blob: 3aafbde8b637f44c0622e8dc9bd224f369c2d065 [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
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
Chris Mason79154b12007-03-22 15:59:16 -040019#include <linux/fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Chris Mason34088782007-06-12 11:36:58 -040021#include <linux/sched.h>
Chris Masond3c2fdcf2007-09-17 10:58:06 -040022#include <linux/writeback.h>
Chris Mason5f39d392007-10-15 16:14:19 -040023#include <linux/pagemap.h>
Chris Mason5f2cc082008-11-07 18:22:45 -050024#include <linux/blkdev.h>
Alexander Block8ea05e32012-07-25 17:35:53 +020025#include <linux/uuid.h>
Chris Mason79154b12007-03-22 15:59:16 -040026#include "ctree.h"
27#include "disk-io.h"
28#include "transaction.h"
Chris Mason925baed2008-06-25 16:01:30 -040029#include "locking.h"
Chris Masone02119d2008-09-05 16:13:11 -040030#include "tree-log.h"
Li Zefan581bb052011-04-20 10:06:11 +080031#include "inode-map.h"
Stefan Behrens733f4fb2012-05-25 16:06:10 +020032#include "volumes.h"
Stefan Behrens8dabb742012-11-06 13:15:27 +010033#include "dev-replace.h"
Josef Bacikfcebe452014-05-13 17:30:47 -070034#include "qgroup.h"
Chris Mason79154b12007-03-22 15:59:16 -040035
Chris Mason0f7d52f2007-04-09 10:42:37 -040036#define BTRFS_ROOT_TRANS_TAG 0
37
Miao Xie4a9d8bd2013-05-17 03:53:43 +000038static 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
Josef Bacik724e2312013-09-30 11:36:38 -040061void btrfs_put_transaction(struct btrfs_transaction *transaction)
Chris Mason79154b12007-03-22 15:59:16 -040062{
Josef Bacik13c5a932011-04-11 15:45:29 -040063 WARN_ON(atomic_read(&transaction->use_count) == 0);
64 if (atomic_dec_and_test(&transaction->use_count)) {
Josef Bacika4abeea2011-04-11 17:25:13 -040065 BUG_ON(!list_empty(&transaction->list));
Liu Boc46effa2013-10-14 12:59:45 +080066 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
Josef Bacik6df9a952013-06-27 13:22:46 -040067 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 }
Chris Mason2c90e5d2007-04-02 10:50:19 -040075 kmem_cache_free(btrfs_transaction_cachep, transaction);
Chris Mason78fae272007-03-25 11:35:08 -040076 }
Chris Mason79154b12007-03-22 15:59:16 -040077}
78
Josef Bacik9e351cc2014-03-13 15:42:13 -040079static noinline void switch_commit_roots(struct btrfs_transaction *trans,
80 struct btrfs_fs_info *fs_info)
Josef Bacik817d52f2009-07-13 21:29:25 -040081{
Josef Bacik9e351cc2014-03-13 15:42:13 -040082 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);
Josef Bacik817d52f2009-07-13 21:29:25 -040094}
95
Miao Xie0860adf2013-05-15 07:48:27 +000096static 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
103static 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
110static 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
116static inline int extwriter_counter_read(struct btrfs_transaction *trans)
117{
118 return atomic_read(&trans->num_extwriters);
Miao Xie178260b2013-02-20 09:16:24 +0000119}
120
Chris Masond352ac62008-09-29 15:18:18 -0400121/*
122 * either allocate a new transaction or hop into the existing one
123 */
Miao Xie0860adf2013-05-15 07:48:27 +0000124static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
Chris Mason79154b12007-03-22 15:59:16 -0400125{
126 struct btrfs_transaction *cur_trans;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200127 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacika4abeea2011-04-11 17:25:13 -0400128
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200129 spin_lock(&fs_info->trans_lock);
Chris Masond43317d2011-11-06 03:26:19 -0500130loop:
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100131 /* The file system has been taken offline. No new transactions. */
Miao Xie87533c42013-01-29 10:14:48 +0000132 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200133 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100134 return -EROFS;
135 }
136
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200137 cur_trans = fs_info->running_transaction;
Josef Bacika4abeea2011-04-11 17:25:13 -0400138 if (cur_trans) {
David Sterba871383b2012-04-02 18:31:37 +0200139 if (cur_trans->aborted) {
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200140 spin_unlock(&fs_info->trans_lock);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100141 return cur_trans->aborted;
David Sterba871383b2012-04-02 18:31:37 +0200142 }
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000143 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
Miao Xie178260b2013-02-20 09:16:24 +0000144 spin_unlock(&fs_info->trans_lock);
145 return -EBUSY;
146 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400147 atomic_inc(&cur_trans->use_count);
Josef Bacik13c5a932011-04-11 15:45:29 -0400148 atomic_inc(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000149 extwriter_counter_inc(cur_trans, type);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200150 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400151 return 0;
Chris Mason79154b12007-03-22 15:59:16 -0400152 }
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200153 spin_unlock(&fs_info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400154
Miao Xie354aa0f2012-09-20 01:54:00 -0600155 /*
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
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000162 /*
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
Josef Bacika4abeea2011-04-11 17:25:13 -0400168 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
169 if (!cur_trans)
170 return -ENOMEM;
Chris Masond43317d2011-11-06 03:26:19 -0500171
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200172 spin_lock(&fs_info->trans_lock);
173 if (fs_info->running_transaction) {
Chris Masond43317d2011-11-06 03:26:19 -0500174 /*
175 * someone started a transaction after we unlocked. Make sure
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000176 * to redo the checks above
Chris Masond43317d2011-11-06 03:26:19 -0500177 */
Josef Bacika4abeea2011-04-11 17:25:13 -0400178 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
Chris Masond43317d2011-11-06 03:26:19 -0500179 goto loop;
Miao Xie87533c42013-01-29 10:14:48 +0000180 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
Dan Carpentere4b50e12012-06-19 13:30:11 +0300181 spin_unlock(&fs_info->trans_lock);
Josef Bacik7b8b92a2012-05-31 15:52:43 -0400182 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
183 return -EROFS;
Josef Bacika4abeea2011-04-11 17:25:13 -0400184 }
Chris Masond43317d2011-11-06 03:26:19 -0500185
Josef Bacika4abeea2011-04-11 17:25:13 -0400186 atomic_set(&cur_trans->num_writers, 1);
Miao Xie0860adf2013-05-15 07:48:27 +0000187 extwriter_counter_init(cur_trans, type);
Josef Bacika4abeea2011-04-11 17:25:13 -0400188 init_waitqueue_head(&cur_trans->writer_wait);
189 init_waitqueue_head(&cur_trans->commit_wait);
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000190 cur_trans->state = TRANS_STATE_RUNNING;
Josef Bacika4abeea2011-04-11 17:25:13 -0400191 /*
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);
Josef Bacika4abeea2011-04-11 17:25:13 -0400196 cur_trans->start_time = get_seconds();
197
Liu Boc46effa2013-10-14 12:59:45 +0800198 cur_trans->delayed_refs.href_root = RB_ROOT;
Josef Bacikd7df2c72014-01-23 09:21:38 -0500199 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
Josef Bacika4abeea2011-04-11 17:25:13 -0400200 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;
Jan Schmidt20b297d2012-05-20 15:43:53 +0200204
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();
Julia Lawall31b1a2b2012-11-03 10:58:34 +0000210 if (!list_empty(&fs_info->tree_mod_seq_list))
Frank Holtonefe120a2013-12-20 11:37:06 -0500211 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
Jan Schmidt20b297d2012-05-20 15:43:53 +0200212 "creating a fresh transaction\n");
Julia Lawall31b1a2b2012-11-03 10:58:34 +0000213 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
Frank Holtonefe120a2013-12-20 11:37:06 -0500214 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
Jan Schmidt20b297d2012-05-20 15:43:53 +0200215 "creating a fresh transaction\n");
Jan Schmidtfc36ed7e2013-04-24 16:57:33 +0000216 atomic64_set(&fs_info->tree_mod_seq, 0);
Jan Schmidt20b297d2012-05-20 15:43:53 +0200217
Josef Bacika4abeea2011-04-11 17:25:13 -0400218 spin_lock_init(&cur_trans->delayed_refs.lock);
219
220 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
Josef Bacik569e0f32013-02-13 11:09:14 -0500221 INIT_LIST_HEAD(&cur_trans->ordered_operations);
Josef Bacik6df9a952013-06-27 13:22:46 -0400222 INIT_LIST_HEAD(&cur_trans->pending_chunks);
Josef Bacik9e351cc2014-03-13 15:42:13 -0400223 INIT_LIST_HEAD(&cur_trans->switch_commits);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200224 list_add_tail(&cur_trans->list, &fs_info->trans_list);
Josef Bacika4abeea2011-04-11 17:25:13 -0400225 extent_io_tree_init(&cur_trans->dirty_pages,
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200226 fs_info->btree_inode->i_mapping);
227 fs_info->generation++;
228 cur_trans->transid = fs_info->generation;
229 fs_info->running_transaction = cur_trans;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100230 cur_trans->aborted = 0;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200231 spin_unlock(&fs_info->trans_lock);
Josef Bacik15ee9bc2007-08-10 16:22:09 -0400232
Chris Mason79154b12007-03-22 15:59:16 -0400233 return 0;
234}
235
Chris Masond352ac62008-09-29 15:18:18 -0400236/*
Chris Masond3977122009-01-05 21:25:51 -0500237 * this does all the record keeping required to make sure that a reference
238 * counted root is properly recorded in a given transaction. This is required
239 * to make sure the old root from before we joined the transaction is deleted
240 * when the transaction commits
Chris Masond352ac62008-09-29 15:18:18 -0400241 */
Chris Mason75857172011-06-13 20:00:16 -0400242static int record_root_in_trans(struct btrfs_trans_handle *trans,
Josef Bacika4abeea2011-04-11 17:25:13 -0400243 struct btrfs_root *root)
Chris Mason6702ed42007-08-07 16:15:09 -0400244{
Miao Xie27cdeb72014-04-02 19:51:05 +0800245 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
246 root->last_trans < trans->transid) {
Chris Mason6702ed42007-08-07 16:15:09 -0400247 WARN_ON(root == root->fs_info->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400248 WARN_ON(root->commit_root != root->node);
Yan Zheng31153d82008-07-28 15:32:19 -0400249
Chris Mason75857172011-06-13 20:00:16 -0400250 /*
Miao Xie27cdeb72014-04-02 19:51:05 +0800251 * see below for IN_TRANS_SETUP usage rules
Chris Mason75857172011-06-13 20:00:16 -0400252 * we have the reloc mutex held now, so there
253 * is only one writer in this function
254 */
Miao Xie27cdeb72014-04-02 19:51:05 +0800255 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
Chris Mason75857172011-06-13 20:00:16 -0400256
Miao Xie27cdeb72014-04-02 19:51:05 +0800257 /* make sure readers find IN_TRANS_SETUP before
Chris Mason75857172011-06-13 20:00:16 -0400258 * they find our root->last_trans update
259 */
260 smp_wmb();
261
Josef Bacika4abeea2011-04-11 17:25:13 -0400262 spin_lock(&root->fs_info->fs_roots_radix_lock);
263 if (root->last_trans == trans->transid) {
264 spin_unlock(&root->fs_info->fs_roots_radix_lock);
265 return 0;
266 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400267 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
268 (unsigned long)root->root_key.objectid,
269 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -0400270 spin_unlock(&root->fs_info->fs_roots_radix_lock);
Chris Mason75857172011-06-13 20:00:16 -0400271 root->last_trans = trans->transid;
272
273 /* this is pretty tricky. We don't want to
274 * take the relocation lock in btrfs_record_root_in_trans
275 * unless we're really doing the first setup for this root in
276 * this transaction.
