blob: 703238ed7337e807c6cbcad49ad1ba1498d64194 [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 Bacik6df9a952013-06-27 13:22:46 -0400221 INIT_LIST_HEAD(&cur_trans->pending_chunks);
Josef Bacik9e351cc2014-03-13 15:42:13 -0400222 INIT_LIST_HEAD(&cur_trans->switch_commits);
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200223 list_add_tail(&cur_trans->list, &fs_info->trans_list);
Josef Bacika4abeea2011-04-11 17:25:13 -0400224 extent_io_tree_init(&cur_trans->dirty_pages,
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200225 fs_info->btree_inode->i_mapping);
226 fs_info->generation++;
227 cur_trans->transid = fs_info->generation;
228 fs_info->running_transaction = cur_trans;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100229 cur_trans->aborted = 0;
Jan Schmidt19ae4e82012-05-20 15:42:19 +0200230 spin_unlock(&fs_info->trans_lock);
Josef Bacik15ee9bc2007-08-10 16:22:09 -0400231
Chris Mason79154b12007-03-22 15:59:16 -0400232 return 0;
233}
234
Chris Masond352ac62008-09-29 15:18:18 -0400235/*
Chris Masond3977122009-01-05 21:25:51 -0500236 * this does all the record keeping required to make sure that a reference
237 * counted root is properly recorded in a given transaction. This is required
238 * to make sure the old root from before we joined the transaction is deleted
239 * when the transaction commits
Chris Masond352ac62008-09-29 15:18:18 -0400240 */
Chris Mason75857172011-06-13 20:00:16 -0400241static int record_root_in_trans(struct btrfs_trans_handle *trans,
Josef Bacika4abeea2011-04-11 17:25:13 -0400242 struct btrfs_root *root)
Chris Mason6702ed42007-08-07 16:15:09 -0400243{
Miao Xie27cdeb72014-04-02 19:51:05 +0800244 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
245 root->last_trans < trans->transid) {
Chris Mason6702ed42007-08-07 16:15:09 -0400246 WARN_ON(root == root->fs_info->extent_root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400247 WARN_ON(root->commit_root != root->node);
Yan Zheng31153d82008-07-28 15:32:19 -0400248
Chris Mason75857172011-06-13 20:00:16 -0400249 /*
Miao Xie27cdeb72014-04-02 19:51:05 +0800250 * see below for IN_TRANS_SETUP usage rules
Chris Mason75857172011-06-13 20:00:16 -0400251 * we have the reloc mutex held now, so there
252 * is only one writer in this function
253 */
Miao Xie27cdeb72014-04-02 19:51:05 +0800254 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
Chris Mason75857172011-06-13 20:00:16 -0400255
Miao Xie27cdeb72014-04-02 19:51:05 +0800256 /* make sure readers find IN_TRANS_SETUP before
Chris Mason75857172011-06-13 20:00:16 -0400257 * they find our root->last_trans update
258 */
259 smp_wmb();
260
Josef Bacika4abeea2011-04-11 17:25:13 -0400261 spin_lock(&root->fs_info->fs_roots_radix_lock);
262 if (root->last_trans == trans->transid) {
263 spin_unlock(&root->fs_info->fs_roots_radix_lock);
264 return 0;
265 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400266 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
267 (unsigned long)root->root_key.objectid,
268 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -0400269 spin_unlock(&root->fs_info->fs_roots_radix_lock);
Chris Mason75857172011-06-13 20:00:16 -0400270 root->last_trans = trans->transid;
271
272 /* this is pretty tricky. We don't want to
273 * take the relocation lock in btrfs_record_root_in_trans
274 * unless we're really doing the first setup for this root in
275 * this transaction.
276 *
277 * Normally we'd use root->last_trans as a flag to decide
278 * if we want to take the expensive mutex.
279 *
280 * But, we have to set root->last_trans before we
281 * init the relocation root, otherwise, we trip over warnings
282 * in ctree.c. The solution used here is to flag ourselves
Miao Xie27cdeb72014-04-02 19:51:05 +0800283 * with root IN_TRANS_SETUP. When this is 1, we're still
Chris Mason75857172011-06-13 20:00:16 -0400284 * fixing up the reloc trees and everyone must wait.
285 *
286 * When this is zero, they can trust root->last_trans and fly
287 * through btrfs_record_root_in_trans without having to take the
288 * lock. smp_wmb() makes sure that all the writes above are
289 * done before we pop in the zero below
290 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400291 btrfs_init_reloc_root(trans, root);
Chris Masonc7548af2014-06-10 13:06:56 -0700292 smp_mb__before_atomic();
Miao Xie27cdeb72014-04-02 19:51:05 +0800293 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
Chris Mason6702ed42007-08-07 16:15:09 -0400294 }
295 return 0;
296}
297
Chris Mason75857172011-06-13 20:00:16 -0400298
299int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
300 struct btrfs_root *root)
301{
Miao Xie27cdeb72014-04-02 19:51:05 +0800302 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
Chris Mason75857172011-06-13 20:00:16 -0400303 return 0;
304
305 /*
Miao Xie27cdeb72014-04-02 19:51:05 +0800306 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
Chris Mason75857172011-06-13 20:00:16 -0400307 * and barriers
308 */
309 smp_rmb();
310 if (root->last_trans == trans->transid &&
Miao Xie27cdeb72014-04-02 19:51:05 +0800311 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
Chris Mason75857172011-06-13 20:00:16 -0400312 return 0;
313
314 mutex_lock(&root->fs_info->reloc_mutex);
315 record_root_in_trans(trans, root);
316 mutex_unlock(&root->fs_info->reloc_mutex);
317
318 return 0;
319}
320
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000321static inline int is_transaction_blocked(struct btrfs_transaction *trans)
322{
323 return (trans->state >= TRANS_STATE_BLOCKED &&
Josef Bacik501407a2013-06-10 16:47:23 -0400324 trans->state < TRANS_STATE_UNBLOCKED &&
325 !trans->aborted);
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000326}
327
Chris Masond352ac62008-09-29 15:18:18 -0400328/* wait for commit against the current transaction to become unblocked
329 * when this is done, it is safe to start a new transaction, but the current
330 * transaction might not be fully on disk.
331 */
Chris Mason37d1aee2008-07-31 10:48:37 -0400332static void wait_current_trans(struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400333{
Chris Masonf9295742008-07-17 12:54:14 -0400334 struct btrfs_transaction *cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400335
Josef Bacika4abeea2011-04-11 17:25:13 -0400336 spin_lock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400337 cur_trans = root->fs_info->running_transaction;
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000338 if (cur_trans && is_transaction_blocked(cur_trans)) {
Josef Bacik13c5a932011-04-11 15:45:29 -0400339 atomic_inc(&cur_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -0400340 spin_unlock(&root->fs_info->trans_lock);
Li Zefan72d63ed2011-07-14 03:17:00 +0000341
342 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -0400343 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
344 cur_trans->aborted);
Josef Bacik724e2312013-09-30 11:36:38 -0400345 btrfs_put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400346 } else {
347 spin_unlock(&root->fs_info->trans_lock);
Chris Masonf9295742008-07-17 12:54:14 -0400348 }
Chris Mason37d1aee2008-07-31 10:48:37 -0400349}
350
Yan, Zhenga22285a2010-05-16 10:48:46 -0400351static int may_wait_transaction(struct btrfs_root *root, int type)
Chris Mason37d1aee2008-07-31 10:48:37 -0400352{
Josef Bacika4abeea2011-04-11 17:25:13 -0400353 if (root->fs_info->log_root_recovering)
354 return 0;
355
356 if (type == TRANS_USERSPACE)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400357 return 1;
Josef Bacika4abeea2011-04-11 17:25:13 -0400358
359 if (type == TRANS_START &&
360 !atomic_read(&root->fs_info->open_ioctl_trans))
361 return 1;
362
Yan, Zhenga22285a2010-05-16 10:48:46 -0400363 return 0;
364}
365
Miao Xie20dd2cb2013-09-25 21:47:45 +0800366static inline bool need_reserve_reloc_root(struct btrfs_root *root)
367{
368 if (!root->fs_info->reloc_ctl ||
Miao Xie27cdeb72014-04-02 19:51:05 +0800369 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
Miao Xie20dd2cb2013-09-25 21:47:45 +0800370 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
371 root->reloc_root)
372 return false;
373
374 return true;
375}
376
Miao Xie08e007d2012-10-16 11:33:38 +0000377static struct btrfs_trans_handle *
Miao Xie0860adf2013-05-15 07:48:27 +0000378start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
Miao Xie08e007d2012-10-16 11:33:38 +0000379 enum btrfs_reserve_flush_enum flush)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400380{
381 struct btrfs_trans_handle *h;
382 struct btrfs_transaction *cur_trans;
Josef Bacikb5009942011-06-07 15:07:51 -0400383 u64 num_bytes = 0;
Arne Jansenc5567232011-09-14 15:44:05 +0200384 u64 qgroup_reserved = 0;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800385 bool reloc_reserved = false;
386 int ret;
liuboacce9522011-01-06 19:30:25 +0800387
Filipe Manana46c4e712014-06-24 17:48:28 +0100388 /* Send isn't supposed to start transactions. */
David Sterba2755a0d2014-07-31 00:43:18 +0200389 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
Filipe Manana46c4e712014-06-24 17:48:28 +0100390
Miao Xie87533c42013-01-29 10:14:48 +0000391 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
liuboacce9522011-01-06 19:30:25 +0800392 return ERR_PTR(-EROFS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400393
Filipe Manana46c4e712014-06-24 17:48:28 +0100394 if (current->journal_info) {
Miao Xie0860adf2013-05-15 07:48:27 +0000395 WARN_ON(type & TRANS_EXTWRITERS);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400396 h = current->journal_info;
397 h->use_count++;
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000398 WARN_ON(h->use_count > 2);
Josef Bacik2a1eb462011-04-13 15:15:59 -0400399 h->orig_rsv = h->block_rsv;
400 h->block_rsv = NULL;
401 goto got_it;
402 }
Josef Bacikb5009942011-06-07 15:07:51 -0400403
404 /*
405 * Do the reservation before we join the transaction so we can do all
406 * the appropriate flushing if need be.
407 */
408 if (num_items > 0 && root != root->fs_info->chunk_root) {
Arne Jansenc5567232011-09-14 15:44:05 +0200409 if (root->fs_info->quota_enabled &&
410 is_fstree(root->root_key.objectid)) {
David Sterba707e8a02014-06-04 19:22:26 +0200411 qgroup_reserved = num_items * root->nodesize;
Arne Jansenc5567232011-09-14 15:44:05 +0200412 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
413 if (ret)
414 return ERR_PTR(ret);
415 }
416
Josef Bacikb5009942011-06-07 15:07:51 -0400417 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
Miao Xie20dd2cb2013-09-25 21:47:45 +0800418 /*
419 * Do the reservation for the relocation root creation
420 */
David Sterbaee39b432014-09-30 01:33:33 +0200421 if (need_reserve_reloc_root(root)) {
Miao Xie20dd2cb2013-09-25 21:47:45 +0800422 num_bytes += root->nodesize;
423 reloc_reserved = true;
424 }
425
Miao Xie08e007d2012-10-16 11:33:38 +0000426 ret = btrfs_block_rsv_add(root,
427 &root->fs_info->trans_block_rsv,
428 num_bytes, flush);
Josef Bacikb5009942011-06-07 15:07:51 -0400429 if (ret)
Miao Xie843fcf32013-01-28 12:36:22 +0000430 goto reserve_fail;
Josef Bacikb5009942011-06-07 15:07:51 -0400431 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400432again:
433 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
Miao Xie843fcf32013-01-28 12:36:22 +0000434 if (!h) {
435 ret = -ENOMEM;
436 goto alloc_fail;
437 }
Yan, Zhenga22285a2010-05-16 10:48:46 -0400438
Josef Bacik98114652012-09-14 11:22:38 -0400439 /*
440 * If we are JOIN_NOLOCK we're already committing a transaction and
441 * waiting on this guy, so we don't need to do the sb_start_intwrite
442 * because we're already holding a ref. We need this because we could
443 * have raced in and did an fsync() on a file which can kick a commit
444 * and then we deadlock with somebody doing a freeze.
