874b3dSMiao Xie 	 * because all the tree which are snapshoted will be forced to COW
157142874b3dSMiao Xie 	 * the nodes and leaves.
157242874b3dSMiao Xie 	 */
157342874b3dSMiao Xie 	ret = btrfs_run_delayed_items(trans, root);
157449b25e05SJeff Mahoney 	if (ret) {
157549b25e05SJeff Mahoney 		mutex_unlock(&root->fs_info->reloc_mutex);
157649b25e05SJeff Mahoney 		goto cleanup_transaction;
157749b25e05SJeff Mahoney 	}
157816cdcec7SMiao Xie 
157956bec294SChris Mason 	ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
158049b25e05SJeff Mahoney 	if (ret) {
158149b25e05SJeff Mahoney 		mutex_unlock(&root->fs_info->reloc_mutex);
158249b25e05SJeff Mahoney 		goto cleanup_transaction;
158349b25e05SJeff Mahoney 	}
158456bec294SChris Mason 
1585e999376fSChris Mason 	/*
1586e999376fSChris Mason 	 * make sure none of the code above managed to slip in a
1587e999376fSChris Mason 	 * delayed item
1588e999376fSChris Mason 	 */
1589e999376fSChris Mason 	btrfs_assert_delayed_root_empty(root);
1590e999376fSChris Mason 
15912c90e5d6SChris Mason 	WARN_ON(cur_trans != trans->transaction);
1592dc17ff8fSChris Mason 
1593a2de733cSArne Jansen 	btrfs_scrub_pause(root);
1594e02119d5SChris Mason 	/* btrfs_commit_tree_roots is responsible for getting the
1595e02119d5SChris Mason 	 * various roots consistent with each other.  Every pointer
1596e02119d5SChris Mason 	 * in the tree of tree roots has to point to the most up to date
1597e02119d5SChris Mason 	 * root for every subvolume and other tree.  So, we have to keep
1598e02119d5SChris Mason 	 * the tree logging code from jumping in and changing any
1599e02119d5SChris Mason 	 * of the trees.
1600e02119d5SChris Mason 	 *
1601e02119d5SChris Mason 	 * At this point in the commit, there can't be any tree-log
1602e02119d5SChris Mason 	 * writers, but a little lower down we drop the trans mutex
1603e02119d5SChris Mason 	 * and let new people in.  By holding the tree_log_mutex
1604e02119d5SChris Mason 	 * from now until after the super is written, we avoid races
1605e02119d5SChris Mason 	 * with the tree-log code.
1606e02119d5SChris Mason 	 */
1607e02119d5SChris Mason 	mutex_lock(&root->fs_info->tree_log_mutex);
16081a40e23bSZheng Yan 
16095d4f98a2SYan Zheng 	ret = commit_fs_roots(trans, root);
161049b25e05SJeff Mahoney 	if (ret) {
161149b25e05SJeff Mahoney 		mutex_unlock(&root->fs_info->tree_log_mutex);
1612871383beSDavid Sterba 		mutex_unlock(&root->fs_info->reloc_mutex);
161349b25e05SJeff Mahoney 		goto cleanup_transaction;
161449b25e05SJeff Mahoney 	}
161554aa1f4dSChris Mason 
16165d4f98a2SYan Zheng 	/* commit_fs_roots gets rid of all the tree log roots, it is now
1617e02119d5SChris Mason 	 * safe to free the root of tree log roots
1618e02119d5SChris Mason 	 */
1619e02119d5SChris Mason 	btrfs_free_log_root_tree(trans, root->fs_info);
1620e02119d5SChris Mason 
16215d4f98a2SYan Zheng 	ret = commit_cowonly_roots(trans, root);
162249b25e05SJeff Mahoney 	if (ret) {
162349b25e05SJeff Mahoney 		mutex_unlock(&root->fs_info->tree_log_mutex);
1624871383beSDavid Sterba 		mutex_unlock(&root->fs_info->reloc_mutex);
162549b25e05SJeff Mahoney 		goto cleanup_transaction;
162649b25e05SJeff Mahoney 	}
162754aa1f4dSChris Mason 