277 *
278 * Normally we'd use root->last_trans as a flag to decide
279 * if we want to take the expensive mutex.
280 *
281 * But, we have to set root->last_trans before we
282 * init the relocation root, otherwise, we trip over warnings
283 * in ctree.c. The solution used here is to flag ourselves
Miao Xie27cdeb72014-04-02 19:51:05 +0800284 * with root IN_TRANS_SETUP. When this is 1, we're still
Chris Mason75857172011-06-13 20:00:16 -0400285 * fixing up the reloc trees and everyone must wait.
286 *
287 * When this is zero, they can trust root->last_trans and fly
288 * through btrfs_record_root_in_trans without having to take the
289 * lock. smp_wmb() makes sure that all the writes above are
290 * done before we pop in the zero below
291 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400292 btrfs_init_reloc_root(trans, root);
Miao Xie27cdeb72014-04-02 19:51:05 +0800293 smp_mb__before_clear_bit();
294 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
Chris Mason6702ed42007-08-07 16:15:09 -0400295 }
296 return 0;
297}
298
Chris Mason75857172011-06-13 20:00:16 -0400299
300int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
301 struct btrfs_root *root)
302{
Miao Xie27cdeb72014-04-02 19:51:05 +0800303 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
Chris Mason75857172011-06-13 20:00:16 -0400304 return 0;
305
306 /*
Miao Xie27cdeb72014-04-02 19:51:05 +0800307 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
Chris Mason75857172011-06-13 20:00:16 -0400308 * and barriers
309 */
310 smp_rmb();
311 if (root->last_trans == trans->transid &&
Miao Xie27cdeb72014-04-02 19:51:05 +0800312 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
Chris Mason75857172011-06-13 20:00:16 -0400313 return 0;
314
315 mutex_lock(&root->fs_info->reloc_mutex);
316 record_root_in_trans(trans, root);
317 mutex_unlock(&root->fs_info->reloc_mutex);
318
319 return 0;
320}
321
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000322static inline int is_transaction_blocked(struct btrfs_transaction *trans)
323{
324 return (trans->state >= TRANS_STATE_BLOCKED &&
Josef Bacik501407a2013-06-10 16:47:23 -0400325 trans->state < TRANS_STATE_UNBLOCKED &&
326 !trans->aborted);
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000327}
328
Chris Masond352ac62008-09-29 15:18:18 -0400329/* wait for commit against the current transaction to become unblocked
330 * when this is done, it is safe to start a new transaction, but the current
331 * transaction might not be fully on disk.
332 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400333static void wait_current_trans(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400334{
Chris Masonf9295742008-07-17 12:54:14 -0400335 struct btrfs_transaction *cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400336
Josef Bacika4abeea2011-04-11 17:25:13 -0400337 spin_lock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400338 cur_trans = root->fs_info->running_transaction;
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000339 if (cur_trans && is_transaction_blocked(cur_trans)) {
Josef Bacik13c5a932011-04-11 15:45:29 -0400340 atomic_inc(&cur_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -0400341 spin_unlock(&root->fs_info->trans_lock);
Li Zefan72d63ed2011-07-14 03:17:00 +0000342
343 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -0400344 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
345 cur_trans->aborted);
Josef Bacik724e2312013-09-30 11:36:38 -0400346 btrfs_put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400347 } else {
348 spin_unlock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400349 }
Chris Mason37d1aee2008-07-31 10:48:37 -0400350}
351
Yan, Zhenga22285a2010-05-16 10:48:46 -0400352static int may_wait_transaction(struct btrfs_root *root, int type)
Chris Mason37d1aee2008-07-31 10:48:37 -0400353{
Josef Bacika4abeea2011-04-11 17:25:13 -0400354 if (root->fs_info->log_root_recovering)
355 return 0;
356
357 if (type == TRANS_USERSPACE)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400358 return 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400359
360 if (type == TRANS_START &&
361 !atomic_read(&root->fs_info->open_ioctl_trans))
362 return 1;
363
Yan, Zhenga22285a2010-05-16 10:48:46 -0400364 return 0;
365}
366
Miao Xie20dd2cb2013-09-25 21:47:45 +0800367static inline bool need_reserve_reloc_root(struct btrfs_root *root)
368{
369 if (!root->fs_info->reloc_ctl ||
Miao Xie27cdeb72014-04-02 19:51:05 +0800370 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
Miao Xie20dd2cb2013-09-25 21:47:45 +0800371 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
372 root->reloc_root)
373 return false;
374
375 return true;
376}
377
Miao Xie08e007d2012-10-16 11:33:38 +0000378static struct btrfs_trans_handle *
Miao Xie0860adf2013-05-15 07:48:27 +0000379start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
Miao Xie08e007d2012-10-16 11:33:38 +0000380 enum btrfs_reserve_flush_enum flush)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400381{
382 struct btrfs_trans_handle *h;
383 struct btrfs_transaction *cur_trans;
Josef Bacikb5009942011-06-07 15:07:51 -0400384 u64 num_bytes = 0;
Arne Jansenc5567232011-09-14 15:44:05 +0200385 u64 qgroup_reserved = 0;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800386 bool reloc_reserved = false;
387 int ret;
liuboacce9522011-01-06 19:30:25 +0800388
Miao Xie87533c42013-01-29 10:14:48 +0000389 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
liuboacce9522011-01-06 19:30:25 +0800390 return ERR_PTR(-EROFS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400391
Josef Bacika26e8c92014-03-28 17:07:27 -0400392 if (current->journal_info &&
393 current->journal_info != (void *)BTRFS_SEND_TRANS_STUB) {
Miao Xie0860adf2013-05-15 07:48:27 +0000394 WARN_ON(type & TRANS_EXTWRITERS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400395 h = current->journal_info;
396 h->use_count++;
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000397 WARN_ON(h->use_count > 2);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400398 h->orig_rsv = h->block_rsv;
399 h->block_rsv = NULL;
400 goto got_it;
401 }
Josef Bacikb5009942011-06-07 15:07:51 -0400402
403 /*
404 * Do the reservation before we join the transaction so we can do all
405 * the appropriate flushing if need be.
406 */
407 if (num_items > 0 && root != root->fs_info->chunk_root) {
Arne Jansenc5567232011-09-14 15:44:05 +0200408 if (root->fs_info->quota_enabled &&
409 is_fstree(root->root_key.objectid)) {
410 qgroup_reserved = num_items * root->leafsize;
411 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
412 if (ret)
413 return ERR_PTR(ret);
414 }
415
Josef Bacikb5009942011-06-07 15:07:51 -0400416 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
Miao Xie20dd2cb2013-09-25 21:47:45 +0800417 /*
418 * Do the reservation for the relocation root creation
419 */
420 if (unlikely(need_reserve_reloc_root(root))) {
421 num_bytes += root->nodesize;
422 reloc_reserved = true;
423 }
424
Miao Xie08e007d2012-10-16 11:33:38 +0000425 ret = btrfs_block_rsv_add(root,
426 &root->fs_info->trans_block_rsv,
427 num_bytes, flush);
Josef Bacikb5009942011-06-07 15:07:51 -0400428 if (ret)
Miao Xie843fcf32013-01-28 12:36:22 +0000429 goto reserve_fail;
Josef Bacikb5009942011-06-07 15:07:51 -0400430 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400431again:
432 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
Miao Xie843fcf32013-01-28 12:36:22 +0000433 if (!h) {
434 ret = -ENOMEM;
435 goto alloc_fail;
436 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400437
Josef Bacik98114652012-09-14 11:22:38 -0400438 /*
439 * If we are JOIN_NOLOCK we're already committing a transaction and
440 * waiting on this guy, so we don't need to do the sb_start_intwrite
441 * because we're already holding a ref. We need this because we could
442 * have raced in and did an fsync() on a file which can kick a commit
443 * and then we deadlock with somebody doing a freeze.
Miao Xie354aa0f2012-09-20 01:54:00 -0600444 *
445 * If we are ATTACH, it means we just want to catch the current
446 * transaction and commit it, so we needn't do sb_start_intwrite().