Miao Xie354aa0f2012-09-20 01:54:00 -0600445 *
446 * If we are ATTACH, it means we just want to catch the current
447 * transaction and commit it, so we needn't do sb_start_intwrite().
Josef Bacik98114652012-09-14 11:22:38 -0400448 */
Miao Xie0860adf2013-05-15 07:48:27 +0000449 if (type & __TRANS_FREEZABLE)
Josef Bacik60376ce2012-09-14 10:34:40 -0400450 sb_start_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +0200451
Yan, Zhenga22285a2010-05-16 10:48:46 -0400452 if (may_wait_transaction(root, type))
Chris Mason37d1aee2008-07-31 10:48:37 -0400453 wait_current_trans(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400454
Josef Bacika4abeea2011-04-11 17:25:13 -0400455 do {
Miao Xie354aa0f2012-09-20 01:54:00 -0600456 ret = join_transaction(root, type);
Miao Xie178260b2013-02-20 09:16:24 +0000457 if (ret == -EBUSY) {
Josef Bacika4abeea2011-04-11 17:25:13 -0400458 wait_current_trans(root);
Miao Xie178260b2013-02-20 09:16:24 +0000459 if (unlikely(type == TRANS_ATTACH))
460 ret = -ENOENT;
461 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400462 } while (ret == -EBUSY);
463
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000464 if (ret < 0) {
Miao Xie354aa0f2012-09-20 01:54:00 -0600465 /* We must get the transaction if we are JOIN_NOLOCK. */
466 BUG_ON(type == TRANS_JOIN_NOLOCK);
Miao Xie843fcf32013-01-28 12:36:22 +0000467 goto join_fail;
Tsutomu Itohdb5b4932011-03-23 08:14:16 +0000468 }
Chris Mason0f7d52f2007-04-09 10:42:37 -0400469
Yan, Zhenga22285a2010-05-16 10:48:46 -0400470 cur_trans = root->fs_info->running_transaction;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400471
472 h->transid = cur_trans->transid;
473 h->transaction = cur_trans;
Chris Mason79154b12007-03-22 15:59:16 -0400474 h->blocks_used = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400475 h->bytes_reserved = 0;
Arne Jansend13603e2011-09-13 11:40:09 +0200476 h->root = root;
Chris Mason56bec292009-03-13 10:10:06 -0400477 h->delayed_ref_updates = 0;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400478 h->use_count = 1;
Josef Bacik0e721102012-06-26 16:13:18 -0400479 h->adding_csums = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -0400480 h->block_rsv = NULL;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400481 h->orig_rsv = NULL;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100482 h->aborted = 0;
Miao Xie4b824902013-02-20 09:13:32 +0000483 h->qgroup_reserved = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200484 h->delayed_ref_elem.seq = 0;
Miao Xiea698d0752012-09-20 01:51:59 -0600485 h->type = type;
Josef Bacikc6b305a2012-12-18 09:16:16 -0500486 h->allocating_chunk = false;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800487 h->reloc_reserved = false;
Josef Bacik5039edd2014-01-15 13:34:13 -0500488 h->sync = false;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200489 INIT_LIST_HEAD(&h->qgroup_ref_list);
Josef Bacikea658ba2012-09-11 16:57:25 -0400490 INIT_LIST_HEAD(&h->new_bgs);
Chris Masonb7ec40d2009-03-12 20:12:45 -0400491
Yan, Zhenga22285a2010-05-16 10:48:46 -0400492 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000493 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
494 may_wait_transaction(root, type)) {
Filipe Mananaabdd2e82014-06-24 17:46:58 +0100495 current->journal_info = h;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400496 btrfs_commit_transaction(h, root);
497 goto again;
498 }
Josef Bacik9ed74f22009-09-11 16:12:44 -0400499
Josef Bacikb5009942011-06-07 15:07:51 -0400500 if (num_bytes) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500501 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -0400502 h->transid, num_bytes, 1);
Josef Bacikb5009942011-06-07 15:07:51 -0400503 h->block_rsv = &root->fs_info->trans_block_rsv;
504 h->bytes_reserved = num_bytes;
Miao Xie20dd2cb2013-09-25 21:47:45 +0800505 h->reloc_reserved = reloc_reserved;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400506 }
Miao Xie4b824902013-02-20 09:13:32 +0000507 h->qgroup_reserved = qgroup_reserved;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400508
Josef Bacik2a1eb462011-04-13 15:15:59 -0400509got_it:
Josef Bacika4abeea2011-04-11 17:25:13 -0400510 btrfs_record_root_in_trans(h, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400511
512 if (!current->journal_info && type != TRANS_USERSPACE)
513 current->journal_info = h;
Chris Mason79154b12007-03-22 15:59:16 -0400514 return h;
Miao Xie843fcf32013-01-28 12:36:22 +0000515
516join_fail:
Miao Xie0860adf2013-05-15 07:48:27 +0000517 if (type & __TRANS_FREEZABLE)
Miao Xie843fcf32013-01-28 12:36:22 +0000518 sb_end_intwrite(root->fs_info->sb);
519 kmem_cache_free(btrfs_trans_handle_cachep, h);
520alloc_fail:
521 if (num_bytes)
522 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
523 num_bytes);
524reserve_fail:
525 if (qgroup_reserved)
526 btrfs_qgroup_free(root, qgroup_reserved);
527 return ERR_PTR(ret);
Chris Mason79154b12007-03-22 15:59:16 -0400528}
529
Chris Masonf9295742008-07-17 12:54:14 -0400530struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
Yan, Zhenga22285a2010-05-16 10:48:46 -0400531 int num_items)
Chris Masonf9295742008-07-17 12:54:14 -0400532{
Miao Xie08e007d2012-10-16 11:33:38 +0000533 return start_transaction(root, num_items, TRANS_START,
534 BTRFS_RESERVE_FLUSH_ALL);
Chris Masonf9295742008-07-17 12:54:14 -0400535}
Miao Xie8407aa42012-09-07 01:43:32 -0600536
Miao Xie08e007d2012-10-16 11:33:38 +0000537struct btrfs_trans_handle *btrfs_start_transaction_lflush(
Miao Xie8407aa42012-09-07 01:43:32 -0600538 struct btrfs_root *root, int num_items)
539{
Miao Xie08e007d2012-10-16 11:33:38 +0000540 return start_transaction(root, num_items, TRANS_START,
541 BTRFS_RESERVE_FLUSH_LIMIT);
Miao Xie8407aa42012-09-07 01:43:32 -0600542}
543
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400544struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
Chris Masonf9295742008-07-17 12:54:14 -0400545{
Miao Xie8407aa42012-09-07 01:43:32 -0600546 return start_transaction(root, 0, TRANS_JOIN, 0);
Chris Masonf9295742008-07-17 12:54:14 -0400547}
548
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400549struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
Josef Bacik0af3d002010-06-21 14:48:16 -0400550{
Miao Xie8407aa42012-09-07 01:43:32 -0600551 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -0400552}
553
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400554struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
Sage Weil9ca9ee02008-08-04 10:41:27 -0400555{
Miao Xie8407aa42012-09-07 01:43:32 -0600556 return start_transaction(root, 0, TRANS_USERSPACE, 0);
Sage Weil9ca9ee02008-08-04 10:41:27 -0400557}
558
Miao Xied4edf392013-02-20 09:17:06 +0000559/*
560 * btrfs_attach_transaction() - catch the running transaction
561 *
562 * It is used when we want to commit the current the transaction, but
563 * don't want to start a new one.
564 *
565 * Note: If this function return -ENOENT, it just means there is no
566 * running transaction. But it is possible that the inactive transaction
567 * is still in the memory, not fully on disk. If you hope there is no
568 * inactive transaction in the fs when -ENOENT is returned, you should
569 * invoke
570 * btrfs_attach_transaction_barrier()
571 */
Miao Xie354aa0f2012-09-20 01:54:00 -0600572struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
Josef Bacik60376ce2012-09-14 10:34:40 -0400573{
Miao Xie354aa0f2012-09-20 01:54:00 -0600574 return start_transaction(root, 0, TRANS_ATTACH, 0);
Josef Bacik60376ce2012-09-14 10:34:40 -0400575}
576
Miao Xied4edf392013-02-20 09:17:06 +0000577/*
Wang Sheng-Hui90b6d282013-06-14 16:21:24 +0800578 * btrfs_attach_transaction_barrier() - catch the running transaction
Miao Xied4edf392013-02-20 09:17:06 +0000579 *
580 * It is similar to the above function, the differentia is this one
581 * will wait for all the inactive transactions until they fully
582 * complete.