162811833d66SYan Zheng 	btrfs_prepare_extent_commit(trans, root);
162911833d66SYan Zheng 
163078fae27eSChris Mason 	cur_trans = root->fs_info->running_transaction;
16315f39d397SChris Mason 
16325d4f98a2SYan Zheng 	btrfs_set_root_node(&root->fs_info->tree_root->root_item,
16335d4f98a2SYan Zheng 			    root->fs_info->tree_root->node);
1634817d52f8SJosef Bacik 	switch_commit_root(root->fs_info->tree_root);
16355d4f98a2SYan Zheng 
16365d4f98a2SYan Zheng 	btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
16375d4f98a2SYan Zheng 			    root->fs_info->chunk_root->node);
1638817d52f8SJosef Bacik 	switch_commit_root(root->fs_info->chunk_root);
16395d4f98a2SYan Zheng 
1640edf39272SJan Schmidt 	assert_qgroups_uptodate(trans);
16415d4f98a2SYan Zheng 	update_super_roots(root);
1642e02119d5SChris Mason 
1643e02119d5SChris Mason 	if (!root->fs_info->log_root_recovering) {
16446c41761fSDavid Sterba 		btrfs_set_super_log_root(root->fs_info->super_copy, 0);
16456c41761fSDavid Sterba 		btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1646e02119d5SChris Mason 	}
1647e02119d5SChris Mason 
16486c41761fSDavid Sterba 	memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
16496c41761fSDavid Sterba 	       sizeof(*root->fs_info->super_copy));
1650ccd467d6SChris Mason 
1651f9295749SChris Mason 	trans->transaction->blocked = 0;
1652a4abeea4SJosef Bacik 	spin_lock(&root->fs_info->trans_lock);
1653a4abeea4SJosef Bacik 	root->fs_info->running_transaction = NULL;
1654a4abeea4SJosef Bacik 	root->fs_info->trans_no_join = 0;
1655a4abeea4SJosef Bacik 	spin_unlock(&root->fs_info->trans_lock);
16567585717fSChris Mason 	mutex_unlock(&root->fs_info->reloc_mutex);
1657b7ec40d7SChris Mason 
1658f9295749SChris Mason 	wake_up(&root->fs_info->transaction_wait);
1659e6dcd2dcSChris Mason 
166079154b1bSChris Mason 	ret = btrfs_write_and_wait_transaction(trans, root);
166149b25e05SJeff Mahoney 	if (ret) {
166249b25e05SJeff Mahoney 		btrfs_error(root->fs_info, ret,
166349b25e05SJeff Mahoney 			    "Error while writing out transaction.");
166449b25e05SJeff Mahoney 		mutex_unlock(&root->fs_info->tree_log_mutex);
166549b25e05SJeff Mahoney 		goto cleanup_transaction;
166649b25e05SJeff Mahoney 	}
166749b25e05SJeff Mahoney 
166849b25e05SJeff Mahoney 	ret = write_ctree_super(trans, root, 0);
166949b25e05SJeff Mahoney 	if (ret) {
167049b25e05SJeff Mahoney 		mutex_unlock(&root->fs_info->tree_log_mutex);
167149b25e05SJeff Mahoney 		goto cleanup_transaction;
167249b25e05SJeff Mahoney 	}
16734313b399SChris Mason 
1674e02119d5SChris Mason 	/*
1675e02119d5SChris Mason 	 * the super is written, we can safely allow the tree-loggers
1676e02119d5SChris Mason 	 * to go about their business
1677e02119d5SChris Mason 	 */
1678e02119d5SChris Mason 	mutex_unlock(&root->fs_info->tree_log_mutex);
1679e02119d5SChris Mason 
168011833d66SYan Zheng 	btrfs_finish_extent_commit(trans, root);
16814313b399SChris Mason 
16822c90e5d6SChris Mason 	cur_trans->commit_done = 1;
1683b7ec40d7SChris Mason 
168415ee9bc7SJosef Bacik 	root->fs_info->last_trans_committed = cur_trans->transid;
1685817d52f8SJosef Bacik 
16862c90e5d6SChris Mason 	wake_up(&cur_trans->commit_wait);