Josef Bacik98114652012-09-14 11:22:38 -0400447 */
Miao Xie0860adf2013-05-15 07:48:27 +0000448 if (type & __TRANS_FREEZABLE)
Josef Bacik60376ce2012-09-14 10:34:40 -0400449 sb_start_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +0200450
Yan, Zhenga22285a2010-05-16 10:48:46 -0400451 if (may_wait_transaction(root, type))
Chris Mason37d1aee2008-07-31 10:48:37 -0400452 wait_current_trans(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400453
Josef Bacika4abeea2011-04-11 17:25:13 -0400454 do {
Miao Xie354aa0f2012-09-20 01:54:00 -0600455 ret = join_transaction(root, type);
Miao Xie178260b2013-02-20 09:16:24 +0000456 if (ret == -EBUSY) {
Josef Bacika4abeea2011-04-11 17:25:13 -0400457 wait_current_trans(root);
Miao Xie178260b2013-02-20 09:16:24 +0000458 if (unlikely(type == TRANS_ATTACH))
459 ret = -ENOENT;
460 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400461 } while (ret == -EBUSY);
462
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000463 if (ret < 0) {
Miao Xie354aa0f2012-09-20 01:54:00 -0600464 /* We must get the transaction if we are JOIN_NOLOCK. */
465 BUG_ON(type == TRANS_JOIN_NOLOCK);
Miao Xie843fcf32013-01-28 12:36:22 +0000466 goto join_fail;
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000467 }
Chris Mason0f7d52f2007-04-09 10:42:37 -0400468
Yan, Zhenga22285a2010-05-16 10:48:46 -0400469 cur_trans = root->fs_info->running_transaction;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400470
471 h->transid = cur_trans->transid;
472 h->transaction = cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400473 h->blocks_used = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400474 h->bytes_reserved = 0;
Arne Jansend13603e2011-09-13 11:40:09 +0200475 h->root = root;
Chris Mason56bec292009-03-13 10:10:06 -0400476 h->delayed_ref_updates = 0;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400477 h->use_count = 1;
Josef Bacik0e721102012-06-26 16:13:18 -0400478 h->adding_csums = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -0400479 h->block_rsv = NULL;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400480 h->orig_rsv = NULL;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100481 h->aborted = 0;
Miao Xie4b824902013-02-20 09:13:32 +0000482 h->qgroup_reserved = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200483 h->delayed_ref_elem.seq = 0;
Miao Xiea698d0752012-09-20 01:51:59 -0600484 h->type = type;
Josef Bacikc6b305a2012-12-18 09:16:16 -0500485 h->allocating_chunk = false;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800486 h->reloc_reserved = false;
Josef Bacik5039edd2014-01-15 13:34:13 -0500487 h->sync = false;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200488 INIT_LIST_HEAD(&h->qgroup_ref_list);
Josef Bacikea658ba2012-09-11 16:57:25 -0400489 INIT_LIST_HEAD(&h->new_bgs);
Chris Masonb7ec40d2009-03-12 20:12:45 -0400490
Yan, Zhenga22285a2010-05-16 10:48:46 -0400491 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000492 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
493 may_wait_transaction(root, type)) {
Yan, Zhenga22285a2010-05-16 10:48:46 -0400494 btrfs_commit_transaction(h, root);
495 goto again;
496 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400497
Josef Bacikb5009942011-06-07 15:07:51 -0400498 if (num_bytes) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500499 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -0400500 h->transid, num_bytes, 1);
Josef Bacikb5009942011-06-07 15:07:51 -0400501 h->block_rsv = &root->fs_info->trans_block_rsv;
502 h->bytes_reserved = num_bytes;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800503 h->reloc_reserved = reloc_reserved;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400504 }
Miao Xie4b824902013-02-20 09:13:32 +0000505 h->qgroup_reserved = qgroup_reserved;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400506
Josef Bacik2a1eb462011-04-13 15:15:59 -0400507got_it:
Josef Bacika4abeea2011-04-11 17:25:13 -0400508 btrfs_record_root_in_trans(h, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400509
510 if (!current->journal_info && type != TRANS_USERSPACE)
511 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400512 return h;
Miao Xie843fcf32013-01-28 12:36:22 +0000513
514join_fail:
Miao Xie0860adf2013-05-15 07:48:27 +0000515 if (type & __TRANS_FREEZABLE)
Miao Xie843fcf32013-01-28 12:36:22 +0000516 sb_end_intwrite(root->fs_info->sb);
517 kmem_cache_free(btrfs_trans_handle_cachep, h);
518alloc_fail:
519 if (num_bytes)
520 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
521 num_bytes);
522reserve_fail:
523 if (qgroup_reserved)
524 btrfs_qgroup_free(root, qgroup_reserved);
525 return ERR_PTR(ret);
Chris Mason79154b12007-03-22 15:59:16 -0400526}
527
Chris Masonf9295742008-07-17 12:54:14 -0400528struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400529 int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400530{
Miao Xie08e007d2012-10-16 11:33:38 +0000531 return start_transaction(root, num_items, TRANS_START,
532 BTRFS_RESERVE_FLUSH_ALL);
Chris Masonf9295742008-07-17 12:54:14 -0400533}
Miao Xie8407aa42012-09-07 01:43:32 -0600534
Miao Xie08e007d2012-10-16 11:33:38 +0000535struct btrfs_trans_handle *btrfs_start_transaction_lflush(
Miao Xie8407aa42012-09-07 01:43:32 -0600536 struct btrfs_root *root, int num_items)
537{
Miao Xie08e007d2012-10-16 11:33:38 +0000538 return start_transaction(root, num_items, TRANS_START,
539 BTRFS_RESERVE_FLUSH_LIMIT);
Miao Xie8407aa42012-09-07 01:43:32 -0600540}
541
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400542struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
Chris Masonf9295742008-07-17 12:54:14 -0400543{
Miao Xie8407aa42012-09-07 01:43:32 -0600544 return start_transaction(root, 0, TRANS_JOIN, 0);
Chris Masonf9295742008-07-17 12:54:14 -0400545}
546
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400547struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
Josef Bacik0af3d002010-06-21 14:48:16 -0400548{
Miao Xie8407aa42012-09-07 01:43:32 -0600549 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -0400550}
551
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400552struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
Sage Weil9ca9ee02008-08-04 10:41:27 -0400553{
Miao Xie8407aa42012-09-07 01:43:32 -0600554 return start_transaction(root, 0, TRANS_USERSPACE, 0);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400555}
556
Miao Xied4edf392013-02-20 09:17:06 +0000557/*
558 * btrfs_attach_transaction() - catch the running transaction
559 *
560 * It is used when we want to commit the current the transaction, but
561 * don't want to start a new one.
562 *
563 * Note: If this function return -ENOENT, it just means there is no
564 * running transaction. But it is possible that the inactive transaction
565 * is still in the memory, not fully on disk. If you hope there is no
566 * inactive transaction in the fs when -ENOENT is returned, you should
567 * invoke
568 * btrfs_attach_transaction_barrier()
569 */
Miao Xie354aa0f2012-09-20 01:54:00 -0600570struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
Josef Bacik60376ce2012-09-14 10:34:40 -0400571{
Miao Xie354aa0f2012-09-20 01:54:00 -0600572 return start_transaction(root, 0, TRANS_ATTACH, 0);
Josef Bacik60376ce2012-09-14 10:34:40 -0400573}
574
Miao Xied4edf392013-02-20 09:17:06 +0000575/*
Wang Sheng-Hui90b6d282013-06-14 16:21:24 +0800576 * btrfs_attach_transaction_barrier() - catch the running transaction
Miao Xied4edf392013-02-20 09:17:06 +0000577 *
578 * It is similar to the above function, the differentia is this one
579 * will wait for all the inactive transactions until they fully
580 * complete.
581 */
582struct btrfs_trans_handle *
583btrfs_attach_transaction_barrier(struct btrfs_root *root)
584{
585 struct btrfs_trans_handle *trans;
586
587 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
588 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
589 btrfs_wait_for_commit(root, 0);
590
591 return trans;
592}
593
Chris Masond352ac62008-09-29 15:18:18 -0400594/* wait for a transaction commit to be fully complete */
Li Zefanb9c83002011-07-14 03:17:14 +0000595static noinline void wait_for_commit(struct btrfs_root *root,
Chris Mason89ce8a62008-06-25 16:01:31 -0400596 struct btrfs_transaction *commit)
597{
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000598 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
Chris Mason89ce8a62008-06-25 16:01:31 -0400599}
600
Sage Weil46204592010-10-29 15:41:32 -0400601int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
602{
603 struct btrfs_transaction *cur_trans = NULL, *t;
Miao Xie8cd28072012-11-26 08:42:07 +0000604 int ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400605
Sage Weil46204592010-10-29 15:41:32 -0400606 if (transid) {
607 if (transid <= root->fs_info->last_trans_committed)
Josef Bacika4abeea2011-04-11 17:25:13 -0400608 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400609
Miao Xie8cd28072012-11-26 08:42:07 +0000610 ret = -EINVAL;
Sage Weil46204592010-10-29 15:41:32 -0400611 /* find specified transaction */
Josef Bacika4abeea2011-04-11 17:25:13 -0400612 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400613 list_for_each_entry(t, &root->fs_info->trans_list, list) {
614 if (t->transid == transid) {
615 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400616 atomic_inc(&cur_trans->use_count);
Miao Xie8cd28072012-11-26 08:42:07 +0000617 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400618 break;
619 }
Miao Xie8cd28072012-11-26 08:42:07 +0000620 if (t->transid > transid) {
621 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400622 break;
Miao Xie8cd28072012-11-26 08:42:07 +0000623 }
Sage Weil46204592010-10-29 15:41:32 -0400624 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400625 spin_unlock(&root->fs_info->trans_lock);
Miao Xie8cd28072012-11-26 08:42:07 +0000626 /* The specified transaction doesn't exist */
Sage Weil46204592010-10-29 15:41:32 -0400627 if (!cur_trans)
Miao Xie8cd28072012-11-26 08:42:07 +0000628 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400629 } else {
630 /* find newest transaction that is committing | committed */
Josef Bacika4abeea2011-04-11 17:25:13 -0400631 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400632 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
633 list) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000634 if (t->state >= TRANS_STATE_COMMIT_START) {
635 if (t->state == TRANS_STATE_COMPLETED)
Josef Bacik3473f3c2011-06-09 10:15:17 -0400636 break;
Sage Weil46204592010-10-29 15:41:32 -0400637 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400638 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400639 break;
640 }
641 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400642 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400643 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400644 goto out; /* nothing committing|committed */
Sage Weil46204592010-10-29 15:41:32 -0400645 }
646
Sage Weil46204592010-10-29 15:41:32 -0400647 wait_for_commit(root, cur_trans);
Josef Bacik724e2312013-09-30 11:36:38 -0400648 btrfs_put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400649out:
Sage Weil46204592010-10-29 15:41:32 -0400650 return ret;
651}
652
Chris Mason37d1aee2008-07-31 10:48:37 -0400653void btrfs_throttle(struct btrfs_root *root)
654{
Josef Bacika4abeea2011-04-11 17:25:13 -0400655 if (!atomic_read(&root->fs_info->open_ioctl_trans))
Sage Weil9ca9ee02008-08-04 10:41:27 -0400656 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400657}
658
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400659static int should_end_transaction(struct btrfs_trans_handle *trans,
660 struct btrfs_root *root)
661{
Josef Bacik1be41b72013-06-12 13:56:06 -0400662 if (root->fs_info->global_block_rsv.space_info->full &&
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500663 btrfs_check_space_for_delayed_refs(trans, root))
Josef Bacik1be41b72013-06-12 13:56:06 -0400664 return 1;
Josef Bacik36ba0222011-10-18 12:15:48 -0400665
Josef Bacik1be41b72013-06-12 13:56:06 -0400666 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400667}
668
669int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
670 struct btrfs_root *root)
671{
672 struct btrfs_transaction *cur_trans = trans->transaction;
673 int updates;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100674 int err;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400675
Josef Bacika4abeea2011-04-11 17:25:13 -0400676 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000677 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
678 cur_trans->delayed_refs.flushing)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400679 return 1;
680
681 updates = trans->delayed_ref_updates;
682 trans->delayed_ref_updates = 0;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100683 if (updates) {
684 err = btrfs_run_delayed_refs(trans, root, updates);
685 if (err) /* Error code will also eval true */
686 return err;
687 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400688
689 return should_end_transaction(trans, root);
690}
691
Chris Mason89ce8a62008-06-25 16:01:31 -0400692static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Miao Xiea698d0752012-09-20 01:51:59 -0600693 struct btrfs_root *root, int throttle)
Chris Mason79154b12007-03-22 15:59:16 -0400694{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400695 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400696 struct btrfs_fs_info *info = root->fs_info;
Josef Bacik1be41b72013-06-12 13:56:06 -0400697 unsigned long cur = trans->delayed_ref_updates;
Miao Xiea698d0752012-09-20 01:51:59 -0600698 int lock = (trans->type != TRANS_JOIN_NOLOCK);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400699 int err = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400700
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500701 if (trans->use_count > 1) {
702 trans->use_count--;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400703 trans->block_rsv = trans->orig_rsv;
704 return 0;
705 }
706
Josef Bacikb24e03d2011-10-14 14:40:17 -0400707 btrfs_trans_release_metadata(trans, root);
Josef Bacik4c13d752011-08-30 11:31:29 -0400708 trans->block_rsv = NULL;
Arne Jansenc5567232011-09-14 15:44:05 +0200709
Josef Bacikea658ba2012-09-11 16:57:25 -0400710 if (!list_empty(&trans->new_bgs))
711 btrfs_create_pending_block_groups(trans, root);
712
Josef Bacik1be41b72013-06-12 13:56:06 -0400713 trans->delayed_ref_updates = 0;
Josef Bacik5039edd2014-01-15 13:34:13 -0500714 if (!trans->sync && btrfs_should_throttle_delayed_refs(trans, root)) {
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500715 cur = max_t(unsigned long, cur, 32);
Chris Masonc3e69d52009-03-13 10:17:05 -0400716 trans->delayed_ref_updates = 0;
Josef Bacik1be41b72013-06-12 13:56:06 -0400717 btrfs_run_delayed_refs(trans, root, cur);
Chris Mason56bec292009-03-13 10:10:06 -0400718 }
Chris Masonbb721702013-01-29 18:44:12 -0500719
Josef Bacikfcebe452014-05-13 17:30:47 -0700720 if (trans->qgroup_reserved) {
721 /*
722 * the same root has to be passed here between start_transaction
723 * and end_transaction. Subvolume quota depends on this.