583 */
584struct btrfs_trans_handle *
585btrfs_attach_transaction_barrier(struct btrfs_root *root)
586{
587 struct btrfs_trans_handle *trans;
588
589 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
590 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
591 btrfs_wait_for_commit(root, 0);
592
593 return trans;
594}
595
Chris Masond352ac62008-09-29 15:18:18 -0400596/* wait for a transaction commit to be fully complete */
Li Zefanb9c83002011-07-14 03:17:14 +0000597static noinline void wait_for_commit(struct btrfs_root *root,
Chris Mason89ce8a62008-06-25 16:01:31 -0400598 struct btrfs_transaction *commit)
599{
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000600 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
Chris Mason89ce8a62008-06-25 16:01:31 -0400601}
602
Sage Weil46204592010-10-29 15:41:32 -0400603int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
604{
605 struct btrfs_transaction *cur_trans = NULL, *t;
Miao Xie8cd28072012-11-26 08:42:07 +0000606 int ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400607
Sage Weil46204592010-10-29 15:41:32 -0400608 if (transid) {
609 if (transid <= root->fs_info->last_trans_committed)
Josef Bacika4abeea2011-04-11 17:25:13 -0400610 goto out;
Sage Weil46204592010-10-29 15:41:32 -0400611
612 /* find specified transaction */
Josef Bacika4abeea2011-04-11 17:25:13 -0400613 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400614 list_for_each_entry(t, &root->fs_info->trans_list, list) {
615 if (t->transid == transid) {
616 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400617 atomic_inc(&cur_trans->use_count);
Miao Xie8cd28072012-11-26 08:42:07 +0000618 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400619 break;
620 }
Miao Xie8cd28072012-11-26 08:42:07 +0000621 if (t->transid > transid) {
622 ret = 0;
Sage Weil46204592010-10-29 15:41:32 -0400623 break;
Miao Xie8cd28072012-11-26 08:42:07 +0000624 }
Sage Weil46204592010-10-29 15:41:32 -0400625 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400626 spin_unlock(&root->fs_info->trans_lock);
Sage Weil42383022014-09-26 08:30:06 -0700627
628 /*
629 * The specified transaction doesn't exist, or we
630 * raced with btrfs_commit_transaction
631 */
632 if (!cur_trans) {
633 if (transid > root->fs_info->last_trans_committed)
634 ret = -EINVAL;
Miao Xie8cd28072012-11-26 08:42:07 +0000635 goto out;
Sage Weil42383022014-09-26 08:30:06 -0700636 }
Sage Weil46204592010-10-29 15:41:32 -0400637 } else {
638 /* find newest transaction that is committing | committed */
Josef Bacika4abeea2011-04-11 17:25:13 -0400639 spin_lock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400640 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
641 list) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000642 if (t->state >= TRANS_STATE_COMMIT_START) {
643 if (t->state == TRANS_STATE_COMPLETED)
Josef Bacik3473f3c2011-06-09 10:15:17 -0400644 break;
Sage Weil46204592010-10-29 15:41:32 -0400645 cur_trans = t;
Josef Bacika4abeea2011-04-11 17:25:13 -0400646 atomic_inc(&cur_trans->use_count);
Sage Weil46204592010-10-29 15:41:32 -0400647 break;
648 }
649 }
Josef Bacika4abeea2011-04-11 17:25:13 -0400650 spin_unlock(&root->fs_info->trans_lock);
Sage Weil46204592010-10-29 15:41:32 -0400651 if (!cur_trans)
Josef Bacika4abeea2011-04-11 17:25:13 -0400652 goto out; /* nothing committing|committed */
Sage Weil46204592010-10-29 15:41:32 -0400653 }
654
Sage Weil46204592010-10-29 15:41:32 -0400655 wait_for_commit(root, cur_trans);
Josef Bacik724e2312013-09-30 11:36:38 -0400656 btrfs_put_transaction(cur_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -0400657out:
Sage Weil46204592010-10-29 15:41:32 -0400658 return ret;
659}
660
Chris Mason37d1aee2008-07-31 10:48:37 -0400661void btrfs_throttle(struct btrfs_root *root)
662{
Josef Bacika4abeea2011-04-11 17:25:13 -0400663 if (!atomic_read(&root->fs_info->open_ioctl_trans))
Sage Weil9ca9ee02008-08-04 10:41:27 -0400664 wait_current_trans(root);
Chris Mason37d1aee2008-07-31 10:48:37 -0400665}
666
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400667static int should_end_transaction(struct btrfs_trans_handle *trans,
668 struct btrfs_root *root)
669{
Josef Bacik1be41b72013-06-12 13:56:06 -0400670 if (root->fs_info->global_block_rsv.space_info->full &&
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500671 btrfs_check_space_for_delayed_refs(trans, root))
Josef Bacik1be41b72013-06-12 13:56:06 -0400672 return 1;
Josef Bacik36ba0222011-10-18 12:15:48 -0400673
Josef Bacik1be41b72013-06-12 13:56:06 -0400674 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400675}
676
677int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
678 struct btrfs_root *root)
679{
680 struct btrfs_transaction *cur_trans = trans->transaction;
681 int updates;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100682 int err;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400683
Josef Bacika4abeea2011-04-11 17:25:13 -0400684 smp_mb();
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000685 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
686 cur_trans->delayed_refs.flushing)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400687 return 1;
688
689 updates = trans->delayed_ref_updates;
690 trans->delayed_ref_updates = 0;
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100691 if (updates) {
692 err = btrfs_run_delayed_refs(trans, root, updates);
693 if (err) /* Error code will also eval true */
694 return err;
695 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400696
697 return should_end_transaction(trans, root);
698}
699
Chris Mason89ce8a62008-06-25 16:01:31 -0400700static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
Miao Xiea698d0752012-09-20 01:51:59 -0600701 struct btrfs_root *root, int throttle)
Chris Mason79154b12007-03-22 15:59:16 -0400702{
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400703 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Masonab78c842008-07-29 16:15:18 -0400704 struct btrfs_fs_info *info = root->fs_info;
Josef Bacik1be41b72013-06-12 13:56:06 -0400705 unsigned long cur = trans->delayed_ref_updates;
Miao Xiea698d0752012-09-20 01:51:59 -0600706 int lock = (trans->type != TRANS_JOIN_NOLOCK);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400707 int err = 0;
Chris Masona79b7d42014-05-22 16:18:52 -0700708 int must_run_delayed_refs = 0;
Chris Masond6e4a422007-04-06 15:37:36 -0400709
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500710 if (trans->use_count > 1) {
711 trans->use_count--;
Josef Bacik2a1eb462011-04-13 15:15:59 -0400712 trans->block_rsv = trans->orig_rsv;
713 return 0;
714 }
715
Josef Bacikb24e03d2011-10-14 14:40:17 -0400716 btrfs_trans_release_metadata(trans, root);
Josef Bacik4c13d752011-08-30 11:31:29 -0400717 trans->block_rsv = NULL;
Arne Jansenc5567232011-09-14 15:44:05 +0200718
Josef Bacikea658ba2012-09-11 16:57:25 -0400719 if (!list_empty(&trans->new_bgs))
720 btrfs_create_pending_block_groups(trans, root);
721
Josef Bacik1be41b72013-06-12 13:56:06 -0400722 trans->delayed_ref_updates = 0;
Chris Masona79b7d42014-05-22 16:18:52 -0700723 if (!trans->sync) {
724 must_run_delayed_refs =
725 btrfs_should_throttle_delayed_refs(trans, root);
Josef Bacik0a2b2a82014-01-23 10:54:11 -0500726 cur = max_t(unsigned long, cur, 32);
Chris Masona79b7d42014-05-22 16:18:52 -0700727
728 /*
729 * don't make the caller wait if they are from a NOLOCK
730 * or ATTACH transaction, it will deadlock with commit
731 */
732 if (must_run_delayed_refs == 1 &&
733 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
734 must_run_delayed_refs = 2;
Chris Mason56bec292009-03-13 10:10:06 -0400735 }
Chris Masonbb721702013-01-29 18:44:12 -0500736
Josef Bacikfcebe452014-05-13 17:30:47 -0700737 if (trans->qgroup_reserved) {
738 /*
739 * the same root has to be passed here between start_transaction
740 * and end_transaction. Subvolume quota depends on this.
741 */
742 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
743 trans->qgroup_reserved = 0;
744 }
745
Josef Bacik0e721102012-06-26 16:13:18 -0400746 btrfs_trans_release_metadata(trans, root);
747 trans->block_rsv = NULL;
Chris Mason56bec292009-03-13 10:10:06 -0400748
Josef Bacikea658ba2012-09-11 16:57:25 -0400749 if (!list_empty(&trans->new_bgs))
750 btrfs_create_pending_block_groups(trans, root);
751
Josef Bacika4abeea2011-04-11 17:25:13 -0400752 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000753 should_end_transaction(trans, root) &&
754 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
755 spin_lock(&info->trans_lock);
756 if (cur_trans->state == TRANS_STATE_RUNNING)
757 cur_trans->state = TRANS_STATE_BLOCKED;
758 spin_unlock(&info->trans_lock);
Josef Bacika4abeea2011-04-11 17:25:13 -0400759 }
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400760
Miao Xie4a9d8bd2013-05-17 03:53:43 +0000761 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500762 if (throttle)
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400763 return btrfs_commit_transaction(trans, root);
Josef Bacik3bbb24b2014-03-06 19:01:07 -0500764 else
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400765 wake_up_process(info->transaction_kthread);
766 }
767
Miao Xie0860adf2013-05-15 07:48:27 +0000768 if (trans->type & __TRANS_FREEZABLE)
Josef Bacik98114652012-09-14 11:22:38 -0400769 sb_end_intwrite(root->fs_info->sb);
Josef Bacik6df78812012-09-05 08:08:30 -0600770
Yan, Zheng8929ecfa2010-05-16 10:49:58 -0400771 WARN_ON(cur_trans != info->running_transaction);
Josef Bacik13c5a932011-04-11 15:45:29 -0400772 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
773 atomic_dec(&cur_trans->num_writers);
Miao Xie0860adf2013-05-15 07:48:27 +0000774 extwriter_counter_dec(cur_trans, trans->type);
Chris Mason89ce8a62008-06-25 16:01:31 -0400775
Sage Weil99d16cb2010-10-29 15:37:34 -0400776 smp_mb();
Chris Mason79154b12007-03-22 15:59:16 -0400777 if (waitqueue_active(&cur_trans->writer_wait))
778 wake_up(&cur_trans->writer_wait);
Josef Bacik724e2312013-09-30 11:36:38 -0400779 btrfs_put_transaction(cur_trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400780
781 if (current->journal_info == trans)
782 current->journal_info = NULL;
Chris Masonab78c842008-07-29 16:15:18 -0400783
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000784 if (throttle)
785 btrfs_run_delayed_iputs(root);
786
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100787 if (trans->aborted ||
Josef Bacik4e121c02013-09-27 16:32:39 -0400788 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
789 wake_up_process(info->transaction_kthread);
Dave Jones4edc2ca2012-04-12 16:03:56 -0400790 err = -EIO;
Josef Bacik4e121c02013-09-27 16:32:39 -0400791 }
Jan Schmidtedf39272012-06-28 18:04:55 +0200792 assert_qgroups_uptodate(trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100793
Dave Jones4edc2ca2012-04-12 16:03:56 -0400794 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Chris Masona79b7d42014-05-22 16:18:52 -0700795 if (must_run_delayed_refs) {
796 btrfs_async_run_delayed_refs(root, cur,
797 must_run_delayed_refs == 1);
798 }
Dave Jones4edc2ca2012-04-12 16:03:56 -0400799 return err;
Chris Mason79154b12007-03-22 15:59:16 -0400800}
801
Chris Mason89ce8a62008-06-25 16:01:31 -0400802int btrfs_end_transaction(struct btrfs_trans_handle *trans,
803 struct btrfs_root *root)
804{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000805 return __btrfs_end_transaction(trans, root, 0);
Chris Mason89ce8a62008-06-25 16:01:31 -0400806}
807
808int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
809 struct btrfs_root *root)
810{
Wang Shilong98ad43be2013-04-14 14:08:49 +0000811 return __btrfs_end_transaction(trans, root, 1);
Miao Xie16cdcec2011-04-22 18:12:22 +0800812}
813
Chris Masond352ac62008-09-29 15:18:18 -0400814/*
815 * when btree blocks are allocated, they have some corresponding bits set for
816 * them in one of two extent_io trees. This is used to make sure all of
Chris Mason690587d2009-10-13 13:29:19 -0400817 * those extents are sent to disk but does not wait on them
Chris Masond352ac62008-09-29 15:18:18 -0400818 */
Chris Mason690587d2009-10-13 13:29:19 -0400819int btrfs_write_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000820 struct extent_io_tree *dirty_pages, int mark)
Chris Mason79154b12007-03-22 15:59:16 -0400821{
Chris Mason777e6bd2008-08-15 15:34:15 -0400822 int err = 0;
Chris Mason7c4452b2007-04-28 09:29:35 -0400823 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400824 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400825 struct extent_state *cached_state = NULL;
Chris Mason777e6bd2008-08-15 15:34:15 -0400826 u64 start = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400827 u64 end;
Chris Mason7c4452b2007-04-28 09:29:35 -0400828
Josef Bacik17283662011-09-26 13:58:47 -0400829 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400830 mark, &cached_state)) {
831 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
832 mark, &cached_state, GFP_NOFS);
833 cached_state = NULL;
Josef Bacik17283662011-09-26 13:58:47 -0400834 err = filemap_fdatawrite_range(mapping, start, end);
835 if (err)
836 werr = err;
837 cond_resched();
838 start = end + 1;
Chris Mason7c4452b2007-04-28 09:29:35 -0400839 }
Chris Mason690587d2009-10-13 13:29:19 -0400840 if (err)
841 werr = err;
842 return werr;
843}
844
845/*
846 * when btree blocks are allocated, they have some corresponding bits set for
847 * them in one of two extent_io trees. This is used to make sure all of
848 * those extents are on disk for transaction or log commit. We wait
849 * on all the pages and clear them from the dirty pages state tree
850 */
851int btrfs_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000852 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400853{
Chris Mason690587d2009-10-13 13:29:19 -0400854 int err = 0;
855 int werr = 0;
Josef Bacik17283662011-09-26 13:58:47 -0400856 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
Josef Bacike6138872012-09-27 17:07:30 -0400857 struct extent_state *cached_state = NULL;
Chris Mason690587d2009-10-13 13:29:19 -0400858 u64 start = 0;
859 u64 end;
Filipe Manana656f30d2014-09-26 12:25:56 +0100860 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
861 bool errors = false;
Chris Mason690587d2009-10-13 13:29:19 -0400862
Josef Bacik17283662011-09-26 13:58:47 -0400863 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -0400864 EXTENT_NEED_WAIT, &cached_state)) {
865 clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
866 0, 0, &cached_state, GFP_NOFS);
Josef Bacik17283662011-09-26 13:58:47 -0400867 err = filemap_fdatawait_range(mapping, start, end);
868 if (err)
869 werr = err;
870 cond_resched();
871 start = end + 1;
Chris Mason777e6bd2008-08-15 15:34:15 -0400872 }
Chris Mason7c4452b2007-04-28 09:29:35 -0400873 if (err)
874 werr = err;
Filipe Manana656f30d2014-09-26 12:25:56 +0100875
876 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
877 if ((mark & EXTENT_DIRTY) &&
878 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
879 &btree_ino->runtime_flags))
880 errors = true;
881
882 if ((mark & EXTENT_NEW) &&
883 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
884 &btree_ino->runtime_flags))
885 errors = true;
886 } else {
887 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
888 &btree_ino->runtime_flags))
889 errors = true;
890 }
891
892 if (errors && !werr)
893 werr = -EIO;
894
Chris Mason7c4452b2007-04-28 09:29:35 -0400895 return werr;
Chris Mason79154b12007-03-22 15:59:16 -0400896}
897
Chris Mason690587d2009-10-13 13:29:19 -0400898/*
899 * when btree blocks are allocated, they have some corresponding bits set for
900 * them in one of two extent_io trees. This is used to make sure all of
901 * those extents are on disk for transaction or log commit
902 */
Sergei Trofimovich171170c2013-08-14 23:27:46 +0300903static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000904 struct extent_io_tree *dirty_pages, int mark)
Chris Mason690587d2009-10-13 13:29:19 -0400905{
906 int ret;
907 int ret2;
Miao Xiec6adc9c2013-05-28 10:05:39 +0000908 struct blk_plug plug;
Chris Mason690587d2009-10-13 13:29:19 -0400909
Miao Xiec6adc9c2013-05-28 10:05:39 +0000910 blk_start_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000911 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
Miao Xiec6adc9c2013-05-28 10:05:39 +0000912 blk_finish_plug(&plug);
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000913 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
Chris Masonbf0da8c2011-11-04 12:29:37 -0400914
915 if (ret)
916 return ret;
917 if (ret2)
918 return ret2;
919 return 0;
Chris Mason690587d2009-10-13 13:29:19 -0400920}
921
Chris Masond0c803c2008-09-11 16:17:57 -0400922int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
923 struct btrfs_root *root)
924{
925 if (!trans || !trans->transaction) {
926 struct inode *btree_inode;
927 btree_inode = root->fs_info->btree_inode;
928 return filemap_write_and_wait(btree_inode->i_mapping);
929 }
930 return btrfs_write_and_wait_marked_extents(root,
Yan, Zheng8cef4e12009-11-12 09:33:26 +0000931 &trans->transaction->dirty_pages,
932 EXTENT_DIRTY);
Chris Masond0c803c2008-09-11 16:17:57 -0400933}
934
Chris Masond352ac62008-09-29 15:18:18 -0400935/*
936 * this is used to update the root pointer in the tree of tree roots.