16873de4586cSChris Mason 
1688a4abeea4SJosef Bacik 	spin_lock(&root->fs_info->trans_lock);
168913c5a93eSJosef Bacik 	list_del_init(&cur_trans->list);
1690a4abeea4SJosef Bacik 	spin_unlock(&root->fs_info->trans_lock);
1691a4abeea4SJosef Bacik 
169279154b1bSChris Mason 	put_transaction(cur_trans);
169378fae27eSChris Mason 	put_transaction(cur_trans);
169458176a96SJosef Bacik 
1695354aa0fbSMiao Xie 	if (trans->type < TRANS_JOIN_NOLOCK)
1696b2b5ef5cSJan Kara 		sb_end_intwrite(root->fs_info->sb);
1697b2b5ef5cSJan Kara 
16981abe9b8aSliubo 	trace_btrfs_transaction_commit(root);
16991abe9b8aSliubo 
1700a2de733cSArne Jansen 	btrfs_scrub_continue(root);
1701a2de733cSArne Jansen 
17029ed74f2dSJosef Bacik 	if (current->journal_info == trans)
17039ed74f2dSJosef Bacik 		current->journal_info = NULL;
17049ed74f2dSJosef Bacik 
17052c90e5d6SChris Mason 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
170624bbcf04SYan, Zheng 
170724bbcf04SYan, Zheng 	if (current != root->fs_info->transaction_kthread)
170824bbcf04SYan, Zheng 		btrfs_run_delayed_iputs(root);
170924bbcf04SYan, Zheng 
171079154b1bSChris Mason 	return ret;
171149b25e05SJeff Mahoney 
171249b25e05SJeff Mahoney cleanup_transaction:
17130e721106SJosef Bacik 	btrfs_trans_release_metadata(trans, root);
17140e721106SJosef Bacik 	trans->block_rsv = NULL;
171549b25e05SJeff Mahoney 	btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
171649b25e05SJeff Mahoney //	WARN_ON(1);
171749b25e05SJeff Mahoney 	if (current->journal_info == trans)
171849b25e05SJeff Mahoney 		current->journal_info = NULL;
17197b8b92afSJosef Bacik 	cleanup_transaction(trans, root, ret);
172049b25e05SJeff Mahoney 
172149b25e05SJeff Mahoney 	return ret;
172279154b1bSChris Mason }
172379154b1bSChris Mason 
1724d352ac68SChris Mason /*
1725d352ac68SChris Mason  * interface function to delete all the snapshots we have scheduled for deletion
1726d352ac68SChris Mason  */
1727e9d0b13bSChris Mason int btrfs_clean_old_snapshots(struct btrfs_root *root)
1728e9d0b13bSChris Mason {
17295d4f98a2SYan Zheng 	LIST_HEAD(list);
17305d4f98a2SYan Zheng 	struct btrfs_fs_info *fs_info = root->fs_info;
1731e9d0b13bSChris Mason 
1732a4abeea4SJosef Bacik 	spin_lock(&fs_info->trans_lock);
17335d4f98a2SYan Zheng 	list_splice_init(&fs_info->dead_roots, &list);
1734a4abeea4SJosef Bacik 	spin_unlock(&fs_info->trans_lock);
17355d4f98a2SYan Zheng 
17365d4f98a2SYan Zheng 	while (!list_empty(&list)) {
17372c536799SJeff Mahoney 		int ret;
17382c536799SJeff Mahoney 
17395d4f98a2SYan Zheng 		root = list_entry(list.next, struct btrfs_root, root_list);
174076dda93cSYan, Zheng 		list_del(&root->root_list);
174176dda93cSYan, Zheng 
174216cdcec7SMiao Xie 		btrfs_kill_all_delayed_nodes(root);
174316cdcec7SMiao Xie 
174476dda93cSYan, Zheng 		if (btrfs_header_backref_rev(root->node) <
174576dda93cSYan, Zheng 		    BTRFS_MIXED_BACKREF_REV)
17462c536799SJeff Mahoney 			ret = btrfs_drop_snapshot(root, NULL, 0, 0);
174776dda93cSYan, Zheng 		else
17482c536799SJeff Mahoney 			ret =btrfs_drop_snapshot(root, NULL, 1, 0);
17492c536799SJeff Mahoney 		BUG_ON(ret < 0);
1750e9d0b13bSChris Mason 	}
1751e9d0b13bSChris Mason 	return 0;
1752e9d0b13bSChris Mason }
1753