724 */
725 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
726 trans->qgroup_reserved = 0;
727 }
728
Josef Bacik0e721102012-06-26 16:13:18 -0400729 btrfs_trans_release_metadata(trans, root);
730 trans->block_rsv = NULL;
Chris Mason56bec292009-03-13 10:10:06 -0400731
Josef Bacikea658ba2012-09-11 16:57:25 -0400732 if (!list_empty(&trans->new_bgs))
733 btrfs_create_pending_block_groups(trans, root);
734
Josef Bacika4abeea2011-04-11 17:25:13 -0400735 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000736 should_end_transaction(trans, root) &&
737 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
738 spin_lock(&info->trans_lock);
739 if (cur_trans->state == TRANS_STATE_RUNNING)
740 cur_trans->state = TRANS_STATE_BLOCKED;
741 spin_unlock(&info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400742 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400743
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000744 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500745 if (throttle)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400746 return btrfs_commit_transaction(trans, root);
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500747 else
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400748 wake_up_process(info->transaction_kthread);
749 }
750
Miao Xie0860adf2013-05-15 07:48:27 +0000751 if (trans->type & __TRANS_FREEZABLE)
Josef Bacik98114652012-09-14 11:22:38 -0400752 sb_end_intwrite(root->fs_info->sb);
Josef Bacik6df78812012-09-05 08:08:30 -0600753
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400754 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400755 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
756 atomic_dec(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000757 extwriter_counter_dec(cur_trans, trans->type);
Chris Mason89ce8a62008-06-25 16:01:31 -0400758
Sage Weil99d16cb2010-10-29 15:37:34 -0400759 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400760 if (waitqueue_active(&cur_trans->writer_wait))
761 wake_up(&cur_trans->writer_wait);
Josef Bacik724e2312013-09-30 11:36:38 -0400762 btrfs_put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400763
764 if (current->journal_info == trans)
765 current->journal_info = NULL;
Chris Masonab78c842008-07-29 16:15:18 -0400766
Yan, Zheng24bbcf042009-11-12 09:36:34 +0000767 if (throttle)
768 btrfs_run_delayed_iputs(root);
769
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100770 if (trans->aborted ||
Josef Bacik4e121c02013-09-27 16:32:39 -0400771 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
772 wake_up_process(info->transaction_kthread);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400773 err = -EIO;
Josef Bacik4e121c02013-09-27 16:32:39 -0400774 }
Jan Schmidtedf39272012-06-28 18:04:55 +0200775 assert_qgroups_uptodate(trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100776
Dave Jones4edc2ca2012-04-12 16:03:56 -0400777 kmem_cache_free(btrfs_trans_handle_cachep, trans);
778 return err;
Chris Mason79154b12007-03-22 15:59:16 -0400779}
780
Chris Mason89ce8a62008-06-25 16:01:31 -0400781int btrfs_end_transaction(struct btrfs_trans_handle *trans,
782 struct btrfs_root *root)
783{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000784 return __btrfs_end_transaction(trans, root, 0);
Chris Mason89ce8a62008-06-25 16:01:31 -0400785}
786
787int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
788 struct btrfs_root *root)
789{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000790 return __btrfs_end_transaction(trans, root, 1);
Miao Xie16cdcec2011-04-22 18:12:22 +0800791}
792
Chris Masond352ac62008-09-29 15:18:18 -0400793/*
794 * when btree blocks are allocated, they have some corresponding bits set for
795 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400796 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400797 */
Chris Mason690587d2009-10-13 13:29:19 -0400798int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000799 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400800{
Chris Mason777e6bd2008-08-15 15:34:15 -0400801 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400802 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400803 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400804 struct extent_state *cached_state = NULL;
Chris Mason777e6bd2008-08-15 15:34:15 -0400805 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400806 u64 end;
Chris Mason7c4452b2007-04-28 09:29:35 -0400807
Josef Bacik17283662011-09-26 13:58:47 -0400808 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400809 mark, &cached_state)) {
810 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
811 mark, &cached_state, GFP_NOFS);
812 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -0400813 err = filemap_fdatawrite_range(mapping, start, end);
814 if (err)
815 werr = err;
816 cond_resched();
817 start = end + 1;
Chris Mason7c4452b2007-04-28 09:29:35 -0400818 }
Chris Mason690587d2009-10-13 13:29:19 -0400819 if (err)
820 werr = err;
821 return werr;
822}
823
824/*
825 * when btree blocks are allocated, they have some corresponding bits set for
826 * them in one of two extent_io trees. This is used to make sure all of
827 * those extents are on disk for transaction or log commit. We wait
828 * on all the pages and clear them from the dirty pages state tree
829 */
830int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000831 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400832{
Chris Mason690587d2009-10-13 13:29:19 -0400833 int err = 0;
834 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400835 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400836 struct extent_state *cached_state = NULL;
Chris Mason690587d2009-10-13 13:29:19 -0400837 u64 start = 0;
838 u64 end;
Chris Mason690587d2009-10-13 13:29:19 -0400839
Josef Bacik17283662011-09-26 13:58:47 -0400840 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400841 EXTENT_NEED_WAIT, &cached_state)) {
842 clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
843 0, 0, &cached_state, GFP_NOFS);
Josef Bacik17283662011-09-26 13:58:47 -0400844 err = filemap_fdatawait_range(mapping, start, end);
845 if (err)
846 werr = err;
847 cond_resched();
848 start = end + 1;
Chris Mason777e6bd2008-08-15 15:34:15 -0400849 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400850 if (err)
851 werr = err;
852 return werr;
Chris Mason79154b12007-03-22 15:59:16 -0400853}
854
Chris Mason690587d2009-10-13 13:29:19 -0400855/*
856 * when btree blocks are allocated, they have some corresponding bits set for
857 * them in one of two extent_io trees. This is used to make sure all of
858 * those extents are on disk for transaction or log commit
859 */
Sergei Trofimovich171170c2013-08-14 23:27:46 +0300860static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000861 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400862{
863 int ret;
864 int ret2;
Miao Xiec6adc9c2013-05-28 10:05:39 +0000865 struct blk_plug plug;
Chris Mason690587d2009-10-13 13:29:19 -0400866
Miao Xiec6adc9c2013-05-28 10:05:39 +0000867 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000868 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
Miao Xiec6adc9c2013-05-28 10:05:39 +0000869 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000870 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Masonbf0da8c2011-11-04 12:29:37 -0400871
872 if (ret)
873 return ret;
874 if (ret2)
875 return ret2;
876 return 0;
Chris Mason690587d2009-10-13 13:29:19 -0400877}
878
Chris Masond0c803c2008-09-11 16:17:57 -0400879int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
880 struct btrfs_root *root)
881{
882 if (!trans || !trans->transaction) {
883 struct inode *btree_inode;
884 btree_inode = root->fs_info->btree_inode;
885 return filemap_write_and_wait(btree_inode->i_mapping);
886 }
887 return btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000888 &trans->transaction->dirty_pages,
889 EXTENT_DIRTY);
Chris Masond0c803c2008-09-11 16:17:57 -0400890}
891
Chris Masond352ac62008-09-29 15:18:18 -0400892/*
893 * this is used to update the root pointer in the tree of tree roots.
894 *
895 * But, in the case of the extent allocation tree, updating the root
896 * pointer may allocate blocks which may change the root of the extent
897 * allocation tree.
898 *
899 * So, this loops and repeats and makes sure the cowonly root didn't
900 * change while the root pointer was being updated in the metadata.
901 */
Chris Mason0b86a832008-03-24 15:01:56 -0400902static int update_cowonly_root(struct btrfs_trans_handle *trans,
903 struct btrfs_root *root)
904{
905 int ret;
906 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000907 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -0400908 struct btrfs_root *tree_root = root->fs_info->tree_root;
909
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000910 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -0400911 btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400912
Chris Masond3977122009-01-05 21:25:51 -0500913 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -0400914 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000915 if (old_root_bytenr == root->node->start &&
916 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -0400917 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400918
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400919 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -0400920 ret = btrfs_update_root(trans, tree_root,
921 &root->root_key,
922 &root->root_item);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100923 if (ret)
924 return ret;
Chris Mason56bec292009-03-13 10:10:06 -0400925
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000926 old_root_used = btrfs_root_used(&root->root_item);
Yan Zheng4a8c9a62009-07-22 10:07:05 -0400927 ret = btrfs_write_dirty_block_groups(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100928 if (ret)
929 return ret;
Chris Mason0b86a832008-03-24 15:01:56 -0400930 }
Yan Zheng276e6802009-07-30 09:40:40 -0400931
Chris Mason0b86a832008-03-24 15:01:56 -0400932 return 0;
933}
934
Chris Masond352ac62008-09-29 15:18:18 -0400935/*
936 * update all the cowonly tree roots on disk
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100937 *
938 * The error handling in this function may not be obvious. Any of the
939 * failures will cause the file system to go offline. We still need
940 * to clean up the delayed refs.