937 *
938 * But, in the case of the extent allocation tree, updating the root
939 * pointer may allocate blocks which may change the root of the extent
940 * allocation tree.
941 *
942 * So, this loops and repeats and makes sure the cowonly root didn't
943 * change while the root pointer was being updated in the metadata.
944 */
Chris Mason0b86a832008-03-24 15:01:56 -0400945static int update_cowonly_root(struct btrfs_trans_handle *trans,
946 struct btrfs_root *root)
947{
948 int ret;
949 u64 old_root_bytenr;
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000950 u64 old_root_used;
Chris Mason0b86a832008-03-24 15:01:56 -0400951 struct btrfs_root *tree_root = root->fs_info->tree_root;
952
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000953 old_root_used = btrfs_root_used(&root->root_item);
Chris Mason0b86a832008-03-24 15:01:56 -0400954 btrfs_write_dirty_block_groups(trans, root);
Chris Mason56bec292009-03-13 10:10:06 -0400955
Chris Masond3977122009-01-05 21:25:51 -0500956 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -0400957 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000958 if (old_root_bytenr == root->node->start &&
959 old_root_used == btrfs_root_used(&root->root_item))
Chris Mason0b86a832008-03-24 15:01:56 -0400960 break;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400961
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400962 btrfs_set_root_node(&root->root_item, root->node);
Chris Mason0b86a832008-03-24 15:01:56 -0400963 ret = btrfs_update_root(trans, tree_root,
964 &root->root_key,
965 &root->root_item);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100966 if (ret)
967 return ret;
Chris Mason56bec292009-03-13 10:10:06 -0400968
Yan, Zheng86b9f2e2009-11-12 09:36:50 +0000969 old_root_used = btrfs_root_used(&root->root_item);
Yan Zheng4a8c9a62009-07-22 10:07:05 -0400970 ret = btrfs_write_dirty_block_groups(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100971 if (ret)
972 return ret;
Chris Mason0b86a832008-03-24 15:01:56 -0400973 }
Yan Zheng276e6802009-07-30 09:40:40 -0400974
Chris Mason0b86a832008-03-24 15:01:56 -0400975 return 0;
976}
977
Chris Masond352ac62008-09-29 15:18:18 -0400978/*
979 * update all the cowonly tree roots on disk
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100980 *
981 * The error handling in this function may not be obvious. Any of the
982 * failures will cause the file system to go offline. We still need
983 * to clean up the delayed refs.
Chris Masond352ac62008-09-29 15:18:18 -0400984 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400985static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
986 struct btrfs_root *root)
Chris Mason79154b12007-03-22 15:59:16 -0400987{
Chris Mason79154b12007-03-22 15:59:16 -0400988 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0b86a832008-03-24 15:01:56 -0400989 struct list_head *next;
Yan Zheng84234f32008-10-29 14:49:05 -0400990 struct extent_buffer *eb;
Chris Mason56bec292009-03-13 10:10:06 -0400991 int ret;
Yan Zheng84234f32008-10-29 14:49:05 -0400992
Chris Mason56bec292009-03-13 10:10:06 -0400993 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100994 if (ret)
995 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -0400996
Yan Zheng84234f32008-10-29 14:49:05 -0400997 eb = btrfs_lock_root_node(fs_info->tree_root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +0100998 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
999 0, &eb);
Yan Zheng84234f32008-10-29 14:49:05 -04001000 btrfs_tree_unlock(eb);
1001 free_extent_buffer(eb);
Chris Mason79154b12007-03-22 15:59:16 -04001002
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001003 if (ret)
1004 return ret;
1005
Chris Mason56bec292009-03-13 10:10:06 -04001006 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001007 if (ret)
1008 return ret;
Chris Mason87ef2bb2008-10-30 11:23:27 -04001009
Stefan Behrens733f4fb2012-05-25 16:06:10 +02001010 ret = btrfs_run_dev_stats(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001011 if (ret)
1012 return ret;
Stefan Behrens8dabb742012-11-06 13:15:27 +01001013 ret = btrfs_run_dev_replace(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001014 if (ret)
1015 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001016 ret = btrfs_run_qgroups(trans, root->fs_info);
Josef Bacikc16ce192013-09-27 16:38:20 -04001017 if (ret)
1018 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001019
1020 /* run_qgroups might have added some more refs */
1021 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Josef Bacikc16ce192013-09-27 16:38:20 -04001022 if (ret)
1023 return ret;
Jan Schmidt546adb02012-06-14 16:37:44 +02001024
Chris Masond3977122009-01-05 21:25:51 -05001025 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
Chris Mason0b86a832008-03-24 15:01:56 -04001026 next = fs_info->dirty_cowonly_roots.next;
1027 list_del_init(next);
1028 root = list_entry(next, struct btrfs_root, dirty_list);
Chris Mason87ef2bb2008-10-30 11:23:27 -04001029
Josef Bacik9e351cc2014-03-13 15:42:13 -04001030 if (root != fs_info->extent_root)
1031 list_add_tail(&root->dirty_list,
1032 &trans->transaction->switch_commits);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001033 ret = update_cowonly_root(trans, root);
1034 if (ret)
1035 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001036 }
Yan Zheng276e6802009-07-30 09:40:40 -04001037
Josef Bacik9e351cc2014-03-13 15:42:13 -04001038 list_add_tail(&fs_info->extent_root->dirty_list,
1039 &trans->transaction->switch_commits);
Stefan Behrens8dabb742012-11-06 13:15:27 +01001040 btrfs_after_dev_replace_commit(fs_info);
1041
Chris Mason79154b12007-03-22 15:59:16 -04001042 return 0;
1043}
1044
Chris Masond352ac62008-09-29 15:18:18 -04001045/*
1046 * dead roots are old snapshots that need to be deleted. This allocates
1047 * a dirty root struct and adds it into the list of dead roots that need to
1048 * be deleted
1049 */
Josef Bacikcfad3922013-07-25 15:11:47 -04001050void btrfs_add_dead_root(struct btrfs_root *root)
Chris Mason5eda7b52007-06-22 14:16:25 -04001051{
Josef Bacika4abeea2011-04-11 17:25:13 -04001052 spin_lock(&root->fs_info->trans_lock);
Josef Bacikcfad3922013-07-25 15:11:47 -04001053 if (list_empty(&root->root_list))
1054 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
Josef Bacika4abeea2011-04-11 17:25:13 -04001055 spin_unlock(&root->fs_info->trans_lock);
Chris Mason5eda7b52007-06-22 14:16:25 -04001056}
1057
Chris Masond352ac62008-09-29 15:18:18 -04001058/*
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001059 * update all the cowonly tree roots on disk
Chris Masond352ac62008-09-29 15:18:18 -04001060 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001061static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1062 struct btrfs_root *root)
Chris Mason0f7d52f2007-04-09 10:42:37 -04001063{
Chris Mason0f7d52f2007-04-09 10:42:37 -04001064 struct btrfs_root *gang[8];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001065 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001066 int i;
1067 int ret;
Chris Mason54aa1f42007-06-22 14:16:25 -04001068 int err = 0;
1069
Josef Bacika4abeea2011-04-11 17:25:13 -04001070 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Masond3977122009-01-05 21:25:51 -05001071 while (1) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001072 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1073 (void **)gang, 0,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001074 ARRAY_SIZE(gang),
1075 BTRFS_ROOT_TRANS_TAG);
1076 if (ret == 0)
1077 break;
1078 for (i = 0; i < ret; i++) {
1079 root = gang[i];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001080 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1081 (unsigned long)root->root_key.objectid,
1082 BTRFS_ROOT_TRANS_TAG);
Josef Bacika4abeea2011-04-11 17:25:13 -04001083 spin_unlock(&fs_info->fs_roots_radix_lock);
Yan Zheng31153d82008-07-28 15:32:19 -04001084
Chris Masone02119d2008-09-05 16:13:11 -04001085 btrfs_free_log(trans, root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001086 btrfs_update_reloc_root(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001087 btrfs_orphan_commit_root(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001088
Li Zefan82d59022011-04-20 10:33:24 +08001089 btrfs_save_ino_cache(root, trans);
1090
Liu Bof1ebcc72011-11-14 20:48:06 -05001091 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001092 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Chris Masonc7548af2014-06-10 13:06:56 -07001093 smp_mb__after_atomic();
Liu Bof1ebcc72011-11-14 20:48:06 -05001094
Yan Zheng978d9102009-06-15 20:01:02 -04001095 if (root->commit_root != root->node) {
Josef Bacik9e351cc2014-03-13 15:42:13 -04001096 list_add_tail(&root->dirty_list,
1097 &trans->transaction->switch_commits);
Yan Zheng978d9102009-06-15 20:01:02 -04001098 btrfs_set_root_node(&root->root_item,
1099 root->node);
1100 }
Chris Mason9f3a7422007-08-07 15:52:19 -04001101
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001102 err = btrfs_update_root(trans, fs_info->tree_root,
Chris Mason0f7d52f2007-04-09 10:42:37 -04001103 &root->root_key,
1104 &root->root_item);
Josef Bacika4abeea2011-04-11 17:25:13 -04001105 spin_lock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001106 if (err)
1107 break;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001108 }
1109 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001110 spin_unlock(&fs_info->fs_roots_radix_lock);
Chris Mason54aa1f42007-06-22 14:16:25 -04001111 return err;
Chris Mason0f7d52f2007-04-09 10:42:37 -04001112}
1113
Chris Masond352ac62008-09-29 15:18:18 -04001114/*
Eric Sandeende78b512013-01-31 18:21:12 +00001115 * defrag a given btree.
1116 * Every leaf in the btree is read and defragged.
Chris Masond352ac62008-09-29 15:18:18 -04001117 */
Eric Sandeende78b512013-01-31 18:21:12 +00001118int btrfs_defrag_root(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001119{
1120 struct btrfs_fs_info *info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001121 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001122 int ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001123
Miao Xie27cdeb72014-04-02 19:51:05 +08001124 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
Chris Masone9d0b132007-08-10 14:06:19 -04001125 return 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001126
Chris Mason6b800532007-10-15 16:17:34 -04001127 while (1) {
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001128 trans = btrfs_start_transaction(root, 0);
1129 if (IS_ERR(trans))
1130 return PTR_ERR(trans);
1131
Eric Sandeende78b512013-01-31 18:21:12 +00001132 ret = btrfs_defrag_leaves(trans, root);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001133
Chris Masone9d0b132007-08-10 14:06:19 -04001134 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00001135 btrfs_btree_balance_dirty(info->tree_root);
Chris Masone9d0b132007-08-10 14:06:19 -04001136 cond_resched();
1137
David Sterba7841cb22011-05-31 18:07:27 +02001138 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
Chris Masone9d0b132007-08-10 14:06:19 -04001139 break;
David Sterba210549e2013-02-09 23:38:06 +00001140
1141 if (btrfs_defrag_cancelled(root->fs_info)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05001142 pr_debug("BTRFS: defrag_root cancelled\n");
David Sterba210549e2013-02-09 23:38:06 +00001143 ret = -EAGAIN;
1144 break;
1145 }
Chris Masone9d0b132007-08-10 14:06:19 -04001146 }
Miao Xie27cdeb72014-04-02 19:51:05 +08001147 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001148 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001149}
1150
Chris Masond352ac62008-09-29 15:18:18 -04001151/*
1152 * new snapshots need to be created at a very specific time in the
Miao Xieaec80302013-03-04 09:44:29 +00001153 * transaction commit. This does the actual creation.
1154 *
1155 * Note:
1156 * If the error which may affect the commitment of the current transaction
1157 * happens, we should return the error number. If the error which just affect
1158 * the creation of the pending snapshots, just return 0.
Chris Masond352ac62008-09-29 15:18:18 -04001159 */
Chris Mason80b67942008-02-01 16:35:04 -05001160static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
Chris Mason3063d292008-01-08 15:46:30 -05001161 struct btrfs_fs_info *fs_info,
1162 struct btrfs_pending_snapshot *pending)
1163{
1164 struct btrfs_key key;
Chris Mason80b67942008-02-01 16:35:04 -05001165 struct btrfs_root_item *new_root_item;
Chris Mason3063d292008-01-08 15:46:30 -05001166 struct btrfs_root *tree_root = fs_info->tree_root;
1167 struct btrfs_root *root = pending->root;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001168 struct btrfs_root *parent_root;
Liu Bo98c99422011-09-11 10:52:24 -04001169 struct btrfs_block_rsv *rsv;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001170 struct inode *parent_inode;
Miao Xie42874b32012-09-06 04:03:32 -06001171 struct btrfs_path *path;
1172 struct btrfs_dir_item *dir_item;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001173 struct dentry *dentry;
Chris Mason3063d292008-01-08 15:46:30 -05001174 struct extent_buffer *tmp;
Chris Mason925baed2008-06-25 16:01:30 -04001175 struct extent_buffer *old;
Alexander Block8ea05e32012-07-25 17:35:53 +02001176 struct timespec cur_time = CURRENT_TIME;
Miao Xieaec80302013-03-04 09:44:29 +00001177 int ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001178 u64 to_reserve = 0;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001179 u64 index = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001180 u64 objectid;
Li Zefanb83cc962010-12-20 16:04:08 +08001181 u64 root_flags;
Alexander Block8ea05e32012-07-25 17:35:53 +02001182 uuid_le new_uuid;
Chris Mason3063d292008-01-08 15:46:30 -05001183
Miao Xie42874b32012-09-06 04:03:32 -06001184 path = btrfs_alloc_path();
1185 if (!path) {
Miao Xieaec80302013-03-04 09:44:29 +00001186 pending->error = -ENOMEM;
1187 return 0;
Miao Xie42874b32012-09-06 04:03:32 -06001188 }
1189
Chris Mason80b67942008-02-01 16:35:04 -05001190 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1191 if (!new_root_item) {
Miao Xieaec80302013-03-04 09:44:29 +00001192 pending->error = -ENOMEM;
Miao Xie6fa97002012-09-06 04:00:32 -06001193 goto root_item_alloc_fail;
Chris Mason80b67942008-02-01 16:35:04 -05001194 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001195
Miao Xieaec80302013-03-04 09:44:29 +00001196 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1197 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001198 goto no_free_objectid;
Chris Mason3063d292008-01-08 15:46:30 -05001199
Yan, Zheng3fd0a552010-05-16 10:49:59 -04001200 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001201
1202 if (to_reserve > 0) {
Miao Xieaec80302013-03-04 09:44:29 +00001203 pending->error = btrfs_block_rsv_add(root,
1204 &pending->block_rsv,
1205 to_reserve,
1206 BTRFS_RESERVE_NO_FLUSH);
1207 if (pending->error)
Miao Xie6fa97002012-09-06 04:00:32 -06001208 goto no_free_objectid;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001209 }
1210
Chris Mason3063d292008-01-08 15:46:30 -05001211 key.objectid = objectid;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001212 key.offset = (u64)-1;
1213 key.type = BTRFS_ROOT_ITEM_KEY;
Chris Mason3063d292008-01-08 15:46:30 -05001214
Miao Xie6fa97002012-09-06 04:00:32 -06001215 rsv = trans->block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001216 trans->block_rsv = &pending->block_rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001217 trans->bytes_reserved = trans->block_rsv->reserved;
Sage Weil6bdb72d2010-03-15 17:27:13 +00001218
Yan, Zhenga22285a2010-05-16 10:48:46 -04001219 dentry = pending->dentry;
Miao Xiee9662f72013-02-28 10:01:15 +00001220 parent_inode = pending->dir;
Yan, Zhenga22285a2010-05-16 10:48:46 -04001221 parent_root = BTRFS_I(parent_inode)->root;
Chris Mason75857172011-06-13 20:00:16 -04001222 record_root_in_trans(trans, parent_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001223
Sage Weil6bdb72d2010-03-15 17:27:13 +00001224 /*
1225 * insert the directory item
1226 */
Sage Weil6bdb72d2010-03-15 17:27:13 +00001227 ret = btrfs_set_inode_index(parent_inode, &index);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001228 BUG_ON(ret); /* -ENOMEM */
Miao Xie42874b32012-09-06 04:03:32 -06001229
1230 /* check if there is a file/dir which has the same name. */
1231 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1232 btrfs_ino(parent_inode),
1233 dentry->d_name.name,
1234 dentry->d_name.len, 0);
1235 if (dir_item != NULL && !IS_ERR(dir_item)) {
Chris Masonfe66a052012-02-20 08:40:56 -05001236 pending->error = -EEXIST;
Miao Xieaec80302013-03-04 09:44:29 +00001237 goto dir_item_existed;
Miao Xie42874b32012-09-06 04:03:32 -06001238 } else if (IS_ERR(dir_item)) {
1239 ret = PTR_ERR(dir_item);
Miao Xie8732d442012-09-17 23:52:38 -06001240 btrfs_abort_transaction(trans, root, ret);
1241 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001242 }
Miao Xie42874b32012-09-06 04:03:32 -06001243 btrfs_release_path(path);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001244
Chris Masone9993762011-06-17 16:14:09 -04001245 /*
1246 * pull in the delayed directory update
1247 * and the delayed inode item
1248 * otherwise we corrupt the FS during
1249 * snapshot
1250 */
1251 ret = btrfs_run_delayed_items(trans, root);
Miao Xie8732d442012-09-17 23:52:38 -06001252 if (ret) { /* Transaction aborted */
1253 btrfs_abort_transaction(trans, root, ret);
1254 goto fail;
1255 }
Chris Masone9993762011-06-17 16:14:09 -04001256
Chris Mason75857172011-06-13 20:00:16 -04001257 record_root_in_trans(trans, root);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001258 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1259 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
Li Zefan08fe4db2011-03-28 02:01:25 +00001260 btrfs_check_and_init_root_item(new_root_item);
Sage Weil6bdb72d2010-03-15 17:27:13 +00001261
Li Zefanb83cc962010-12-20 16:04:08 +08001262 root_flags = btrfs_root_flags(new_root_item);
1263 if (pending->readonly)
1264 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1265 else
1266 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1267 btrfs_set_root_flags(new_root_item, root_flags);
1268
Alexander Block8ea05e32012-07-25 17:35:53 +02001269 btrfs_set_root_generation_v2(new_root_item,
1270 trans->transid);
1271 uuid_le_gen(&new_uuid);
1272 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1273 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1274 BTRFS_UUID_SIZE);
Stefan Behrens70023da2013-04-17 09:11:47 +00001275 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1276 memset(new_root_item->received_uuid, 0,
1277 sizeof(new_root_item->received_uuid));
1278 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1279 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1280 btrfs_set_root_stransid(new_root_item, 0);
1281 btrfs_set_root_rtransid(new_root_item, 0);
1282 }
Qu Wenruo3cae2102013-07-16 11:19:18 +08001283 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1284 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
Alexander Block8ea05e32012-07-25 17:35:53 +02001285 btrfs_set_root_otransid(new_root_item, trans->transid);
Alexander Block8ea05e32012-07-25 17:35:53 +02001286
Chris Mason925baed2008-06-25 16:01:30 -04001287 old = btrfs_lock_root_node(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001288 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001289 if (ret) {
1290 btrfs_tree_unlock(old);
1291 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001292 btrfs_abort_transaction(trans, root, ret);
1293 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001294 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001295
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001296 btrfs_set_lock_blocking(old);
Chris Mason3063d292008-01-08 15:46:30 -05001297
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001298 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001299 /* clean up in any case */
Chris Mason925baed2008-06-25 16:01:30 -04001300 btrfs_tree_unlock(old);
1301 free_extent_buffer(old);
Miao Xie8732d442012-09-17 23:52:38 -06001302 if (ret) {
1303 btrfs_abort_transaction(trans, root, ret);
1304 goto fail;
1305 }
Chris Mason3063d292008-01-08 15:46:30 -05001306
Josef Bacikfcebe452014-05-13 17:30:47 -07001307 /*
1308 * We need to flush delayed refs in order to make sure all of our quota
1309 * operations have been done before we call btrfs_qgroup_inherit.