Chris Masond352ac62008-09-29 15:18:18 -0400941 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400942static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
943 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400944{
Chris Mason79154b12007-03-22 15:59:16 -0400945 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0b86a832008-03-24 15:01:56 -0400946 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -0400947 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -0400948 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -0400949
Chris Mason56bec292009-03-13 10:10:06 -0400950 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100951 if (ret)
952 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400953
Yan Zheng84234f32008-10-29 14:49:05 -0400954 eb = btrfs_lock_root_node(fs_info->tree_root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100955 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
956 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -0400957 btrfs_tree_unlock(eb);
958 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -0400959
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100960 if (ret)
961 return ret;
962
Chris Mason56bec292009-03-13 10:10:06 -0400963 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100964 if (ret)
965 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400966
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200967 ret = btrfs_run_dev_stats(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -0400968 if (ret)
969 return ret;
Stefan Behrens8dabb742012-11-06 13:15:27 +0100970 ret = btrfs_run_dev_replace(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -0400971 if (ret)
972 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +0200973 ret = btrfs_run_qgroups(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -0400974 if (ret)
975 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +0200976
977 /* run_qgroups might have added some more refs */
978 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Josef Bacikc16ce192013-09-27 16:38:20 -0400979 if (ret)
980 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +0200981
Chris Masond3977122009-01-05 21:25:51 -0500982 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -0400983 next = fs_info->dirty_cowonly_roots.next;
984 list_del_init(next);
985 root = list_entry(next, struct btrfs_root, dirty_list);
Chris Mason87ef2bb2008-10-30 11:23:27 -0400986
Josef Bacik9e351cc2014-03-13 15:42:13 -0400987 if (root != fs_info->extent_root)
988 list_add_tail(&root->dirty_list,
989 &trans->transaction->switch_commits);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100990 ret = update_cowonly_root(trans, root);
991 if (ret)
992 return ret;
Chris Mason79154b12007-03-22 15:59:16 -0400993 }
Yan Zheng276e6802009-07-30 09:40:40 -0400994
Josef Bacik9e351cc2014-03-13 15:42:13 -0400995 list_add_tail(&fs_info->extent_root->dirty_list,
996 &trans->transaction->switch_commits);
Stefan Behrens8dabb742012-11-06 13:15:27 +0100997 btrfs_after_dev_replace_commit(fs_info);
998
Chris Mason79154b12007-03-22 15:59:16 -0400999 return 0;
1000}
1001
Chris Masond352ac62008-09-29 15:18:18 -04001002/*
1003 * dead roots are old snapshots that need to be deleted. This allocates
1004 * a dirty root struct and adds it into the list of dead roots that need to
1005 * be deleted
1006 */
Josef Bacikcfad3922013-07-25 15:11:47 -04001007void btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -04001008{
Josef Bacika4abeea2011-04-11 17:25:13 -04001009 spin_lock(&root->fs_info->trans_lock);
Josef Bacikcfad3922013-07-25 15:11:47 -04001010 if (list_empty(&root->root_list))
1011 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
Josef Bacika4abeea2011-04-11 17:25:13 -04001012 spin_unlock(&root->fs_info->trans_lock);
Chris Mason5eda7b52007-06-22 14:16:25 -04001013}
1014
Chris Masond352ac62008-09-29 15:18:18 -04001015/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001016 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -04001017 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001018static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1019 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -04001020{
Chris Mason0f7d52f2007-04-09 10:42:37 -04001021 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001022 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001023 int i;
1024 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -04001025 int err = 0;
1026
Josef Bacika4abeea2011-04-11 17:25:13 -04001027 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Masond3977122009-01-05 21:25:51 -05001028 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001029 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1030 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001031 ARRAY_SIZE(gang),
1032 BTRFS_ROOT_TRANS_TAG);
1033 if (ret == 0)
1034 break;
1035 for (i = 0; i < ret; i++) {
1036 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001037 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1038 (unsigned long)root->root_key.objectid,
1039 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -04001040 spin_unlock(&fs_info->fs_roots_radix_lock);
Yan Zheng31153d82008-07-28 15:32:19 -04001041
Chris Masone02119d2008-09-05 16:13:11 -04001042 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001043 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001044 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001045
Li Zefan82d59022011-04-20 10:33:24 +08001046 btrfs_save_ino_cache(root, trans);
1047
Liu Bof1ebcc72011-11-14 20:48:06 -05001048 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001049 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1050 smp_mb__after_clear_bit();
Liu Bof1ebcc72011-11-14 20:48:06 -05001051
Yan Zheng978d9102009-06-15 20:01:02 -04001052 if (root->commit_root != root->node) {
Josef Bacik9e351cc2014-03-13 15:42:13 -04001053 list_add_tail(&root->dirty_list,
1054 &trans->transaction->switch_commits);
Yan Zheng978d9102009-06-15 20:01:02 -04001055 btrfs_set_root_node(&root->root_item,
1056 root->node);
1057 }
Chris Mason9f3a7422007-08-07 15:52:19 -04001058
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001059 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001060 &root->root_key,
1061 &root->root_item);
Josef Bacika4abeea2011-04-11 17:25:13 -04001062 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001063 if (err)
1064 break;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001065 }
1066 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001067 spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001068 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001069}
1070
Chris Masond352ac62008-09-29 15:18:18 -04001071/*
Eric Sandeende78b512013-01-31 18:21:12 +00001072 * defrag a given btree.
1073 * Every leaf in the btree is read and defragged.
Chris Masond352ac62008-09-29 15:18:18 -04001074 */
Eric Sandeende78b512013-01-31 18:21:12 +00001075int btrfs_defrag_root(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001076{
1077 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001078 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001079 int ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001080
Miao Xie27cdeb72014-04-02 19:51:05 +08001081 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
Chris Masone9d0b132007-08-10 14:06:19 -04001082 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001083
Chris Mason6b800532007-10-15 16:17:34 -04001084 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001085 trans = btrfs_start_transaction(root, 0);
1086 if (IS_ERR(trans))
1087 return PTR_ERR(trans);
1088
Eric Sandeende78b512013-01-31 18:21:12 +00001089 ret = btrfs_defrag_leaves(trans, root);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001090
Chris Masone9d0b132007-08-10 14:06:19 -04001091 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00001092 btrfs_btree_balance_dirty(info->tree_root);
Chris Masone9d0b132007-08-10 14:06:19 -04001093 cond_resched();
1094
David Sterba7841cb22011-05-31 18:07:27 +02001095 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -04001096 break;
David Sterba210549e2013-02-09 23:38:06 +00001097
1098 if (btrfs_defrag_cancelled(root->fs_info)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05001099 pr_debug("BTRFS: defrag_root cancelled\n");
David Sterba210549e2013-02-09 23:38:06 +00001100 ret = -EAGAIN;
1101 break;
1102 }
Chris Masone9d0b132007-08-10 14:06:19 -04001103 }
Miao Xie27cdeb72014-04-02 19:51:05 +08001104 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001105 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001106}
1107
Chris Masond352ac62008-09-29 15:18:18 -04001108/*
1109 * new snapshots need to be created at a very specific time in the
Miao Xieaec80302013-03-04 09:44:29 +00001110 * transaction commit. This does the actual creation.
1111 *
1112 * Note:
1113 * If the error which may affect the commitment of the current transaction
1114 * happens, we should return the error number. If the error which just affect
1115 * the creation of the pending snapshots, just return 0.
Chris Masond352ac62008-09-29 15:18:18 -04001116 */
Chris Mason80b67942008-02-01 16:35:04 -05001117static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -05001118 struct btrfs_fs_info *fs_info,
1119 struct btrfs_pending_snapshot *pending)
1120{
1121 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -05001122 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -05001123 struct btrfs_root *tree_root = fs_info->tree_root;
1124 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001125 struct btrfs_root *parent_root;
Liu Bo98c99422011-09-11 10:52:24 -04001126 struct btrfs_block_rsv *rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001127 struct inode *parent_inode;
Miao Xie42874b32012-09-06 04:03:32 -06001128 struct btrfs_path *path;
1129 struct btrfs_dir_item *dir_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001130 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -05001131 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -04001132 struct extent_buffer *old;
Alexander Block8ea05e32012-07-25 17:35:53 +02001133 struct timespec cur_time = CURRENT_TIME;
Miao Xieaec80302013-03-04 09:44:29 +00001134 int ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001135 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001136 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001137 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +08001138 u64 root_flags;
Alexander Block8ea05e32012-07-25 17:35:53 +02001139 uuid_le new_uuid;
Chris Mason3063d292008-01-08 15:46:30 -05001140
Miao Xie42874b32012-09-06 04:03:32 -06001141 path = btrfs_alloc_path();
1142 if (!path) {
Miao Xieaec80302013-03-04 09:44:29 +00001143 pending->error = -ENOMEM;
1144 return 0;
Miao Xie42874b32012-09-06 04:03:32 -06001145 }
1146
Chris Mason80b67942008-02-01 16:35:04 -05001147 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1148 if (!new_root_item) {
Miao Xieaec80302013-03-04 09:44:29 +00001149 pending->error = -ENOMEM;
Miao Xie6fa97002012-09-06 04:00:32 -06001150 goto root_item_alloc_fail;
Chris Mason80b67942008-02-01 16:35:04 -05001151 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001152
Miao Xieaec80302013-03-04 09:44:29 +00001153 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1154 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001155 goto no_free_objectid;
Chris Mason3063d292008-01-08 15:46:30 -05001156
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001157 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001158
1159 if (to_reserve > 0) {
Miao Xieaec80302013-03-04 09:44:29 +00001160 pending->error = btrfs_block_rsv_add(root,
1161 &pending->block_rsv,
1162 to_reserve,
1163 BTRFS_RESERVE_NO_FLUSH);
1164 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001165 goto no_free_objectid;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001166 }
1167
Chris Mason3063d292008-01-08 15:46:30 -05001168 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001169 key.offset = (u64)-1;
1170 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -05001171
Miao Xie6fa97002012-09-06 04:00:32 -06001172 rsv = trans->block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001173 trans->block_rsv = &pending->block_rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001174 trans->bytes_reserved = trans->block_rsv->reserved;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001175
Yan, Zhenga22285a2010-05-16 10:48:46 -04001176 dentry = pending->dentry;
Miao Xiee9662f72013-02-28 10:01:15 +00001177 parent_inode = pending->dir;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001178 parent_root = BTRFS_I(parent_inode)->root;
Chris Mason75857172011-06-13 20:00:16 -04001179 record_root_in_trans(trans, parent_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001180
Sage Weil6bdb72d2010-03-15 17:27:13 +00001181 /*
1182 * insert the directory item
1183 */
Sage Weil6bdb72d2010-03-15 17:27:13 +00001184 ret = btrfs_set_inode_index(parent_inode, &index);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001185 BUG_ON(ret); /* -ENOMEM */
Miao Xie42874b32012-09-06 04:03:32 -06001186
1187 /* check if there is a file/dir which has the same name. */
1188 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1189 btrfs_ino(parent_inode),
1190 dentry->d_name.name,
1191 dentry->d_name.len, 0);
1192 if (dir_item != NULL && !IS_ERR(dir_item)) {
Chris Masonfe66a052012-02-20 08:40:56 -05001193 pending->error = -EEXIST;
Miao Xieaec80302013-03-04 09:44:29 +00001194 goto dir_item_existed;
Miao Xie42874b32012-09-06 04:03:32 -06001195 } else if (IS_ERR(dir_item)) {
1196 ret = PTR_ERR(dir_item);
Miao Xie8732d442012-09-17 23:52:38 -06001197 btrfs_abort_transaction(trans, root, ret);
1198 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001199 }
Miao Xie42874b32012-09-06 04:03:32 -06001200 btrfs_release_path(path);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001201
Chris Masone9993762011-06-17 16:14:09 -04001202 /*
1203 * pull in the delayed directory update
1204 * and the delayed inode item
1205 * otherwise we corrupt the FS during
1206 * snapshot
1207 */
1208 ret = btrfs_run_delayed_items(trans, root);
Miao Xie8732d442012-09-17 23:52:38 -06001209 if (ret) { /* Transaction aborted */
1210 btrfs_abort_transaction(trans, root, ret);
1211 goto fail;
1212 }
Chris Masone9993762011-06-17 16:14:09 -04001213
Chris Mason75857172011-06-13 20:00:16 -04001214 record_root_in_trans(trans, root);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001215 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1216 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +00001217 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001218
Li Zefanb83cc962010-12-20 16:04:08 +08001219 root_flags = btrfs_root_flags(new_root_item);
1220 if (pending->readonly)
1221 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1222 else
1223 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1224 btrfs_set_root_flags(new_root_item, root_flags);
1225
Alexander Block8ea05e32012-07-25 17:35:53 +02001226 btrfs_set_root_generation_v2(new_root_item,
1227 trans->transid);
1228 uuid_le_gen(&new_uuid);
1229 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1230 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1231 BTRFS_UUID_SIZE);
Stefan Behrens70023da2013-04-17 09:11:47 +00001232 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1233 memset(new_root_item->received_uuid, 0,
1234 sizeof(new_root_item->received_uuid));
1235 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1236 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1237 btrfs_set_root_stransid(new_root_item, 0);
1238 btrfs_set_root_rtransid(new_root_item, 0);
1239 }
Qu Wenruo3cae2102013-07-16 11:19:18 +08001240 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1241 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
Alexander Block8ea05e32012-07-25 17:35:53 +02001242 btrfs_set_root_otransid(new_root_item, trans->transid);
Alexander Block8ea05e32012-07-25 17:35:53 +02001243
Chris Mason925baed2008-06-25 16:01:30 -04001244 old = btrfs_lock_root_node(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001245 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001246 if (ret) {
1247 btrfs_tree_unlock(old);
1248 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001249 btrfs_abort_transaction(trans, root, ret);
1250 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001251 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001252
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001253 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -05001254
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001255 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001256 /* clean up in any case */
Chris Mason925baed2008-06-25 16:01:30 -04001257 btrfs_tree_unlock(old);
1258 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001259 if (ret) {
1260 btrfs_abort_transaction(trans, root, ret);
1261 goto fail;
1262 }
Chris Mason3063d292008-01-08 15:46:30 -05001263
Josef Bacikfcebe452014-05-13 17:30:47 -07001264 /*
1265 * We need to flush delayed refs in order to make sure all of our quota
1266 * operations have been done before we call btrfs_qgroup_inherit.