1310 */
1311 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1312 if (ret) {
1313 btrfs_abort_transaction(trans, root, ret);
1314 goto fail;
1315 }
1316
Eric Sandeen47a306a2014-06-12 00:53:44 -05001317 ret = btrfs_qgroup_inherit(trans, fs_info,
1318 root->root_key.objectid,
1319 objectid, pending->inherit);
1320 if (ret) {
1321 btrfs_abort_transaction(trans, root, ret);
1322 goto fail;
1323 }
Josef Bacikfcebe452014-05-13 17:30:47 -07001324
Liu Bof1ebcc72011-11-14 20:48:06 -05001325 /* see comments in should_cow_block() */
Miao Xie27cdeb72014-04-02 19:51:05 +08001326 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
Liu Bof1ebcc72011-11-14 20:48:06 -05001327 smp_wmb();
1328
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001329 btrfs_set_root_node(new_root_item, tmp);
Yan, Zhenga22285a2010-05-16 10:48:46 -04001330 /* record when the snapshot was created in key.offset */
1331 key.offset = trans->transid;
1332 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
Chris Mason925baed2008-06-25 16:01:30 -04001333 btrfs_tree_unlock(tmp);
Chris Mason3063d292008-01-08 15:46:30 -05001334 free_extent_buffer(tmp);
Miao Xie8732d442012-09-17 23:52:38 -06001335 if (ret) {
1336 btrfs_abort_transaction(trans, root, ret);
1337 goto fail;
1338 }
Chris Mason0660b5a2008-11-17 20:37:39 -05001339
Yan, Zhenga22285a2010-05-16 10:48:46 -04001340 /*
1341 * insert root back/forward references
1342 */
1343 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1344 parent_root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001345 btrfs_ino(parent_inode), index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04001346 dentry->d_name.name, dentry->d_name.len);
Miao Xie8732d442012-09-17 23:52:38 -06001347 if (ret) {
1348 btrfs_abort_transaction(trans, root, ret);
1349 goto fail;
1350 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04001351
1352 key.offset = (u64)-1;
1353 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001354 if (IS_ERR(pending->snap)) {
1355 ret = PTR_ERR(pending->snap);
Miao Xie8732d442012-09-17 23:52:38 -06001356 btrfs_abort_transaction(trans, root, ret);
1357 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001358 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04001359
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001360 ret = btrfs_reloc_post_snapshot(trans, pending);
Miao Xie8732d442012-09-17 23:52:38 -06001361 if (ret) {
1362 btrfs_abort_transaction(trans, root, ret);
1363 goto fail;
1364 }
Miao Xie361048f2012-09-06 04:00:57 -06001365
1366 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Miao Xie8732d442012-09-17 23:52:38 -06001367 if (ret) {
1368 btrfs_abort_transaction(trans, root, ret);
1369 goto fail;
1370 }
Miao Xie42874b32012-09-06 04:03:32 -06001371
1372 ret = btrfs_insert_dir_item(trans, parent_root,
1373 dentry->d_name.name, dentry->d_name.len,
1374 parent_inode, &key,
1375 BTRFS_FT_DIR, index);
1376 /* We have check then name at the beginning, so it is impossible. */
Chris Mason9c520572012-12-17 14:26:57 -05001377 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
Miao Xie8732d442012-09-17 23:52:38 -06001378 if (ret) {
1379 btrfs_abort_transaction(trans, root, ret);
1380 goto fail;
1381 }
Miao Xie42874b32012-09-06 04:03:32 -06001382
1383 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1384 dentry->d_name.len * 2);
1385 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
Josef Bacikbe6aef62012-10-22 15:43:12 -04001386 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001387 if (ret) {
Miao Xie8732d442012-09-17 23:52:38 -06001388 btrfs_abort_transaction(trans, root, ret);
Stefan Behrensdd5f9612013-08-15 17:11:20 +02001389 goto fail;
1390 }
1391 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1392 BTRFS_UUID_KEY_SUBVOL, objectid);
1393 if (ret) {
1394 btrfs_abort_transaction(trans, root, ret);
1395 goto fail;
1396 }
1397 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1398 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1399 new_root_item->received_uuid,
1400 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1401 objectid);
1402 if (ret && ret != -EEXIST) {
1403 btrfs_abort_transaction(trans, root, ret);
1404 goto fail;
1405 }
1406 }
Chris Mason3063d292008-01-08 15:46:30 -05001407fail:
Miao Xieaec80302013-03-04 09:44:29 +00001408 pending->error = ret;
1409dir_item_existed:
Liu Bo98c99422011-09-11 10:52:24 -04001410 trans->block_rsv = rsv;
Liu Bo2382c5c2013-02-22 04:33:36 +00001411 trans->bytes_reserved = 0;
Miao Xie6fa97002012-09-06 04:00:32 -06001412no_free_objectid:
1413 kfree(new_root_item);
1414root_item_alloc_fail:
Miao Xie42874b32012-09-06 04:03:32 -06001415 btrfs_free_path(path);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001416 return ret;
Chris Mason3063d292008-01-08 15:46:30 -05001417}
1418
Chris Masond352ac62008-09-29 15:18:18 -04001419/*
1420 * create all the snapshots we've scheduled for creation
1421 */
Chris Mason80b67942008-02-01 16:35:04 -05001422static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1423 struct btrfs_fs_info *fs_info)
Chris Mason3063d292008-01-08 15:46:30 -05001424{
Miao Xieaec80302013-03-04 09:44:29 +00001425 struct btrfs_pending_snapshot *pending, *next;
Chris Mason3063d292008-01-08 15:46:30 -05001426 struct list_head *head = &trans->transaction->pending_snapshots;
Miao Xieaec80302013-03-04 09:44:29 +00001427 int ret = 0;
Chris Mason3de45862008-11-17 21:02:50 -05001428
Miao Xieaec80302013-03-04 09:44:29 +00001429 list_for_each_entry_safe(pending, next, head, list) {
1430 list_del(&pending->list);
1431 ret = create_pending_snapshot(trans, fs_info, pending);
1432 if (ret)
1433 break;
1434 }
1435 return ret;
Chris Mason3de45862008-11-17 21:02:50 -05001436}
1437
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001438static void update_super_roots(struct btrfs_root *root)
1439{
1440 struct btrfs_root_item *root_item;
1441 struct btrfs_super_block *super;
1442
David Sterba6c417612011-04-13 15:41:04 +02001443 super = root->fs_info->super_copy;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001444
1445 root_item = &root->fs_info->chunk_root->root_item;
1446 super->chunk_root = root_item->bytenr;
1447 super->chunk_root_generation = root_item->generation;
1448 super->chunk_root_level = root_item->level;
1449
1450 root_item = &root->fs_info->tree_root->root_item;
1451 super->root = root_item->bytenr;
1452 super->generation = root_item->generation;
1453 super->root_level = root_item->level;
Josef Bacik73bc1872011-10-03 14:07:49 -04001454 if (btrfs_test_opt(root, SPACE_CACHE))
Josef Bacik0af3d002010-06-21 14:48:16 -04001455 super->cache_generation = root_item->generation;
Stefan Behrens70f80172013-08-15 17:11:23 +02001456 if (root->fs_info->update_uuid_tree_gen)
1457 super->uuid_tree_generation = root_item->generation;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001458}
1459
Chris Masonf36f3042009-07-30 10:04:48 -04001460int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1461{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001462 struct btrfs_transaction *trans;
Chris Masonf36f3042009-07-30 10:04:48 -04001463 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001464
Josef Bacika4abeea2011-04-11 17:25:13 -04001465 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001466 trans = info->running_transaction;
1467 if (trans)
1468 ret = (trans->state >= TRANS_STATE_COMMIT_START);
Josef Bacika4abeea2011-04-11 17:25:13 -04001469 spin_unlock(&info->trans_lock);
Chris Masonf36f3042009-07-30 10:04:48 -04001470 return ret;
1471}
1472
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001473int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1474{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001475 struct btrfs_transaction *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001476 int ret = 0;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001477
Josef Bacika4abeea2011-04-11 17:25:13 -04001478 spin_lock(&info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001479 trans = info->running_transaction;
1480 if (trans)
1481 ret = is_transaction_blocked(trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001482 spin_unlock(&info->trans_lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04001483 return ret;
1484}
1485
Sage Weilbb9c12c2010-10-29 15:37:34 -04001486/*
1487 * wait for the current transaction commit to start and block subsequent
1488 * transaction joins
1489 */
1490static void wait_current_trans_commit_start(struct btrfs_root *root,
1491 struct btrfs_transaction *trans)
1492{
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001493 wait_event(root->fs_info->transaction_blocked_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001494 trans->state >= TRANS_STATE_COMMIT_START ||
1495 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001496}
1497
1498/*
1499 * wait for the current transaction to start and then become unblocked.
1500 * caller holds ref.
1501 */
1502static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1503 struct btrfs_transaction *trans)
1504{
Li Zefan72d63ed2011-07-14 03:17:00 +00001505 wait_event(root->fs_info->transaction_wait,
Josef Bacik501407a2013-06-10 16:47:23 -04001506 trans->state >= TRANS_STATE_UNBLOCKED ||
1507 trans->aborted);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001508}
1509
1510/*
1511 * commit transactions asynchronously. once btrfs_commit_transaction_async
1512 * returns, any subsequent transaction will not be allowed to join.
1513 */
1514struct btrfs_async_commit {
1515 struct btrfs_trans_handle *newtrans;
1516 struct btrfs_root *root;
Miao Xie7892b5a2012-11-15 08:14:47 +00001517 struct work_struct work;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001518};
1519
1520static void do_async_commit(struct work_struct *work)
1521{
1522 struct btrfs_async_commit *ac =
Miao Xie7892b5a2012-11-15 08:14:47 +00001523 container_of(work, struct btrfs_async_commit, work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001524
Sage Weil6fc4e35482012-08-30 16:26:15 -06001525 /*
1526 * We've got freeze protection passed with the transaction.
1527 * Tell lockdep about it.
1528 */
Liu Bob1a06a42013-11-06 16:57:55 +08001529 if (ac->newtrans->type & __TRANS_FREEZABLE)
Miao Xieff7c1d32012-11-26 08:41:29 +00001530 rwsem_acquire_read(
1531 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1532 0, 1, _THIS_IP_);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001533
Sage Weile209db72012-08-30 16:26:16 -06001534 current->journal_info = ac->newtrans;
1535
Sage Weilbb9c12c2010-10-29 15:37:34 -04001536 btrfs_commit_transaction(ac->newtrans, ac->root);
1537 kfree(ac);
1538}
1539
1540int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1541 struct btrfs_root *root,
1542 int wait_for_unblock)
1543{
1544 struct btrfs_async_commit *ac;
1545 struct btrfs_transaction *cur_trans;
1546
1547 ac = kmalloc(sizeof(*ac), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001548 if (!ac)
1549 return -ENOMEM;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001550
Miao Xie7892b5a2012-11-15 08:14:47 +00001551 INIT_WORK(&ac->work, do_async_commit);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001552 ac->root = root;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001553 ac->newtrans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001554 if (IS_ERR(ac->newtrans)) {
1555 int err = PTR_ERR(ac->newtrans);
1556 kfree(ac);
1557 return err;
1558 }
Sage Weilbb9c12c2010-10-29 15:37:34 -04001559
1560 /* take transaction reference */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001561 cur_trans = trans->transaction;
Josef Bacik13c5a932011-04-11 15:45:29 -04001562 atomic_inc(&cur_trans->use_count);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001563
1564 btrfs_end_transaction(trans, root);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001565
1566 /*
1567 * Tell lockdep we've released the freeze rwsem, since the
1568 * async commit thread will be the one to unlock it.
1569 */
Liu Bob1a06a42013-11-06 16:57:55 +08001570 if (ac->newtrans->type & __TRANS_FREEZABLE)
Miao Xieff7c1d32012-11-26 08:41:29 +00001571 rwsem_release(
1572 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1573 1, _THIS_IP_);
Sage Weil6fc4e35482012-08-30 16:26:15 -06001574
Miao Xie7892b5a2012-11-15 08:14:47 +00001575 schedule_work(&ac->work);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001576
1577 /* wait for transaction to start and unblock */
Sage Weilbb9c12c2010-10-29 15:37:34 -04001578 if (wait_for_unblock)
1579 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1580 else
1581 wait_current_trans_commit_start(root, cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001582
Sage Weil38e88052011-06-10 18:43:13 +00001583 if (current->journal_info == trans)
1584 current->journal_info = NULL;
1585
Josef Bacik724e2312013-09-30 11:36:38 -04001586 btrfs_put_transaction(cur_trans);
Sage Weilbb9c12c2010-10-29 15:37:34 -04001587 return 0;
1588}
1589
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001590
1591static void cleanup_transaction(struct btrfs_trans_handle *trans,
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001592 struct btrfs_root *root, int err)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001593{
1594 struct btrfs_transaction *cur_trans = trans->transaction;
Liu Bof094ac32013-02-27 13:28:25 +00001595 DEFINE_WAIT(wait);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001596
1597 WARN_ON(trans->use_count > 1);
1598
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001599 btrfs_abort_transaction(trans, root, err);
1600
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001601 spin_lock(&root->fs_info->trans_lock);
Liu Bo66b61352013-03-04 16:25:41 +00001602
Miao Xie25d8c282013-05-15 07:48:26 +00001603 /*
1604 * If the transaction is removed from the list, it means this
1605 * transaction has been committed successfully, so it is impossible
1606 * to call the cleanup function.
1607 */
1608 BUG_ON(list_empty(&cur_trans->list));
Liu Bo66b61352013-03-04 16:25:41 +00001609
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001610 list_del_init(&cur_trans->list);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001611 if (cur_trans == root->fs_info->running_transaction) {
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001612 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Liu Bof094ac32013-02-27 13:28:25 +00001613 spin_unlock(&root->fs_info->trans_lock);
1614 wait_event(cur_trans->writer_wait,
1615 atomic_read(&cur_trans->num_writers) == 1);
1616
1617 spin_lock(&root->fs_info->trans_lock);
Josef Bacikd7096fc2012-05-31 15:49:57 -04001618 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001619 spin_unlock(&root->fs_info->trans_lock);
1620
1621 btrfs_cleanup_one_transaction(trans->transaction, root);
1622
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001623 spin_lock(&root->fs_info->trans_lock);
1624 if (cur_trans == root->fs_info->running_transaction)
1625 root->fs_info->running_transaction = NULL;
1626 spin_unlock(&root->fs_info->trans_lock);
1627
Josef Bacike0228282013-09-20 22:26:29 -04001628 if (trans->type & __TRANS_FREEZABLE)
1629 sb_end_intwrite(root->fs_info->sb);
Josef Bacik724e2312013-09-30 11:36:38 -04001630 btrfs_put_transaction(cur_trans);
1631 btrfs_put_transaction(cur_trans);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001632
1633 trace_btrfs_transaction_commit(root);
1634
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001635 if (current->journal_info == trans)
1636 current->journal_info = NULL;
Wang Shilongc0af8f02014-02-19 19:24:18 +08001637 btrfs_scrub_cancel(root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001638
1639 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1640}
1641
Miao Xie82436612013-05-15 07:48:28 +00001642static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1643{
1644 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
Miao Xie6c255e62014-03-06 13:55:01 +08001645 return btrfs_start_delalloc_roots(fs_info, 1, -1);
Miao Xie82436612013-05-15 07:48:28 +00001646 return 0;
1647}
1648
1649static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1650{
1651 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
Miao Xieb0244192013-11-04 23:13:25 +08001652 btrfs_wait_ordered_roots(fs_info, -1);
Miao Xie82436612013-05-15 07:48:28 +00001653}
1654
Chris Mason79154b12007-03-22 15:59:16 -04001655int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1656 struct btrfs_root *root)
1657{
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001658 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason8fd17792007-04-19 21:01:03 -04001659 struct btrfs_transaction *prev_trans = NULL;
Filipe Manana656f30d2014-09-26 12:25:56 +01001660 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
Miao Xie25287e02012-10-25 09:31:03 +00001661 int ret;
Chris Mason79154b12007-03-22 15:59:16 -04001662
Miao Xie8d25a082013-01-15 06:27:25 +00001663 /* Stop the commit early if ->aborted is set */
1664 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
Miao Xie25287e02012-10-25 09:31:03 +00001665 ret = cur_trans->aborted;
Josef Bacike4a2bca2013-02-06 16:55:41 -05001666 btrfs_end_transaction(trans, root);
1667 return ret;
Miao Xie25287e02012-10-25 09:31:03 +00001668 }
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001669
Chris Mason56bec292009-03-13 10:10:06 -04001670 /* make a pass through all the delayed refs we have so far
1671 * any runnings procs may add more while we are here
1672 */
1673 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001674 if (ret) {
1675 btrfs_end_transaction(trans, root);
1676 return ret;
1677 }
Chris Mason56bec292009-03-13 10:10:06 -04001678
Josef Bacik0e721102012-06-26 16:13:18 -04001679 btrfs_trans_release_metadata(trans, root);
1680 trans->block_rsv = NULL;
Miao Xie272d26d2013-02-20 14:16:39 +00001681 if (trans->qgroup_reserved) {
1682 btrfs_qgroup_free(root, trans->qgroup_reserved);
1683 trans->qgroup_reserved = 0;
1684 }
Josef Bacik0e721102012-06-26 16:13:18 -04001685
Chris Masonb7ec40d2009-03-12 20:12:45 -04001686 cur_trans = trans->transaction;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001687
Chris Mason56bec292009-03-13 10:10:06 -04001688 /*
1689 * set the flushing flag so procs in this transaction have to
1690 * start sending their work down.
1691 */
Chris Masonb7ec40d2009-03-12 20:12:45 -04001692 cur_trans->delayed_refs.flushing = 1;
Josef Bacik1be41b72013-06-12 13:56:06 -04001693 smp_wmb();
Chris Mason56bec292009-03-13 10:10:06 -04001694
Josef Bacikea658ba2012-09-11 16:57:25 -04001695 if (!list_empty(&trans->new_bgs))
1696 btrfs_create_pending_block_groups(trans, root);
1697
Chris Masonc3e69d52009-03-13 10:17:05 -04001698 ret = btrfs_run_delayed_refs(trans, root, 0);
Josef Bacike4a2bca2013-02-06 16:55:41 -05001699 if (ret) {
1700 btrfs_end_transaction(trans, root);
1701 return ret;
1702 }
Chris Mason56bec292009-03-13 10:10:06 -04001703
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001704 spin_lock(&root->fs_info->trans_lock);
1705 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1706 spin_unlock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001707 atomic_inc(&cur_trans->use_count);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001708 ret = btrfs_end_transaction(trans, root);
Chris Masonccd467d2007-06-28 15:57:36 -04001709
Li Zefanb9c83002011-07-14 03:17:14 +00001710 wait_for_commit(root, cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001711
Josef Bacik724e2312013-09-30 11:36:38 -04001712 btrfs_put_transaction(cur_trans);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001713
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001714 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001715 }
Chris Mason4313b392008-01-03 09:08:48 -05001716
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001717 cur_trans->state = TRANS_STATE_COMMIT_START;
Sage Weilbb9c12c2010-10-29 15:37:34 -04001718 wake_up(&root->fs_info->transaction_blocked_wait);
1719
Chris Masonccd467d2007-06-28 15:57:36 -04001720 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1721 prev_trans = list_entry(cur_trans->list.prev,
1722 struct btrfs_transaction, list);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001723 if (prev_trans->state != TRANS_STATE_COMPLETED) {
Josef Bacik13c5a932011-04-11 15:45:29 -04001724 atomic_inc(&prev_trans->use_count);
Josef Bacika4abeea2011-04-11 17:25:13 -04001725 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001726
1727 wait_for_commit(root, prev_trans);
Chris Masonccd467d2007-06-28 15:57:36 -04001728
Josef Bacik724e2312013-09-30 11:36:38 -04001729 btrfs_put_transaction(prev_trans);
Josef Bacika4abeea2011-04-11 17:25:13 -04001730 } else {
1731 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001732 }
Josef Bacika4abeea2011-04-11 17:25:13 -04001733 } else {
1734 spin_unlock(&root->fs_info->trans_lock);
Chris Masonccd467d2007-06-28 15:57:36 -04001735 }
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001736
Miao Xie0860adf2013-05-15 07:48:27 +00001737 extwriter_counter_dec(cur_trans, trans->type);
1738
Miao Xie82436612013-05-15 07:48:28 +00001739 ret = btrfs_start_delalloc_flush(root->fs_info);
1740 if (ret)
1741 goto cleanup_transaction;
1742
Chris Mason8d875f92014-08-12 10:47:42 -07001743 ret = btrfs_run_delayed_items(trans, root);
Miao Xie581227d2013-05-15 07:48:30 +00001744 if (ret)
1745 goto cleanup_transaction;
Chris Mason89573b92009-03-12 20:12:45 -04001746
Miao Xie581227d2013-05-15 07:48:30 +00001747 wait_event(cur_trans->writer_wait,
1748 extwriter_counter_read(cur_trans) == 0);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001749
Miao Xie581227d2013-05-15 07:48:30 +00001750 /* some pending stuffs might be added after the previous flush. */
Chris Mason8d875f92014-08-12 10:47:42 -07001751 ret = btrfs_run_delayed_items(trans, root);
Miao Xieca469632012-11-01 07:33:14 +00001752 if (ret)
1753 goto cleanup_transaction;
1754
Miao Xie82436612013-05-15 07:48:28 +00001755 btrfs_wait_delalloc_flush(root->fs_info);
Wang Shilongcb7ab022013-12-04 21:16:53 +08001756
1757 btrfs_scrub_pause(root);
Chris Mason75857172011-06-13 20:00:16 -04001758 /*
Josef Baciked0ca142011-06-14 16:22:15 -04001759 * Ok now we need to make sure to block out any other joins while we
1760 * commit the transaction. We could have started a join before setting
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001761 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
Josef Baciked0ca142011-06-14 16:22:15 -04001762 */
1763 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001764 cur_trans->state = TRANS_STATE_COMMIT_DOING;
Josef Baciked0ca142011-06-14 16:22:15 -04001765 spin_unlock(&root->fs_info->trans_lock);
1766 wait_event(cur_trans->writer_wait,
1767 atomic_read(&cur_trans->num_writers) == 1);
1768
Miao Xie2cba30f2013-01-15 06:29:12 +00001769 /* ->aborted might be set after the previous check, so check it */
1770 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1771 ret = cur_trans->aborted;
Wang Shilong6cf7f772014-02-19 19:24:16 +08001772 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00001773 }
Chris Masone038dca2011-06-17 14:16:13 -04001774 /*
Chris Mason75857172011-06-13 20:00:16 -04001775 * the reloc mutex makes sure that we stop
1776 * the balancing code from coming in and moving
1777 * extents around in the middle of the commit
1778 */
1779 mutex_lock(&root->fs_info->reloc_mutex);
1780
Miao Xie42874b32012-09-06 04:03:32 -06001781 /*
1782 * We needn't worry about the delayed items because we will
1783 * deal with them in create_pending_snapshot(), which is the
1784 * core function of the snapshot creation.
1785 */
1786 ret = create_pending_snapshots(trans, root->fs_info);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001787 if (ret) {
1788 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001789 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001790 }
Chris Mason3063d292008-01-08 15:46:30 -05001791
Miao Xie42874b32012-09-06 04:03:32 -06001792 /*
1793 * We insert the dir indexes of the snapshots and update the inode
1794 * of the snapshots' parents after the snapshot creation, so there
1795 * are some delayed items which are not dealt with. Now deal with
1796 * them.
1797 *
1798 * We needn't worry that this operation will corrupt the snapshots,
1799 * because all the tree which are snapshoted will be forced to COW
1800 * the nodes and leaves.
1801 */
1802 ret = btrfs_run_delayed_items(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001803 if (ret) {
1804 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001805 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001806 }
Miao Xie16cdcec2011-04-22 18:12:22 +08001807
Chris Mason56bec292009-03-13 10:10:06 -04001808 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001809 if (ret) {
1810 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001811 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001812 }
Chris Mason56bec292009-03-13 10:10:06 -04001813
Chris Masone9993762011-06-17 16:14:09 -04001814 /*
1815 * make sure none of the code above managed to slip in a
1816 * delayed item
1817 */
1818 btrfs_assert_delayed_root_empty(root);
1819
Chris Mason2c90e5d2007-04-02 10:50:19 -04001820 WARN_ON(cur_trans != trans->transaction);
Chris Masondc17ff82008-01-08 15:46:30 -05001821
Chris Masone02119d2008-09-05 16:13:11 -04001822 /* btrfs_commit_tree_roots is responsible for getting the
1823 * various roots consistent with each other. Every pointer
1824 * in the tree of tree roots has to point to the most up to date
1825 * root for every subvolume and other tree. So, we have to keep
1826 * the tree logging code from jumping in and changing any
1827 * of the trees.