1267 */
1268 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1269 if (ret) {
1270 btrfs_abort_transaction(trans, root, ret);
1271 goto fail;
1272 }
1273
1274 pending->error = btrfs_qgroup_inherit(trans, fs_info,
1275 root->root_key.objectid,
1276 objectid, pending->inherit);
1277 if (pending->error)
1278 goto no_free_objectid;
1279
Liu Bof1ebcc72011-11-14 20:48:06 -05001280 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001281 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Liu Bof1ebcc72011-11-14 20:48:06 -05001282 smp_wmb();
1283
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001284 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001285 /* record when the snapshot was created in key.offset */
1286 key.offset = trans->transid;
1287 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -04001288 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -05001289 free_extent_buffer(tmp);
Miao Xie8732d442012-09-17 23:52:38 -06001290 if (ret) {
1291 btrfs_abort_transaction(trans, root, ret);
1292 goto fail;
1293 }
Chris Mason0660b5a2008-11-17 20:37:39 -05001294
Yan, Zhenga22285a2010-05-16 10:48:46 -04001295 /*
1296 * insert root back/forward references
1297 */
1298 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1299 parent_root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001300 btrfs_ino(parent_inode), index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04001301 dentry->d_name.name, dentry->d_name.len);
Miao Xie8732d442012-09-17 23:52:38 -06001302 if (ret) {
1303 btrfs_abort_transaction(trans, root, ret);
1304 goto fail;
1305 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001306
1307 key.offset = (u64)-1;
1308 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001309 if (IS_ERR(pending->snap)) {
1310 ret = PTR_ERR(pending->snap);
Miao Xie8732d442012-09-17 23:52:38 -06001311 btrfs_abort_transaction(trans, root, ret);
1312 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001313 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04001314
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001315 ret = btrfs_reloc_post_snapshot(trans, pending);
Miao Xie8732d442012-09-17 23:52:38 -06001316 if (ret) {
1317 btrfs_abort_transaction(trans, root, ret);
1318 goto fail;
1319 }
Miao Xie361048f2012-09-06 04:00:57 -06001320
1321 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Miao Xie8732d442012-09-17 23:52:38 -06001322 if (ret) {
1323 btrfs_abort_transaction(trans, root, ret);
1324 goto fail;
1325 }
Miao Xie42874b32012-09-06 04:03:32 -06001326
1327 ret = btrfs_insert_dir_item(trans, parent_root,
1328 dentry->d_name.name, dentry->d_name.len,
1329 parent_inode, &key,
1330 BTRFS_FT_DIR, index);
1331 /* We have check then name at the beginning, so it is impossible. */
Chris Mason9c520572012-12-17 14:26:57 -05001332 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
Miao Xie8732d442012-09-17 23:52:38 -06001333 if (ret) {
1334 btrfs_abort_transaction(trans, root, ret);
1335 goto fail;
1336 }
Miao Xie42874b32012-09-06 04:03:32 -06001337
1338 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1339 dentry->d_name.len * 2);
1340 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
Josef Bacikbe6aef62012-10-22 15:43:12 -04001341 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001342 if (ret) {
Miao Xie8732d442012-09-17 23:52:38 -06001343 btrfs_abort_transaction(trans, root, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001344 goto fail;
1345 }
1346 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1347 BTRFS_UUID_KEY_SUBVOL, objectid);
1348 if (ret) {
1349 btrfs_abort_transaction(trans, root, ret);
1350 goto fail;
1351 }
1352 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1353 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1354 new_root_item->received_uuid,
1355 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1356 objectid);
1357 if (ret && ret != -EEXIST) {
1358 btrfs_abort_transaction(trans, root, ret);
1359 goto fail;
1360 }
1361 }
Chris Mason3063d292008-01-08 15:46:30 -05001362fail:
Miao Xieaec80302013-03-04 09:44:29 +00001363 pending->error = ret;
1364dir_item_existed:
Liu Bo98c99422011-09-11 10:52:24 -04001365 trans->block_rsv = rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001366 trans->bytes_reserved = 0;
Miao Xie6fa97002012-09-06 04:00:32 -06001367no_free_objectid:
1368 kfree(new_root_item);
1369root_item_alloc_fail:
Miao Xie42874b32012-09-06 04:03:32 -06001370 btrfs_free_path(path);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001371 return ret;
Chris Mason3063d292008-01-08 15:46:30 -05001372}
1373
Chris Masond352ac62008-09-29 15:18:18 -04001374/*
1375 * create all the snapshots we've scheduled for creation
1376 */
Chris Mason80b67942008-02-01 16:35:04 -05001377static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1378 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001379{
Miao Xieaec80302013-03-04 09:44:29 +00001380 struct btrfs_pending_snapshot *pending, *next;
Chris Mason3063d292008-01-08 15:46:30 -05001381 struct list_head *head = &trans->transaction->pending_snapshots;
Miao Xieaec80302013-03-04 09:44:29 +00001382 int ret = 0;
Chris Mason3de45862008-11-17 21:02:50 -05001383
Miao Xieaec80302013-03-04 09:44:29 +00001384 list_for_each_entry_safe(pending, next, head, list) {
1385 list_del(&pending->list);
1386 ret = create_pending_snapshot(trans, fs_info, pending);
1387 if (ret)
1388 break;
1389 }
1390 return ret;
Chris Mason3de45862008-11-17 21:02:50 -05001391}
1392
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001393static void update_super_roots(struct btrfs_root *root)
1394{
1395 struct btrfs_root_item *root_item;
1396 struct btrfs_super_block *super;
1397
David Sterba6c417612011-04-13 15:41:04 +02001398 super = root->fs_info->super_copy;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001399
1400 root_item = &root->fs_info->chunk_root->root_item;
1401 super->chunk_root = root_item->bytenr;
1402 super->chunk_root_generation = root_item->generation;
1403 super->chunk_root_level = root_item->level;
1404
1405 root_item = &root->fs_info->tree_root->root_item;
1406 super->root = root_item->bytenr;
1407 super->generation = root_item->generation;
1408 super->root_level = root_item->level;
Josef Bacik73bc1872011-10-03 14:07:49 -04001409 if (btrfs_test_opt(root, SPACE_CACHE))
Josef Bacik0af3d002010-06-21 14:48:16 -04001410 super->cache_generation = root_item->generation;
Stefan Behrens70f80172013-08-15 17:11:23 +02001411 if (root->fs_info->update_uuid_tree_gen)
1412 super->uuid_tree_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001413}
1414
Chris Masonf36f3042009-07-30 10:04:48 -04001415int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1416{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001417 struct btrfs_transaction *trans;
Chris Masonf36f3042009-07-30 10:04:48 -04001418 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001419
Josef Bacika4abeea2011-04-11 17:25:13 -04001420 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001421 trans = info->running_transaction;
1422 if (trans)
1423 ret = (trans->state >= TRANS_STATE_COMMIT_START);
Josef Bacika4abeea2011-04-11 17:25:13 -04001424 spin_unlock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001425 return ret;
1426}
1427
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001428int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1429{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001430 struct btrfs_transaction *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001431 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001432
Josef Bacika4abeea2011-04-11 17:25:13 -04001433 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001434 trans = info->running_transaction;
1435 if (trans)
1436 ret = is_transaction_blocked(trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001437 spin_unlock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001438 return ret;
1439}
1440
Sage Weilbb9c12c2010-10-29 15:37:34 -04001441/*
1442 * wait for the current transaction commit to start and block subsequent
1443 * transaction joins
1444 */
1445static void wait_current_trans_commit_start(struct btrfs_root *root,
1446 struct btrfs_transaction *trans)
1447{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001448 wait_event(root->fs_info->transaction_blocked_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001449 trans->state >= TRANS_STATE_COMMIT_START ||
1450 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001451}
1452
1453/*
1454 * wait for the current transaction to start and then become unblocked.
1455 * caller holds ref.
1456 */
1457static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1458 struct btrfs_transaction *trans)
1459{
Li Zefan72d63ed2011-07-14 03:17:00 +00001460 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001461 trans->state >= TRANS_STATE_UNBLOCKED ||
1462 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001463}
1464
1465/*
1466 * commit transactions asynchronously. once btrfs_commit_transaction_async
1467 * returns, any subsequent transaction will not be allowed to join.
1468 */
1469struct btrfs_async_commit {
1470 struct btrfs_trans_handle *newtrans;
1471 struct btrfs_root *root;
Miao Xie7892b5a2012-11-15 08:14:47 +00001472 struct work_struct work;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001473};
1474
1475static void do_async_commit(struct work_struct *work)
1476{
1477 struct btrfs_async_commit *ac =
Miao Xie7892b5a2012-11-15 08:14:47 +00001478 container_of(work, struct btrfs_async_commit, work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001479
Sage Weil6fc4e35482012-08-30 16:26:15 -06001480 /*
1481 * We've got freeze protection passed with the transaction.
1482 * Tell lockdep about it.
1483 */
Liu Bob1a06a42013-11-06 16:57:55 +08001484 if (ac->newtrans->type & __TRANS_FREEZABLE)
Miao Xieff7c1d32012-11-26 08:41:29 +00001485 rwsem_acquire_read(
1486 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1487 0, 1, _THIS_IP_);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001488
Sage Weile209db72012-08-30 16:26:16 -06001489 current->journal_info = ac->newtrans;
1490
Sage Weilbb9c12c2010-10-29 15:37:34 -04001491 btrfs_commit_transaction(ac->newtrans, ac->root);
1492 kfree(ac);
1493}
1494
1495int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1496 struct btrfs_root *root,
1497 int wait_for_unblock)
1498{
1499 struct btrfs_async_commit *ac;
1500 struct btrfs_transaction *cur_trans;
1501
1502 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001503 if (!ac)
1504 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001505
Miao Xie7892b5a2012-11-15 08:14:47 +00001506 INIT_WORK(&ac->work, do_async_commit);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001507 ac->root = root;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001508 ac->newtrans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001509 if (IS_ERR(ac->newtrans)) {
1510 int err = PTR_ERR(ac->newtrans);
1511 kfree(ac);
1512 return err;
1513 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001514
1515 /* take transaction reference */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001516 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001517 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001518
1519 btrfs_end_transaction(trans, root);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001520
1521 /*
1522 * Tell lockdep we've released the freeze rwsem, since the
1523 * async commit thread will be the one to unlock it.