1828 *
1829 * At this point in the commit, there can't be any tree-log
1830 * writers, but a little lower down we drop the trans mutex
1831 * and let new people in. By holding the tree_log_mutex
1832 * from now until after the super is written, we avoid races
1833 * with the tree-log code.
1834 */
1835 mutex_lock(&root->fs_info->tree_log_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04001836
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001837 ret = commit_fs_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001838 if (ret) {
1839 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02001840 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001841 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001842 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001843
Qu Wenruo3818aea2014-01-13 13:36:06 +08001844 /*
David Sterba7e1876a2014-02-05 15:26:17 +01001845 * Since the transaction is done, we can apply the pending changes
1846 * before the next transaction.
Qu Wenruo3818aea2014-01-13 13:36:06 +08001847 */
David Sterba572d9ab2014-02-05 15:26:17 +01001848 btrfs_apply_pending_changes(root->fs_info);
1849
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001850 /* commit_fs_roots gets rid of all the tree log roots, it is now
Chris Masone02119d2008-09-05 16:13:11 -04001851 * safe to free the root of tree log roots
1852 */
1853 btrfs_free_log_root_tree(trans, root->fs_info);
1854
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001855 ret = commit_cowonly_roots(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001856 if (ret) {
1857 mutex_unlock(&root->fs_info->tree_log_mutex);
David Sterba871383b2012-04-02 18:31:37 +02001858 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001859 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001860 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001861
Miao Xie2cba30f2013-01-15 06:29:12 +00001862 /*
1863 * The tasks which save the space cache and inode cache may also
1864 * update ->aborted, check it.
1865 */
1866 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1867 ret = cur_trans->aborted;
1868 mutex_unlock(&root->fs_info->tree_log_mutex);
1869 mutex_unlock(&root->fs_info->reloc_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001870 goto scrub_continue;
Miao Xie2cba30f2013-01-15 06:29:12 +00001871 }
1872
Yan Zheng11833d62009-09-11 16:11:19 -04001873 btrfs_prepare_extent_commit(trans, root);
1874
Chris Mason78fae272007-03-25 11:35:08 -04001875 cur_trans = root->fs_info->running_transaction;
Chris Mason5f39d392007-10-15 16:14:19 -04001876
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001877 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1878 root->fs_info->tree_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04001879 list_add_tail(&root->fs_info->tree_root->dirty_list,
1880 &cur_trans->switch_commits);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001881
1882 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1883 root->fs_info->chunk_root->node);
Josef Bacik9e351cc2014-03-13 15:42:13 -04001884 list_add_tail(&root->fs_info->chunk_root->dirty_list,
1885 &cur_trans->switch_commits);
1886
1887 switch_commit_roots(cur_trans, root->fs_info);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001888
Jan Schmidtedf39272012-06-28 18:04:55 +02001889 assert_qgroups_uptodate(trans);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001890 update_super_roots(root);
Chris Masone02119d2008-09-05 16:13:11 -04001891
Josef Bacik60e7cd32013-09-30 14:10:43 -04001892 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1893 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
David Sterba6c417612011-04-13 15:41:04 +02001894 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1895 sizeof(*root->fs_info->super_copy));
Chris Masonccd467d2007-06-28 15:57:36 -04001896
Miao Xie935e5cc2014-09-03 21:35:33 +08001897 btrfs_update_commit_device_size(root->fs_info);
Miao Xiece7213c2014-09-03 21:35:34 +08001898 btrfs_update_commit_device_bytes_used(root, cur_trans);
Miao Xie935e5cc2014-09-03 21:35:33 +08001899
Filipe Manana656f30d2014-09-26 12:25:56 +01001900 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
1901 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
1902
Josef Bacika4abeea2011-04-11 17:25:13 -04001903 spin_lock(&root->fs_info->trans_lock);
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001904 cur_trans->state = TRANS_STATE_UNBLOCKED;
Josef Bacika4abeea2011-04-11 17:25:13 -04001905 root->fs_info->running_transaction = NULL;
Josef Bacika4abeea2011-04-11 17:25:13 -04001906 spin_unlock(&root->fs_info->trans_lock);
Chris Mason75857172011-06-13 20:00:16 -04001907 mutex_unlock(&root->fs_info->reloc_mutex);
Chris Masonb7ec40d2009-03-12 20:12:45 -04001908
Chris Masonf9295742008-07-17 12:54:14 -04001909 wake_up(&root->fs_info->transaction_wait);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001910
Chris Mason79154b12007-03-22 15:59:16 -04001911 ret = btrfs_write_and_wait_transaction(trans, root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001912 if (ret) {
1913 btrfs_error(root->fs_info, ret,
David Sterba08748812013-03-12 14:46:08 +00001914 "Error while writing out transaction");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001915 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001916 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001917 }
1918
1919 ret = write_ctree_super(trans, root, 0);
1920 if (ret) {
1921 mutex_unlock(&root->fs_info->tree_log_mutex);
Wang Shilong6cf7f772014-02-19 19:24:16 +08001922 goto scrub_continue;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001923 }
Chris Mason4313b392008-01-03 09:08:48 -05001924
Chris Masone02119d2008-09-05 16:13:11 -04001925 /*
1926 * the super is written, we can safely allow the tree-loggers
1927 * to go about their business
1928 */
1929 mutex_unlock(&root->fs_info->tree_log_mutex);
1930
Yan Zheng11833d62009-09-11 16:11:19 -04001931 btrfs_finish_extent_commit(trans, root);
Chris Mason4313b392008-01-03 09:08:48 -05001932
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001933 root->fs_info->last_trans_committed = cur_trans->transid;
Miao Xie4a9d8bd2013-05-17 03:53:43 +00001934 /*
1935 * We needn't acquire the lock here because there is no other task
1936 * which can change it.
1937 */
1938 cur_trans->state = TRANS_STATE_COMPLETED;
Chris Mason2c90e5d2007-04-02 10:50:19 -04001939 wake_up(&cur_trans->commit_wait);
Chris Mason3de45862008-11-17 21:02:50 -05001940
Josef Bacika4abeea2011-04-11 17:25:13 -04001941 spin_lock(&root->fs_info->trans_lock);
Josef Bacik13c5a932011-04-11 15:45:29 -04001942 list_del_init(&cur_trans->list);
Josef Bacika4abeea2011-04-11 17:25:13 -04001943 spin_unlock(&root->fs_info->trans_lock);
1944
Josef Bacik724e2312013-09-30 11:36:38 -04001945 btrfs_put_transaction(cur_trans);
1946 btrfs_put_transaction(cur_trans);
Josef Bacik58176a92007-08-29 15:47:34 -04001947
Miao Xie0860adf2013-05-15 07:48:27 +00001948 if (trans->type & __TRANS_FREEZABLE)
Miao Xie354aa0f2012-09-20 01:54:00 -06001949 sb_end_intwrite(root->fs_info->sb);
Jan Karab2b5ef52012-06-12 16:20:45 +02001950
liubo1abe9b82011-03-24 11:18:59 +00001951 trace_btrfs_transaction_commit(root);
1952
Arne Jansena2de7332011-03-08 14:14:00 +01001953 btrfs_scrub_continue(root);
1954
Josef Bacik9ed74f22009-09-11 16:12:44 -04001955 if (current->journal_info == trans)
1956 current->journal_info = NULL;
1957
Chris Mason2c90e5d2007-04-02 10:50:19 -04001958 kmem_cache_free(btrfs_trans_handle_cachep, trans);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001959
1960 if (current != root->fs_info->transaction_kthread)
1961 btrfs_run_delayed_iputs(root);
1962
Chris Mason79154b12007-03-22 15:59:16 -04001963 return ret;
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001964
Wang Shilong6cf7f772014-02-19 19:24:16 +08001965scrub_continue:
1966 btrfs_scrub_continue(root);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001967cleanup_transaction:
Josef Bacik0e721102012-06-26 16:13:18 -04001968 btrfs_trans_release_metadata(trans, root);
1969 trans->block_rsv = NULL;
Miao Xie272d26d2013-02-20 14:16:39 +00001970 if (trans->qgroup_reserved) {
1971 btrfs_qgroup_free(root, trans->qgroup_reserved);
1972 trans->qgroup_reserved = 0;
1973 }
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00001974 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001975 if (current->journal_info == trans)
1976 current->journal_info = NULL;
Josef Bacik7b8b92a2012-05-31 15:52:43 -04001977 cleanup_transaction(trans, root, ret);
Jeff Mahoney49b25e02012-03-01 17:24:58 +01001978
1979 return ret;
Chris Mason79154b12007-03-22 15:59:16 -04001980}
1981
Chris Masond352ac62008-09-29 15:18:18 -04001982/*
David Sterba9d1a2a32013-03-12 15:13:28 +00001983 * return < 0 if error
1984 * 0 if there are no more dead_roots at the time of call
1985 * 1 there are more to be processed, call me again
1986 *
1987 * The return value indicates there are certainly more snapshots to delete, but
1988 * if there comes a new one during processing, it may return 0. We don't mind,
1989 * because btrfs_commit_super will poke cleaner thread and it will process it a
1990 * few seconds later.
Chris Masond352ac62008-09-29 15:18:18 -04001991 */
David Sterba9d1a2a32013-03-12 15:13:28 +00001992int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
Chris Masone9d0b132007-08-10 14:06:19 -04001993{
David Sterba9d1a2a32013-03-12 15:13:28 +00001994 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001995 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone9d0b132007-08-10 14:06:19 -04001996
Josef Bacika4abeea2011-04-11 17:25:13 -04001997 spin_lock(&fs_info->trans_lock);
David Sterba9d1a2a32013-03-12 15:13:28 +00001998 if (list_empty(&fs_info->dead_roots)) {
1999 spin_unlock(&fs_info->trans_lock);
2000 return 0;
2001 }
2002 root = list_first_entry(&fs_info->dead_roots,
2003 struct btrfs_root, root_list);
Josef Bacikcfad3922013-07-25 15:11:47 -04002004 list_del_init(&root->root_list);
Josef Bacika4abeea2011-04-11 17:25:13 -04002005 spin_unlock(&fs_info->trans_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002006
Frank Holtonefe120a2013-12-20 11:37:06 -05002007 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
Jeff Mahoney2c536792011-10-03 23:22:41 -04002008
David Sterba9d1a2a32013-03-12 15:13:28 +00002009 btrfs_kill_all_delayed_nodes(root);
Yan, Zheng76dda932009-09-21 16:00:26 -04002010
David Sterba9d1a2a32013-03-12 15:13:28 +00002011 if (btrfs_header_backref_rev(root->node) <
2012 BTRFS_MIXED_BACKREF_REV)
2013 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2014 else
2015 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
David Sterba32471dc2014-02-05 02:03:47 +01002016
Josef Bacik6596a9282013-07-31 10:28:05 -04002017 return (ret < 0) ? 0 : 1;
Chris Masone9d0b132007-08-10 14:06:19 -04002018}
David Sterba572d9ab2014-02-05 15:26:17 +01002019
2020void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2021{
2022 unsigned long prev;
2023 unsigned long bit;
2024
2025 prev = cmpxchg(&fs_info->pending_changes, 0, 0);
2026 if (!prev)
2027 return;
2028
David Sterba7e1876a2014-02-05 15:26:17 +01002029 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2030 if (prev & bit)
2031 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2032 prev &= ~bit;
2033
2034 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2035 if (prev & bit)
2036 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2037 prev &= ~bit;
2038
David Sterba572d9ab2014-02-05 15:26:17 +01002039 if (prev)
2040 btrfs_warn(fs_info,
2041 "unknown pending changes left 0x%lx, ignoring", prev);
2042}