1524 */
Liu Bob1a06a42013-11-06 16:57:55 +08001525 if (ac->newtrans->type & __TRANS_FREEZABLE)
Miao Xieff7c1d32012-11-26 08:41:29 +00001526 rwsem_release(
1527 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1528 1, _THIS_IP_);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001529
Miao Xie7892b5a2012-11-15 08:14:47 +00001530 schedule_work(&ac->work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001531
1532 /* wait for transaction to start and unblock */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001533 if (wait_for_unblock)
1534 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1535 else
1536 wait_current_trans_commit_start(root, cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001537
Sage Weil38e88052011-06-10 18:43:13 +00001538 if (current->journal_info == trans)
1539 current->journal_info = NULL;
1540
Josef Bacik724e2312013-09-30 11:36:38 -04001541 btrfs_put_transaction(cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001542 return 0;
1543}
1544
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001545
1546static void cleanup_transaction(struct btrfs_trans_handle *trans,
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001547 struct btrfs_root *root, int err)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001548{
1549 struct btrfs_transaction *cur_trans = trans->transaction;
Liu Bof094ac32013-02-27 13:28:25 +00001550 DEFINE_WAIT(wait);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001551
1552 WARN_ON(trans->use_count > 1);
1553
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001554 btrfs_abort_transaction(trans, root, err);
1555
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001556 spin_lock(&root->fs_info->trans_lock);
Liu Bo66b61352013-03-04 16:25:41 +00001557
Miao Xie25d8c282013-05-15 07:48:26 +00001558 /*
1559 * If the transaction is removed from the list, it means this
1560 * transaction has been committed successfully, so it is impossible
1561 * to call the cleanup function.
1562 */
1563 BUG_ON(list_empty(&cur_trans->list));
Liu Bo66b61352013-03-04 16:25:41 +00001564
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001565 list_del_init(&cur_trans->list);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001566 if (cur_trans == root->fs_info->running_transaction) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001567 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Liu Bof094ac32013-02-27 13:28:25 +00001568 spin_unlock(&root->fs_info->trans_lock);
1569 wait_event(cur_trans->writer_wait,
1570 atomic_read(&cur_trans->num_writers) == 1);
1571
1572 spin_lock(&root->fs_info->trans_lock);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001573 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001574 spin_unlock(&root->fs_info->trans_lock);
1575
1576 btrfs_cleanup_one_transaction(trans->transaction, root);
1577
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001578 spin_lock(&root->fs_info->trans_lock);
1579 if (cur_trans == root->fs_info->running_transaction)
1580 root->fs_info->running_transaction = NULL;
1581 spin_unlock(&root->fs_info->trans_lock);
1582
Josef Bacike0228282013-09-20 22:26:29 -04001583 if (trans->type & __TRANS_FREEZABLE)
1584 sb_end_intwrite(root->fs_info->sb);
Josef Bacik724e2312013-09-30 11:36:38 -04001585 btrfs_put_transaction(cur_trans);
1586 btrfs_put_transaction(cur_trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001587
1588 trace_btrfs_transaction_commit(root);
1589
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001590 if (current->journal_info == trans)
1591 current->journal_info = NULL;
Wang Shilongc0af8f02014-02-19 19:24:18 +08001592 btrfs_scrub_cancel(root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001593
1594 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1595}
1596
Miao Xieca469632012-11-01 07:33:14 +00001597static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root)
1599{
Miao Xieca469632012-11-01 07:33:14 +00001600 int ret;
1601
Miao Xieca469632012-11-01 07:33:14 +00001602 ret = btrfs_run_delayed_items(trans, root);
Miao Xieca469632012-11-01 07:33:14 +00001603 /*
1604 * running the delayed items may have added new refs. account
1605 * them now so that they hinder processing of more delayed refs
1606 * as little as possible.
1607 */
Filipe David Borba Manana80d94fb2013-10-07 12:04:28 +01001608 if (ret)
1609 return ret;
Miao Xieca469632012-11-01 07:33:14 +00001610
1611 /*
1612 * rename don't use btrfs_join_transaction, so, once we
1613 * set the transaction to blocked above, we aren't going
1614 * to get any new ordered operations. We can safely run
1615 * it here and no for sure that nothing new will be added
1616 * to the list
1617 */
Josef Bacik569e0f32013-02-13 11:09:14 -05001618 ret = btrfs_run_ordered_operations(trans, root, 1);
Miao Xieca469632012-11-01 07:33:14 +00001619
Miao Xieeebc6082013-01-22 10:50:01 +00001620 return ret;
Miao Xieca469632012-11-01 07:33:14 +00001621}
1622
Miao Xie82436612013-05-15 07:48:28 +00001623static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1624{
1625 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
Miao Xie6c255e62014-03-06 13:55:01 +08001626 return btrfs_start_delalloc_roots(fs_info, 1, -1);
Miao Xie82436612013-05-15 07:48:28 +00001627 return 0;
1628}
1629
1630static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1631{
1632 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
Miao Xieb0244192013-11-04 23:13:25 +08001633 btrfs_wait_ordered_roots(fs_info, -1);
Miao Xie82436612013-05-15 07:48:28 +00001634}
1635
Chris Mason79154b12007-03-22 15:59:16 -04001636int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1637 struct btrfs_root *root)
1638{
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001639 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason8fd17792007-04-19 21:01:03 -04001640 struct btrfs_transaction *prev_trans = NULL;
Miao Xie25287e02012-10-25 09:31:03 +00001641 int ret;
Chris Mason79154b12007-03-22 15:59:16 -04001642
Josef Bacik569e0f32013-02-13 11:09:14 -05001643 ret = btrfs_run_ordered_operations(trans, root, 0);
Miao Xie25287e02012-10-25 09:31:03 +00001644 if (ret) {
1645 btrfs_abort_transaction(trans, root, ret);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001646 btrfs_end_transaction(trans, root);
1647 return ret;
Miao Xie25287e02012-10-25 09:31:03 +00001648 }
1649
Miao Xie8d25a082013-01-15 06:27:25 +00001650 /* Stop the commit early if ->aborted is set */
1651 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
Miao Xie25287e02012-10-25 09:31:03 +00001652 ret = cur_trans->aborted;
Josef Bacike4a2bca2013-02-06 16:55:41 -05001653 btrfs_end_transaction(trans, root);
1654 return ret;
Miao Xie25287e02012-10-25 09:31:03 +00001655 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001656
Chris Mason56bec292009-03-13 10:10:06 -04001657 /* make a pass through all the delayed refs we have so far
1658 * any runnings procs may add more while we are here
1659 */
1660 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001661 if (ret) {
1662 btrfs_end_transaction(trans, root);
1663 return ret;
1664 }
Chris Mason56bec292009-03-13 10:10:06 -04001665
Josef Bacik0e721102012-06-26 16:13:18 -04001666 btrfs_trans_release_metadata(trans, root);
1667 trans->block_rsv = NULL;
Miao Xie272d26d2013-02-20 14:16:39 +00001668 if (trans->qgroup_reserved) {
1669 btrfs_qgroup_free(root, trans->qgroup_reserved);
1670 trans->qgroup_reserved = 0;
1671 }
Josef Bacik0e721102012-06-26 16:13:18 -04001672
Chris Masonb7ec40d2009-03-12 20:12:45 -04001673 cur_trans = trans->transaction;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001674
Chris Mason56bec292009-03-13 10:10:06 -04001675 /*
1676 * set the flushing flag so procs in this transaction have to
1677 * start sending their work down.
1678 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001679 cur_trans->delayed_refs.flushing = 1;
Josef Bacik1be41b72013-06-12 13:56:06 -04001680 smp_wmb();
Chris Mason56bec292009-03-13 10:10:06 -04001681
Josef Bacikea658ba2012-09-11 16:57:25 -04001682 if (!list_empty(&trans->new_bgs))
1683 btrfs_create_pending_block_groups(trans, root);
1684
Chris Masonc3e69d52009-03-13 10:17:05 -04001685 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001686 if (ret) {
1687 btrfs_end_transaction(trans, root);
1688 return ret;
1689 }
Chris Mason56bec292009-03-13 10:10:06 -04001690
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001691 spin_lock(&root->fs_info->trans_lock);
1692 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1693 spin_unlock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001694 atomic_inc(&cur_trans->use_count);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001695 ret = btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001696
Li Zefanb9c83002011-07-14 03:17:14 +00001697 wait_for_commit(root, cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001698
Josef Bacik724e2312013-09-30 11:36:38 -04001699 btrfs_put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001700
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001701 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001702 }
Chris Mason4313b392008-01-03 09:08:48 -05001703
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001704 cur_trans->state = TRANS_STATE_COMMIT_START;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001705 wake_up(&root->fs_info->transaction_blocked_wait);
1706
Chris Masonccd467d2007-06-28 15:57:36 -04001707 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1708 prev_trans = list_entry(cur_trans->list.prev,
1709 struct btrfs_transaction, list);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001710 if (prev_trans->state != TRANS_STATE_COMPLETED) {
Josef Bacik13c5a932011-04-11 15:45:29 -04001711 atomic_inc(&prev_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -04001712 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001713
1714 wait_for_commit(root, prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001715
Josef Bacik724e2312013-09-30 11:36:38 -04001716 btrfs_put_transaction(prev_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001717 } else {
1718 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001719 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001720 } else {
1721 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001722 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001723
Miao Xie0860adf2013-05-15 07:48:27 +00001724 extwriter_counter_dec(cur_trans, trans->type);
1725
Miao Xie82436612013-05-15 07:48:28 +00001726 ret = btrfs_start_delalloc_flush(root->fs_info);
1727 if (ret)
1728 goto cleanup_transaction;
1729
Miao Xie581227d2013-05-15 07:48:30 +00001730 ret = btrfs_flush_all_pending_stuffs(trans, root);
1731 if (ret)
1732 goto cleanup_transaction;
Chris Mason89573b92009-03-12 20:12:45 -04001733
Miao Xie581227d2013-05-15 07:48:30 +00001734 wait_event(cur_trans->writer_wait,
1735 extwriter_counter_read(cur_trans) == 0);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001736
Miao Xie581227d2013-05-15 07:48:30 +00001737 /* some pending stuffs might be added after the previous flush. */
Miao Xieca469632012-11-01 07:33:14 +00001738 ret = btrfs_flush_all_pending_stuffs(trans, root);
1739 if (ret)
1740 goto cleanup_transaction;
1741
Miao Xie82436612013-05-15 07:48:28 +00001742 btrfs_wait_delalloc_flush(root->fs_info);
Wang Shilongcb7ab022013-12-04 21:16:53 +08001743
1744 btrfs_scrub_pause(root);
Chris Mason75857172011-06-13 20:00:16 -04001745 /*
Josef Baciked0ca142011-06-14 16:22:15 -04001746 * Ok now we need to make sure to block out any other joins while we
1747 * commit the transaction. We could have started a join before setting
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001748 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
Josef Baciked0ca142011-06-14 16:22:15 -04001749 */
1750 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001751 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Josef Baciked0ca142011-06-14 16:22:15 -04001752 spin_unlock(&root->fs_info->trans_lock);
1753 wait_event(cur_trans->writer_wait,
1754 atomic_read(&cur_trans->num_writers) == 1);
1755
Miao Xie2cba30f2013-01-15 06:29:12 +00001756 /* ->aborted might be set after the previous check, so check it */
1757 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1758 ret = cur_trans->aborted;
Wang Shilong6cf7f772014-02-19 19:24:16 +08001759 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00001760 }
Chris Masone038dca2011-06-17 14:16:13 -04001761 /*
Chris Mason75857172011-06-13 20:00:16 -04001762 * the reloc mutex makes sure that we stop
1763 * the balancing code from coming in and moving
1764 * extents around in the middle of the commit
1765 */
1766 mutex_lock(&root->fs_info->reloc_mutex);
1767
Miao Xie42874b32012-09-06 04:03:32 -06001768 /*
1769 * We needn't worry about the delayed items because we will
1770 * deal with them in create_pending_snapshot(), which is the
1771 * core function of the snapshot creation.
1772 */
1773 ret = create_pending_snapshots(trans, root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001774 if (ret) {
1775 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001776 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001777 }
Chris Mason3063d292008-01-08 15:46:30 -05001778
Miao Xie42874b32012-09-06 04:03:32 -06001779 /*
1780 * We insert the dir indexes of the snapshots and update the inode
1781 * of the snapshots' parents after the snapshot creation, so there
1782 * are some delayed items which are not dealt with. Now deal with
1783 * them.
1784 *
1785 * We needn't worry that this operation will corrupt the snapshots,
1786 * because all the tree which are snapshoted will be forced to COW
1787 * the nodes and leaves.
1788 */
1789 ret = btrfs_run_delayed_items(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001790 if (ret) {
1791 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001792 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001793 }
Miao Xie16cdcec2011-04-22 18:12:22 +08001794
Chris Mason56bec292009-03-13 10:10:06 -04001795 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001796 if (ret) {
1797 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001798 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001799 }
Chris Mason56bec292009-03-13 10:10:06 -04001800
Chris Masone9993762011-06-17 16:14:09 -04001801 /*
1802 * make sure none of the code above managed to slip in a
1803 * delayed item
1804 */
1805 btrfs_assert_delayed_root_empty(root);
1806
Chris Mason2c90e5d2007-04-02 10:50:19 -04001807 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05001808
Chris Masone02119d2008-09-05 16:13:11 -04001809 /* btrfs_commit_tree_roots is responsible for getting the
1810 * various roots consistent with each other. Every pointer
1811 * in the tree of tree roots has to point to the most up to date
1812 * root for every subvolume and other tree. So, we have to keep
1813 * the tree logging code from jumping in and changing any
1814 * of the trees.
1815 *
1816 * At this point in the commit, there can't be any tree-log
1817 * writers, but a little lower down we drop the trans mutex
1818 * and let new people in. By holding the tree_log_mutex
1819 * from now until after the super is written, we avoid races
1820 * with the tree-log code.
1821 */
1822 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04001823
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001824 ret = commit_fs_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001825 if (ret) {
1826 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02001827 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001828 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001829 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001830
Qu Wenruo3818aea2014-01-13 13:36:06 +08001831 /*
1832 * Since the transaction is done, we should set the inode map cache flag
1833 * before any other comming transaction.
1834 */
1835 if (btrfs_test_opt(root, CHANGE_INODE_CACHE))
1836 btrfs_set_opt(root->fs_info->mount_opt, INODE_MAP_CACHE);
1837 else
1838 btrfs_clear_opt(root->fs_info->mount_opt, INODE_MAP_CACHE);
1839
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001840 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04001841 * safe to free the root of tree log roots
1842 */
1843 btrfs_free_log_root_tree(trans, root->fs_info);
1844
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001845 ret = commit_cowonly_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001846 if (ret) {
1847 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02001848 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001849 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001850 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001851
Miao Xie2cba30f2013-01-15 06:29:12 +00001852 /*
1853 * The tasks which save the space cache and inode cache may also
1854 * update ->aborted, check it.
1855 */
1856 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1857 ret = cur_trans->aborted;
1858 mutex_unlock(&root->fs_info->tree_log_mutex);
1859 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001860 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00001861 }
1862
Yan Zheng11833d62009-09-11 16:11:19 -04001863 btrfs_prepare_extent_commit(trans, root);
1864
Chris Mason78fae272007-03-25 11:35:08 -04001865 cur_trans = root->fs_info->running_transaction;
Chris Mason5f39d392007-10-15 16:14:19 -04001866
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001867 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1868 root->fs_info->tree_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04001869 list_add_tail(&root->fs_info->tree_root->dirty_list,
1870 &cur_trans->switch_commits);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001871
1872 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1873 root->fs_info->chunk_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04001874 list_add_tail(&root->fs_info->chunk_root->dirty_list,
1875 &cur_trans->switch_commits);
1876
1877 switch_commit_roots(cur_trans, root->fs_info);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001878
Jan Schmidtedf39272012-06-28 18:04:55 +02001879 assert_qgroups_uptodate(trans);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001880 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04001881
Josef Bacik60e7cd32013-09-30 14:10:43 -04001882 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1883 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
David Sterba6c417612011-04-13 15:41:04 +02001884 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1885 sizeof(*root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04001886
Josef Bacika4abeea2011-04-11 17:25:13 -04001887 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001888 cur_trans->state = TRANS_STATE_UNBLOCKED;
Josef Bacika4abeea2011-04-11 17:25:13 -04001889 root->fs_info->running_transaction = NULL;
Josef Bacika4abeea2011-04-11 17:25:13 -04001890 spin_unlock(&root->fs_info->trans_lock);
Chris Mason75857172011-06-13 20:00:16 -04001891 mutex_unlock(&root->fs_info->reloc_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001892
Chris Masonf9295742008-07-17 12:54:14 -04001893 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001894
Chris Mason79154b12007-03-22 15:59:16 -04001895 ret = btrfs_write_and_wait_transaction(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001896 if (ret) {
1897 btrfs_error(root->fs_info, ret,
David Sterba08748812013-03-12 14:46:08 +00001898 "Error while writing out transaction");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001899 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001900 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001901 }
1902
1903 ret = write_ctree_super(trans, root, 0);
1904 if (ret) {
1905 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001906 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001907 }
Chris Mason4313b392008-01-03 09:08:48 -05001908
Chris Masone02119d2008-09-05 16:13:11 -04001909 /*
1910 * the super is written, we can safely allow the tree-loggers
1911 * to go about their business
1912 */
1913 mutex_unlock(&root->fs_info->tree_log_mutex);
1914
Yan Zheng11833d62009-09-11 16:11:19 -04001915 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05001916
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001917 root->fs_info->last_trans_committed = cur_trans->transid;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001918 /*
1919 * We needn't acquire the lock here because there is no other task
1920 * which can change it.
1921 */
1922 cur_trans->state = TRANS_STATE_COMPLETED;
Chris Mason2c90e5d2007-04-02 10:50:19 -04001923 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05001924
Josef Bacika4abeea2011-04-11 17:25:13 -04001925 spin_lock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001926 list_del_init(&cur_trans->list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001927 spin_unlock(&root->fs_info->trans_lock);
1928
Josef Bacik724e2312013-09-30 11:36:38 -04001929 btrfs_put_transaction(cur_trans);
1930 btrfs_put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04001931
Miao Xie0860adf2013-05-15 07:48:27 +00001932 if (trans->type & __TRANS_FREEZABLE)
Miao Xie354aa0f2012-09-20 01:54:00 -06001933 sb_end_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +02001934
liubo1abe9b82011-03-24 11:18:59 +00001935 trace_btrfs_transaction_commit(root);
1936
Arne Jansena2de7332011-03-08 14:14:00 +01001937 btrfs_scrub_continue(root);
1938
Josef Bacik9ed74f22009-09-11 16:12:44 -04001939 if (current->journal_info == trans)
1940 current->journal_info = NULL;
1941
Chris Mason2c90e5d2007-04-02 10:50:19 -04001942 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf042009-11-12 09:36:34 +00001943
1944 if (current != root->fs_info->transaction_kthread)
1945 btrfs_run_delayed_iputs(root);
1946
Chris Mason79154b12007-03-22 15:59:16 -04001947 return ret;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001948
Wang Shilong6cf7f772014-02-19 19:24:16 +08001949scrub_continue:
1950 btrfs_scrub_continue(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001951cleanup_transaction:
Josef Bacik0e721102012-06-26 16:13:18 -04001952 btrfs_trans_release_metadata(trans, root);
1953 trans->block_rsv = NULL;
Miao Xie272d26d2013-02-20 14:16:39 +00001954 if (trans->qgroup_reserved) {
1955 btrfs_qgroup_free(root, trans->qgroup_reserved);
1956 trans->qgroup_reserved = 0;
1957 }
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00001958 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001959 if (current->journal_info == trans)
1960 current->journal_info = NULL;
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001961 cleanup_transaction(trans, root, ret);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001962
1963 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001964}
1965
Chris Masond352ac62008-09-29 15:18:18 -04001966/*
David Sterba9d1a2a32013-03-12 15:13:28 +00001967 * return < 0 if error
1968 * 0 if there are no more dead_roots at the time of call
1969 * 1 there are more to be processed, call me again
1970 *
1971 * The return value indicates there are certainly more snapshots to delete, but
1972 * if there comes a new one during processing, it may return 0. We don't mind,
1973 * because btrfs_commit_super will poke cleaner thread and it will process it a
1974 * few seconds later.
Chris Masond352ac62008-09-29 15:18:18 -04001975 */
David Sterba9d1a2a32013-03-12 15:13:28 +00001976int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001977{
David Sterba9d1a2a32013-03-12 15:13:28 +00001978 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001979 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001980
Josef Bacika4abeea2011-04-11 17:25:13 -04001981 spin_lock(&fs_info->trans_lock);
David Sterba9d1a2a32013-03-12 15:13:28 +00001982 if (list_empty(&fs_info->dead_roots)) {
1983 spin_unlock(&fs_info->trans_lock);
1984 return 0;
1985 }
1986 root = list_first_entry(&fs_info->dead_roots,
1987 struct btrfs_root, root_list);
Josef Bacikcfad3922013-07-25 15:11:47 -04001988 list_del_init(&root->root_list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001989 spin_unlock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001990
Frank Holtonefe120a2013-12-20 11:37:06 -05001991 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
Jeff Mahoney2c536792011-10-03 23:22:41 -04001992
David Sterba9d1a2a32013-03-12 15:13:28 +00001993 btrfs_kill_all_delayed_nodes(root);
Yan, Zheng76dda932009-09-21 16:00:26 -04001994
David Sterba9d1a2a32013-03-12 15:13:28 +00001995 if (btrfs_header_backref_rev(root->node) <
1996 BTRFS_MIXED_BACKREF_REV)
1997 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1998 else
1999 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2000 /*
2001 * If we encounter a transaction abort during snapshot cleaning, we
2002 * don't want to crash here
2003 */
Josef Bacik6596a9282013-07-31 10:28:05 -04002004 return (ret < 0) ? 0 : 1;
Chris Masone9d0b132007-08-10 14:06:19 -04002005}